CN103618585B - A joint multi-user detecting and decoding method based on a belief propagation algorithm - Google Patents

Abstract

The invention relates to a joint multi-user detecting and decoding method based on a belief propagation algorithm and belongs to communication technical field, and in particular, relates to a joint multi-user detecting and decoding method for eliminating interference among multiple users based on a belief propagation algorithm in a multi-user communication system. The extrinsic information of bits and the extrinsic information of symbols of a user and are repeatedly iterated at encoding/decoding nodes and mapping/inverse mapping nodes. After the number of the iteration is satisfied, the joint multi-user detecting and decoding method may achieve joint interference elimination and decoding among multiple users. Compared with a conventional multi-user interference eliminating method, the joint multi-user detecting and decoding method is capable of effectively decreasing the computation complexity of an interference eliminating algorithm and increasing the performance of the interference elimination.

Description

A kind of associating Multiuser Detection based on belief propagation algorithm and interpretation method

Technical field

A communication technical field is belonged to, the associating Multiuser Detection disturbed between a kind of elimination multi-user based on belief propagation algorithm particularly in multi-user comm and interpretation method based on the associating Multiuser Detection of belief propagation algorithm and interpretation method.

Background technology

In order to reduce costs and utilize efficiently the resource of communication system, existing communication system allows multiple common-user system resource mostly.Generally can adopt time division multiple access (Time Division Multiple Access, TDMA), frequency division multiple access (Frequency Division Multiple Access, FDMA), the different modes such as code division multiple access (Code Division Multiple Access, CDMA) distinguish different users.These multi-access modes theoretic can fully distinguish each user, interference can not be produced between different user, but in practical communication system, ambient noise or synchronous error will cause there is interference between user, and inter-user interference can reduce the performance of system.

Inter-user interference is regarded as the additive noise in environment by conventional method, detects the information of each user independently.Along with the increase of number of users, inter-user interference can increase, and the performance of whole communication system can decline.Compared with conventional method, multiuser detection is no longer used as the interference of other users noise to process, but combine all users considering busy channel, weaken or eliminate the interference of other user to arbitrary user, detecting the information of certain user or all these users simultaneously.Traditional multi-user test method is divided into Linear Multiuser Detection method and Multi-User Detection method.Multiuser Detection linear algorithm adopts linear operator L to dock collection of letters breath and processes.The basic thought of Multi-User Detection algorithm reconstructs the subscriber signal detected, detects other users again after it being eliminated from Received signal strength.These two kinds of methods all can relate to matrix inversion operation in processing procedure, and computation complexity is higher; In addition, detection and decoding are separated independent process by traditional receiving handling method, namely first detect decoding again, cannot realize combined optimization.

The present invention proposes a kind of associating Multiuser Detection based on belief propagation algorithm and interpretation method.The present invention adopts belief propagation algorithm, utilize the structure of factor graph that symbol and intersymbol external information, the external information between symbol and bit and the external information between bit and bit are carried out iteration and renewal repeatedly, eliminate and decoding to obtain maximum a posteriori probability to carry out combining interference.Under linear computation complexity, the present invention can realize good interference elimination performance.

Summary of the invention

The object of this invention is to provide a kind of associating Multiuser Detection based on belief propagation algorithm and interpretation method.The present invention uses belief propagation algorithm, by the external information of the bit of user and the external information of symbol at " coding/decoding node " with " mapping/anti-mapping node " iterates, after meeting iterations, the interference of combining that the present invention can realize between multi-user is eliminated and decoding.Compare with traditional multi-user interference eliminating method, the present invention effectively can reduce the computation complexity of interference cancellation algorithm and improve the performance of interference elimination.

A kind of associating Multiuser Detection based on belief propagation algorithm and interpretation method are characterised in that, described method detailed process realizes according to following steps successively:

Step (1), initialization:

I-th symbol sent first set in iteration detection method is x _i, x _iin average and the variance use of the t time iteration with represent, iterative algorithm initial phase t=1, arrange,

${\mathrm{\ζ}}_{x_i}^{t-1}=0,{\mathrm{\γ}}_{x_i}^{t-1}=1,$

Meanwhile, adopt QPSK modulation owing to sending, one sends symbol and correspond to 2 coded-bits, in the t time iteration, and a kth coded-bit c of i-th transmission symbol _i,kexternal information be expressed as p ^t(c _i,k) and this method to c _i,kc when being approximately Gaussian Profile _i,kexternal information be expressed as be initialized as,

$p^t\left(c_{i,k}\right)=\frac{1}{2},$

${\tilde{p}}^t\left(c_{i,k}\right)=\frac{1}{2};$

Step (2), propagate and upgrade external information:

In the iterations required, in the t time iterative process, N is total number of users, performs following steps:

Step (2.1), for numbering i, has 0 < i < N+1, represent and send symbol node x _ibe delivered to observation symbol node f _jexternal information, be an approximate Gaussian Profile, average and variance be expressed as with χ represents the symbol field of QPSK, performs following steps and upgrades with

${\hat{x}}_{x_i}^t=\underset{{\mathrm{\&alpha;}}_i\&Element;\mathrm{\&chi;}}{\mathrm{\&Sigma;}}{\mathrm{\&alpha;}}_i{\tilde{p}}^t(x_i={\mathrm{\&alpha;}}_i),$

${\widehat{\mathrm{\&tau;}}}_{x_i}^t=\underset{{\mathrm{\&alpha;}}_i\&Element;\mathrm{\&chi;}}{\mathrm{\&Sigma;}}{|{\mathrm{\&alpha;}}_i-{\hat{x}}_{x_i}^t|}^2{\tilde{p}}^t(x_i={\mathrm{\&alpha;}}_i),$

Step (2.2), based on upgrade after with for numbering j, there is 0 < j < N+1, h _i,jrepresent the coefficient of channel transfer matrix Η, for the variance of noise, y _jfor a jth receiving symbol, upgrade with

${\mathrm{\&tau;}}_{f_j}^t={\mathrm{\&sigma;}}_n^2+\underset{k}{\mathrm{\&Sigma;}}{\left|h_{j,k}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_k}^t,$

$z_{f_j}^t=y_j-\underset{k}{\mathrm{\&Sigma;}}{h}_{j,k}{\hat{x}}_{x_k,}^t$

Meanwhile, for same j, for numbering i, there is 0 < i < N+1, will be approximately a Gaussian Profile after, average and variance be respectively,

${\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t={\mathrm{\&tau;}}_{f_j}^t-{\left|h_{j,i}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_i,}^t$

$z_{f_j\&RightArrow;x_i}^t=z_{f_j}^t+h_{j,i}{\hat{x}}_{x_i}^t,$

Step (2.3), for numbering i, has 0 < i < N+1, calculates x _iaverage in iteration MMSE detection algorithm and variance

${\mathrm{\&gamma;}}_{x_i}^t\&LeftArrow;{\left(\underset{j}{\mathrm{\&Sigma;}}\frac{{\left|h_{j,i}\right|}^2}{{\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t}\right)}^{-1},$

${\mathrm{\&zeta;}}_{x_i}^t\&LeftArrow;{\mathrm{\&gamma;}}_{x_i}^t\underset{j}{\mathrm{\&Sigma;}}\frac{h_{j,i}^{*}{z}_{f_j\&RightArrow;x_i}^t}{{\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t},$

Meanwhile, for numbering q, there is 0 < q < Q+1, calculate coded-bit c _i,qexternal information

Step (3), utilizes decoder completes an iterative process:

To l=1,2 ... L, q=1,2 ... Q, generates the coded-bit c of next iteration _i,qexternal information,

${\tilde{p}}^{t+1}\left(c_{i,q}\right[l\left]\right),p^{t+1}\left(c_{i,q}\right[l\left]\right);$

Step (4), repeats above step until meet iterations, exports decode results.

To compare with conventional multi-user interference elimination method based on the associating Multiuser Detection of belief propagation algorithm and interpretation method in the present invention and there are following two notable features:

● jointly eliminate interference;

● interference is eliminated and decoding Combined Treatment;

Respectively these two features are described below:

● jointly eliminate interference;

Based on different with traditional multiuser detection in the algorithm principle of Multiuser Detection part in associating Multiuser Detection and the interpretation method of belief propagation algorithm.Traditional Multi-User Detection method is the every iteration of serial interference elimination or parallel interference canceller is all once to eliminate and to rebuild, and can realize all users based on the associating Multiuser Detection of belief propagation algorithm and interpretation method and side by side jointly disturb elimination.

● interference is eliminated and decoding Combined Treatment;

As shown in Figure 2, the present invention contains " coding/decoding node ", " mapping/anti-mapping node " and " Multiuser Detection node ", feature due to this algorithm be external information in each node and each internodal propagation, three nodes closely link together the entire infrastructure of composition algorithm.The interference that the present invention realizes between different user at Multiuser Detection node is eliminated, and realizes decoding at coding and decoding node, namely achieves interference and eliminates and decoding Combined Treatment.

Accompanying drawing explanation

The associating Multiuser Detection of Fig. 1 based on belief propagation algorithm and the factor graph of interpretation method composition;

Fig. 2 is based on the associating Multiuser Detection of belief propagation algorithm and interpretation method algorithm flow chart;

Fig. 3 is based on the associating Multiuser Detection of belief propagation algorithm and interpretation method receiver schematic diagram;

Fig. 4 is based on the associating Multiuser Detection of belief propagation algorithm and interpretation method and the Performance comparision of MMSE decoding algorithm under RICE channel;

Embodiment

As shown in Figure 2, the associating Multiuser Detection that the present invention proposes and interpretation method comprise the following steps for the associating Multiuser Detection based on belief propagation algorithm that the present invention proposes and interpretation method:

When this method starts, the external information of first initialization top " coding/decoding node " as shown in Figure 2, the corresponding relation of information bit and coded-bit depends on coded system.The symbol sebolic addressing received is input to " the Multiuser Detection node " shown in accompanying drawing 1, and initialization associating Multiuser Detection and code translator, now arrange iterations and be initialized as 0.Then, do not meet iterations condition is set time, Multiuser Detection node carries out the special delivery of external information, calculating and renewal based on belief propagation algorithm; After meeting iterations, as shown in Figure 2, " Multiuser Detection node " exports the external information of the coded-bit obtained, and through " mapping/anti-mapping node ", " coding/decoding node " obtains decode results.

Principle and the arthmetic statement of the method for the invention are as follows:

1) initialization:

Iterative algorithm initial phase t=1, is arranged,

${\mathrm{\&zeta;}}_{x_i}^{t-1}=0,{\mathrm{\&gamma;}}_{x_i}^{t-1}=1,$

$p^t\left(c_{i,k}\right)=\frac{1}{2},$

${\tilde{p}}^t\left(c_{i,k}\right)=\frac{1}{2};$

2) propagate and upgrade external information:

In the t time iterative process, perform following steps:

A) for numbering i, perform following steps and upgrade with

${\hat{x}}_{x_i}^t=\underset{{\mathrm{\&alpha;}}_i\&Element;\mathrm{\&chi;}}{\mathrm{\&Sigma;}}{\mathrm{\&alpha;}}_i{\tilde{p}}^t(x_i={\mathrm{\&alpha;}}_i),$

${\widehat{\mathrm{\&tau;}}}_{x_i}^t=\underset{{\mathrm{\&alpha;}}_i\&Element;\mathrm{\&chi;}}{\mathrm{\&Sigma;}}{|{\mathrm{\&alpha;}}_i-{\hat{x}}_{x_i}^t|}^2{\tilde{p}}^t(x_i={\mathrm{\&alpha;}}_i),$

B) based on upgrade after with for numbering j, there is 0 < j < N+1, upgrade with

${\mathrm{\&tau;}}_{f_j}^t={\mathrm{\&sigma;}}_n^2+\underset{k}{\mathrm{\&Sigma;}}{\left|h_{j,k}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_k}^t,$

$z_{f_j}^t=y_j-\underset{k}{\mathrm{\&Sigma;}}{h}_{j,k}{\hat{x}}_{x_k,}^t$

Meanwhile, for same j, for numbering i, there is 0 < i < N+1, upgrade and variance

${\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t={\mathrm{\&tau;}}_{f_j}^t-{\left|h_{j,i}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_i,}^t$

$z_{f_j\&RightArrow;x_i}^t=z_{f_j}^t+h_{j,i}{\hat{x}}_{x_i}^t,$

C) for numbering i, there is 0 < i < N+1, calculate and variance

${\mathrm{\&gamma;}}_{x_i}^t\&LeftArrow;{\left(\underset{j}{\mathrm{\&Sigma;}}\frac{{\left|h_{j,i}\right|}^2}{{\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t}\right)}^{-1},$

${\mathrm{\&zeta;}}_{x_i}^t\&LeftArrow;{\mathrm{\&gamma;}}_{x_i}^t\underset{j}{\mathrm{\&Sigma;}}\frac{h_{j,i}^{*}{z}_{f_j\&RightArrow;x_i}^t}{{\mathrm{\&tau;}}_{f_j\&RightArrow;x_i}^t},$

Meanwhile, for numbering q, there is 0 < q < Q+1, calculate coded-bit c _i,qexternal information

3) utilize complete decode procedure:

To l=1,2 ... L, q=1,2 ... Q, generates the coded-bit c of next iteration _i,qexternal information,

${\tilde{p}}^{t+1}\left(c_{i,q}\right[l\left]\right),p^{t+1}\left(c_{i,q}\right[l\left]\right);$

4) repeat above step until meet iterations, export decode results.

A specific embodiment of the present invention

A kind of associating Multiuser Detection based on belief propagation algorithm that the present invention proposes and interpretation method, for multi-user's cdma spread spectrum communication system.The reverse link of this multi-user cdma spread spectrum communication system uses L frequency range, and the length of the spreading code that cdma spread spectrum communication system uses is 16, and Installed System Memory is 8 users, and modulation system is QPSK, and coding uses LDPC code, under RICE channel, and K=10.Below provide workflow, be convenient to understand object of the present invention, feature and advantage.

(1) LDPC code is adopted to encode to the information bit of each user;

(2), after each subscriber-coded bit stream carries out QPSK constellation point, complex data flow is formed.

(3) output time-domain data flow after complex data flow parallel-serial conversion;

(4) data that Base-Band Processing is later, through digital-to-analogue (D/A) conversion, form continuous analog signal, then through up-conversion, are modulated to radio frequency;

(5) finally modulation signal is launched.

(6) the radio frequency unit Received signal strength of terminal, realizes down-conversion and analog to digital conversion, produces digital baseband signal;

(7) external information calculated by the digital baseband information of multiple user substitutes into Multiuser Detection node, is transmitted and upgrades, draw the external information of multiple user symbol by belief propagation algorithm;

(8) external information of updated different user symbol is passed through " mapping/anti-mapping node ", calculate the external information of each subscriber-coded bit stream;

(9) external information of updated different user coded-bit is passed through " coding/decoding node ", calculate the external information of each user profile bit stream;

(10) by the external information of message bit stream that obtained according to the direction going down of schematic structure, successively pass through " coding/decoding node ", " coding/decoding node " and " Multiuser Detection node ", recalculates and upgrades external information;

(11) said process is repeated, until meet iterations.

Computer Simulation test is carried out to said system, the result below having drawn:

Accompanying drawing 4 and table 1, compare based on the associating Multiuser Detection of belief propagation algorithm and interpretation method and traditional MMSE ber curve.Can see that the associating Multiuser Detection based on belief propagation algorithm has the lower error rate with interpretation method under equal signal to noise ratio condition compared with additive method.

The performance of table 1. two kinds of decoding algorithms under RICE channel

?	E b/N 0＝2dB	E b/N 0＝3dB	E b/N 0＝4dB
				IteraMMSE-QPSK	0.0350	0.00038	0.0005
ABP-QPSK	0.0175	0.0011	0.0001

Visible, adopt this method can obtain the lower error rate under identical signal to noise ratio condition, improve laser propagation effect.

Effect of the present invention is, propagation on the factor graph formed at Multiuser Detection node by utilizing external information and renewal make external information feature more reliably, effectively lower ambient noise and the impact of interference on signal message, obtain lower error code result and more reliable laser propagation effect, this algorithm can realize multi-user interference elimination and carry out combining of decoding simultaneously, is conducive to the realization of parallel organization.Therefore, eliminate for multi-user interference, this method is obviously better than additive method.

Claims (1)

1. the associating Multiuser Detection based on belief propagation algorithm and interpretation method are characterised in that, described method detailed process realizes according to following steps successively:

Step (1), initialization:

I-th symbol sent first set in iteration detection method is x _i, x _iin average and the variance use of the t time iteration with represent, iterative algorithm initial phase t=1, arrange,

${\mathrm{\&zeta;}}_{x_i}^{t-1}=0,{\mathrm{\&gamma;}}_{x_i}^{t-1}=1,$

Meanwhile, adopt QPSK modulation owing to sending, one sends symbol and correspond to 2 coded-bits, in the t time iteration, and a kth coded-bit c of i-th transmission symbol _i,kexternal information be expressed as p ^t(c _i,k) and this method to c _i,kc when being approximately Gaussian Profile _i,kexternal information be expressed as be initialized as,

$p^t\left(c_{i,k}\right)=\frac{1}{2},$

${\tilde{p}}^t\left(c_{i,k}\right)=\frac{1}{2},$

Step (2), propagate and upgrade external information:

In the iterations required, in the t time iterative process, N is total number of users, performs following steps:

Step (2.1), for numbering i, has 0<i<N+1, represent and send symbol node x _ibe delivered to the external information of observation symbol node fj, be an approximate Gaussian Profile, average and variance be expressed as with χ represents the symbol field of QPSK, performs following steps and upgrades with

${\hat{x}}_{x_i}^t=\underset{{\mathrm{\&alpha;}}_i\&Element;\mathrm{\&chi;}}{\mathrm{\&Sigma;}}{\mathrm{\&alpha;}}_i{\tilde{p}}^t(x_i={\mathrm{\&alpha;}}_i),$

Step (2.2), based on upgrade after with for numbering j, there is 0<j<N+1, h _i,jrepresent the coefficient of channel transfer matrix Η, for the variance of noise, y _jfor a jth receiving symbol, upgrade variance and average

${\mathrm{\&tau;}}_{f_i}^t={\mathrm{\&sigma;}}_n^2\text{+}\underset{k}{\mathrm{\&Sigma;}}{\left|h_{j,k}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_k}^t,$

$z_{f_i}^t=y_i-\underset{k}{\mathrm{\&Sigma;}}{h}_{j,k}{\hat{x}}_{x_k}^t,$

Meanwhile, for same j, for numbering i, there is 0<i<N+1, will be approximately a Gaussian Profile

after, average and variance be respectively,

${\mathrm{\&tau;}}_{f_i\&RightArrow;x_i}^t={\mathrm{\&tau;}}_{f_i}^t-{\left|h_{j,i}\right|}^2{\widehat{\mathrm{\&tau;}}}_{x_i}^t,$

$z_{f_i\&RightArrow;x_i}^t=z_{f_i}^t+h_{j,i}{\hat{x}}_{x_i}^t,$

Step (2.3), for numbering i, has 0<i<N+1, calculates x _iaverage in iteration MMSE detection algorithm and variance

${\mathrm{\&gamma;}}_{x_i}^t\&LeftArrow;{\left(\underset{j}{\mathrm{\&Sigma;}}\frac{{\left|h_{j,i}\right|}^2}{{\mathrm{\&tau;}}_{f_i\&RightArrow;x_i}^t}\right)}^{-1},$

${\mathrm{\&zeta;}}_{x_i}^t\&LeftArrow;{\mathrm{\&gamma;}}_{x_i}^t\underset{j}{\mathrm{\&Sigma;}}\frac{h_{j,i}^{*}{z}_{f_i\&RightArrow;x_i}^t}{{\mathrm{\&tau;}}_{f_i\&RightArrow;x_i}^t},$

Meanwhile, for numbering q, there is 0<q<Q+1, calculate coded-bit c _i,qexternal information

Wherein, for about variable x _iaverage be variance is gauss of distribution function.

Step (3), utilizes decoder completes an iterative process:

To l=1,2 ... L, q=1,2 ... Q, generates the coded-bit c of next iteration _i,qexternal information,

${\tilde{p}}^{t+1}\left(c_{i,q}\right[l\left]\right),p^{t+1}\left(c_{i,q}\right[l\left]\right);$

Step (4), repeats above step until meet iterations, exports decode results.