CN1200527C - Simplified maximum likelihood multi-user detecting method based on sensitive bit - Google Patents

Abstract

The present invention relates to a simplified maximum likelihood multi-user detection method based on sensitive bits, which belongs to the wireless communication field. The present invention is characterized in that in the method, sensitive bits are distinguished in multi-user signals estimated by the traditional non-optimal multi-user detection arithmetic with low complexity, and mistaken bits are corrected by the maximal likelihood detection (MLD) method in small search subsets corresponding to the sensitive bits. The method decreases mistaken bit rates, and simultaneously greatly simplifies the complexity of the maximal likelihood multi-user detection arithmetic to make the arithmetic become possible in practical application. The method can approach to the performance of the MLD arithmetic by increasing the number of the sensitive bits or by an iterative method. The present invention can be used in CDMA, SDM and SDMA systems.

Description

Simplification maximum likelihood Multiuser Detection method based on sensitive bit

Technical field

Wireless communication system: comprise that space division multiple access inserts (Space Division Multiple Access, SDMA), code division multiple access inserts (Code Division Multiple Access, CDMA) and space division multiplexing (Space Division Multiplexing, system such as SDM).

Background technology

Along with the develop rapidly of wireless communications industry, the release of wireless traffics such as especially video conference, broadband the Internet access and wide area network access.Wireless communication system faces the pressure of bigger custom system capacity requirement.Spread spectrum and intelligent antenna technology are the methods that effectively overcomes the limited spectral bandwidth of channel.At present, code division multiple access access technology (CDMA) has been applied to various Digital Cellular Systems and PCS Personal Communications System.Intelligent antenna technology can improve the performance of system greatly under the situation that does not increase system spectral resources, the application that space division multiple access inserts (SDMA), space division multiplexing (SDM) system more and more receives publicity.The degree of freedom of utilizing spread spectrum and antenna system to increase that multiuser detection can be imitated very much, and improve the capacity of system greatly.Than suboptimal multiuser detection algorithm (decorrelation, least mean-square error detection and interference eliminated etc.), traditional Maximum Likelihood Detection (MLD) algorithm has optimum BER (Bit Error Rate) performance.Yet the algorithm complex of traditional MLD algorithm is exponential relationship with detecting number of users and order of modulation, is difficult in the practical application realizing.We can keep very little BER performance loss by the sensitive bit multiuser detection algorithm of invention under the situation of the algorithm complex of simplifying traditional multiuser detection algorithm greatly, and its algorithm complex follows the algorithm complex of suboptimal multiuser detection algorithm with the order of magnitude.

Summary of the invention

The present invention has purpose to be to provide little and algorithm complex of a kind of bit error rate and suboptimal multiuser detection algorithm to belong to the simplification maximum likelihood Multiuser Detection method based on sensitive bit with the order of magnitude.

Before the sensitive bit multiuser detection algorithm of introducing our invention, be necessary to introduce simply traditional MLD algorithm.Optimum MLD algorithm is the signal that estimates all users with the method for maximum likelihood (ML) simultaneously.Consider simple system signal model

$\stackrel{\→}{R}=H\·\stackrel{\→}{S}+\stackrel{\→}{N}.----\left(1\right)$

Wherein,

H and

Be respectively received signal vector, matrix channel and the white Gauss noise vector in SDM and SDMA system; In addition, in the direct sequence cdma system, represent matched filtering output signal vector, correlation matrix and matched filtering output noise respectively.

Be the symbolic vector of multi-user's transmission, its optimal estimation is separated, Can obtain by search

Suppose that order of modulation is that Q and terminal number are N.For searching optimal solution, search completely needs to calculate all Q ^NIndividual possible symbolic vector, thus in real system when Q and N become big, this algorithm can not be realized too greatly because of complexity.For this reason, we proposed simply, effective sensitive bit multiuser detection algorithm

Basic thought based on the sensitive bit multiuser detection algorithm is to realize the branch two-stage: first starting stage, estimate multiple user signals with the non-optimum multiuser detection algorithm such as least mean-square error LMMSE (the Linear Minimum Mean Square Error) algorithm of traditional low complex degree with orderly iteration interference eliminated SIC (Successive Interference Cancellation) algorithm.Second stage, we tell some particular bit, are referred to as " sensitive bit ", and they are the bits in very possible misjudgment of phase I.Then, we correct make mistakes bit with maximum likelihood MLD algorithm in corresponding littler search with sensitive bit in collection.Particularly, be described below based on the sensitive bit multiuser detection algorithm:

The first step: initialization

Be to realize " sensitive bit " algorithm, estimate that as far as possible correctly multi-user's bit is is important to the effectiveness that improves the after-stage algorithm.Starting stage can be utilized simple detection algorithm such as LMMSE and orderly SIC.Then obtain the symbolic vector of N user's transmission This symbolic vector can be mapped to the vector of 2N bit

Wherein, i=1 ..., N and j=I, Q, and the sine and the cosine component of I and Q difference conventional letter.Second step: sensitive bit algorithm

According to formula (1),

Conditional Distribution Density Functions can be expressed as

$P\left(\stackrel{\→}{R}\right|\stackrel{\→}{S})=\frac{1}{{\left(2\mathrm{\π}\right)}^{M/2}{\left|\right|{\mathrm{\σ}}^{2}I\left|\right|}^{1/2}}\exp \{-\frac{{(\stackrel{\→}{R}-H\·\stackrel{\→}{S})}^H\·(\stackrel{\→}{R}-H\·\stackrel{\→}{S})}{{\mathrm{\σ}}^{2}I}\}.----\left(3\right)$

Maximization likelihood probability (3) then is equivalent to and minimizes yardstick

$\mathrm{\Ψ}\left(\stackrel{\→}{S}\right)={(\stackrel{\→}{R}-H\·\stackrel{\→}{S})}^H(\stackrel{\→}{R}-H\·\stackrel{\→}{S}).----\left(4\right)$

Definition

Wherein,

It is the estimation of starting stage multi-user transmission symbol vector.Obviously,

It or not the optimal detection value.Yet, because In only have a few bits mistake and estimate, thereby Ψ ₀Can be very near smallest dimension $\underset{\stackrel{\→}{s}}{\arg \min }\mathrm{\Ψ}\left(\stackrel{\→}{S}\right).$ In fact, we have confirmed that this situation exists really in emulation experiment.Consider bit vectors

And hypothesis

In have and only have ij element bit mistake estimate (i=1 ..., N, j=I, Q).Then, we adjust

In ij element bit, the polarity of the ij element bit that promptly reverses, and to adjust the new bit vectors in back The corresponding symbolic vector of adjusting,

Also obtain thereupon.And satisfy

Wherein,

And have

Note, at this moment It is multi-user's symbolic vector of entirely true estimation.In appendix .1, we have proved existence

Therefore, it means that working as us corrects the initial estimation bit vectors

In an error bit after and obtain new bit vectors and estimate

On the expectation meaning, Will be less than Ψ ₀Therefore, such bit is most possibly to be error detection, if i.e. our bit vectors of reversing

In this bit polarity (as, 0 → 1 or 1 → 0), and resulting new symbolic vector Tolerance less than Ψ ₀

In general, the error bit number in the symbolic vector of initial estimation is very limited.Such as, if bit error rate is 10 in the initial estimation ^-1, mean then on the average meaning that it is wrongly to estimate that a bit is only arranged in ten bits.Therefore, generally speaking, the number of sensitive bit can be very not big.In addition, new symbolic vector Tolerance More little, then the bit of its corresponding reversed polarity just may be estimated in the starting stage mistake more.Consider In all 2N bit and obtain corresponding institute might the new symbolic vector of 2N Then, we are with the new symbolic vector of all 2N

Tolerance By descending, then we define corresponding to all tolerance

2 of middle minimum ^f(bit of reversed polarity is " sensitive bit " in the new vector of f≤2N), because these bits are that most probable is made mistakes in all bits.Provide this f sensitive bit, we can suppose that other remaining bit all is correctly to estimate.Therefore, for correcting bit wrong in f the sensitive bit, adopt the MLD algorithm can only consider 2 of a corresponding f sensitive bit this moment ^fIndividual possible symbolic vector combination.As seen we the MLD algorithm of optimum from 4 ^NIndividual may symbolic vector collection narrow down to very little by 2 ^fSearch in the subclass of individual symbolic vector element.Thereby 2 ^fThe MLD algorithm of searching in the subclass of individual symbolic vector element can be used to do the last suboptimal maximal possibility estimation of sensitive bit algorithm.This estimation may be defined as

The invention is characterized in: it is to tell sensitive bit by a kind of of computer in the above-mentioned multiple user signals that estimates, the method of in corresponding to the littler search subset of sensitive bit, correcting the bit that makes mistake again with the MLD algorithm, it contains successively and has the following steps:

(1). with traditional linear least mean-square difference is that LMMSE and orderly successive interference eliminated are that the SIC algorithm is realized initial Multiuser Detection;

(2). preset initial multiple user signals vector;

(3). in subclass, use the MLD algorithm corresponding to sensitive bit:

(3.1) definition f is maximum sensitive bit number;

(3.2) symbolic vector of transmitting N the user that step (1) is obtained

Be mapped to the vector of 2N bit

After, wherein: i=1,2...N, j=I, Q, J, Q is the sine and the cosine component of conventional letter respectively; By counter-rotating

In the polarity of ij bit obtain new symbolic vector It is corresponding to initial estimation There is and only have a bit to change polarity;

(3.3) all bits that reverse obtain 2N new symbolic vector Calculate all new symbolic vectors

2N

(i=1,2 ... ..N, j=I, Q);

(3.4) 2N tolerance that newly obtains

Middle search 2 ^fIndividual minimum

Define corresponding this 2 ^fF in the individual vector counter-rotating bit is " sensitive bit ";

(3.5) it is constant to fix other 2N-f bit, corresponding to 2 of f sensitive bit ^fMake Maximum Likelihood Detection in the subclass of individual symbolic vector;

(4). judge that iteration finishes not?:

If: do not finish, then get back to step (2);

If: finish, then export the Multiuser Detection signal.

Operation instruction: the sensitive bit algorithm has greatly reduced the bit error rate of starting stage receiver, and algorithm complex is compared more simplification with optimum MLD algorithm complex.And, when total bit number of all users can improve performance with iteration sensitive bit algorithm repeatedly during much larger than the sensitive bit number.

Description of drawings

Fig. 1: the program circuit of sensitive bit algorithm is gripped figure.

Fig. 2: based on the sensitive bit algorithm performance (4 users, 6 reception antennas) of LMMSE initial estimation.

The sensitive bit algorithm, f=5

The sensitive bit algorithm, f=4

The sensitive bit algorithm, f=3

The tradition least-mean-square error algorithm

The V-BLAST algorithm

Maximum likelihood algorithm

Fig. 3: based on the sensitive bit algorithm performance (4 users, 6 reception antennas) of orderly SIC initial estimation.

The sensitive bit algorithm, f=5

The sensitive bit algorithm, f=4

The sensitive bit algorithm, f=3

The tradition least-mean-square error algorithm

The V-BLAST algorithm

Fig. 4: based on the sensitive bit algorithm performance (8 users, 12 reception antennas) of LMMSE initial estimation.

The sensitive bit algorithm, f=5

The sensitive bit algorithm, f=4

The sensitive bit algorithm, f=3

The tradition least-mean-square error algorithm

The V-BLAST algorithm

Fig. 5: based on the sensitive bit algorithm performance (8 users, 12 reception antennas) of SIC initial estimation.

The sensitive bit algorithm, f=5

The sensitive bit algorithm, f=4

The sensitive bit algorithm, f=3

The sensitive bit algorithm, f=2 (iteration)

The tradition least-mean-square error algorithm

The V-BLAST algorithm

Concrete execution mode

The sensitive bit multiuser detection algorithm of implementing contains following steps successively:

● the first step: realize initial Multiuser Detection with traditional LMMSE algorithm and orderly SIC multiuser detection algorithm.

● second step: sensitive bit algorithm

Definition f is maximum sensitive bit number

1. by counter-rotating In the polarity of ij bit obtain new symbolic vector

Corresponding to initial estimation There is and only have a bit to change polarity.All bits and calculate the tolerance of all new symbolic vectors reverse (i=1 ..., N; J=I, Q).

2. 2N tolerance that newly arrives Middle search 2 ^fIndividual minimum

And define corresponding this 2 ^fF in the individual vector counter-rotating bit is " sensitive bit "

3. it is constant to fix other 2N-f bit, corresponding to 2 of f sensitive bit ^fDo Maximum Likelihood Detection in the subclass of individual symbolic vector

In addition, should be noted that the computation complexity of sensitive bit multiuser detection algorithm depends primarily on the number of sensitive bit.If there is too many user to insert simultaneously simultaneously, such as cdma system, and the sensitive bit number f that sets is not enough to correct simultaneously the error bit of initial examination and measurement.Therefore, we propose iteration type sensitive bit multiuser detection algorithm and avoid increasing the sensitive bit number.Be about to the initial estimation of last sensitive bit testing result as sensitive bit detection next time.In real system, coordinate the iterative operation number of sensitive bit number and sensitive bit algorithm according to actual conditions.Realization block diagram such as Fig. 1 of sensitive bit algorithm show.

Comparison shows that by Computer Simulation and simulation result the sensitive bit multiuser detection algorithm that we invent has very strong application potential.In the experiment simulation, we only consider the SDMA system, the individual antenna of base station apparatus M=6 (or 12) wherein, and the individual user of N=4 (or 8) that single antenna arranged is in indoor random distribution.Definition sensitive bit number f=3,4 and 5.Definition space average received signal to noise ratio (snr) is $\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{SNR}}_i,$ SNR wherein _iBe all N users' the received signal energy and the ratio of noise power on i antenna.In order to compare, we have listed traditional LMMSE multiuser receiver and have been similar to the BER performance of the V-BLAST receiver of orderly SIC algorithm.

At first, we consider Fig. 2 and Fig. 3, and the BER performance that they have provided the sensitive bit detection algorithm of our invention and traditional LMMSE algorithm, V-BLAST receiver and optimum MLD algorithm relatively.As can be seen, the sensitive bit algorithm has greatly improved the BER performance of starting stage LMMSE or V-BLAST receiver.For example, when sensitive bit is counted f=5, compare, can obtain the BER performance gain of 3dB and 1dB based on the MLD algorithm of sensitive bit respectively with traditional LMMSE or V-BLAST receiver.In addition, along with the increase of sensitive bit number, the performance of this algorithm will be further near optimum MLD algorithm.From Fig. 2 as seen, the performance difference of sensitive bit algorithm and optimum MLD algorithm is 0.3dB in all average received SNR scopes.Though sensitive bit algorithm simulation performance when initially adopting LMMSE with the V-BLAST receiver respectively is similar, in high SNR district, both simulation results are variant.In Fig. 2, initial estimation has intersection based on the sensitive bit algorithm and the orderly SIC algorithm of LMMSE algorithm at high SNR place.This is because the performance of LMMSE algorithm far is worse than orderly SIC algorithm.

In Fig. 4 and Fig. 5, we have provided and Fig. 2 and the similar simulation result of Fig. 3.Different is that base station reception antenna number and number of users at this moment is respectively M=12 and N=8.Because optimum MLD algorithm is too complicated and be omitted under such simulation parameter.More manifest the importance of initial estimation from Fig. 4 and Fig. 5, in high SNR district, based on the BER performance of the sensitive bit algorithm of orderly SIC initial estimation significantly better than sensitive bit algorithm based on the LMMSE initial estimation.In Fig. 5, when the sensitive bit number of setting during much smaller than the total number of bits that detects, it is effective adopting iteration type sensitive bit algorithm as can be seen.Particularly, sensitive bit is counted f=2, and number of repetition is that the performance of 1 time iteration sensitive bit algorithm is better than that sensitive bit is counted f=3 but the sensitive bit algorithm that do not have repetitive operation.

Except the BER performance, another important factor of check algorithm quality is an algorithm complex.8 users, under the situation of QPSK, finish that optimum MLD algorithm will calculate and relatively all 4 ⁸The tolerance of=65,536 possibility symbolic vectors.Yet the sensitive bit algorithm of our invention realizes that except make simple LMMSE or orderly SIC algorithm in the starting stage we only need make 2N+2 in second stage the estimation just ^fThe calculating of=32 possibility symbolic vectors is relatively just passable.Thereby simplified the algorithm complex of optimum MLD widely and the BER performance loss is very little.The complexity of sensitive bit algorithm and optimum MLD algorithm is listed in table 1, and wherein we have only considered multiplying.What note is that the computing of initialization section only needs do once in the invariable time at channel, and sensitive bit algorithm part is carried out computing by the speed of symbol rate.

At last, we have provided particularly and have worked as M=12, N=8, under the situation of Q=4 and f=5, corresponding algorithm complex relatively list in table 2.As seen the algorithm complex of sensitive bit algorithm is 4 * 10 of an optimum MLD algorithm complex ^-4This shows that the MLD algorithm based on sensitive bit is effective, the feasible multiuser detection algorithm of practical application.It should be noted that when total bit number of all users during much larger than the sensitive bit number, we can improve performance by iteration sensitive bit algorithm repeatedly, are about to the initial estimation of the output of sensitive bit algorithm last time as this sensitive bit algorithm.

Table 1: algorithm complex relatively

Table 2: work as M=12, N=8, under the situation of Q=4 and f=5, algorithm complex is relatively

The algorithm computation complexity	Initialization	The SB-MLD algorithm
			LMMSE-SB-MLD	12912(Multiplication)	2820×M
SIC-SB-MLD	5232×M	3062×M
			Exhaustive-MLD		6815744×M

Claims (1)

1, based on the simplification maximum likelihood multi-user test method of sensitive bit, the non-optimum multiuser detection algorithm that contains useful traditional low complex degree is estimated multiple user signals, and be the step that the MLD algorithm is corrected the bit of makeing mistakes with Maximum Likelihood Detection, it is characterized in that: it is to tell sensitive bit by a kind of of computer in the above-mentioned multiple user signals that estimates, the method of in corresponding to the littler search subset of sensitive bit, correcting the bit that makes mistake again with the MLD algorithm, it contains successively and has the following steps:

(1). with traditional linear least mean-square difference is that LMMSE and orderly successive interference eliminated are that the SIC algorithm is realized initial Multiuser Detection;

(2). preset initial multiple user signals vector;

(3). in subclass, use the MLD algorithm corresponding to sensitive bit:

(3.1) definition f is maximum sensitive bit number;

(3.2) symbolic vector of transmitting N the user that step (1) is obtained

Be mapped to the vector of 2N bit After, wherein:

I=1,2...N, j=I, Q, J, Q is the sine and the cosine component of conventional letter respectively; By counter-rotating In the polarity of ij bit obtain new symbolic vector It is corresponding to initial estimation

There is and only have a bit to change polarity;

(3.3) all bits that reverse obtain 2N new symbolic vector Calculate all new symbolic vectors 2N (i=1,2 ... ..N, j=I, Q);

(3.4) 2N tolerance that newly obtains

Middle search 2 ^fIndividual minimum Define corresponding this 2 ^fF in the individual vector counter-rotating bit is " sensitive bit ";

(3.5) it is constant to fix other 2N-f bit, corresponding to 2 of f sensitive bit ^fMake Maximum Likelihood Detection in the subclass of individual symbolic vector;

(4). judge that iteration finishes not?:

If: do not finish, then get back to step (2);

If: finish, then export the Multiuser Detection signal.

Images