PCS signal separation in a one dimensional channel

[0002] The present invention relates to wireless communications, more specifically, to a method and apparatus for detecting users at a time division multiple access (“TDMA”) receiver.

[0003] Prior art TDMA cellular/personal communications system (“PCS”) receivers generally include a series of processing stages for exploiting the features of a one dimensional signal stream such as the output signal of a single analog to digital converter.

[0004] With reference to FIG. 1, a receiver may include in its processing stages a spatial filter 11, a polarization filter 13, a channelizer filter 15, and a multi-user detection filter 17. Each stage acts as a user-population filter to separate the users in the receive environment based on polarization, spatial, and frequency characteristics as illustrated by the concentric subsets shown in FIG. 2.

[0005] The performance of each stage is a factor in designing a workable receiver architecture. Cost and complexity are dominant features. Frequently, receiver design is also constrained by the number of available observation array elements due to limited access to the geographic areas selected for deploying antenna assets.

[0006] Multi-user detection is the last stage of the receiver. A one-dimensional channelized signal stream which contains a cluster of cochannel signals is fed to the last stage. The signal stream may be TDMA-staggered (e.g., GSM, IS-54, IS-136) but contains multiple signals within a common time-slot.

[0007] In dense wireless environments, cochannel interference and intersymbol interference are significant problems for the detection of multiple users. Improving the performance of the last stage in detecting users may simplify the processing burden on preceding stages.

[0008] Generally, cochannel interference is caused by a multitude of other signals impinging upon the communication receiver. For TDMA systems, numerous techniques exist for cancelling cochannel interference. One known cancellation technique attempts to eliminate interference through a filtering process using the demodulation-remodulation technique and another method uses the cyclostationary signal processing technique.

[0009] The demod-remod technique uses successive stages of cancellation to eliminate one signal at a time. One problem with this technique is that signals must be widely spaced in power from each other or the error-rate is too high to perform effective remodulation/cancellation. Acquisition time may also be an issue because of the time needed to allow each stage to pull in.

[0010] Under certain conditions cyclostationary signal processing known as frequency-shift filtering (“FRESH”) can be used to separate interfering signals. The disadvantages of FRESH are that the symbol rates and carrier-frequencies must be distinct, i.e., the frequencies must be known with high accuracy and the channel transfer function must be known for each signal. Finally, the pulsed nature of TDMA is not very amenable to the FRESH architecture.

[0011] Another known cancellation technique does not attempt to eliminate the interference, but jointly detects all signals simultaneously. Joint maximum likelihood sequence estimation (“JMLSE”) is an example of a joint detection technique.

[0012] JMLSE is known in the art to be the highest performing technique for removing co-channel interference. JMLSE directly embraces the signaling probability functions and is capable of exploiting carrier-frequency and symbol-rate differences. However, the main drawback of JMLSE is its complexity.

[0013] Intersymbol interference (“ISI”), also known as multipath interference, as opposed to cochannel interference is generally caused by multiple signal propagation paths. These multiple paths result from the signal being scattered by a myriad of objects prior to reaching the receiver. The resulting interference is a form of self-interference which lowers the performance of the system.

[0014] Modern communication theory has devised multiple techniques for combating ISI-induced performance loss. In sequential order of best-to-worst performance, they are: the matched-filter bound technique, the maximum likelihood sequence estimator (MLSE), the decision feedback equalizer (“DFE”) and the linear equalizer.

[0015] DFE's may be either fractionally-spaced or symbol-spaced, where the fractionally-spaced equalizer is higher performing. DFE operation is often several dB better than a linear equalizer which may amplify noise when channel nulls are present.

[0016] The matched-filter bound technique is not practical in application, because isolated pulses are sent at a relatively-slow symbol rate.

[0017] The MLSE receiver is theoretically the top-performing technique. However, MLSE is usually too complex to be used in many applications. Nevertheless, most TDMA cellular receivers use MLSE because the multipath spread usually only extends across a few symbols.

[0018] In systems employing an MLSE receiver, a transmitter 10 such as shown in FIG. 3 may generate a 3-tap channel 4-ary pulse-amplitude-modulation (“PAM”) signal with a transmit symbol A_kat time k taken from the signal set {−3, −1, 1, 3}. The noise-free signal arriving at the receiver can be represented by a finite state machine (“FSM”) as shown in FIG. 4. Since there are only two delay registers 11 and 12, the FSM only remembers the last two inputs, A_kand A_k−1(A_k, A_k−1) which define the state. The channel transfer function for this example is:

[0019] H(z)=−0.5+z⁻¹−0.25z⁻²  (1)

[0020] Given a new input A_k+1, a new output voltage is generated and a new state results as shown in FIG. 5. The output voltage and new state are a function of the current state and the new input. Since the new input A_k+1can be one of four values, each state has four exiting branches.

[0021] Generally, given an M-ary constellation with an (L+1)-tap channel, the number of states is equal to M^L, and the number of branches entering and exiting each state is M.

[0022] For each noisy received sample, M*(M^L) many branch metrics are computed and added to the preceding state metric. All branches entering the output state are eliminated except for the one corresponding to the minimum cumulative distance metric. The output state with the best (i.e., smallest) metric is used to make a symbol decision by tracing back into the trellis a certain depth (the “traceback depth”). The traceback depth is usually about five times the channel length.

[0023] Therefore, the complexity is roughly proportional to M^L. The complexity can become prohibitively large for implementation with high-dimensional modulations where M is large (e.g.,M=16) and channel impulse responses which are long (e.g., L=8). Thus, both the storage demands and the operation complexity may become unmanageable.

[0024] It is object of the present invention to provide a novel joint MLSE/DFE technique for user detection in a TDMA system.

[0025] It is yet another object of the present invention to provide a novel method and-apparatus for detecting multiple users in a TDMA system with an MLSE operating on a signal stream having a cancelled postcursor portion.

[0026] These and many other objects and advantages of the present invention will be readily apparent to one skilled in the art to which the invention pertains from a perusal of the claims when read in connection with the appended drawings, and the following detailed description of the preferred embodiments.

[0027] FIG. 1 is a functional block diagram illustrating the processing stages of a prior art TDMA receiver.

[0028] FIG. 2 is a concentric signal-filter plot illustrating an example of spacial, polarized, and frequency filtered subsets of cellular signals received by the prior art TDMA receiver of FIG. 1.

[0029] FIG. 3 is a functional block diagram illustrating a channel memory effect on a signal in a TDMA system.

[0030] FIG. 4 is a signal versus time graph illustrating the channel impulse response of FIG. 3.

[0031] FIG. 5 is a state trellis diagram illustrating the trellis for the receiver of the TDMA system of FIG. 3.

[0032] FIG. 6 is a functional block diagram illustrating one embodiment of the user-detector of the present invention.

[0033] FIG. 7 is a signal-filter plot illustrating postcursor and precursor separation of a TDMA signal for the embodiment of FIG. 6.

[0034] FIG. 8 is a graph illustrating the impulse response of a TDMA signal.

[0035] FIG. 9 is a simplified functional block diagram of FIG. 6 in a communication system where the signal sequence is noiseless.

[0036] FIG. 10 is a simplified functional block diagram of FIG. 9 where the postcursor is cancelled.

[0037] FIG. 11 is a simplified state trellis diagram of the state trellis of FIG. 5.

[0038] FIG. 12 is a path metrics diagram illustrating state metrics links for an MLSE-VA stage.

[0039] FIG. 13 is a path metrics diagram illustrating the path decision metrics of an embodiment of the present invention.

[0040] FIG. 14 is a functional block diagram illustrating a second embodiment of the user-detector of the present invention.

[0041] With reference now FIG. 6, a single user detector 50 includes a receive filter 20, an analog-to-digital converter 21, a channel estimator 22, a joint MLSE/DFE stage 26, and a combiner 25. The joint MLSE/DFE stage 26 includes an MLSE-VA stage 27 and a DFE filter 28.

[0042] In operation, a signal stream may be received at the receiver filter 20. The receiver filter 20 may perform noise limiting and bandwidth limiting operation on the signal stream. The receive filter 20 may provide the filtered signal to the analog-to-digital converter 21. The analog-to-digital converter 21 may perform an analog-to-digital signal conversion and provide a digital signal to the channel estimator 22 and the combiner 25. The channel estimator 22 may include an impulse response filter 23 and an impulse response separator 24. The impulse response filter 23 provides the end-to-end channel impulse response.

[0043] The impulse response calculator separates the impulse response estimate into a precursor and postcursor portion. Generally, and as shown in FIG. 8, channel taps following the largest tap are the postcursor components, while taps preceding the largest tap are the precursor components. The impulse response separation may decompose the signal into precursor and postcursor components by detecting the largest tap. However, separating the channel impulse response based on the largest tap is not required.

[0044] The channel estimator 22 may provide the postcursor and precursor portions of a signal stream to the joint MLSE-VA 26. The MLSE-VA 27 estimates a likely symbol from the precursor portion using full traceback depth decision and provides an estimate of the postcursor portion to the DFE 28 using partial traceback depth decisions. The DFE 28 may receive the postcursor portion from the channel estimator 22 and the estimate of the postcursor portion from the MLSE-VA 27. The DFE 28 provides a signal to the combiner 25 for truncating the signal stream. The MLSE-VA 27 estimates a likely symbol sequence from the truncated channel impulse response using full traceback depth decisions. The DFE-truncated signal stream is simpler for the MLSE-VA to process for user detection. For an M-ary modulation with a channel postcursor response of N_Btaps, the overall complexity of the MLSE-VA user detector of the receiver is reduced by a factor M^NB.

[0045] The 4-ary PAM and a 3-tap channel system of FIG. 3 may illustrate the reduction in complexity. The roots of the channel transfer function (1) are computed:

[0046] H(z)=−0.5+z⁻¹−0.25z⁻².=−0.5z⁻¹(z−0.29289)(z−1.7071).  (2)

[0047] The first zero (z=0) is a phase shift. The second zero (z=0.29289) is a postcursor. The third zero (z=1.7071) is a precursor.

[0048] The system of FIG. 3 with a receiver having the user detector.of FIG. 6 simplifies to the system of FIG. 9. With reference to FIG. 9, the DFE 28 operates on the only postcursor component of the channel c₁and an estimate of the postcursor component A_k−1from the MLSE-VA 27. The rest of the channel components may be used to form a new finite-state-machine with a simplified MLSE trellis.

[0049] The system of FIG. 9 may be simplified for illustration to the system of FIG. 10. With reference to FIG. 10, the channel impulse response is only two taps long (c₀and c₋₁) Consequently, the number of states in the Viterbi trellis is four as shown in FIG. 11, instead of 16 (see FIG. 5).

[0050] A comparison of the four state trellis of FIG. 11 with the sixteen state trellis of FIG. 5 illustrates the great reduction in complexity. The complexity has been reduced by a factor of four. In a multi-user system the complexity may be reduced by a factor of 100 or more.

[0051] In operation, the MLSE-VA of FIG. 6 may have state metric links illustrated in FIG. 12. Branch metrics are computed after each receive sample arrives and the best cumulative branch metric is retained for each new output state. The best (lowest) metric then identifies the highest-probable transmit-symbol-sequence.

[0052] With reference to FIG. 13, the DFE 28 cannot function with decision delays, so the closest N_Bdecisions identified by the traceback path of the MLSE-VA 27 are loaded in the DFE 28. If these partial traceback decisions are correct, the postcursor component is eliminated. If a partial traceback decision is wrong, a DFE-induced noise burst occurs which the full traceback decision may correct. High performance is realized because maximum-likelihood decisions are used in all cases.

[0053] Channel impulse response is important for user detection. To identify the channel impulse response, TDMA cellular type signals are designed with overhead patterns.

[0054] For example, GSM uses eight different midamble sequences, each having impulsive autocorrelation properties over a span of 11 lags. One of the eight sequences is allocated to a user corresponding to their time-slot within a frame. The receiver correlates the received signal with the known midamble pattern to derive a channel estimate.

[0055] With reference to FIG. 14, a multi-user detector 70 includes a receive filter 40, an analog-to-digital converter 41, plural channel estimators 42, a joint MLSE/DFE stage 44, and a combiner 45. The joint MLSE/DFE stage 44 includes an MLSE-VA stage 47 and a DFE filter 48.

[0056] An advantage of the multi-user technique is that the end-to-end multi-channel impulse response contains a natural form of waveform coding (e.q., a unique multipath profile, time-of-arrival (TOA), amplitude level, and carrier phase).

[0057] In operation, the channel estimators 42 may jointly estimate the number of users and associated channels using correlated waveform training sequences and other signal features. Initially, the multiple DFE's 48 are set to a zero-data state. The impulse response separator 43 in each of the channel estimators may decompose each channel into postcursor and precursor portions. The postcursor portions are loaded into the multiple DFE's 48. The precursor portions are loaded into the MLSE-VA 47. The MLSE-VA 47 constructs a new trellis representing the reduced complexity multi-user finite state machine. The MLSE-VA 47 may detect multiple users from a received signal sample. The multiple DFE's 48 are updated by a partial traceback decisions of the MLSE-VA 47. The combiner 45 truncates the incoming signal by subtracting the output of the multiple DFE's 48 from the signal stream.

[0058] While preferred embodiments of the present invention have been described, it is to be understood that the embodiments described are illustrative only and the scope of the invention is to be defined solely by the appended claims when accorded a full range of equivalence, many variations and modifications naturally occurring to those of skill in the art from a perusal hereof.