Compensation of I/Q gain mismatch in a communications receiver

[0001] The subject matter of the present application is related to and may be advantageously combined with the subject matter of copending and commonly owned application U.S. patent application Ser. No. 09/616,161, to Yakhnich et at, filed Jul. 14, 2000, entitled “Method of Channel Order Selection and Channel Order Estimation in a Wireless Communication System,” incorporated herein by reference in its entirety.

[0002] The present invention relates generally to communications systems and more particularly relates to an apparatus for and a method of compensating for nonlinear effects or noise impairments including I and Q gain mismatch in a communications system.

[0003] In recent years, the world has witnessed explosive growth in the demand for wireless communications and it is predicted that this demand will increase in the future. In the case of cellular services, for example, there are already over 500 million users that subscribe to cellular telephone services and the this number is continually increasing. Eventually, in the not too distant future, the number of cellular subscribers will exceed the number of fixed line telephone installations. In many cases, the revenues from mobile services already exceeds that for fixed line services even though the amount of traffic generated through mobile phones is much less than in fixed networks.

[0004] Other related wireless technologies have experienced growth similar to that of cellular. For example, cordless telephony, two way radio trunking systems, paging (one way and two way), messaging, wireless local area networks (WLANs) and wireless local loops (WLLs). In addition, new broadband communication schemes are rapidly being deployed to provide users with increased bandwidth and faster access to the Internet. Broadband services such as xDSL, short range high speed wireless connections, high rate satellite downlink (and the uplink in some cases) are being offered to users in more and more locations.

[0005] In connection with cellular services, the majority of users currently subscribe to digital cellular networks. Almost all new cellular handsets sold to customers are based on digital technology, typically second generation digital technology. Currently, third generation digital networks are being designed and tested which will be able to support data packet networks and much higher data rates. The first generation analog systems comprise the well known protocols AMPS, TACS, etc. The digital systems comprise GSM, TDMA (IS-136) or CDMA (IS-95), for example.

[0006] The recent explosion in demand for portable wireless communication devices has prompted a large amount of research in the design of efficient radio frequency (RF) receivers. The requirement of receiver portability, however, limits the available battery power and consequently places severe constraints on the power consumption, physical size and weight of such devices. Therefore, miniature radio receivers that dissipate low power are highly desirable. The search for efficient RF receivers has focused on the development of simplified radio receiver architectures.

[0007] For decades, nearly all RF receivers have been based on the well known superheterodyne architecture. In a superheterodyning receiver, an antenna feeds the received RF signal to an RF amplifier. The amplified RF signal is then converted to an intermediate frequency (IF) by mixing it with a signal produced by an offset local oscillator. The resulting IF signal is then amplified by an IF amplifier and then shifted to baseband by mixing it with a signal from a second local oscillator. The baseband signal is then quantized in an analog-to-digital (A/D) converter and demodulated by a digital signal processing (DSP) based demodulator. Alternatively, the radio may use conventional hardware techniques to demodulate the baseband signal and extract the original information signal.

[0008] There are several disadvantages of the superheterodyne architecture, however, which make it impractical for low-power implementation. One disadvantage is that in order for an IF amplifier to produce sufficient gain in the IF signal, IF filters biased at large currents are needed which consume substantial power. Furthermore, these IF filters require numerous passive components which cannot be integrated onto a single chip with the rest of the receiver, adding to the size and cost of the receiver. Another significant drawback of the superheterodyne architecture results from the symmetry in mixing the RF signal with the signal from the offset local oscillator. The mixing generates not only the desired RF signal but also undesired image signals at an intermediate frequency above or below the offset local oscillator frequency. Removing the image signals, however, requires a very selective analog RF filter which complicates the design of the receiver and increases its cost.

[0009] The direct conversion receiver architecture avoids many of the above difficulties of the superheterodyne architecture. A block diagram illustrating the typical structure of a direct conversion receiver front end circuit is shown in FIG. 1. The receiver, generally referenced 10, comprises an antenna 12 coupled to an analog front end circuit 14 and a DSP based baseband signal processor 16.

[0010] The antenna 12 couples the received RF signal to an RF amplifier 18. The amplified RF signal is then converted directly to baseband (hence the term ‘direct conversion’) by mixing it with a signal produced by a local oscillator 22. Both I and Q baseband signals are generated via I and Q mixers 20, 26, respectively. The LO signal without any phase shift is input to the I mixer 20 while the LO signal phase is shifted 90 degrees via phase shifter 24 and is input to the Q mixer 26. The resulting baseband signals are then filtered by low pass filters 28, 32. The filtered analog I and Q signals are then converted to digital by A/D converters 30, 34. The digital I and Q signals are then processed digitally by the baseband processor 16 which demodulates the received signal using a DSP based demodulator.

[0011] Note that because the down-converted signal in the direct conversion design is centered at frequency zero, there is no image signal to be rejected making the filtering task relatively easy and thus permitting the use of active low pass filters. In addition, the direct conversion architecture relaxes the selectivity requirements of RF filters and eliminates all IF analog components previously required, allowing for a highly integrated, low-cost and low-power receiver. Due to these and other potential advantages, many vendors have been focusing efforts to designing direct conversion radio receivers.

[0012] Several well-known problems associated with direct conversion designs, however, include 1/f noise, multiplicative impairments and additive impairments (also referred to as DC-offset noise). These noise sources result in severe performance degradation of the receiver and reduce the detectability of the transmitted signal.

[0013] The 1/f noise (also known as flicker noise or pink noise) is an intrinsic noise phenomenon found in semiconductor devices, with a power spectral density inversely proportional to frequency. The coupling of 1/f noise with the received signal takes place primarily after down-conversion at the baseband amplifier. Since the baseband signal could be in the range of hundreds of microvolts RMS, the 1/f noise can potentially comprise a substantial fraction of the signal power, resulting in large signal distortion.

[0014] Note that in a superheterodyne architecture the IF signal is substantially amplified by an IF amplifier. Since the IF frequency is high enough that 1/f noise is negligible, the 1/f noise then becomes relatively insignificant when the IF signal is mixed down to baseband.

[0015] The DC-offset noise impairment is an offset voltage that appears in the signal spectrum at DC when an RF signal is converted directly to baseband. This offset value typically may dominate the desired signal and can substantially degrade the signal to noise ratio (SNR) if it is not removed.

[0016] The DC offset is derived from three main sources: transistor mismatch in the signal path, the LO signal leaking back to the antenna due to poor reverse isolation through the mixer and RF amplifier and then reflecting off the antenna and self-downconverting to DC via the mixer, a large near channel interferer leaking into the LO port of the mixer and self-downconverting to DC, 1/f noise from any operational amplifiers in the circuit and bias in the filters and amplifiers.

[0017] The first source is transistor mismatch in the signal path between the mixer and the I and Q inputs of the detector. With careful circuit design this effect could be largely minimized. The second cause of DC-offset occurs when the signal from the local oscillator, which is at the same frequency as the RF signal, leaks from the antenna and reflects off an external object and self-converts to DC. This local oscillator radiation also interferes with other nearby receivers tuned to the same frequency. Since this radiation may be one or more orders of magnitude stronger than the RF signal, this self-rectification and nearby interference introduce tremendous DC-offset noise after direct-conversion. Furthermore, the amount of DC-offset generated by the local oscillator radiation is difficult to predict since its magnitude changes with receiver location and orientation. Good circuit isolation techniques could reduce this effect to a certain extent, but it cannot be eliminated entirely. The DC offset problem is particularly acute in GSM systems. GSM uses Gaussian minimum shift keying (GMSK) modulation which has a peak at DC. The offsets generated, directly add to the spectral peak of the downconverted signal. Generally, these offsets are much larger than the RMS front-end noise, severely degrading the SNR of the receiver. One way of removing the offsets is to use AC coupling. This, however, would require impractically large capacitors in order not to remove the signal bearing spectrum around DC.

[0018] In one approach to eliminating the DC offset, an analog DC notch filter implemented as simple capacitive coupling can be used to remove most of the DC offset noise. This approach, however, can only be used when the information bearing signal is not at or near DC as is the case, for example, of simple two-tone signaling, i.e. FSK modulation whose spectrum comprises little energy at DC.

[0019] When a portion of the information signal is, however, at or near DC, this approach cannot be used. Many of the modulation schemes used in personal communications require low frequency information to achieve low bit error rates. Thus, simple two-tone signaling is not suitable for most RF communications applications since these applications require modulation schemes that are more spectrally efficient than FSK. In these other modulation schemes, the method of capacitively coupling the baseband signal before sampling fails because of the large signal-bearing spectrum near DC, and a notch filter at DC will remove significant portions of the signal.

[0020] Multiplicative impairments manifest as distortion of the I and Q phasor results when the components of the I and Q channels are either uniformly distorted or differentially distorted. To avoid this type of distortion requires that all the components in the I and Q channel paths have constant gain and phase characteristics across their dynamic range. To assure such constant gain and phase characteristics is difficult and costly.

[0021] Another source of multiplicative type noise is the sampling jitter of the sample and hold used at the input to the A/D converters which introduces differential phase distortion between the I and Q channels.

[0022] A block diagram illustrating the impairment model of the channel and the receiver front end circuit is shown in FIG. 2. The channel is modeled as a dispersive, fading channel 42. The phase/frequency offset manifests as multiplicative noise 44. The I/Q mismatch 46 results from nonsymmetrical low pass filtering. The impairment is modeled as multiplying 54 the imaginary signal Q by an attenuation mismatch g while the real signal is considered the unimpaired reference. The real and impaired imaginary signals are added via adder 58. The DC offset additive impairment is then added to the signal via adder 60.

[0023] A diagram illustrating the effects of the noise impairments including I/Q gain mismatch and DC offset on an 8-PSK constellation is shown in FIG. 3. The effects of the multiplicative and additive impairments are illustrated with respect to an 8-PSK signal. The ideal unimpaired 8-PSK constellation 70 is a circle centered at the origin in the I/Q plane. The multiplicative impairments (i.e. attenuation mismatch by the LPFs, sampling jitter, etc.) impart an elliptical shape to the constellation. DC offset impairments further impart a linear shift to the constellation making the center of the constellation a point other than the origin.

[0024] It is therefore desirable to have a mechanism for compensating for the multiplicative noise impairments, particularly I and Q gain mismatch, in a communications receiver that is efficient, relatively easy to implement either in hardware, software or a combination of the two and is not costly to implement. An I/Q gain mismatch correction mechanism would permit the use of protocols and modulation schemes that have significant portions of their spectrum around or near DC such as GMSK and 8-PSK in EDGE based GSM systems.

[0025] The present invention is a novel and useful apparatus for and method of multiplicative noise compensation for use in a communications receiver. The invention can be used to compensate for 1 and Q gain mismatch impairments generated in the hardware front-end portion of communication receivers. Such impairments may be caused, for example, by attenuation mismatch of the I and Q low pass filters and by any sampling jitter introduced by the sample and hold circuits before the A/D converters.

[0026] The invention can be used in a variety of applications including, but not limited to, communications receivers whose baseband processing is implemented in hardware, software or a combination of hardware and software. The present invention is useful in compensating for the I/Q gain mismatch in the hardware front-end portion of direct conversion receivers. The gain mismatch impairments must be corrected in order to achieve adequate performance, especially in the case of an EDGE system signal which uses GMSK or 8-PSK modulation.

[0027] The invention is operative to calculate an estimate of the I/Q gain mismatch by which each input sample is multiplied with to generate compensated samples. Rather than attempt to pinpoint the actual source of any particular multiplicative noise impairment, the invention works backwards to compensate for any and all multiplicative impairments. In the case where the receiver also compensates for DC offsets, the DC offsets are compensated for first, followed by compensation for the multiplicative (i.e. nonlinear) impairments. It is noted that compensating in the reverse order, i.e. first I/Q gain mismatch and then DC offset, yields incorrect results, assuming that order of addition of the impairments is as given in the model described hereinbelow.

[0028] The training sequence portion of the uncompensated input samples is used to generate the I/Q gain mismatch estimate. In accordance with the present invention, the convolution matrix H, which when multiplied with the channel estimate yields sample estimates of the training sequence, is used to calculate the gain mismatch estimate and is pre-calculated for several channel lengths and stored in memory. Alternatively, the convolution H matrix can be built on the fly for any desired channel length. In either case, a plurality of matrices can be pre-calculated or built depending on the application.

[0029] Throughout this document, references to pre-calculating one or more convolution H matrices are intended to encompass not only the pre-calculation of the convolution H matrix but also the re-building on the fly of one or more convolution H matrices from scratch for each channel length L. Note that the size of the convolution H matrix varies depending on the channel length and the number of training symbols transmitted. The scope of the invention is not limited by the technique used to obtain the H matrix.

[0030] For example, in the case of EDGE, the dimensions of the H matrix varies from 3×24 (channel length L=3) to 7×20 (L=7). Since the H matrix is complex, sufficient memory, e.g., ROM, etc., must be allocated to store both real and imaginary parts. Moreover, since the EDGE standard specifies 8 different training sequences, the amount of required memory must be multiplied by a factor of 8.

[0031] An estimate of the channel is generated which provides the channel length and the location in the input sample buffer of the first training sequence sample to be used in calculating the gain mismatch estimate. The channel length is used to determine the number of training sequence samples to be used and to select one of the previously calculated matrices.

[0032] The method of the invention can be performed in either hardware or software. An I/Q gain mismatch compensation module is presented which is operative to pre-compute a plurality of H matrices, one of which is subsequently selected for each received message in accordance with the estimated channel length. The gain mismatch estimate g is calculated based on the received training sequence samples and the uncompensated input samples are subsequently multiplied therewith to yield compensated output samples.

[0033] The I/Q gain mismatch compensation technique of the present invention can be implemented in either hardware, software or a combination of both. In the case of software implementation, a computer comprising a processor, memory, etc. is disclosed which is operative to execute software adapted to perform the I/Q gain mismatch compensation method of the present invention.

[0034] There is therefore provided in accordance with the present invention a method of compensating I and Q gain mismatch in a sequence of input samples received over a channel, the method comprising the steps of splitting the input samples into a real component and an imaginary component, calculating a gain estimate g given by $g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}$
wherein y is a vector of uncompensated training sequence input samples, H is a training sequence convolution matrix, h is a vector of channel estimation taps and multiplying the imaginary component by 1/g and summing the real component therewith to yield a sequence of compensated input samples.

[0035] There is also provided in accordance with the present invention a method of compensating I and Q gain mismatch in a sequence of input samples received over a channel, the method comprising the steps of pre-calculating a plurality of training sequence convolution H matrices, each H matrix corresponding to a different channel length L, calculating an estimate of the channel and determining a channel length L therefrom, selecting a number of uncompensated training sequence input samples y and one of the pre-calculated plurality of H matrices in accordance with the channel length, calculating a gain estimate g given by $g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}$
where h is a vector of channel estimation taps, splitting the input samples into a real component and an imaginary component and multiplying the imaginary component by 1/g and summing the real component therewith to yield a sequence of compensated input samples.

[0036] There is further provided in accordance with the present invention an apparatus for compensating I and Q gain mismatch in a sequence of input samples received over a channel comprising means for pre-calculating a plurality of training sequence convolution H matrices, each H matrix corresponding to a different channel length L, a channel estimation module adapted to generate an estimate of the channel including a channel length L, means for selecting a number of uncompensated training sequence input samples y and one of the pre-calculated plurality of H matrices in accordance with the channel length, means for splitting the input samples into a real component and an imaginary component, means for calculating a gain estimate g given by $g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}$
where h is a vector of channel estimation taps and means for multiplying the imaginary component by 1/g and summing the real component therewith to yield a sequence of compensated input samples.

[0037] There is also provided in accordance with the present invention a communications receiver for receiving and decoding an M-ary transmitted signal comprising a radio frequency (RF) front end circuit for receiving and converting the M-ary transmitted signal to a plurality of baseband input samples, a DC offset compensation module comprising processing means programmed to: split the input samples into a real component and an imaginary component, calculate a gain estimate g given by $g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}$
where y is a vector of uncompensated training sequence input samples, H is a training sequence convolution matrix, h is a vector of channel estimation taps, multiply the imaginary component by 1/g and summing the real component therewith to yield a sequence of compensated input samples, a demodulator adapted to receive the compensated baseband input samples and to generate a received signal therefrom in accordance with the M-ary modulation scheme used to generate the transmitted signal, a first decoder operative to receive the received signal and to generate a sequence of soft symbol decisions therefrom, a second decoder adapted to receive the soft bit values and to generate binary received data therefrom and wherein M is a positive integer.

[0038] There is also provided in accordance with the present invention a computer readable storage medium having a computer program embodied thereon for causing a suitably programmed system to compensate for 1 and Q gain mismatch in a sequence of input samples received over a channel by performing the following steps when such program is executed on the system: splitting the input samples into a real component and an imaginary component, calculating a gain estimate g given by $g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}$
where y is a vector of uncompensated training sequence input samples, H is a training sequence convolution matrix, h is a vector of channel estimation taps and multiplying the imaginary component by 1/g and summing the real component therewith to yield a sequence of compensated input samples.

[0050] The following notation is used throughout this document.

[0051]

Term	Definition
AC	Alternating Current
AMPS	Advanced Mobile Telephone System
ASIC	Application Specific Integrated Circuit
BER	Bit Error Rate
CDMA	Code Division Multiple Access
CPU	Central Processing Unit
CRC	Cyclic Redundancy Code
DC	Direct Current
DFE	Decision Feedback Equalizer
DSL	Digital Subscriber Line
DSP	Digital Signal Processor
EDGE	Enhanced Data rates for GSM and TDMA/136 Evolution
EEPROM	Electrically Erasable Programmable Read Only Memory
EEROM	Electrically Erasable Read Only Memory
EGPRS	Enhanced General Packet Radio System
EPROM	Electrically Programmable Read Only Memory
FEC	Forward Error Correction
FIR	Finite Impulse Response
FPGA	Field Programmable Gate Array
FSK	Frequency Shift Keying
GERAN	GSM EDGE Radio Access Network
GMSK	Gaussian Minimum Shift Keying
GSM	Global System for Mobile Communication
IF	Intermediate Frequency
ISDN	Integrated Services Digital Network
ISI	Intersymbol Interference
LO	Local Oscillator
LPF	Low Pass Filter
MAP	Maximum A Posteriori
MLSE	Maximum Likelihood Sequence Estimation
PSK	Phase Shift Keying
RAM	Random Access Memory
RF	Radio Frequency
RMS	Root Mean Square
ROM	Read Only Memory
SNR	Signal to Noise Ratio
SOVA	Soft Output Viterbi Algorithm
TACS	Total Access Communications Systems
TDMA	Time Division Multiple Access
VA	Viterbi Algorithm
WLAN	Wireless Local Area Network
WLL	Wireless Local Loop

[0052] The present invention is a novel and useful apparatus for and method of I/Q gain mismatch compensation for use in a communications receiver. The invention can be used to compensate for I/Q gain mismatch impairments generated in the hardware front-end portion of communications receivers. Such impairments may be caused, for example, by attenuation mismatch of the I and Q low pass filters and by any sampling jitter introduced by the sample and hold circuits before the A/D converters.

[0053] The invention can be used in a variety of applications including, but not limited to, communications receivers whose baseband processing is implemented in hardware, software or a combination of hardware and software. The present invention is useful in compensating for the I/Q gain mismatch impairment generated in the hardware front-end portion of direct conversion receivers. The gain mismatch noise impairments must be corrected in order to achieve adequate performance, especially in the case of GMSK and 8-PSK modulations in EDGE based GSM systems.

[0054] To aid in understanding the principles of the present invention, the method is described in the context of an I/Q gain mismatch compensation module. As described in more detail below, the I/Q gain mismatch compensation module is adapted to perform the following functions: calculate an estimate of the channel, pre-calculate a plurality of H matrices subsequently used to generate the estimate of the gain mismatch, calculate the gain mismatch estimate and multiply each input sample thereby.

[0055] Note that the receiver configuration presented herein is intended for illustration purposes only and is not meant to limit the scope of the present invention. It is appreciated that one skilled in the communication signal processing arts can apply the method of the invention to numerous other receiver topologies and modulation schemes as well.

[0056] A concatenated communications system wherein the receiver employs the I/Q gain mismatch compensation module of the present invention to reduce nonlinear impairments present in input samples is shown in FIG. 4. The communications system, generally referenced 80, comprises a concatenated encoder transmitter 82 coupled to a channel 94, and a concatenated decoder receiver 96. The transmitter comprises an encoder 84, interleaver 86, bit to symbol mapper 88, modulator 90 and transmit circuit 92.

[0057] Input data bits to be transmitted are input to the encoder which may comprise a forward error correction (FEC) encoder such as Reed Solomon, convolutional encoder, parity bit generator, etc. The encoder functions to add redundancy bits to enable errors in transmission to be located and fixed.

[0058] The bits output of the encoder are then input to the interleaver which functions to rearrange the order of the bits in order to more effectively combat error bursts in the channel. The bits output of the interleaver are then mapped to symbols by the symbol mapper. The bit to symbol mapper functions to transform bits to modulator symbols drawn from an M-ary alphabet. For example, an 8-PSK modulator converts input bits into one of eight symbols. Thus, the mapper in this case generates a symbol for every three input bits.

[0059] The symbols output from the mapper are input to the modulator which functions to receive symbols in the M-ary alphabet and to generate an analog signal therefrom. The transmit circuit filters and amplifies this signal before transmitting it over the channel. The transmit circuit comprises coupling circuitry required to optimally interface the signal to the channel medium.

[0060] The channel may comprise a mobile wireless channel, e.g., cellular, cordless, fixed wireless channel, satellite, or may comprise a wired channel, e.g., xDSL, ISDN, Ethernet, etc. It is assumed that noise is present and added to the signal in the channel. The transmitter is adapted to generate a signal that can be transmitted over the channel so as to provide robust, error free detection by the receiver.

[0061] It is noted that both the inner and outer decoders in the receiver have complimentary encoders in the system. The outer encoder in the system comprises the encoder, e.g., convolutional, etc. The inner encoder comprises the channel, which in one embodiment can be modeled as an L-symbol long FIR-type channel.

[0062] At the receiver, the analog signal from the channel is input to Rx front end circuitry 98 which demodulates and samples the received signal to generate uncompensated input samples 136. The linear impairments in the input signal which manifest as a DC offset in the input samples are removed by the DC offset detection and compensation module 100 which is operative to output DC offset compensated input samples. Multiplicative or nonlinear impairments in the input signal which manifest as an I/Q gain mismatch in the input samples are removed by the nonlinear impairment detection and compensation module 102 which is operative to output compensated input samples y(k) 134. The samples are then input to an inner decoder (i.e. equalizer) 106 and channel estimation block 104.

[0063] The DC offset impairments may be removed using any technique. One technique suitable for use with the present invention is described in U.S. patent application Ser. No. 09/789,437, filed Feb. 20, 2001, entitled “Compensation Of DC Offset Impairment In A Communications Receiver,” similarly assigned and incorporated herein by reference in its entirety.

[0064] The channel estimation is operative to generate a channel estimate h(k) that is used by the inner decoder equalizer. The channel estimation is generated using the compensated received input samples y(k) and the known training sequence trn(k).

[0065] Several methods of channel estimation that are known in the art and suitable for use with the present invention include, for example, a correlation method and a least squares method. The correlation method is described in detail in “GSM System Engineering,” A. Mehrotra, 1997, Chapter 6 and in the article “On the Minimization of Overhead in Channel Impulse response Measurement,” Y. Han, IEEE Transactions on Vehicular Technology, Vol. 47, No. 2, May 1998, pages 631-636. The least squares method of channel estimation is described in more detail in the articles “Improved Channel Estimation With Side Information,” A. S. Khayrallah, R. Ramesh, G. E. Bottomley, D. Koilpillai, IEEE, March 1997, pages 1049-1051 and “Impact of Blind versus Non-Blind Channel Estimation on the BER Performance of GSM Receivers,” D. Boss, T. Petermann, K. Kammeyer, IEEE Signal Processing Workshop on Higher-Order Statistics, Jul. 21, 1997, pages 62-67 and in the book “Adaptive Filter Theory,” S. Haykin, 1996, Chapter 11 (Method of Least Squares).

[0066] Another channel estimation technique suitable for use with the present invention is described in U.S. patent application Ser. No. 09/616,161, to Yakhnich et al, filed Jul. 14, 2000, entitled “Method of Channel Order Selection and Channel Order Estimation in a Wireless Communication System,”? similarly assigned and incorporated herein by reference in its entirety.

[0067] The equalizer is operative to use the channel estimate h(k) in generating the hard decisions. Note that a hard decision is one of the possible values a symbol s(k) can take. An example of an inner decoder is an equalizer which compensates for the ISI caused by the delay and time spreading of the channel. The function of the equalizer is to attempt to detect the symbols that were originally transmitted by the modulator.

[0068] Note that the equalizer is adapted to output hard symbol decisions and may comprise, for example the well known maximum likelihood sequence estimation (MLSE) based equalizer that utilizes the Viterbi Algorithm (VA), linear equalizer or decision feedback equalizer (DFE). Since the equalizer generates hard decisions only, soft decisions must subsequently be generated from the hard decisions.

[0069] Equalization is a well known technique used to combat intersymbol interference whereby the receiver attempts to compensate for the effects of the channel on the transmitted symbols. An equalizer attempts to determine the transmitted data from the received distorted symbols using an estimate of the channel that caused the distortions. In communications systems where ISI arises due to partial response modulation or a frequency selective channel, a maximum likelihood sequence estimation (MLSE) equalizer is optimal. This is the form of equalizer generally used in GSM, EDGE and GERAN systems.

[0070] The MLSE technique is a nonlinear equalization technique which is applicable when the radio channel can be modeled as a Finite Impulse Response (FIR) system. Such a FIR system requires knowledge of the channel impulse response tap values. The channel estimate is obtained using a known training symbol sequence to estimate the channel impulse response. Other equalization techniques such as DFE or linear equalization require precise knowledge of channel.

[0071] The hard decisions are input to a bitwise de-interleaver 108 which is operative to reconstruct the original order of the bits input to the transmitter. The output of the de-interleaver comprises the soft bit values in the original bit order. The de-interleaved soft bit values are subsequently input to the outer decoder 110 which generates binary receive data therefrom.

[0072] Note that if soft decisions are desired, the hard decisions output of the equalizer may be input to an optional soft output generator (not shown) which functions to generate soft bit decisions as a function of the hard decisions. A soft decision, in the ideal case, comprises the reliabilities of each possible symbol value. Soft decision information may be needed by the de-interleaver and soft decoder. Internally, the soft output computation module first computes soft symbol values which are subsequently converted to soft bit decisions.

[0073] Any suitable technique may be used to generate soft symbol information from hard decisions. One technique for generating the soft decisions is described in U.S. Pat. No. 5,457,704 to Hoeher et al., entitled “Post Processing Method and Apparatus for Symbol Reliability Generation,” incorporated herein by reference in its entirety. Another suitable soft symbol generation technique is described in U.S. Pat. No. 6,731,700, entitled “Soft Decision Output Generator,” incorporated herein by reference in its entirety. The technique of converting soft symbols to soft bits is described in detail in U.S. Publication No. US2002/0122510A1, entitled “Apparatus For And Method Of Converting Soft Symbol Information To Soft Bit Information,” incorporated herein by reference in its entirety.

[0074] The outer decoder typically comprises a soft decision decoder, i.e. it takes soft bit values as input, which functions to detect and correct errors using the redundancy bits inserted by the encoder. The outer decoder generates the binary receive data utilizing the soft bit input. Examples of the outer decoder include convolutional decoders utilizing the Viterbi Algorithm, etc. Soft decision Viterbi decoders have the advantage of efficiently processing soft decision information and providing optimum performance in the sense of minimum sequence error probability.

[0075] In addition to the technique of DC offset detection and compensation, the invention specifies a model of the hardware impairments and the order of compensation of linear and nonlinear impairments. A block diagram illustrating the order to the addition of and compensation of linear (i.e. DC offset) and nonlinear impairments in a communications receiver is shown in FIG. 5.

[0076] The hardware model, generally referenced 250, is divided into three portions including channel, front end and baseband signal processing. The channel portion comprises the FIR channel model 252. The front end portion comprises the hardware components making up the front end of the receiver and the baseband signal processing comprises the signal processing functions performed digitally in software.

[0077] In accordance with the invention, the total impairments are separated into nonlinear and linear types of impairments. The components in the front end including the antenna 253, RF amplifier 254, I/Q mixers 256, 258, local oscillator 262 and I/Q low pass filters 264, 266, contribute the nonlinear effects such as I/Q gain mismatch. The A/D converters 268, 270 contribute linear effects including the addition of DC offset. Note that other components in the front end may also contribute linear effects. In accordance with the hardware model, however, the linear effects are considered to be added via the A/D converters.

[0078] Accurate compensation of nonlinear impairments is difficult to achieve without first compensating for linear impairments. The order of compensation also effects the channel estimation subsequently performed. Therefore, in accordance with the present invention, channel estimation and hardware impairment compensation is separated into three stages: first, detection and compensation of linear impairments 100; second, detection and compensation of nonlinear impairments 102; and third, computation of the channel estimate 104.

[0079] The operation of the I/Q gain mismatch compensation module 100 will now be described in more detail. The mathematical derivation of the expressions for the estimate of the gain mismatch is presented hereinbelow. A block diagram illustrating the signal model used to derive the I/Q gain mismatch estimate is shown in FIG. 6.

[0080] The following terms defined below are used in the derivation of the expression for the DC offset estimate.

• • [0081] u and y are vector representations of the uncompensated received samples;
  • [0082] ŷ is the vector representation of the estimated received samples;
  • [0083] LS is the Least Square matrix which when multiplied with the received training sequence samples is used to derive an estimate of the channel;
  • [0084] H represents the convolution matrix H comprising the known training symbols, as transmitted on the channel, which when multiplied with the channel estimate yields sample estimates of the training sequence;
  • [0085] h is the vector of channel estimation taps;
  • [0086] g is the real scalar gain mismatch estimate we are seeking;
  • [0087] d is the difference vector;
  • [0088] C is the cost function;
    Note that u, y, ŷ, h, d and C are complex values.

[0089] The model of the multiplicative impairment problem, generally referenced 140, uses closed loop feedback to determine an estimate g of the gain mismatch. The goal is to find the g that yields a minimum cost function C. The model comprises real component 142, imaginary component 144, 1/g function 138, multiplier 146, j multiplier 148, summer 141, LS matrix multiplication 143, H matrix multiplication and summer 147 configured in a negative feedback loop wherein the gain mismatch compensation block 149 functions to calculate the gain mismatch estimate g. Note that it is assumed that a known training sequence, as transmitted on the channel, is used in determining the estimate. The training sequence is commonly sent as a preamble or midamble in the messages of many different communications schemes, e.g., GSM burst midamble.

[0090] In the model, the imaginary portion of each received training sequence sample is multiplied by the estimated gain mismatch g. The compensated imaginary portion of the training samples are summed with the real portion and the sum y is multiplied by the LS matrix to generate the channel estimate. The channel estimate is then multiplied by the H matrix resulting in estimates ŷ of the training sequence samples. The training sequence sample estimates are subtracted from the compensated received samples to generate the difference vector d. This difference vector is used in determining the gain mismatch estimate g.

[0091] Mathematically, the difference vector is represented by the following
d=y−ŷ  (1)
where
ŷ=Re{H·h}+j·g·Im{H·h}  (2)
The goal is to find the g that gives a minimum cost function C. C being a cost function defined as follows $\begin{array}{cc}\begin{array}{c}C={\left(\underset{\_}{y}-\underset{\_}{\hat{y}}\right)}^H\left(\underset{\_}{y}-\underset{\_}{\hat{y}}\right)\\ =\left({\underset{\_}{y}}^H-{\left(H_R{\underset{\_}{h}}_R\right)}^H+{\left(H_I{\underset{\_}{h}}_I\right)}^H+{\mathrm{jg}\left(H_I{\underset{\_}{h}}_R\right)}^H+{\mathrm{jg}\left(H_R{\underset{\_}{h}}_I\right)}^H\right)\\ \left(\underset{\_}{y}-H_R{\underset{\_}{h}}_R+H_I{\underset{\_}{h}}_I-{\mathrm{jgH}}_I{\underset{\_}{h}}_R-{\mathrm{jgH}}_R{\underset{\_}{h}}_I\right)\end{array}& \left(3\right)\end{array}$
where [ ]_R denotes the real part of the vector or matrix;

• • [0092] [ ]_l denotes the imaginary part of the vector or matrix;
  • [0093] [ ]^T denotes the transpose of the vector or matrix;
  • [0094] [ ]^H denotes the Hermetian of the vector or matrix;
    We then take the partial derivative with respect to g and set the result equal to zero and solve for g as follows $\begin{array}{cc}\frac{\partial C}{\partial g}=g{\underset{\_}{h}}_R^T{H}_I^T{H}_I{\underset{\_}{h}}_R+g{\underset{\_}{h}}_R^T{H}_I^T{H}_R{\underset{\_}{h}}_I+g{\underset{\_}{h}}_I^T{H}_R^T{H}_I{\underset{\_}{h}}_R+g{\underset{\_}{h}}_I^T{H}_R^T{H}_R{\underset{\_}{h}}_I-{\underset{\_}{y}}_I^T{H}_I{\underset{\_}{h}}_R-{\underset{\_}{y}}_I^T{H}_R{\underset{\_}{h}}_I=0& \left(4\right)\end{array}$
    Setting equal to zero yields the following expression for the gain mismatch estimate g $\begin{array}{cc}g=\frac{\mathrm{Im}\left({\underset{\_}{y}}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}{\mathrm{Im}\left({\underset{\_}{h}}^H{H}^H\right)\mathrm{Im}\left(H\underset{\_}{h}\right)}& \left(5\right)\end{array}$
    where y is a vector of uncompensated training sequence input samples, H is the convolution matrix derived from the known training sequence, h is a vector of channel estimation taps.

[0095] It is noted that the convolution matrix H can be computed a priori since it is a function of the known training sequence symbols, as transmitted on the channel. The dimensions of the H matrix, however, are different depending on the number of training sequence symbols and the channel length L. The H matrix is constructed by creating a matrix having (T−L+1) rows and L columns and filling it with the known training samples as shown below, where T is the number of symbols in the training sequence and t represents the training symbols. $\left[\begin{array}{ccccc}{t}_{L-1}& \dots & t_2& t_1& t_0\\ t_L& \dots & t_3& t_2& t_1\\ t_{L+1}& \dots & t_4& t_3& t_2\\ ⋮& \dots & ⋮& ⋮& ⋮\\ t_{T-1}& \dots & t_{T-L+2}& t_{T-L+1}& t_{T-L}\end{array}\right]\hspace{1em}$

[0096] The dimensions of the H and LS matrices are given by (T−L+1)×L and L×(T−L+1), respectively. In the case of a GSM burst, T equals 26. Therefore, a plurality of H matrices can be calculated beforehand, each corresponding to a different channel length, and stored in memory. During operation, the channel length, as determined by the channel estimation block, can be used to select the appropriate H matrix to use in calculating the gain mismatch estimate.

[0097] For example, in the case of a GSM EDGE based communications system, the channel length can range between 3 and 7. Thus, five H matrices can be calculated beforehand and stored in memory for use during operation of the receiver. The appropriate matrix for each received burst is selected using the channel length determined from the channel estimate.

[0098] The I/Q gain mismatch compensation module will now be described in more detail. A block diagram illustrating the I/Q gain mismatch compensation module of the present invention in more detail is shown in FIG. 7. Note that the implementation shown is one possible embodiment of the I/Q gain mismatch compensation module of the present invention. It is intended to operate with protocols wherein data is transmitted using messages comprising a training sequence, e.g., GSM, etc. It is appreciated that one skilled in the art can apply the principles of the present invention, in particular the expression for the gain mismatch estimate g, to construct alternative implementations of the I/Q gain mismatch compensation module depending on the application requirements without departing from the scope of the present invention.

[0099] The gain mismatch compensation module, generally referenced 102, comprises a channel estimate 120, gain estimation block 122, H matrix pre-calculation block 128, H matrix storage memory 126 and H matrix selection block 124. In addition, the module 102 comprises means 130, 132 for splitting the received input samples into real and imaginary components, respectively, inversion block 135, multiplier 139 and adder 137. In operation, uncompensated samples u 136 from the receiver front end 98 (FIG. 4) enter the compensation module that is operative to generate an estimate g of the gain mismatch. The estimate g is inverted and the result is multiplied by the imaginary part of the input sample. The resulting product is added to the real part to yield compensated output samples 134.

[0100] It is appreciated that although the expression for the gain mismatch estimate is derived for the imaginary part of the input samples, it can also be derived in similar fashion for the real part as well. In this case, the multiplier is in the real signal path instead of the imaginary signal path.

[0101] As described hereinabove, the calculation of the gain mismatch requires the known training sequence, as transmitted on the channel, and the received training sequence samples. In accordance with the invention, the H matrix is optionally pre-calculated for a plurality of different channel lengths via the H matrix pre-calculation block 128. The resulting set of H matrices is stored in the H matrix memory storage 126 which may comprise any suitable memory storage type such as RAM, ROM, etc. or any other volatile or non-volatile type memory. The pre-calculation unit utilizes the known training sequence symbols 127, as transmitted on the channel, to calculate the H matrices.

[0102] During the receipt of messages comprising data, one of the H matrices previously calculated is selected and used in generating the estimate of the gain mismatch. As described supra, the particular matrix is chosen in accordance with the channel length L. The channel length is determined by the channel estimate block 120 that generates an estimate of the channel using the uncompensated input samples 136 and the known training sequence symbols 127, as transmitted on the channel. The channel estimate is generated using any of the known techniques described supra.

[0103] The output of the estimation block comprises the vector or channel estimation taps h, the channel length L 125 and the sample within the buffer of input samples corresponding to the first training sequence sample to be used in generating the gain mismatch estimate. The channel length is also used by the H matrix selection block to determine which of the pre-calculated H matrices to use in generating the gain mismatch estimate. The gain estimation block 122 extracts the appropriate number of received training sequence samples u using the channel length. In the case of an EDGE system, for example, the number of samples extracted is equal to (26−L+1).

[0104] The gain mismatch estimate g is then calculated using the received training sequence samples, appropriate H matrix and vector channel estimation taps h in accordance with Equations 5 and 6. The gain mismatch estimate g is then inverted to yield 1/g via inversion block 135. The imaginary part of each input sample is then multiplied by 1/g thus compensating for I/Q gain mismatch. The compensated output samples 134 are subsequently input to the equalizer 106 (FIG. 4).

[0105] A GSM Enhanced General Packet Radio System (EGPRS) mobile station constructed to comprise means for performing the I/Q gain mismatch compensation method of the present invention is presented. A block diagram illustrating the functional processing blocks in a GSM EGPRS mobile radio station is shown in FIG. 8. The EGPRS system is a Time Division Multiple Access (TDMA) system wherein eight users are able to share the same carrier frequency.

[0106] The GSM EGPRS mobile station, generally referenced 150, comprises a transmitter and receiver divided into the following sections: signal processing circuitry 180, baseband codec 178 and RF circuitry section 176.

[0107] In the transmit direction, the signal processing portion functions to protect the data so as to provide reliable communications from the transmitter to the base station 164 over the channel 162. Several processes performed by the channel coding block 154 are used to protect the user data 152 including cyclic redundancy code (CRC) check, convolutional coding, interleaving and burst assembly. An EGPRS radio station is designed to provide reliable data communications at rates of up to 480 kbit/s. In the transmitter, the data bits are encoded with a rate ⅓ convolutional encoder, interleaved and mapped to 8-ary symbols. The resultant data is assembled into bursts of 142 symbols as shown in FIG. 9. Head and tail (i.e. guard) symbols are added in addition to a training sequence midamble that is added to the middle of the burst. Note that both the user data and the signaling information go through similar processing. The assembled burst is then differentially encoded and modulated in accordance with the GSM standard by a modulator 156 which may be implemented as a 3π/8 offset 8-PSK modulator or π/2 offset GMSK modulator.

[0108] In GSM, the training sequence is sent in the middle of each burst. Each fixed length burst 190 consists of 142 symbols preceded by a 3 symbol tail 192 and followed by a 3 symbol tail 200 and 8.25 symbol guard 202. The 142 symbols include a 58 symbol left data portion 194, 26 symbol training sequence 196 and another 58 symbol right data portion 198. Since the training sequence is sent in the middle of the burst, it is referred to as a midamble. It is inserted in the middle of the burst in order to minimize the maximum distance to a data bit thus minimizing the time varying effects at the beginning and end of the burst.

[0109] In the receive direction, the output of the baseband codec is demodulated using a complementary demodulator 170. Several processes performed by the channel decoding block 172 in the signal processing section are then applied to the demodulated output. The processes performed include burst disassembly, linear effects compensation followed by I/Q gain mismatch compensation in accordance with the present invention, equalization, de-interleaving, convolutional decoding and CRC check. Optinally, soft symbol generation and soft symbol to soft bit conversion may be performed depending on the particular implementation.

[0110] The baseband codec converts the transmit and receive data into analog and digital signals, respectively, via D/A converter 158 and A/D converter 168. The transmit D/A converter provides analog baseband I and Q signals to the transmitter 160 in the RF circuitry section. The I and Q signals are used to modulate the carrier for transmission over the channel.

[0111] In the receive direction, the signal transmitted by the base station over the channel is received by the receiver circuitry 166. The analog signals I and Q output from the receiver are converted back into a digital data stream via the A/D converter. This I and Q digital data stream is filtered and demodulated by the demodulator before being input to the channel decoding block 172. Several processes performed by signal processing block are then applied to the demodulated output as described supra.

[0112] In addition, the mobile station performs other functions that may be considered higher level such as synchronization, frequency and time acquisition and tracking, monitoring, measurements of received signal strength and control of the radio. Other functions include handling the user interface, signaling between the mobile station and the network, the SIM interface, etc.

[0113] In another embodiment, a computer is operative to execute software adapted to perform the I/Q gain mismatch compensation method of the present invention. A block diagram illustrating an example computer processing system adapted to perform the I/Q gain mismatch compensation method of the present invention is shown in FIG. 10. The system may be incorporated within a communications device such as a receiver or transceiver, part of which is implemented in software.

[0114] The computer system, generally referenced 220, comprises a processor 222 which may be implemented as a microcontroller, microprocessor, microcomputer, ASIC core, central processing unit (CPU) or digital signal processor (DSP). The system further comprises static read only memory (ROM) 224 and dynamic main memory (RAM) 230 all in communication with the processor. The processor is also in communication, via bus 226, with a number of peripheral devices that are also included in the computer system. An A/D converter 238 functions to sample the baseband signal output of the front end circuit 234 coupled to the channel 232. Samples generated by the processor are input to the front end circuit via D/A converter 236. The front end circuit comprises RF circuitry, including receiver, transmitter and channel coupling circuitry.

[0115] One or more communication lines 242 are optionally connected to the system via I/O interface 240. A user interface 244 responds to user inputs and provides feedback and other status information. An optional host interface 248 connects a host device 246 to the system. The host is adapted to configure, control and maintain the operation of the system. The system may also comprise magnetic storage device 228 for storing application programs and data. The system comprises computer readable storage medium which may include any suitable memory means including but not limited to magnetic storage, optical storage, semiconductor volatile or non-volatile memory, biological memory devices, or any other memory storage device.

[0116] The I/Q gain mismatch method software is adapted to reside on a computer readable medium, such as a magnetic disk within a disk drive unit. Alternatively, the computer readable medium may comprise a floppy disk, Flash memory card, EEROM, EPROM or EEPROM based memory, bubble memory storage, ROM storage, etc. The software adapted to perform the I/Q gain mismatch compensation method of the present invention may also reside, in whole or in part, in the static or dynamic main memories or in firmware within the processor of the computer system (i.e. within microcontroller, microprocessor, microcomputer, DSP, etc. internal memory).

[0117] In alternative embodiments, the method of the present invention may be applicable to implementations of the invention in integrated circuits, field programmable gate arrays (FPGAs), chip sets or application specific integrated circuits (ASICs), wireless implementations and other communication system products.

[0118] It is intended that the appended claims cover all such features and advantages of the invention that fall within the spirit and scope of the present invention. As numerous modifications and changes will readily occur to those skilled in the art, it is intended that the invention not be limited to the limited number of embodiments described herein. Accordingly, it will be appreciated that all suitable variations, modifications and equivalents may be resorted to, falling within the spirit and scope of the present invention.