CN1252959C - Method for testing DS. CDMA system multi-user developed based on weighting - Google Patents

Abstract

The present invention relates to a multi-user detection method for a code division multiple access system of a multipath channel in wireless communication. The present invention is characterized in that a multinomial expansion type basic matrix is obtained through optimization design by using the minimized spectral radius as a target; a weighting sequence e<0>, e<1>... e<N> us designed according to the principle that the error output by the corresponding linear multi-user detector is minimum; intermediate signal vectors generated by multistage iteration are added by using the weighting sequence e<0>, e<1>... e<N> so as to obtain a final multi-user joint detection output result with excellent performance. In the condition of maintaining low realizing complexity, the present invention effectively reduces the interference resulted from multiple accesses and multiple paths in the code division multiple access system in the multipath channel.

Description

DS-CDMA system multi-user detection method based on the weighting expansion

Technical field

The present invention relates to code division multiple access system (CDMA) multi-user test method under the radio communication multipath channel.Present invention is specifically related in the big capacity wireless communication system in broadband, keep under the prerequisite than low implementation complexity, reduce in the multipath channel code division multiple access system because the method for the interference that multiple access inserts and multidiameter delay causes and raising power system capacity and performance.

Background technology

The performance of code division multiple access system (CDMA) mainly is subject to interference, disturbs the signal by other user in the system to cause, reason is because non-orthogonal between the subscriber signal waveform of the non-orthogonal or channel of user's spreading code transmission back in the code division multiple access system.A non-orthogonal example of spreading code be the Walsh transformation scrambled code that adopts in the code division multiple access system spectrum spreading method, in fact be equivalent to spread spectrum at random, non-orthogonal usually between the random sequence that each user's spread spectrum is used in this code division multiple access method, therefore introduced the interference between a plurality of users.Have again, also can be between the subscriber signal waveform of channel transmission back owing to multipath transmi quadrature no longer, at this moment multiuser detection just seems particularly crucial for cdma system.Consider two users' example, their spreading code or spread spectrum random sequence are respectively that 1 ,-1,1 ,-1 and 1,1 ,-1 ,-1, two code book body is a quadrature.If two users' the channel that signal experienced does not all have phase deviation, also be quadrature at receiving terminal user waveform so.But if equivalent spreading code had become 1 ,-1,1 ,-1 after the user experienced channel, and equivalent spreading code of the user 2 has become 1,1 ,-1,1.Obvious two equivalent spreading codes become non-orthogonal, because the dot product of two equivalent spreading codes is 1.2 couples of users' 1 of user despreading meeting causes interference, and 1 couple of user's 2 of user despreading also can cause interference, and this interference is corresponding to the dot product of two equivalent spreading codes.

American scholar S.Verdu has at first proposed to utilize the difference of different user signal structure, effectively reduce phase mutual interference between the user by carry out Combined Treatment for all users' received signal, and then the thought of raising entire system performance, just so-called multi-user association detects thought.Traditional linear multi-user joint detection algorithm mainly comprises decorrelation multi-user association detector and least mean-square error multi-user association detector.Linear multi-user detectors has good performance, but because their implementation complexity is all very high, therefore than difficulty actual use in engineering, especially for the big capacity wireless communication system in broadband.At this situation, American scholar S.Moshavi has proposed to utilize alternative manner to approach the thought of linear multi-user detectors.See S.Moshavi. " Multistage linear Detectors for DS-CDMA systems, " Int ' I.J.Wireless info.Networks for details, vol3, no.1, Jan.1996.This multi-user detector can effectively be avoided nonlinear operation, thus the simplified system design.Yet the alternative manner of Moshavi needs to estimate in advance quantity of parameters, and complicated emulation and calculating are passed through in the acquisition of these parameters again in advance.The flexibility that has below all caused algorithm to be realized is relatively poor, needs a large amount of in advance stores system parameters, thereby actual Project Realization has been caused huge difficulty.At the shortcoming of Moshavi method, people such as Singapore scholar Lei have proposed the iteration multi-subscriber test method based on the Taylor expansion.See Z.D.Lei and T.J.Lim. " simplified polynomial-expansion linear detectors for DS-CDMA systems, " Electronics Letters.Vol.34 for details, no.16, PP.1561-1563, Aug.1998.This multi-user detector carries out polynomial expansion by the inversion operation with correlation matrix, thereby has obtained avoiding the multi-user detector structure of calculation of parameter.People such as the Chen Ming of Southeast China University according to the same principle application of the method for Lei patent " multi-stage iteration multi-subscriber detector ", number of patent application 98111378.8.According to our discovering, the method of Lei exists following significant drawback in the multipath hicap: because the convergence rate that Taylor launches is slower, cause the algorithmic statement performance relatively poor, especially it is bigger to work as power system capacity, number of users more for a long time, algorithm needs a large amount of iteration just can reach better performance, thereby causes problems such as implementation complexity height, real-time difference.

In the big capacity wireless communication system in broadband, use the significant difficulties that exists for solving existing method, the present invention proposes a kind of new multi-user detector based on the weighting multinomial expansion, this new multi-user detector can be under the prerequisite that keeps low time complexity and space complexity, approach linear multi-user detectors very efficiently, thereby effectively overcome many deficiencies that existing correlation technique exists.Studies show that this novel multi-use family detector can effectively be suitable for various multi-user DS/CDMA application systems.

Summary of the invention

The present invention is a kind of novel multi-user detector based on the weighting expansion.When using multi-user test method involved in the present invention, receiver should at first pass through down-conversion, molding filtration, operations such as sampling obtain baseband signal vector y, and generate comprehensive channel response matrix A according to different user channel impulse response vector that obtains and corresponding frequency expansion sequence.Then with y by receiving matched filter banks to generate the signal phasor y after the matched filtering _MF, at last to y _MFCarrying out the specially designed weighting multinomial expansion of the present invention Multiuser Detection handles to obtain final detected value.Multi-user test method involved in the present invention mainly comprises following operation: generate the single-stage iterative processing matrix G and the corresponding intermediate variable α that can quicken the expansion convergence according to special algorithm 1..Algorithm is the optimal design target with the spectral radius that reduces the G matrix, thereby can effectively improve the constringency performance of multi-user detector involved in the present invention, and then the time and the space complexity of the realization of reduction system; 2. with the output signal vector y of matched filter banks _MFInput iterative processing unit, iterative process uses matrix G to multiply by α times of y for many times _MFThereby, obtain the M signal vector sequence: (α y _MF, G α y _MF, G ²α y _MFG ^Nα y _MF); 3. for making the corresponding linear multi-user detector that approaches that multi-user detector output result involved in the present invention can be best export result's (theoretic optimal result), the present invention is a target with the mean error square minimum with linear multi-user detectors output result, designs the weight coefficient sequence (e through optimizing ₀, e ₁... e _k); 4. be weighted summation by M signal vector sequence and the weight sequence of using preceding step to generate, can obtain the final output result of multi-user detector involved in the present invention $\hat{d}=\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{e}_i{G}^i{y}_{\mathrm{MF}}.$ This result can export the decoding decision unit to and be further processed.

In order to prove absolutely principle of the present invention, our frequency expansion sequence of at first arranging different user is expressed as respectively here:

The user 1: $C^{\left(1\right)}=(c_1^{\left(1\right)},c_2^{\left(1\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(1\right)})\&CenterDot;\&CenterDot;\&CenterDot;\left(1\right)$

The user 2: $C^{\left(2\right)}=(c_1^{\left(2\right)},c_2^{\left(2\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(2\right)})$



User K: $C^{\left(K\right)}=(c_1^{\left(K\right)},c_2^{\left(K\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(K\right)})$

K representative of consumer number, Q represents the length of frequency expansion sequence.The channel impulse response sequence table of different user is shown user 1:

$h^{\left(1\right)}=(h_1^{\left(1\right)},h_2^{\left(1\right)},\&CenterDot;\&CenterDot;\&CenterDot;{,h}_W^{\left(1\right)})\&CenterDot;\&CenterDot;\&CenterDot;\left(2\right)$

The user 2: $h^{\left(2\right)}=(h_1^{\left(2\right)},h_2^{\left(2\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h_W^{\left(2\right)})$



User K: $h^{\left(K\right)}=(h_1^{\left(K\right)},h_2^{\left(K\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h_W^{\left(K\right)})$

W represents the length of channel impulse response.Thus, frequency expansion sequence and the corresponding user's impulse response with different user carries out convolution to obtain each user's mixed channel response sequence

b ^(k)＝c ^(k)*h ^(k)，k＝1，2…K (3)

According to definition, b ^(k)The length of sequence is W+Q-1, just

$b^{\left(k\right)}=(b_1^{\left(k\right)},b_2^{\left(k\right)}\&CenterDot;\&CenterDot;\&CenterDot;b_{Q+W-1}^{\left(k\right)}),k=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;K$

Suppose that each frame transmission information symbol number of each user is N _s, then can further arrange symbol a _n ^(k), A ⁽ⁿ⁾With A be

In addition, all users information transmitted symbol in n transmission intercal is concentrated in together, write as total symbolic vector $d^{\left(n\right)}={(d_n^{\left(1\right)},d_n^{\left(2\right)}\&CenterDot;\&CenterDot;\&CenterDot;d_n^{\left(K\right)})}^T,$ Thereby all symbols that all users are transmitted in the transmission time at a frame can be unified to be expressed as

d＝(d ^(1)T，d ^(2)T…d ^(N)T) ^T (5)

Here T represents the matrix transpose computing, and H will represent the matrix transpose conjugate operation.By above agreement, the model of whole system can be represented with following formula:

y＝Ad+n (6)

N is the additive white Gaussian noise vector, has noise power σ ²(can obtain by measuring in advance), y is the received signal vector.At first y being passed through matched filter, is y thereby obtain output vector _MF=A ^HY.Y then _MFTo further carry out Multiuser Detection according to different multi-user detector types handles.For linear multi-user detectors, the estimated signal vector of final output can be unified to be expressed as (the R+ σ of L=in the minimum Mean Square Error Linear multi-user detector ²I) ^-1L=R in the decorrelation linear multi-user detectors ^-1, R=A here ^HA.We are example with the minimum Mean Square Error Linear multi-user detector all in following analysis, only need make σ in the least mean-square error multi-user detector if be concerned about the pairing situation of decorrelation linear multi-user detectors ²=0, all results can continue to use.

Make G=I-α (R+ σ ²I), λ _j(G) be j the characteristic value of matrix G, ρ is the spectral radius of matrix.The multiuser detection algorithm based on the Taylor expansion that traditional people such as Lei propose is only required arbitrarily and is chosen $\mathrm{\&alpha;}\&Element;(0,\frac{2}{\mathrm{\&rho;}})$ Thereby, launch principle L=α (I-G) according to the Taylor that knows ^-1≈ α (1+G+ ... G ^N) (N is enough big), by intercepting the output that limited N item summation can approach the minimum Mean Square Error Linear multi-user detector.The Taylor multi-user detector is final output result can be expressed as

$\hat{d}=\mathrm{\&alpha;}(y_{\mathrm{MF}}+{\mathrm{Gy}}_{\mathrm{MF}}\&CenterDot;\&CenterDot;\&CenterDot;G^N{y}_{\mathrm{MF}})\&CenterDot;\&CenterDot;\&CenterDot;\left(7\right)$

This method need expansion item number N very big, so complexity is higher, detects poor-performing because the convergence rate that Taylor launches is slower, and not being suitable for the 3-G (Generation Three mobile communication system) is the big capacity wireless communication system of high-performance of representative.

In our invention at first according to launching the minimum principle optimal design of basic matrix G spectral radius matrix G, thereby according to the character that Taylor launches, effectively reach the purpose of accelerating algorithm convergence.The specific implementation algorithm is as follows:

$\left\{\begin{array}{c}{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)=\max \left\{\min \right\{r_{i,t}-\underset{s=1,s\&NotEqual;t}{\overset{m}{\mathrm{\&Sigma;}}}\left|r_{t,s}\right|\},0\}+{\mathrm{\&sigma;}}^2,t\&Element;(1,m)\\ {\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)=\max \left\{\underset{s=1}{\overset{m}{\mathrm{\&Sigma;}}}\right|r_{t,s}\left|\right\}+{\mathrm{\&sigma;}}^2,t\&Element;(1,m)\\ \mathrm{\&alpha;}=\frac{2}{{\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)+{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)}\\ G=I-\mathrm{\&alpha;}(R+{\mathrm{\&sigma;}}^{2}I)\end{array}\right.\&CenterDot;\&CenterDot;\&CenterDot;\left(8\right)$

R wherein _{T, s}For the t of correlation matrix R is capable, the element of s row, m is the dimension of R, I is a m dimension unit matrix, σ ²Be the power of additive white Gaussian noise, With

For for matrix R+ σ ²The minimum of I and the estimation of eigenvalue of maximum.Estimate directly to make σ if can not carry out noise power for the simplified system design ²=0.

By (7) formula is configured to following iteration form

$\left\{\begin{array}{c}{X}^{\left(k\right)}=G{X}^{(k-1)}+\mathrm{\&alpha;}{y}_{\mathrm{MF}},k=\mathrm{1,2}\&CenterDot;\&CenterDot;\&CenterDot;N\\ X^{\left(0\right)}=\mathrm{\&alpha;}{y}_{\mathrm{MF}}\end{array}\right.---\left(9\right)$

Traditional multi-user detector based on the Taylor expansion is exported the output vector X that is equivalent to the N time iteration ^(N), promptly $\hat{d}=X^{\left(N\right)}.$ (vector X in the middle of other iteration ^(k), k=0,2 ... N-1 will no longer cut any ice).In our invention, introduced the notion of weighting first, by all middle vector X that (9) formula is produced in iterative process ⁽⁰⁾, X ⁽¹⁾... X ^(N)The weighted sum that is optimized can produce the excellent more final output result of performance, promptly $\hat{d}=\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{a}_i^{\left(N\right)}{X}^{\left(i\right)}$ 。Here optimal design weight coefficient a ₀ ^(N), a ₁ ^(N)... a _N ^(N)Be the most key technology, we adopt output result and theoretical optimum linearity detector output mean error square minimum as design criterion, can be described as seeking one group of weight coefficient a with mathematical formulae ₀ ^(N), a ₁ ^(N)... a _N ^(N), make

$\left\{\begin{array}{c}\min \left|\right|\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{a}_i^{\left(N\right)}{X}^{\left(i\right)}-X_{\mathrm{MMSE}}\left|\right|\\ s.t.a_0^{\left(N\right)}+a_1^{\left(N\right)}+\&CenterDot;\&CenterDot;\&CenterDot;+a_N^{\left(N\right)}=1\end{array}\right.\&CenterDot;\&CenterDot;\&CenterDot;\left(10\right)$

X wherein _MMSEBe the output result of the minimum Mean Square Error Linear multi-user detector of correspondence (limiting performance that can provide just) based on the multiuser detection algorithm of multinomial expansion, || ● || be norm of matrix.

Analysis can obtain a through mathematical optimization ₀ ^(N), a ₁ ^(N)... a _N ^(N)Just should be the multinomial coefficient of N order polynomial P (z) from the low order to the high order.(8) formula of utilization, P (z) can obtain with the following method

$\left\{\begin{array}{c}f=-1\\ g=1-\frac{{2\mathrm{\&sigma;}}^2}{{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)+{\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)}\\ w\left(z\right)=(2z-f-g)/(g-f)\\ P\left(z\right)=T_N\left(w\left(z\right)\right)/T_N\left(w\left(1\right)\right)\end{array}\right.\&CenterDot;\&CenterDot;\&CenterDot;\left(11\right)$

T wherein _N(z)=t _Nz ^N+ t _N-1z ^N-1T ₀Expression N rank Chebyshev multinomials (can obtain) by consulting the mathematical tool handbook.

By with set of vectors X ⁽⁰⁾, X ⁽¹⁾... X ^(N)Power time difference according to G merges similar terms, and multinomial P (z) is launched, and can obtain final weight coefficient e ₀, e ₁E _NExpression formula:

$\left\{\begin{array}{c}\mathrm{\&gamma;}=2/(g-f),\mathrm{\&beta;}=-(g+f)/(g-f)\\ a_i^{\left(N\right)}={\mathrm{\&Sigma;}}_{l=i}^N{t}_l{C}_l^i{\mathrm{\&gamma;}}^i{\mathrm{\&beta;}}^{(l-i)}/{\mathrm{\&Sigma;}}_{l=0}^N{t}_l{(\mathrm{\&gamma;}+\mathrm{\&beta;})}^l,i=\mathrm{0,1}\&CenterDot;\&CenterDot;\&CenterDot;N\\ {e_i=\underset{j=i}{\overset{N}{\mathrm{\&Sigma;}}}}_{}{a}_j^{\left(N\right)},i=\mathrm{0,1}\&CenterDot;\&CenterDot;\&CenterDot;N\end{array}\right.\&CenterDot;\&CenterDot;\&CenterDot;\left(12\right)$

Here C _l ⁱBe illustrated in the common combinations of getting the i kind in the l kind sample and count computing.Multi-user detector among the present invention can be write as following final expression formula thus:

$\hat{d}=\mathrm{\&alpha;}\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{e}_i{G}^i{y}_{\mathrm{MF}}\&CenterDot;\&CenterDot;\&CenterDot;\left(13\right)$

The present invention has the following advantages with respect to existing relevant multi-user detector:

1. with respect to linear multi-user detectors (comprising minimum Mean Square Error Linear multi-user detector reconciliation dependent linearity multi-user detector), the present invention is owing to avoided the inversion operation of correlation matrix cleverly, thereby reduce under the prerequisite of systematic function at a little, greatly reduce computing and enforcement complexity that real system is realized.

2. the multi-user detector that proposes with respect to Moshavi, the present invention is owing to adopted the expansion base multinomial of special tectonic and new high-performance simultaneously, low complex degree iteration weight coefficient computational algorithm, thereby the real-time online that can guarantee weight coefficient calculates, it is desired in advance according to user profile to have overcome the Moshavi method, carry out complex simulation, off-line is estimated the shortcoming of quantity of parameters.The present invention can reduce the time and the space complexity of real system keeping effectively improving the system implementation flexibility under the high performance prerequisite.

3. with respect to the multi-user detector based on the Taylor expansion of people such as Lei proposition, the present invention is owing to adopted the minimum principle optimal design expansion of spectral radius basic matrix G (spectral radius of ρ representing matrix R, G=I-α (R+ σ in the method for Lei on the one hand ²I), only require to choose arbitrarily $\mathrm{\&alpha;}\&Element;(0,\frac{2}{\mathrm{\&rho;}})$ , to specifically choosing of α without any optimizing); Adopted on the other hand and be different from the novel weighting method of deploying that Taylor launches fully, according to the mean error square minimum criteria optimal design of corresponding linear multi-user detectors (theoretical optimum detector) output weight coefficient, thereby under the prerequisite that keeps than low computational complexity, effectively quickened the convergence rate of expansion and and then improved systematic function.Fig. 5 has provided in the outdoor low speed mobile environment of 3-G (Generation Three mobile communication system) TD-SCDMA up link people such as using the present invention and Lei and has proposed the simulation performance contrast (simulation model adopt COST201 TU model) of method under multipath channel.Can see that the inventive method has surpassed the performance of Lei method under 30 rank iteration in the performance under the 10 rank iteration.And when the iterations of the inventive method reached 15 times, performance had begun the approximation theory limit in the error rate.Prove that thus the present invention can improve systematic function with respect to traditional Lei method greatly under the prerequisite that keeps low implementation complexity.

Of the present invention hold to levy be:

To minimize spectral radius is that target obtains matrix multinomial expansion basic matrix G by optimal design; Design weight sequence e according to mean error square minimum for principle with corresponding linear multi-user detectors output ₀, e ₁E _NM signal vector right to use repeated order row e for the multi-stage iteration generation ₀, e ₁E _NBe weighted summation, thereby obtain the output result that final multi-user association detects.

Description of drawings

Fig. 1. existing linear multi-user detectors structure,

The multi-user detector structrual description figure that people such as Fig. 2 .Lei propose,

Fig. 3. the multi-user association detector arrangement figure based on the weighting multinomial expansion of the present invention,

Fig. 4. algorithm implementing procedure figure of the present invention,

Fig. 5. people such as the present invention and Lei propose the simulation performance comparison diagram of method.

Embodiment

Hereinafter the implication of related letter please refer to above for inventive principle introduction partly in the formula.The specific embodiment of the present invention can be divided into following a few step:

1. the accuracy of detection that requires according to real system is determined to launch item number N, and carries out channel estimating and to the estimation of system's additive white noise energy, simultaneously by down-conversion, and filtering, the baseband equivalence signal y of circuit such as sampling generation different user.

2. use method mentioned above, utilize formula (3) (4) to come structural matrix A, B, y _MFAnd then, according to (8) formula calculate required for the present invention want be the specially designed expansion polynomial basis of accelerating ated test matrix G and intermediate variable α.

3. according to the weight coefficient after the each iteration of following algorithm computation

At first according to (8) (11) generator polynomial P (z), the different powers time coefficient of correspondence of getting multinomial P (z) then obtains sequence a according to order from low to high ₀ ^(N), a ₁ ^(N)... a _N ^(N), further obtain the required weight coefficient of weighted sum by (12) at last

$e_i=\underset{j=i}{\overset{N}{\mathrm{\&Sigma;}}}{a}_j^{\left(N\right)},i=\mathrm{0,1}\&CenterDot;\&CenterDot;\&CenterDot;N$

4. Reference Design figure (3) has provided the hardware circuit schematic diagram of concrete enforcement algorithm of the present invention, reference diagram (4) has provided corresponding software realization flow figure, according to figure (3) (4) the basic matrix G that calculates in the step 2 is used for the loop iteration processing unit, generates signal phasor group (α y as the signal processing matrix of iteration each time _MF, α Gy _MF... α G ^Ny _MF)

5. utilize the weight coefficient that obtains in the step 3, the M signal vector that generates for step 4 is weighted summation, and the result of final output can describe with following formula:

$\hat{d}={\mathrm{\&alpha;y}}_{\mathrm{MF}}\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{e}_i{G}^i$

With the application of the present invention in 3-G (Generation Three mobile communication system) TD-SCDMA up link, specify this novel high-performance, the low-complexity multi-user detection method below based on the weighting multinomial expansion.Suppose that having 8 users in the sub-district of TD-SCDMA system communicates simultaneously, 8 the quadrature Hadamard sequences that every user uses separately that a length is is used c here as spreading code _q ^(k), q=1 ... Q, k=1 ... K, Q=8, K=8 represents.The data symbol number of each user's one frame transmission is N=22, and uses the QPSK mode to modulate.The radiofrequency signal that receiving terminal receives is amplified in process, down-conversion, and filtering finally is converted into baseband digital signal after a series of processing such as base band moulding and sampling.At this moment, at first carry out channel estimating, obtain the channel impulse response sequences h of different user _w ^k, k=1,2 ... K, w=1,2 ... W.(distinguishable multipath number W=8 under the application scenario of this TD-SCDMA) calculates each user's channel impulse response h on this basis _w ^kWith corresponding frequency expansion sequence c _q ^(k)Convolution, to obtain the mixed channel response of different user correspondence $b_i^{\left(k\right)}=c_i^{\left(k\right)}*h_i^{\left(k\right)},$ K=1,2 ... k, i=1,2 ... W+Q-1, (4) formula of utilization is come structural matrix A.By using (8) formula, obtain iterative processing matrix G; By using (8) and (11) (12) to obtain iteration weight coefficient e ₀, e ₁E _N

With the equivalent baseband signal vector y that receives at first by receiving matched filter banks to obtain through the signal phasor y after the matched filtering _MF=A ^HY is then with y _MFVector y in the middle of input iterative processing unit generates _MF, α Gy _MF... α G ^Ny _MFThe output signal vector of iterative processing unit (is comprised y _MF), according to the weight coefficient e that calculates in advance ₀, e ₁E _NBe weighted summation, finally export the result thereby obtain multi-user detector involved in the present invention.

The concrete digital signal processor embodiment of realization the inventive method has multiple, and introduce a kind of general digital signal processing (DSP) chip implementation of being convenient to practical application here.Concrete steps are as follows:

The first step: select to be suitable for realizing the dsp chip of the inventive method, the TS201 chip that we have selected U.S. ADI company to produce in the TD-SCDMA engineering practice.

Second step: supporting TS201 chip development and the debugging acid of buying ADI company built the system hardware platform.

The 3rd step: according to the software and hardware structure that Reference Design figure (3) and (4) provide, use C or assembler language or other integrated circuit design tool are finished the exploitation of wanting required for the present invention according to the step that provides in the embodiment.

For highly skilled Electronics Engineer, the present invention can have multiple variation and modification in actual applications, for example will directly ask weight coefficient to change into and use certain alternative manner etc., but these can not deviate from the spirit and scope of the present invention.Like this, have only improvement and variation and claim equivalent or in its scope, these improvement and to change all be that the intent of the present invention is included.

Claims (1)

1. based on the DS-CDMA system multi-user detection method of weighting expansion, it is characterized in that it is realized according to following steps in DSP or FPGH chip:

The received signal of step 1. receiver output is passed through down-conversion successively, and molding filtration obtains baseband signal y behind each circuit of sampling;

Step 2. is the baseband signal that obtains in the step 1, the channel impulse response of the different user that records in advance and frequency expansion sequence are sent into the first multiplicaton addition unit circuit, the white noise power that records is in advance sent into the 3rd multiplicaton addition unit circuit, by first, the second and the 3rd multiplicaton addition unit circuit generates M signal vector and matrix according to following steps more successively:

Step 2.1 first is taken advantage of and is added circuit the frequency expansion sequence of different user and corresponding user's impulse response are carried out convolution to obtain each user's mixed channel response sequence, and its process is as follows:

b ^(k)＝c ^(k)*h ^(k)， k＝1，2…K

$=(b_1^{\left(k\right)},b_2^{\left(k\right)},\&CenterDot;\&CenterDot;\&CenterDot;b_{W+Q-1}^{\left(k\right)})$

C wherein ^(k)Be the general formula of the frequency expansion sequence of expression different user:

The user 1: $C^{\left(1\right)}=(c_1^{\left(1\right)},c_2^{\left(1\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(1\right)})$

The user 2: $C^{\left(2\right)}=(c_1^{\left(2\right)},c_2^{\left(2\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(2\right)})$



User K: $C^{\left(K\right)}=(c_1^{\left(K\right)},c_2^{\left(K\right)},\&CenterDot;\&CenterDot;\&CenterDot;c_Q^{\left(K\right)})$

K representative of consumer number, Q represents the length of frequency expansion sequence;

h ^(k)Be the general formula of the channel impulse response sequence of different user, be expressed as follows

The user 1: $h^{\left(1\right)}=(h_1^{\left(1\right)},h_2^{\left(1\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h_W^{\left(1\right)})$

The user 2: $h^{\left(2\right)}=(h_1^{\left(2\right)},h_2^{\left(2\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h_W^{\left(2\right)})$



User K: $h^{\left(K\right)}=(h_1^{\left(K\right)},h_2^{\left(K\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h_W^{\left(K\right)}),$

Subscript W represents the length of channel impulse response sequence;

Each user's that step 2.2 obtains step 2.1 mixed channel response sequence b ^(k), k=1,2 ... K generates comprehensive channel response matrix A according to following formula

Here M _sRepresent the data number of each user's one-time detection;

Step 2.3 at the first multiplicaton addition unit circuit according to step 2.2 generator matrix R

R＝A ^HA

A ^HIt is the transpose conjugate matrix that step 2.2 generates comprehensive channel response matrix A;

The above-mentioned first multiplicaton addition unit circuit of step 2.4 is delivered to the comprehensive channel response matrix A that is generated the transpose conjugate matrix A of calculating A in the second multiplicaton addition unit circuit ^H, and obtain the received signal vector y that the process matched filtering is operated thus _MF, represent with following formula:

y _MF＝A ^Hy；

The above-mentioned first multiplicaton addition unit circuit of step 2.5 is formed matrix A, R and the white noise power that records are in advance delivered to the 3rd multiplicaton addition unit circuit, in described the 3rd multiplicaton addition unit circuit,, generate by following steps successively and launch basic matrix G according to launching the minimum principle of basic matrix G spectral radius:

$\left\{\begin{array}{c}{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)=\max \left\{\min \right\{r_{t,s}-\underset{s=1,s\&NotEqual;1}{\overset{m}{\mathrm{\&Sigma;}}}\left|r_{t,s}\right|\},0\}+{\mathrm{\&sigma;}}^{2}t\&Element;(1,m)\\ {\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)=\max \left\{\underset{s=1}{\overset{m}{\mathrm{\&Sigma;}}}\right|r_{t,s}\left|\right\}+{\mathrm{\&sigma;}}^{2}t\&Element;(1,m)\\ \mathrm{\&alpha;}=\frac{2}{{\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)+{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)}\\ G=I-\mathrm{\&alpha;}(R+{\mathrm{\&sigma;}}^{2}I)\end{array}\right.$

R wherein _{T, s}For the t of correlation matrix R is capable, the element of s row, m is the dimension of R, I is a m dimension unit matrix, σ ²Be the power of additive white noise, ${\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)$ With ${\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)$ For for matrix R+ σ ²The minimum of I and the estimation of eigenvalue of maximum estimate directly to make σ if can not carry out noise power for the simplified system design ²=0;

The matrix A that step 3. generates the above-mentioned first multiplicaton addition unit circuit, the maximum that R and the 3rd multiplicaton addition unit circuit generate descend characteristic value to estimate most ${\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I)$ $,{\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)$ And the white noise power σ that records in advance ²Be input to the 4th multiplicaton addition unit circuit, by calling the N rank Chebyshey multinomial coefficient t that arranges to low order by high order that is stored in advance in the ROM circuit _N, t _N-1T ₀, generate iteration weight coefficient e according to following steps ₀, e ₁E _N:

$1.f=-1,g=1-2{\mathrm{\&sigma;}}^2/({\widehat{\mathrm{\&lambda;}}}_{\min }(R+{\mathrm{\&sigma;}}^{2}I)+{\widehat{\mathrm{\&lambda;}}}_{\max }(R+{\mathrm{\&sigma;}}^{2}I))$

2.γ＝2/(g-f)，β＝-(g+f)/(g-f)

$3.a_i^{\left(N\right)}={\mathrm{\&Sigma;}}_{l=i}^N{t}_l{C}_l^i{\mathrm{\&gamma;}}^i{\mathrm{\&beta;}}^{(l-i)}/{\mathrm{\&Sigma;}}_{l=0}^N{t}_l{(\mathrm{\&gamma;}+\mathrm{\&beta;})}^l,i=\mathrm{0,1}\&CenterDot;\&CenterDot;\&CenterDot;N$

$4.e_i=\underset{j=i}{\overset{N}{\mathrm{\&Sigma;}}}{a}_j^{\left(N\right)},i=\mathrm{0,1}\&CenterDot;\&CenterDot;\&CenterDot;N$

Here C _l ⁱBe illustrated in the number of combinations computing of getting the i kind in the l kind sample, the iterations that the N representative is determined according to actual requirement of engineering is generally weighed implementation complexity and accuracy of detection, and N desirable about 10;

Step 4. is with the signal phasor y through matched filtering of second multiplicaton addition unit circuit output _MFThe expansion basic matrix G that generates with the 3rd multiplicaton addition unit circuit, and intermediate variable α imports the 5th multiplicaton addition unit circuit, by iteration phase multiplication, generates signal phasor group α y _MF, α Gy _MF... α G ^Ny _MF

The weight coefficient e of step 5. by the 4th multiplicaton addition unit circuit is generated ₀, e ₁E _NSignal phasor group α y with the generation of the 5th multiplicaton addition unit circuit _MF, α Gy _MF... α G ^Ny _MFSend into the 6th multiplicaton addition unit circuit, the signal phasor that generates for each iteration use the corresponding weights coefficient be weighted summation can the amount of obtaining the multi-user detector output vector that launches based on weighting at end:

$\hat{d}=\mathrm{\&alpha;}\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{e}_i{G}^i{y}_{\mathrm{MF}}.$

Images