HK1112336B - A mimo jdgrake receiver and its data processing method - Google Patents

Description

MIMO JDGRAKE receiver and data processing method thereof

Background

Technical Field

The present invention relates generally to the field of wireless telecommunications, and more particularly to a mobile terminal including a Reduced Complexity (RC) Joint Detection (JD) Generalized R AKE (GRAKE) receiver that utilizes a cumulative metric (cumulative metric) across transmit antennas to enable the use of a reduced complexity tree search technique to determine soft bit values representing coded bits received from transmit antennas in a base station.

Background

Today, there is a great interest in developing ways to enhance data rates in multiple-input multiple-output (MIMO) antenna structures used in mobile communication systems, which enable High Speed Downlink Packet Access (HSDPA) provisioning of the Wideband Code Division Multiple Access (WCDMA) standard. For example, Code Reuse (CR) -BLAST (similar to V-BLAST) and per-antenna rate control (PARC) are two such techniques that may be used to enhance data rates in MIMO antenna systems. These two techniques are described in detail in the following two articles:

foschini et al, "Simplified Processing for High spectral efficiency Wireless Communication using a Multi-Element array," IEEEjournal on Selected Area of Communications, vol.17, p.1841-1852, 11 months 1999.

Chung et al, "adaptive Eigenmode BLAST ChannelCapacity Using V-BLAST with Rate and Power feed (BLAST channel capacity Using V-BLAST with Rate and Power Feedback Approaching Eigenmode)" Proc. IEEE VTC' 02-Fall, Atlantic City, NJ, 10 months 2001.

When applied to HSDPA systems, both CR-BLAST and PARC techniques employ multi-code transmission to take advantage of large capacity MIMO channels and thus deliver very high data rates to mobile terminals. CR-BLAST is a spatial multiplexing technique, meaning that one coded bit stream is interleaved over all transmit antennas, while PARC transmits a separate coded bit stream from each transmit antenna. Initially, receiver designs for MIMO systems employing such techniques often focused on the case of flat fading channels. However, in practice the channel is often dispersive (dispersive), thus leading to Multiple Access Interference (MAI) and intersymbol interference (ISI). Moreover, self-coherent interference occurs even in flat fading channels because the multi-codes used in HSDPA are reused (reuse) across transmit antennas to avoid code restriction issues.

Since the memory and/or processing power of mobile terminals is typically quite limited, a challenge in receiver design for dispersive MIMO scenarios is to achieve a good balance between performance and complexity in the receiver. This is particularly true because the number of signals that the receiver needs to demodulate is large due to multi-code and multi-antenna transmission. On one extreme of complexity, the scaling is a conventional RAKE receiver that performs poorly because RAKE is designed for white noise and ISI and MAI are colored. Moreover, conventional RAKE receivers perform poorly due to the auto-coherent interference from code reuse. The other extreme is occupied by a fully jointly demodulated receiver, which performs well but is extremely complex. Somewhere in the middle are for example those like Minimum Mean Square (MMSE) -GRAKE receivers, which employ some form of equalization either linear or decision feedback. Detailed descriptions of MMSE-GRAKE receivers are provided in the following articles:

"Generalized RAKE Receivers for MIMO Systems (RAKE Receivers for generalization of MIMO Systems)" in proc.vtc' 03-Fall, olando, FL, 2003, 10 months, by s.j.grant et al.

Although the MMSE-GRAKE receiver works well in frequency selective fading, it does not perform well severely in the case of slight dispersion or near flatness. Thus, Joint Detection (JD) -generalized RAKE receivers (JD-GRAKE receivers) have recently been developed to recover performance in such situations and are also described by Grant et al in the above reference article. The JD-GRAKE receiver may also be used in a MIMO configuration where the number of receive antennas is less than the number of transmit antennas. In these cases, JD-GRAKE receivers outperform MMSE-GRAKE at various levels of dispersion.

JD-GRAKE receivers are able to handle various types of interference by forming signal recall (reminiscent) partitions for group detection of CDMA. In particular, subsets of signals sharing the same channelization code are formed and joint detection is applied to the M signals in each subset, where M is the number of transmit antennas at the base station. This addresses interference due to code reuse. ISI and MAI from signals outside each subset are suppressed in a manner similar to that in a conventional single antenna GRAKE receiver. That is, ISI and MAI are treated as colored gaussian noise and exploit the correlation of interference across fingers on multiple receive antennas in adapting the finger (finger) delays and combining weights. This detection process is performed separately for each of the K channelization codes.

However, one problem with JD-GRAKE receivers is: when higher order modulation is used with a larger number of transmit antennas, the number of metrics to be calculated in forming the soft bit values required by the decoder becomes large. In particular, for M transmit antennas and a signal constellation comprising Q points, the number of metrics to be calculated per symbol period is Q^M. For example, for 16-QAM (Q-16) and M-4 transmit antennas, the number of metrics is 65,536, which is really a large number. One way to address this problem and avoid exponential complexity in JD-GRAKE receivers is to use a continuous Cancellation based receiver, as described in U.S. patent application serial No. 10/795,101 entitled "continuous interference Cancellation in Generalized RAKE receiver Architecture", filed 5.5.2004, which is incorporated by reference. By continuously detecting the M transmitted signals in a multi-stage scheme, exponential complexity in the continuous cancellation receiver is avoided.At the same time, the present invention addresses the complexity increase by introducing a technique that significantly reduces the number of metric computations for JD-GRAKE receivers, thereby allowing near-optimal joint detection to be performed in a single stage.

Disclosure of Invention

The present invention includes a technique for reducing the number of metric calculations that need to be performed by a JD-GRAKE receiver by reforming a metric such that the metric is cumulative over the transmit antennas, which in turn enables the use of a tree search technique (e.g., m-algorithm) of reduced complexity to calculate soft bit values that are then processed to determine the coded bits received from the transmit antennas. The resulting reduced complexity JD-GRAKE receiver is referred to herein as an RC-JD-GRAKE receiver.

Drawings

A more complete understanding of the present invention may be had by reference to the following detailed description when taken in conjunction with the accompanying drawings wherein:

FIG. 1 is a block diagram of a MIMO wireless communication network including a base station and a mobile terminal incorporating an RC-JD-GRAKE receiver in accordance with the present invention;

FIG. 2 is a block diagram illustrating in more detail the RC-JD-GRAKE receiver architecture shown in FIG. 1;

fig. 3 is an exemplary tree generated by the RC-JD-GRAKE receiver shown in fig. 2, illustrating 8 reserved branches in the case of Quadrature Phase Shift Keying (QPSK) (Q4) and M3 transmit antennas;

FIG. 4 is a flow chart illustrating the steps of a preferred method for reducing the complexity of processing a signal received at a mobile terminal by using the RC-JD-GRAKE receiver of FIG. 2 in accordance with the present invention; and

FIG. 5 is a block diagram illustrating in greater detail the components of one embodiment of a reduced search soft value generator for use in the RC-JD-GRAKE receiver of FIG. 2 in accordance with the present invention.

Detailed description of the invention

The details of the exemplary embodiments of the present invention discussed and illustrated must be of a certain level of complexity. Such complexity is studied in the exemplary details presented later herein, but an initial understanding of the broader aspects of the invention can be obtained with reference to the relatively simple diagram presented in FIG. 1. However, before discussing fig. 1, it should be understood that the present invention broadly relates to the use of GRAKE-based signal detection in conjunction with joint detection techniques.

As used herein, the term "GRAKE" means a RAKE combining circuit and/or combining method that calculates impairment correlations (impairment correlations) between streams of despread values that are RAKE combined by the circuit. Such impairments may be caused, for example, by MAI (multiple access interference), excessive spreading code reuse, channel fading conditions, and the like. It is noted that the MAI of the own cell (owncell) may be treated as interference, while MAIs of other cells may be treated as noise and/or interference. Regardless, however, as will be explained later herein, the GRAKE processing used in the present invention is suitable for joint detection processing and, as noted, the generation of soft bit values to be communicated to the decoder is by a reduced complexity tree search technique.

Referring to fig. 1, one sees a typical MIMO wireless communication network 100 comprising a base station 110 and a mobile terminal 120 incorporating an RC-JD-GRAKE receiver 125 configured in accordance with the present invention. It should be appreciated that many of the components and details associated with the base station 110 and the mobile terminal 120 described herein are well known in the industry. Accordingly, the description provided below omits those well-known components and details that are not necessary for an understanding of the present invention for the sake of clarity.

At the base station 110, the information bit stream 102 is first encoded and interleaved by the encoder/interleaver 104. The encoded bits are spatially tapped by a spatial tap 106 and a plurality of mapping spreaders 108 and then distributed to a plurality of transmit antennas 109. In particular, the spatial demultiplexer 106 and the plurality of mapping spreaders 108 process the bit stream 102 such that the coded bits (data stream) on each transmit antenna 109 are mapped to K modulation symbols and each symbol is spread using one of K channelization codes according to the WCDMA scheme. It should be appreciated that different data streams are transmitted from each transmit antenna 109. It should also be appreciated that the spatial tapping scheme described above may be replaced by using more than one encoder/interleaver, possibly with different coding rates, and assigning the resulting coded bit streams to one or more transmit antennas. The modulated and spread coded bits on each transmit antenna 109 are then converted to RF signals and transmitted to the mobile terminal 120 over a spatial channel represented by the term G. Channel G represents the medium response (medium response) from each transmit antenna 109 to each receive antenna 122 in the mobile terminal 120, and may be dispersive and time-varying. For the sake of simplicity, a static, time-invariant channel is considered, which is dispersive.

At the mobile terminal 120, a received signal r is included₁(t) to r_L(t) the received composite signal, which has been received by the L receive antennas 122, is processed by a receiver 124 comprising an RC-JD-GRAKE receiver 125 and a decoder/deinterleaver 126. The RC-JD-GRAKE receiver 125 processes the received signal and outputs soft bit values representing the coded bits. A decoder/deinterleaver 126 then processes the soft bit values and outputs a bit stream 128 that is representative of the bit stream 102 at the base station 110. A more detailed discussion of the different components of the RC-JD-GRAKE receiver 125 and how these components function to generate soft bit values using the techniques of the present invention is provided below with reference to fig. 2.

Referring to fig. 2, there is a block diagram illustrating in more detail the structure of the RC-JD-GRAKE receiver 125 for the kth multi-code (multi-code). RC-JD-GRThe first stage of AKE receiver 125 includes a number of correlators 202 (fingers 202) tuned to the kth multi-code, where the total number of correlators 202 across all receive antennas 122 is N. Each correlator bank (bank) on each receive antenna contains several correlators (fingers), and the actual number of correlators is typically larger than the number of channel taps. It should be observed that when code reuse MIMO is applied to HSDPA, K multi-codes are transmitted on each of the M transmit antennas 109 and the same code is reused on all transmit antennas 109. The output of the correlator 202 is sampled at time T-iT, where T is the symbol period. The despread values are grouped into length-N vectors y_k(i) (see FIG. 2) wherein the q-th element corresponds to a delay T_qAnd is given by:

here E_TIs the total transmitted energy per symbol, which is evenly divided among the multicodes and transmit antennas 109. Function r_knij(T) is the spreading waveform u during the ith symbol period_ki(t) and spreading waveform u during the (i-j) th symbol period_n，i-j(t) cross correlation function between. The waveform correlation varies from symbol to symbol due to the long code scrambling, so the subscript is "i". Noise component of correlator output is composed ofIt is given. Although not explicitly shown, receive antenna index 1 refers to a function of index q, as finger 202 spans multiple receive antennas 122. Time delay tau_lmpAnd complex gain coefficient g_lmpThe P-th tap of the P-th tap channel impulse response between the m-th transmit antenna 109 and the 1 st receive antenna 122 is depicted. One for despreading vector y_k(i) More useful forms are:

y_k(i)＝Hc_k(i)+x_k(i) (1B)

transmitted signal vector c_k(i)＝[c_1k(i)c_1k(i) L c_MK(i)]^TIncluding M symbols during the ith symbol period sharing the kth multicode. N × M gain matrix H ═ H₁h₂Λh_M]The MIMO channel is fully described, where each gain vector h_mThe channel between the mth transmit antenna 109 and the fingers of the multiple receive antennas 122 is depicted. Vector x_k(i) The impairment process consisting of ISI, MAI, and noise, i.e., the despreading vector y, is described_k(i) And c_k(i) The portion that is not relevant. The impairment covariance matrix is represented asThe matrix captures the correlation of interference across both the finger and multiple receive antennas. The expressions H and R are provided by Grant et al in the appendix to the previously mentioned article entitled "Generalized RAKE receivers for MIMO Systems_xFor a more detailed discussion.

Similar to the JD-GRAKB receiver described in S.J.Grant et al, the operation of the RC-JD-GRAKE receiver 125 is based on the vector c_k(i) Using a despreading vector y conditioned on a hypothesized (hypothesiis) symbol vector c and a gain matrix H_k(i) Knowledge of the Probability Density Function (PDF). For example, using despread values from the pilot channel transmitted on each antenna 109, the gain matrix H may be estimated. Further assuming that the impairment vector is Gaussian, the required PDF is compared to exp λ_k(c)]In proportion of where_k(c) Is a measure for the hypothesis c. After discarding the hypothesis-independent terms, the detection metric is given by:

λ_k(c)＝2Re[c⁺z_k(i)]-c⁺Sc(2)

by weighting the despreading vector y_k(i) The decision statistic z is generated and output by the combiner 204_k(i) The following were used:

z_k(i)＝W⁺y_k(i) (3)

wherein the GRAKE weight matrix is given by:

the M × M matrix, referred to herein as the S-parameter matrix and represented as the matrix symbol S, is given by:

and it represents the mixing between symbols (mixing) in the detection metric described in equation 2. The s-parameter matrix is the product of the combining weights and the channel gain matrix and is similar to the s-parameters in MLSE type equalizers.

As described in Grant et al, a conventional JD-GRAKE receiver uses an exhaustive (explicit) tree search to compute all Q' s^MA measure of the symbol vector hypothesis_k(c) In that respect Again, the number of computations that need to be performed by a conventional JD-GRAKE receiver can be very large. For example, for 16-QAM (Q-16) and M-4 transmit antennas, a calculated λ is needed_k(c) The number of measurements of (a) is 65536-this is really a huge number.

The RC-JD-GRAKE receiver 125 of the present invention solves the computational complexity problem of JD-GRAKE receivers by using a technique that significantly reduces the number of metrics that need to be computed. The RC-JD-GRAKE receiver 125 passes the metric λ_k(c) This complexity is reduced by assigning a special form (see equation 10) that allows the computation of fewer metrics λ using a non-exhaustive tree search technique_k(c) In that respect In this embodiment, the m-algorithm is used to reduce the complexity of the tree search. Specifically, the key to reducing this complexity by the m-algorithm is to use equation 2 as shownMeasure of (a)_k(c) Rewritten to a form accumulated on the transmitting antenna 109.

For this reason, the metric described by equation 2 is considered. Due to lambda_k(c) Is a measure for detecting c, so a term not containing c can be added to λ_k(c) Since this does not affect the detection process. Thus, a term is added to "complete the square" to get the following new metric:

equation 6 can be factored into the following form:

λ_k(c)＝-(c-S^-1z_k(i))⁺S(c-S^-1z_k(i)). (7)

order to

Equation 7 can be rewritten as:

it is noted that the itemsIs the first symbol estimate of the transmitted symbol vector c. First signal estimationSimilar to the estimate obtained from the Zero Forcing block linear equalizer described in "Zero Forcing and Minimum Mean-Square-Error Equalization for code-Division Multiple-Access Channels (Zero Forcing and Minimum Mean-Square Error Equalization for code Division Multiple Access Channels)" by Anja Klein et al, published 5.1996 at paragraph 5, volume 45, phase 2, p.276-. The contents of this article are incorporated herein by reference.

The final step of rewriting the metric is to perform a decomposition of the s-parameter matrix so that the metric in equation 9 can be represented as the sum of the branch (branch) metrics, which is causal (cause) over the set of transmit antennas. To this end, a Cholesky decomposition of the S-parameter matrix is performed, such that the S-parameter matrix S is factorized into the product of another matrix and its hermitian transpose. Watch (A)The matrix representing this factorization is referred to herein as the causal factorization (cause factorization) of the s-parameter matrix, and is denoted by the matrix symbol L₈Representation, which is an M lower triangular matrix with the (M, n) th element represented as l_mn. This factorization is represented as

Because L is_sIs of lower triangular shape, so_mnOnly when n ≦ m is non-zero. This allows the metric in equation 9 to be expressed asAlternatively, equation 9 is written as:

and the left-hand side of equation 10 is referred to herein as the cumulative metric. The terms in the right-hand sum, referred to herein as the branch metric, are denoted as λ_km(c) And is given as:

symbol c_nAndthe nth element and the first symbol estimate, respectively, of a hypothesis c of length MAs can be seen, the metrics in equation 10 are accumulated over the transmit antennas 109. Furthermore, each branch measures λ_km(c) Strictly positive and only dependent on the symbol hypotheses of the current and previous antennas 109. This fact allows to greatly reduce the number of metric calculations by using tree search techniques like the m-algorithm.

It should be appreciated that there are two reasons for performing the Cholesky decomposition on S to establish equation 10. The first is to make the metric in equation 9 writable as a vectorSquare norm of (d). The reason why the squared norm of this vector is desired is to make the branch metric λ_km(c) Is strictly positive. The second reason is the Cholesky factor L_mIs lower triangular, so that the norm of the square isWhen multiplied (i.e., the results of equations 10-11), one finds the branch metric λ_km(c) Relying only on symbols from the current and previous transmit antennas 109. In this way, the metric λ is accumulated_k(c) Is causal across transmit antennas. Those skilled in the art will understand that: there may be other possible factorizations of the s-parameter matrix that result in a cumulative metric λ_k(c) A cause and effect version of (1).

The m-algorithm operates on a tree with levels corresponding to the transmit antennas 109. Fig. 3 shows an example for QPSK and 3 transmit antennas 109, where only the 8 best cumulative metrics are retained. Of course, for the first level where there are fewer than 8 nodes in the tree, all cumulative metrics are retained. For a more detailed description of the m-algorithm, reference may be made to "Iterative Tree Search Detection for MIMO Wireless Systems" by y.l.c de long et al, proc.vtc' 02-Fall, wingowski, canada, 2002, "Iterative Tree Search Detection for MIMO Wireless Systems". The contents of this article are incorporated herein by reference.

After the GRAKB merge, the soft value generator 206 calculates the soft bit values needed by the decoder 126 by using the m-algorithm. As can be seen, vector c_k(i) Of the mth symbol (denoted as b)_mkr(i) Is expressed as Λ) soft value_mkr(i) In that respect To calculate it, the subsets C need to be separately assigned_mr1And C_mr0Is defined as containing b to it_mkr(i) Those subsets of the hypotheses of 1 and 0. The size of each of these subsets is Q^M/2. Further, F_kDefined as the set of candidate symbol vectors that contains the branch that remains after the m algorithm ends. This set and subset C_mr1And C_mr0The intersection of (B) contains all those bits b of interest (interest) respectively_mkr(i) A symbol vector hypothesis equal to 1 or 0. The standard max-log-MAP approximation is used, giving bit b by_mkr(i) Soft value of (d):

the measure used therein lambda_k(c) Is that in equation 10.

What may happen is that: f_kAnd or C_mr1Or C_mr0The intersection of (but not both) is an empty set. This typically occurs when the number of branches remaining in the tree is small. The physical meaning of this case is: for bit b_mkr(i) Neither 1 nor 0 is possible. FIG. 3 shows c for symbol hypotheses₃An example of this case for the second bit (last level of the tree). In this position all reserved branches have a bit value of 1, which means that bit value 1 is most likely. In this case, the soft value Λ_mkr(i) Can be set to a positive constant. Conversely, if all the reserved branches have a bit value of 0 in a given position, then Λ_mkr(i) Can be set to a negative constant, indicating that 0 is most likely.

A detailed description is provided next to summarize the operation of the mobile terminal 120 according to the present invention. At the mobile terminal 120, the signal r₁(t) to r_L(t) are received at the receive antennas 122 (step 402 in fig. 4) and processed by the correlator 202. For a channelization code k, each received signal r₁(t) to r_L(t) are individually correlated by one of the correlators 202 at different Rake finger delays (step 404 in fig. 4). The output of the correlator 202 is the despread value y described in equation 1B_k(i) In that respect It should be observed that the channel is now represented as a matrix H that includes samples of the medium response G and the contributions from the transmit and receive filter pulse shapes (see fig. 1). Despread value y of code k_k(i) Followed byProcessed by JD-GRAKE combiner 204, which applies (step 406 in FIG. 4) a weighting matrix W to the despread values y according to equation 3_k(i) In that respect This has at least despread values y of the signal from each transmit antenna 109_k(i) Canceling out some of the effects of MAI, ISI, and other antenna interference. The output of the combiner 204 is a decision statistic z_k(i)。

Reduced search soft value generator 206 then uses a reduced tree search technique, such as the m-algorithm, to determine the representation of the received signal r₁(t) to r_L(t) soft bit values of the coded bits, and the metric λ in equation 10 is determined_k(c) In that respect The input to the reduced search soft value generator 206 is a decision statistic z_k(i) S-parameter matrix S, and search parameters that govern the search technique.

FIG. 5 is a block diagram illustrating components in one embodiment of reduced search soft value generator 206 in greater detail. As shown, the S-parameter matrix S is input to block 510, a Cholesky decomposition is performed on S, and the result L is output₈To the m algorithm processor 530. First symbol estimator 520 calculates a first symbol estimateWhich is input to the m-algorithm processor 530. Search parameters, such as the number of hypotheses to retain P, are also input to the m-algorithm processor 530, which calculates a total cumulative metric λ based on equation 10_k(c)。

Within the m-algorithm processor 530 is a hypothesis generator 532 that generates hypothesis c that is input to a branch metric calculator 534. Branch metric calculator 534 uses hypotheses c, L according to equation 11₈And first symbol estimateTo calculate the branch metric lambda_km(c) In that respect The branch metrics are output to the cumulative metric processor 536, which determines which of the P cumulative metrics to retain for the mth step according to equation 10. Typically, the cumulative metrics are collated and the best P of these values are stored andused in the subsequent (m +1) th step. After the last (mth) step, the best P cumulative metrics are output from the M-algorithm processor 530 to the soft value generator 540.

Soft value generator 540 calculates soft values according to equation 12 and outputs soft bit values to decoder 126. The decoder 126 then processes the soft bit values and outputs a bit stream 128 that represents the bit stream 102 (see fig. 1) in the base station 110.

As can be seen, the techniques of the present invention effectively reduce the number of metric calculations that need to be performed by the RC-JD-GRAKE receiver 125. To demonstrate this improved performance, complexity estimates per symbol period, per multicode, are given as follows. First, P is expressed as an m-algorithm parameter, i.e., the maximum number of branches remaining in the tree. In the level M ∈ {1, 2, Λ, M }, QL is calculated_mA branch metric, wherein L_m＝min(Q^m-1P). The min (minimum) function accounts for the fact that: during the first few levels, the number of nodes in the tree may be less than P. QL_mThe calculated branch metrics are then added to the L retained from the previous stage_mThe maximum cumulative metric. The resulting cumulative metrics are then sorted and only the largest L_m+1One is reserved for the next stage. The total number of branch metric calculations is capped at MQP for the M stages. The upper bound is assumed to be L during the first few stages_mP. Clearly, the number of metric calculations is linear with M, as compared to the exponential correlation for an exhaustive search.

Table 1 shows the actual complexity (rather than the upper limit) for various combinations of M and P for 16QAM (Q ═ 16). The complexity is expressed as the sum of the computation of Q at each stage^mThe full complexity of the metric is reduced by a percentage of the number of branch metric calculations compared to a conventional JD-GRAKE receiver. Obviously, for 1 and 2 antennas, the complexity is not reduced for the parameter settings considered. For a larger number of antennas (3 or 4), a significant reduction may be obtained. As can be seen, by retaining only 32 branches, a dramatic 98% reduction in complexity can be achieved. Moreover, compared to the traditional JD-GRAKE of full complexityThis reduction is performed at little cost to the receiver and still yields better performance than the conventional MMSE-GRAKE receiver. While the complexity savings are partially offset by the grooming operations required by the m-algorithm, the overall complexity reduction is still significant.

TABLE 1

Additional embodiments of the invention are described in detail below:

● while the preferred embodiment described above employs the m algorithm, another embodiment could instead employ List Sphere Decoding (LSD) to reduce the complexity of JD-GRAKE tree searching. For example, ball decoding is described by an article entitled "Soft-input, Soft-output lattice sphere decoder for Linear channels" published by J.Boutros et al, IEEE Global communications Conf, at pages 1583-87, 2003, and which is incorporated herein by reference. It is believed that: with LSD, a similar reduction in complexity can be achieved with minimal impact on performance. Furthermore, it should be observed that the LSD scheme is designed primarily for flat fading, while the preferred JD-GRAKE receiver 125 is designed for dispersion propagation as seen in practical systems.

● at each stage of the tree search, it is possible to predict which branch metrics λ are possible using the Cholesky decomposed structure_km(c) For the cumulative metric λ_k(c) There is little contribution. These branches λ are then_km(c) Can be discarded to reduce the collation load and thus further reduce the complexity of the tree search.

From the above, it can be easily understood that: an important aspect of being able to use the m-algorithm for the RC-JD-GRAKE tree search is to re-form the receiver metrics so that it accumulates on the transmit antennas. This is achieved using a two-step process. The first step includes finding a first symbol estimate of the transmitted symbol vector, the first symbol estimate comprising a matrix transformation of decision statistics at an output of the RC-JD-GRAKE receiver. The second step includes Cholesky decomposition of the s-parameter matrix, i.e., the product of the channel gain matrix and the RC-JD-GRAKE combining weight matrix. Because the Cholesky factor is downward triangular, each term of the cumulative metric depends only on the symbol hypotheses for the current and previous antennas. This fact allows the tree to be searched sequentially from its root and to maintain only a limited number of branches based on the best metric on each node. When the tree search ends, the soft value of each constituent bit of the transmitted symbol vector is calculated using the metric of the most likely hypothesis.

Several advantages of the present invention include the following:

● large reduction in receiver complexity. For example, for 4 transmit antennas and 16QAM modulation, where the number of branches in the tree is 65,536, the number of branches retained can be as small as 32, while still achieving performance very close to a full tree search. This is a dramatic 98% reduction in metric calculation. While this reduction is partially offset by the grooming operation required by the m-algorithm, the overall complexity reduction is still significant.

● the large reduction in complexity results in minimal degradation in performance. Using the example above, where only 32 branches are reserved, the degradation is about 0.33dB compared to a full tree search.

● the reduction in complexity of the RC-JD-GRAKE demodulator makes possible an iterative demodulation/decoding scheme. Since the complexity of the demodulation step is kept to a minimum, the overall complexity becomes manageable when several iterations are performed.

Although several embodiments of the present invention have been illustrated in the accompanying drawings and described in the foregoing detailed description, it should be understood that the invention is not limited to the embodiments disclosed, but is capable of numerous rearrangements, modifications, and substitutions without departing from the spirit of the invention as set forth and defined by the following claims.

Claims (15)

1. A method of processing a received composite communication signal (r) comprising two or more signals of interest₁(t)-_rL (t)), the method (400) of comprising the steps of:

generating (406) a despread value y as the input signal_k(i) The merge weight of the impairment correlation function of (1);

combining (406) the despread values according to the combined weights; and

jointly detecting (408) the signal of interest based on the combined despread values, and applying a detection metric λ_k(c) Symbol of (1)Line blending, wherein the joint detection step further comprises:

jointly demodulating the combined despread values to obtain soft bit values representing estimates of detected bits in the signal of interest, wherein the jointly demodulating step further comprises:

generating soft bit values by a non-exhaustive tree search technique comprising the steps of:

causal factorization L to generate s-parameter matrices_s；

Generating a first symbol estimate of a transmitted symbol vector c

Generating a reduced cumulative detection metric λ based on a causal factorization of an s-parameter matrix and a first symbol estimate of a transmitted symbol vector by using a set of cumulative detection metrics produced by an m-algorithm processor (530)_k(c) Group (d); and

soft bit values are generated based on the set of cumulative detection metrics produced by the m-algorithm processor (530).

2. The method of claim 1, further comprising providing the soft bit values to a decoder circuit (126) and decoding the soft bit values to obtain detected bits (128) for each signal of interest.

3. The method of claim 1, wherein generating, by the m-algorithm processor, the set of cumulative detection metrics comprises:

hypothesis lambda generating complete branches_km(c) Group (d);

hypothesis lambda based on complete branches_km(c) Set to calculate the detection metric lambda of a complete branch_k(c) Group (d);

based on the hypothesis lambda of the complete branch_km(c) To calculate a cumulative detection metric lambda_k(c)；

Selecting a cumulative detection metric λ_k(c) Group (d); and

outputting a cumulative detection metric λ_k(c) And (4) grouping.

4. The method of claim 1, wherein the signals of interest share the same channelization or spreading code.

5. The method of claim 1, wherein the signals of interest represent signals transmitted by different transmit antennas (109).

6. The method of claim 1, wherein the signal of interest is a subset of a larger set of signals, and wherein interference caused by those signals within the larger set that are not signals of interest is cancelled by combining weights.

7. A method of processing a received composite communication signal (r)₁(t)-r_L(t)) a receiver (124) of the composite communication signal containing two or more signals of interest, said receiver comprising:

a plurality of correlator banks (202) and a combiner (204) for generating despread values y as input signals_k(i) Combining weights of the impairment correlation functions and combining the despread values according to the combining weights; and

a reduced search soft bit value generator (206) for jointly detecting the signal of interest based on the combined despread values and for detecting a metric λ_k(c) Mixing the symbols in (1);

the reduced search soft bit value generator jointly detects the signal of interest by jointly demodulating the combined despread values to obtain soft bit values representing estimates of detected bits in the signal of interest;

the reduced search soft bit value generator generates the soft bit values using a non-exhaustive tree search technique by performing the steps of:

causal factorization L to generate s-parameter matrices_s；

Generating a first symbol estimate of a transmitted symbol vector c

Generating a reduced cumulative detection metric λ based on a causal factorization of an s-parameter matrix and a first symbol estimate of a transmitted symbol vector by using a set of cumulative detection metrics produced by an m-algorithm processor (530)_k(c) Group (d); and

soft bit values are generated based on the sets of cumulative detection metrics produced by the m-algorithm processor (530).

8. The receiver of claim 7, further comprising a decoder circuit that receives the soft bit values and decodes the soft bit values to obtain detected bits for each signal of interest.

9. The receiver of claim 7, wherein said m-algorithm processor generates the set of cumulative detection metrics by performing the steps of:

hypothesis lambda generating complete branches_km(c) Group (d);

hypothesis lambda based on complete branches_km(c) Set to calculate the detection metric lambda of a complete branch_k(c) Group (d);

based on the hypothesis lambda of the complete branch_km(c) To calculate a cumulative detection metric lambda_k(c)；

Selecting a cumulative detection metric λ_k(c) Group (d); and

outputting the cumulative detection metric λ_k(c) And (4) grouping.

10. The receiver of claim 7, wherein the signals of interest share the same channelization or spreading code.

11. The receiver of claim 7, wherein the signals of interest represent signals transmitted by different transmit antennas (109).

12. The receiver of claim 7, wherein the signal of interest is a subset of a further set of signals, and wherein interference caused by those signals within the further set that are not the signal of interest is cancelled by combining weights.

13. A wireless communication system (100), comprising:

a base station (110) for transmitting a composite communication signal;

a mobile terminal (120) for receiving and processing a composite communication signal (r) comprising two or more signals of interest by performing the following steps₁(t)-r_L(t))：

Generating (406) a despread value y as the input signal_k(i) The merge weight of the impairment correlation function of (1);

combining (406) the despread values in accordance with the combining weights; and

jointly detecting the signals of interest based on the combined despread values, and applying a detection metric λ_k(c) Wherein the joint detection step further comprises:

jointly demodulating the combined despread values to obtain soft bit values representing estimates of detected bits in the signal of interest, wherein the step of jointly demodulating further comprises:

generating soft bit values by a non-exhaustive tree search technique comprising the steps of:

causal factorization L to generate s-parameter matrices_s；

Generating a first symbol estimate of a transmitted symbol vector c

Generating a reduced cumulative detection metric λ based on a causal factorization of an s-parameter matrix and a first symbol estimate of a transmitted symbol vector by using a set of cumulative detection metrics produced by an m-algorithm processor (530)_k(c) Group (d); and

soft bit values are generated based on the set of cumulative detection metrics produced by the m-algorithm processor (530).

14. A mobile terminal (120), comprising:

a plurality of receiving antennas (122) for receiving a composite communication signal (r)₁(t)-r_L(t)); and

a receiver (124), comprising:

a plurality of correlator banks (202) and a combiner (204) for generating despread values y as input signals_k(i) And combining the despread values according to the combining weights; and

a reduced search soft bit value generator (206) for jointly detecting the signal of interest based on the combined despread values and for detecting a metric λ_k(c) Mixing the symbols in (1);

the reduced search soft bit value generator jointly detects the signal of interest by jointly demodulating the combined despread values to obtain soft bit values representing estimates of detected bits in the signal of interest;

the reduced search soft bit value generator generates the soft bit values using a non-exhaustive tree search technique by performing the steps of:

causal factorization L to generate s-parameter matrices_s；

Generating a first symbol estimate of a transmitted symbol vector c

Generating a reduced cumulative detection metric λ based on a causal factorization of an s-parameter matrix and a first symbol estimate of a transmitted symbol vector using a set of cumulative detection metrics produced by an m-algorithm processor (530)_k(c) Group (d); and

soft bit values are generated based on the sets of cumulative detection metrics produced by the m-algorithm processor (530).

15. A base station (110), comprising:

a plurality of transmit antennas (109) for transmitting a composite communication signal to a mobile terminal (120), the mobile terminal comprising:

a plurality of receiving antennas (122) for receivingA composite communication signal (r)₁(t)-r_L(t)); and

a receiver (124), comprising:

a plurality of correlator banks (202) and a combiner (204) for generating despread values y as input signals_k(i) And combining the despread values according to the combining weights; and

a reduced search soft bit value generator (206) for jointly detecting the signal of interest based on the combined despread values and for detecting a metric λ_k(c) Mixing the symbols in (1);

the reduced search soft bit value generator jointly detects the signals of interest by jointly demodulating the combined despread values to obtain soft bit values representing estimates of detected bits in the signals of interest;

the reduced search soft bit value generator generates the soft bit values using a non-exhaustive tree search technique by performing the steps of:

causal factorization L to generate s-parameter matrices_s；

Generating a first symbol estimate of a transmitted symbol vector c

Generating a reduced cumulative detection metric λ based on a causal factorization of an s-parameter matrix and a first symbol estimate of a transmitted symbol vector by using a set of cumulative detection metrics produced by an m-algorithm processor (530)_k(c) Group (d); and

soft bit values are generated based on the sets of cumulative detection metrics produced by the m-algorithm processor (530).