WO2010139008A1

WO2010139008A1 - Detection of a communication signal

Info

Publication number: WO2010139008A1
Application number: PCT/AU2010/000678
Authority: WO
Inventors: Mark Reed; Zhenning Shi; Ming Zhao
Original assignee: National Ict Australia Limited
Priority date: 2009-06-04
Filing date: 2010-06-02
Publication date: 2010-12-09
Also published as: WO2010139008A8

Abstract

The present disclosure relates generally to communication, and more specifically detection of a communication signal in a multiuser or multiple-input-multiple-output (MIMO) receiver of a communication system. A reliable signal subset (72) and a less reliable signal subset (74) of the communication signal (66) is identified using a reliability threshold test (50). Soft output detection (54) is performed on the less reliable signal subset (74) to produce a first output. Interference cancellation (52) is performed on the reliable signal subset (72), where the interference cancellation is less computationally complex than the soft output detection (54) performed on the less reliable signal subset (74). Then a soft output (56) of the communication signal (66) is determined for use by a channel decoder (42) based on the interference cancelled reliable signal subset and the first output. Aspects include methods, devices and software.

Description

DETECTION OF A COMMUNICATION SIGNAL

Technical Field The disclosure relates generally to communication, and more specifically detection of a communication signal in a multiuser or multiple-input-multiple-output (MIMO) receiver of a communication system. Aspects include a method, device and software.

Background Art Communication systems employing multiple antennas at the transmitter and receiver, known as Multiple-Input-Multiple-Output (MIMO) spatial multiplexing systems, have become popular in recent years. Similar to a multiuser system, each transmit antenna may be viewed as a user and the channel gains between one transmit antenna and multiple receive antennas can be viewed as spreading code for the corresponding user. Many detection algorithms have been proposed for MIMO systems. The maximum likelihood (ML) detection is optimal. However, the complexity of ML detection grows exponentially with the number of antennas. Hence, conventional suboptimal MIMO detection methods, e.g., zero forcing (ZF) detector, decision feedback equalizer (DFE), minimum mean square error (MMSE) detector, and sphere decoder (SD) have been proposed to perform deterministic search of the near ML candidate in the signal space with a reduced complexity.

Recently, a statistical method called Gibbs Sampler (GS) that is a particular realization of Markov Chain Monte Carlo (MCMC) simulation has been applied to MIMO detection. In MCMC, statistical inferences are developed by simulating the underlying process through Markov chain. The GS is a particular Markov chain process that searches the state space defined by the transmitted signal. The basic idea is to draw random samples of unknown transmitted signal from their conditional posterior distribution and then to calculate the marginal a posteriori distribution by averaging over the random samples. Hence, GS starts from uniformly distributed samples and walks through the transmitted signal space in a stochastic manner to look for the important/significant samples close to the transmitted signal.

H. Zhu, B. Farhang-Boroujeny, and R.-R. Chen, "On performance of sphere decoding and Markov Chain Monte Carlo detection methods," IEEE Signal Processing Lett., vol.

12, no. 10, pp. 669-672, Oct 2005 make the comparison between the MCMC detector and the SD detector. See also B. Hochwald and S. ten Brink, "Achieving near capacity on a multiple-antenna channel," /EEE Trans. Commun., vol. 51, no. 3, pp. 389-399, Mar 2003. The results in Zhu et al show that the MCMC detector outperforms the SD detector in low signal-to-noise (SNR) region with significant reduction in the complexity. However, it has been found that the MCMC detector has degraded performance as SNR increases and suffers from error floor at high SNR. B. Farhang- Boroujeny, H. Zhu, and Z. Shi, "Markov chain monte carlo algorithm for CDMA and MIMO communication systems," /EEE Trans. Signal processing, vol. 54, no. 5, pp. 1896-1909, May 2006 proposes two solutions:

(1) run parallel independent GS Markov chains; or

(2) assuming noise variance an increasing level in the GS process.

These solutions show performance improvement in medium SNRs. X. Mao, P. Amini, and B. Farhang-Boroujeny, "Markov chain monte carlo MIMO detection methods for high signal-to-noise ratio regimes," in Proc. IEEE GLOBECOM, Washington, DC, 26-

30 Nov 2007, pp. 3979-3983 investigates the source of the undesired behaviour and proposed a number of ad-hoc methods, such as using ZF or MMSΕ solution to initialize the GS, and run more iterations with increased number of samples if error occurs after Cyclic Redundancy Check (CRC).

Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is solely for the purpose of providing a context for the present disclosure. It is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each claim of this application.

Summary A first aspect provides a method of detecting a communication signal, the method comprising the step of:

(a) identifying a reliable signal subset and a less reliable signal subset of the communication signal;

(b) performing soft output detection on the less reliable signal subset to produce a first output; and (c) determining a soft output of the communication signal for use by a channel decoder based on the reliable signal subset and the first output.

It is an advantage that this method is computationally less complex than a conventional detector, such as a detector utilising Markov Chain Monte Carlo (MCMC) simulation on the entire communication signal. By using this method the signal processing computations are directed more efficiently at the unreliable signals, while the reliable signals are detected with a low complexity interference cancellation approach. By dividing the optimisation into two parts there are simultaneous gains in power efficiency and reduction in complexity. Reduction in complexity leads to reduced chip die area and cost in a realisation.

The step of identifying the reliable signal subset and the less reliable signal subset may comprise assigning bits of the communication signal that meet a reliability test to the reliable signal subset and the remaining bits to the less reliable signal subset.

The reliability test may comprise testing whether for at least one previous iteration, coded bits of the communication signal with absolute log likelihood ratio (LLR) values greater than a determined threshold.

The method may further comprise adapting the threshold based on a predefined error probability, a first order and second order statistics of received log likelihood ratio (LLR) values received from a decoder based on previous iterations of the method.

The reliability test may comprise testing whether the log likelihood ratio (LLR) values received from the decoder for the coded bits enhance the Maximum Likelihood (ML) LLR over iterations.

The less reliable signal subset may be dependent on the reliable signal subset. The method may further comprise performing soft output detection on the reliable signal subset to produce a second output. This may comprise performing interference cancellation on the communication signal by interference cancelling out the reliable signal subset. Step (b) of performing the soft detection on the less reliable signal subset may be based on the second output.

The step of performing soft output detection on the reliable signal subset (i.e. interference cancellation) may have less computational complexity than the soft detection of step (b). For example the interference cancellation step may have linear or quadratic complexity and step (b) may be non-linear in complexity, such as maximum a posteriori (MAP), sphere detection, or statistical methods.

The step of identifying the signal subsets may further comprise identifying a third signal subset that is less reliable than the reliable signal subset and more reliable than the less reliable signal subset.

The step of identifying the third signal subset may comprise assigning bits of the communication signal that do not meet the reliability test but meet a further reliability test to the third signal subset.

The method may further comprise the step of performing soft output detection on the third reliable signal subset to produce a second output representative of the third signal subset; and step (c) may further be based on the second output.

The soft detection of step (b) may comprise performing Markov Chain Monte Carlo (MCMC) simulation, such as Gibbs Sampler (GS), by drawing samples for bits of the less reliable set.

The first output may consist of coded bits representative of the less reliable signal set and may be important samples drawn during the Markov Chain Monte Carlo (MCMC) simulation. Determining the soft input of the decoder according to step (c) may comprise computing extrinsic log likelihood ratio (LLR) values for the first output and the less reliable signal subset.

The method may be performed iteratively by repeating steps (a) to (b) such that the coded bits within the reliable signal subset and the less reliable signal subset may change between at least two iterations. The number of bits in the reliable signal subset may increases between at least two iterations. In this way the complexity of the method of detection is able to reduce over time as more bits become more reliable as determined from previous iterations.

The method may be performed by a Multiple-Input-Multiple-Output (MIMO) receiver. The communication signal may be a decoded communication signal.

A further aspect is a device for detecting a communication signal having a processor operable to:

(a) identify a reliable signal subset and a less reliable signal subset of the communication signal;

(b) perform soft detection on the less reliable signal subset to produce a first output representative of the less reliable signal subset; and

(c) determine a soft output of the communication signal for use by a channel decoder based on the reliable signal subset and the first output.

The device may be realised in an application specific integrated circuit (ASIC).

The device may form part of a multiple-user receiver or a multiple-input multiple- output (MIMO) receiver. In one embodiment the receiver is a 3G DS-CDMA wireless multi-user receiver used for base station signal detection.

Yet a further aspect, is software, that when installed on a receiver is able to perform the method described above.

Yet a further aspect provides a method of detecting a communication signal, the method comprising the step of: identifying a reliable signal subset and a less reliable signal subset of the communication signal; performing soft output detection on the reliable signal subset using a first technique; and performing soft output detection on the less reliable signal subset using a second technique, wherein the first technique is less computationally complex than the first technique.

Also provided is software and/or device for performing the method directly above.

For the third signal subset as described above, a third technique may be performed for soft output detection, wherein the third technique is more complex than the first technique and less complex than the second technique.

Advantages and/or features of at least one example includes: the ability to use smaller less expensive chips that consume less power dynamic and static scheduling of receiver tasks (detection/decoding) that together with an iterative receive approach has improvements in communication range extension, terminal battery life extension, communication link throughput improvement for uplink improving the receiver sensitivity at high signal to noise ratios

Brief Description of the Drawings

An example will now be described with reference to the accompanying drawings in which:

Fig. 1 is a schematic drawing of a multiple-input multiple-output (MIMO) spatial multiplexing transmitter with iterative receiver;

Fig. 2(a) is a flow chart of the method of the example;

Fig. 2(b) is a schematic diagram of this example method as performed by a receiver;

Fig. 3 is a graph showing Bit Error Rate (BER) performance in the iterative receivers with the conventional Markov Chain Monte Carlo (MCMC) detector, and with the detector of this example in a 4 x 4 MIMO spatial multiplexing system with QPSK and 16QAM modulation; Fig. 4 is a graph showing the complexity reduction over SNR with QPSK and

16QAM modulation; Fig. 5 is a graph showing the complexity reduction over iterations with QPSK modulation;

Fig. 6 is a graph showing the complexity reduction over iterations with 16QAM modulation; and Fig. 7 is a table showing the total complexity reductions.

Best Modes

Notations used in this specification are as follows. Matrices and vectors are denoted by symbols in bold face and (•) ,(•) and (•) represent complex conjugate, transpose and Hermitian transpose. £{•} denotes the statistical expectation. [X]₍ indicates the

(i,j)' elements of a matrix x and similarly, [x₍] indicates the element i in a vector x . Finally, {x) represents the sequence x .

An example using MIMO spatial multiplexing system and iterative detection and decoding will now be described with reference to Fig. 1. In this example, computational power is re-allocated to ensure all received signals achieve similar performance (e.g. frame error rate). There are N₁. 20 transmitting antennas and N_R 22 receiving antennas. The information bits |b] 24 are first encoded 26 and interleaved 28 to coded bits sequences {d} 30. These coded bits are mapped into a sequence of

_r f M - ary complex symbols { x} represented by a vector ^{x ~} [_-*Ό » %ι > " ^'» ^XN_Γ-\ J 32.

Assuming that each transmitting and receiving antenna link undergoes independent flat fading, the system model can be expressed as:

y = hx+n, (1)

where y 38 is the received signal arranged as N_R x \ vector

y ⁼ I >O » ^i 9 ^{* * *} > -V_A^_7J -_I J • h is N_R x N_T channel matrix, n is N_R ^χ \ additive white Gaussian noise (AWGN) vector with co variance σ²l_N . It is convenient to reformulate the system model from a complex value domain to a real value domain by defining the 2N_R x 1 vector Y 38, 2N_R x 1 2N_T x 1 vector X 36, 2N_R x 1 vector N as: Y = [*(y)^r3(y)^r]^r

X = [*(x)^r3(x)^r]^r N = [*(n)^r 3(n)^r]^r

and 2N_R x 2N₇- matrix H as:

*(h) -3(h)^"

H = 3(h) «(h)

where 5l(-) and 3(-) denote the real and imaginary parts of the argument. Then the system can be expressed in as:

Y = HX + N . (2)

In the conventional non-iterative receiver, the MIMO detector tries to maximize the likelihood of the transmitted signal, that is, to minimize the Euclidean Distance to the received signal given that the channel is known, which is given by:

X = arg max p ( Y x) = arg min Il Y - H X|| (4)

XeA XeA

where β is the signal set of dimension 2N_T . On the other hand, in the iterative decoding, the MIMO detector aim to maximize a posteriori (MAP) probability of transmitted signal sequence, which is given by:

X ₍₅₎

In (5), p (X) 40 is considered as the a priori information fed back from channel decoder 42. In the first iteration, the a priori information 40 is not available, the ML metric (4) is employed in the MIMO detector 44. From the second iteration onwards, MAP detection metric (5) is employed. In each iteration, the MIMO detector 44 selects points I Xj from the completed signal constellation set to form a list U to compute the extrinsic log likelihood ratio (LLR) of the coded bit as:

where d_k (x) is the &'^Λ coded bit in sequence {d} representing the MCMC GS decoder enumerated transmitted symbol vector X . Ul and U^~ denote the subset of U for which d_k (x \ is +1 and -1 respectively. d__k is obtained from sequence d by removing the k'^h coded bit. ^i -_k is the extrinsic LLR of sequence d_^ from the channel decoder 42. After MIMO detection, the sequence of extrinsic LLR j /L₁ ⁶J of coded bits is deinterleaved 46 and passed on to the channel decoder 42 to complete one iteration.

At high SNR, a high portion of the coded bits can be detected in the first few iterations. The LLRs of these bits have large value such that some of the transition probabilities in the underlying Markov chain may become very small. As a result, the Markov chain may be divided into a number of nearly disjoint chains because the high probability associated with large LLR value prevent the transitions among the disjoint chains. Hence, the GS has less chance to visit sufficient points. This phenomenon is undesirable for the stochastic approach of MCMC, which requires a large number of samples in order to cover the whole state space defined by the transmitted signal.

In the reduced-state-space MCMC (RSS-MCMC) detector of this example, of the interference from the bits with large LLR values is removed. Then we draw random samples only for unreliable bits associated with small LLR values in the system with less interference.

This method will now be described in more detail with reference to the flow chart of

Fig. 2(a) and the schematic diagram of Fig. 2(b). This method can be performed by a chip installed in the MIMO detector component of the receiver. The MIMO detector having a processor and associated memory to execute the instruction set of software on the chip.

We divide the transmitted signal sequence X 36 into two sets, namely the reliable signal subset X_R and a less reliable (unreliable) signal subset X₁₁ 50. The reliable signal subset X_β contains the bits that satisfy reliability constraint (discussed below), and are considered as correctly detected in the previous iteration. Similarly, the unreliable signal subset X_y contains the bits that not satisfy the reliability constraint and are considered as unknown. Therefore, the system can be expressed in as:

Y = H -diag(e) - X + H(l_2;Vr - diag(e))X + N

= HX_C/ +HX_Λ + N,

where X_j, = diag (e) X, X_Λ = \ 1_2NT - diag(e)j X, diag (-) is the diagonal function, and e is the 2N_τ χ l vector which contains the position of unreliable bits. Here, the modified MCMC GS detector treats the reliable signal subset X_βas the interference and that can be deterministically removed.

We first propose the reliability constraints 50(a) and 50(b) to construct the reliable signal set X_β 72 of the communication signal 66. Intuitively, the LLRs 40 fed back from channel decoder 42 (a priori information) are the direct measure of the reliability of the coded bits (shown here as a soft signal distribution). Unfortunately, the analytical treatment of the soft output of channel decoder 42 is difficult. Nevertheless, we make use of the Gaussian consistent assumption based on the empirical observations from simulations in S. ten Brink, "Convergence behaviour of iteratively decoded parallel concatenated codes," IEEE Trans. Commun., vol. 49, no. 10, pp. 1727-1737, Oct 2001. The Gaussian consistent assumption states as follows: 1) For large interleavers the a priori information remain fairly uncorrelated from channel observation over many iterations. 2) The probability density function of the extrinsic information of channel decoder 42 {a priori information of detector respectively) approach Gaussian-like distribution with increasing number of iterations. Gaussian consistent assumption suggests that the a priori 40 information as input of MIMO detector 44 can be modelled by applying an independent Gaussian random variable η with variance G_η and zero mean in conjunction with the known coded bit d_k G {—1,1} as follows: A,^e (d_k) = μ-d_k +η, (7)

where the mean value μ satisfies the relation [9]

_ η

M = (8)

To compute μ and σ , we can compute the 2 nⁿd order statistics of LLRs 40, which is given by:

The solution of μ can be obtained by taking the positive root of equation (9):

if E { K J I J is significant larger than one over iterations. And the solution of

.2 is obtained as:

Therefore, the conditional distribution of the a priori 40 information given that d_k = 1 and d_k = —\ is Gaussian with

respectively. Then we can obtain a threshold p from the error probability of the a priori 40 information as follows:

Case d. = — 1 ; P{error\d_k

where Q (•) is the Q-function. Then we have

where Q (") is the inverse of Q-function. Hence, we can obtain the threshold p as: p = σ_η • Q-' (P(error \d_k = - 1)) - μ . (14)

Case d,, = \ :

P (error d, = l

Then we have

Hence, we can obtain the threshold p as:

Combine the cases of d_k = -1 and d_k = \, d_k = \ . p as first reliability constraint 50(a), which can be obtain as:

P = σ_η ■ Q ^x (P (error \d_k = ±\))-μ (18)

It can be seen that this threshold p to be applied by the threshold detector 68 is adaptive to the predefined error probability, and first order and second order statistics of the a priori information over iterations. The coded bits with absolute LLR values greater than this threshold will be considered as reliable.

Furthermore, LLRs with large values result in disjoint Markov chains with less chance to visit the significant samples. The insufficient number of samples may generate ill- conditioned LLRs, which has large value but sign flipped. This can be explained by equation (6). We can rewrite the LLR in (6) as the summation of the ML LLR and the a priori LLR:

4' K(X)) = 4^ (^«/,(*)) ⁺ 4^ (*-(*)), 09)

where

(20)

and

The ML LLR is the measure of the Euclidean Distance between the enumerated coded bits and the actual transmitted coded bits, and the a priori LLR is the measure of likelihood of the coded bit of interest as seen by other interfering bits. The computation of LLR in (6) relies on ML LLR initially, then is joint determined by ML LLR and the a priori LLR over iterations, and finally dominated by the a priori LLR. If we purely rely on the threshold as unique reliability constraint, once LLR error occurs in the reliable signal set, the large but sign flipped LLR value may dominate the computation of the a priori LLR for current bit of interest, even though the ML LLR has the correct sign. This phenomenon is especially undesirable in MCMC, because the later drawn samples are influenced by the earlier drawn samples. The receiver suffers from error propagation as in the hard decision DFE. Therefore, we propose the second reliability constraint 50(b) that the sign of the a priori LLR should be the same as the ML LLR as:

Λ:«4^J*(^X)H_"K(^X))>°- rø

This second reliability constraint ensures that only the coded bits with the a priori LLR enhancing the ML LLR over iterations are considered to be reliable. In practical implementation, the ML LLR from first iteration can be stored and doesn't have to be computed in every iteration. The a priori LLR in each iteration can be obtained by subtracting the ML LLR from first iteration in equation (6).

In the RSS-MCMC detector of this example, the interference from the bits in the reliable signal subset X_Λ =

,d_{r 2} - - -J 72 as constructed by the two reliability constraints is first cancelled 52 from received signal. Extended from equation (3), the output of interference canceller can be obtain as: Y = Y -HX R

The residual interference and noise have zero mean and covariance matrix C I 2N_R

The next step of the RSS-MCMC detector is to draw random samples 54 for the coded bits in the unreliable signal subset Xy =

IA. Given Y' and the a priori information A^ , the a posterior probability is evaluated as follows:

The GS initializes the samples in X_y with equal probability of 0.5, and proceeds with drawing 54(a) one sample in X₁₇ at a time. The procedure is summarized as follows:

Initialize X_y randomly with equal probability

For H = I to N

draw«_{u l} from-' ^M,! j"_u,2 >"κ,3 >^•••> * ^J

draw<»from/'(<_t|<₁ ⁾-,^7i!,...,Y'Λ')

It can be seen that in the n^l loop, the k^lh sample is drawn based on the probability P\d_uk = ±1 Y'jXj __A,/^^e J that partially depends on the (k -l) samples drawn in the n'^h loop, and partially depends on the rest samples drawn in the (n-l)' loop. That is ^Λι/,-* ^~

' • • • ( • This probability is obtained by first computing the a posterior LLR:

The computation of P Y' γ(") ^v

[/,-* > ^βM,yfc ^~ — ^A I follows the equation (23) as follows:

= «)

where Λ^" is a constant. Once ^₁ ^"_U>^ J is obtained, we have

See B. Fahang-Boroujeny, H. Zhu, and Z. Shi, "Markov chain monte carlo algorithm for CDMA and MIMO communication systems." /EEE Trans. Signal Processing, vol. 54, no. 5, pp. 1896-1909, May 2006. The important samples are chosen from a uniform distribution with probability shown in (27). The reasons to use uniform distribution to draw random samples are

1) theoretically simplify the Markov chain integral in (24) such that the total number of samples are minimum and contributes significantly to the integral in (24);

2) mathematically obtain a better approximation of infinite Markov chain integral; and

3) practically easy way in hardware implementation through shift register realization. After going through the above procedure, important samples are drawn 54(b) for the bits in the unreliable set 74 and they form the list U together with the bits in the reliable set 72. The final step is to compute 56 soft input being the extrinsic LLR 56 for each bit by applying equation (6) based on list U . This soft input 56 is then provided to the decoder 42.

The multiplexer 76 recombines the soft values from both the low complexity interference canceller 52 and the advanced signal detector 54 providing a single bit stream for the decoder 42. The decoder 42 outputs 80 are shown as a signal distribution 56. The change in the distribution provides a significantly improved distinction between the data symbols of +1 and -1. This is due to the receiver scheme and the ability to overcome the effects of interference and noise.

It is important to note that this example does not concern the splitting of soft information but the signal itself. Statistics of the soft information (LLR distribution) for each value to determine if each value is a reliable r unreliable, thereby creating two signal sets.

The most significant difference between the RSS-MCMC of this example and conventional MCMC detectors is that the significant samples can be drawn in an interference reduced system rather than a fully loaded system. In the RSS-MCMC detector of this example, interference from reliable bits are removed in (23), which results in a MIMO system with less interference, where the GS performance can be improved. If all the bits are reliable, the example operates as an interference canceler. On the other hand, if all the bits are unreliable, the RSS-MCMC of this example is the same as conventional MCMC detector. Otherwise, if the bits are partially reliable, the example is a hybrid conventional MCMC detector and interference canceler.

As the detector is an iterative detector in stable conditions the reliable bit subset will increase in size (and in turn the unreliable subset will decrease). This also causes the example to reduce in computational complexity over time.

Simulation results of the example will now be described. We consider a 4x4 MIMO spatial multiplexing system and compare the Bit Error Rate (BER) and complexity reduction for the iterative receivers with the conventional MCMC detector and the iterative receiver with the RSS-MCMC detector. We run 4 parallel Markov chains, and within a Markov chain, there are 5 samples for each coded bit. The complexity reduction is measured by the number of the drawn samples in the RSS-MCMC detector over the total number of samples drawn in the conventional MCMC detector. The channel model for each transmit and receive antenna are independent flat Rayleigh fading channel. A rate- 1/2 (171, 133)₈ convolutional code is used for channel coding. The modulation includes QPSK and 16QAM. We refer to the conventional MCMC GS detector as "Conventional MCMC" and the RSS-MCMC detector proposed in this specification as "RSS-MCMC". "itr 2", "itr 3", and "itr 4" denote the 2^nd, 3^rd and 4^th iteration.

Fig. 3 shows the BER performance for the iterative receivers with the conventional MCMC GS detector and with the RSS-MCMC detector over 4 iterations. It can be seen that in QPSK, the RSS-MCMC detector has slightly better performance than the conventional MCMC detector, although it is not noticeable in the 4^th iteration. In the case of 16QAM, the RSS-MCMC detector improves the performance over the conventional MCMC detector. This is because the RSS-MCMC detector performs detection on the undetermined bits in a MIMO system with less interference after cancelling the interference from the reliable bits. At high SNR, the conventional MCMC detector shows degraded performance in the 3^rd and 4^th iteration. Nevertheless, the RSS-MCMC does not show the error floor.

Fig. 4 shows the complexity reduction per SNR for the RSS-MCMC detector over the conventional MCMC detector. In both QPSK and 16QAM modulation, the complexity reduction increases as the SNR increases. There is no complexity reduction in the first iteration as the proposed and conventional MCMC detectors are the same. In the 2^nd iteration, the complexity reduction reaches 30% for QPSK and 8% for 16QAM at 6dB. As iteration goes, more bits satisfy the reliability constraints. This can be observed in the 4th iteration that more that 50% and 30% of the bits in QPSK and 16QAM respectively at 6dB are considered as reliable bits and canceled as interference in the RSS-MCMC detector rather than continuously drawn samples in the conventional MCMC detector.

Fig. 5 and Fig. 6 show the complexity reduction for the RSS-MCMC detector over iterations for QPSK and 16QAM modulation respectively. In both modulation schemes, there is a significant increase of the complexity reduction from the 2^nd and the 3^th iteration, while the complexity reduction saturates in 4^th iteration. This observation indicates the evolution of the reliability of LLRs over iterations. It also indicates that after 4 iterations, the LLRs converge to equilibrium and the further complexity reduction is marginal.

Finally, Table I of Fig. 7 summarizes the total complexity reductions. The total complexity reduction is the summation of complexity in each iteration including the first iteration. Generally, the complexity reduction in QPSK is more than that in 16QAM. It can be seen that 35% computation power can be saved in QPSK at 6dB, while 26% computation power can be saved in 16QAM at 12dB.

It will be appreciated by persons skilled in the art that numerous variations and/or modifications may be made to the example as described and/or shown in the specific embodiments without departing from the scope as broadly described.

It should be understood that the techniques of the present disclosure might be implemented using a variety of technologies. For example, the methods described herein may be implemented by a series of computer executable instructions residing on a suitable computer readable medium. Suitable computer readable media may include volatile (e.g. RAM) and/or non-volatile (e.g. ROM, disk) memory, carrier waves and transmission media. Exemplary carrier waves may take the form of electrical, electromagnetic or optical signals conveying digital data streams along a local network or a publicly accessible network.

It should also be understood that, unless specifically stated otherwise as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as "identifying" or "processing" or "computing" or "calculating", "optimizing" or "determining" or "displaying" or "performing" or the like, refer to the action and processes of a computer system, or similar electronic computing device, that processes and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.

Claims

CLAIMS:

1. A method of detecting a communication signal, the method comprising the step of:

(b) performing soft output detection on the less reliable signal subset to produce a first output; and

(c) determining a soft output of the communication signal for use by a channel decoder based on the reliable signal subset and the first output.

2. The method of claim 1, wherein the step of identifying the reliable signal subset and the less reliable signal subset comprises assigning bits of the communication signal that meet a reliability test to the reliable signal subset and the remaining bits of the communication signal to the less reliable signal subset.

3. The method of claim 2, wherein the reliability test comprises testing whether for at least one previous iteration coded bits of the communication signal have absolute log likelihood ratio (LLR) values greater than a determined threshold.

4. The method of claim 3, the method further comprises adapting the threshold based on a predefined error probability, a first order and second order statistics of received log likelihood ratio (LLR) values received based on a decoder from previous iterations of the method.

5. The method of claim 2, 3 or 4, wherein the reliability test comprises testing whether the log likelihood ratio (LLR) values received from the decoder for the coded bits enhance the Maximum Likelihood (ML) LLR over iterations.

6. The method of any one of the preceding claims, wherein the method further comprises performing soft output detection on the reliable signal subset to produce a second output.

7. The method of claim 6, wherein the soft output detection on the reliable signal subset comprises performing interference cancellation on the communication signal by interference cancelling out the reliable signal subset.

8. The method of claim 6 and 7, wherein step (b) of performing the soft detection on the less reliable signal subset is based on the second output.

9. The method according to any one of claims 6, 7 or 8 when limited by 6, wherein the step of performing output detection on the reliable signal subset has less computational complexity than the soft detection of step (b).

10. The method according to any one of the preceding claims, wherein the step (a) further comprises identifying a third signal subset that is less reliable than the reliable signal subset and more reliable than the less reliable signal subset.

11. The method according to claim 10, wherein the step of identifying the third signal subset comprises assigning bits of the communication signal that do not meet the reliability test but meet a further reliability test to the third signal subset.

12. The method of claim 10 or 11, wherein the method further comprises the step of performing soft output detection on the third reliable signal subset to produce a second output representative of the third signal subset; and step (c) is further based on the second output.

13. The method according to any one of the preceding claims, wherein determining the soft input of the decoder according to step (c) comprises computing extrinsic log likelihood ratio (LLR) values for the first output and the less reliable signal subset.

14. The method according to any one of the preceding claims, wherein the method is performed iteratively by repeating steps (a) to (b) such that coded bits within the reliable signal subset and the less reliable signal subset change between at least two iterations.

15. A device for detecting a communication signal having a processor operable to:

16. The device of claim 15 is an application specific integrated circuit (ASIC).

17. The device of claim 15 or 16 is part of a multiple-user receiver or a multiple- input multiple-output (MIMO) receiver.

18. Software, being computer readable instructions stored on computer readable media, that when installed on a receiver is able to perform the method according to any one of claims 1 to 14.

19. A method of detecting a communication signal, the method comprising the step of: identifying a reliable signal subset and a less reliable signal subset of the communication signal; performing soft output detection on the reliable signal subset using a first technique; and performing soft output detection on the less reliable signal subset using a second technique, wherein the first technique is less computationally complex than the first technique.