System and method for managing handoff of a client between different distributed-input-distributed-output (DIDO) networks based on detected velocity of the client

[0001] This application is a continuation-in-part of the following co-pending U.S. patent applications:

[0002] U.S. application Ser. No. 12/630,627, filed Dec. 3, 2009, entitled “System and Method For Distributed Antenna Wireless Communications”

[0003] U.S. application Ser. No. 12/143,503, filed Jun. 20, 2008 entitled “System and Method For Distributed Input-Distributed Output Wireless Communications”;

[0004] U.S. application Ser. No. 11/894,394, filed Aug. 20, 2007 entitled, “System and Method for Distributed Input Distributed Output Wireless Communications”;

[0005] U.S. application Ser. No. 11/894,362, filed Aug. 20, 2007 entitled, “System and method for Distributed Input-Distributed Wireless Communications”;

[0006] U.S. application Ser. No. 11/894,540, filed Aug. 20, 2007 entitled “System and Method For Distributed Input-Distributed Output Wireless Communications”

[0007] U.S. application Ser. No. 11/256,478, filed Oct. 21, 2005 entitled “System and Method For Spatial-Multiplexed Tropospheric Scatter Communications”;

[0008] U.S. application Ser. No. 10/817,731, filed Apr. 2, 2004 entitled “System and Method For Enhancing Near Vertical Incidence Skywave (“NVIS”) Communication Using Space-Time Coding.

[0009] Prior art multi-user wireless systems may include only a single base station or several base stations.

[0010] A single WiFi base station (e.g., utilizing 2.4 GHz 802.11b, g or n protocols) attached to a broadband wired Internet connection in an area where there are no other WiFi access points (e.g. a WiFi access point attached to DSL within a rural home) is an example of a relatively simple multi-user wireless system that is a single base station that is shared by one or more users that are within its transmission range. If a user is in the same room as the wireless access point, the user will typically experience a high-speed link with few transmission disruptions (e.g. there may be packet loss from 2.4 GHz interferers, like microwave ovens, but not from spectrum sharing with other WiFi devices), If a user is a medium distance away or with a few obstructions in the path between the user and WiFi access point, the user will likely experience a medium-speed link. If a user is approaching the edge of the range of the WiFi access point, the user will likely experience a low-speed link, and may be subject to periodic drop-outs if changes to the channel result in the signal SNR dropping below usable levels. And, finally, if the user is beyond the range of the WiFi base station, the user will have no link at all.

[0011] When multiple users access the WiFi base station simultaneously, then the available data throughput is shared among them. Different users will typically place different throughput demands on a WiFi base station at a given time, but at times when the aggregate throughput demands exceed the available throughput from the WiFi base station to the users, then some or all users will receive less data throughput than they are seeking. In an extreme situation where a WiFi access point is shared among a very large number of users, throughput to each user can slow down to a crawl, and worse, data throughput to each user may arrive in short bursts separated by long periods of no data throughput at all, during which time other users are served. This “choppy” data delivery may impair certain applications, like media streaming.

[0012] Adding additional WiFi base stations in situations with a large number of users will only help up to a point. Within the 2.4 GHz ISM band in the U.S., there are 3 non-interfering channels that can be used for WiFi, and if 3 WiFi base stations in the same coverage area are configured to each use a different non-interfering channel, then the aggregate throughput of the coverage area among multiple users will be increased up to a factor of 3. But, beyond that, adding more WiFi base stations in the same coverage area will not increase aggregate throughput, since they will start sharing the same available spectrum among them, effectually utilizing time-division multiplexed access (TDMA) by “taking turns” using the spectrum. This situation is often seen in coverage areas with high population density, such as within multi-dwelling units. For example, a user in a large apartment building with a WiFi adapter may well experience very poor throughput due to dozens of other interfering WiFi networks (e.g. in other apartments) serving other users that are in the same coverage area, even if the user's access point is in the same room as the client device accessing the base station. Although the link quality is likely good in that situation, the user would be receiving interference from neighbor WiFi adapters operating in the same frequency band, reducing the effective throughput to the user.

[0013] Current multiuser wireless systems, including both unlicensed spectrum, such as WiFi, and licensed spectrum, suffer from several limitations. These include coverage area, downlink (DL) data rate and uplink (UL) data rate. Key goals of next generation wireless systems, such as WiMAX and LTE, are to improve coverage area and DL and UL data rate via multiple-input multiple-output (MIMO) technology. MIMO employs multiple antennas at transmit and receive sides of wireless links to improve link quality (resulting in wider coverage) or data rate (by creating multiple non-interfering spatial channels to every user). If enough data rate is available for every user (note, the terms “user” and “client” are used herein interchangeably), however, it may be desirable to exploit channel spatial diversity to create non-interfering channels to multiple users (rather than single user), according to multiuser MIMO (MU-MIMO) techniques. See, e.g., the following references:

• [0014] G. Caire and S. Shamai, “On the achievable throughput of a multiantenna Gaussian broadcast channel,” IEEE Trans. Info. Th., vol. 49, pp. 1691-1706, July 2003.
• [0015] P. Viswanath and D. Tse, “Sum capacity of the vector Gaussian broadcast channel and uplink-downlink duality,” IEEE Trans. Info. Th., vol. 49, pp. 1912-1921, August 2003.
• [0016] S. Vishwanath, N. Jindal, and A. Goldsmith, “Duality, achievable rates, and sum-rate capacity of Gaussian MIMO broadcast channels,” IEEE Trans. Info. Th., vol. 49, pp. 2658-2668, October 2003.
• [0017] W. Yu and J. Cioffi, “Sum capacity of Gaussian vector broadcast channels,” IEEE Trans. Info. Th., vol. 50, pp. 1875-1892, September 2004.
• [0018] M. Costa, “Writing on dirty paper,” IEEE Transactions on Information Theory, vol. 29, pp. 439-441, May 1983.
• [0019] M. Bengtsson, “A pragmatic approach to multi-user spatial multiplexing,” Proc. of Sensor Array and Multichannel Sign. Proc. Workshop, pp. 130-134, August 2002.
• [0020] K.-K. Wong, R. D. Murch, and K. B. Letaief, “Performance enhancement of multiuser MIMO wireless communication systems,” IEEE Trans. Comm., vol. 50, pp. 1960-1970, December 2002.
• [0021] M. Sharif and B. Hassibi, “On the capacity of MIMO broadcast channel with partial side information,” IEEE Trans. Info. Th., vol. 51, pp. 506-522, February 2005.

[0022] For example, in MIMO 4×4 systems (i.e., four transmit and four receive antennas), 10 MHz bandwidth, 16-QAM modulation and forward error correction (FEC) coding with rate ¾ (yielding spectral efficiency of 3 bps/Hz), the ideal peak data rate achievable at the physical layer for every user is 4×30 Mbps=120 Mbps, which is much higher than required to deliver high definition video content (which may only require ˜10 Mbps). In MU-MIMO systems with four transmit antennas, four users and single antenna per user, in ideal scenarios (i.e., independent identically distributed, i.i.d., channels) downlink data rate may be shared across the four users and channel spatial diversity may be exploited to create four parallel 30 Mbps data links to the users. Different MU-MIMO schemes have been proposed as part of the LTE standard as described, for example, in 3GPP, “Multiple Input Multiple Output in UTRA”, 3GPP TR 25.876 V7.0.0, March 2007; 3GPP, “Base Physical channels and modulation”, TS 36.211, V8.7.0, May 2009; and 3GPP, “Multiplexing and channel coding”, TS 36.212, V8.7.0, May 2009. However, these schemes can provide only up to 2× improvement in DL data rate with four transmit antennas. Practical implementations of MU-MIMO techniques in standard and proprietary cellular systems by companies like ArrayComm (see, e.g., ArrayComm, “Field-proven results”, http://www.arraycomm.com/serve.php?page=proof) have yielded up to a ˜3× increase (with four transmit antennas) in DL data rate via space division multiple access (SDMA). A key limitation of MU-MIMO schemes in cellular networks is lack of spatial diversity at the transmit side. Spatial diversity is a function of antenna spacing and multipath angular spread in the wireless links. In cellular systems employing MU-MIMO techniques, transmit antennas at a base station are typically clustered together and placed only one or two wavelengths apart due to limited real estate on antenna support structures (referred to herein as “towers,” whether physically tall or not) and due to limitations on where towers may be located. Moreover, multipath angular spread is low since cell towers are typically placed high up (10 meters or more) above obstacles to yield wider coverage.

[0023] Other practical issues with cellular system deployment include excessive cost and limited availability of locations for cellular antenna locations (e.g. due to municipal restrictions on antenna placement, cost of real-estate, physical obstructions, etc.) and the cost and/or availability of network connectivity to the transmitters (referred to herein as “backhaul”). Further, cellular systems often have difficulty reaching clients located deeply in buildings due to losses from walls, ceilings, floors, furniture and other impediments.

[0024] Indeed, the entire concept of a cellular structure for wide-area network wireless presupposes a rather rigid placement of cellular towers, an alternation of frequencies between adjacent cells, and frequently sectorization, so as to avoid interference among transmitters (either base stations or users) that are using the same frequency. As a result, a given sector of a given cell ends up being a shared block of DL and UL spectrum among all of the users in the cell sector, which is then shared among these users primarily in only the time domain. For example, cellular systems based on Time Division Multiple Access (TDMA) and Code Division Multiple Access (CDMA) both share spectrum among users in the time domain. By overlaying such cellular systems with sectorization, perhaps a 2-3× spatial domain benefit can be achieved. And, then by overlaying such cellular systems with a MU-MIMO system, such as those described previously, perhaps another 2-3× space-time domain benefit can be achieved. But, given that the cells and sectors of the cellular system are typically in fixed locations, often dictated by where towers can be placed, even such limited benefits are difficult to exploit if user density (or data rate demands) at a given time does not match up well with tower/sector placement. A cellular smart phone user often experiences the consequence of this today where the user may be talking on the phone or downloading a web page without any trouble at all, and then after driving (or even walking) to a new location will suddenly see the voice quality drop or the web page slow to a crawl, or even lose the connection entirely. But, on a different day, the user may have the exact opposite occur in each location. What the user is probably experiencing, assuming the environmental conditions are the same, is the fact that user density (or data rate demands) is highly variable, but the available total spectrum (and thereby total data rate, using prior art techniques) to be shared among users at a given location is largely fixed.

[0025] Further, prior art cellular systems rely upon using different frequencies in different adjacent cells, typically 3 different frequencies. For a given amount of spectrum, this reduces the available data rate by 3×.

[0026] So, in summary, prior art cellular systems may lose perhaps 3× in spectrum utilization due to cellularization, and may improve spectrum utilization by perhaps 3× through sectorization and perhaps 3× more through MU-MIMO techniques, resulting in a net 3*3/3=3× potential spectrum utilization. Then, that bandwidth is typically divided up among users in the time domain, based upon what sector of what cell the users fall into at a given time. There are even further inefficiencies that result due to the fact that a given user's data rate demands are typically independent of the user's location, but the available data rate varies depending on the link quality between the user and the base station. For example, a user further from a cellular base station will typically have less available data rate than a user closer to a base station. Since the data rate is typically shared among all of the users in a given cellular sector, the result of this is that all users are impacted by high data rate demands from distant users with poor link quality (e.g. on the edge of a cell) since such users will still demand the same amount of data rate, yet they will be consuming more of the shared spectrum to get it.

[0027] Other proposed spectrum sharing systems, such as that used by WiFi (e.g., 802.11b, g, and n) and those proposed by the White Spaces Coalition, share spectrum very inefficiently since simultaneous transmissions by base stations within range of a user result in interference, and as such, the systems utilize collision avoidance and sharing protocols. These spectrum sharing protocols are within the time domain, and so, when there are a large number of interfering base stations and users, no matter how efficient each base station itself is in spectrum utilization, collectively the base stations are limited to time domain sharing of the spectrum among each other. Other prior art spectrum sharing systems similarly rely upon similar methods to mitigate interference among base stations (be they cellular base stations with antennas on towers or small scale base stations, such as WiFi Access Points (APs)). These methods include limiting transmission power from the base station so as to limit the range of interference, beamforming (via synthetic or physical means) to narrow the area of interference, time-domain multiplexing of spectrum and/or MU-MIMO techniques with multiple clustered antennas on the user device, the base station or both. And, in the case of advanced cellular networks in place or planned today, frequently many of these techniques are used at once.

[0028] But, what is apparent by the fact that even advanced cellular systems can achieve only about a 3× increase in spectrum utilization compared to a single user utilizing the spectrum is that all of these techniques have done little to increase the aggregate data rate among shared users for a given area of coverage. In particular, as a given coverage area scales in terms of users, it becomes increasingly difficult to scale the available data rate within a given amount of spectrum to keep pace with the growth of users. For example, with cellular systems, to increase the aggregate data rate within a given area, typically the cells are subdivided into smaller cells (often called nano-cells or femto-cells). Such small cells can become extremely expensive given the limitations on where towers can be placed, and the requirement that towers must be placed in a fairly structured pattern so as to provide coverage with a minimum of “dead zones”, yet avoid interference between nearby cells using the same frequencies. Essentially, the coverage area must be mapped out, the available locations for placing towers or base stations must be identified, and then given these constraints, the designers of the cellular system must make do with the best they can. And, of course, if user data rate demands grow over time, then the designers of the cellular system must yet again remap the coverage area, try to find locations for towers or base stations, and once again work within the constraints of the circumstances. And, very often, there simply is no good solution, resulting in dead zones or inadequate aggregate data rate capacity in a coverage area. In other words, the rigid physical placement requirements of a cellular system to avoid interference among towers or base stations utilizing the same frequency results in significant difficulties and constraints in cellular system design, and often is unable to meet user data rate and coverage requirements.

[0029] So-called prior art “cooperative” and “cognitive” radio systems seek to increase the spectral utilization in a given area by using intelligent algorithms within radios such that they can minimize interference among each other and/or such that they can potentially “listen” for other spectrum use so as to wait until the channel is clear. Such systems are proposed for use particularly in unlicensed spectrum in an effort to increase the spectrum utilization of such spectrum.

[0030] A mobile ad hoc network (MANET) (see http://en.wikipedia.org/wiki/Mobile_ad_hoc_network) is an example of a cooperative self-configuring network intended to provide peer-to-peer communications, and could be used to establish communication among radios without cellular infrastructure, and with sufficiently low-power communications, can potentially mitigate interference among simultaneous transmissions that are out of range of each other. A vast number of routing protocols have been proposed and implemented for MANET systems (see http://en.wikipedia.org/wiki/List_of_ad-hoc_routing_protocols for a list of dozens of routing protocols in a wide range of classes), but a common theme among them is they are all techniques for routing (e.g. repeating) transmissions in such a way to minimize transmitter interference within the available spectrum, towards the goal of particular efficiency or reliability paradigms.

[0031] All of the prior art multi-user wireless systems seek to improve spectrum utilization within a given coverage area by utilizing techniques to allow for simultaneous spectrum utilization among base stations and multiple users. Notably, in all of these cases, the techniques utilized for simultaneous spectrum utilization among base stations and multiple users achieve the simultaneous spectrum use by multiple users by mitigating interference among the waveforms to the multiple users. For example, in the case of 3 base stations each using a different frequency to transmit to one of 3 users, there interference is mitigated because the 3 transmissions are at 3 different frequencies. In the case of sectorization from a base station to 3 different users, each 180 degrees apart relative to the base station, interference is mitigated because the beamforming prevents the 3 transmissions from overlapping at any user.

[0032] When such techniques are augmented with MU-MIMO, and, for example, each base station has 4 antennas, then this has the potential to increase downlink throughput by a factor of 4, by creating four non-interfering spatial channels to the users in given coverage area. But it is still the case that some technique must be utilized to mitigate the interference among multiple simultaneous transmissions to multiple users in different coverage areas.

[0033] And, as previously discussed, such prior art techniques (e.g. cellularization, sectorization) not only typically suffer from increasing the cost of the multi-user wireless system and/or the flexibility of deployment, but they typically run into physical or practical limitations of aggregate throughput in a given coverage area. For example, in a cellular system, there may not be enough available locations to install more base stations to create smaller cells. And, in an MU-MIMO system, given the clustered antenna spacing at each base station location, the limited spatial diversity results in asymptotically diminishing returns in throughput as more antennas are added to the base station.

[0034] And further, in the case of multi-user wireless systems where the user location and density is unpredictable, it results in unpredictable (with frequently abrupt changes) in throughput, which is inconvenient to the user and renders some applications (e.g. the delivery of services requiring predictable throughput) impractical or of low quality. Thus, prior art multi-user wireless systems still leave much to be desired in terms of their ability to provide predictable and/or high-quality services to users.

[0035] Despite the extraordinary sophistication and complexity that has been developed for prior art multi-user wireless systems over time, there exist common themes: transmissions are distributed among different base stations (or ad hoc transceivers) and are structured and/or controlled so as to avoid the RF waveform transmissions from the different base stations and/or different ad hoc transceivers from interfering with each other at the receiver of a given user.

[0036] Or, to put it another way, it is taken as a given that if a user happens to receive transmissions from more than one base station or ad hoc transceiver at the same time, the interference from the multiple simultaneous transmissions will result in a reduction of the SNR and/or bandwidth of the signal to the user which, if severe enough, will result in loss of all or some of the potential data (or analog information) that would otherwise have been received by the user.

[0037] Thus, in a multiuser wireless system, it is necessary to utilize one or more spectrum sharing approaches or another to avoid or mitigate such interference to users from multiple base stations or ad hoc transceivers transmitting at the same frequency at the same time. There are a vast number of prior art approaches to avoiding such interference, including controlling base stations' physical locations (e.g. cellularization), limiting power output of base stations and/or ad hoc transceivers (e.g. limiting transmit range), beamforming/sectorization, and time domain multiplexing. In short, all of these spectrum sharing systems seek to address the limitation of multiuser wireless systems that when multiple base stations and/or ad hoc transceivers transmitting simultaneously at the same frequency are received by the same user, the resulting interference reduces or destroys the data throughput to the affected user. If a large percentage, or all, of the users in the multi-user wireless system are subject to interference from multiple base stations and/or ad hoc transceivers (e.g. in the event of the malfunction of a component of a multi-user wireless system), then it can result in a situation where the aggregate throughput of the multi-user wireless system is dramatically reduced, or even rendered non-functional.

[0038] Prior art multi-user wireless systems add complexity and introduce limitations to wireless networks and frequently result in a situation where a given user's experience (e.g. available bandwidth, latency, predictability, reliability) is impacted by the utilization of the spectrum by other users in the area. Given the increasing demands for aggregate bandwidth within wireless spectrum shared by multiple users, and the increasing growth of applications that can rely upon multi-user wireless network reliability, predictability and low latency for a given user, it is apparent that prior art multi-user wireless technology suffers from many limitations. Indeed, with the limited availability of spectrum suitable for particular types of wireless communications (e.g. at wavelengths that are efficient in penetrating building walls), it may be the case that prior art wireless techniques will be insufficient to meet the increasing demands for bandwidth that is reliable, predictable and low-latency.

[0039] Prior art related to the current invention describes beamforming systems and methods for null-steering in multiuser scenarios. Beamforming was originally conceived to maximize received signal-to-noise ratio (SNR) by dynamically adjusting phase and/or amplitude of the signals (i.e., beamforming weights) fed to the antennas of the array, thereby focusing energy toward the user's direction [14-20]. In multiuser scanarios, beamforming can be used to suppress interfering sources and maximize signal-to-interference-plus-noise ratio (SINR) [21-23]. For example, when beamforming is used at the receiver of a wireless link, the weights are computed to create nulls in the direction of the interfering sources [15]. When beamforming is used at the transmitter in multiuser downlink scenarios, the weights are calculated to pre-cancel inter-user interfence and maximize the SINR to every user [21-23]. Alternative techniques for multiuser systems, such as BD precoding [24-25], compute the precoding weights to maximize throughput in the downlink broadcast channel.

[0151] One solution to overcome many of the above prior art limitations is an embodiment of Distributed-Input Distributed-Output (DIDO) technology. DIDO technology is described in the following patents and patent applications, all of which are assigned the assignee of the present patent and are incorporated by reference. These patents and applications are sometimes referred to collectively herein as the “related patents and applications”:

[0152] U.S. application Ser. No. 12/630,627, filed Dec. 2, 2009, entitled “System and Method For Distributed Antenna Wireless Communications”

[0153] U.S. Pat. No. 7,599,420, filed Aug. 20, 2007, issued Oct. 6, 2009, entitled “System and Method for Distributed Input Distributed Output Wireless Communication”;

[0154] U.S. Pat. No. 7,633,994, filed Aug. 20, 2007, issued Dec. 15, 2009, entitled “System and Method for Distributed Input Distributed Output Wireless Communication”;

[0155] U.S. Pat. No. 7,636,381, filed Aug. 20, 2007, issued Dec. 22, 2009, entitled “System and Method for Distributed Input Distributed Output Wireless Communication”;

[0156] U.S. application Ser. No. 12/143,503, filed Jun. 20, 2008 entitled, “System and Method For Distributed Input-Distributed Output Wireless Communications”;

[0157] U.S. application Ser. No. 11/256,478, filed Oct. 21, 2005 entitled “System and Method For Spatial-Multiplexed Tropospheric Scatter Communications”;

[0158] U.S. Pat. No. 7,418,053, filed Jul. 30, 2004, issued Aug. 26, 2008, entitled “System and Method for Distributed Input Distributed Output Wireless Communication”;

[0159] U.S. application Ser. No. 10/817,731, filed Apr. 2, 2004 entitled “System and Method For Enhancing Near Vertical Incidence Skywave (“NVIS”) Communication Using Space-Time Coding.

[0160] For organization purposes, the present detailed description is separated into the following sections:

[0161] I. Disclosure of the Present Application which includes new matter introduced in the present application and includes FIGS. 80-113 and associated text;

[0162] II. Disclosure From Certain Related Applications which includes matter previously disclosed in certain related applications and which includes FIGS. 1-60 and associated text; and

[0163] III. Disclosure From U.S. application Ser. No. 12/630,627 which includes new matter introduced in the most recent related application in this series (having Ser. No. 12/630,627) and which includes FIGS. 61-79 and associated text.

[0164] Note that section I (Disclosure of the Present Application) utilizes its own set of endnotes which refer to prior art references and prior applications assigned to the assignee of the present application. The endnote citations are listed at the end of section I (just prior to the heading for Section II). Citations in Sections II and III may have numerical designations which overlap with those used in Section I even through these numerical designations identify different references (which are identified within each respective section). Thus, references identified by a particular numerical designation may be identified within the section in which the numerical designation is used.

[0165] Described below are wireless radio frequency (RF) communication systems and methods employing a plurality of distributed transmitting antennas to create locations in space with zero RF energy. When M transmit antennas are employed, it is possible to create up to (M−1) points of zero RF energy in predefined locations. In one embodiment of the invention, the points of zero RF energy are wireless devices and the transmit antennas are aware of the channel state information (CSI) between the transmitters and the receivers. In one embodiment, the CSI is computed at the receivers and fed back to the transmitters. In another embodiment, the CSI is computed at the transmitter via training from the receivers, assuming channel reciprocity is exploited. The transmitters may utilize the CSI to determine the interfering signals to be simultaneously transmitted. In one embodiment, block diagonalization (BD) precoding is employed at the transmit antennas to generate points of zero RF energy.

[0166] The system and methods described herein differ from the conventional receive/transmit beamforming techniques described above. In fact, receive beamforming computes the weights to suppress interference at the receive side (via null-steering), whereas some embodiments of the invention described herein apply weights at the transmit side to create interference patters that result in one or multiple locations in space with “zero RF energy.” Unlike conventional transmit beamforming or BD precoding designed to maximize signal quality (or SINR) to every user or downlink throughput, respectively, the systems and methods described herein minimize signal quality under certain conditions and/or from certain transmitters, thereby creating points of zero RF energy at the client devices (sometimes referred to herein as “users”). Moreover, in the context of distributed-input distributed-output (DIDO) systems (described in our related patents and applications), transmit antennas distributed in space provide higher degrees of freedom (i.e., higher channel spatial diversity) that can be exploited to create multiple points of zero RF energy and/or maximum SINR to different users. For example, with M transmit antennas it is possible to create up to (M−1) points of RF energy. By contrast, practical beamforming or BD multiuser systems are typically designed with closely spaced antennas at the transmit side that limit the number of simultaneous users that can be serviced over the wireless link, for any number of transmit antennas M.

[0167] Consider a system with M transmit antennas and K users, with K<M. We assume the transmitter is aware of the CSI (Hε⊂^K×M) between the M transmit antennas and K users. For simplicity, every user is assumed to be equipped with single antenna, but the same method can be extended to multiple receive antennas per user. The precoding weights (wε⊂^M×1) that create zero RF energy at the K users' locations are computed to satisfy the following condition

Hw=0^K×1

• • [0168] where 0^K×1 is the vector with all zero entries and H is the channel matrix obtained by combining the channel vectors (h_k ε⊂^1×M) from the M transmit antennas to the K users as

$H=\left[\begin{array}{c}{h}_1\\ ⋮\\ h_k\\ ⋮\\ h_K\end{array}\right].$

• • [0169] In one embodiment, singular value decomposition (SVD) of the channel matrix H is computed and the precoding weight w is defined as the right singular vector corresponding to the null subspace (identified by zero singular value) of H. The transmit antennas employ the weight vector defined above to transmit RF energy, while creating K points of zero RF energy at the locations of the K users such that the signal received at the k^th user is given by

r_k=h_kws_k+n_k=0+n_k

• • [0170] where n_k ε⊂^1×1 is the additive white Gaussian noise (AWGN) at the k^th user.
    In one embodiment, singular value decomposition (SVD) of the channel matrix H is computed and the precoding weight w is defined as the right singular vector corresponding to the null subspace (identified by zero singular value) of H.

[0171] In another embodiment, the wireless system is a DIDO system and points of zero RF energy are created to pre-cancel interference to the clients between different DIDO coverage areas. In U.S. application Ser. No. 12/630,627, a DIDO system is described which includes:

• • [0172] DIDO clients
  • [0173] DIDO distributed antennas
  • [0174] DIDO base transceiver stations (BTS)
  • [0175] DIDO base station network (BSN)
    Every BTS is connected via the BSN to multiple distributed antennas that provide service to given coverage area called DIDO cluster. In the present patent application we describe a system and method for removing interference between adjacent DIDO clusters. As illustrated in FIG. 77, we assume the main DIDO cluster hosts the client (i.e. a user device served by the multi-user DIDO system) affected by interference (or target client) from the neighbor clusters.

[0176] In one embodiment, neighboring clusters operate at different frequencies according to frequency division multiple access (FDMA) techniques similar to conventional cellular systems. For example, with frequency reuse factor of 3, the same carrier frequency is reused every third DIDO cluster as illustrated in FIG. 78. In FIG. 78, the different carrier frequencies are identified as F₁, F₂ and F₃. While this embodiment may be used in some implementations, this solution yields loss in spectral efficiency since the available spectrum is divided in multiple subbands and only a subset of DIDO clusters operate in the same subband. Moreover, it requires complex cell planning to associate different DIDO clusters to different frequencies, thereby preventing interference. Like prior art cellular systems, such cellular planning requires specific placement of antennas and limiting of transmit power to as to avoid interference between clusters using the same frequency.

[0177] In another embodiment, neighbor clusters operate in the same frequency band, but at different time slots according to time division multiple access (TDMA) technique. For example, as illustrated in FIG. 79 DIDO transmission is allowed only in time slots T₁, T₂, and T₃ for certain clusters, as illustrated. Time slots can be assigned equally to different clusters, such that different clusters are scheduled according to a Round-Robin policy. If different clusters are characterized by different data rate requirements (i.e., clusters in crowded urban environments as opposed to clusters in rural areas with fewer number of clients per area of coverage), different priorities are assigned to different clusters such that more time slots are assigned to the clusters with larger data rate requirements. While TDMA as described above may be employed in one embodiment of the invention, a TDMA approach may require time synchronization across different clusters and may result in lower spectral efficiency since interfering clusters cannot use the same frequency at the same time.

[0178] In one embodiment, all neighboring clusters transmit at the same time in the same frequency band and use spatial processing across clusters to avoid interference. In this embodiment, the multi-cluster DIDO system: (i) uses conventional DIDO precoding within the main cluster to transmit simultaneous non-interfering data streams within the same frequency band to multiple clients (such as described in the related patents and applications, including U.S. Pat. Nos. 7,599,420; 7,633,994; 7,636,381; and application Ser. No. 12/143,503); (ii) uses DIDO precoding with interference cancellation in the neighbor clusters to avoid interference to the clients lying in the interfering zones 8010 in FIG. 80, by creating points of zero radio frequency (RF) energy at the locations of the target clients. If a target client is in an interfering zone 8010, it will receive the sum of the RF containing the data stream from the main cluster 8011 and the zero RF energy from the interfering cluster 8012-8013, which will simply be the RF containing the data stream from the main cluster. Thus, adjacent clusters can utilize the same frequency simultaneously without target clients in the interfering zone suffering from interference.

[0179] In practical systems, the performance of DIDO precoding may be affected by different factors such as: channel estimation error or Doppler effects (yielding obsolete channel state information at the DIDO distributed antennas); intermodulation distortion (IMD) in multicarrier DIDO systems; time or frequency offsets. As a result of these effects, it may be impractical to achieve points of zero RF energy. However, as long as the RF energy at the target client from the interfering clusters is negligible compared to the RF energy from the main cluster, the link performance at the target client is unaffected by the interference. For example, let us assume the client requires 20 dB signal-to-noise ratio (SNR) to demodulate 4-QAM constellations using forward error correction (FEC) coding to achieve target bit error rate (BER) of 10⁻⁶. If the RF energy at the target client received from the interfering cluster is 20 dB below the RF energy received from the main cluster, the interference is negligible and the client can demodulate data successfully within the predefined BER target. Thus, the term “zero RF energy” as used herein does not necessarily mean that the RF energy from interfering RF signals is zero. Rather, it means that the RF energy is sufficiently low relative to the RF energy of the desired RF signal such that the desired RF signal may be received at the receiver. Moreover, while certain desirable thresholds for interfering RF energy relative to desired RF energy are described, the underlying principles of the invention are not limited to any particular threshold values.

[0180] There are different types of interfering zones 8010 as shown in FIG. 80. For example, “type A” zones (as indicated by the letter “A” in FIG. 80) are affected by interference from only one neighbor cluster, whereas “type B” zones (as indicated by the letter “B”) account for interference from two or multiple neighbor clusters.

[0181] FIG. 81 depicts a framework employed in one embodiment of the invention. The dots denote DIDO distributed antennas, the crosses refer to the DIDO clients and the arrows indicate the directions of propagation of RF energy. The DIDO antennas in the main cluster transmit precoded data signals to the clients MC 8101 in that cluster. Likewise, the DIDO antennas in the interfering cluster serve the clients IC 8102 within that cluster via conventional DIDO precoding. The green cross 8103 denotes the target client TC 8103 in the interfering zone. The DIDO antennas in the main cluster 8011 transmit precoded data signals to the target client (black arrows) via conventional DIDO precoding. The DIDO antennas in the interfering cluster 8012 use precoding to create zero RF energy towards the directions of the target client 8103 (green arrows).

[0182] The received signal at target client k in any interfering zone 8010A, B in FIG. 80 is given by

$\begin{array}{cc}{r}_k=H_kW_ks_k+H_k\sum _{\underset{u\ne k}{u=1}}^UW_us_u+\sum _{c=1}^CH_{c,k}\sum _{i=1}^{I_c}W_{c,i}s_{c,i}+n_k& \left(1\right)\end{array}$

where k=1, . . . , K, with K being the number of clients in the interfering zone 8010A, B, U is the number of clients in the main DIDO cluster, C is the number of interfering DIDO clusters 8012-8013 and I_c is the number of clients in the interfering cluster c. Moreover, r_k ε⊂^N×M is the vector containing the receive data streams at client k, assuming M transmit DIDO antennas and N receive antennas at the client devices; S_k ε⊂^N×1 is the vector of transmit data streams to client k in the main DIDO cluster; s_u ε⊂^N×1 is the vector of transmit data streams to client u in the main DIDO cluster; s_c,i ε⊂^N×1 is the vector of transmit data streams to client i in the c^th interfering DIDO cluster; n_k ε⊂^N×1 is the vector of additive white Gaussian noise (AWGN) at the N receive antennas of client k; H_k ε⊂^N×M is the DIDO channel matrix from the M transmit DIDO antennas to the N receive antennas at client k in the main DIDO cluster; H_c,k ε⊂^N×M is the DIDO channel matrix from the M transmit DIDO antennas to the N receive antennas t client k in the cth interfering DIDO cluster; W_k ε⊂^M×N is the matrix of DIDO precoding weights to client k in the main DIDO cluster; W_k ε⊂^M×N is the matrix of DIDO precoding weights to client u in the main DIDO cluster; W_c,i ε⊂^M×N is the matrix of DIDO precoding weights to client i in the c^th interfering DIDO cluster.

[0183] To simplify the notation and without loss of generality, we assume all clients are equipped with N receive antennas and there are M DIDO distributed antennas in every DIDO cluster, with M≧(N·U) and M≧(N·I_c), ∀c=1, . . . , C. If M is larger than the total number of receive antennas in the cluster, the extra transmit antennas are used to pre-cancel interference to the target clients in the interfering zone or to improve link robustness to the clients within the same cluster via diversity schemes described in the related patents and applications, including U.S. Pat. Nos. 7,599,420; 7,633,994; 7,636,381; and application Ser. No. 12/143,503.

[0184] The DIDO precoding weights are computed to pre-cancel inter-client interference within the same DIDO cluster. For example, block diagonalization (BD) precoding described in the related patents and applications, including U.S. Pat. Nos. 7,599,420; 7,633,994; 7,636,381; and application Ser. No. 12/143,503 and [7] can be used to remove inter-client interference, such that the following condition is satisfied in the main cluster

H_kW_u=0^N×N; ∀u=1, . . . , U; with u≠k.  (2)

The precoding weight matrices in the neighbor DIDO clusters are designed such that the following condition is satisfied

H_c,kW_c,i=0^N×N; ∀c=1, . . . , C and ∀i=1, . . . , I_c.  (3)

To compute the precoding matrices W_c,i, the downlink channel from the M transmit antennas to the I_c clients in the interfering cluster as well as to client k in the interfering zone is estimated and the precoding matrix is computed by the DIDO BTS in the interfering cluster. If BD method is used to compute the precoding matrices in the interfering clusters, the following effective channel matrix is built to compute the weights to the i^th client in the neighbor clusters

$\begin{array}{cc}{\stackrel{\_}{H}}_{c,i}=\left[\begin{array}{c}{H}_{c,k}\\ {\tilde{H}}_{c,i}\end{array}\right]& \left(4\right)\end{array}$

where {tilde over (H)}_c,i is the matrix obtained from the channel matrix H_cε⊂^(N−Ic)×M for the interfering cluster c, where the rows corresponding to the i^th client are removed. Substituting conditions (2) and (3) into (1), we obtain the received data streams for target client k, where intra-cluster and inter-cluster interference is removed

r_k=H_kW_ks_k+n_k.  (5)

The precoding weights W_c,i in (1) computed in the neighbor clusters are designed to transmit precoded data streams to all clients in those clusters, while pre-cancelling interference to the target client in the interfering zone. The target client receives precoded data only from its main cluster. In a different embodiment, the same data stream is sent to the target client from both main and neighbor clusters to obtain diversity gain. In this case, the signal model in (5) is expressed as

r_k=(H_kW_k+Σ_c=1^CH_c,kW_c,k)s_k+n_k  (6)

where W_c,k is the DIDO precoding matrix from the DIDO transmitters in the c^th cluster to the target client k in the interfering zone. Note that the method in (6) requires time synchronization across neighboring clusters, which may be complex to achieve in large systems, but nonetheless, is quite feasible if the diversity gain benefit justifies the cost of implementation.

[0185] We begin by evaluating the performance of the proposed method in terms of symbol error rate (SER) as a function of the signal-to-noise ratio (SNR). Without loss of generality, we define the following signal model assuming single antenna per client and reformulate (1) as

r_k=√{square root over (SNR)}+h_kw_ks_k+√{square root over (INR)}h_c,kΣ_i=1^Iw_c,is_c,i+n_k  (7)

where INR is the interference-to-noise ratio defined as INR=SNR/SIR and SIR is the signal-to-interference ratio.

[0186] FIG. 82 shows the SER as a function of the SNR, assuming SIR=10 dB for the target client in the interfering zone. Without loss of generality, we measured the SER for 4-QAM and 16-QAM without forwards error correction (FEC) coding. We fix the target SER to 1% for uncoded systems. This target corresponds to different values of SNR depending on the modulation order (i.e., SNR=20 dB for 4-QAM and SNR=28 dB for 16-QAM). Lower SER targets can be satisfied for the same values of SNR when using FEC coding due to coding gain. We consider the scenario of two clusters (one main cluster and one interfering cluster) with two DIDO antennas and two clients (equipped with single antenna each) per cluster. One of the clients in the main cluster lies in the interfering zone. We assume flat-fading narrowband channels, but the following results can be extended to frequency selective multicarrier (OFDM) systems, where each subcarrier undergoes flat-fading. We consider two scenarios: (i) one with inter-DIDO-cluster interference (IDCI) where the precoding weights w_c,i are computed without accounting for the target client in the interfering zone; and (ii) the other where the IDCI is removed by computing the weights w_c,i to cancel IDCI to the target client. We observe that in presence of IDCI the SER is high and above the predefined target. With IDCI-precoding at the neighbor cluster the interference to the target client is removed and the SER targets are reached for SNR>20 dB.

[0187] The results in FIG. 82 assumes IDCI-precoding as in (5). If IDCI-precoding at the neighbor clusters is also used to precode data streams to the target client in the interfering zone as in (6), additional diversity gain is obtained. FIG. 83 compares the SER derived from two techniques: (i) “Method 1” using the IDCI-precoding in (5); (ii) “Method 2” employing IDCI-precoding in (6) where the neighbor clusters also transmit precoded data stream to the target client. Method 2 yields ˜3 dB gain compared to conventional IDCI-precoding due to additional array gain provided by the DIDO antennas in the neighbor cluster used to transmit precoded data stream to the target client. More generally, the array gain of Method 2 over Method 1 is proportional to 10*log 10(C+1), where C is the number of neighbor clusters and the factor “1” refers to the main cluster.

[0188] Next, we evaluate the performance of the above method as a function of the target client's location with respect to the interfering zone. We consider one simple scenario where a target client 8401 moves from the main DIDO cluster 8402 to the interfering cluster 8403, as depicted in FIG. 84. We assume all DIDO antennas 8412 within the main cluster 8402 employ BD precoding to cancel intra-cluster interference to satisfy condition (2). We assume single interfering DIDO cluster, single receiver antenna at the client device 8401 and equal pathloss from all DIDO antennas in the main or interfering cluster to the client (i.e., DIDO antennas placed in circle around the client). We use one simplified pathloss model with pathloss exponent 4 (as in typical urban environments) [11].

[0189] The analysis hereafter is based on the following simplified signal model that extends (7) to account for pathloss

$\begin{array}{cc}{r}_k=\sqrt{\frac{\mathrm{SNR}·D_o^4}{D^4}}h_kw_ks_k+\sqrt{\frac{\mathrm{SNR}·D_o^4}{{\left(1-D\right)}^4}}h_{c,k}\sum _{i=1}^Iw_{c,i}s_{c,i}+n_k& \left(8\right)\end{array}$

where the signal-to-interference (SIR) is derived as SIR=((1−D)/D)⁴. In modeling the IDCI, we consider three scenarios: i) ideal case with no IDCI; ii) IDCI pre-cancelled via BD precoding in the interfering cluster to satisfy condition (3); iii) with IDCI, not pre-cancelled by the neighbor cluster.

[0190] FIG. 85 shows the signal-to-interference-plus-noise ratio (SINR) as a function of D (i.e., as the target client moves from the main cluster 8402 towards the DIDO antennas 8413 in the interfering cluster 8403). The SINR is derived as the ratio of signal power and interference plus noise power using the signal model in (8). We assume that D_o=0.1 and SNR=50 dB for D=D_o. In absence of IDCI the wireless link performance is only affected by noise and the SINR decreases due to pathloss. In presence of IDCI (i.e., without IDCI-precoding) the interference from the DIDO antennas in the neighbor cluster contributes to reduce the SINR.

[0191] FIG. 86 shows the symbol error rate (SER) performance of the three scenarios above for 4-QAM modulation in flat-fading narrowband channels. These SER results correspond to the SINR in FIG. 85. We assume SER threshold of 1% for uncoded systems (i.e., without FEC) corresponding to SINR threshold SINR_T=20 dB in FIG. 85. The SINR threshold depends on the modulation order used for data transmission. Higher modulation orders are typically characterized by higher SINR_T to achieve the same target error rate. With FEC, lower target SER can be achieved for the same SINR value due to coding gain. In case of IDCI without precoding, the target SER is achieved only within the range D<0.25. With IDCI-precoding at the neighbor cluster the range that satisfies the target SER is extended up to D<0.6. Beyond that range, the SINR increases due to pathloss and the SER target is not satisfied.

[0192] One embodiment of a method for IDCI precoding is shown in FIG. 87 and consists of the following steps:

• • [0193] SIR estimate 8701: Clients estimate the signal power from the main DIDO cluster (i.e., based on received precoded data) and the interference-plus-noise signal power from the neighbor DIDO clusters. In single-carrier DIDO systems, the frame structure can be designed with short periods of silence. For example, periods of silence can be defined between training for channel estimation and precoded data transmissions during channel state information (CSI) feedback. In one embodiment, the interference-plus-noise signal power from neighbor clusters is measured during the periods of silence from the DIDO antennas in the main cluster. In practical DIDO multicarrier (OFDM) systems, null tones are typically used to prevent direct current (DC) offset and attenuation at the edge of the band due to filtering at transmit and receive sides. In another embodiment employing multicarrier systems, the interference-plus-noise signal power is estimated from the null tones. Correction factors can be used to compensate for transmit/receive filter attenuation at the edge of the band. Once the signal-plus-interference-and-noise power (P_S) from the main cluster and the interference-plus-noise power from neighbor clusters (P_IN) are estimated, the client computes the SINR as

$\begin{array}{cc}\mathrm{SINR}=\frac{P_S-P_{\mathrm{IN}}}{P_{\mathrm{IN}}}.& \left(9\right)\end{array}$

• • [0194] Alternatively, the SINR estimate is derived from the received signal strength indication (RSSI) used in typical wireless communication systems to measure the radio signal power.
  • [0195] We observe the metric in (9) cannot discriminate between noise and interference power level. For example, clients affected by shadowing (i.e., behind obstacles that attenuate the signal power from all DIDO distributed antennas in the main cluster) in interference-free environments may estimate low SINR even though they are not affected by inter-cluster interference. A more reliable metric for the proposed method is the SIR computed as

$\begin{array}{cc}\mathrm{SIR}=\frac{P_S-P_{\mathrm{IN}}}{P_{\mathrm{IN}}-P_N}& \left(10\right)\end{array}$

• • [0196] where P_N is the noise power. In practical multicarrier OFDM systems, the noise power P_N in (10) is estimated from the null tones, assuming all DIDO antennas from main and neighbor clusters use the same set of null tones. The interference-plus-noise power (P_IN), is estimated from the period of silence as mentioned above. Finally, the signal-plus-interference-and-noise power (P_S) is derived from the data tones. From these estimates, the client computes the SIR in (10).
  • [0197] Channel estimation at neighbor clusters 8702-8703: If the estimated SIR in (10) is below predefined threshold (SIR_T), determined at 8702 in FIG. 87, the client starts listening to training signals from neighbor clusters. Note that SIR_T depends on the modulation and FEC coding scheme (MCS) used for data transmission. Different SIR targets are defined depending on the client's MCS. When DIDO distributed antennas from different clusters are time-synchronized (i.e., locked to the same pulse-per-second, PPS, time reference), the client exploits the training sequence to deliver its channel estimates to the DIDO antennas in the neighbor clusters at 8703. The training sequence for channel estimation in the neighbor clusters are designed to be orthogonal to the training from the main cluster. Alternatively, when DIDO antennas in different clusters are not time-synchronized, orthogonal sequences (with good cross-correlation properties) are used for time synchronization in different DIDO clusters. Once the client locks to the time/frequency reference of the neighbor clusters, channel estimation is carried out at 8703.
  • [0198] IDCI Precoding 8704: Once the channel estimates are available at the DIDO BTS in the neighbor clusters, IDCI-precoding is computed to satisfy the condition in (3). The DIDO antennas in the neighbor clusters transmit precoded data streams only to the clients in their cluster, while pre-cancelling interference to the clients in the interfering zone 8010 in FIG. 80. We observe that if the client lies in the type B interfering zone 8010 in FIG. 80, interference to the client is generated by multiple clusters and IDCI-precoding is carried out by all neighbor clusters at the same time.

[0199] Hereafter, we describe different handoff methods for clients that move across DIDO clusters populated by distributed antennas that are located in separate areas or that provide different kinds of services (i.e., low- or high-mobility services).

[0200] a. Handoff Between Adjacent DIDO Clusters

[0201] In one embodiment, the IDCI-precoder to remove inter-cluster interference described above is used as a baseline for handoff methods in DIDO systems. Conventional handoff in cellular systems is conceived for clients to switch seamlessly across cells served by different base stations. In DIDO systems, handoff allows clients to move from one cluster to another without loss of connection.

[0202] To illustrate one embodiment of a handoff strategy for DIDO systems, we consider again the example in FIG. 84 with only two clusters 8402 and 8403. As the client 8401 moves from the main cluster (C1) 8402 to the neighbor cluster (C2) 8403, one embodiment of a handoff method dynamically calculates the signal quality in different clusters and selects the cluster that yields the lowest error rate performance to the client.

[0203] FIG. 88 shows the SINR variation as a function of the client's distance from the center of clusters C1. For 4-QAM modulation without FEC coding, we consider target SINR=20 dB. The line identified by circles represents the SINR for the target client being served by the DIDO antennas in C1, when both C1 and C2 use DIDO precoding without interference cancellation. The SINR decreases as a function of D due to pathloss and interference from the neighboring cluster. When IDCI-precoding is implemented at the neighboring cluster, the SINR loss is only due to pathloss (as shown by the line with triangles), since interference is completely removed. Symmetric behavior is experienced when the client is served from the neighboring cluster. One embodiment of the handoff strategy is defined such that, as the client moves from C1 to C2, the algorithm switches between different DIDO schemes to maintain the SINR above predefined target.

[0204] From the plots in FIG. 88, we derive the SER for 4-QAM modulation in FIG. 89. We observe that, by switching between different precoding strategies, the SER is maintained within predefined target.

[0205] One embodiment of the handoff strategy is as follows.

• • [0206] C1-DIDO and C2-DIDO precoding: When the client lies within C1, away from the interfering zone, both clusters C1 and C2 operate with conventional DIDO precoding independently.
  • [0207] C1-DIDO and C2-IDCI precoding: As the client moves towards the interfering zone, its SIR or SINR degrades. When the target SINR_T1 is reached, the target client starts estimating the channel from all DIDO antennas in C2 and provides the CSI to the BTS of C2. The BTS in C2 computes IDCI-precoding and transmits to all clients in C2 while preventing interference to the target client. For as long as the target client is within the interfering zone, it will continue to provide its CSI to both C1 and C2.
  • [0208] C1-IDCI and C2-DIDO precoding: As the client moves towards C2, its SIR or SINR keeps decreasing until it again reaches a target. At this point the client decides to switch to the neighbor cluster. In this case, C1 starts using the CSI from the target client to create zero interference towards its direction with IDCI-precoding, whereas the neighbor cluster uses the CSI for conventional DIDO-precoding. In one embodiment, as the SIR estimate approaches the target, the clusters C1 and C2 try both DIDO- and IDCI-precoding schemes alternatively, to allow the client to estimate the SIR in both cases. Then the client selects the best scheme, to maximize certain error rate performance metric. When this method is applied, the cross-over point for the handoff strategy occurs at the intersection of the curves with triangles and rhombus in FIG. 88. One embodiment uses the modified IDCI-precoding method described in (6) where the neighbor cluster also transmits precoded data stream to the target client to provide array gain. With this approach the handoff strategy is simplified, since the client does not need to estimate the SINR for both strategies at the cross-over point.
  • [0209] C1-DIDO and C2-DIDO precoding: As the client moves out of the interference zone towards C2, the main cluster C1 stops pre-cancelling interference towards that target client via IDCI-precoding and switches back to conventional DIDO-precoding to all clients remaining in C1. This final cross-over point in our handoff strategy is useful to avoid unnecessary CSI feedback from the target client to C1, thereby reducing the overhead over the feedback channel. In one embodiment a second target SINR_T2 is defined. When the SINR (or SIR) increases above this target, the strategy is switched to C1-DIDO and C2-DIDO. In one embodiment, the cluster C1 keeps alternating between DIDO- and IDCI-precoding to allow the client to estimate the SINR. Then the client selects the method for C1 that more closely approaches the target SINR_T1 from above.

[0210] The method described above computes the SINR or SIR estimates for different schemes in real time and uses them to select the optimal scheme. In one embodiment, the handoff algorithm is designed based on the finite-state machine illustrated in FIG. 90. The client keeps track of its current state and switches to the next state when the SINR or SIR drops below or above the predefined thresholds illustrated in FIG. 88. As discussed above, in state 8801, both clusters C1 and C2 operate with conventional DIDO precoding independently and the client is served by cluster C1; in state 8802, the client is served by cluster C1, the BTS in C2 computes IDCI-precoding and cluster C1 operates using conventional DIDO precoding; in state 8803, the client is served by cluster C2, the BTS in C1 computes IDCI-precoding and cluster C2 operates using conventional DIDO precoding; and in state 8804, the client is served by cluster C2, and both clusters C1 and C2 operate with conventional DIDO precoding independently.

[0211] In presence of shadowing effects, the signal quality or SIR may fluctuate around the thresholds as shown in FIG. 91, causing repetitive switching between consecutive states in FIG. 90. Changing states repetitively is an undesired effect, since it results in significant overhead on the control channels between clients and BTSs to enable switching between transmission schemes. FIG. 91 depicts one example of a handoff strategy in the presence of shadowing. In one embodiment, the shadowing coefficient is simulated according to log-normal distribution with variance 3 [3]. Hereafter, we define some methods to prevent repetitive switching effect during DIDO handoff.

[0212] One embodiment of the invention employs a hysteresis loop to cope with state switching effects. For example, when switching between “C1-DIDO,C2-IDCI” 9302 and “C1-IDCI,C2-DIDO” 9303 states in FIG. 90 (or vice versa) the threshold SINR_T1 can be adjusted within the range A₁. This method avoids repetitive switches between states as the signal quality oscillates around SINR_T1. For example, FIG. 92 shows the hysteresis loop mechanism when switching between any two states in FIG. 90. To switch from state B to A the SIR must be larger than (SIR_T1+A₁/2), but to switch back from A to B the SIR must drop below (SIR_T1−A₁/2).

[0213] In a different embodiment, the threshold SINR_T2 is adjusted to avoid repetitive switching between the first and second (or third and fourth) states of the finite-state machine in FIG. 90. For example, a range of values A₂ may be defined such that the threshold SINR_T2 is chosen within that range depending on channel condition and shadowing effects.

[0214] In one embodiment, depending on the variance of shadowing expected over the wireless link, the SINR threshold is dynamically adjusted within the range [SINR_T2, SINR_T2+A₂]. The variance of the log-normal distribution can be estimated from the variance of the received signal strength (or RSSI) as the client moves from its current cluster to the neighbor cluster.

[0215] The methods above assume the client triggers the handoff strategy. In one embodiment, the handoff decision is deferred to the DIDO BTSs, assuming communication across multiple BTSs is enabled.

[0216] For simplicity, the methods above are derived assuming no FEC coding and 4-QAM. More generally, the SINR or SIR thresholds are derived for different modulation coding schemes (MCSs) and the handoff strategy is designed in combination with link adaptation (see, e.g., U.S. Pat. No. 7,636,381) to optimize downlink data rate to each client in the interfering zone.

[0217] b. Handoff Between Low- and High-Doppler DIDO Networks

[0218] DIDO systems employ closed-loop transmission schemes to precode data streams over the downlink channel. Closed-loop schemes are inherently constrained by latency over the feedback channel. In practical DIDO systems, computational time can be reduced by transceivers with high processing power and it is expected that most of the latency is introduced by the DIDO BSN, when delivering CSI and baseband precoded data from the BTS to the distributed antennas. The BSN can be comprised of various network technologies including, but not limited to, digital subscriber lines (DSL), cable modems, fiber rings, T1 lines, hybrid fiber coaxial (HFC) networks, and/or fixed wireless (e.g., WiFi). Dedicated fiber typically has very large bandwidth and low latency, potentially less than a millisecond in local region, but it is less widely deployed than DSL and cable modems. Today, DSL and cable modem connections typically have between 10-25 ms in last-mile latency in the United States, but they are very widely deployed.

[0219] The maximum latency over the BSN determines the maximum Doppler frequency that can be tolerated over the DIDO wireless link without performance degradation of DIDO precoding. For example, in [1] we showed that at the carrier frequency of 400 MHz, networks with latency of about 10 msec (i.e., DSL) can tolerate clients' velocity up to 8 mph (running speed), whereas networks with 1 msec latency (i.e., fiber ring) can support speed up to 70 mph (i.e., freeway traffic).

[0220] We define two or multiple DIDO sub-networks depending on the maximum Doppler frequency that can be tolerated over the BSN. For example, a BSN with high-latency DSL connections between the DIDO BTS and distributed antennas can only deliver low mobility or fixed-wireless services (i.e., low-Doppler network), whereas a low-latency BSN over a low-latency fiber ring can tolerate high mobility (i.e., high-Doppler network). We observe that the majority of broadband users are not moving when they use broadband, and further, most are unlikely to be located near areas with many high speed objects moving by (e.g., next to a highway) since such locations are typically less desirable places to live or operate an office. However, there are broadband users who will be using broadband at high speeds (e.g., while in a car driving on the highway) or will be near high speed objects (e.g., in a store located near a highway). To address these two differing user Doppler scenarios, in one embodiment, a low-Doppler DIDO network consists of a typically larger number of DIDO antennas with relatively low power (i.e., 1 W to 100 W, for indoor or rooftop installation) spread across a wide area, whereas a high-Doppler network consists of a typically lower number of DIDO antennas with high power transmission (i.e., 100 W for rooftop or tower installation). The low-Doppler DIDO network serves the typically larger number of low-Doppler users and can do so at typically lower connectivity cost using inexpensive high-latency broadband connections, such as DSL and cable modems. The high-Doppler DIDO network serves the typically fewer number of high-Doppler users and can do so at typically higher connectivity cost using more expensive low-latency broadband connections, such as fiber.

[0221] To avoid interference across different types of DIDO networks (e.g. low-Doppler and high-Doppler), different multiple access techniques can be employed such as: time division multiple access (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA).

[0222] Hereafter, we propose methods to assign clients to different types of DIDO networks and enable handoff between them. The network selection is based on the type of mobility of each client. The client's velocity (v) is proportional to the maximum Doppler shift according to the following equation [6]

$\begin{array}{cc}{f}_d=\frac{v}{\lambda }\sin \theta & \left(11\right)\end{array}$

where f_d is the maximum Doppler shift, λ is the wavelength corresponding to the carrier frequency and θ is the angle between the vector indicating the direction transmitter-client and the velocity vector.

[0223] In one embodiment, the Doppler shift of every client is calculated via blind estimation techniques. For example, the Doppler shift can be estimated by sending RF energy to the client and analyzing the reflected signal, similar to Doppler radar systems.

[0224] In another embodiment, one or multiple DIDO antennas send training signals to the client. Based on those training signals, the client estimates the Doppler shift using techniques such as counting the zero-crossing rate of the channel gain, or performing spectrum analysis. We observe that for fixed velocity v and client's trajectory, the angular velocity v sin θ in (11) may depend on the relative distance of the client from every DIDO antenna. For example, DIDO antennas in the proximity of a moving client yield larger angular velocity and Doppler shift than faraway antennas. In one embodiment, the Doppler velocity is estimated from multiple DIDO antennas at different distances from the client and the average, weighted average or standard deviation is used as an indicator for the client's mobility. Based on the estimated Doppler indicator, the DIDO BTS decides whether to assign the client to low- or high-Doppler networks.

[0225] The Doppler indicator is periodically monitored for all clients and sent back to the BTS. When one or multiple clients change their Doppler velocity (i.e., client riding in the bus versus client walking or sitting), those clients are dynamically re-assigned to different DIDO network that can tolerate their level of mobility.

[0226] Although the Doppler of low-velocity clients can be affected by being in the vicinity of high-velocity objects (e.g. near a highway), the Doppler is typically far less than the Doppler of clients that are in motion themselves. As such, in one embodiment, the velocity of the client is estimated (e.g. by using a means such as monitoring the clients position using GPS), and if the velocity is low, the client is assigned to a low-Doppler network, and if the velocity if high, the client is assigned to a high-Doppler network.

[0227] The block diagram of DIDO systems with power control is depicted in FIG. 93. One or multiple data streams (s_k) for every client (1, . . . , U) are first multiplied by the weights generated by the DIDO precoding unit. Precoded data streams are multiplied by power scaling factor computed by the power control unit, based on the input channel quality information (CQI). The CQI is either fed back from the clients to DIDO BTS or derived from the uplink channel assuming uplink-downlink channel reciprocity. The U precoded streams for different clients are then combined and multiplexed into M data streams (t_m), one for each of the M transmit antennas. Finally, the streams t_m are sent to the digital-to-analog converter (DAC) unit, the radio frequency (RF) unit, power amplifier (PA) unit and finally to the antennas.

[0228] The power control unit measures the CQI for all clients. In one embodiment, the CQI is the average SNR or RSSI. The CQI varies for different clients depending on pathloss or shadowing. Our power control method adjusts the transmit power scaling factors P_k for different clients and multiplies them by the precoded data streams generated for different clients. Note that one or multiple data streams may be generated for every client, depending on the number of clients' receive antennas.

[0229] To evaluate the performance of the proposed method, we defined the following signal model based on (5), including pathloss and power control parameters

r_k=√{square root over (SNRP_kα_k)}H_kW_ks_k+n_k  (12)

where k=1, . . . , U, U is the number of clients, SNR=P_o/N_o, with P_o being the average transmit power, N_o the noise power and α_k the pathloss/shadowing coefficient. To model pathloss/shadowing, we use the following simplified model

$\begin{array}{cc}{\alpha }_k={}^{-a\frac{k-1}{U}}& \left(13\right)\end{array}$

where a=4 is the pathloss exponent and we assume the pathloss increases with the clients' index (i.e., clients are located at increasing distance from the DIDO antennas).

[0230] FIG. 94 shows the SER versus SNR assuming four DIDO transmit antennas and four clients in different scenarios. The ideal case assumes all clients have the same pathloss (i.e., a=0), yielding P_k=1 for all clients. The plot with squares refers to the case where clients have different pathloss coefficients and no power control. The curve with dots is derived from the same scenario (with pathloss) where the power control coefficients are chosen such that P_k=1/α_k. With the power control method, more power is assigned to the data streams intended to the clients that undergo higher pathloss/shadowing, resulting in 9 dB SNR gain (for this particular scenario) compared to the case with no power control.

[0231] The Federal Communications Commission (FCC) (and other international regulatory agencies) defines constraints on the maximum power that can be transmitted from wireless devices to limit the exposure of human body to electromagnetic (EM) radiation. There are two types of limits [2]: i) “occupational/controlled” limit, where people are made fully aware of the radio frequency (RF) source via fences, warnings or labels; ii) “general population/uncontrolled” limit where there is no control over the exposure.

[0232] Different emission levels are defined for different types of wireless devices. In general, DIDO distributed antennas used for indoor/outdoor applications qualify for the FCC category of “mobile” devices, defined as [2]:

“transmitting devices designed to be used in other than fixed locations that would normally be used with radiating structures maintained 20 cm or more from the body of the user or nearby persons.”

[0233] The EM emission of “mobile” devices is measured in terms of maximum permissible exposure (MPE), expressed in mW/cm². FIG. 95 shows the MPE power density as a function of distance from the source of RF radiation for different values of transmit power at 700 MHz carrier frequency. The maximum allowed transmit power to meet the FCC “uncontrolled” limit for devices that typically operate beyond 20 cm from the human body is 1 W.

[0234] Less restrictive power emission constraints are defined for transmitters installed on rooftops or buildings, away from the “general population”. For these “rooftop transmitters” the FCC defines a looser emission limit of 1000 W, measured in terms of effective radiated power (ERP).

[0235] Based on the above FCC constraints, in one embodiment we define two types of DIDO distributed antennas for practical systems:

• • [0236] Low-power (LP) transmitters: located anywhere (i.e., indoor or outdoor) at any height, with maximum transmit power of 1 W and 5 Mbps consumer-grade broadband (e.g. DSL, cable modem, Fibe To The Home (FTTH)) backhaul connectivity.
  • [0237] High-power (HP) transmitters: rooftop or building mounted antennas at height of approximately 10 meters, with transmit power of 100 W and a commercial-grade broadband (e.g. optical fiber ring) backhaul (with effectively “unlimited” data rate compared to the throughput available over the DIDO wireless links).

[0238] Note that LP transmitters with DSL or cable modem connectivity are good candidates for low-Doppler DIDO networks (as described in the previous section), since their clients are mostly fixed or have low mobility. HP transmitters with commercial fiber connectivity can tolerate higher client's mobility and can be used in high-Doppler DIDO networks.

[0239] To gain practical intuition on the performance of DIDO systems with different types of LP/HP transmitters, we consider the practical case of DIDO antenna installation in downtown Palo Alto, Calif. FIG. 96a shows a random distribution of N_LP=100 low-power DIDO distributed antennas in Palo Alto. In FIG. 96b, 50 LP antennas are substituted with N_HP=50 high-power transmitters.

[0240] Based on the DIDO antenna distributions in FIGS. 96a-b, we derive the coverage maps in Palo Alto for systems using DIDO technology. FIGS. 97a and 97b show two power distributions corresponding to the configurations in FIG. 96a and FIG. 96b, respectively. The received power distribution (expressed in dBm) is derived assuming the pathloss/shadowing model for urban environments defined by the 3GPP standard [3] at the carrier frequency of 700 MHz. We observe that using 50% of HP transmitters yields better coverage over the selected area.

[0241] FIGS. 98a-b depict the rate distribution for the two scenarios above. The throughput (expressed in Mbps) is derived based on power thresholds for different modulation coding schemes defined in the 3GPP long-term evolution (LTE) standard in [4,5]. The total available bandwidth is fixed to 10 MHz at 700 MHz carrier frequency. Two different frequency allocation plans are considered: i) 5 MHz spectrum allocated only to the LP stations; ii) 9 MHz to HP transmitters and 1 MHz to LP transmitters. Note that lower bandwidth is typically allocated to LP stations due to their DSL backhaul connectivity with limited throughput. FIGS. 98a-b shows that when using 50% of HP transmitters it is possible to increase significantly the rate distribution, raising the average per-client data rate from 2.4 Mbps in FIGS. 98a to 38M bps in FIG. 98b.

[0242] Next, we defined algorithms to control power transmission of LP stations such that higher power is allowed at any given time, thereby increasing the throughput over the downlink channel of DIDO systems in FIG. 98b. We observe that the FCC limits on the power density is defined based on average over time as [2]

$\begin{array}{cc}S=\frac{\sum _{n=1}^NS_nt_n}{T_{\mathrm{MPE}}}& \left(14\right)\end{array}$

where T_MPE=Σ_n=1^Nt_n is the MPE averaging time, t_n is the period of time of exposure to radiation with power density S_n. For “controlled” exposure the average time is 6 minutes, whereas for “uncontrolled” exposure it is increased up to 30 minutes. Then, any power source is allowed to transmit at larger power levels than the MPE limits, as long as the average power density in (14) satisfies the FCC limit over 30 minute average for “uncontrolled” exposure.

[0243] Based on this analysis, we define adaptive power control methods to increase instantaneous per-antenna transmit power, while maintaining average power per DIDO antenna below MPE limits. We consider DIDO systems with more transmit antennas than active clients. This is a reasonable assumption given that DIDO antennas can be conceived as inexpensive wireless devices (similar to WiFi access points) and can be placed anywhere there is DSL, cable modem, optical fiber, or other Internet connectivity.

[0244] The framework of DIDO systems with adaptive per-antenna power control is depicted in FIG. 99. The amplitude of the digital signal coming out of the multiplexer 994 is dynamically adjusted with power scaling factors S₁, . . . , S_m, before being sent to the DAC units 995. The power scaling factors are computed by the power control unit 992 based on the CQI 993.

[0245] In one embodiment, N_g DIDO antenna groups are defined. Every group contains at least as many DIDO antennas as the number of active clients (K). At any given time, only one group has N_a>K active DIDO antennas transmitting to the clients at larger power level (S_o) than MPE limit ( MPE). One method iterates across all antenna groups according to Round-Robin scheduling policy depicted in FIG. 100. In another embodiment, different scheduling techniques (i.e., proportional-fair scheduling [8]) are employed for cluster selection to optimize error rate or throughput performance.

[0246] Assuming Round-Robin power allocation, from (14) we derive the average transmit power for every DIDO antenna as

$\begin{array}{cc}S=S_o\frac{t_o}{T_{\mathrm{MPE}}}\le \stackrel{\_}{\mathrm{MPE}}& \left(15\right)\end{array}$

where t_o is the period of time over which the antenna group is active and T_MPE=30 min is the average time defined by the FCC guidelines [2]. The ratio in (15) is the duty factor (DF) of the groups, defined such that the average transmit power from every DIDO antenna satisfies the MPE limit (MPE). The duty factor depends on the number of active clients, the number of groups and active antennas per-group, according to the following definition

$\begin{array}{cc}\mathrm{DF}\stackrel{\Delta }{=}\frac{\kappa }{N_gN_a}=\frac{t_o}{T_{\mathrm{MPE}}}.& \left(16\right)\end{array}$

The SNR gain (in dB) obtained in DIDO systems with power control and antenna grouping is expressed as a function of the duty factor as

$\begin{array}{cc}{G}_{dB}=10{\log }_{10}\left(\frac{1}{\mathrm{DF}}\right).& \left(17\right)\end{array}$

We observe the gain in (17) is achieved at the expense of G_dB additional transmit power across all DIDO antennas.
In general, the total transmit power from all N_a of all N_g groups is defined as

P=Σ_j=1^NgΣ_i=1^NaP_ij  (18)

where the P_ij is the average per-antenna transmit power given by

$\begin{array}{cc}{P}_{\mathrm{ij}}=\frac{1}{T_{\mathrm{MPE}}}{\int }_0^{T_{\mathrm{MPE}}}S_{\mathrm{ij}}\left(t\right)t\le \stackrel{\_}{\mathrm{MPE}}& \left(19\right)\end{array}$

and S_ij(t) is the power spectral density for the i^th transmit antenna within the j^th group. In one embodiment, the power spectral density in (19) is designed for every antenna to optimize error rate or throughput performance.

[0247] To gain some intuition on the performance of the proposed method, consider 400 DIDO distributed antennas in a given coverage area and 400 clients subscribing to a wireless Internet service offered over DIDO systems. It is unlikely that every Internet connection will be fully utilized all the time. Let us assume that 10% of the clients will be actively using the wireless Internet connection at any given time. Then, 400 DIDO antennas can be divided in N_g=10 groups of N_a=40 antennas each, every group serving K=40 active clients at any given time with duty factor DF=0.1. The SNR gain resulting from this transmission scheme is G_dB=10 log₁₀(1/DF)=10 dB, provided by 10 dB additional transmit power from all DIDO antennas. We observe, however, that the average per-antenna transmit power is constant and is within the MPE limit.

[0248] FIG. 101 compares the (uncoded) SER performance of the above power control with antenna grouping against conventional eigenmode selection in U.S. Pat. No. 7,636,381. All schemes use BD precoding with four clients, each client equipped with single antenna. The SNR refers to the ratio of per-transmit-antenna power over noise power (i.e., per-antenna transmit SNR). The curve denoted with DIDO 4×4 assumes four transmit antenna and BD precoding. The curve with squares denotes the SER performance with two extra transmit antennas and BD with eigenmode selection, yielding 10 dB SNR gain (at 1% SER target) over conventional BD precoding. Power control with antenna grouping and DF=1/10 yields 10 dB gain at the same SER target as well. We observe that eigenmode selection changes the slope of the SER curve due to diversity gain, whereas our power control method shifts the SER curve to the left (maintaining the same slope) due to increased average transmit power. For comparison, the SER with larger duty factor DF=1/50 is shown to provide additional 7 dB gain compared to DF=1/10.

[0249] Note that our power control may have lower complexity than conventional eigenmode selection methods. In fact, the antenna ID of every group can be pre-computed and shared among DIDO antennas and clients via lookup tables, such that only K channel estimates are required at any given time. For eigenmode selection, (K+2) channel estimates are computed and additional computational processing is required to select the eigenmode that minimizes the SER at any given time for all clients.

[0250] Next, we describe another method involving DIDO antenna grouping to reduce CSI feedback overhead in some special scenarios. FIG. 102a shows one scenario where clients (dots) are spread randomly in one area covered by multiple DIDO distributed antennas (crosses). The average power over every transmit-receive wireless link can be computed as

A={|H|²}.  (20)

where H is the channel estimation matrix available at the DIDO BTS.

[0251] The matrices A in FIGS. 102a-c are obtained numerically by averaging the channel matrices over 1000 instances. Two alternative scenarios are depicted in FIG. 102b and FIG. 102c, respectively, where clients are grouped together around a subset of DIDO antennas and receive negligible power from DIDO antennas located far away. For example, FIG. 102b shows two groups of antennas yielding block diagonal matrix A. One extreme scenario is when every client is very close to only one transmitter and the transmitters are far away from one another, such that the power from all other DIDO antennas is negligible. In this case, the DIDO link degenerates in multiple SISO links and A is a diagonal matrix as in FIG. 102c.

[0252] In all three scenarios above, the BD precoding dynamically adjusts the precoding weights to account for different power levels over the wireless links between DIDO antennas and clients. It is convenient, however, to identify multiple groups within the DIDO cluster and operate DIDO precoding only within each group. Our proposed grouping method yields the following advantages:

• • [0253] Computational gain: DIDO precoding is computed only within every group in the cluster. For example, if BD precoding is used, singular value decomposition (SVD) has complexity O(n³), where n is the minimum dimension of the channel matrix H. If H can be reduced to a block diagonal matrix, the SVD is computed for every block with reduced complexity. In fact, if the channel matrix is divided into two block matrices with dimensions n₁ and n₂ such that n=n₁+n₂, the complexity of the SVD is only O(n₁³)+O(n₂³)<O(n³). In the extreme case, if H is diagonal matrix, the DIDO link reduce to multiple SISO links and no SVD calculation is required.
  • [0254] Reduced CSI feedback overhead: When DIDO antennas and clients are divided into groups, in one embodiment, the CSI is computed from the clients to the antennas only within the same group. In TDD systems, assuming channel reciprocity, antenna grouping reduces the number of channel estimates to compute the channel matrix H. In FDD systems where the CSI is fed back over the wireless link, antenna grouping further yields reduction of CSI feedback overhead over the wireless links between DIDO antennas and clients.

[0255] In one embodiment of the invention, different multiple access techniques are defined for the DIDO uplink channel. These techniques can be used to feedback the CSI or transmit data streams from the clients to the DIDO antennas over the uplink. Hereafter, we refer to feedback CSI and data streams as uplink streams.

• • [0256] Multiple-input multiple-output (MIMO): the uplink streams are transmitted from the client to the DIDO antennas via open-loop MIMO multiplexing schemes. This method assumes all clients are time/frequency synchronized. In one embodiment, synchronization among clients is achieved via training from the downlink and all DIDO antennas are assumed to be locked to the same time/frequency reference clock. Note that variations in delay spread at different clients may generate jitter between the clocks of different clients that may affect the performance of MIMO uplink scheme. After the clients send uplink streams via MIMO multiplexing schemes, the receive DIDO antennas may use non-linear (i.e., maximum likelihood, ML) or linear (i.e., zeros-forcing, minimum mean squared error) receivers to cancel co-channel interference and demodulate the uplink streams individually.
  • [0257] Time division multiple access (TDMA): Different clients are assigned to different time slots. Every client sends its uplink stream when its time slot is available.
  • [0258] Frequency division multiple access (FDMA): Different clients are assigned to different carrier frequencies. In multicarrier (OFDM) systems, subsets of tones are assigned to different clients that transmit the uplink streams simultaneously, thereby reducing latency.
  • [0259] Code division multiple access (CDMA): Every client is assigned to a different pseudo-random sequence and orthogonality across clients is achieved in the code domain.

[0260] In one embodiment of the invention, the clients are wireless devices that transmit at much lower power than the DIDO antennas. In this case, the DIDO BTS defines client sub-groups based on the uplink SNR information, such that interference across sub-groups is minimized. Within every sub-group, the above multiple access techniques are employed to create orthogonal channels in time, frequency, space or code domains thereby avoiding uplink interference across different clients.

[0261] In another embodiment, the uplink multiple access techniques described above are used in combination with antenna grouping methods presented in the previous section to define different client groups within the DIDO cluster.

[0262] Link adaptation methods for DIDO systems exploiting time, frequency and space selectivity of wireless channels were defined in U.S. Pat. No. 7,636,381. Described below are embodiments of the invention for link adaptation in multicarrier (OFDM) DIDO systems that exploit time/frequency selectivity of wireless channels.

[0263] We simulate Rayleigh fading channels according to the exponentially decaying power delay profile (PDP) or Saleh-Valenzuela model in [9]. For simplicity, we assume single-cluster channel with multipath PDP defined as

P_n=e^−βn  (21)

where n=0, . . . , L−1, is the index of the channel tap, L is the number of channel taps and β=1/σ_DS is the PDP exponent that is an indicator of the channel coherence bandwidth, inverse proportional to the channel delay spread (σ_DS). Low values of β yield frequency-flat channels, whereas high values of β produce frequency selective channels. The PDP in (21) is normalized such that the total average power for all L channel taps is unitary

$\begin{array}{cc}\stackrel{\_}{P_n}=\frac{P_n}{\sum _{i=0}^{L-1}P_i}.& \left(22\right)\end{array}$

FIG. 103 depicts the amplitude of low frequency selective channels (assuming β=1) over delay domain or instantaneous PDP (upper plot) and frequency domain (lower plot) for DIDO 2×2 systems. The first subscript indicates the client, the second subscript the transmit antenna. High frequency selective channels (with β=0.1) are shown in FIG. 104.

[0264] Next, we study the performance of DIDO precoding in frequency selective channels. We compute the DIDO precoding weights via BD, assuming the signal model in (1) that satisfies the condition in (2). We reformulate the DIDO receive signal model in (5), with the condition in (2), as

r_k=H_eks_k+n_k.  (23)

[0265] where H_ek=H_kW_k is the effective channel matrix for user k. For DIDO 2×2, with a single antenna per client, the effective channel matrix reduces to one value with a frequency response shown in FIG. 105 and for channels characterized by high frequency selectivity (e.g., with β=0.1) in FIG. 104. The continuous line in FIG. 105 refers to client 1, whereas the line with dots refers to client 2. Based on the channel quality metric in FIG. 105 we define time/frequency domain link adaptation (LA) methods that dynamically adjust MCSs, depending on the changing channel conditions.

[0266] We begin by evaluating the performance of different MCSs in AWGN and Rayleigh fading SISO channels. For simplicity, we assume no FEC coding, but the following LA methods can be extended to systems that include FEC.

[0267] FIG. 106 shows the SER for different QAM schemes (i.e., 4-QAM, 16-QAM, 64-QAM). Without loss of generality, we assume target SER of 1% for uncoded systems. The SNR thresholds to meet that target SER in AWGN channels are 8 dB, 15.5 dB and 22 dB for the three modulation schemes, respectively. In Rayleigh fading channels, it is well known the SER performance of the above modulation schemes is worse than AWGN [13] and the SNR thresholds are: 18.6 dB, 27.3 dB and 34.1 dB, respectively. We observe that DIDO precoding transforms the multi-user downlink channel into a set of parallel SISO links. Hence, the same SNR thresholds as in FIG. 106 for SISO systems hold for DIDO systems on a client-by-client basis. Moreover, if instantaneous LA is carried out, the thresholds in AWGN channels are used.

[0268] The key idea of the proposed LA method for DIDO systems is to use low MCS orders when the channel undergoes deep fades in the time domain or frequency domain (depicted in FIG. 104) to provide link-robustness. Contrarily, when the channel is characterized by large gain, the LA method switches to higher MCS orders to increase spectral efficiency. One contribution of the present application compared to U.S. Pat. No. 7,636,381 is to use the effective channel matrix in (23) and in FIG. 105 as a metric to enable adaptation.

[0269] The general framework of the LA methods is depicted in FIG. 107 and defined as follows:

• • [0270] CSI estimation: At 1071 the DIDO BTS computes the CSI from all users. Users may be equipped with single or multiple receive antennas.
  • [0271] DIDO precoding: At 1072, the BTS computes the DIDO precoding weights for all users. In one embodiment, BD is used to compute these weights. The precoding weights are calculated on a tone-by-tone basis.
  • [0272] Link-quality metric calculation: At 1073 the BTS computes the frequency-domain link quality metrics. In OFDM systems, the metrics are calculated from the CSI and DIDO precoding weights for every tone. In one embodiment of the invention, the link-quality metric is the average SNR over all OFDM tones. We define this method as LA1 (based on average SNR performance). In another embodiment, the link quality metric is the frequency response of the effective channel in (23). We define this method as LA2 (based on tone-by-tone performance to exploit frequency diversity). If every client has single antenna, the frequency-domain effective channel is depicted in FIG. 105. If the clients have multiple receive antennas, the link-quality metric is defined as the Frobenius norm of the effective channel matrix for every tone. Alternatively, multiple link-quality metrics are defined for every client as the singular values of the effective channel matrix in (23).
  • [0273] Bit-loading algorithm: At 1074, based on the link-quality metrics, the BTS determines the MCSs for different clients and different OFDM tones. For LA1 method, the same MCS is used for all clients and all OFDM tones based on the SNR thresholds for Rayleigh fading channels in FIG. 106. For LA2, different MCSs are assigned to different OFDM tones to exploit channel frequency diversity.
  • [0274] Precoded data transmission: At 1075, the BTS transmits precoded data streams from the DIDO distributed antennas to the clients using the MCSs derived from the bit-loading algorithm. One header is attached to the precoded data to communicate the MCSs for different tones to the clients. For example, if eight MCSs are available and the OFDM symbols are defined with N=64 tone, log₂(8)*N=192 bits are required to communicate the current MCS to every client. Assuming 4-QAM (2 bits/symbol spectral efficiency) is used to map those bits into symbols, only 192/2/N=1.5 OFDM symbols are required to map the MCS information. In another embodiment, multiple subcarriers (or OFDM tones) are grouped into subbands and the same MCS is assigned to all tones in the same subband to reduce the overhead due to control information. Moreover, the MCS are adjusted based on temporal variations of the channel gain (proportional to the coherence time). In fixed-wireless channel (characterized by low Doppler effect) the MCS are recalculated every fraction of the channel coherence time, thereby reducing the overhead required for control information.

[0275] FIG. 108 shows the SER performance of the LA methods described above. For comparison, the SER performance in Rayleigh fading channels is plotted for each of the three QAM schemes used. The LA2 method adapts the MCSs to the fluctuation of the effective channel in the frequency domain, thereby providing 1.8 bps/Hz gain in spectral efficiency for low SNR (i.e., SNR=20 dB) and 15 dB gain in SNR (for SNR>35 dB) compared to LA1.

[0276] The computational complexity of DIDO systems is mostly localized at the centralized processor or BTS. The most computationally expensive operation is the calculation of the precoding weights for all clients from their CSI. When BD precoding is employed, the BTS has to carry out as many singular value decomposition (SVD) operations as the number of clients in the system. One way to reduce complexity is through parallelized processing, where the SVD is computed on a separate processor for every client.

[0277] In multicarrier DIDO systems, each subcarrier undergoes flat-fading channel and the SVD is carried out for every client over every subcarrier. Clearly the complexity of the system increases linearly with the number of subcarriers. For example, in OFDM systems with 1 MHz signal bandwidth, the cyclic prefix (L₀) must have at least eight channel taps (i.e., duration of 8 microseconds) to avoid intersymbol interference in outdoor urban macrocell environments with large delay spread [3]. The size (N_FFT) of the fast Fourier transform (FFT) used to generate the OFDM symbols is typically set to multiple of L₀ to reduce loss of data rate. If N_FFT=64, the effective spectral efficiency of the system is limited by a factor N_FFT/(N_FFT+L₀)=89%. Larger values of N_FFT yield higher spectral efficiency at the expense of higher computational complexity at the DIDO precoder.

[0278] One way to reduce computational complexity at the DIDO precoder is to carry out the SVD operation over a subset of tones (that we call pilot tones) and derive the precoding weights for the remaining tones via interpolation. Weight interpolation is one source of error that results in inter-client interference. In one embodiment, optimal weight interpolation techniques are employed to reduce inter-client interference, yielding improved error rate performance and lower computational complexity in multicarrier systems. In DIDO systems with M transmit antennas, U clients and N receive antennas per clients, the condition for the precoding weights of the k^th client (W_k) that guarantees zero interference to the other clients u is derived from (2) as

H_uW_k=0^N×N; ∀u=1, . . . , U; with u≠k  (24)

where H_u are the channel matrices corresponding to the other DIDO clients in the system.

[0279] In one embodiment of the invention, the objective function of the weight interpolation method is defined as

$\begin{array}{cc}f\left({\theta }_k\right)=\sum _{\underset{u\ne k}{u=1}}^U{H_u{\hat{W}}_k\left({\theta }_k\right)}_F& \left(25\right)\end{array}$

where θ_k is the set of parameters to be optimized for user k, Ŵ_k(θ_k) is the weight interpolation matrix and ∥·∥_F denotes the Frobenius norm of a matrix. The optimization problem is formulated as

θ_k,opt=arg min_θkΘkf(θ_k)  (26)

where Θ_k is the feasible set of the optimization problem and θ_{k, opt} is the optimal solution.

[0280] The objective function in (25) is defined for one OFDM tone. In another embodiment of the invention, the objective function is defined as linear combination of the Frobenius norm in (25) of the matrices for all the OFDM tones to be interpolated. In another embodiment, the OFDM spectrum is divided into subsets of tones and the optimal solution is given by

θ_k,opt=arg min_θkΘkmax_nAf(n,θ_k)  (27)

where n is the OFDM tone index and A is the subset of tones.

[0281] The weight interpolation matrix W_k(θ_k) in (25) is expressed as a function of a set of parameters θ_k. Once the optimal set is determined according to (26) or (27), the optimal weight matrix is computed. In one embodiment of the invention, the weight interpolation matrix of given OFDM tone n is defined as linear combination of the weight matrices of the pilot tones. One example of weight interpolation function for beamforming systems with single client was defined in [11]. In DIDO multi-client systems we write the weight interpolation matrix as

Ŵ_k(lN₀+n,θ_k)=(1−c_n)·W(l)+c_ne^jθk·W(l+1)  (28)

where 0≦l≦(L₀−1), L₀ is the number of pilot tones and c_n=(n−1)/N₀, with N₀=N_FFT/L₀. The weight matrix in (28) is then normalized such that ∥Ŵ_k∥_F=√{square root over (NM)} to guarantee unitary power transmission from every antenna. If N=1 (single receive antenna per client), the matrix in (28) becomes a vector that is normalized with respect to its norm. In one embodiment of the invention, the pilot tones are chosen uniformly within the range of the OFDM tones. In another embodiment, the pilot tones are adaptively chosen based on the CSI to minimize the interpolation error.

[0282] We observe that one key difference of the system and method in [11] against the one proposed in this patent application is the objective function. In particular, the systems in [11] assumes multiple transmit antennas and single client, so the related method is designed to maximize the product of the precoding weight by the channel to maximize the receive SNR for the client. This method, however, does not work in multi-client scenarios, since it yields inter-client interference due to interpolation error. By contrast, our method is designed to minimize inter-client interference thereby improving error rate performance to all clients.

[0283] FIG. 109 shows the entries of the matrix in (28) as a function of the OFDM tone index for DIDO 2×2 systems with N_FFT=64 and L₀=8. The channel PDP is generated according to the model in (21) with β=1 and the channel consists of only eight channel taps. We observe that L₀ must be chosen to be larger than the number of channel taps. The solid lines in FIG. 109 represent the ideal functions, whereas the dotted lines are the interpolated ones. The interpolated weights match the ideal ones for the pilot tones, according to the definition in (28). The weights computed over the remaining tones only approximate the ideal case due to estimation error.

[0284] One way to implement the weight interpolation method is via exhaustive search over the feasible set Θ_k in (26). To reduce the complexity of the search, we quantize the feasible set into P values uniformly in the range [0,2π]. FIG. 110 shows the SER versus SNR for L₀=8, M=N_t=2 transmit antennas and variable number of P. As the number of quantization levels increases, the SER performance improves. We observe the case P=10 approaches the performance of P=100 for much lower computational complexity, due to reduced number of searches.

[0285] FIG. 111 shows the SER performance of the interpolation method for different DIDO orders and L₀=16. We assume the number of clients is the same as the number of transmit antennas and every client is equipped with single antenna. As the number of clients increases the SER performance degrades due to increase inter-client interference produced by weight interpolation errors.

[0286] In another embodiment of the invention, weight interpolation functions other than those in (28) are used. For example, linear prediction autoregressive models [12] can be used to interpolate the weights across different OFDM tones, based on estimates of the channel frequency correlation.

• [0287] [1] A. Forenza and S. G. Perlman, “System and method for distributed antenna wireless communications”, U.S. application Ser. No. 12/630,627, filed Dec. 2, 2009, entitled “System and Method For Distributed Antenna Wireless Communications”
• [0288] [2] FCC, “Evaluating compliance with FCC guidelines for human exposure to radiofrequency electromagnetic fields,” OET Bulletin 65, Ed. 97-01, August 1997
• [0289] [3] 3GPP, “Spatial Channel Model AHG (Combined ad-hoc from 3GPP & 3GPP2)”, SCM Text V6.0, Apr. 22, 2003
• [0290] [4] 3GPP TR 25.912, “Feasibility Study for Evolved UTRA and UTRAN”, V9.0.0 (2009-10)
• [0291] [5] 3GPP TR 25.913, “Requirements for Evolved UTRA (E-UTRA) and Evolved UTRAN (E-UTRAN)”, V8.0.0 (2009-01)
• [0292] [6] W. C. Jakes, Microwave Mobile Communications, IEEE Press, 1974
• [0293] [7] K. K. Wong, et al., “A joint channel diagonalization for multiuser MIMO antenna systems,” IEEE Trans. Wireless Comm., vol. 2, pp. 773-786, July 2003;
• [0294] [8] P. Viswanath, et al., “Opportunistic beamforming using dump antennas,” IEEE Trans. On Inform. Theory, vol. 48, pp. 1277-1294, June 2002.
• [0295] [9] A. A. M. Saleh, et al., “A statistical model for indoor multipath propagation,” IEEE Jour. Select. Areas in Comm., vol. 195 SAC-5, no. 2, pp. 128-137, February 1987.
• [0296] [10] A. Paulraj, et al., Introduction to Space-Time Wireless Communications, Cambridge University Press, 40 West 20th Street, New York, N.Y., USA, 2003.
• [0297] [11] J. Choi, et al., “Interpolation Based Transmit Beamforming for MIMO-OFDM with Limited Feedback,” IEEE Trans. on Signal Processing, vol. 53, no. 11, pp. 4125-4135, November 2005.
• [0298] [12] I. Wong, et al., “Long Range Channel Prediction for Adaptive OFDM Systems,” Proc. of the IEEE Asilomar Conf. on Signals, Systems, and Computers, vol. 1, pp. 723-736, Pacific Grove, Calif., USA, Nov. 7-10, 2004.
• [0299] [13] J. G. Proakis, Communication System Engineering, Prentice Hall, 1994
• [0300] [14] B. D. Van Veen, et al., “Beamforming: a versatile approach to spatial filtering,” IEEE ASSP Magazine, April 1988.
• [0301] [15] R. G. Vaughan, “On optimum combining at the mobile,” IEEE Trans. On Vehic. Tech., vol 37, n. 4, pp. 181-188, November 1988
• [0302] [16] F. Qian, “Partially adaptive beamforming for correlated interference rejection,” IEEE Trans. On Sign. Proc., vol. 43, n. 2, pp. 506-515, February 1995
• [0303] [17]H. Krim, et. al., “Two decades of array signal processing research,” IEEE Signal Proc. Magazine, pp. 67-94, July 1996
• [0304] [19] W. R. Remley, “Digital beamforming system”,U.S. Pat. No. 4,003,016, January 1977
• [0305] [18] R. J. Masak, “Beamforming/null-steering adaptive array”, U.S. Pat. No. 4,771,289, September 1988
• [0306] [20] K.-B.Yu, et. al., “Adaptive digital beamforming architecture and algorithm for nulling mainlobe and multiple sidelobe radar jammers while preserving monopulse ratio angle estimation accuracy”, U.S. Pat. No. 5,600,326, February 1997
• [0307] [21]H. Boche, et al., “Analysis of different precoding/decoding strategies for multiuser beamforming”, IEEE Vehic. Tech. Conf., vol. 1, April 2003
• [0308] [22] M. Schubert, et al., “Joint ‘dirty paper’ pre-coding and downlink beamforming,” vol. 2, pp. 536-540, December 2002
• [0309] [23]H. Boche, et al. “A general duality theory for uplink and downlink beamformingc”, vol. 1, pp. 87-91, December 2002
• [0310] [24] K. K. Wong, R. D. Murch, and K. B. Letaief, “A joint channel diagonalization for multiuser MIMO antenna systems,” IEEE Trans. Wireless Comm., vol. 2, pp. 773-786, July 2003;
• [0311] [25] Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zero forcing methods for downlink spatial multiplexing in multiuser MIMO channels,” IEEE Trans. Sig. Proc., vol. 52, pp. 461-471, February 2004.

[0312] FIG. 1 shows a prior art MIMO system with transmit antennas 104 and receive antennas 105. Such a system can achieve up to 3× the throughput that would normally be achievable in the available channel. There are a number of different approaches in which to implement the details of such a MIMO system which are described in published literature on the subject, and the following explanation describes one such approach.

[0313] Before data is transmitted in the MIMO system of FIG. 1, the channel is “characterized.” This is accomplished by initially transmitting a “training signal” from each of the transmit antennas 104 to each of the receivers 105. The training signal is generated by the coding and modulation subsystem 102, converted to analog by a D/A converter (not shown), and then converted from baseband to RF by each transmitter 103, in succession. Each receive antenna 105 coupled to its RF Receiver 106 receives each training signal and converts it to baseband. The baseband signal is converted to digital by a D/A converter (not shown), and the signal processing subsystem 107 characterizes the training signal. Each signal's characterization may include many factors including, for example, phase and amplitude relative to a reference internal to the receiver, an absolute reference, a relative reference, characteristic noise, or other factors. Each signal's characterization is typically defined as a vector that characterizes phase and amplitude changes of several aspects of the signal when it is transmitted across the channel. For example, in a quadrature amplitude modulation (“QAM”)-modulated signal the characterization might be a vector of the phase and amplitude offsets of several multipath images of the signal. As another example, in an orthogonal frequency division multiplexing (“OFDM”)-modulated signal, it might be a vector of the phase and amplitude offsets of several or all of the individual sub-signals in the OFDM spectrum.

[0314] The signal processing subsystem 107 stores the channel characterization received by each receiving antenna 105 and corresponding receiver 106. After all three transmit antennas 104 have completed their training signal transmissions, then the signal processing subsystem 107 will have stored three channel characterizations for each of three receiving antennas 105, resulting in a 3×3 matrix 108, designated as the channel characterization matrix, “H.” Each individual matrix element is the channel characterization (which is typically a vector, as described above) of the training signal transmission of transmit antenna 104i as received by the receive antenna 105j.

[0315] At this point, the signal processing subsystem 107 inverts the matrix H 108, to produce H⁻¹, and awaits transmission of actual data from transmit antennas 104. Note that various prior art MIMO techniques described in available literature, can be utilized to ensure that the H matrix 108 can be inverted.

[0316] In operation, a payload of data to be transmitted is presented to the data Input subsystem 100. It is then divided up into three parts by splitter 101 prior to being presented to coding and modulation subsystem 102. For example, if the payload is the ASCII bits for “abcdef,” it might be divided up into three sub-payloads of ASCII bits for “ad,” “be,” and “cf” by Splitter 101. Then, each of these sub-payloads is presented individually to the coding and modulation subsystem 102.

[0317] Each of the sub-payloads is individually coded by using a coding system suitable for both statistical independence of each signal and error correction capability. These include, but are not limited to Reed-Solomon coding, Viterbi coding, and Turbo Codes. Finally, each of the three coded sub-payloads is modulated using an appropriate modulation scheme for the channel. Examples of modulation schemes are differential phase shift key (“DPSK”) modulation, 64-QAM modulation and OFDM. It should be noted here that the diversity gains provided by MIMO allow for higher-order modulation constellations that would otherwise be feasible in a SISO (Single Input-Single Output) system utilizing the same channel. Each coded and modulated signal is then transmitted through its own antenna 104 following D/A conversion by a D/A conversion unit (not shown) and RF generation by each transmitter 103.

[0318] Assuming that adequate spatial diversity exists amongst the transmit and receive antennas, each of the receiving antennas 105 will receive a different combination of the three transmitted signals from antennas 104. Each signal is received and converted down to baseband by each RF receiver 106, and digitized by an ND converter (not shown). If y_n is the signal received by the nth receive antenna 105, and x_n is the signal transmitted by nth transmit antenna 104, and N is noise, this can be described by the following three equations:

y₁=x₁H₁₁+x₂H₁₂+x₃H₁₃+N

y₂=x₁H₂₁+x₂H₂₂+x₃H₂₃+N

y₃=x₁H₃₁+x₂H₃₂+x₃H₃₃+N

[0319] Given that this is a system of three equations with three unknowns, it is a matter of linear algebra for the signal processing subsystem 107 to derive x₁, x₂, and x₃ (assuming that N is at a low enough level to permit decoding of the signals):

x₁=y₁H⁻¹₁₁+y₂H⁻¹₁₂+y₃H⁻¹₁₃

x₂=y₁H⁻¹₂₁+y₂H⁻¹₂₂+y₃H⁻¹₂₃

x₃=y₁H⁻¹₃₁+y₂H⁻¹₃₂+y₃H⁻¹₃₃

[0320] Once the three transmitted signals x_n are thus derived, they are then demodulated, decoded, and error-corrected by signal processing subsystem 107 to recover the three bit streams that were originally separated out by splitter 101. These bit streams are combined in combiner unit 108, and output as a single data stream from the data output 109. Assuming the robustness of the system is able to overcome the noise impairments, the data output 109 will produce the same bit stream that was introduced to the data Input 100.

[0321] Although the prior art system just described is generally practical up to four antennas, and perhaps up to as many as 10, for the reasons described in the Background section of this disclosure, it becomes impractical with large numbers of antennas (e.g. 25, 100, or 1000).

[0322] Typically, such a prior art system is two-way, and the return path is implemented exactly the same way, but in reverse, with each side of the communications channels having both transmit and receive subsystems.

[0323] FIG. 2 illustrates one embodiment of the invention in which a Base Station (BS) 200 is configured with a Wide Area Network (WAN) interface (e.g. to the Internet through a T1 or other high speed connection) 201 and is provisioned with a number (N) of antennas 202. For the time being, we use the term “Base Station” to refer to any wireless station that communicates wirelessly with a set of clients from a fixed location. Examples of Base Stations are access points in wireless local area networks (WLANs) or WAN antenna tower or antenna array. There are a number of Client Devices 203-207, each with a single antenna, which are served wirelessly from the Base Station 200. Although for the purposes of this example it is easiest to think about such a Base Station as being located in an office environment where it is serving Client Devices 203-207 that are wireless-network equipped personal computers, this architecture will apply to a large number of applications, both indoor and outdoor, where a Base Station is serving wireless clients. For example, the Base Station could be based at a cellular phone tower, or on a television broadcast tower. In one embodiment, the Base Station 200 is positioned on the ground and is configured to transmit upward at HF frequencies (e.g., frequencies up to 24 MHz) to bounce signals off the ionosphere as described in co-pending application entitled SYSTEM AND METHOD FOR ENHANCING NEAR VERTICAL INCIDENCE SKYWAVE (“NVIS”) COMMUNICATION USING SPACE-TIME CODING, Ser. No. 10/817,731, Filed Apr. 2, 2004, which is assigned to the assignee of the present application and which is incorporated herein by reference.

[0324] Certain details associated with the Base Station 200 and Client Devices 203-207 set forth above are for the purpose of illustration only and are not required for complying with the underlying principles of the invention. For example, the Base Station may be connected to a variety of different types of wide area networks via WAN interface 201 including application-specific wide area networks such as those used for digital video distribution. Similarly, the Client Devices may be any variety of wireless data processing and/or communication devices including, but not limited to cellular phones, personal digital assistants (“PDAs”), receivers, and wireless cameras.

[0325] In one embodiment, the Base Station's n Antennas 202 are separated spatially such that each is transmitting and receiving signals which are not spatially correlated, just as if the Base Station was a prior art MIMO transceiver. As described in the Background, experiments have been done where antennas placed within λ/6 (i.e. ⅙ wavelength) apart successfully achieve an increase in throughput from MIMO, but generally speaking, the further apart these Base Station antennas are placed, the better the system performance, and λ/2 is a desirable minimum. Of course, the underlying principles of the invention are not limited to any particular separation between antennas.

[0326] Note that a single Base Station 200 may very well have its antennas located very far apart. For example, in the HF spectrum, the antennas may be 10 meters apart or more (e.g., in an NVIS implementation mentioned above). If 100 such antennas are used, the Base Station's antenna array could well occupy several square kilometers.

[0327] In addition to spatial diversity techniques, one embodiment of the invention polarizes the signal in order to increase the effective throughput of the system. Increasing channel capacity through polarization is a well known technique which has been employed by satellite television providers for years. Using polarization, it is possible to have multiple (e.g., three) Base Station or users' antennas very close to each other, and still be not spatially correlated. Although conventional RF systems usually will only benefit from the diversity of two dimensions (e.g. x and y) of polarization, the architecture described herein may further benefit from the diversity of three dimensions of polarization (x, y and z).

[0328] In addition to space and polarization diversity, one embodiment of the invention employs antennas with near-orthogonal radiation patterns to improve link performance via pattern diversity. Pattern diversity can improve the capacity and error-rate performance of MIMO systems and its benefits over other antenna diversity techniques have been shown in the following papers:

• [0329] [13] L. Dong, H. Ling, and R. W. Heath Jr., “Multiple-input multiple-output wireless communication systems using antenna pattern diversity,” Proc. IEEE Glob. Telecom. Conf., vol. 1, pp. 997-1001, November 2002.
• [0330] [14] R. Vaughan, “Switched parasitic elements for antenna diversity,” IEEE Trans. Antennas Propagat., vol. 47, pp. 399-405, February 1999.
• [0331] [15] P. Mattheijssen, M. H. A. J. Herben, G. Dolmans, and L. Leyten, “Antenna-pattern diversity versus space diversity for use at handhelds,” IEEE Trans. on Veh. Technol., vol. 53, pp. 1035-1042, July 2004.
• [0332] [16] C. B. Dietrich Jr, K. Dietze, J. R. Nealy, and W. L. Stutzman, “Spatial, polarization, and pattern diversity for wireless handheld terminals,” Proc. IEEE Antennas and Prop. Symp., vol. 49, pp. 1271-1281, September 2001.
• [0333] [17] A. Forenza and R. W. Heath, Jr., “Benefit of Pattern Diversity Via 2-element Array of Circular Patch Antennas in Indoor Clustered MIMO Channels”, IEEE Trans. on Communications, vol. 54, no. 5, pp. 943-954, May 2006.

[0334] Using pattern diversity, it is possible to have multiple Base Station or users' antennas very close to each other, and still be not spatially correlated.

[0335] FIG. 3 provides additional detail of one embodiment of the Base Station 200 and Client Devices 203-207 shown in FIG. 2. For the purposes of simplicity, the Base Station 300 is shown with only three antennas 305 and only three Client Devices 306-308. It will be noted, however, that the embodiments of the invention described herein may be implemented with a virtually unlimited number of antennas 305 (i.e., limited only by available space and noise) and Client Devices 306-308.

[0336] FIG. 3 is similar to the prior art MIMO architecture shown in FIG. 1 in that both have three antennas on each sides of a communication channel. A notable difference is that in the prior art MIMO system the three antennas 105 on the right side of FIG. 1 are all a fixed distance from one another (e.g., integrated on a single device), and the received signals from each of the antennas 105 are processed together in the Signal Processing subsystem 107. By contrast, in FIG. 3, the three antennas 309 on the right side of the diagram are each coupled to a different Client Device 306-308, each of which may be distributed anywhere within range of the Base Station 305.

As such, the signal that each Client Device receives is processed independently from the other two received signals in its Coding, Modulation, Signal Processing subsystem 311. Thus, in contrast to a Multiple-Input (i.e. antennas 105) Multiple-Output (i.e. antennas 104) “MIMO” system, FIG. 3 illustrates a Multiple Input (i.e. antennas 305) Distributed Output (i.e. antennas 305) system, referred to hereinafter as a “MIDO” system.

[0337] Note that this application uses different terminology than previous applications, so as to better conform with academic and industry practices. In previously cited co-pending application, SYSTEM AND METHOD FOR ENHANCING NEAR VERTICAL INCIDENCE SKYWAVE (“NVIS”) COMMUNICATION USING SPACE-TIME CODING, Ser. No. 10/817,731, Filed Apr. 2, 2004, and application Ser. No. 10/902,978 filed Jul. 30, 2004 for which this is application is a continuation-in-part, the meaning of “Input” and “Output” (in the context of SIMO, MISO, DIMO and MIDO) is reversed from how the terms are used in this application. In the prior applications, “Input” referred to the wireless signals as they are input to the receiving antennas (e.g. antennas 309 in FIG. 3), and “Output” referred to the wireless signals as they are output by the transmitting antennas (e.g. antennas 305). In academia and the wireless industry, the reverse meaning of “Input” and “Output” is commonly used, in which “Input” refers to the wireless signals as they are input to the channel (i.e. the transmitted wireless signals from antennas 305) and “Output” refers to the wireless signals as they are output from the channel (i.e. wireless signals received by antennas 309). This application adopts this terminology, which is the reverse of the applications cited previously in this paragraph. Thus, the following terminology equivalences shall be drawn between applications:

10/817,731 and 10/902,978		Current Application
SIMO	=	MISO
MISO	=	SIMO
DIMO	=	MIDO
MIDO	=	DIMO

[0338] The MIDO architecture shown in FIG. 3 achieves a similar capacity increase as MIMO over a SISO system for a given number of transmitting antennas. However, one difference between MIMO and the particular MIDO embodiment illustrated in FIG. 3 is that, to achieve the capacity increase provided by multiple base station antennas, each MIDO Client Device 306-308 requires only a single receiving antenna, whereas with MIMO, each Client Device requires as least as many receiving antennas as the capacity multiple that is hoped to be achieved. Given that there is usually a practical limit to how many antennas can be placed on a Client Device (as explained in the Background), this typically limits MIMO systems to between four to ten antennas (and 4× to 10× capacity multiple). Since the Base Station 300 is typically serving many Client Devices from a fixed and powered location, is it practical to expand it to far more antennas than ten, and to separate the antennas by a suitable distance to achieve spatial diversity. As illustrated, each antenna is equipped with a transceiver 304 and a portion of the processing power of a Coding, Modulation, and Signal Processing section 303. Significantly, in this embodiment, no matter how much Base Station 300 is expanded, each Client Device 306-308 only will require one antenna 309, so the cost for an individual user Client Device 306-308 will be low, and the cost of Base Station 300 can be shared among a large base of users.

[0339] An example of how a MIDO transmission from the Base Station 300 to the Client Devices 306-308 can be accomplished is illustrated in FIGS. 4 through 6.

[0340] In one embodiment of the invention, before a MIDO transmission begins, the channel is characterized. As with a MIMO system, a training signal is transmitted (in the embodiment herein described), one-by-one, by each of the antennas 405. FIG. 4 illustrates only the first training signal transmission, but with three antennas 405 there are three separate transmissions in total. Each training signal is generated by the Coding, Modulation, and Signal Processing subsystem 403, converted to analog through a D/A converter, and transmitted as RF through each RF Transceiver 404. Various different coding, modulation and signal processing techniques may be employed including, but not limited to, those described above (e.g., Reed Solomon, Viterbi coding; QAM, DPSK, QPSK modulation, . . . etc).

[0341] Each Client Device 406-408 receives a training signal through its antenna 409 and converts the training signal to baseband by Transceiver 410. An ND converter (not shown) converts the signal to digital where is it processed by each Coding, Modulation, and Signal Processing subsystem 411. Signal characterization logic 320 then characterizes the resulting signal (e.g., identifying phase and amplitude distortions as described above) and stores the characterization in memory. This characterization process is similar to that of prior art MIMO systems, with a notable difference being that the each client device only computes the characterization vector for its one antenna, rather than for n antennas. For example, the Coding Modulation and Signal Processing subsystem 420 of client device 406 is initialized with a known pattern of the training signal (either at the time of manufacturing, by receiving it in a transmitted message, or through another initialization process). When antenna 405 transmits the training signal with this known pattern, Coding Modulation and Signal Processing subsystem 420 uses correlation methods to find the strongest received pattern of the training signal, it stores the phase and amplitude offset, then it subtracts this pattern from the received signal. Next, it finds then second strongest received pattern that correlates to the training signal, it stores the phase and amplitude offset, then it subtracts this second strongest pattern from the received signal. This process continues until either some fixed number of phase and amplitude offsets are stored (e.g. eight), or a detectable training signal pattern drops below a given noise floor. This vector of phase/amplitude offsets becomes element H₁₁ of the vector 413. Simultaneously, Coding Modulation and Signal Processing subsystems for Client Devices 407 and 408 implement the produce their vector elements H₂₁ and H₃₁.

[0342] The memory in which the characterization is stored may be a non-volatile memory such as a Flash memory or a hard drive and/or a volatile memory such as a random access memory (e.g., SDRAM, RDAM). Moreover, different Client Devices may concurrently employ different types of memories to store the characterization information (e.g., PDA's may use Flash memory whereas notebook computers may use a hard drive). The underlying principles of the invention are not limited to any particular type of storage mechanism on the various Client Devices or the Base Station.

[0343] As mentioned above, depending on the scheme employed, since each Client Device 406-408 has only one antenna, each only stores a 1×3 row 413-415 of the H matrix. FIG. 4 illustrates the stage after the first training signal transmission where the first column of 1×3 rows 413-415 has been stored with channel characterization information for the first of the three Base Station antennas 405. The remaining two columns are stored following the channel characterization of the next two training signal transmissions from the remaining two base station antennas. Note that for the sake of illustration the three training signals are transmitted at separate times. If the three training signal patterns are chosen such as not to be correlated to one another, they may be transmitted simultaneously, thereby reducing training time.

[0344] As indicated in FIG. 5, after all three pilot transmissions are complete, each Client Device 506-508 transmits back to the Base Station 500 the 1×3 row 513-515 of matrix H that it has stored. To the sake of simplicity, only one Client Device 506 is illustrated transmitting its characterization information in FIG. 5. An appropriate modulation scheme (e.g. DPSK, 64QAM, OFDM) for the channel combined with adequate error correction coding (e.g. Reed Solomon, Viterbi, and/or Turbo codes) may be employed to make sure that the Base Station 500 receives the data in the rows 513-515 accurately.

[0345] Although all three antennas 505 are shown receiving the signal in FIG. 5, it is sufficient for a single antenna and transceiver of the Base Station 500 to receive each 1×3 row 513-515 transmission. However, utilizing many or all of antennas 505 and Transceivers 504 to receive each transmission (i.e., utilizing prior art Single-Input Multiple-Output (“SIMO”) processing techniques in the Coding, Modulation and Signal Processing subsystem 503) may yield a better signal-to-noise ratio (“SNR”) than utilizing a single antenna 505 and Transceiver 504 under certain conditions.

[0346] As the Coding, Modulation and Signal Processing subsystem 503 of Base Station 500 receives the 1×3 row 513-515, from each Client Device 507-508, it stores it in a 3×3H matrix 516. As with the Client Devices, the Base Station may employ various different storage technologies including, but not limited to non-volatile mass storage memories (e.g., hard drives) and/or volatile memories (e.g., SDRAM) to store the matrix 516. FIG. 5 illustrates a stage at which the Base Station 500 has received and stored the 1×3 row 513 from Client Device 509. The 1×3 rows 514 and 515 may be transmitted and stored in H matrix 516 as they are received from the remaining Client Devices, until the entire H matrix 516 is stored.

[0347] One embodiment of a MIDO transmission from a Base Station 600 to Client Devices 606-608 will now be described with reference to FIG. 66. Because each Client Device 606-608 is an independent device, typically each device is receiving a different data transmission. As such, one embodiment of a Base Station 600 includes a Router 602 communicatively positioned between the WAN Interface 601 and the Coding, Modulation and Signal Processing subsystem 603 that sources multiple data streams (formatted into bit streams) from the WAN interface 601 and routes them as separate bit streams u₁-u₃ intended for each Client Device 606-608, respectively. Various well known routing techniques may be employed by the router 602 for this purpose.

[0348] The three bit streams, u₁-u₃, shown in FIG. 6 are then routed into the Coding, Modulation and Signal Processing subsystem 603 and coded into statistically distinct, error correcting streams (e.g. using Reed Solomon, Viterbi, or Turbo Codes) and modulated using an appropriate modulation scheme for the channel (such as DPSK, 64QAM or OFDM). In addition, the embodiment illustrated in FIG. 6 includes signal precoding logic 630 for uniquely coding the signals transmitted from each of the antennas 605 based on the signal characterization matrix 616. More specifically, rather than routing each of the three coded and modulated bit streams to a separate antenna (as is done in FIG. 1), in one embodiment, the precoding logic 630 multiplies the three bit streams u₁-u₃ in FIG. 6 by the inverse of the H matrix 616, producing three new bit streams, u′₁-u′₃. The three precoded bit streams are then converted to analog by D/A converters (not shown) and transmitted as RF by Transceivers 604 and antennas 605.

[0349] Before explaining how the bit streams are received by the Client Devices 606-608, the operations performed by the precoding module 630 will be described. Similar to the MIMO example from FIG. 1 above, the coded and modulated signal for each of the three source bit streams will be designated with u_n. In the embodiment illustrated in FIG. 6, each u_i contains the data from one of the three bit streams routed by the Router 602, and each such bit stream is intended for one of the three Client Devices 606-608.

[0350] However, unlike the MIMO example of FIG. 1, where each x_i is transmitted by each antenna 104, in the embodiment of the invention illustrated in FIG. 6, each u_i is received at each Client Device antenna 609 (plus whatever noise N there is in the channel). To achieve this result, the output of each of the three antennas 605 (each of which we will designate as v) is a function of u_i and the H matrix that characterizes the channel for each Client Device. In one embodiment, each v_i is calculated by the precoding logic 630 within the Coding, Modulation and Signal Processing subsystem 603 by implementing the following formulas:

v₁=u₁H⁻¹₁₁+u₂H⁻¹₁₂+u₃H⁻¹₁₃

v₂=u₁H⁻¹₂₁+u₂H⁻¹₂₂+u₃H⁻¹₂₃

v₃=u₁H⁻¹₃₁+u₂H⁻¹₃₂+u₃H⁻¹₃₃

[0351] Thus, unlike MIMO, where each x, is calculated at the receiver after the signals have been transformed by the channel, the embodiments of the invention described herein solve for each v_i at the transmitter before the signals have been transformed by the channel. Each antenna 609 receives u_i already separated from the other u_n−1 bit streams intended for the other antennas 609. Each Transceiver 610 converts each received signal to baseband, where it is digitized by an ND converter (now shown), and each Coding, Modulation and Signal Processing subsystem 611, demodulates and decodes the x_i bit stream intended for it, and sends its bit stream to a Data Interface 612 to be used by the Client Device (e.g., by an application on the client device).

[0352] The embodiments of the invention described herein may be implemented using a variety of different coding and modulation schemes. For example, in an OFDM implementation, where the frequency spectrum is separated into a plurality of sub-bands, the techniques described herein may be employed to characterize each individual sub-band. As mentioned above, however, the underlying principles of the invention are not limited to any particular modulation scheme.

[0353] If the Client Devices 606-608 are portable data processing devices such as PDAs, notebook computers, and/or wireless telephones the channel characterization may change frequently as the Client Devices may move from one location to another. As such, in one embodiment of the invention, the channel characterization matrix 616 at the Base Station is continually updated. In one embodiment, the Base Station 600 periodically (e.g., every 250 milliseconds) sends out a new training signal to each Client Device, and each Client Device continually transmits its channel characterization vector back to the Base Station 600 to ensure that the channel characterization remains accurate (e.g. if the environment changes so as to affect the channel or if a Client Device moves). In one embodiment, the training signal is interleaved within the actual data signal sent to each client device. Typically, the training signals are much lower throughput than the data signals, so this would have little impact on the overall throughput of the system. Accordingly, in this embodiment, the channel characterization matrix 616 may be updated continuously as the Base Station actively communicates with each Client Device, thereby maintaining an accurate channel characterization as the Client Devices move from one location to the next or if the environment changes so as to affect the channel.

[0354] One embodiment of the invention illustrated in FIG. 7 employs MIMO techniques to improve the upstream communication channel (i.e., the channel from the Client Devices 706-708 to the Base Station 700). In this embodiment, the channel from each of the Client Devices is continually analyzed and characterized by upstream channel characterization logic 741 within the Base Station. More specifically, each of the Client Devices 706-708 transmits a training signal to the Base Station 700 which the channel characterization logic 741 analyzes (e.g., as in a typical MIMO system) to generate an N×M channel characterization matrix 741, where N is the number of Client Devices and M is the number of antennas employed by the Base Station. The embodiment illustrated in FIG. 7 employs three antennas 705 at the Base Station and three Client Devices 706-608, resulting in a 3×3 channel characterization matrix 741 stored at the Base Station 700. The MIMO upstream transmission illustrated in FIG. 7 may be used by the Client Devices both for transmitting data back to the Base Station 700, and for transmitting channel characterization vectors back to the Base Station 700 as illustrated in FIG. 5. But unlike the embodiment illustrated in FIG. 5 in which each Client Device's channel characterization vector is transmitted at a separate time, the method shown in FIG. 7 allows for the simultaneous transmission of channel characterization vectors from multiple Client Devices back to the Base Station 700, thereby dramatically reducing the channel characterization vectors' impact on return channel throughput.

[0355] As mentioned above, each signal's characterization may include many factors including, for example, phase and amplitude relative to a reference internal to the receiver, an absolute reference, a relative reference, characteristic noise, or other factors. For example, in a quadrature amplitude modulation (“QAM”)-modulated signal the characterization might be a vector of the phase and amplitude offsets of several multipath images of the signal. As another example, in an orthogonal frequency division multiplexing (“OFDM”)-modulated signal, it might be a vector of the phase and amplitude offsets of several or all of the individual sub-signals in the OFDM spectrum. The training signal may be generated by each Client Device's coding and modulation subsystem 711, converted to analog by a D/A converter (not shown), and then converted from baseband to RF by each Client Device's transmitter 709. In one embodiment, in order to ensure that the training signals are synchronized, Client Devices only transmit training signals when requested by the Base Station (e.g., in a round robin manner). In addition, training signals may be interleaved within or transmitted concurrently with the actual data signal sent from each client device. Thus, even if the Client Devices 706-708 are mobile, the training signals may be continuously transmitted and analyzed by the upstream channel characterization logic 741, thereby ensuring that the channel characterization matrix 741 remains up-to-date.

[0356] The total channel capacity supported by the foregoing embodiments of the invention may be defined as min (N, M) where M is the number of Client Devices and N is the number of Base Station antennas. That is, the capacity is limited by the number of antennas on either the Base Station side or the Client side. As such, one embodiment of the invention employs synchronization techniques to ensure that no more than min (N, M) antennas are transmitting/receiving at a given time.

[0357] In a typical scenario, the number of antennas 705 on the Base Station 700 will be less than the number of Client Devices 706-708. An exemplary scenario is illustrated in FIG. 8 which shows five Client Devices 804-808 communicating with a base station having three antennas 802. In this embodiment, after determining the total number of Client Devices 804-808, and collecting the necessary channel characterization information (e.g., as described above), the Base Station 800 chooses a first group of three clients 810 with which to communicate (three clients in the example because min (N, M)=3). After communicating with the first group of clients 810 for a designated period of time, the Base Station then selects another group of three clients 811 with which to communicate. To distribute the communication channel evenly, the Base Station 800 selects the two Client Devices 807, 808 which were not included in the first group. In addition, because an extra antenna is available, the Base Station 800 selects an additional client device 806 included in the first group. In one embodiment, the Base Station 800 cycles between groups of clients in this manner such that each client is effectively allocated the same amount of throughput over time. For example, to allocate throughput evenly, the Base Station may subsequently select any combination of three Client Devices which excludes Client Device 806 (i.e., because Client Device 806 was engaged in communication with the Base Station for the first two cycles).

[0358] In one embodiment, in addition to standard data communications, the Base Station may employ the foregoing techniques to transmit training signals to each of the Client Devices and receive training signals and signal characterization data from each of the Client Devices.

[0359] In one embodiment, certain Client Devices or groups of client devices may be allocated different levels of throughput. For example, Client Devices may be prioritized such that relatively higher priority Client Devices may be guaranteed more communication cycles (i.e., more throughput) than relatively lower priority client devices. The “priority” of a Client Device may be selected based on a number of variables including, for example, the designated level of a user's subscription to the wireless service (e.g., user's may be willing to pay more for additional throughput) and/or the type of data being communicated to/from the Client Device (e.g., real-time communication such as telephony audio and video may take priority over non-real time communication such as email).

[0360] In one embodiment of the Base Station dynamically allocates throughput based on the Current Load required by each Client Device. For example, if Client Device 804 is streaming live video and the other devices 805-808 are performing non-real time functions such as email, then the Base Station 800 may allocate relatively more throughput to this client 804. It should be noted, however, that the underlying principles of the invention are not limited to any particular throughput allocation technique.

[0361] As illustrated in FIG. 9, two Client Devices 907, 908 may be so close in proximity, that the channel characterization for the clients is effectively the same. As a result, the Base Station will receive and store effectively equivalent channel characterization vectors for the two Client Devices 907, 908 and therefore will not be able to create unique, spatially distributed signals for each Client Device. Accordingly, in one embodiment, the Base Station will ensure that any two or more Client Devices which are in close proximity to one another are allocated to different groups. In FIG. 9, for example, the Base Station 900 first communicates with a first group 910 of Client Devices 904, 905 and 908; and then with a second group 911 of Client Devices 905, 906, 907, ensuring that Client Devices 907 and 908 are in different groups.

[0362] Alternatively, in one embodiment, the Base Station 900 communicates with both Client Devices 907 and 908 concurrently, but multiplexes the communication channel using known channel multiplexing techniques. For example, the Base Station may employ time division multiplexing (“TDM”), frequency division multiplexing (“FDM”) or code division multiple access (“CDMA”) techniques to divide the single, spatially-correlated signal between Client Devices 907 and 908.

[0363] Although each Client Device described above is equipped with a single antenna, the underlying principles of the invention may be employed using Client Devices with multiple antennas to increase throughput. For example, when used on the wireless systems described above, a client with 2 antennas will realize a 2× increase in throughput, a client with 3 antennas will realize a 3× increase in throughput, and so on (i.e., assuming that the spatial and angular separation between the antennas is sufficient). The Base Station may apply the same general rules when cycling through Client Devices with multiple antennas. For example, it may treat each antenna as a separate client and allocate throughput to that “client” as it would any other client (e.g., ensuring that each client is provided with an adequate or equivalent period of communication).

[0364] As mentioned above, one embodiment of the invention employs the MIDO and/or MIMO signal transmission techniques described above to increase the signal-to-noise ratio and throughput within a Near Vertical Incidence Skywave (“NVIS”) system. Referring to FIG. 10, in one embodiment of the invention, a first NVIS station 1001 equipped with a matrix of N antennas 1002 is configured to communicate with M client devices 1004. The NVIS antennas 1002 and antennas of the various client devices 1004 transmit signals upward to within about 15 degrees of vertical in order to achieve the desired NVIS and minimize ground wave interference effects. In one embodiment, the antennas 1002 and client devices 1004, support multiple independent data streams 1006 using the various MIDO and MIMO techniques described above at a designated frequency within the NVIS spectrum (e.g., at a carrier frequency at or below 23 MHz, but typically below 10 MHz), thereby significantly increasing the throughput at the designated frequency (i.e., by a factor proportional to the number of statistically independent data streams).

[0365] The NVIS antennas serving a given station may be physically very far apart from each other. Given the long wavelengths below 10 MHz and the long distance traveled for the signals (as much as 300 miles round trip), physical separation of the antennas by 100s of yards, and even miles, can provide advantages in diversity. In such situations, the individual antenna signals may be brought back to a centralized location to be processed using conventional wired or wireless communications systems. Alternatively, each antenna can have a local facility to process its signals, then use conventional wired or wireless communications systems to communicate the data back to a centralized location. In one embodiment of the invention, NVIS Station 1001 has a broadband link 1015 to the Internet 1010 (or other wide area network), thereby providing the client devices 1003 with remote, high speed, wireless network access.

[0366] In one embodiment, the Base Station and/or users may exploit polarization/pattern diversity techniques described above to reduce the array size and/or users' distance while providing diversity and increased throughput. As an example, in MIDO systems with HF transmissions, the users may be in the same location and yet their signals be uncorrelated because of polarization/pattern diversity. In particular, by using pattern diversity, one user may be communicating to the Base Station via groundwave whereas the other user via NVIS.

[0367] One embodiment of the invention employs a system and method to compensate for in-phase and quadrature (I/Q) imbalance in distributed-input distributed-output (DIDO) systems with orthogonal frequency division multiplexing (OFDM). Briefly, according to this embodiment, user devices estimate the channel and feedback this information to the Base Station; the Base Station computes the precoding matrix to cancel inter-carrier and inter-user interference caused by I/Q imbalance; and parallel data streams are transmitted to multiple user devices via DIDO precoding; the user devices demodulate data via zero-forcing (ZF), minimum mean-square error (MMSE) or maximum likelihood (ML) receiver to suppress residual interference.

[0368] As described in detail below, some of the significant features of this embodiment of the invention include, but are not limited to:

[0369] Precoding to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch) in OFDM systems;

[0370] Precoding to cancel inter-user interference and ICI (due to I/Q mismatch) in DIDO-OFDM systems;

[0371] Techniques to cancel ICI (due to I/Q mismatch) via ZF receiver in DIDO-OFDM systems employing block diagonalization (BD) precoder;

[0372] Techniques to cancel inter-user interference and ICI (due to I/Q mismatch) via precoding (at the transmitter) and a ZF or MMSE filter (at the receiver) in DIDO-OFDM systems;

[0373] Techniques to cancel inter-user interference and ICI (due to I/Q mismatch) via pre-coding (at the transmitter) and a nonlinear detector like a maximum likelihood (ML) detector (at the receiver) in DIDO-OFDM systems;

[0374] The use of pre-coding based on channel state information to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch) in OFDM systems;

[0375] The use of pre-coding based on channel state information to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch) in DIDO-OFDM systems;

[0376] The use of an I/Q mismatch aware DIDO precoder at the station and an IQ-aware DIDO receiver at the user terminal;

[0377] The use of an I/Q mismatch aware DIDO precoder at the station, an I/Q aware DIDO receiver at the user terminal, and an I/Q aware channel estimator;

[0378] The use of an I/Q mismatch aware DIDO precoder at the station, an I/Q aware DIDO receiver at the user terminal, an I/Q aware channel estimator, and an I/Q aware DIDO feedback generator that sends channel state information from the user terminal to the station;

[0379] The use of an I/Q mismatch-aware DIDO precoder at the station and an I/Q aware DIDO configurator that uses I/Q channel information to perform functions including user selection, adaptive coding and modulation, space-time-frequency mapping, or precoder selection;

[0380] The use of an I/Q aware DIDO receiver that cancels ICI (due to I/Q mismatch) via ZF receiver in DIDO-OFDM systems employing block diagonalization (BD) precoder;

[0381] The use of an I/Q aware DIDO receiver that cancels ICI (due to I/Q mismatch) via pre-coding (at the transmitter) and a nonlinear detector like a maximum likelihood detector (at the receiver) in DIDO-OFDM systems; and

[0382] The use of an I/Q aware DIDO receiver that cancels ICI (due to I/Q mismatch) via ZF or MMSE filter in DIDO-OFDM systems.

[0383] The transmit and receive signals of typical wireless communication systems consist of in-phase and quadrature (I/Q) components. In practical systems, the inphase and quadrature components may be distorted due to imperfections in the mixing and baseband operations. These distortions manifest as I/Q phase, gain and delay mismatch. Phase imbalance is caused by the sine and cosine in the modulator/demodulator not being perfectly orthogonal. Gain imbalance is caused by different amplifications between the inphase and quadrature components. There may be an additional distortion, called delay imbalance, due to difference in delays between the I- and Q-rails in the analog circuitry.

[0384] In orthogonal frequency division multiplexing (OFDM) systems, I/Q imbalance causes inter-carrier interference (ICI) from the mirror tones. This effect has been studied in the literature and methods to compensate for I/Q mismatch in single-input single-output SISO-OFDM systems have been proposed in M. D. Benedetto and P. Mandarini, “Analysis of the effect of the I/Q baseband filter mismatch in an OFDM modem,” Wireless personal communications, pp. 175-186, 2000; S. Schuchert and R. Hasholzner, “A novel I/Q imbalance compensation scheme for the reception of OFDM signals,” IEEE Transaction on Consumer Electronics, August 2001; M. Valkama, M. Renfors, and V. Koivunen, “Advanced methods for I/Q imbalance compensation in communication receivers,” IEEE Trans. Sig. Proc., October 2001; R. Rao and B. Daneshrad, “Analysis of I/Q mismatch and a cancellation scheme for OFDM systems,” IST Mobile Communication Summit, June 2004; A. Tarighat, R. Bagheri, and A. H. Sayed, “Compensation schemes and performance analysis of IQ imbalances in OFDM receivers,” Signal Processing, IEEE Transactions on [see also Acoustics, Speech, and Signal Processing, IEEE Transactions on], vol. 53, pp. 3257-3268, August 2005.

[0385] An extension of this work to multiple-input multiple-output MIMO-OFDM systems was presented in R. Rao and B. Daneshrad, “I/Q mismatch cancellation for MIMO OFDM systems,” in Personal, Indoor and Mobile Radio Communications, 2004; PIMRC 2004. 15th IEEE International Symposium on, vol. 4, 2004, pp. 2710-2714. R. M. Rao, W. Zhu, S. Lang, C. Oberli, D. Browne, J. Bhatia, J. F. Frigon, J. Wang, P; Gupta, H. Lee, D. N. Liu, S. G. Wong, M. Fitz, B. Daneshrad, and O. Takeshita, “Multi-antenna testbeds for research and education in wireless communications,” IEEE Communications Magazine, vol. 42, no. 12, pp. 72-81, December 2004; S. Lang, M. R. Rao, and B. Daneshrad, “Design and development of a 5.25 GHz software defined wireless OFDM communication platform,” IEEE Communications Magazine, vol. 42, no. 6, pp. 6-12, June 2004, for spatial multiplexing (SM) and in A. Tarighat and A. H. Sayed, “MIMO OFDM receivers for systems with IQ imbalances,” IEEE Trans. Sig. Proc., vol. 53, pp. 3583-3596, September 2005, for orthogonal space-time block codes (OSTBC).

[0386] Unfortunately, there is currently no literature on how to correct for I/Q gain and phase imbalance errors in a distributed-input distributed-output (DIDO) communication system. The embodiments of the invention described below provide a solution to these problems.

[0387] DIDO systems consist of one Base Station with distributed antennas that transmits parallel data streams (via pre-coding) to multiple users to enhance downlink throughput, while exploiting the same wireless resources (i.e., same slot duration and frequency band) as conventional SISO systems. A detailed description of DIDO systems was presented in S. G. Perlman and T. Cotter, “System and Method for Distributed Input-Distributed Output Wireless Communications,” Ser. No. 10/902,978, filed Jul. 30, 2004 (“Prior Application”), which is assigned to the assignee of the present application and which is incorporated herein by reference.

[0388] There are many ways to implement DIDO precoders. One solution is block diagonalization (BD) described in Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zero forcing methods for downlink spatial multiplexing in multiuser MIMO channels,” IEEE Trans. Sig. Proc., vol. 52, pp. 461-471, February 2004. K. K. Wong, R. D. Murch, and K. B. Letaief, “A joint channel diagonalization for multiuser MIMO antenna systems,” IEEE Trans. Wireless Comm., vol. 2, pp. 773-786, July 2003; L. U. Choi and R. D. Murch, “A transmit preprocessing technique for multiuser MIMO systems using a decomposition approach,” IEEE Trans. Wireless Comm., vol. 3, pp. 20-24, January 2004; Z. Shen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Low complexity user selection algorithms for multiuser MIMO systems with block diagonalization,” accepted for publication in IEEE Trans. Sig. Proc., September 2005; Z. Shen, R. Chen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Sum capacity of multiuser MIMO broadcast channels with block diagonalization,” submitted to IEEE Trans. Wireless Comm., October 2005; R. Chen, R. W. Heath, and J. G. Andrews, “Transmit selection diversity for unitary precoded multiuser spatial multiplexing systems with linear receivers,” accepted to IEEE Trans. on Signal Processing, 2005. The methods for I/Q compensation presented in this document assume BD precoder, but can be extended to any type of DIDO precoder.

[0389] In DIDO-OFDM systems, I/Q mismatch causes two effects: ICI and inter-user interference. The former is due to interference from the mirror tones as in SISO-OFDM systems. The latter is due to the fact that I/Q mismatch destroys the orthogonality of the DIDO precoder yielding interference across users. Both of these types of interference can be cancelled at the transmitter and receiver through the methods described herein. Three methods for I/Q compensation in DIDO-OFDM systems are described and their performance is compared against systems with and without I/Q mismatch. Results are presented based both on simulations and practical measurements carried out with the DIDO-OFDM prototype.

[0390] The present embodiments are an extension of the Prior Application. In particular, these embodiments relate to the following features of the Prior Application:

[0391] The system as described in the prior application, where the I/Q rails are affected by gain and phase imbalance;

[0392] The training signals employed for channel estimation are used to calculate the DIDO precoder with I/Q compensation at the transmitter; and

[0393] The signal characterization data accounts for distortion due to I/Q imbalance and is used at the transmitter to compute the DIDO precoder according to the method proposed in this document.

[0394] First, the mathematical model and framework of the invention will be described.

[0395] Before presenting the solution, it is useful to explain the core mathematical concept. We explain it assuming I/Q gain and phase imbalance (phase delay is not included in the description but is dealt with automatically in the DIDO-OFDM version of the algorithm). To explain the basic idea, suppose that we want to multiply two complex numbers s=s_I+js_Q and h=h_I+jh_Q and let x=h*s. We use the subscripts to denote inphase and quadrature components. Recall that

x_I=s_Ih_I−s_Qh_Q

and

x_Q=s_Ih_Q−s_Qh_I

[0396] In matrix form this can be rewritten as

$\left[\begin{array}{c}{x}_I\\ x_Q\end{array}\right]=\left[\begin{array}{cc}{h}_I& -h_Q\\ h_Q& h_I\end{array}\right]\left[\begin{array}{c}{s}_I\\ s_Q\end{array}\right].$

[0397] Note the unitary transformation by the channel matrix (H). Now suppose that s is the transmitted symbol and h is the channel. The presence of I/Q gain and phase imbalance can be modeled by creating a non-unitary transformation as follows

$\begin{array}{cc}\left[\begin{array}{c}{x}_I\\ x_Q\end{array}\right]=\left[\begin{array}{cc}{h}_{11}& h_{12}\\ h_{21}& h_{22}\end{array}\right]\left[\begin{array}{c}{s}_I\\ s_Q\end{array}\right].& \left(A\right)\end{array}$

[0398] The trick is to recognize that it is possible to write

$\begin{array}{c}\left[\begin{array}{cc}{h}_{11}& h_{12}\\ h_{21}& h_{22}\end{array}\right]=\frac{1}{2}\left[\begin{array}{cc}{h}_{11}+h_{22}& h_{12}-h_{21}\\ -\left(h_{12}-h_{21}\right)& h_{11}+h_{22}\end{array}\right]+\frac{1}{2}\left[\begin{array}{cc}{h}_{11}-h_{22}& h_{12}+h_{21}\\ h_{12}-h_{21}& h_{22}-h_{11}\end{array}\right]-\\ \frac{1}{2}\left[\begin{array}{cc}{h}_{11}+h_{22}& h_{12}-h_{21}\\ -\left(h_{12}-h_{21}\right)& h_{11}+h_{22}\end{array}\right]+\\ \frac{1}{2}\left[\begin{array}{cc}{h}_{11}-h_{22}& -\left(h_{12}+h_{21}\right)\\ h_{12}+h_{21}& h_{11}-h_{22}\end{array}\right]\left[\begin{array}{cc}1& 0\\ 0& -1\end{array}\right].\end{array}$

[0399] Now, rewriting (A)

$\begin{array}{cc}\begin{array}{c}\left[\begin{array}{c}{x}_I\\ x_Q\end{array}\right]=\frac{1}{2}\left[\begin{array}{cc}{h}_{11}+h_{22}& h_{12}-h_{21}\\ -\left(h_{12}-h_{21}\right)& h_{11}+h_{22}\end{array}\right]\left[\begin{array}{c}{s}_I\\ s_Q\end{array}\right]+\\ \frac{1}{2}\left[\begin{array}{cc}{h}_{11}-h_{22}& -\left(h_{12}+h_{21}\right)\\ h_{12}+h_{21}& h_{11}-h_{22}\end{array}\right]\left[\begin{array}{cc}1& 0\\ 0& -1\end{array}\right]\left[\begin{array}{c}{s}_I\\ s_Q\end{array}\right]\\ =\frac{1}{2}\left[\begin{array}{cc}{h}_{11}+h_{22}& h_{12}-h_{21}\\ -\left(h_{12}-h_{21}\right)& h_{11}+h_{22}\end{array}\right]\left[\begin{array}{c}{s}_I\\ s_Q\end{array}\right]+\\ \frac{1}{2}\left[\begin{array}{cc}{h}_{11}-h_{22}& -\left(h_{12}+h_{21}\right)\\ h_{12}+h_{21}& h_{11}-h_{22}\end{array}\right]\left[\begin{array}{c}{s}_I\\ -s_Q\end{array}\right]\end{array}& \left(5\right)\end{array}$

[0400] Let us define

${\mathscr{H}}_e=\frac{1}{2}\left[\begin{array}{cc}{h}_{11}+h_{22}& h_{12}-h_{21}\\ -\left(h_{12}-h_{21}\right)& h_{11}+h_{22}\end{array}\right]$ $\mathrm{and}$ ${\mathscr{H}}_c=\frac{1}{2}\left[\begin{array}{cc}{h}_{11}-h_{22}& -\left(h_{12}+h_{21}\right)\\ h_{12}+h_{21}& h_{11}-h_{22}\end{array}\right].$

[0401] Both of these matrices have a unitary structure thus can be equivalently represented by complex scalars as and

h_e=h₁₁+h₂₂+j(h₂₁−h₁₂)

and

h_c=h₁₁−h₂₂+j(h₂₁−h₁₂).

[0402] Using all of these observations, we can put the effective equation back in a scalar form with two channels: the equivalent channel h_e and the conjugate channel h_c. Then the effective transformation in (5) becomes

x=h_es+h_cs*.

[0403] We refer to the first channel as the equivalent channel and the second channel as the conjugate channel. The equivalent channel is the one you would observe if there were no I/Q gain and phase imbalance.

[0404] Using similar arguments, it can be shown that the input-output relationship of a discrete-time MIMO N×M system with I/Q gain and phase imbalance is (using the scalar equivalents to build their matrix counterparts)

$x\left[t\right]=\sum _{=0}^Lh_e\left[\right]s\left[t-\right]+h_c\left[\right]s^{*}\left[t-\right]$

where t is the discrete time index, h_e, h_cεC^M×N, s=[s₁, . . . , s_N], x=[x₁, . . . , x_M] and L is the number of channel taps.

[0405] In DIDO-OFDM systems, the received signal in the frequency domain is represented. Recall from signals and systems that if

FFT_K{s[t]}=S[k] then FFT_K{s*[t]}=S*[(−k)]=S*[K−k] for k=0, 1, . . . , K−1.

With OFDM, the equivalent input-output relationship for a MIMO-OFDM system for subcarrier k is

x[k]=H_e[k] s[k]+H_c[k] s*[K−k]  (1)

where k=0, 1, . . . , K−1 is the OFDM subcarrier index, H_e and H_c denote the equivalent and conjugate channel matrices, respectively, defined as

$H_e\left[k\right]=\sum _{=0}^Lh_e\left[\right]{}^{-j\frac{2\pi k}{K}}$ $\mathrm{and}$ $H_c\left[k\right]=\sum _{=0}^Lh_c\left[\right]{}^{-j\frac{2\pi k}{K}}.$

[0406] The second contribution in (1) is interference from the mirror tone. It can be dealt with by constructing the following stacked matrix system (note carefully the conjugates)

$\left[\begin{array}{c}\stackrel{\_}{x}\left[k\right]\\ {\stackrel{\_}{x}}^{*}\left[K-k\right]\end{array}\right]=\left[\begin{array}{cc}{H}_e\left[k\right]& H_c\left[k\right]\\ H_c^{*}\left[K-k\right]& H_e^{*}\left[K-k\right]\end{array}\right]\left[\begin{array}{c}\stackrel{\_}{s}\left[k\right]\\ {\stackrel{\_}{s}}^{*}\left[K-k\right]\end{array}\right]$

where s=[ s₁, s₂]^T and x=[ x₁, x₂]^T are the vectors of transmit and receive symbols in the frequency domain, respectively.

[0407] Using this approach, an effective matrix is built to use for DIDO operation. For example, with DIDO 2×2 the input-output relationship (assuming each user has a single receive antenna) the first user device sees (in the absence of noise)

$\begin{array}{cc}\left[\begin{array}{c}{\stackrel{\_}{x}}_1\left[k\right]\\ \stackrel{\_}{x_1^{*}}\left[K-k\right]\end{array}\right]=\left[\begin{array}{cc}{H}_e^{\left(1\right)}\left[k\right]& H_c^{\left(1\right)}\left[k\right]\\ H_c^{\left(1\right)*}\left[K-k\right]& H_e^{\left(1\right)*}\left[K-k\right]\end{array}\right]W\left[\begin{array}{c}{\stackrel{\_}{s}}_1\left[k\right]\\ {\stackrel{\_}{s}}_1^{*}\left[K-k\right]\\ {\stackrel{\_}{s}}_2\left[k\right]\\ {\stackrel{\_}{s}}_2^{*}\left[K-k\right]\end{array}\right]& \left(2\right)\end{array}$

while the second user observes

$\begin{array}{cc}\left[\begin{array}{c}{\stackrel{\_}{x}}_2\left[k\right]\\ \stackrel{\_}{x_2^{*}}\left[K-k\right]\end{array}\right]=\left[\begin{array}{cc}{H}_e^{\left(2\right)}\left[k\right]& H_c^{\left(2\right)}\left[k\right]\\ H_c^{\left(2\right)*}\left[K-k\right]& H_e^{\left(2\right)*}\left[K-k\right]\end{array}\right]W\left[\begin{array}{c}{\stackrel{\_}{s}}_1\left[k\right]\\ {\stackrel{\_}{s}}_1^{*}\left[K-k\right]\\ {\stackrel{\_}{s}}_2\left[k\right]\\ {\stackrel{\_}{s}}_2^{*}\left[K-k\right]\end{array}\right]& \left(3\right)\end{array}$

where H_e^(m), H_c^(m)εC^1×2 denote the m-th row of the matrices H_e and H_c, respectively, and WεC^4×4 is the DIDO pre-coding matrix. From (2) and (3) it is observed that the received symbol x_m[k] of user m is affected by two sources of interference caused by I/Q imbalance: inter-carrier interference from the mirror tone (i.e., s*_m[K−k]) and inter-user interference (i.e., s_p[k] and s*_p[K−k] with p≠m). The DIDO precoding matrix W in (3) is designed to cancel these two interference terms.

[0408] There are several different embodiments of the DIDO precoder that can be used here depending on joint detection applied at the receiver. In one embodiment, block diagonalization (BD) is employed (see, e.g., Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zeroforcing methods for downlink spatial multiplexing in multiuser MIMO channels,” IEEE Trans. Sig. Proc., vol. 52, pp. 461-471, February 2004. K. K. Wong, R. D. Murch, and K. B. Letaief, “A joint channel diagonalization for multiuser MIMO antenna systems,” IEEE Trans. Wireless Comm., vol. 2, pp. 773-786, July 2003. L. U. Choi and R. D. Murch, “A transmit preprocessing technique for multiuser MIMO systems using a decomposition approach,” IEEE Trans. Wireless Comm., vol. 3, pp. 20-24, January 2004. Z. Shen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Low complexity user selection algorithms for multiuser MIMO systems with block diagonalization,” accepted for publication in IEEE Trans. Sig. Proc., September 2005. Z. Shen, R. Chen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Sum capacity of multiuser MIMO broadcast channels with block diagonalization,” submitted to IEEE Trans. Wireless Comm., October 2005, computed from the composite channel [H_e^(m), H_c^(m)] (rather than H_e^(m)). So, the current DIDO system chooses the precoder such that

$\begin{array}{cc}{H}_w\stackrel{\Delta }{=}\left[\begin{array}{cc}{H}_e^{\left(1\right)}\left[k\right]& H_c^{\left(1\right)}\left[k\right]\\ H_c^{\left(1\right)*}\left[K-k\right]& H_e^{\left(1\right)*}\left[K-k\right]\\ H_e^{\left(2\right)}\left[k\right]& H_c^{\left(2\right)}\left[k\right]\\ H_c^{\left(2\right)*}\left[K-k\right]& H_e^{\left(2\right)*}\left[K-k\right]\end{array}\right]W=\left[\begin{array}{cccc}{\alpha }_{1,1}& 0& 0& 0\\ 0& {\alpha }_{1,2}& 0& 0\\ 0& 0& {\alpha }_{2,1}& 0\\ 0& 0& 0& {\alpha }_{2,2}\end{array}\right]\stackrel{\Delta }{=}\left[\begin{array}{cc}{H}_w^{\left(1,1\right)}& H_w^{\left(1,2\right)}\\ H_w^{\left(2,1\right)}& H_w^{\left(2,2\right)}\end{array}\right]& \left(4\right)\end{array}$

where α_i,j are constants and H_w^(i,j)εC^2×2. This method is beneficial because using this precoder, it is possible to keep other aspects of the DIDO precoder the same as before, since the effects of I/Q gain and phase imbalance are completely cancelled at the transmitter.

[0409] It is also possible to design DIDO precoders that pre-cancel inter-user interference, without pre-cancelling ICI due to 10 imbalance. With this approach, the receiver (instead of the transmitter) compensates for the IQ imbalance by employing one of the receive filters described below. Then, the pre-coding design criterion in (4) can be modified as

$\begin{array}{cc}{H}_w\stackrel{\Delta }{=}\left[\begin{array}{cc}{H}_e^{\left(1\right)}\left[k\right]& H_c^{\left(1\right)}\left[k\right]\\ H_c^{\left(1\right)*}\left[K-k\right]& H_e^{\left(1\right)*}\left[K-k\right]\\ H_e^{\left(2\right)}\left[k\right]& H_c^{\left(2\right)}\left[k\right]\\ H_c^{\left(2\right)*}\left[K-k\right]& H_e^{\left(2\right)*}\left[K-k\right]\end{array}\right]W=\left[\begin{array}{cccc}{\alpha }_{1,1}& {\alpha }_{1,2}& 0& 0\\ {\alpha }_{2,1}& {\alpha }_{2,2}& 0& 0\\ 0& 0& {\alpha }_{3,3}& {\alpha }_{3,4}\\ 0& 0& {\alpha }_{4,3}& {\alpha }_{4,4}\end{array}\right]\stackrel{\Delta }{=}\left[\begin{array}{cc}{H}^{\left(1,1\right)}& H_w^{\left(1,2\right)}\\ H_w^{\left(2,1\right)}& H_w^{\left(2,2\right)}\end{array}\right]& \left(5\right)\\ {\stackrel{\_}{x}}_1\left[k\right]=\left[H_w^{\left(1,1\right)}H_w^{\left(1,2\right)}\right]\left[\begin{array}{c}{\stackrel{\_}{s}}_1\left[k\right]\\ {\stackrel{\_}{s}}_2\left[k\right]\end{array}\right]\mathrm{and}& \left(6\right)\\ {\stackrel{\_}{x}}_2\left[k\right]=\left[H_w^{\left(2,1\right)}H_w^{\left(2,2\right)}\right]\left[\begin{array}{c}{\stackrel{\_}{s}}_1\left[k\right]\\ {\stackrel{\_}{s}}_2\left[k\right]\end{array}\right]& \left(7\right)\end{array}$

[0410] where s_m[k]=[ s_m[k], s*_m[K−k]]^T for the m-th transmit symbol and x_m[k]=[ x_m[k], x*_m[K−k]]^T is the receive symbol vector for user m.

[0411] At the receive side, to estimate the transmit symbol vector s_m[k], user m employs ZF filter and the estimated symbol vector is given by

ŝ_m^(ZF)[k]=[(H_w^(m,m)†H_w^(m,m))⁻¹H_w^(m,m)†] x_m[k]  (8)

[0412] While the ZF filter is the easiest to understand, the receiver may apply any number of other filters known to those skilled in the art. One popular choice is the MMSE filter where

ŝ_m^(MMSE)[k]=(H_w^(m,m)†+p⁻¹I)⁻¹H_w^(m,m)H_w^(m,m)† x_m[k]  (9)

and ρ is the signal-to-noise ratio. Alternatively, the receiver may perform a maximum likelihood symbol detection (or sphere decoder or iterative variation). For example, the first user might use the ML receiver and solve the following optimization

$\begin{array}{cc}{\hat{s}}_m^{\left(\mathrm{ML}\right)}\left[k\right]=\arg \underset{s_1s_2\in S}{\min }{\stackrel{\_}{y}}_1\left[k\right]-\left[H_w^{\left(1,1\right)}H_w^{\left(1,2\right)}\right]\left[\begin{array}{c}{s}_1\left[k\right]\\ s_2\left[k\right]\end{array}\right]& \left(10\right)\end{array}$

where S is the set of all possible vectors s and depends on the constellation size. The ML receiver gives better performance at the expense of requiring more complexity at the receiver. A similar set of equations applies for the second user.

[0413] Note that H_w^(1,2) and H_w^(2,1) in (6) and (7) are assumed to have zero entries. This assumption holds only if the transmit precoder is able to cancel completely the inter-user interference as for the criterion in (4). Similarly, H_w^(1,1) and H_w^(2,2) are diagonal matrices only if the transmit precoder is able to cancel completely the inter-carrier interference (i.e., from the mirror tones).

[0414] FIG. 13 illustrates one embodiment of a framework for DIDO-OFDM systems with I/Q compensation including IQ-DIDO precoder 1302 within a Base Station (BS), a transmission channel 1304, channel estimation logic 1306 within a user device, and a ZF, MMSE or ML receiver 1308. The channel estimation logic 1306 estimates the channels H_e^(m) and H_c^(m) via training symbols and feedbacks these estimates to the precoder 1302 within the AP. The BS computes the DIDO precoder weights (matrix W) to pre-cancel the interference due to I/Q gain and phase imbalance as well as inter-user interference and transmits the data to the users through the wireless channel 1304. User device m employs the ZF, MMSE or ML receiver 1308, by exploiting the channel estimates provided by the unit 1304, to cancel residual interference and demodulates the data.

[0415] The following three embodiments may be employed to implement this I/Q compensation algorithm:

[0416] Method 1—TX compensation: In this embodiment, the transmitter calculates the pre-coding matrix according to the criterion in (4). At the receiver, the user devices employ a “simplified” ZF receiver, where H_w^(1,1) and H_w^(2,2) are assumed to be diagonal matrices. Hence, equation (8) simplifies as

$\begin{array}{cc}{\hat{s}}_m\left[k\right]=\left[\begin{array}{cc}1/{\alpha }_{m,1}& 0\\ 0& 1/{\alpha }_{m,2}\end{array}\right]{\stackrel{\_}{x}}_m\left[k\right].& \left(10\right)\end{array}$

[0417] Method 2—RX compensation: In this embodiment, the transmitter calculates the pre-coding matrix based on the conventional BD method described in R. Chen, R. W. Heath, and J. G. Andrews, “Transmit selection diversity for unitary precoded multiuser spatial multiplexing systems with linear receivers,” accepted to IEEE Trans. on Signal Processing, 2005, without canceling inter-carrier and inter-user interference as for the criterion in (4). With this method, the pre-coding matrix in (2) and (3) simplifies as

$\begin{array}{cc}W=\left[\begin{array}{cccc}{w}_{1,1}\left[k\right]& 0& w_{1,2}\left[k\right]& 0\\ 0& w_{1,1}^{*}\left[K-k\right]& 0& w_{1,2}^{*}\left[K-k\right]\\ w_{2,1}\left[k\right]& 0& w_{2,2}\left[k\right]& 0\\ 0& w_{2,1}^{*}\left[K-k\right]& 0& w_{2,2}^{*}\left[K-k\right]\end{array}\right].& \left(12\right)\end{array}$

[0418] At the receiver, the user devices employ a ZF filter as in (8). Note that this method does not pre-cancel the interference at the transmitter as in the method 1 above. Hence, it cancels the inter-carrier interference at the receiver, but it is not able to cancel the inter-user interference. Moreover, in method 2 the users only need to feedback the vector H_e^(m) for the transmitter to compute the DIDO precoder, as opposed to method 1 that requires feedback of both H_e^(m) and H_c^(m). Therefore, method 2 is particularly suitable for DIDO systems with low rate feedback channels. On the other hand, method 2 requires slightly higher computational complexity at the user device to compute the ZF receiver in (8) rather than (11).

[0419] Method 3—TX-RX compensation: In one embodiment, the two methods described above are combined. The transmitter calculates the pre-coding matrix as in (4) and the receivers estimate the transmit symbols according to (8).

[0420] I/Q imbalance, whether phase imbalance, gain imbalance, or delay imbalance, creates a deleterious degradation in signal quality in wireless communication systems. For this reason, circuit hardware in the past was designed to have very low imbalance. As described above, however, it is possible to correct this problem using digital signal processing in the form of transmit pre-coding and/or a special receiver. One embodiment of the invention comprises a system with several new functional units, each of which is important for the implementation of I/Q correction in an OFDM communication system or a DIDO-OFDM communication system.

[0421] One embodiment of the invention uses pre-coding based on channel state information to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch) in an OFDM system. As illustrated in FIG. 11, a DIDO transmitter according to this embodiment includes a user selector unit 1102, a plurality of coding modulation units 1104, a corresponding plurality of mapping units 1106, a DIDO IQ-aware precoding unit 1108, a plurality of RF transmitter units 1114, a user feedback unit 1112 and a DIDO configurator unit 1110.

[0422] The user selector unit 1102 selects data associated with a plurality of users U₁-U_M, based on the feedback information obtained by the feedback unit 1112, and provides this information each of the plurality of coding modulation units 1104. Each coding modulation unit 1104 encodes and modulates the information bits of each user and send them to the mapping unit 1106. The mapping unit 1106 maps the input bits to complex symbols and sends the results to the DIDO IQ-aware precoding unit 1108. The DIDO IQ-aware precoding unit 1108 exploits the channel state information obtained by the feedback unit 1112 from the users to compute the DIDO IQ-aware precoding weights and precoding the input symbols obtained from the mapping units 1106. Each of the precoded data streams is sent by the DIDO IQ-aware precoding unit 1108 to the OFDM unit 1115 that computes the IFFT and adds the cyclic prefix. This information is sent to the D/A unit 1116 that operates the digital to analog conversion and send it to the RF unit 1114. The RF unit 1114 upconverts the baseband signal to intermediate/radio frequency and send it to the transmit antenna.

[0423] The precoder operates on the regular and mirror tones together for the purpose of compensating for I/Q imbalance. Any number of precoder design criteria may be used including ZF, MMSE, or weighted MMSE design. In a preferred embodiment, the precoder completely removes the ICI due to I/Q mismatch thus resulting in the receiver not having to perform any additional compensation.

[0424] In one embodiment, the precoder uses a block diagonalization criterion to completely cancel inter-user interference while not completely canceling the I/Q effects for each user, requiring additional receiver processing. In another embodiment, the precoder uses a zero-forcing criterion to completely cancel both inter-user interference and ICI due to I/Q imbalance. This embodiment can use a conventional DIDO-OFDM processor at the receiver.

[0425] One embodiment of the invention uses pre-coding based on channel state information to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch) in a DIDO-OFDM system and each user employs an IQ-aware DIDO receiver. As illustrated in FIG. 12, in one embodiment of the invention, a system including the receiver 1202 includes a plurality of RF units 1208, a corresponding plurality of A/D units 1210, an IQ-aware channel estimator unit 1204 and a DIDO feedback generator unit 1206.

[0426] The RF units 1208 receive signals transmitted from the DIDO transmitter units 1,14 downconverts the signals to baseband and provide the downconverted signals to the A/D units 1210. The A/D units 1210 then convert the signal from analog to digital and send it to the OFDM units 1213. The OFDM units 1213 remove the cyclic prefix and operates the FFT to report the signal to the frequency domain. During the training period the OFDM units 1213 send the output to the IQ-aware channel estimate unit 1204 that computes the channel estimates in the frequency domain. Alternatively, the channel estimates can be computed in the time domain. During the data period the OFDM units 1213 send the output to the IQ-aware receiver unit 1202. The IQ-aware receiver unit 1202 computes the IQ receiver and demodulates/decodes the signal to obtain the data 1214. The IQ-aware channel estimate unit 1204 sends the channel estimates to the DIDO feedback generator unit 1206 that may quantize the channel estimates and send it back to the transmitter via the feedback control channel 1112.

[0427] The receiver 1202 illustrated in FIG. 12 may operate under any number of criteria known to those skilled in the art including ZF, MMSE, maximum likelihood, or MAP receiver. In one preferred embodiment, the receiver uses an MMSE filter to cancel the ICI caused by IQ imbalance on the mirror tones. In another preferred embodiment, the receiver uses a nonlinear detector like a maximum likelihood search to jointly detect the symbols on the mirror tones. This method has improved performance at the expense of higher complexity.

[0428] In one embodiment, an IQ-aware channel estimator 1204 is used to determine the receiver coefficients to remove ICI. Consequently we claim a DIDO-OFDM system that uses pre-coding based on channel state information to cancel inter-carrier interference (ICI) from mirror tones (due to I/Q mismatch), an IQ-aware DIDO receiver, and an IQ-aware channel estimator. The channel estimator may use a conventional training signal or may use specially constructed training signals sent on the inphase and quadrature signals. Any number of estimation algorithms may be implemented including least squares, MMSE, or maximum likelihood. The IQ-aware channel estimator provides an input for the IQ-aware receiver.

[0429] Channel state information can be provided to the station through channel reciprocity or through a feedback channel. One embodiment of the invention comprises a DIDO-OFDM system, with I/Q-aware precoder, with an I/Q-aware feedback channel for conveying channel state information from the user terminals to the station. The feedback channel may be a physical or logical control channel. It may be dedicated or shared, as in a random access channel. The feedback information may be generated using a DIDO feedback generator at the user terminal, which we also claim. The DIDO feedback generator takes as an input the output of the I/Q aware channel estimator. It may quantize the channel coefficients or may use any number of limited feedback algorithms known in the art.

[0430] The allocation of users, modulation and coding rate, mapping to space-time-frequency code slots may change depending on the results of the DIDO feedback generator. Thus, one embodiment comprises an IQ-aware DIDO configurator that uses an IQ-aware channel estimate from one or more users to configure the DIDO IQ-aware precoder, choose the modulation rate, coding rate, subset of users allowed to transmit, and their mappings to space-time-frequency code slots.

[0431] To evaluate the performance of the proposed compensation methods, three DIDO 2×2 systems will be compared:

[0432] 1. With I/Q mismatch: transmit over all the tones (except DC and edge tones), without compensation for I/Q mismatch;

[0433] 2. With I/Q compensation: transmit over all the tones and compensate for I/Q mismatch by using the “method 1” described above;

[0434] 3. Ideal: transmit data only over the odd tones to avoid inter-user and inter-carrier (i.e., from the mirror tones) interference caused to I/Q mismatch.

[0435] Hereafter, results obtained from measurements with the DIDO-OFDM prototype in real propagation scenarios are presented. FIG. 14 depicts the 64-QAM constellations obtained from the three systems described above. These constellations are obtained with the same users' locations and fixed average signal-to-noise ratio (˜45 dB). The first constellation 1401 is very noisy due to interference from the mirror tones caused by I/Q imbalance. The second constellation 1402 shows some improvements due to I/Q compensations. Note that the second constellation 1402 is not as clean as the ideal case shown as constellation 1403 due to possible phase noise that yields inter-carrier interference (ICI).

[0436] FIG. 15 shows the average SER (Symbol Error Rate) 1501 and per-user goodput 1502 performance of DIDO 2×2 systems with 64-QAM and coding rate ¾, with and without I/Q mismatch. The OFDM bandwidth is 250 KHz, with 64 tones and cyclic prefix length L_cp=4. Since in the ideal case we transmit data only over a subset of tones, SER and goodput performance is evaluated as a function of the average per-tone transmit power (rather than total transmit power) to guarantee a fair comparison across different cases. Moreover, in the following results, we use normalized values of transmit power (expressed in decibel), since our goal here is to compare the relative (rather than absolute) performance of different schemes. FIG. 15 shows that in presence of I/Q imbalance the SER saturates, without reaching the target SER (˜10⁻²), consistently to the results reported in A. Tarighat and A. H. Sayed, “MIMO OFDM receivers for systems with IQ imbalances,” IEEE Trans. Sig. Proc., vol. 53, pp. 3583-3596, September 2005. This saturation effect is due to the fact that both signal and interference (from the mirror tones) power increase as the TX power increases. Through the proposed I/Q compensation method, however, it is possible to cancel the interference and obtain better SER performance. Note that the slight increase in SER at high SNR is due to amplitude saturation effects in the DAC, due to the larger transmit power required for 64-QAM modulations.

[0437] Moreover, observe that the SER performance with I/Q compensation is very close to the ideal case. The 2 dB gap in TX power between these two cases is due to possible phase noise that yields additional interference between adjacent OFDM tones. Finally, the goodput curves 1502 show that it is possible to transmit twice as much data when the I/Q method is applied compared to the ideal case, since we use all the data tones rather than only the odd tones (as for the ideal case).

[0438] FIG. 16 graphs the SER performance of different QAM constellations with and without I/Q compensation. We observe that, in this embodiment, the proposed method is particularly beneficial for 64-QAM constellations. For 4-QAM and 16-QAM the method for I/Q compensation yields worse performance than the case with I/Q mismatch, possibly because the proposed method requires larger power to enable both data transmission and interference cancellation from the mirror tones. Moreover, 4-QAM and 16-QAM are not as affected by I/Q mismatch as 64-QAM due to the larger minimum distance between constellation points. See A. Tarighat, R. Bagheri, and A. H. Sayed, “Compensation schemes and performance analysis of IQ imbalances in OFDM receivers,” IEEE Transactions on Signal Processing, vol. 53, pp. 3257-3268, August 2005. This can be also observed in FIG. 16 by comparing the I/Q mismatch against the ideal case for 4-QAM and 16-QAM. Hence, the additional power required by the DIDO precoder with interference cancellation (from the mirror tones) does not justify the small benefit of the I/Q compensation for the cases of 4-QAM and 16-QAM. Note that this issue may be fixed by employing the methods 2 and 3 for I/Q compensation described above.

[0439] Finally, the relative SER performance of the three methods described above is measured in different propagation conditions. For reference, also described is the SER performance in presence of I/Q mismatch. FIG. 17 depicts the SER measured for a DIDO 2×2 system with 64-QAM at carrier frequency of 450.5 MHz and bandwidth of 250 KHz, at two different users' locations. In Location 1 the users are ˜6λ from the BS in different rooms and NLOS (Non-Line of Sight)) conditions. In Location 2 the users are ˜λ from the BS in LOS (Line of Sight).

[0440] FIG. 17 shows that all three compensation methods always outperform the case of no compensation. Moreover, it should be noted that method 3 outperforms the other two compensation methods in any channel scenario. The relative performance of method 1 and 2 depends on the propagation conditions. It is observed through practical measurement campaigns that method 1 generally outperforms method 2, since it pre-cancels (at the transmitter) the inter-user interference caused by I/Q imbalance. When this inter-user interference is minimal, method 2 may outperform method 1 as illustrated in graph 1702 of FIG. 17, since it does not suffer from power loss due to the I/Q compensation precoder.

[0441] So far, different methods have been compared by considering only a limited set of propagation scenarios as in FIG. 17. Hereafter, the relative performance of these methods in ideal i.i.d. (independent and identically-distributed) channels is measured. DIDO-OFDM systems are simulated with I/Q phase and gain imbalance at the transmit and receive sides. FIG. 18 shows the performance of the proposed methods with only gain imbalance at the transmit side (i.e., with 0.8 gain on the I rail of the first transmit chain and gain 1 on the other rails). It is observed that method 3 outperforms all the other methods. Also, method 1 performs better than method 2 in i.i.d. channels, as opposed to the results obtained in Location 2 in graph 1702 of FIG. 17.

[0442] Thus, given the three novel methods to compensate for I/Q imbalance in DIDO-OFDM systems described above, Method 3 outperforms the other proposed compensation methods. In systems with low rate feedback channels, method 2 can be used to reduce the amount of feedback required for the DIDO precoder, at the expense of worse SER performance.

[0443] Another embodiment of a system and method to enhance the performance of distributed-input distributed-output (DIDO) systems will now be described. This method dynamically allocates the wireless resources to different user devices, by tracking the changing channel conditions, to increase throughput while satisfying certain target error rate. The user devices estimate the channel quality and feedback it to the Base Station (BS); the Base Station processes the channel quality obtained from the user devices to select the best set of user devices, DIDO scheme, modulation/coding scheme (MCS) and array configuration for the next transmission; the Base Station transmits parallel data to multiple user devices via pre-coding and the signals are demodulated at the receiver.

[0444] A system that efficiently allocates resources for a DIDO wireless link is also described. The system includes a DIDO Base Station with a DIDO configurator, which processes feedback received from the users to select the best set of users, DIDO scheme, modulation/coding scheme (MCS) and array configuration for the next transmission; a receiver in a DIDO system that measures the channel and other relevant parameters to generate a DIDO feedback signal; and a DIDO feedback control channel for conveying feedback information from users to the Base Station.

[0445] As described in detail below, some of the significant features of this embodiment of the invention include, but are not limited to:

[0446] Techniques to adaptively select number of users, DIDO transmission schemes (i.e., antenna selection or multiplexing), modulation/coding scheme (MCS) and array configurations based on the channel quality information, to minimize SER or maximize per-user or downlink spectral efficiency;

[0447] Techniques to define sets of DIDO transmission modes as combinations of DIDO schemes and MCSs;

[0448] Techniques to assign different DIDO modes to different time slots, OFDM tones and DIDO substreams, depending on the channel conditions;

[0449] Techniques to dynamically assign different DIDO modes to different users based on their channel quality;

[0450] Criteria to enable adaptive DIDO switching based on link quality metrics computed in the time, frequency and space domains;

[0451] Criteria to enable adaptive DIDO switching based on lookup tables.

[0452] A DIDO system with a DIDO configurator at the Base Station as in FIG. 19 to adaptively select the number of users, DIDO transmission schemes (i.e., antenna selection or multiplexing), modulation/coding scheme (MCS) and array configurations based on the channel quality information, to minimize SER or maximize per user or downlink spectral efficiency;

[0453] A DIDO system with a DIDO configurator at the Base Station and a DIDO feedback generator at each user device as in FIG. 20, which uses the estimated channel state and/or other parameters like the estimated SNR at the receiver to generate a feedback message to be input into the DIDO configurator.

[0454] A DIDO system with a DIDO configurator at the Base Station, DIDO feedback generator, and a DIDO feedback control channel for conveying DIDO-specific configuration information from the users to the Base Station.

[0455] In multiple-input multiple-output (MIMO) systems, diversity schemes such as orthogonal space-time block codes (OSTBC) (See V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Spacetime block codes from orthogonal designs,” IEEE Trans. Info. Th., vol. 45, pp. 1456-467, July 1999) or antenna selection (See R. W. Heath Jr., S. Sandhu, and A. J. Paulraj, “Antenna selection for spatial multiplexing systems with linear receivers,” IEEE Trans. Comm., vol. 5, pp. 142-144, April 2001) are conceived to combat channel fading, providing increased link robustness that translates in better coverage. On the other hand, spatial multiplexing (SM) enables transmission of multiple parallel data streams as a means to enhance systems throughput. See G. J. Foschini, G. D. Golden, R. A. Valenzuela, and P. W. Wolniansky, “Simplified processing for high spectral efficiency wireless communication employing multielement arrays,” IEEE Jour. Select. Areas in Comm., vol. 17, no. 11, pp. 1841-1852, November 1999. These benefits can be simultaneously achieved in MIMO systems, according to the theoretical diversity/multiplexing tradeoffs derived in L. Zheng and D. N. C. Tse, “Diversity and multiplexing: a fundamental tradeoff in multiple antenna channels,” IEEE Trans. Info. Th., vol. 49, no. 5, pp. 1073-1096, May 2003. One practical implementation is to adaptively switch between diversity and multiplexing transmission schemes, by tracking the changing channel conditions.

[0456] A number of adaptive MIMO transmission techniques have been proposed thus far. The diversity/multiplexing switching method in R. W. Heath and A. J. Paulraj, “Switching between diversity and multiplexing in MIMO systems,” IEEE Trans. Comm., vol. 53, no. 6, pp. 962-968, June 2005, was designed to improve BER (Bit Error Rate) for fixed rate transmission, based on instantaneous channel quality information. Alternatively, statistical channel information can be employed to enable adaptation as in S. Catreux, V. Erceg, D. Gesbert, and R. W. Heath. Jr., “Adaptive modulation and MIMO coding for broadband wireless data networks,” IEEE Comm. Mag., vol. 2, pp. 108-115, June 2002 (“Catreux”), resulting in reduced feedback overhead and number of control messages. The adaptive transmission algorithm in Catreux was designed to enhance spectral efficiency for predefined target error rate in orthogonal frequency division multiplexing (OFDM) systems, based on channel time/frequency selectivity indicators. Similar low feedback adaptive approaches have been proposed for narrowband systems, exploiting the channel spatial selectivity to switch between diversity schemes and spatial multiplexing. See, e.g., A. Forenza, M. R. McKay, A. Pandharipande, R. W. Heath. Jr., and I. B. Collings, “Adaptive MIMO transmission for exploiting the capacity of spatially correlated channels,” accepted to the IEEE Trans. on Veh. Tech., March 2007; M. R. McKay, I. B. Collings, A. Forenza, and R. W. Heath. Jr., “Multiplexing/beamforming switching for coded MIMO in spatially correlated Rayleigh channels,” accepted to the IEEE Trans. on Veh. Tech., December 2007; A. Forenza, M. R. McKay, R. W. Heath. Jr., and I. B. Collings, “Switching between OSTBC and spatial multiplexing with linear receivers in spatially correlated MIMO channels,” Proc. IEEE Veh. Technol. Conf., vol. 3, pp. 1387-1391, May 2006; M. R. McKay, I. B. Collings, A. Forenza, and R. W. Heath Jr., “A throughput-based adaptive MIMO BICM approach for spatially correlated channels,” to appear in Proc. IEEE ICC, June 2006

[0457] In this document, we extend the scope of the work presented in various prior publications to DIDO-OFDM systems. See, e.g., R. W. Heath and A. J. Paulraj, “Switching between diversity and multiplexing in MIMO systems,” IEEE Trans. Comm., vol. 53, no. 6, pp. 962-968, June 2005. S. Catreux, V. Erceg, D. Gesbert, and R. W. Heath Jr., “Adaptive modulation and MIMO coding for broadband wireless data networks,” IEEE Comm. Mag., vol. 2, pp. 108-115, June 2002; A. Forenza, M. R. McKay, A. Pandharipande, R. W. Heath Jr., and I. B. Collings, “Adaptive MIMO transmission for exploiting the capacity of spatially correlated channels,” IEEE Trans. on Veh. Tech., vol. 56, n. 2, pp. 619-630, March 2007. M. R. McKay, I. B. Collings, A. Forenza, and R. W. Heath Jr., “Multiplexing/beamforming switching for coded MIMO in spatially correlated Rayleigh channels,” accepted to the IEEE Trans. on Veh. Tech., December 2007; A. Forenza, M. R. McKay, R. W. Heath Jr., and I. B. Collings, “Switching between OSTBC and spatial multiplexing with linear receivers in spatially correlated MIMO channels,” Proc. IEEE Veh. Technol. Conf., vol. 3, pp. 1387-1391, May 2006. M. R. McKay, I. B. Collings, A. Forenza, and R. W. Heath Jr., “A throughput-based adaptive MIMO BICM approach for spatially correlated channels,” to appear in Proc. IEEE ICC, June 2006.

[0458] A novel adaptive DIDO transmission strategy is described herein that switches between different numbers of users, numbers of transmit antennas and transmission schemes based on channel quality information as a means to improve the system performance. Note that schemes that adaptively select the users in multiuser MIMO systems were already proposed in M. Sharif and B. Hassibi, “On the capacity of MIMO broadcast channel with partial side information,” IEEE Trans. Info. Th., vol. 51, p. 506522, February 2005; and W. Choi, A. Forenza, J. G. Andrews, and R. W. Heath Jr., “Opportunistic space division multiple access with beam selection,” to appear in IEEE Trans. on Communications. The opportunistic space division multiple access (OSDMA) schemes in these publications, however, are designed to maximize the sum capacity by exploiting multi-user diversity and they achieve only a fraction of the theoretical capacity of dirty paper codes, since the interference is not completely pre-canceled at the transmitter. In the DIDO transmission algorithm described herein block diagonalization is employed to pre-cancel inter-user interference. The proposed adaptive transmission strategy, however, can be applied to any DIDO system, independently on the type of pre-coding technique.

[0459] The present patent application describes an extension of the embodiments of the invention described above and in the Prior Application, including, but not limited to the following additional features:

[0460] 1. The training symbols of the Prior Application for channel estimation can be employed by the wireless client devices to evaluate the link-quality metrics in the adaptive DIDO scheme;

[0461] 2. The base station receives signal characterization data from the client devices as described in the Prior Application. In the current embodiment, the signal characterization data is defined as link-quality metric used to enable adaptation;

[0462] 3. The Prior Application describes a mechanism to select the number of transmit antennas and users as well as defines throughput allocation. Moreover, different levels of throughput can be dynamically assigned to different clients as in the Prior Application. The current embodiment of the invention defines novel criteria related to this selection and throughput allocation.

[0463] The goal of the proposed adaptive DIDO technique is to enhance per-user or downlink spectral efficiency by dynamically allocating the wireless resource in time, frequency and space to different users in the system. The general adaptation criterion is to increase throughput while satisfying the target error rate. Depending on the propagation conditions, this adaptive algorithm can also be used to improve the link quality of the users (or coverage) via diversity schemes. The flowchart illustrated in FIG. 21 describes steps of the adaptive DIDO scheme.

[0464] The Base Station (BS) collects the channel state information (CSI) from all the users in 2102. From the received CSI, the BS computes the link quality metrics in time/frequency/space domains in 2104. These link quality metrics are used to select the users to be served in the next transmission as well as the transmission mode for each of the users in 2106. Note that the transmission modes consist of different combinations of modulation/coding and DIDO schemes. Finally, the BS transmits data to the users via DIDO precoding as in 2108.

[0465] At 2102, the Base Station collects the channel state information (CSI) from all the user devices. The CSI is used by the Base Station to determine the instantaneous or statistical channel quality for all the user devices at 2104. In DIDO-OFDM systems the channel quality (or link quality metric) can be estimated in the time, frequency and space domains. Then, at 2106, the Base Station uses the link quality metric to determine the best subset of users and transmission mode for the current propagation conditions. A set of DIDO transmission modes is defined as combinations of DIDO schemes (i.e., antenna selection or multiplexing), modulation/coding schemes (MCSs) and array configuration. At 2108, data is transmitted to user devices using the selected number of users and transmission modes.

[0466] In one embodiment, the mode selection is enabled by lookup tables (LUTs) pre-computed based on error rate performance of DIDO systems in different propagation environments. These LUTs map channel quality information into error rate performance. To construct the LUTs, the error rate performance of DIDO systems is evaluated in different propagation scenarios as a function of the SNR. From the error rate curves, it is possible to compute the minimum SNR required to achieve certain pre-defined target error rate. We define this SNR requirement as SNR threshold. Then, the SNR thresholds are evaluated in different propagation scenarios and for different DIDO transmission modes and stored in the LUTs. For example, the SER results in FIGS. 24 and 26 can be used to construct the LUTs. Then, from the LUTs, the Base Station selects the transmission modes for the active users that increase throughput while satisfying predefined target error rate. Finally, the Base Station transmits data to the selected users via DIDO pre-coding. Note that different DIDO modes can be assigned to different time slots, OFDM tones and DIDO substreams such that the adaptation may occur in time, frequency and space domains.

[0467] One embodiment of a system employing DIDO adaptation is illustrated in FIGS. 19-20. Several new functional units are introduced to enable implementation of the proposed DIDO adaptation algorithms. Specifically, in one embodiment, a DIDO configurator 1910 performs a plurality of functions including selecting the number of users, DIDO transmission schemes (i.e., antenna selection or multiplexing), modulation/coding scheme (MCS), and array configurations based on the channel quality information 1912 provided by user devices.

[0468] The user selector unit 1902 selects data associated with a plurality of users U₁-U_M, based on the feedback information obtained by the DIDO configurator 1910, and provides this information each of the plurality of coding modulation units 1904. Each coding modulation unit 1904 encodes and modulates the information bits of each user and sends them to the mapping unit 1906. The mapping unit 1906 maps the input bits to complex symbols and sends it to the precoding unit 1908. Both the coding modulation units 1904 and the mapping unit 1906 exploit the information obtained from the DIDO configurator unit 1910 to choose the type of modulation/coding scheme to employ for each user. This information is computed by the DIDO configurator unit 1910 by exploiting the channel quality information of each of the users as provided by the feedback unit 1912. The DIDO precoding unit 1908 exploits the information obtained by the DIDO configurator unit 1910 to compute the DIDO precoding weights and precoding the input symbols obtained from the mapping units 1906. Each of the precoded data streams are sent by the DIDO precoding unit 1908 to the OFDM unit 1915 that computes the IFFT and adds the cyclic prefix. This information is sent to the D/A unit 1916 that operates the digital to analog conversion and sends the resulting analog signal to the RF unit 1914. The RF unit 1914 upconverts the baseband signal to intermediate/radio frequency and send it to the transmit antenna.

[0469] The RF units 2008 of each client device receive signals transmitted from the DIDO transmitter units 1914, downconverts the signals to baseband and provide the downconverted signals to the ND units 2010. The ND units 2010 then convert the signal from analog to digital and send it to the OFDM units 2013. The OFDM units 2013 remove the cyclic prefix and carries out the FFT to report the signal to the frequency domain. During the training period the OFDM units 2013 send the output to the channel estimate unit 2004 that computes the channel estimates in the frequency domain. Alternatively, the channel estimates can be computed in the time domain. During the data period the OFDM units 2013 send the output to the receiver unit 2002 which demodulates/decodes the signal to obtain the data 2014. The channel estimate unit 2004 sends the channel estimates to the DIDO feedback generator unit 2006 that may quantize the channel estimates and send it back to the transmitter via the feedback control channel 1912.

[0470] The DIDO configurator 1910 may use information derived at the Base Station or, in a preferred embodiment, uses additionally the output of a DIDO Feedback Generator 2006 (see FIG. 20), operating at each user device. The DIDO Feedback Generator 2006 uses the estimated channel state 2004 and/or other parameters like the estimated SNR at the receiver to generate a feedback message to be input into the DIDO Configurator 1910. The DIDO Feedback Generator 2006 may compress information at the receiver, may quantize information, and/or use some limited feedback strategies known in the art.

[0471] The DIDO Configurator 1910 may use information recovered from a DIDO Feedback Control Channel 1912. The DIDO Feedback Control Channel 1912 is a logical or physical control channel that is used to send the output of the DIDO Feedback Generator 2006 from the user to the Base Station. The control channel 1912 may be implemented in any number of ways known in the art and may be a logical or a physical control channel. As a physical channel it may comprise a dedicated time/frequency slot assigned to a user. It may also be a random access channel shared by all users. The control channel may be pre-assigned or it may be created by stealing bits in a predefined way from an existing control channel.

[0472] In the following discussion, results obtained through measurements with the DIDO-OFDM prototype are described in real propagation environments. These results demonstrate the potential gains achievable in adaptive DIDO systems. The performance of different order DIDO systems is presented initially, demonstrating that it is possible to increase the number of antennas/user to achieve larger downlink throughput. The DIDO performance as a function of user device's location is then described, demonstrating the need for tracking the changing channel conditions. Finally, the performance of DIDO systems employing diversity techniques is described.

[0473] i. Performance of Different Order DIDO Systems

[0474] The performance of different DIDO systems is evaluated with increasing number of transmit antennas N=M, where M is the number of users. The performance of the following systems is compared: SISO, DIDO 2×2, DIDO 4×4, DIDO 6×6 and DIDO 8×8. DIDO N×M refers to DIDO with N transmit antennas at the BS and M users.

[0475] FIG. 22 illustrates the transmit/receive antenna layout in a exemplary residential floor plan. The transmit antennas 2201 are placed in squared array configuration and the users are located around the transmit array. In FIG. 22, T indicates the “transmit” antennas and U refers to the “user devices” 2202.

[0476] Different antenna subsets are active in the 8-element transmit array, depending on the value of N chosen for different measurements. For each DIDO order (N) the subset of antennas that covers the largest real estate for the fixed size constraint of the 8-element array was chosen. This criterion is expected to enhance the spatial diversity for any given value of N.

[0477] FIG. 23 shows the array configurations for different DIDO orders that fit the available real estate (i.e., dashed line and outer walls). The squared dashed box has dimensions of 24″×24″, corresponding to ˜λ×λ at the carrier frequency of 450 MHz.

[0478] Based on the comments related to FIG. 23 and with reference to FIG. 22, the performance of each of the following systems will now be defined and compared:

[0479] SISO with T1 and U1 (2301)

[0480] DIDO 2×2 with T1,2 and U1,2 (2302)

[0481] DIDO 4×4 with T1,2,3,4 and U1,2,3,4 (2303)

[0482] DIDO 6×6 with T1,2,3,4,5,6 and U1,2,3,4,5,6 (2304)

[0483] DIDO 8×8 with T1,2,3,4,5,6,7,8 and U1,2,3,4,5,6,7,8 (2305)

[0484] FIG. 24 shows the SER, BER, SE (Spectral Efficiency) and goodput performance as a function of the transmit (TX) power for the DIDO systems described above, with 4-QAM and FEC (Forward Error Correction) rate of ½. Observe that the SER and BER performance degrades for increasing values of N. This effect is due to two phenomena: for fixed TX power, the input power to the DIDO array is split between increasing number of users (or data streams); the spatial diversity decreases with increasing number of users in realistic (spatially correlated) DIDO channels.

[0485] To compare the relative performance of different order DIDO systems the target BER is fixed to 10⁻⁴ (this value may vary depending on the system) that corresponds approximately to SER=10⁻² as shown in FIG. 24. We refer to the TX power values corresponding to this target as TX power thresholds (TPT). For any N, if the TX power is below the TPT, we assume it is not possible to transmit with DIDO order N and we need to switch to lower order DIDO. Also, in FIG. 24, observe that the SE and goodput performance saturate when the TX power exceeds the TPTs for any value of N. From these results, an adaptive transmission strategy may be designed that switches between different order DIDO to enhance SE or goodput for fixed predefined target error rate.

[0486] ii. Performance with Variable User Location

[0487] The goal of this experiment is to evaluate the DIDO performance for different users' location, via simulations in spatially correlated channels. DIDO 2×2 systems are considered with 4QAM and an FEC rate of ½. User 1 is at a broadside direction from the transmit array, whereas user 2 changes locations from broadside to endfire directions as illustrated in FIG. 25. The transmit antennas are spaced ˜λ/2 and separated ˜2.5λ from the users.

[0488] FIG. 26 shows the SER and per-user SE results for different locations of user device 2. The user device's angles of arrival (AOAs) range between 0° and 90°, measured from the broadside direction of the transmit array. Observe that, as the user device's angular separation increases, the DIDO performance improves, due to larger diversity available in the DIDO channel. Also, at target SER=10⁻² there is a 10 dB gap between the cases AOA2=0° and AOA2=90°. This result is consistent to the simulation results obtained in FIG. 35 for an angle spread of 10°. Also, note that for the case of AOA1=AOA2=0° there may be coupling effects between the two users (due to the proximity of their antennas) that may vary their performance from the simulated results in FIG. 35.

[0489] iii. Preferred Scenario for DIDO 8×8

[0490] FIG. 24 illustrated that DIDO 8×8 yields a larger SE than lower order DIDO at the expense of higher TX power requirement. The goal of this analysis is to show there are cases where DIDO 8×8 outperforms DIDO 2×2, not only in terms of peak spectral efficiency (SE), but also in terms of TX power requirement (or TPT) to achieve that peak SE.

[0491] Note that, in i.i.d. (ideal) channels, there is ˜6 dB gap in TX power between the SE of DIDO 8×8 and DIDO 2×2. This gap is due to the fact that DIDO 8×8 splits the TX power across eight data streams, whereas DIDO 2×2 only between two streams. This result is shown via simulation in FIG. 32.

[0492] In spatially correlated channels, however, the TPT is a function of the characteristics of the propagation environment (e.g., array orientation, user location, angle spread). For example, FIG. 35 shows ˜15 dB gap for low angle spread between two different user device's locations. Similar results are presented in FIG. 26 of the present application.

[0493] Similarly to MIMO systems, the performance of DIDO systems degrades when the users are located at endfire directions from the TX array (due to lack of diversity). This effect has been observed through measurements with the current DIDO prototype. Hence, one way to show that DIDO 8×8 outperforms DIDO 2×2 is to place the users at endfire directions with respect to the DIDO 2×2 arrays. In this scenario, DIDO 8×8 outperforms DIDO 2×2 due to the higher diversity provided by the 8-antenna array.

[0494] In this analysis, consider the following systems:

[0495] System 1: DIDO 8×8 with 4-QAM (transmit 8 parallel data streams every time slot);

[0496] System 2: DIDO 2×2 with 64-QAM (transmit to users X and Y every 4 time slots). For this system we consider four combinations of TX and RX antenna locations: a) T1,T2 U1,2 (endfire direction); b) T3,T4 U3,4 (endfire direction); c) T5,T6 U5,6 (˜30° from the endfire direction); d) T7,T8 U7,8 (NLOS (Non-Line of Sight));

[0497] System 3: DIDO 8×8 with 64-QAM; and

[0498] System 4: MISO 8×1 with 64-QAM (transmit to user X every 8 time slots).

[0499] For all these cases, an FEC rate of ¾ was used.

[0500] The users' locations are depicted in FIG. 27.

[0501] In FIG. 28 the SER results show a ˜15 dB gap between Systems 2a and 2c due to different array orientations and user locations (similar to the simulation results in FIG. 35). The first subplot in the second row shows the values of TX power for which the SE curves saturate (i.e. corresponding to BER 1e-4). We observe that System 1 yields larger per-user SE for lower TX power requirement (˜5 dB less) than System 2. Also, the benefits of DIDO 8×8 versus DIDO 2×2 are more evident for the DL (downlink) SE and DL goodput due to multiplexing gain of DIDO 8×8 over DIDO 2×2. System 4 has lower TX power requirement (8 dB less) than System 1, due to the array gain of beamforming (i.e., MRC with MISO 8×1). But System 4 yields only ⅓ of per-user SE compared to System 1. System 2 performs worse than System 1 (i.e., yields lower SE for larger TX power requirement). Finally, System 3 yields much larger SE (due to larger order modulations) than System 1 for larger TX power requirement (˜15 dB).

[0502] From these results, the following conclusions may be drawn:

[0503] One channel scenario was identified for which DIDO 8×8 outperforms DIDO 2×2 (i.e., yields larger SE for lower TX power requirement);

[0504] In this channel scenario, DIDO 8×8 yields larger per user SE and DL SE than DIDO 2×2 and MISO 8×1; and

[0505] It is possible to further increase the performance of DIDO 8×8 by using higher order modulations (i.e., 64-QAM rather than 4-QAM) at the expense of larger TX power requirements (˜15 dB more).

[0506] iv. DIDO with Antenna Selection

[0507] Hereafter, we evaluate the benefit of the antenna selection algorithm described in R. Chen, R. W. Heath, and J. G. Andrews, “Transmit selection diversity for unitary precoded multiuser spatial multiplexing systems with linear receivers,” accepted to IEEE Trans. on Signal Processing, 2005. We present the results for one particular DIDO system with two users, 4-QAM and FEC rate of ½. The following systems are compared in FIG. 27:

[0508] DIDO 2×2 with T1,2 and U1,2; and

[0509] DIDO 3×2 using antenna selection with T1,2,3 and U1,2.

[0510] The transmit antenna's and user device locations are the same as in FIG. 27.

[0511] FIG. 29 shows that DIDO 3×2 with antenna selection may provide ˜5 dB gain compared to DIDO 2×2 systems (with no selection). Note that the channel is almost static (i.e., no Doppler), so the selection algorithms adapts to the path-loss and channel spatial correlation rather than the fast-fading. We should be seeing different gains in scenarios with high Doppler. Also, in this particular experiment it was observed that the antenna selection algorithm selects antennas 2 and 3 for transmission.

[0512] iv. SNR Thresholds for the LUTs

[0513] In section [0171] we stated that the mode selection is enabled by LUTs. The LUTs can be pre-computed by evaluating the SNR thresholds to achieve certain predefined target error-rate performance for the DIDO transmission modes in different propagation environments. Hereafter, we provide the performance of DIDO systems with and without antenna selection and variable number of users that can be used as guidelines to construct the LUTs. While FIGS. 24, 26, 28, 29 were derived from practical measurements with the DIDO prototype, the following Figures are obtained through simulations. The following BER results assume no FEC.

[0514] FIG. 30 shows the average BER performance of different DIDO precoding schemes in i.i.d. channels. The curve labeled as ‘no selection’ refers to the case when BD is employed. In the same figure the performance of antenna selection (ASel) is shown for different number of extra antennas (with respect to the number of users). It is possible to see that as the number of extra antennas increases, ASel provides better diversity gain (characterized by the slope of the BER curve in high SNR regime), resulting in better coverage. For example, if we fix the target BER to 10⁻² (practical value for uncoded systems), the SNR gain provided by ASel increases with the number of antennas.

[0515] FIG. 31 shows the SNR gain of ASel as a function of the number of extra transmit antennas in i.i.d. channels, for different targets BER. It is possible to see that, just by adding 1 or 2 antennas, ASel yields significant SNR gains compared to BD. In the following sections, we will evaluate the performance of ASel only for the cases of 1 or 2 extra antennas and by fixing the target BER to 10⁻² (for uncoded systems).

[0516] FIG. 32 depicts the SNR thresholds as a function of the number of users (M) for BD and ASel with 1 and 2 extra antennas in i.i.d. channels. We observe that the SNR thresholds increase with M due to the larger receive SNR requirement for larger number of users. Note that we assume fixed total transmit power (with variable number of transmit antennas) for any number of users. Moreover, FIG. 32 shows that the gain due to antenna selection is constant for any number of users in i.i.d. channels.

[0517] Hereafter, we show the performance of DIDO systems in spatially correlated channels. We simulate each user's channel through the COST-259 spatial channel model described in X. Zhuang, F. W. Vook, K. L. Baum, T. A. Thomas, and M. Cudak, “Channel models for link and system level simulations,” IEEE 802.16 Broadband Wireless Access Working Group, September 2004. We generate single-cluster for each user. As a case study, we assume NLOS channels, uniform linear array (ULA) at the transmitter, with element spacing of 0.5 lambda. For the case of 2-user system, we simulate the clusters with mean angles of arrival AOA1 and AOA2 for the first and second user, respectively. The AOAs are measured with respect to the broadside direction of the ULA. When more than two users are in the system, we generate the users' clusters with uniformly spaced mean AOAs in the range [−φ_m,φ_m], where we define

$\begin{array}{cc}{\Phi }_M=\frac{\mathrm{\Delta \phi }\left(M-1\right)}{2}& \left(13\right)\end{array}$

with K being the number of users and Δφ is the angular separation between the users' mean AOAs. Note that the angular range [−φ_m,φ_m] is centered at the 0° angle, corresponding to the broadside direction of the ULA. Hereafter, we study the BER performance of DIDO systems as a function of the channel angle spread (AS) and angular separation between users, with BD and ASel transmission schemes and different numbers of users.

[0518] FIG. 33 depicts the BER versus per-user average SNR for two users located at the same angular direction (i.e., AOA1=AOA2=0°, with respect to the broadside direction of the ULA), with different values of AS. It is possible to see that as the AS increases the BER performance improves and approaches the i.i.d. case. In fact, higher AS yields statistically less overlapping between the eigenmodes of the two users and better performance of the BD precoder.

[0519] FIG. 34 shows similar results as FIG. 33, but with higher angular separation between the users. We consider AOA1=0° and AOA2=90° (i.e., 90° angular separation). The best performance is now achieved in the low AS case. In fact, for the case of high angle separation, there is less overlapping between the users' eigenmodes when the angular spread is low. Interestingly, we observe that the BER performance in low AS is better than i.i.d. channels for the same reasons just mentioned.

[0520] Next, we compute the SNR thresholds, for target BER of 10⁻² in different correlation scenarios. FIG. 35 plots the SNR thresholds as a function of the AS for different values of the mean AOAs of the users. For low users' angular separation reliable transmissions with reasonable SNR requirement (i.e., 18 dB) are possible only for channels characterized by high AS. On the other hand, when the users are spatially separated, less SNR is required to meet the same target BER.

[0521] FIG. 36 shows the SNR threshold for the case of five users. The users' mean AOAs are generated according to the definition in (13), with different values of angular separation Δφ. We observe that for Δφ=0° and AS<15°, BD performs poorly due to the small angle separation between users, and the target BER is not satisfied. For increasing AS the SNR requirement to meet the fixed target BER decreases. On the other end, for Δφ=30°, the smallest SNR requirement is obtained at low AS, consistently to the results in FIG. 35. As the AS increases, the SNR thresholds saturate to the one of i.i.d. channels. Note that Δφ=30″ with 5 users corresponds to the AOA range of [−60°, 60°], that is typical for base stations in cellular systems with 120° sectorized cells.

[0522] Next, we study the performance of ASel transmission scheme in spatially correlated channels. FIG. 37 compares the SNR threshold of BD and ASel, with 1 and 2 extra antennas, for two user case. We consider two different cases of angular separation between users: {AOA1=0°,AOA2=0°} and {AOA1=0°,AOA2=90°}. The curves for BD scheme (i.e., no antenna selection) are the same as in FIG. 35. We observe that ASel yields 8 dB and 10 dB SNR gains with 1 and 2 extra antennas, respectively, for high AS. As the AS decreases, the gain due to ASel over BD becomes smaller due to the reduced number of degrees of freedom in the MIMO broadcast channel. Interestingly, for AS=0° (i.e., close to LOS channels) and the case {AOA1=0°,AOA2=90°}, ASel does not provide any gain due to the luck of diversity in the space domain. FIG. 38 shows similar results as FIG. 37, but for five user case.

[0523] We compute the SNR thresholds (assuming usual target BER of 10⁻²) as a function of the number of users in the system (M), for both BD and ASel transmission schemes. The SNR thresholds correspond to the average SNR, such that the total transmit power is constant for any M. We assume maximum separation between the mean AOAs of each user's cluster within the azimuth range [−φ_m,φ_m]=[−60°, 60°]. Then, the angular separation between users is Δφ=120°/(M−1).

[0524] FIG. 39 shows the SNR thresholds for BD scheme with different values of AS. We observe that the lowest SNR requirement is obtained for AS=0.1° (i.e., low angle spread) with relatively small number of users (i.e., K≦20), due to the large angular separation between users. For M>50, however, the SNR requirement is way above 40 dB, since Δφ is very small, and BD is impractical. Moreover, for AS>10° the SNR thresholds remain almost constant for any M, and the DIDO system in spatially correlated channels approaches the performance of i.i.d. channels.

[0525] To reduce the values of the SNR thresholds and improve the performance of the DIDO system we apply ASel transmission scheme. FIG. 40 depicts the SNR thresholds in spatially correlated channels with AS=0.1° for BD and ASel with 1 and 2 extra antennas. For reference we report also the curves for the i.i.d. case shown in FIG. 32. It is possible to see that, for low number of users (i.e., M≦10), antenna selection does not help reducing the SNR requirement due to the lack of diversity in the DIDO broadcast channel. As the number of users increases, ASel benefits from multiuser diversity yielding SNR gains (i.e., 4 dB for M=20). Moreover, for M≦20, the performance of ASel with 1 or 2 extra antennas in highly spatially correlated channels is the same.

[0526] We then compute the SNR thresholds for two more channel scenarios: AS=5° in FIG. 41 and AS=10° in FIG. 42. FIG. 41 shows that ASel yields SNR gains also for relatively small number of users (i.e., M≦10) as opposed to FIG. 40, due to the larger angle spread. For AS=10° the SNR thresholds reduce further and the gains due to ASel get higher, as reported in FIG. 42.

[0527] Finally, we summarize the results presented so far for correlated channels. FIG. 43 and FIG. 44 show the SNR thresholds as a function of the number of users (M) and angle spread (AS) for BD and ASel schemes, with 1 and 2 extra antennas, respectively. Note that the case of AS=30° corresponds actually to i.i.d. channels, and we used this value of AS in the plot only, for graphical representation. We observe that, while BD is affected by the channel spatial correlation, ASel yields almost the same performance for any AS. Moreover, for AS=0.1°, ASel performs similarly to BD for low M, whereas outperforms BD for large M (i.e., M≧20), due to multiuser diversity.

[0528] FIG. 49 compares the performance of different DIDO schemes in terms of SNR thresholds. The DIDO schemes considered are: BD, ASel, BD with eigenmode selection (BD-ESel) and maximum ratio combining (MRC). Note that MRC, does not pre-cancel interference at the transmitter (unlike the other methods), but does provide larger gain in case the users are spatially separated. In FIG. 49 we plot the SNR threshold for target BER=10−2 for DIDO N×2 systems when the two users are located at −30° and 30° from the broadside direction of the transmit array, respectively. We observe that for low AS the MRC scheme provides 3 dB gain compared to the other schemes since the users' spatial channels are well separated and the effect of inter-user interference is low. Note that the gain of MRC over DIDO N×2 are due to array gain. For AS larger than 20° the QR-ASel scheme outperforms the other and yields about 10 dB gain compared to BD 2×2 with no selection. QR-ASel and BD-ESel provide about the same performance for any value of AS.

[0529] Described above is a novel adaptive transmission technique for DIDO systems. This method dynamically switches between DIDO transmission modes to different users to enhance throughput for fixed target error rate. The performance of different order DIDO systems was measured in different propagation conditions and it was observed that significant gains in throughput may be achieved by dynamically selecting the DIDO modes and number of users as a function of the propagation conditions.

[0530] As previously described, wireless communication systems use carrier waves to convey information. These carrier waves are usually sinusoids that are amplitude and/or phase modulated in response to information to be transmitted. The nominal frequency of the sinusoid is known as the carrier frequency. To create this waveform, the transmitter synthesizes one or more sinusoids and uses upconversion to create a modulated signal riding on a sinusoid with the prescribed carrier frequency. This may be done through direct conversion where the signal is directly modulated on the carrier or through multiple upconversion stages. To process this waveform, the receiver must demodulate the received RF signal and effectively remove the modulating carrier. This requires that the receiver synthesize one or more sinusoidal signals to reverse the process of modulation at the transmitter, known as downconversion. Unfortunately, the sinusoidal signals generated at the transmitter and receiver are derived from different reference oscillators. No reference oscillator creates a perfect frequency reference; in practice there is always some deviation from the true frequency.

[0531] In wireless communication systems, the differences in the outputs of the reference oscillators at the transmitter and receivers create the phenomena known as carrier frequency offset, or simply frequency offset, at the receiver. Essentially there is some residual modulation in the received signal (corresponding to the difference in the transmit and receive carriers), which occurs after downconversion. This creates distortion in the received signal resulting in higher bit error rates and lower throughput.

[0532] There are different techniques for dealing with carrier frequency offset. Most approaches estimate the carrier frequency offset at the receiver and then apply a carrier frequency offset correction algorithm. The carrier frequency offset estimation algorithm may be blind using offset QAM (T. Fusco and M. Tanda, “Blind Frequency-offset Estimation for OFDM/OQAM Systems,” IEEE Transactions on Signal Processing, vol. 55, pp. 1828-1838,2007); periodic properties (E. Serpedin, A. Chevreuil, G. B. Giannakis, and P. Loubaton, “Blind channel and carrier frequency offset estimation using periodic modulation precoders,” IEEE Transactions on Signal Processing, vol. 48, no. 8, pp. 2389-2405, August 2000); or the cyclic prefix in orthogonal frequency division multiplexing (OFDM) structure approaches (J. J. van de Beek, M. Sandell, and P. O. Borjesson, “ML estimation of time and frequency offset in OFDM systems,” IEEE Transactions on Signal Processing, vol. 45, no. 7, pp. 1800-1805, July 1997; U. Tureli, H. Liu, and M. D. Zoltowski, “OFDM blind carrier offset estimation: ESPRIT,” IEEE Trans. Commun., vol. 48, no. 9, pp. 1459-1461, September 2000; M. Luise, M. Marselli, and R. Reggiannini, “Low-complexity blind carrier frequency recovery for OFDM signals over frequency-selective radio channels,” IEEE Trans. Commun., vol. 50, no. 7, pp. 1182-1188, July 2002).

[0533] Alternatively special training signals may be utilized including a repeated data symbol (P. H. Moose, “A technique for orthogonal frequency division multiplexing frequency offset correction,” IEEE Trans. Commun., vol. 42, no. 10, pp. 2908-2914, October 1994); two different symbols (T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997); or periodically inserted known symbol sequences (M. Luise and R. Reggiannini, “Carrier frequency acquisition and tracking for OFDM systems,” IEEE Trans. Commun., vol. 44, no. 11, pp. 1590-1598, November 1996). The correction may occur in analog or in digital. The receiver can also use carrier frequency offset estimation to precorrect the transmitted signal to eliminate offset. Carrier frequency offset correction has been studied extensively for multicarrier and OFDM systems due to their sensitivity to frequency offset (J. J. van de Beek, M. Sandell, and P. O. Borjesson, “ML estimation of time and frequency offset in OFDM systems,” Signal Processing, IEEE Transactions on [see also Acoustics, Speech, and Signal Processing, IEEE Transactions on], vol. 45, no. 7, pp. 1800-1805, July 1997; U. Tureli, H. Liu, and M. D. Zoltowski, “OFDM blind carrier offset estimation: ESPRIT,” IEEE Trans. Commun., vol. 48, no. 9, pp. 1459-1461, September 2000; T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997; M. Luise, M. Marselli, and R. Reggiannini, “Low-complexity blind carrier frequency recovery for OFDM signals over frequency-selective radio channels,” IEEE Trans. Commun., vol. 50, no. 7, pp. 1182-1188, July 2002).

[0534] Frequency offset estimation and correction is an important issue for multi-antenna communication systems, or more generally MIMO (multiple input multiple output) systems. In MIMO systems where the transmit antennas are locked to one frequency reference and the receivers are locked to another frequency reference, there is a single offset between the transmitter and receiver. Several algorithms have been proposed to tackle this problem using training signals (K. Lee and J. Chun, “Frequency-offset estimation for MIMO and OFDM systems using orthogonal training sequences,” IEEE Trans. Veh. Technol., vol. 56, no. 1, pp. 146-156, January 2007; M. Ghogho and A. Swami, “Training design for multipath channel and frequency offset estimation in MIMO systems,” IEEE Transactions on Signal Processing, vol. 54, no. 10, pp. 3957-3965, October 2006, and adaptive tracking C. Oberli and B. Daneshrad, “Maximum likelihood tracking algorithms for MIMOOFDM,” in Communications, 2004 IEEE International Conference on, vol. 4, Jun. 20-24, 2004, pp. 2468-2472). A more severe problem is encountered in MIMO systems where the transmit antennas are not locked to the same frequency reference but the receive antennas are locked together. This happens practically in the uplink of a spatial division multiple access (SDMA) system, which can be viewed as a MIMO system where the different users correspond to different transmit antennas. In this case the compensation of frequency offset is much more complicated. Specifically, the frequency offset creates interference between the different transmitted MIMO streams. It can be corrected using complex joint estimation and equalization algorithms (A. Kannan, T. P. Krauss, and M. D. Zoltowski, “Separation of cochannel signals under imperfect timing and carrier synchronization,” IEEE Trans. Veh. Technol., vol. 50, no. 1, pp. 79-96, January 2001), and equalization followed by frequency offset estimation (T. Tang and R. W. Heath, “Joint frequency offset estimation and interference cancellation for MIMO-OFDM systems [mobile radio],” 2004. VTC2004-Fall. 2004 IEEE 60^th Vehicular Technology Conference, vol. 3, pp. 1553-1557, Sep. 26-29, 2004; X. Dai, “Carrier frequency offset estimation for OFDM/SDMA systems using consecutive pilots,” IEEE Proceedings-Communications, vol. 152, pp. 624-632, Oct. 7, 2005). Some work has dealt with the related problem of residual phase off-set and tracking error, where residual phase offsets are estimated and compensated after frequency offset estimation, but this work only consider the uplink of an SDMA OFDMA system (L. Haring, S. Bieder, and A. Czylwik, “Residual carrier and sampling frequency synchronization in multiuser OFDM systems,” 2006. VTC 2006-Spring. IEEE 63rd Vehicular Technology Conference, vol. 4, pp. 1937-1941, 2006). The most severe case in MIMO systems occurs when all transmit and receive antennas have different frequency references. The only available work on this topic only deals with asymptotic analysis of estimation error in flat fading channels (O. Besson and P. Stoica, “On parameter estimation of MIMO flat-fading channels with frequency offsets,” Signal Processing, IEEE Transactions on [see also Acoustics, Speech, and Signal Processing, IEEE Transactions on], vol. 51, no. 3, pp. 602-613, March 2003).

[0535] A case that has not been significantly investigated occurs when the different transmit antennas of a MIMO system do not have the same frequency reference and the receive antennas process the signals independently. This happens in what is known as a distributed input distributed-output (DIDO) communication system, also called the MIMO broadcast channel in the literature. DIDO systems consist of one access point with distributed antennas that transmit parallel data streams (via precoding) to multiple users to enhance downlink throughput, while exploiting the same wireless resources (i.e., same slot duration and frequency band) as conventional SISO systems. Detailed description of DIDO systems was presented in, S. G. Perlman and T. Cotter, “System and method for distributed input-distributed output wireless communications,” United States Patent Application 20060023803, July 2004. There are many ways to implement DIDO precoders. One solution is block diagonalization (BD) described in, for example, Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zero-forcing methods for downlink spatial multiplexing in multiuser MIMO channels,” IEEE Trans. Sig. Proc., vol. 52, pp. 461-471, February 2004; K. K. Wong, R. D. Murch, and K. B. Letaief, “A joint-channel diagonalization for multiuser MIMO antenna systems,” IEEE Trans. Wireless Comm., vol. 2, pp. 773-786, July 2003; L. U. Choi and R. D. Murch, “A transmit preprocessing technique for multiuser MIMO systems using a decomposition approach,” IEEE Trans. Wireless Comm., vol. 3, pp. 20-24, January 2004; Z. Shen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Low complexity user selection algorithms for multiuser MIMO systems with block diagonalization,” accepted for publication in IEEE Trans. Sig. Proc., September 2005; Z. Shen, R. Chen, J. G. Andrews, R. W. Heath, and B. L. Evans, “Sum capacity of multiuser MIMO broadcast channels with block diagonalization,” submitted to IEEE Trans. Wireless Comm., October 2005; R. Chen, R. W. Heath, and J. G. Andrews, “Transmit selection diversity for unitary precoded multiuser spatial multiplexing systems with linear receivers,” accepted to IEEE Trans. on Signal Processing, 2005.

[0536] In DIDO systems, transmit precoding is used to separate data streams intended for different users. Carrier frequency offset causes several problems related to the system implementation when the transmit antenna radio frequency chains do not share the same frequency reference. When this happens, each antenna is effectively transmits at a slightly different carrier frequency. This destroys the integrity of the DIDO precoder resulting in each user experiencing extra interference. Proposed below are several solutions to this problem. In one embodiment of the solution, the DIDO transmit antennas share a frequency reference through a wired, optical, or wireless network. In another embodiment of the solution, one or more users estimate the frequency offset differences (the relative differences in the offsets between pairs of antennas) and send this information back to the transmitter. The transmitter then precorrects for the frequency offset and proceeds with the training and precoder estimation phase for DIDO. There is a problem with this embodiment when there are delays in the feedback channel. The reason is that there may be residual phase errors created by the correction process that are not accounted for in the subsequent channel estimation. To solve this problem, one additional embodiment uses a novel frequency offset and phase estimator that can correct this problem by estimating the delay. Results are presented based both on simulations and practical measurements carried out with a DIDO-OFDM prototype.

[0537] The frequency and phase offset compensation method proposed in this document may be sensitive to estimation errors due to noise at the receiver. Hence, one additional embodiment proposes methods for time and frequency offset estimation that are robust also under low SNR conditions.

[0538] There are different approaches for performing time and frequency offset estimation. Because of its sensitivity to synchronization errors, many of these approaches were proposed specifically for the OFDM waveform.

[0539] The algorithms typically do not exploit the structure of the OFDM waveform thus they are generic enough for both single carrier and multicarrier waveforms. The algorithm described below is among a class of techniques that employ known reference symbols, e.g. training data, to aid in synchronization. Most of these methods are extensions of Moose's frequency offset estimator (see P. H. Moose, “A technique for orthogonal frequency division multiplexing frequency offset correction,” IEEE Trans. Commun., vol. 42, no. 10, pp. 2908-2914, October 1994.). Moose proposed to use two repeated training signals and derived the frequency offset using the phase difference between both received signals. Moose's method can only correct for the fractional frequency offset. An extension of the Moose method was proposed by Schmidl and Cox (T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997.). Their key innovation was to use one periodic OFDM symbol along with an additional differentially encoded training symbol. The differential encoding in the second symbol enables integer offset correction. Coulson considered a similar setup as described in T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997, and provided a detailed discussion of algorithms and analysis as described in A. J. Coulson, “Maximum likelihood synchronization for OFDM using a pilot symbol: analysis,” IEEE J. Select. Areas Commun., vol. 19, no. 12, pp. 2495-2503, December 2001; A. J. Coulson, “Maximum likelihood synchronization for OFDM using a pilot symbol: algorithms,” IEEE J. Select. Areas Commun., vol. 19, no. 12, pp. 2486-2494, December 2001. One main difference is that Coulson uses repeated maximum length sequences to provide good correlation properties. He also suggests using chirp signals because of their constant envelope properties in the time and frequency domains. Coulson considers several practical details but does not include integer estimation. Multiple repeated training signals were considered by Minn et. al. in H. Minn, V. K. Bhargava, and K. B. Letaief, “A robust timing and frequency synchronization for OFDM systems,” IEEE Trans. Wireless Commun., vol. 2, no. 4, pp. 822-839, July 2003, but the structure of the training was not optimized. Shi and Serpedin show that the training structure has some optimality form the perspective of frame synchronization (K. Shi and E. Serpedin, “Coarse frame and carrier synchronization of OFDM systems: a new metric and comparison,” IEEE Trans. Wireless Commun., vol. 3, no. 4, pp. 1271-1284, July 2004). One embodiment of the invention uses the Shi and Serpedin approach to perform frame synchronization and fractional frequency offset estimation.

[0540] Many approaches in the literature focus on frame synchronization and fractional frequency offset correction. Integer offset correction is solved using an additional training symbol as in T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997. For example, Morrelli et. al. derived an improved version of T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613-1621, December 1997, in M. Morelli, A. N. D'Andrea, and U. Mengali, “Frequency ambiguity resolution in OFDM systems,” IEEE Commun. Lett., vol. 4, no. 4, pp. 134-136, April 2000. An alternative approach using a different preamble structure was suggested by Morelli and Mengali (M. Morelli and U. Mengali, “An improved frequency offset estimator for OFDM applications,” IEEE Commun. Lett., vol. 3, no. 3, pp. 75-77, March 1999). This approach uses the correlations between M repeated identical training symbols to increase the range of the fractional frequency offset estimator by a factor of M. This is the best linear unbiased estimator and accepts a large offset (with proper design) but does not provide good timing synchronization.

[0541] One embodiment of the invention uses pre-coding based on channel state information to cancel frequency and phase offsets in DIDO systems. See FIG. 11 and the associated description above for a description of this embodiment.

[0542] In one embodiment of the invention, each user employs a receiver equipped with frequency offset estimator/compensator. As illustrated in FIG. 45, in one embodiment of the invention, a system including the receiver includes a plurality of RF units 4508, a corresponding plurality of ND units 4510, a receiver equipped with a frequency offset estimator/compensator 4512 and a DIDO feedback generator unit 4506.

[0543] The RF units 4508 receive signals transmitted from the DIDO transmitter units, downconvert the signals to baseband and provide the downconverted signals to the ND units 4510. The ND units 4510 then convert the signal from analog to digital and send it to the frequency offset estimator/compensator units 4512. The frequency offset estimator/compensator units 4512 estimate the frequency offset and compensate for it, as described herein, and then send the compensated signal to the OFDM units 4513. The OFDM units 4513 remove the cyclic prefix and operate the Fast Fourier Transform (FFT) to report the signal to the frequency domain. During the training period the OFDM units 4513 send the output to the channel estimate unit 4504 that computes the channel estimates in the frequency domain. Alternatively, the channel estimates can be computed in the time domain. During the data period the OFDM units 4513 send the output to the DIDO receiver unit 4502 which demodulates/decodes the signal to obtain the data. The channel estimate unit 4504 sends the channel estimates to the DIDO feedback generator unit 4506 that may quantize the channel estimates and send them back to the transmitter via the feedback control channel, as illustrated.

[0544] Described below are embodiments of an algorithm for frequency/phase offset compensation in DIDO systems. The DIDO system model is initially described with and without frequency/phase offsets. For the sake of the simplicity, the particular implementation of a DIDO 2×2 system is provided. However, the underlying principles of the invention may also be implemented on higher order DIDO systems.

[0545] The received signals of DIDO 2×2 can be written for the first user as

r₁[t]=h₁₁(w₁₁x₁[t]+w₂₁x₂[t])+h₁₂(w₁₂x₁[t]+w₂₂x₂[t])  (1)

and for the second user as

r₂[t]=h₂₁(w₁₁x₁[t]+w₂₁x₂[t])+h₂₂(w₁₂x₁[t]+w₂₂x₂[t])  (2)

where t is the discrete time index, h_mn hand w_mn are the channel and the DIDO precoding weights between the m-th user and n-th transmit antenna, respectively, and x_m is the transmit signal to user m. Note that h_mn and w_mn are not a function of t since we assume the channel is constant over the period between training and data transmission.

[0546] In the presence of frequency and phase offset, the received signals are expressed as

r₁[t]=e^{j(ωU1−ωT1)Ts(t−t11)}h₁₁(w₁₁x₁[t]+w₂₁x₂[t])+e^{j(ωU1−ωT2)Ts(t−t12)}h₁₂(w₁₂x₁[t]+w₂₂x₂[t])  (3)

and

r₂[t]=e^{j(ωU1−ωT1)Ts(t−t21)}h₂₁(w₁₁x₁[t]+w₂₁x₂[t])+e^{j(ωU1−ωT2)Ts(t−t22)}h₂₂(w₁₂x₁[t]+w₂₂x₂[t])  (4)

where T_s is the symbol period, ω_Tn=2Πf_Tn for the n-th transmit antenna, ω_Um=2Πf_Um for the m-th user, and f_Tn and f_Um are the actual carrier frequencies (affected by offset) for the n-th transmit antenna and m-th user, respectively. The values t_mn denote random delays that cause phase offset over the channel h_mn. FIG. 46 depicts the DIDO 2×2 system model.

[0547] For the time being, we use the following definitions:

Δω_mn=ω_Um−ω_Tn  (5)

to denote the frequency offset between the m-th user and the n-th transmit antenna.

[0548] A method according to one embodiment of the invention is illustrated in FIG. 47. The method includes the following general steps (which include sub-steps, as illustrated): training period for frequency offset estimation 4701; training period for channel estimation 4702; data transmission via DIDO precoding with compensation 4703. These steps are described in detail below.

[0549] (a) Training Period for Frequency Offset Estimation (4701)

[0550] During the first training period the base station sends one or more training sequences from each transmit antennas to one of the users (4701a). As described herein “users” are wireless client devices. For the DIDO 2×2 case, the signal received by the m-th user is given by

r_m[t]=e^{jΔωm1Ts(t−tm1)}H_m1p₁[t]+e^{jΔωm2Ts(t−tms)}h_m2p₂[t]  (6)

where p₁ and p₂ are the training sequences transmitted from the first and second antennas, respectively.

[0551] The m-th user may employ any type of frequency offset estimator (i.e., convolution by the training sequences) and estimates the offsets Δω_m1 and Δω_m2. Then, from these values the user computes the frequency offset between the two transmit antennas as

Δω_T=Δω_m2−Δω_m1=ω_T1−ω_T2  (7)

Finally, the value in (7) is fed back to the base station (4701b).

[0552] Note that p₁ and p₂ in (6) are designed to be orthogonal, so that the user can estimate Δω_m1 and Δω_m2. Alternatively, in one embodiment, the same training sequence is used over two consecutive time slots and the user estimates the offset from there. Moreover, to improve the estimate of the offset in (7) the same computations described above can be done for all users of the DIDO systems (not just for the m-th user) and the final estimate may be the (weighted) average of the values obtained from all users. This solution, however, requires more computational time and amount of feedback. Finally, updates of the frequency offset estimation are needed only if the frequency offset varies over time. Hence, depending on the stability of the clocks at the transmitter, this step 4701 of the algorithm can be carried out on a long-term basis (i.e., not for every data transmission), resulting in reduction of feedback overhead.

• • [0553] (a) Training Period for Channel Estimation (4702)
  • [0554] (b) During the second training period, the base station first obtains the frequency offset feedback with the value in (7) from the m-th user or from the plurality of users. The value in (7) is used to pre-compensate for the frequency offset at the transmit side. Then, the base station sends training data to all the users for channel estimation (4702a).

[0555] For DIDO 2×2 systems, the signal received at the first user is given by

r₁[t]=e^{jΔω11Ts(t−{tilde over (t)}11)}h₁₁p₁[t]+e^{jΔω12Ts(t−{tilde over (t)}12)}h₁₂e^−jΔωTTstp₂[t]  (8)

and at the second user by

r₂[t]=e^{jΔω21Ts(t−{tilde over (t)}21)}h₂₁p₁[t]+e^{jΔω22Ts(t−{tilde over (t)}22)}h₂₂e^−jΔωTTstp₂[t]  (9)

where {tilde over (t)}_mn=t_mn+Δt and Δt is random or known delay between the first and second transmissions of the base station. Moreover, p₁ and p₂ are the training sequences transmitted from the first and second antennas, respectively, for frequency offset and channel estimation.

[0556] Note that the pre-compensation is applied only to the second antennas in this embodiment.

[0557] Expanding (8) we obtain

r₁[t]=e^jΔω11Tste^jθ11[h₁₁p₁[t]+e^{j(θ12−θ11)}h₁₂p₂[t]]  (10)

and similarly for the second user

r₂[t]=e^jΔω21Tste^jθ21[h₂₁p₁[t]+e^{j(θ22−θ21)}h₂₂p₂[t]]  (11)

where θ_mn=−Δω_mnT_s{tilde over (t)}_mn.

[0558] At the receive side, the users compensate for the residual frequency offset by using the training sequences p₁ and p₂. Then the users estimate via training the vector channels (4702b)

$\begin{array}{cc}{h}_1=\left[\begin{array}{c}{h}_{11}\\ {}^{j\left({\theta }_{12}-{\theta }_{11}\right)}h_{12}\end{array}\right]h_2=\left[\begin{array}{c}{h}_{21}\\ {}^{j\left({\theta }_{22}-{\theta }_{21}\right)}h_{22}\end{array}\right]& \left(12\right)\end{array}$

[0559] These channel in (12) or channel state information (CSI) is fed back to the base station (4702b) that computes the DIDO precoder as described in the following subsection.

• • [0560] (c) DIDO Precoding with Pre-compensation (4703)

[0561] The base station receives the channel state information (CSI) in (12) from the users and computes the precoding weights via block diagonalization (BD) (4703a), such that

w₁^Th₂=0, w₂^Th₁=0  (13)

where the vectors h₁ are defined in (12) and w_m=[w_m1,w_m2]. Note that the invention presented in this disclosure can be applied to any other DIDO precoding method besides BD. The base station also pre-compensates for the frequency offset by employing the estimate in (7) and phase offset by estimating the delay (Δt₀) between the second training transmission and the current transmission (4703a). Finally, the base station sends data to the users via the DIDO precoder (4703b)

[0562] After this transmit processing, the signal received at user 1 is given by

$\begin{array}{cc}\begin{array}{c}{r}_1\left[t\right]={}^{{\mathrm{j\Delta \omega }}_{11}T_s\left(t-{\tilde{t}}_{11}-\Delta t_o\right)}h_{11}\left[w_{11}x_1\left[t\right]+w_{21}x_2\left[t\right]\right]\\ ={}^{{\mathrm{j\Delta \omega }}_{12}T_s\left(t-{\tilde{t}}_{12}-\Delta t_o\right)}h_{12}{}^{-{\mathrm{j\Delta \omega }}_TT_s\left(t-\Delta t_o\right)}\left[w_{12}x_1\left[t\right]+w_{22}x_2\left[t\right]\right]\\ ={\gamma }_1\left[t\right]\lfloor h_{11}\left(w_{11}x_1\left[t\right]+w_{21}x_2\left[t\right]\right)+\\ {}^{j\left({\mathrm{\Delta \omega }}_{11}t_{11}-{\mathrm{\Delta \omega }}_{12}t_{12}\right)T_s}h_{12}\left(w_{12}x_1\left[t\right]+w_{22}x_2\left[t\right]\right)\rfloor \\ ={\gamma }_1\left[t\right][\left(h_{11}w_{11}+{}^{j\left({\theta }_{12}-{\theta }_{11}\right)}h_{12}w_{12}\right)x_1\left[t\right]+\\ \left(h_{11}w_{21}+{}^{j\left({\theta }_{12}-{\theta }_{11}\right)}h_{12}w_{22}\right)x_2\left[t\right]]\end{array}& \left(14\right)\end{array}$