Apparatus and method for timing synchronization in OFDM-based wireless systems

[0002] This invention relates to timing synchronization and, more particularly, to timing synchronization in wireless communications systems.

[0003] Recently, orthogonal frequency division multiplexing (OFDM) has been applied to communications systems and, in particular, to wireless communications systems. In OFDM a plurality of frequency multiplexed relatively narrow band signals are employed to form a wide band, “high” speed radio transmission link. In such an arrangement timing synchronization is required to position the signal in an optimum sampling window and to ensure proper alignment of the phases of the plurality of narrow band signals. To this end, it is required to obtain a timing adjustment signal that is employed to adjust timing in order to obtain the properly alignment.

[0004] Indeed, in multiple access systems based on OFDM, it is critical that there be synchronous arrival of mobile user symbols at a base station. If a particular user's arriving symbols are offset in time relative to other synchronized user's arriving symbols, the result is interference to the arriving symbols of all the mobile users.

[0005] In prior arrangements, attempts have been made to generate a timing adjustment signal through the estimation of all of the signal transmission channel parameters.

[0006] One prior known arrangement for estimating signal parameters is described in U.S. Pat. No. 4,750,147 issued on Jun. 7, 1988 which is commonly referred to as “ESPRIT” (Estimation of Signal Parameters using Rotational Invariance Techniques). The ESPRIT technique is based on maximum likelihood entropy principles, or subspaces, for the estimation of sinusoid signal parameters. As such, ESPRIT cannot be directly applied to estimation of signal parameters in an OFDM system because ESPRIT is directed toward extracting all of the signal transmission channel parameters and there is insufficient data to realize this. Moreover, extracting all the signal transmission channel parameters generally requires a large amount of measurement data, which either is not available or results in significant system overhead. Indeed, in an OFDM system there are just too many parameters for all of them to be uniquely identified as is required in the prior ESPRIT arrangement.

[0007] Problems and limitations of prior known arrangements and techniques for signal transmission channel parameter estimation are overcome by employing a unique arrangement and/or method that does not require the estimation of all the signal transmission channel parameters.

[0008] Specifically, this is realized by employing a rotational invariance arrangement or technique that utilizes either spectral, and/or temporal, and/or spatial diversity present in the wireless system to generate a single prescribed parameter of the mobile signal transmission channel.

[0009] The single prescribed signal transmission channel parameter is obtained by forcing the rotational invariance arrangement or technique to yield the single largest eigenvector from a prescribed relationship of signal data vectors of the signal transmission channel.

[0010] In one embodiment of the invention the prescribed parameter of the signal transmission channel is the delay that is substantially constant over the diversity, while the gain of each signal transmission channel path may vary. The estimated signal transmission channel delay is advantageously employed in a mobile station to adjust the timing alignment of the mobile signal being transmitted over the channel to the base station, thereby ensuring proper positioning of the transmitted mobile signal in the optimum sampling window at the base station.

[0011] FIG. 1 shows, in simplified block diagram form, details of the timing synchronization portion of a wireless system base station including an embodiment of the invention;

[0012] FIG. 2 shows, in simplified block diagram form, details of the timing synchronization portion of a wireless system mobile station useful in describing an embodiment of the invention;

[0013] FIG. 3 shows, in simplified block diagram form, details of the delay estimation unit of FIG. 1 including an embodiment of the invention; and

[0014] FIG. 4 is a flow chart illustrating the steps of the inventive delay estimation process employed in the estimate processing unit of FIG. 3.

[0015] It is noted that there is possible interference between same cell mobile users that can be avoided. To realize this, it is crucial that all the mobile users' symbol intervals are time-aligned. Indeed, because of multipath signal transmission, this timing control task does not reduce to simply estimating the arrival of a first signal from each of the same cell mobile users. Rather, it is necessary to estimate the time range of the signal transmission channel power-delay profile. The ultimate goal is to correctly place the channel cyclic prefix window, with no significant channel energy outside of it. Simply stated, the channel cyclic prefix window is placed to cover the signal transmission channel impulse response.

[0016] To realize timing control, a mobile station transmits a plurality of (pure) single frequency tones, i.e., carrier frequencies, for a prescribed interval to a base station. Then, the base station determines where the multipath profile is located, with the ultimate goal being the placement of the cyclic prefix. To this end, the base station needs to inform each of the mobile users a time interval by which to advance or delay transmission. A prescribed timing control interval, {overscore (τ)}+T, is determined in advance. Note that the center frequency of the ith tone is f_i=f_c+iΔ, and T is the symbol period, i.e., the period of the frequency tones. The first carrier frequency is f_cand the carrier frequencies are spaced apart by Δ=W/N=1/Thz, where W is the total bandwidth available to the system and N is the number of tones. Furthermore, {overscore (τ)} is analogous to the cyclic prefix, and in this instance is in the order of, for example, 100 μs, so that it covers propagation delays for all mobile users regardless of their distance to the serving base station. It is assumed that the signals from mobile users to be timing controlled arrive at the serving base station between time −{overscore (τ)}/2 and {overscore (τ)}/2, where time “0”is defined arbitrarily as the instance the sampling window starts. This relatively large cyclic prefix {overscore (τ)}, as well as the multiple tone transmission, are what distinguish the timing control from ordinary data transfer.

[0017] The signal transmission channels are assumed to have multiple attenuated and delayed paths. To clarify, the signal transmission channel, let α_n(t), τ_n(t) be the nth path complex attenuation and delay, respectively. If a mobile user transmits M pure tones, $\begin{array}{cc}s\left(t\right)=\underset{i^{\prime }=1}{\sum ^M}{e}^{\mathrm{j2\pi }{f}_{i^{\prime }}t}.& \left(1\right)\end{array}$

[0018] Then, the signal received at the serving base station is $\begin{array}{cc}r\left(t\right)=\underset{i^{\prime }=1}{\sum ^M}\sum _n{\alpha }_n\left(t\right)e^{\mathrm{j2\pi }\left(f_{i^{\prime }}+j_{D,n}\right)\left(t-{\tau }_n\left(t\right)\right)}+w\left(t\right),& \left(2\right)\end{array}$

[0019] where −{overscore (τ)}/2≦τ_n(t)≦{overscore (τ)}/2, and w(t) is the noise and interference, and f_D,nthe Doppler shift of each path. In the following, for simplicity of notation the time dependency is dropped of α_n(t), τ_n(t) and θ_n(t).

[0020] The windowing and Discrete Fourier Transform (DFT) yield: $\begin{array}{cc}{x}_i=\frac{1}{T}{\int }_0^{T}r\left(t\right)e^{-\mathrm{j2\pi }{f}_{i}t}dt\approx \sum _n{\alpha }_n{e}^{-\mathrm{j2\pi }{f}_i{\tau }_n}+w_i,& \left(3\right)\end{array}$

[0021] where i=1, . . . , M. Note that the approximation results from neglecting the effect of the Doppler shift, and w_iis the sampled frequency response of the noise. Since the path attenuation does not vary from tone to tone, as in the case when the total, null-to-null bandwidth of the tones is less than the coherence bandwidth of the signal transmission channel, then $\begin{array}{cc}{x}_i=\sum _n{e}^{-{\mathrm{j\omega }}_i{\tau }_n}{\alpha }_n+w_i,& \left(4\right)\end{array}$

[0022] where i=1, . . . , M and ω_i=2πf_i. In matrix form, the data satisfies $\begin{array}{cc}\left[\begin{array}{c}{x}_1\\ ⋮\\ x_M\end{array}\right]=\left[\begin{array}{ccc}{e}^{-{\mathrm{j\omega }}_i{\tau }_1}& e^{-{\mathrm{j\omega }}_1{\tau }_2}& \dots \\ ⋮& ⋮& \dots \\ e^{-{\mathrm{j\omega }}_M{\tau }_1}& e^{-{\mathrm{j\omega }}_M{\tau }_2}& \dots \end{array}\right]\left[\begin{array}{c}{\alpha }_1\\ {\alpha }_2\\ ⋮\end{array}\right]+\left[\begin{array}{c}{w}_1\\ w_2\\ ⋮\end{array}\right],& \left(5\right)\end{array}$

[0023] namely,

[0024] x=Ha+w.  (6)

[0025] It is assumed that the wireless system has L diversity antennas. For example, the base station has L diversity avenues, i.e., channels. Practical values of L are small. Indeed, most current wireless systems employ two diversity antennas at each base station sector, i.e., cell. Then, L sets of data vectors x are obtained according to equation (6) above. The purpose of diversity antennas is to obtain substantially uncorrelated signal transmission channels. We have recognized that what has changed in the signal transmission channel is the path gains and not the path delays. Stated another way, the path delay is substantially constant over diversity, while the gain of each channel path may vary. Therefore, the ith tone data at the Ith antenna, X_i(l), becomes $\begin{array}{cc}{x}_i\left(l\right)=\sum _n{\alpha }_n\left(l\right)e^{-{\mathrm{j\omega }}_i{\tau }_n},& \left(7\right)\end{array}$

[0026] where i =1, . . . , M, l=1, . . . , L, and α_n(l), w_i(l) are substantially uncorrelated over diversity. Now since the delay matrix H remains unchanged, then the data model is

[0027] x(l)=Ha(l)+w(l),  (8)

[0028] where l=1, . . . , L, and α(l), w(l) are substantially uncorrelated over diversity.

[0029] The example described above is for spatial diversity. Spectral (frequency) and temporal (time) diversity may also be exploited. In the case of spectral diversity, L sets of tones are transmitted over the mobile signal transmission channel. The sets of tones should be separated in frequency by greater than the coherence bandwidth of the mobile signal transmission channel. To realize temporal diversity, the above noted set of tones is transmitted over the mobile signal transmission channel during L timing control intervals separated in time by greater than the coherence time of the mobile signal transmission channel. Indeed, combinations of diversity are possible, for example, when the mobile user transmits one set of tones during a timing control interval, then another set of tones in a timing control interval some time later, and the received signal is captured during each instance by two diversity antennas. Note that when employing temporal diversity, a requirement is that the transmission path delays stay substantially constant during the two timing control intervals.

[0030] FIG. 1 shows, in simplified block diagram form, details of the timing synchronization portion of a wireless system base station including an embodiment of the invention. Specifically, a received signal from a user mobile station 200 (FIG. 2), as indicated in Equation (2) and described above, is supplied to input terminal 101 and, then, to synchronization signal extraction unit 102. Unit 102 includes an optimum sampling window, as described above, and extracts the timing synchronization signal in well known fashion. The extracted signal is down-converted to yield a continuous time down converted and windowed signal. This signal is supplied to delay estimation unit 103. In certain applications the received signal may be modulated by synchronization symbols assigned to the particular user mobile station. In such an instance, it is necessary to remove the synchronization symbols from the received signal. To this end, the assigned symbols for the particular received signal are supplied from synchronization symbol database 104 to delay estimation unit 103 where they are used to remove the synchronization symbols from the received signal, as described below in conjunction with FIG. 3. Note if or when the mobile user station 200 is transmitting only pure tones, no synchronization symbols are required from database 104. Indeed, in certain applications synchronization symbol database 104 may not be used at all or even included in the implementation. Delay estimation unit 103 generates a delay estimate τ that is supplied as an output as signal timing control parameter {circumflex over (τ)} that, in turn, is supplied to format of timing error feedback signal unit 105. It is noted that the estimated delay parameter of the signal transmission channel can be either positive or negative. When the delay is negative, a timing control signal is generated having a corresponding positive adjustment interval, and when the delay is positive, a timing control signal is generated having a corresponding negative adjustment interval. Then, unit 105 formats the signal, including the synchronization timing control signal, into a timing error feedback signal to be transmitted to the particular mobile user station 200. The timing error feedback signal indicates the time interval that the transmit symbol clock has to be either advanced or delayed. Thereafter, the formatted signal perhaps in the form of a message is supplied as an output via output terminal 106.

[0031] FIG. 2 shows, in simplified block diagram form, details of the timing synchronization portion of a wireless system mobile user station 200 useful in describing an embodiment of the invention. A received timing error feedback signal from base station 100 is supplied via input terminal 201 to clock offset adjustment unit 202. Unit 202 extracts the transmit offset value from the received timing error feedback signal in known fashion. In one example, the interval may be in microseconds that the transmit clock has to be advanced or delayed. If the offset value is in a message the interval is typically expressed in time increments. The transmit clock offset is supplied to variable time adjustment unit 203. Also supplied to variable time adjustment unit 203 is the transmit symbol clock from transmit symbol clock unit 204. Variable time adjustment unit 203, in response to the transmit clock offset, adjusts the transmit clock accordingly, in well known fashion. As indicated above the transmit clock is advanced or delayed by the transmit offset interval so that the transmitted signal from user mobile user station 200 will be substantially centered in the optimum sampling window in base station 100. The adjusted transmit symbol clock is supplied to synchronization signal insertion unit 205 where it is used to properly insert the synchronization signal for transmission to the base station in any other current data signal.

[0032] FIG. 3 shows, in simplified block diagram form, details of the delay estimation unit 103 of FIG. 1 including an embodiment of the invention. The continuous-time and down-converted and windowed signal synchronization signal extraction unit 102 (FIG. 1) is supplied via input terminal 301 to Inverse Fast Fourier Transform (IFFT) unit 302. Unit 302 yields a set of complex values 303-1 through 301-N, one for each simultaneously received tone or sinusoid. Thus, N tones or sinusoids are received at base station 100. Complex values 303-1 through 303-N are simultaneously supplied to symbol divide unit 304. Also supplied to unit 304 are assigned symbols from synchronization symbol database 104 (FIG. 1). Each of the complex values 303-1 through 303-N is equal to the product of the frequency response of the signal transmission channel (sampled at the frequency/tone), and the assigned data symbol plus any noise introduced by transmit channel and/or by the transmitter and receiver. Unit 304 removes the assigned symbols by dividing them out of the complex values 303-1 through 303-N. The resulting output from unit 304 are noisy channel frequency response samples 305-1 through 305-N. Note that in some systems pure tones are transmitted and, therefore, there is no need to remove any symbols form the complex values and unit 304 is not necessary. Channel frequency response samples 305-1 through 305-N are supplied 30 to estimate processing unit 306. Unit 306 may be for example a digital signal processor (DSP). Estimate processing unit 306 utilizes channel frequency response samples 305-1 through 305-N, as described below in FIG. 4, to generate the single timing delay estimate {circumflex over (τ)} that is supplied as an output via terminal 307. Note that if the mobile station 200 is transmitting other tones shortly after the simultaneously transmitted tones or sinusoids employed above, as part of the same synchronization interval, other tones are also processed in the same fashion as the channel frequency response samples 305-1 through 305-N above, and the aggregate data is supplied concurrently to estimate processing unit 306. Similarly, if the base station 100 is receiving data simultaneously on two or more antennas, this received data is also processed in a similar fashion as described above and, then, supplied to estimate processing unit 306.

[0033] FIG. 4 is a flow chart illustrating the steps of the inventive delay estimation process employed in the estimate processing unit 306 of FIG. 3.

[0034] Complex values, i.e., channel frequency response samples, 305-1 through 305-N are N complex scalars that are the baseband signal transmission channel estimates. These estimates are grouped according to the coherence bandwidth M of the signal transmission channel and the diversity order such that N=ML. The coherence bandwidth is a relatively loosely defined quantity, indicating that the signal transmission channel frequency response is strongly correlated across M contiguous tones, or MΔHz, where Δ is the tone spacing in Hz. The diversity order is the number of such sets of M contiguous tones that are sufficiently decorrelated among them. As indicated above the diversity can be spatial, i.e., data is obtained from widely separated antennas, temporal, i.e., data is obtained from transmissions at different points in time, or spectral, i.e., data is obtained from transmissions in different parts of the transmission frequency band. Then, each diversity set contains M tones, and can be thought as an M-dimensional vector x_i, where i=1, . . . , L and L is the order of diversity, as described above. By way of an example, if the signal transmission channel is deemed to be substantially flat in a frequency band containing M contiguous tones, then obtained from two diversity antennas are two M-long vectors x₁and x₂, which are substantially uncorrelated. Note that diversity L=2. Further, note that not all of the L x_ivectors need to be mutually decorrelated or independent in order for the proper operation of estimate processing unit 306.

[0035] Referring now to the flowchart of FIG. 4, the delay estimation process in started in step 401 by inputting the complex scalars 305-1 through 305-N. The, step 402 causes the complex scalars to be arranged into vectors x_I, . . . , X_Las described above. Thereafter, step 403 forms the autocorrelation matrix $R=\frac{1}{L}\sum _{i=1}^L{x}_i{x}_i^{*},$

[0036] where “*” indicates the complex conjugate. Step 404 causes the computation of the eigenvector of matrix R of step 403 corresponding to the single largest eigenvalue, in accordance with the invention. This is realized, for example, by employing the so-called power method. See for example a book authored by G. Golub and C. Van Loan entitled Matrix Computations, The John Hopkins University Press, 1989. The found largest single eigenvector is denoted “b” of R. In step 405, vectors b₁and b₂are formed. This is realized by extracting the first M−K entries of vector b and denoting the resulting vector b₁, and similarly extracting the last M−K entries of b and denoting the resulting vector b₂. “K” is an implementation dependent overlap parameter, which is an integer chosen to be greater than or equal to “1” and strictly less than M/2. Typically, K is chosen such that K=1. Step 406 causes a complex scalar ψ to be found that exactly or approximately satisfies b₁ψ=b₂. In one example, this is realized by employing the least squares method or more appropriately the total least squares method also described in the Matrix Computationbook noted above. Finally, step 407 causes the delay estimate {circumflex over (τ)} is obtained. This is realized by dividing the angle of ψ by −2πΔwhere Δ is the tone spacing. In this example, the “angle” of a complex number ψ is θ, a number between −π and π, where ψ=re^jθ, r is the magnitude of ψ and j={square root over (−1)}. Then, in step 408 delay estimate τ is supplied as an output as timing control signal {circumflex over (τ)}.

[0037] The above-described methods and apparatus are, of course, merely illustrative of the principles of the invention. Indeed, numerous other methods or apparatus may be devised by those skilled in the art without departing from the spirit and scope of the invention.