CN101753169B - Method for equalizing ordered block decision feedback in TD-SCDMA - Google Patents

Abstract

The invention discloses a method for equalizing ordered block decision feedback in TD-SCDMA and comprises the following steps: users are arranged by sequence according to the magnitude of the spreading factors when the spreading factors of the users are different, and the combined channel impulse responses (CCIR) of the users with large spreading factors are arranged behind and are reflected in matrix A to conduct block decision feedback equalization, zero forcing block decision feedback equalization (ZF-BDFE) and minimum mean square block error block decision feedback equalization (MMSE-BDFE). Through arranging the users by sequence according to the magnitude of the spreading factors, since the CCIR of the users with large spreading factors are arranged behind and are reflected in the matrix A to conduct common block decision feedback equalization (ZF-BDRE and MMSE-BDFE), the decision gain is improved.

Description

The piece Decision-Feedback Equalization that sorts among a kind of TD-SCDMA

Technical field

The invention belongs to a kind of optimization implementation method of associated detection technique among the TD-SCDMA, in particular the channel equalization in a kind of TD-SCDMA communication and the Multiuser Detection processing method of cdma system.

Background technology

In the prior art, associated detection technique mainly for be the such piece of TD-SCDMA transmission (being also referred to as burst mode) system.Block transmission system has its distinctive matrix form equalization methods, for the TD-SCDMA system equalization methods and the Data Detection of matrix form is combined, Here it is so-called joint detection algorithm.Associated detection technique can overcome near-far interference, can alleviate again the intersymbol interference (ISI, Inter Symbol Interference) and the multiple access that are produced by multipath channel and disturb (MAI, Multiple-access interference).

Cdma system can be divided into asynchronous CDMA and synchronization CDMA.Asynchronous CDMA generally can not guarantee between the spreading code in the orthogonality that does not line up in the situation, therefore just has in essence MAI.Spreading code in the synchronous CDMA system generally is quadrature, selects to there is no ISI, also without MAI in the Selective Fading Channel in non-frequency.But select in the Selective Fading Channel in frequency, the signal of a plurality of propagated has produced ISI, and has destroyed the orthogonality between the spreading code, has caused MAI.The channel that experiences in the broadband system reality is frequency selective fading, so ISI and MAI are the principal elements that affects the performance of wideband CDMA.

Comprised equalization methods in the associated detection technique, its linear algorithm generally can be divided into based on ZF linear block balance (ZF-BLE) with based on two kinds of methods of minimum Mean Square Error Linear block balance (MMSE-BLE).

Usually, the impulse response length of the channel of supposing is no more than 16 chips, and therefore the data of a upper time slot can not affect the data of next time slot.

Suppose that the data that k user sends are $d^{\left(k\right)}={(d_1^{\left(k\right)},d_2^{\left(k\right)},...,d_N^{\left(k\right)})}^T.$ Code after k user's frequency expansion sequence, channel code specify multiplier (Channelisation Code Specific Multiplier), scrambler compound is called spreading code.In order to oversimplify, the spreading code length of supposing K user here all is Q.K user's spreading code is $c^{\left(k\right)}={(c_1^{\left(k\right)},c_2^{\left(k\right)},..,c_Q^{\left(k\right)})}^T.$ K user's channel impulse response is $h^{\left(k\right)}={(h_1^{\left(k\right)},h_2^{\left(k\right)},...,h_W^{\left(k\right)})}^T.$

Obtain k user's spreading code c ^(k)And channel impulse response h ^(k)Carrying out the channel impulse response (Combined Channel Impulse Response, CCIR) that convolution obtains making up is

$b^{\left(k\right)}={(b_1^{\left(k\right)},b_2^{\left(k\right)},...,b_{Q+W-1}^{\left(k\right)})}^T=c^{\left(k\right)}*h^{\left(k\right)}$

b ^(k)Can be arranged in a matrix V:

$V=\left[\begin{array}{ccc}{b}_1^{\left(1\right)}& & b_1^{\left(K\right)}\\ \·& & \·\\ \·& & \·\\ \·& & \·\\ b_Q^{\left(1\right)}& \·\·\·& b_Q^{\left(K\right)}\\ b_{Q+1}^{\left(1\right)}& & b_{Q+1}^{\left(K\right)}\\ \·& & \·\\ \·& & \·\\ \·& & \·\\ b_{Q+W-1}^{\left(1\right)}& & b_{Q+W-1}^{\left(K\right)}\end{array}\right]$

Matrix V generator matrix A:

Like this, the chip sequence that receives of receiver can be expressed as

e＝Ad+n

D=(d wherein ₁, d ₂..., d _N) ^T∈ C ^{KN * 1}The cascade of the vector that becomes for n data symbols of all users, $d_n={(d_n^{\left(1\right)},d_n^{\left(2\right)},...,d_n^{\left(K\right)})}^T$ Be the vector that n data symbols of all users becomes, e=e ⁽¹⁾+ e ⁽²⁾+ ...+e ^(K)∈ C ^{(NQ+W-1) * 1}Be the data that all users that receive mix, n=n ⁽¹⁾+ n ⁽²⁾+ ...+n ^(K)∈ C ^{(NQ+W-1) * 1}For the noise of all users' channel and vector.

Can be clearly seen that by matrix A MAI (multiple access interference) produces reason, the orthogonality between the column vector in piecemeal is destroyed, has introduced MAI.ISI (intersymbol interference) is subject to the interference between each chip after the channel convolution, as long as all this interference can be arranged through the signal of multipath channel; There is the length of W-1 to interfere with each other between the piecemeal, introduced the interference between the spread symbol.

Associated detection technique based on BLE (piece linear equalization) comprises ZF linear equalization and the balanced two kinds of algorithms of minimum Mean Square Error Linear:

(1) ZF linear block balance (ZF-BLE):

The purpose of ZF-BLE is to make ${(e-A\hat{d})}^H\·R_n^{-1}\·(e-A\hat{d})$ Minimize.

Therefore the data that detect are:

${\hat{d}}_{\mathrm{ZF}-\mathrm{BLE}}={\left(A^H{R}_n^{-1}A\right)}^{-1}{A}^H{R}_n^{-1}e=d+{\left(A^H{R}_n^{-1}A\right)}^{-1}{A}^H{R}_n^{-1}n$

R wherein _nThe covariance matrix of n, when n is white Gaussian noise, R _n=σ ²I, following formula is reduced to:

${\hat{d}}_{\mathrm{ZF}-\mathrm{BLE}}={\left(A^{H}A\right)}^{-1}{A}^{H}e$

This shows that ZF-BLE has eliminated MAI and ISI fully, but caused the amplification of noise.

(2) linear equalization of least mean-square error

The purpose of MMSE-BLE is to make $E\left({(d-{\hat{d}}_{\mathrm{MMSE}-\mathrm{BLE}})}^H(d-{\hat{d}}_{\mathrm{MMSE}-\mathrm{BLE}})\right)$ Minimize.

Therefore the data that detect are

${\hat{d}}_{\mathrm{MMSE}-\mathrm{BLE}}={(A^H{R}_n^{-1}A+R_d^{-1})}^{-1}{A}^H{R}_n^{-1}e$

$=\mathrm{diag}\left(W_0\right)d+\stackrel{\‾}{\mathrm{diag}\left(W_0\right)}d+{(A^H{R}_n^{-1}A+R_d^{-1})}^{-1}{A}^H{R}_n^{-1}n$

Wherein $W_0={(I+{\left(R_d{A}^H{R}_n^{-1}A\right)}^{-1})}^{-1},$ First symbol for expectation in this formula, second is MAI and ISI, contrast ZF-BLE, MAI and ISI are not eliminated, but the amplification of noise is suppressed.When n is white Gaussian noise, R _n=σ ²I, and work as R _dDuring=I, following formula is reduced to:

${\hat{d}}_{\mathrm{MMSE}-\mathrm{BLE}}={(A^{H}A+{\mathrm{\σ}}^{2}I)}^{-1}{A}^{H}e$

In ZF-BLE and MMSE-BLE, the step of inverting of a large-sized Toeplitz matrix is arranged, often adopt some fast algorithms to invert avoiding.It is exactly a kind of method of avoiding matrix inversion that Cholesky decomposes:

$A^H{R}_n^{-1}A={\left(\mathrm{\ΣH}\right)}^H\mathrm{\ΣH}$

Wherein H is that diagonal element is 1 upper triangular matrix, and ∑ is diagonal matrix.

The processing of ZF-BLE just becomes like this

${\hat{d}}_{\mathrm{ZF}-\mathrm{BLE}}={\left(\mathrm{\ΣH}\right)}^{-1}{\left(H^H\mathrm{\Σ}\right)}^{-1}{A}^H{R}_n^{-1}e$

Definition

$z=\mathrm{\ΣH}{\hat{d}}_{\mathrm{ZF}-\mathrm{BLE}}$

Then

H ^H∑z＝y

Wherein $y=A^H{R}_n^{-1}e.$

Because H ^H∑ is lower triangular matrix, therefore can solve z by forward substitution.Because ∑ H is upper triangular matrix, therefore can solve by backward replacement

The principle that BDFE (piece decision feedback equalization) realizes is that backward innovation is become the decision-feedback mode.At first see the processing procedure of ZF-BDFE (ZF-piece decision feedback equalization), definition

$z^{\′}=H{\hat{d}}_{\mathrm{ZF}-\mathrm{BLE}}$

Then

H ^H∑∑z′＝y

Substitution in the expression formula of z '

Expression formula, obtain

$z^{\′}=d+(H-I)d+{\mathrm{\Σ}}^{-1}{\mathrm{\Σ}}^{-1}{\left(H^H\right)}^{-1}{A}^H{R}_n^{-1}n$

Therefore can solve z ' by forward substitution, then by deducting the item about decision-feedback Obtain detecting data, namely

${\hat{d}}_{\mathrm{ZF}-\mathrm{BDFE},\mathrm{KN}}=Q\left\{z_{\mathrm{KN}}^{\′}\right\}$

$d_{\mathrm{ZF}-\mathrm{BDFE},\mathrm{KN}-j^{\′}}=Q\{z_{\mathrm{KN}-j^{\′})}^{\′}-\underset{j^{\′\′}=1}{\overset{j^{\′}}{\mathrm{\Σ}}}{[H-I]}_{\mathrm{KN}-j^{\′},\mathrm{KN}-j^{\′}+j^{\′\′}}{\hat{d}}_{\mathrm{ZF}-\mathrm{BDFE},\mathrm{KN}-j^{\′}+j^{\′\′}}\}$ j′＝1，...，(KN-1)

Wherein Q{} is a judgement operator, and is relevant with the planisphere of modulation symbol.

Because decision-feedback has the judgement gain, so the performance of ZF-BDFE is better than ZF-BLE.

Similarly, can transform MMSE-BLE as MMSE-BDFE, as long as with the A among the ZF-BDFE ^HR _n ^-1A replaces with A _HR ^N-1A+R _d ^-1Carrying out the Cholesky decomposition gets final product.About ZF-BLE, MMSE-BLE, detailed derivation and the analysis of ZF-BDFE and MMSE-BDFE are known by prior art, no longer illustrate at this.

Can find out, decision-feedback is actually a from back to front process of feedback, the symbol of namely having adjudicated

Sequence number along with j ' reduces from 1 to KN-1.Therefore, the symbol that decision error is less more is arranged in front, can be so that little decision error be propagated far, and large decision error is propagated closely.But prior art does not have adjudicating the further technical development of gain raising, therefore has yet to be improved and developed.

Summary of the invention

The object of the present invention is to provide the piece Decision-Feedback Equalization that sorts among a kind of TD-SCDMA, at user's spreading factor not simultaneously, realize raising to gain by ordering.

Technical scheme of the present invention comprises:

The piece Decision-Feedback Equalization that sorts among a kind of TD-SCDMA, it may further comprise the steps:

A, at user's spreading factor not simultaneously sorts to the user by the size of spreading factor, after the user's that spreading factor is large aggregate channel impulse response CCIR comes, and is reflected in the matrix A:

Q wherein _Max=16 maximum spreading factors for system's support are 16 in TD-SCDMA; $K_v=\underset{k}{\mathrm{\Σ}}\frac{Q_{\max }}{Q_k}$ For Q _MaxEach user's symbolic number sum is called again empty code channel number in the individual chip; Q _kIt is the spreading factor of k the actual use of user; $N_{\min }=\frac{N_{\mathrm{chip}}}{Q_{\max }}$ Be 16 o'clock symbolic numbers in the data block for getting spreading factor, N _ChipBe the number of chips of a data block, be generally 352; K is the counting natural number;

Matrix A is generated by matrix V:

Wherein k user's spreading code is $c^{\left(k\right)}={(c_1^{\left(k\right)},c_2^{\left(k\right)},...,c_Q^{\left(k\right)})}^T,$ K user's channel impulse response is $h^{\left(k\right)}={(h_1^{\left(k\right)},h_2^{\left(k\right)},...,h_W^{\left(k\right)})}^T,$ Obtain k user's spreading code c ^(k)And channel impulse response h ^(k)Carrying out convolution obtains aggregate channel impulse response CCIR and is:

$b^{\left(k\right)}={(b_1^{\left(k\right)},b_2^{\left(k\right)},...,b_{Q+W-1}^{\left(k\right)})}^T=c^{\left(k\right)}*h^{\left(k\right)}$

By b ^(k)Be arranged in matrix V;

B, carry out the piece decision feedback equalization, the linear equalization MMSE-BDFE of ZF linear block balance ZF-BDFE and least mean-square error.

Described method wherein, also comprises among the described step B:

The chip sequence that receiver receives is expressed as

e＝Ad+n

Wherein $d={(d_1,{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="H2008102179519E00011.png,H2008102179519E00021.png"\; state="\{\{state\}\}">d</figure-callout>}}_2,...,d_{N_{\min }})}^T\&Element;C^{K_v{N}_{\min }\&times;1}$ The cascade of the vector that becomes for n data symbols of all empty code channels, $d_n={(d_n^{\left(1\right)},d_n^{\left(2\right)},...,d_n^{\left(K_v\right)})}^T$ The vector that becomes for n data symbols of all empty code channels, $e=e^{\left(1\right)}+e^{\left(2\right)}+...+e^{\left(K_v\right)}\&Element;C^{(N_{\min }{Q}_{\max }+W-1)\&times;1}$ The data of mixing for all empty code channels that receive, $n=n^{\left(1\right)}+n^{\left(2\right)}+...+n^{\left(K_v\right)}\&Element;C^{(N_{\min }{Q}_{\max }+W-1)\&times;1}$ For the noise of the channel of all empty code channels and vector.

The piece Decision-Feedback Equalization that sorts among a kind of TD-SCDMA provided by the present invention, the user is sorted by the spreading factor size by adopting, the user's that spreading factor is large CCIR comes rear and is reflected in the matrix A, process to carry out general piece decision feedback equalization (ZF-BDFE and MMSE-BDFE), realized the raising of judgement gain.

Description of drawings

Fig. 1 is the contrast effect schematic diagram of the inventive method and prior art, shows Q={16, and during 16,16,16,8,8,8,8}, unsorted Performance Ratio with ordering is illustrated;

Fig. 2 is the contrast effect schematic diagram of the inventive method and prior art, shows Q={16, and during 16,16,16,4,4}, unsorted Performance Ratio with ordering is illustrated.

Embodiment

Below in conjunction with accompanying drawing, will be described in more detail each preferred embodiment of the present invention.

The piece Decision-Feedback Equalization that sorts among the TD-SCDMA of the present invention, at the spreading factor of working as each user not simultaneously, by the spreading factor size user is sorted first, the user's that spreading factor is large CCIR comes rear and is reflected in the matrix A, then carries out general piece decision feedback equalization (ZF-BDFE and MMSE-BDFE).The decision error of the symbol that in general spreading factor is large is less, therefore works as each user's spreading factor not simultaneously, and it is to obtain gain that the judgement symbol is sorted.Other processing procedure is consistent with prior art, does not repeat them here.When each user's spreading factor was identical, the decision error of which symbol is little to be unknown for detector, so the inventive method will not be paid close attention to.

Sort block Decision-Feedback Equalization among the TD-SCDMA of the present invention, it may further comprise the steps:

A, at user's spreading factor not simultaneously sorts to the user by the size of spreading factor, after the user's that spreading factor is large aggregate channel impulse response CCIR comes, and is reflected in the matrix A:

Q wherein _Max=16 maximum spreading factors for system's support; $K_v=\underset{k}{\mathrm{\&Sigma;}}\frac{Q_{\max }}{Q_k}$ For Q _MaxEach user's symbolic number sum, i.e. empty code channel number in the individual chip; Q _kIt is the spreading factor of k the actual use of user; $N_{\min }=\frac{N_{\mathrm{chip}}}{Q_{\max }}$ Be 16 o'clock symbolic numbers in the data block for getting spreading factor, N _ChipIt is the number of chips of a data block; K is the counting natural number;

Matrix A is generated by matrix V:

Wherein, k user's spreading code is $c^{\left(k\right)}={(c_1^{\left(k\right)},c_2^{\left(k\right)},...,c_Q^{\left(k\right)})}^T,$ K user's channel impulse response is $h^{\left(k\right)}={(h_1^{\left(k\right)},h_2^{\left(k\right)},...,h_W^{\left(k\right)})}^T,$ Obtain k user's spreading code c ^(k)And channel impulse response h ^(k)Carrying out convolution obtains aggregate channel impulse response CCIR and is:

$b^{\left(k\right)}={(b_1^{\left(k\right)},b_2^{\left(k\right)},...,b_{Q+W-1}^{\left(k\right)})}^T=c^{\left(k\right)}*h^{\left(k\right)}$

By b ^(k)Be arranged in matrix V;

B, carry out the piece decision feedback equalization, the linear equalization MMSE-BDFE of ZF linear block balance ZF-BDFE and least mean-square error.

The chip sequence that receiver receives in the inventive method is expressed as

e＝Ad+n

D=(d wherein ₁, d ₂..., d _N) ^T∈ C ^{KN * 1}The cascade of the vector that becomes for n data symbols of all empty code channels, $d_n={(d_n^{\left(1\right)},d_n^{\left(2\right)},...,d_n^{\left(K\right)})}^T$ Be the vector that n data symbols of all empty code channels becomes, e=e ⁽¹⁾+ e ⁽²⁾+ ...+e ^(K)∈ C ^{(NQ+W-1) * 1}Be the data that all empty code channels that receive mix, n=n ⁽¹⁾+ n ⁽²⁾+ ...+n ^(K)∈ C ^{(NQ+W-1) * 1}For the noise of the channel of all empty code channels and vector.

Below with scheme and the beneficial effect thereof of specific embodiment explanation the inventive method.

For example when K=2, user 1 spreading factor is 16, and user 2 spreading factor is 8, then with b ⁽¹⁾Come b ⁽²⁾After, obtain matrix V:

W=4 wherein.Can obtain matrix A according to matrix V, then carry out general piece decision feedback equalization.

Effect after the below's checking the inventive method sorts by the spreading factor size to the data symbol.When being configured to K=8, when unsorted, K user's spreading factor is respectively Q={16,16,16,16,8,8,8,8}, and correspondingly, the data symbol number is N={22,22,22,22,44,44,44,44}.As shown in Figure 1, compare with the treatment effect after the ordering of the inventive method, the gain effect of the inventive method is obvious.

When being configured to K=6, as shown in Figure 2, when unsorted, K user's spreading factor is respectively Q={16,16,16,16,4,4}, and correspondingly, the data symbol number is N={22,22,22,22,88,88}.The inventive method is compared without ordering with prior art after sorting, and the gain effect of the inventive method is obvious.

Obviously be better than unsorted performance by the performance after Fig. 1 and the visible ordering of Fig. 2, when the spreading factor difference was larger between the user, the gain that ordering obtains was larger, and the inventive method is applied to have better technique effect in the associated detection technique of TD-SCDMA.

Should be understood that, above-mentioned description for preferred embodiment of the present invention is comparatively detailed, can not therefore think the restriction to scope of patent protection of the present invention, and scope of patent protection of the present invention should be as the criterion with claims.

Claims (2)

1. the piece Decision-Feedback Equalization that sorts among the TD-SCDMA, it may further comprise the steps:

A, at user's spreading factor not simultaneously sorts to the user by the size of spreading factor, after the user's that spreading factor is large aggregate channel impulse response CCIR comes, and is reflected in the following matrix A:

Q wherein _Max=16 maximum spreading factors for system's support; For Q _MaxEach user's symbolic number sum, i.e. empty code channel number in the individual chip; Q _kIt is the spreading factor of k the actual use of user;

Be 16 o'clock symbolic numbers in the data block for getting spreading factor, N _ChipIt is the number of chips of a data block; K is the counting natural number;

Matrix A is generated by matrix V:

Wherein, k user's spreading code is

K user's channel impulse response is

Obtain k user's spreading code c ^(k)And channel impulse response h ^(k)Carrying out convolution obtains aggregate channel impulse response CCIR and is:

$b^{\left(k\right)}={(b_1^{\left(k\right)},b_2^{\left(k\right)},...,b_{Q+W-1}^{\left(k\right)})}^T=c^{\left(k\right)}*h^{\left(k\right)}$

By b ^(k)Be arranged in matrix V;

B, carry out the piece decision feedback equalization, the linear equalization of ZF linear block balance and least mean-square error.

2. method according to claim 1 is characterized in that, also comprises among the described step B:

The chip sequence that receiver receives is expressed as

e＝Ad+n

D=(d wherein ₁, d ₂..., d _N) ^T∈ C ^{KN * 1}The cascade of the vector that becomes for n data symbols of all empty code channels,

Be the vector that n data symbols of all empty code channels becomes, e=e ⁽¹⁾+ e ⁽²⁾+ ...+e ^(K)∈ C ^{(NQ+W-1) * 1}Be the data that all empty code channels that receive mix, n=n ⁽¹⁾+ n ⁽²⁾+ ...+n ^(K)∈ C ^{(NQ+W-1) * 1}For the noise of the channel of all empty code channels and vector.

Images