CN1885727B - Simplified zero-forcing equalization filter calculating method - Google Patents

Abstract

The present invention relates to two short-cut calculation methods of zero-forcing equalization filter (ZF-BLE) used in a communication receiver. In TD-SCDMA communication system, with the first method, a simplified system matrix B is calculated so that a correlation matrix B<H>B of B and an inverse matrix (B<H>B)<-1> of B<H>B are calculated, then (B<H>B)<-1> is used for taking MAI balance directly to data MF through matched filtering and realignment for obtaining estimated data d. With the second method, the simplified matrix B is used to calculate ara; an inverse ara<-1> of ara is directly calculated; partitioned matrix a, b, c are used for taking ISI and MAI balance directly to the data MF through matched filtering and realignment, thus adding to obtain estimated data. The simplified calculation methods can reduce calculated amount and accelerate calculation speed.

Description

The zero-forcing equalization filter calculating method of simplifying

Technical field

The present invention relates to the simplification computational methods of the zero-forcing equalization filter (ZF-BLE) that uses in two kinds of communication control processors, relate in particular to two kinds of improved ZF-BLE quick calculation methods in TD-SCDMA receiver or other communication control processors.

Background technology

In communication system, the recipient will carry out filtering to received signal, and filter out noise is disturbed, and recovers the data of transmission.Zero-forcing equalization filter (ZF-BLE) is exactly a kind of method that can select.The TD-SCDMA system has used ZF-BLE, and in order to realize the ZF-BLE process, receiver has passed through the following processing:

Carry out separating of data division and training sequence part to received signal, and the combined channel of selecting training sequence part to carry out the multi-user is estimated.

Utilize multi-user's combined channel response, the tectonic system matrix.Sytem matrix reflected the process that multi-user's information source experiences from the transmitting terminal to the receiving terminal, comprised the transmission of spectrum-spreading and scrambling, wireless channel.

Implement the zero-forcing equalization filter algorithm, computing formula:

$\hat{d}{|}_{R_n={\mathrm{\&sigma;}}^{2}I}={\left(A^{H}A\right)}^{-1}{A}^{H}y$

Expression transmits the estimation of data, is the column vector of [NK * 1], and N represents each user's data quantity, and K represents number of users;

R _n=σ ²I is the covariance matrix of multi-user's white noise, is [K * K] matrix;

A is a sytem matrix, and it is [(LN+W-1) * NK] matrix, and L represents scrambler length, and W represents that window is long;

Y represents received signal, under one times of spreading rate situation, is the column vector of [NK * 1]; H represents conjugate transpose;

A ^HA is the covariance matrix of sytem matrix, and it is [NK * NK] matrix.

A in the formula ^HY represents matched filtering, can be correlated with slip and realize.(A ^HA) ^-1The expression equalizer need be found the solution [NK * NK] inverse of a matrix, and general method is to use lower triangular matrix to decompose, and the method for nonzero element recursion is inverted.Because [NK * NK] matrix size is bigger, make amount of calculation big, error is easily dispersed.

Because the existence of radio communication channel noise and interference makes channel response estimate to have comprised error, causes sytem matrix A distortion.If use the method for recursion to invert, the signal to noise ratio of the signal processing of back sequential signal is further worsened, the error rate increases.

Summary of the invention

The purpose of this invention is to provide two kinds of simple ZF-BLE computational methods, these two kinds of methods are that effectively amount of calculation is little, improved ZF-BLE computational methods.Can reduce amount of calculation, accelerate computational speed, reduce equipment amount, reduce the receiver cost.This technology both can be applied in system's one side, also can be applied in travelling carriage one side.

The present invention proposes the computational methods of two kinds of simplification,

Method one is carried out equilibrium to MAI, abandons the equilibrium to ISI.

This method comprises following 6 steps:

Each user's of step 1. channel response h _kScrambler channel code ScrambleOVSF with each user _kConvolution obtains B _kVector:

B _k＝h _k

ScrambleOVSF _k

B _kIt is [(L+W-1) * 1] several column vector;

Step 2.K user's B _kBe arranged in the new sytem matrix B of simplification, be [(L+W-1) * and K] matrix;

Step 3. is asked the covariance matrix B of B matrix ^HB is [K * K] matrix;

Step 4. is asked matrix B ^HContrary (the B of B ^HB) ^-1, be [K * K] matrix;

[NK * 1] column vector A after the step 5. pair matched filtering ^HY rearranges:

MF＝reshape(A ^Hy，K，N)

It is [K * N] matrix, and reshape represents to rearrange;

Step 6. is implemented balanced,

$\hat{d}{|}_{R_n={\mathrm{\&sigma;}}^{2}I}={\left(B^{H}B\right)}^{-1}\mathrm{MF}.$

Method two carries out equilibrium simultaneously to MAI and ISI.

This method comprises following 8 steps:

Each user's of step 1. channel response h _kScrambler channel code ScrambleOVSF with each user _kConvolution obtains B _kVector:

B _k＝h _k ScrambleOVSF _k，

B _kIt is [(L+W-1) * 1] several column vector;

Step 2.K user's B _kBe arranged in the sytem matrix B of simplification, it is [(L+W-1) * K] matrix;

Step 3. pair matrix B is decomposed:

B _u＝B(1：W-1，1：K)，

B _d＝B(L+1：L+W-1，1：K)，

Wherein: the arrangement of representing matrix row, column;

Step 4. directly calculates A ^HMatrix in block form ara among the A, it is [2K * 2K] matrix

$\mathrm{ara}=\left[\begin{array}{cc}{B}^{H}B& {B_d}^H{B}_u\\ {B_u}^H{B}_d& B^{H}B\end{array}\right];$

Step 5. is calculated the inverse matrix ara of ara ^-1, be [2K * 2K] matrix in block form,

${\mathrm{ara}}^{-1}=\left[\begin{array}{cc}a& b\\ c& a\end{array}\right];$

[NK * 1] column vector A after the step 6. pair matched filtering ^HY rearranges:

MF＝reshape(A ^Hy，K，N)，

It is [K * N] matrix, and reshape represents to rearrange;

The concrete computational methods of step 7. equilibrium are, remove multiply by data M F respectively with matrix a, b, c,

${\hat{d}}_a=\mathrm{aMF},$

${\hat{d}}_b=\mathrm{bMF},$

${\hat{d}}_c=\mathrm{cMF},$

It all is [K * N] matrix;

Step 8.

The 1st row put 0, obtain N row put 0, obtain Then With The respective items addition obtains the estimation of data

$\hat{d}={\hat{d}}_a+{\hat{d}}_{\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="S06189637820060728E000031.png,S06189637820060728E000041.png"\; state="\{\{state\}\}">b</figure-callout>}}^0+{\hat{d}}_{\mathrm{<figure-callout\; id="0"\; label="c"\; filenames="S06189637820060728E000031.png,S06189637820060728E000041.png"\; state="\{\{state\}\}">c</figure-callout>}}^0.$

Description of drawings

Fig. 1 is the position view of zero-forcing equalization filter 110 in communication system;

Fig. 2 is the schematic diagram of the zero-forcing equalization filter in first method 210 of the present invention and the second method 410;

Fig. 3 is first method 210 related matrixes of the present invention and transformation relation schematic diagram;

Fig. 4 is second method 410 related matrixes of the present invention and transformation relation schematic diagram.

Embodiment

Principle of the present invention is at first described.The TD-SCDMA system has adopted zero-forcing equalization filter (ZF-BLE) associated detecting method, through A ^HSymbol after the y matched filtering will be by (A ^HA) ^-1ISI and MAI are eliminated in equilibrium.ISI is the cross interference of front and back symbol, is that the multipath interference causes, and MAI is a multi-user interference.The thinking of first method of the present invention is the equilibrium of abandoning less important interference ISI, only to main interference MAI equilibrium, does not use traditional sytem matrix A during concrete operations, but has defined the new sytem matrix B that simplifies, and its small scale has reduced amount of calculation.This method can be used in the suburb, and multipath disturbs the base station and the mobile phone of few environment.Second method of the present invention is to ISI and MAI interference carrying out simultaneously equilibrium.Used the matrix ara that simplifies ^-1, its small scale has reduced amount of calculation, but having brought certain error, these errors is the scopes of allowing.This method can be used in the city, and multipath disturbs the base station and the mobile phone of big environment.

Traditional method is with B matrix construction system matrix A, calculates the correlation matrix A of A ^HA at nonzero element in wherein, asks it contrary with recurrence method, and its shortcoming as described above.First method of the present invention is directly to calculate the contrary (B of the correlation matrix of B ^HB) ^-1Second method of the present invention is directly to calculate correlation matrix A ^HThe contrary ara of the matrix in block form ara of A ^-1

First method (B ^HB) ^-1After matched filtering, the data M F of the formula that rearranges (2) carries out the MAI equilibrium.Second method ara ^-1To MF, carry out ISI and MAI equilibrium.

Such processing mode has reduced amount of calculation.

Below in conjunction with accompanying drawing the present invention is elaborated.

First method of the present invention is at first described.Each user's channel response h _kScrambler channel code ScrambleOVSF with each user _kConvolution obtains B _kVector: B _k=h _k

ScrambleOVSF _k, B _kIt is [(L+W-1) * 1] several column vector.

Follow K user's B _kBe arranged in the new sytem matrix B of simplification, it is [(L+W-1) * K] matrix, and obtains the covariance matrix B of B matrix ^HB, it is [K * K] matrix.

In Fig. 3, correlation matrix B ^HB _Kron310, the covariance matrix B of the sytem matrix B of the simplification that forms during the input previous step is rapid ^HB arranges B by data volume N on diagonal ^HB forms matrix in block form B ^HB _KronBe formulated,

B ^HB _kron＝kron(eye(N)，B ^HB)，

It is [NK * NK] matrix, and eye is [N * N] unit matrix, and kron represents repeated arrangement, is the function of MATLAB language.This process is not concrete enforcement in the step of first method, but explanation B ^HB _KronQuite traditional sytem matrix A's, covariance A ^HA.

B ^HThe B device 320 of inverting is to the B of above-mentioned formation ^HThe B operation of inverting obtains (B ^HB) ^-1

Correlation matrix B ^HB _KronInverse matrix 330 because (B ^HB _Kron) ^-1=(B ^HB) ^-1 _KronCan on diagonal, arrange (B by data volume N ^HB) ^-1, form matrix in block form (B ^HB) ^-1 _KronBe formulated (B ^HB) ^-1 _Kron=kron (eye (N), (B ^HB) ^-1), it is [NK * NK] matrix.This process is not concrete enforcement in the step of method one, but explanation (B ^HB) ^-1 _KronQuite traditional covariance A ^HInverse matrix (the A of A ^HA) ^-1

By to (B ^HB) ^-1 _KronAnalysis, to [NK * 1] the column vector A after the matched filtering ^HY rearranges operation:

MF＝reshape(A ^Hy，K，N)，

Then, can directly implement equalization operation: $\hat{d}{|}_{R_n={\mathrm{\&sigma;}}^{2}I}={\left(B^{H}B\right)}^{-1}\mathrm{MF},$ Obtain final data estimator

Second method of the present invention is described below.[(L+W-1) * K] matrix B that obtains in the first method is decomposed: B _u=B (1:W-1,1:K), B _d=B (L+1:L+W-1,1:K), wherein: the arrangement of representing matrix row, column.

Among Fig. 4, the correlation matrix 510 of sytem matrix A directly calculates A ^HMatrix in block form ara among the A, $\mathrm{ara}=\left[\begin{array}{cc}{B}^{H}B& {B_d}^H{B}_u\\ {B_u}^H{B}_d& B^{H}B\end{array}\right],$ A ^HA is arranged on diagonal of a matrix by matrix in block form ara to constitute.A ^HA is not concrete calculating in the step of second method, but is used for illustrating traditional sytem matrix A, covariance A ^HThe formation of A.

Invert device 520 directly to the ara operation of inverting, obtain ara ^-1It also is a matrix in block form, constitutes by matrix in block form a, b, c, ${\mathrm{ara}}^{-1}=\left[\begin{array}{cc}a& b\\ c& a\end{array}\right].$

The A that simplifies ^HThe inverse matrix 530 of A according to each user's data quantity N, is arranged matrix in block form ara on the diagonal of matrix ^-1, the equalizer matrix (A that obtains simplifying ^HA) ^-1 _KronThis process is not concrete enforcement in the step of method two, but explanation (A ^HA) ^-1 _KronQuite traditional covariance A ^HInverse matrix (the A of A ^HA) ^-1

By to (A ^HA) ^-1 _KronAnalysis, can utilize little matrix in block form a, b, c directly the later data M F of matched filtering to be carried out equalization operation, ${\hat{d}}_a=\mathrm{aMF},$ ${\hat{d}}_b=\mathrm{bMF},$ ${\hat{d}}_c=\mathrm{cMF},$

It all is [K * N] matrix.Then, The 1st row put 0, obtain

N row put 0, obtain Then

With

The respective items addition obtains the estimation of data

$\hat{d}={\hat{d}}_a+{\hat{d}}_{\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="S06189637820060728E000031.png,S06189637820060728E000041.png"\; state="\{\{state\}\}">b</figure-callout>}}^0+{\hat{d}}_c^0,$ Thereby obtain final data estimator

The above two kinds of shortcut calculation, general Communication System Engineer just can realize in equipment smoothly.

Claims (2)

1. a zero-forcing equalization filter is simplified Calculation Method, comprising:

Each user's of step 1. channel response h _kScrambler channel code ScrambleOVSF with each user _kConvolution obtains B _kVector:

$B_k=h_k\&CircleTimes;\mathrm{Scramble}{\mathrm{OVSF}}_k,$

B _kIt is [(L+W-1) * 1] several column vector;

Step 2.K user's B _kBe arranged in the sytem matrix B of simplification, it is [(L+W-1) * K] matrix;

Step 3. is asked the covariance matrix B of B matrix ^HB, it is [K * K] matrix;

Step 4. is asked matrix B ^HContrary (the B of B ^HB) ^-1, be [K * K] matrix;

[NK * 1] column vector A after the step 5. pair matched filtering ^HY rearranges:

MF＝reshape(A ^Hy，K，N)，

It is [K * N] matrix, and reshape represents to rearrange;

Step 6. is implemented balanced,

${\hat{d}|}_{R_n={\mathrm{\&sigma;}}^{2}I}={\left(B^{H}B\right)}^{-1}\mathrm{MF},$

A is a sytem matrix, and it is [(LN+W-1) * NK] matrix, and y represents received signal, under one times of spreading rate situation, is the column vector of [NK * 1], A ^HA is the covariance matrix of sytem matrix, and it is [NK * NK] matrix, A in the formula ^HY represents matched filtering,

Expression transmits the estimation of data, is the column vector of [NK * 1], R _n=σ ²I is the covariance matrix of multi-user's white noise, is [K * K] matrix,

N represents each user's data quantity, and K represents number of users, and L represents scrambler length, and W represents that window is long, and H represents conjugate transpose.

2. a zero-forcing equalization filter is simplified computational methods, comprising:

Each user's of step 1. channel response h _kScrambler channel code ScrambleOVSF with each user _kConvolution obtains B _kVector:

$B_k=h_k\&CircleTimes;\mathrm{Scramble}{\mathrm{OVSF}}_k,$

B _kIt is [(L+W-1) * 1] several column vector;

Step 2.K user's B _kBe arranged in the sytem matrix B of simplification, it is [(L+W-1) * K] matrix;

Step 3. pair matrix B is decomposed:

B _u＝B(1∶W-1，1∶K)，

B _d＝B(L+1∶L+W-1，1∶K)，

Wherein: the arrangement of representing matrix row, column;

Step 4. directly calculates A ^HMatrix in block form ara among the A, it is [2K * 2K] matrix

$\mathrm{ara}=\left[\begin{array}{cc}{B}^{H}B& {B_d}^H{B}_u\\ {B_u}^H{B}_d& B^{H}B\end{array}\right],$

A is a sytem matrix, and it is [(LN+W-1) * NK] matrix, A ^HA is the covariance matrix of sytem matrix, and it is [NK * NK] matrix;

Step 5. is calculated the inverse matrix ara of ara ^-1, be [2K * 2K] matrix in block form,

${\mathrm{ara}}^{-1}=\left[\begin{array}{cc}a& b\\ c& a\end{array}\right],$

A, b, c represent matrix in block form respectively;

[NK * 1] column vector A after the step 6. pair matched filtering ^HY rearranges:

MF＝reshape(A ^Hy，K，N)，

It is [K * N] matrix, and reshape represents to rearrange,

Y represents received signal, under one times of spreading rate situation, is the column vector of [NK * 1], A in the formula ^HY represents matched filtering;

The concrete computational methods of step 7. equilibrium are, remove multiply by data M F respectively with matrix a, b, c,

${\hat{d}}_a=\mathrm{aMF},$

${\hat{d}}_b=\mathrm{bMF},$

${\hat{d}}_c=\mathrm{cMF},$

It all is [K * N] matrix;

Step 8.

The 1st row put 0, obtain

N row put 0, obtain

Then With

The respective items addition obtains the estimation of data

$\hat{d}={\hat{d}}_a+{\hat{d}}_b^0+{\hat{d}}_c^0,$

N represents each user's data quantity, and K represents number of users, and L represents scrambler length, and W represents that window is long, and H represents conjugate transpose.

Images