CN101001092A - Multi-user detection method and device using low complexity solution - Google Patents

Abstract

This invention provides two kinds of low complexity decorrelation multi-user test methods and devices, which avoid the computation of inversion to matrix by QR analysis and circulation iteration, and introduce adaptive regulated factors to improve the algorithm performance. Since the itration method reduces rounding error in computation, it increases its accuracy at the time when reducing computing volume of data greatly, reduces noises resulted in MAI and resumes multi-user information finally.

Description

The decorrelation multi-user test method and the device of low complex degree

Technical field:

The present invention relates to a kind of decorrelation multi-user test method and device of low complex degree, belong to the digital communicating field that uses the electromagnetic wave technology, relate in particular to the apparatus and method of Multiuser Detection in direct sequence CDMA (DS-CDMA) system, these apparatus and method can be removed the multiple access that is present in the DS-CDMA system in real time and disturb (MAI).

Background technology:

In cdma communication system, the principal element of system for restricting performance comprises cochannel interference, intersymbol interference etc.Intersymbol interference has retrained the speed of system communication, mainly adopts equilibrium or diversity technique to suppress; Mainly by disturbing the capacity that has then limited system communication with the travelling carriage of sub-district, the same frequency base station of neighbor cell or the cochannel that multiple access such as travelling carriage and other communication systems disturb (MAI) to produce.At present, the technology of eliminating the multiple access interference is a multiuser detection, this technology is on the traditional detection technical foundation, all the subscriber signal information that multiple access disturbs that produce have been made full use of, thereby has good interference free performance, reduced the requirement of system, improved power system capacity significantly the power control precision.

Multiuser detection algorithm mainly is divided into four classes: Optimum Detection, linearity test algorithm, interference cancellation detection algorithm, Blind Detect Algorithm.Optimum multiuser detection algorithm is based on the maximal possibility estimation criterion, though performance the best, computational complexity is too high; The corresponding algorithm that improves has also reduced amount of calculation really to a certain extent, but still is difficult to practicality.The interference cancellation detection algorithm mainly is divided into counteracting serial interference algorithm and Parallel Interference Cancellation algorithm, its basic thought is by reproducing part multiple access interference (PMAI) thereby also offsets the purpose that reaches inhibition MAI that the performance of interference offset device depends on accuracy and the interference-bucking value that reproduces PMAI.In the time can't improving channel estimation accuracy, the performance of partial interference cancellation receiver is interfered to a great extent and offsets the constraint of weights.The blind Detecting device mainly comprise based on linear least mean-square (the MMSE)/recurrence two of least energy output (MOE) criterion take advantage of (RLS)/QR-RLS blind Detecting device, based on blind Detecting device, blind Detecting device, mixed type half-blindness detector four classes of channel subspace search based on narrowband systems and ARRAY PROCESSING.The main thought of blind Detecting device is by subspace tracking technology picked up signal subspace and utilizes it to eliminate the interference that the unknown subscriber causes.Generally speaking, the blind Detecting device has good detection performance under awgn channel.The linearity test algorithm of performance suboptimum mainly is divided into decorrelation detection algorithm and linear minimum mean-squared error (LMMSE) detection algorithm two big classes.The decorrelation algorithm can be eliminated MAI fully, but noise is amplified to some extent.The LMMSE algorithm has reduced the amplification of Linear Mapping to noise, but the MAI elimination not exclusively, must estimate the amplitude of receiver signal simultaneously.

Influencing these two kinds of linear multi-user detection algorithms, to become practical key reason be the problem of matrix inversion.On the one hand, in changeable mobile environment, the inverse matrix of finding the solution correlation matrix R is difficult for.Especially in cdma system, to disturb in order reducing better, generally all will to adopt voice activation and variable rate coding technology, this makes being correlated with between the subscriber signal change at any time, thereby the inverse matrix of R becomes when also being, this makes the inverse matrix of finding the solution R in real time very difficulty that becomes.On the other hand, the multiplying amount of the fast algorithm of finding the inverse matrix also is K ³Rank, when number of users K more for a long time, its computation complexity is difficult to accept especially.

Summary of the invention:

The objective of the invention is:, provide a kind of computation complexity lower, the new quick QR decorrelation multi-user test method that is applicable to code division multiple access system at the deficiencies in the prior art.

For achieving the above object, technical scheme of the present invention is: the method that adopts QR to decompose replaces the inversion operation that correlation matrix is done of the prior art, thereby computation complexity is reduced greatly.This algorithm decomposes the inhibition that has realized apace the multiple access interference by simple QR.

Introduce existing linear multi-user detection method and principle of the present invention below:

1, existing decorrelation multi-user test method

The matrix notation of correlator output discrete signal is:

y＝RAb+N (1)

Wherein, R=SS ^H, the size of correlation matrix R is K * K, it represents two correlations between the code word.

Conventional linear multi-user detection algorithm comprises three kinds of conventional multiuser detection algorithm, decorrelation algorithm and MMSE multiuser detection algorithms, all can samely be expressed as:

$\hat{b}=\mathrm{sgn}\left(\left(\mathrm{Ly}\right)\right)---\left(2\right)$

Wherein, L is K * K matrix.

In conventional multiuser detection algorithm, suppose that I is the unit matrix of K * K size, L=I is arranged, at this moment:

$\hat{b}=\mathrm{sgn}\left(\left(\mathrm{Ly}\right)\right)=\mathrm{sgn}\left(y\right)---\left(3\right)$

In decorrelation detector, adopt inverse matrix L=R ^-1Multiply by mutually and demodulate subscriber signal, have:

$\hat{b}=\mathrm{sgn}\left(\mathrm{Ly}\right)=\mathrm{sgn}\left(R^{-1}y\right)=\mathrm{sgn}(\mathrm{Ab}+R^{-1}N)---\left(4\right)$

2, quick QR decorrelation multiuser detection algorithm one

Conventional decorrelation multi-user detector must be inverted to relevant battle array R, and this has caused algorithm computation complexity height, cannot realize real-time.At this shortcoming, quick QR decorrelation multiuser detection algorithm has been proposed, this algorithm not only computation complexity is low, and performance is more excellent.

Suppose:

L＝(R+α·I) ^-1

d＝(R+αI) ^-1y＝(R+αI) ^-1RAb+(R+αI) ^-1N (5)

In the formula, α is selectable constant.

Then input is obtained by following formula:

$\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}\left({(R+\mathrm{\αI})}^{-1}y\right)---\left(6\right)$

Wherein, related (R+ α I) ^-1Calculating can be according to the fast algorithm implementation of back 4 introductions.

3, improved fast linear multiuser detection algorithm two

Suppose:

$L={(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}$

$\mathrm{Ly}={(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}y---\left(7\right)$

In the formula, p is selectable constant.

Then, input is obtained by following formula:

$\hat{b}=\mathrm{sgn}\left(\mathrm{Ly}\right)=\mathrm{sgn}\left({(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}y\right)---\left(8\right)$

Wherein, related (R+p (σ _n ²/ σ _s ²) I) ^-1Calculating can be according to 3.4 fast algorithm implementation of introducing.

4, the quick implementation algorithm of inverting in the linear multi-user detection algorithm

The linear multi-user detection algorithm all relates to matrix inversion, and its amount of calculation is long along with the increase of the order of inverting is the index multiplication, thereby can increase the requirement to hardware, and cannot accomplish that this is unpractical for the bigger cdma system of number of users in real time.For this reason, finding one, to improve algorithm fast be necessary.Introduce a kind of method of inverting fast below, this method can make two algorithm operation quantities of advising previously reduce greatly.

For relevant battle array R ∈ C ^{K * K}And nonsingular, then R can be decomposed into

R＝PQ (9)

Wherein P is unitary matrix, that is: PP ^H=E, Q are upper triangular matrixs, and E is a unit matrix, Q ^HIt is the conjugate transpose of Q.

R ^-1＝Q ^-1·PH (10)

R wherein ^-1, Q ^-1Be respectively the inverse matrix of R and Q.

If V=Q ^-1=(v _Ij) _{K * K}Be upper right triangle battle array Q=(q _Ij) _{K * K}Inverse matrix, then its element must satisfy:

$v_{\mathrm{ij}}=\left\{\begin{array}{cc}0& i>j\\ 1/q_{\mathrm{ij}}& i=1,...,K\\ -\left(\underset{k=i+1}{\overset{j}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}{v}_{\mathrm{kj}}\right)/q_{\mathrm{ij}}& i=K-1,...,1;j=i+1,...,K\end{array}\right.---\left(11\right)$

Formula (9), (10), (11) substitution formula (5) or formula (6) are just had:

d＝Ly＝Q ^-1·P ^H·y＝V·P ^H·y (12)

At this moment, $\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}(V\&CenterDot;P^H\&CenterDot;y)---\left(13\right)$

Know that by formula (13) improved decorrelation algorithm must not calculate the inverse matrix of correlation matrix R, R is decomposed into matrix P and matrix Q, very easily obtain matrix V by formula (11) then and need only decompose by QR.

Concrete operations step of the present invention is as follows:

Step 1:, can obtain the vector of k matched filter banks output at receiving terminal:

y＝RAb+N (14)

Wherein, R=SS ^H, the size of correlation matrix R is K * K, it represents two correlations between the code word.

Step 2: for the decorrelation multi-user detector, selected self adaptation is adjusted factor-alpha, supposes that I is the unit matrix of K * K size, for matrix (R+ α I) ∈ C ^{K * K}Nonsingular, then matrix (R+ α I) can be decomposed into:

R+αI＝PQ (15)

Wherein, P is unitary matrix, that is: PP ^H=E, Q are upper triangular matrixs, and E is a unit matrix, Q ^HIt is the conjugate transpose of Q.

L＝(R+α·I) ^-1＝Q ^-1·P ^H (16)

d＝(R+αI) ^-1y＝(R+αI) ^-1RAb+(R+αI) ^-1N (17)

Wherein, R ^-1, Q ^-1Be respectively the inverse matrix of R and Q.

Step 3: establish V=Q ^-1=(v _Ij) _{K * K}Be upper right triangle battle array Q=(q _Ij) _{K * K}Inverse matrix, then its element must satisfy:

$v_{\mathrm{ij}}=\left\{\begin{array}{cc}0& i>j\\ 1/q_{\mathrm{ij}}& i=1,...,K\\ -\left(\underset{k=i+1}{\overset{j}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}{v}_{\mathrm{kj}}\right)/q_{\mathrm{ij}}& i=K-1,...,1;j=i+1,...,K\end{array}\right.---\left(18\right)$

Step 4: formula (15), (16), (18) substitution formula (17) are just had:

d＝L ^-1y＝Q ^-1·P ^H·y＝V·P ^H·y (19)

Step 5: can demodulate subscriber signal this moment:

$\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}(V\&CenterDot;P^H\&CenterDot;y)---\left(20\right)$

It more than is improved low complex degree QR decorrelation multi-user test method one.

Second improved low complex degree QR decorrelation multi-user test method comprises the steps:

Step 1:, can obtain the vector of k matched filter banks output at receiving terminal:

y＝RAb+N (21)

Wherein, R=SS ^H, the size of correlation matrix R is K * K, it represents two correlations between the code word.

Step 2: for the decorrelation multi-user detector, selected self adaptation is adjusted factor-alpha, supposes that I is the unit matrix of K * K size, for matrix (R+ α I) ∈ C ^{K * K}Nonsingular, then matrix (R+ α I) can be decomposed into:

$(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)=\mathrm{PQ}---\left(22\right)$

Wherein, P is unitary matrix, that is: PP ^H=E, Q are upper triangular matrixs, and E is a unit matrix, Q ^HIt is the conjugate transpose of Q.

$L={(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}---\left(23\right)$

$\mathrm{Ly}={(R+p\&CenterDot;({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="A20061003272800111.png"\; state="\{\{state\}\}">n</figure-callout>}}^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}y---\left(24\right)$

In the formula, p is selectable constant.

Step 3: establish V=Q ^-1=(v _Ij) _{K * K}Be upper right triangle battle array Q=(q _Ij) _{K * K}Inverse matrix, then its element must satisfy:

$v_{\mathrm{ij}}=\left\{\begin{array}{cc}0& i>j\\ 1/q_{\mathrm{ij}}& i=1,...,K\\ -\left(\underset{k=i+1}{\overset{j}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}{v}_{\mathrm{kj}}\right)/q_{\mathrm{ij}}& i=K-1,...,1;j=i+1,...,K\end{array}\right.---\left(25\right)$

Step 4: formula (22), (23), (25) substitution formula (24) are just had:

d＝Ly＝Q ^-1·P ^H·y＝V·P ^H·y (26)

Step 5: can demodulate subscriber signal this moment:

$\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}(V\&CenterDot;P^H\&CenterDot;y)---\left(27\right)$

It more than is improved low complex degree QR decorrelation multi-user test method two.

The present invention decomposes the inversion operation that replaces correlation matrix at the receiving terminal of cdma communication system by QR, especially the user more for a long time, greatly reduce computation complexity, reached the purpose of quick realization.

Simultaneously, prove also through the cdma communication system link simulation, compare that this algorithm employing QR decomposition is inverted computational complexity is reduced greatly with other conventional methods, and loop iteration reduced round-off error in calculation, thereby makes the performance of bit error rate and anti-near-far interference all more excellent.

Description of drawings:

Fig. 1 is the block diagram of explanation according to multi-user detection device in the cdma system of the present invention.

Fig. 2, Fig. 3 are user 1 bit error rate situation.As shown in the figure, it is better than decorrelation detector and MMSE detector false bit-rate performance that improved fast linear iterates the performance of detector, has stronger counteracting multiple access interference capability.

Fig. 4 is the situation of change of user 1 bit error rate with " distance " effect.As shown in the figure, as long as suitable selection constant coefficient, improved fast linear iterates detector performance and has better anti-" distance " effect performance than decorrelation detector and MMSE detector.

Embodiment:

Below by concrete enforcement technical scheme of the present invention is further described.

Adopt the emulation of Matlab simulation software, analogue system is a synchronous/asynchronous DS-CDMA system, awgn channel, the BPSK modulation, carrier phase is 0, with length is 10 pseudo random sequence spread spectrum, and the decorrelation multi-user detector before and after revising is analyzed from bit error rate, anti-" distance " effect and three aspects of computation complexity respectively.Suppose that user 1 is desired user, number of users is 10.Just can demodulate signal according to foregoing five steps.

Claims (5)

1, the present invention relates to the decorrelation multi-user test method and the device of two kinds of low complex degrees, it is characterized in that wherein a kind of method comprises the steps:

Step 1:, can obtain the vector of k matched filter banks output at receiving terminal:

y＝RAb+N (1)

Wherein, R=SS ^H, the size of correlation matrix R is K * K, it represents two correlations between the code word.A is the diagonal matrix that comprises all user's amplitude modulation, and b is the output bit, and N is the noise vector.

Step 2: for the decorrelation multi-user detector, selected self adaptation is adjusted factor-alpha, supposes that I is the unit matrix of K * K size, for matrix (R+ α I) ∈ C ^{K * K}Nonsingular, then matrix (R+ α I) can be decomposed into:

R+αI＝PQ (2)

Wherein, P is unitary matrix, that is: PP ^H=E, Q are upper triangular matrixs, and E is a unit matrix, Q ^HIt is the conjugate transpose of Q.

L＝(R+α·I) ^-1＝Q ^-1·P ^H (3)

d＝(R+αI) ^-1y＝(R+αI) ^-1RAb+(R+αI) ^-1N (4)

Wherein, R ^-1, Q ^-1Be respectively the inverse matrix of R and Q.

Step 3: establish V=Q ^-1=(v _Ij) _{K * K}Be upper right triangle battle array Q=(q _Ij) _{K * K}Inverse matrix, then its element must satisfy:

$v_{\mathrm{ij}}=\{\begin{array}{cc}0& i>j\\ 1/q_{\mathrm{ij}}& i=1,...,K\\ -\left(\underset{k=i+1}{\overset{j}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}{v}_{\mathrm{kj}}\right)/q_{\mathrm{ij}}& i=K-1,...,1;j=i+1,...,K\end{array}---\left(5\right)$

Step 4: formula (2), (3), (5) substitution formula (4) are just had:

d＝L ^-1y＝Q ^-1·P ^H·y＝V·P ^H·y (6)

Step 5: at this moment:

$\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}(V\&CenterDot;P^H\&CenterDot;y)---\left(7\right)$

Can demodulate subscriber signal by (7) formula.

2, the present invention relates to the decorrelation multi-user test method and the device of two kinds of low complex degrees, it is characterized in that another kind of method comprises the steps:

Step 1:, can obtain the vector of k matched filter banks output at receiving terminal:

y＝RAb+N (8)

Wherein, R=SS ^H, the size of correlation matrix R is K * K, it represents two correlations between the code word.A is the diagonal matrix that comprises all user's amplitude modulation, and b is the output bit, and N is the noise vector.

Step 2: for the decorrelation multi-user detector, selected self adaptation is adjusted factor-alpha, supposes that I is the unit matrix of K * K size, for matrix (R+ α I) ∈ C ^{K * K}Nonsingular, then matrix (R+ α I) can be decomposed into:

$(R+p\&CenterDot;({\mathrm{\&sigma;}}_n^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)=\mathrm{PQ}---\left(9\right)$

Wherein, P is unitary matrix, that is: PP ^H=E, Q are upper triangular matrixs, and E is a unit matrix, Q ^HThe conjugation that is Q is changeed stern.

$L={(R+p\&CenterDot;({\mathrm{\&sigma;}}_n^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}---\left(10\right)$

$\mathrm{Ly}={(R+p\&CenterDot;({\mathrm{\&sigma;}}_n^2/{\mathrm{\&sigma;}}_s^2)\&CenterDot;I)}^{-1}y---\left(11\right)$

In the formula, p is selectable constant.

Step 3: establish V=Q ^-1=(v _Ij) _{K * K}Be upper right triangle battle array Q=(q _Ij) _{K * K}Inverse matrix, then its element must satisfy:

$v_{\mathrm{ij}}=\{\begin{array}{cc}0& i>j\\ 1/q_{\mathrm{ij}}& i=1,...,K\\ -\left(\underset{k=i+1}{\overset{j}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}{v}_{\mathrm{kj}}\right)/q_{\mathrm{ij}}& i=K-1,...,1;j=i+1,...,K\end{array}---\left(12\right)$

Step 4: formula (9), (10), (12) substitution formula (11) are just had:

d＝Ly＝Q ^-1·P ^H·y＝V·P ^H·y (13)

Step 5: at this moment:

$\hat{b}=\mathrm{sgn}\left(d\right)=\mathrm{sgn}(V\&CenterDot;P^H\&CenterDot;y)---\left(14\right)$

Can demodulate subscriber signal by formula (14).

3, said as claim 1 and claim 2, the linear iteraction multiple users detecting device of two kinds of low complex degrees and method is characterized in that: this method has adopted simple QR to decompose to replace in the prior art inversion operation to correlation matrix.

4, said as claim 1 and claim 2, the linear iteraction multiple users detecting device of two kinds of low complex degrees and method is characterized in that: self adaptation adjusts factor-alpha and p can be selected according to actual conditions, and generally this factor is less than 1.

5, said as claim 1 and claim 2, the linear iteraction multiple users detecting device of two kinds of low complex degrees and method is characterized in that: can be used for Multiuser Detection removal multiple access interference in DS-CDMA (direct sequence CDMA) system or the cdma system.

Images