CN101048934B - Reduced complexity sliding window based equalizer - Google Patents

Abstract

The present invention has many aspects. One aspect of the invention is to perform equalization using a sliding window approach. A second aspect reuses information derived for each window for use by a subsequent window. A third aspect utilizes a discrete Fourier transform based approach for the equalization. A fourth aspect relates to handling oversampling of the received signals and channel responses. A fifth aspect relates to handling multiple reception antennas. A sixth embodiment relates to handling both oversampling and multiple reception antennas.

Description

To reduce the equalizer of the sliding window of complicated property

Technical field

The relevant wireless telecommunication system that the present invention is total.Especially, the present invention is the Data Detection in relevant this kind system.

Background technology

Owing to improve the increase of receiver performance requirement, many advanced persons' receiver uses ZF (zero-forcing, ZF) block linear equalizer (block linear equalizer) and least mean-square error (minimum mean square error, MMSE) equalizer.

In these two kinds of methods, received signals has the model of program 1 usually.

R=Hd+n program 1

R is the vector that receives, and comprises the sample of received signals.H is that channel rings matrix.D is with evaluated data vector.In (spread spectrum) system that looses frequently, code division multiple access (CDMA) system for example, d can representative data symbol unit (symbol) or synthetic growth data vector.For synthetic growth data vector, each independently the used data symbols unit of sign indicating number produce by this evaluated data vector d with this yard being gone expand.N is a noise vector.

Assessment data vector in ZF block linear equalizer is for example with program 2.

D=(H ^HH) ^-1H ^HR program 2

() ^HIt is complex-conjugate transpose (or Hermetian) computing.In MMSE block linear equalizer, for example according to program 3 data assessment vectors.

D=(H ^HH+ σ ²I) ^-1H ^HR program 3

In the multipath (multipath) of wireless channel experience is propagated, correctly detect data for using these methods, need to use the sample that is received of a myriad of, this is unpractiaca.Therefore, wish to use approximate technology.Wherein a kind of method is sliding window (sliding window) method.In sliding window method, the window and the channel response of predetermined reception sample are used to Data Detection.After Preliminary detection, slide to the next window of sample under this window.This program continues to carry out till communication is ended.

By not using the sample number of a myriad of, in symbol unit (symbol) model shown in the program 1, import an error, and therefore cause incorrect Data Detection.This error window begin and finish the most significant, wherein infinite sequence effectively deletes to cut and partly has a maximum influence.A kind of method that reduces these errors is to use big window size and begins and finish to block the result at window.The intercepted part of window is being determined before and in the follow-up window.The method has suitable complicated property, especially when big channel delay expansion.This big window size causes employed big data matrix size and vector in the assessment.In addition, the method is because beginning and the efficient of detection of end data on abandoning these data then therefore tool calculating at window.

Therefore, hope can have other Data Detection method.

Summary of the invention

The present invention has many forms.A kind of form of the present invention is to use sliding window method to carry out equalizer.Second kind of form reused to each window and derived by the employed information of a follow-up window.The method based on discrete Fourier conversion (discrete Fourier transform) of usefulness is changed in the third form uses etc.The 4th kind of form is about handling the excessive sampling of received signal and channel response.The 5th kind of form is about handling multiple receive antennas.The 6th embodiment be about overtreating sampling and multiple receive antennas the two.

Description of drawings

Fig. 1 is banded channel response matrix.

Partly new in the banded channel response matrix of Fig. 2.

Fig. 3 is the data vector window of cutting apart with possibility.

Fig. 4 is the explanation of divided signal model.

The flow chart of the sliding window Data Detection of Fig. 5 correction factor of being to use over.

The receiver of the sliding window Data Detection of Fig. 6 correction factor of being to use over.

Fig. 7 is to use the flow chart of the sliding window Data Detection of noise auto-associating correction factor.

Fig. 8 is to use the receiver of the sliding window Data Detection of noise auto-associating correction factor.

Fig. 9 is the graphic representative of sliding window flow process.

Figure 10 is to use sliding window flow process graphic of circulation approximation method (circulant approximation).

Figure 11 is to use discrete Fourier conversion (DFTs) to detect the embodiment circuit diagram of data.

Embodiment

Though feature of the present invention and assembly are described with particular combinations in a particular embodiment, each feature or assembly can be used (further feature and the assembly that do not need preferred embodiment) separately, or are used in the various combination that has or do not have further feature of the present invention and assembly.

Below, wireless receiving/transmission unit (WTRU) includes but not limited to user's equipment, mobile radio station, fixing or moving user unit, calling set, or the device that can operate in wireless environment of any other kenel.When with reference to hereinafter the time, the base station includes but not limited to a B, positioner, the interface arrangement in wireless environment of access point or any kenel.

Though reducing the sliding window equalizer of complicated property is to be described in conjunction with preferable code division multiple access communication system, for example CDMA2000 and general mobile land system (UMTS) frequency division duplexing (FDD), time division duplex (TDD) pattern and timesharing synchronization CDMA (TD-SCDMA), it is applicable to different communication systems, and especially, various wireless telecommunication system.In wireless telecommunication system, it can be applied to being received from a base station by a WTRU, receive from one or more WTRUs by a base station, or by a WTRU from the transmission that another WTRU received, for example in (ad hoc) pattern that moves arbitrarily of running.

Below describe to use preferable MMSE algorithm to reduce the equalizer of complicated sliding window.Yet, also can use other algorithm, for example zero forcing algorithm.H () is the pulse of a channel.D () is to use extended code by expanding k the sample that is transmitted that a symbol unit is produced.It also can be and uses a group code, and orthogonal code for example is by expanding the sum total of the chip (chi p) that one group of symbol unit produced.R () is the signal that receives.The pattern of this system can be expressed as program 4.

$r\left(t\right)=\underset{k=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}d\left(k\right)h(t-k{T}_c)+n\left(t\right),-\&infin;<t<\&infin;$ Program 4

M (t) is noise and the interference (between cell interior (intra-cel 1) and cell (inter-cell)) that adds.For simplicity, be described as below supposing that the spreading rate sampling is to use at receiver, though also can use other sampling rate, for example several times of spreading rate.Sampled received signal can program 5 expressions.

$r\left(j\right)=\underset{k=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}d\left(k\right)h(j-k)+n\left(j\right)=\underset{k=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}d(j-k)h\left(k\right)+n\left(j\right),j\&Element;\{...,-2,-\mathrm{1,0,1,2},...\}$

Program 5

Tc is dropped in mark for the event of simplifying.

Suppose that h () is that limited support does not become in time.This is illustrated in and has pointer L in the discrete time-domain, so h ()=0, for i＜0 and i＞L.Therefore, program 5 can be rewritten as program 6.

$r\left(j\right)=\underset{k=0}{\overset{L-1}{\mathrm{\&Sigma;}}}h\left(k\right)d(j-k)+n\left(j\right),j\&Element;\{...,-2,-\mathrm{1,0,1,2},...\}$ Program 6

Suppose received signal have M received signal r (0) ... .r (M-1), generating routine 7.

r＝Hd+n

Wherein

r＝[r(0)，...，r(M-1)] ^T∈C ^M，

d＝[d(-L+1)，d(-L+2)，...，d(0)，d(1)，...，d(M-1)] ^T∈C ^M+L＝1

n＝[n(0)，...，n(M-1)] ^T∈C ^M

Program 7

At program 7, C ^MExpression has the space of all a plurality of vectors of dimension M.

The part of vector d can be used approximate procedure and be supposed M＞L and definition N=M-L+1 by decision, and vectorial d obtains from program 8.

Program 8

H matrix in the program 7 is a band matrix, and it is graphic that it can be represented as Fig. 1.At Fig. 1, each the row representation vector in the shadow region [h (L-1), h (L-2) ... h (1), h (0)], shown in program 7.

Replace all elements among the assessment d, only middle N element among the d is evaluated.

Be depicted as middle N as program 9.

$\tilde{d}={[d\left(0\right),...,d(N-1)]}^T$ Program 9

R is used identical observation, r with

Between linear approximate relationship according to program 10.

$r=\tilde{H}\tilde{d}+n$ Program 10

Matrix

Can be represented as graphic among Fig. 2 or shown in program 11.

Program 11

As shown, first L-1 of r and last L-1 element are not equal to the right-hand side of program 10.Therefore, at vector

The element of two ends will be than little near the element of central authorities with evaluated correctness.Because this specific character, better be used in the transmission sample as follow-up described sliding window method, chip (chip) for example, assessment.

In each k step of sliding window method, the sample that is received that ascertains the number is maintained at the r[k with N+L-1 dimension] in.They are used to one group of data with transmission of dimension N of service routine 10 assessments At vector

After evaluated, evaluated vector is only arranged

Be used in further data processing, for example by going diffusion (de-spread). Lower part (or after a while timely partly) is evaluated once more in the next step of sliding window processing, wherein r[k+1] have some element r[k] and the sample of some receptions, that is it is r[k] the version of skew (slip).

Though the size N of this window of preferably and slip step size are design parameter (based on the delay expansion of channel (L), the complicated property restrictions of the accurate demand of data assessment and enforcement), are the window size of illustrative purposes in following service routine 12.

N=4N s x SF program 12

SF is an invasin.Typical window size is 5 to 20 times of channel impulse response, though also can use other size.

The slip step size that is of a size of the basis with the window of program 12 is preferably, 2N _s* SF.N _s∈ 1,2 ... }, the preferably stays and does a design parameter.In addition, in each slip step, the evaluated chip that is transferred into despreader is evaluated The element 2N of central authorities _s* SF.This program description is at Fig. 3.

In sliding window method described above, this system model is similar to by abandoning some project in the model.In a kind of technology of following description, the information that project is wherein assessed by the step of sliding before using or make being characterized as noise in the model of described project and kept.This system model uses and keeps/characterization project and being corrected.

A kind of Data Detection Algorithm is used the MMSE algorithm with model error calibration, uses the system model of a sliding window as the method and the program 10 on basis.

Because approximate, the assessment of data, for example chip has error, especially in each slip step (beginning and finishing) at two ends of data vector.For proofreading and correct this error, the matrix H in the program 7 is split into a block column matrix, as program 13 (step 50).

$H=\left[H_p\right|\tilde{H}\left|H_f\right]$ Program 13

Subscript " p " expression " past ", and " f " expression " future ". From program 10.H _pAs program 14.

Program 14

H _fAs program 15.

Program 15

Vector d also is split into block, as program 16.

$d={\left[d_p^T\right|{\tilde{d}}^T\left|d_f^T\right]}^T$ Program 16

Identical with program 8, and d _pAccording to program 17.

d _p=[d (and L+1) d (L+2) ... d (1)] ^T∈ C ^L-1Program 17

d _fAccording to program 18.

d _f=[d (N) d (N+1) ... d (N+L-2)] ^T∈ C ^L-1Program 18

Original system model is subsequently according to program 19 and be illustrated in Fig. 4.

$r=H_p{d}_p+\tilde{H}\tilde{d}+H_f{d}_f+n$ Program 19

A kind of method such as program 20 to model program 19.

$\tilde{r}=\tilde{H}\tilde{d}+{\tilde{n}}_1$

Wherein $\tilde{r}=r-H_p{d}_p\mathrm{and}{\tilde{n}}_1=H_f{d}_f+\mathrm{<figure-callout\; id="20"\; label="n\; Program"\; filenames="H04815584420070530E000041.png"\; state="\{\{state\}\}">n</figure-callout>}$ Program 20

Use the MMSE algorithm, evaluated data vector

As program 21.

$\hat{\tilde{d}}=g_d{\tilde{H}}^H{(g_d\tilde{H}{\tilde{H}}^H+{\mathrm{\&Sigma;}}_1)}^{-1}\hat{\tilde{r}}$ Program 21

At program 21, g _dIt is chip energy according to program 22.

E{d (i) d ^*(j) }=g _dδ _IjProgram 22

Be according to program 23 gained.

$\hat{\tilde{r}}=r-H_p{\hat{d}}_p$ Program 23

Be in the advancing slip window step of elder generation Assessment.∑ ₁Be

The active incidence matrices, that is ${\mathrm{\&Sigma;}}_1=E\left\{{\tilde{n}}_1{{\tilde{n}}_1}^H\right\}.$ If suppose H _fd _fAnd n is not associated generating routine 24.

${\mathrm{\&Sigma;}}_1=g_d{H}_f{H}_f^H+E\left\{n{n}^H\right\}$ Program 24

Reliability decide according to the size of sliding window (channel delay time L relatively) and slip step size.

The method also is illustrated in conjunction with the receiver assembly of Fig. 5 and preferably Fig. 6, its may be implemented on WTRU or base station in.The circuit of Fig. 6 may be implemented on (IC) on the single IC for both, Application Specific Integrated Circuit (ASIC) for example, and on multiple ICs, the combination of for example Li San assembly, or IC and discrete component.

The generation channel assessment matrix H of portion is handled and received to channel estimation device 20 _p,

And H _fVectorial r (step 50).One following noise is the following noise active of associated apparatus 24 decisions association factor g initiatively _dH _fH _f ^H(step 52).One noise initiatively associated apparatus 22 determines initiatively association factor of a noise, E{nn ^H(step 54).One adder 26 adds up two-factor together to produce ∑ 1, (step 56).

One imports the part that means for correcting 28 is got the past of channel response matrix Hp in the past, and data vector

One in the past partly so that produce a correction factor in the past

(step 58).One subtracter 30 deducts this past correction factor and produces a reception vector of revising from the vector that receives

(step 60).MMSE device 34 uses ∑ ₁,

And

Data vector central part with the decision reception

For example according to program 21 (step 62).Next window uses in next window decision in an identical manner Some as

(step 64).As described in the method, only think the data of part

Determined, reduced the complicated property that Data Detection and the undesired part of amputation data vector are comprised.

In another method, only there is the noise project to be corrected about Data Detection.In the method, this system model is according to program 25.

$r=\tilde{H}\tilde{d}+{\tilde{n}}_n,,$ Wherein ${\tilde{n}}_2=H_p{d}_p+H_f{d}_f+n$ Program 25

Use the MMSE algorithm, evaluated data vector Be according to program 26.

$\hat{\tilde{d}}=g_d{\tilde{H}}^H{(g_d\tilde{H}{\tilde{H}}^H+{\mathrm{\&Sigma;}}_2)}^{-1}\mathrm{<figure-callout\; id="26"\; label="r\; Program"\; filenames="H04815584420070530E000041.png"\; state="\{\{state\}\}">r</figure-callout>}$ Program 26

Suppose Hpdp, Hfdf is not corrected, and then generating routine 27.

${\mathrm{\&Sigma;}}_2=g_d{H}_p{H}_p^H+g_d{H}_f{H}_f^H+E\left\{{\mathrm{nn}}^H\right\}$ Program 27

Separate the complicated property of program 26 for reducing service routine 27, do not need H _pH _p ^HAnd H _fH _f ^HThe complete matrix multiplication because H is only arranged usually _pTop and H _fLower corners be non-0.

The preferable receiver assembly that the method also may be implemented on WTRU or base station in conjunction with flow chart and Fig. 8 of Fig. 7 and being illustrated.The circuit of Fig. 8 may be implemented on a single IC for both (IC), and for example the long-pending body (ASICs) of special applications is implemented on the multiple ICs, as discrete assembly, or the combination of ICs and discrete component.

Channel estimation device 36 is handled received vector and is produced partly H of channel assessment matrix _p,

And H _f(step 70).The initiatively related means for correcting 38 of one noise uses the following of channel response matrix and partly determines initiatively related correction factor of a noise, g in the past _dH _pH _p ^H+ g _dH _fH _f ^H, (step 72).One noise initiatively associated apparatus 40 determines initiatively association factor E{nn of a noise ^H, (step 74).One adder is added to noise active association factor to produce ∑ with the related correction factor of noise active ₂, (step 76).One MMSE device 44 uses central part or channel response matrix

The vectorial r and the ∑ that receive ₂Central part with the assessment data vector

(step 78).The advantage of the method is not need to use the feedback loop of this detected data.Therefore, different sliding window version can be by simultaneously but not decision in regular turn.

Discrete Fourier is converted to the gradeization on basis

Above-described sliding window method needs a matrix to reverse (reverse), and this is the process of a complexity.The embodiment that implements sliding window uses following discrete Fourier conversion (DFTs).Though this method based on DFT is to use the MMSE algorithm, it can use other algorithm, for example based on the algorithm of ZF.

For some Integer N, matrix A _Cir∈ C ^{N * N}Be circular matrix, if it has the form of program 28.

Program 28

The matrix of this type uses DFT and IDFT operand and is expressed, and for example program 29.

A wherein _Cir[:, 1]=(a ₀, a ₁..., a _N) ^T∈ C ^N, that is it is a matrix A _CirFirst the row

Program 29

If suitably displacement is arranged, use the row outside first row.F _NBe N point DFT matrix, it is to any x ∈ C ^N, be defined as program 30.

${\left(F_{N}x\right)}_k=\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}x\left(n\right){\mathrm{<figure-callout\; id="2"\; label="e\; j"\; filenames="H04815584420070530E000061.png"\; state="\{\{state\}\}">e</figure-callout>}}^j,k=0,...,N-1$ Program 30

F _N ^-1Be N-1 point DFT matrix, it is to any x ∈ C ^N, be defined as program 31.

${\left(F_N^{-1}x\right)}_k=\frac{1}{N}{\left(F_N^{*}x\right)}_k=\frac{1}{N}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}x\left(n\right)e^{-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="H04815584420070530E000061.png"\; state="\{\{state\}\}">j</figure-callout>}\frac{2\mathrm{\&pi;kn}}{N}},k=0,...,N-1$ Program 31

Λ _N() is diagonal matrix, and it is to any x ∈ C ^N, be defined as program 32.

Λ _N(x)=diag (F _NX) program 32

Matrix A _CirReverse be expressed according to program 33.

$A_{\mathrm{cir}}^{-1}=F_N^{-1}{\mathrm{\&Lambda;}}_N^{-1}\left(A_{\mathrm{cir}}\right[:,1\left]\right)F_N$ Program 33

Below be about using method based on DFT based on the data assessment processing of the accurate position of the chip equalizer of sliding window.First embodiment uses single reception antenna.Follow-up embodiment uses multiple receive antenna.

This recipient system forms model according to program 34.

$r\left(t\right)=\underset{k=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}d\left(k\right)h(t-k{T}_c)+n\left(t\right),-\&infin;<t<\&infin;$ Program 34

H () is the pulse wave response of channel.K the chip sample that is transmitted that d (k) is to use diffuse code to produce by diffusion symbol unit.R () is the signal that receives.N () is the noise that adds and the summation (between cell inside (intra-cell) and the cell (inter-cell)) of interference.

Use spreading rate sampling and h () to have limited support, this is illustrated in the discrete time domain, has an integer L to make h (i)=0, for i＜0 and i 〉=L, j ∈ ... ,-2 ,-1,0,1,2 ... } sampled received signal can be expressed according to program 35.(Tc is the former of simplification sign thereby gives up).

$r\left(j\right)=\underset{k=0}{\overset{L-1}{\mathrm{\&Sigma;}}}h\left(k\right)d(j-k)+n\left(j\right)$ Program 35

Based on the received signal of M (M＞L), r (0) ..., r (M-1), generating routine 36.

r＝Hd+n

Wherein

r＝[r(0)，...，r(M-1)] ^T∈C ^M，

d＝[d(-L+1)，d(-L+2)，...，d(0)，d(1)，...，d(M-1)] ^T∈C ^M+L＝1

n＝[n(0)，...，n(M-1)] ^T∈C ^M

Program 36

Shown in program 36, the H matrix is several complex variable matrix (Toeplitz matrix).Described in the application of follow-up multi-chip rate sampling and/or multiple receive antenna, the H matrix is block several complex variable (block Toeplitz).Use block several complex variable characteristic, use the discrete Fourier switch technology.Several complex variable/block several complex variable nature are and the folding long-pending (convolution) of a channel or the result who amasss with the effective parallel channel folding with limit quantity.The appearance of effective parallel channel is the result of excessively sampling or multiple receive antennas.For a channel, single row must down be slid into the right to produce a several complex variable matrix.

The statistics of noise vector is by processed as having the active associate feature, according to program 37.

E{n n ^H}=σ ²I program 37

The left side of program (5) can be considered to be of continuous input signal string " window (window) ".For assessing this data, use the model that is fit to.In this approximate model, first L-1 of vectorial d and last L-1 element were assumed to be 0 before applying the MMSE algorithm, and the residue M-L+1 element of d forms new vector $\tilde{d}={[d\left(0\right),...,d(M-L+1)]}^T.$ This approximate model can be represented as program 38.

$r=\tilde{H}\tilde{d}+n$

Program 38

At vector

After evaluated, only there be intermediate portion part to be carried out and separate diffusion.Then, the window of observation (being received signal) is by (M-L+1)/2 element that slided, and repeats this flow process.Fig. 9 is the graphic of sliding window flow process as described above.

Use the MMSE algorithm, evaluated data are with program 39 expressions.

$\hat{\tilde{d}}=R^{-1}{\tilde{H}}^{H}r$

Wherein $R={\tilde{H}}^H\tilde{H}+{\mathrm{\&sigma;}}^{2}I$

Program 39

In program 39, matrix R and matrix

Can not be recycled to help DFT to implement.For helping DFT to implement, to each slip step, the approximation system model of service routine 40.

Wherein

Program 40

In program 40, it is the approximate of program 36 elements that first L-1 element [program] is only arranged.

Matrix

Replaced with a circular matrix (circulant matrix), for example according to program 41.

Program 41

This system model for each slip step, is according to program 42.

r＝H _cird+n

D=[d (0) wherein ..., d (M-1)] ^T∈ C ^{M * 1}

Program 42

Vectorial d in the program 42 is because new model and different with vectorial d in the program 36.Program 42 is added to extra distortion first L-1 element of program 39.This distortion makes that two ends of evaluated vectorial d are incorrect.Figure 10 is the graphic representation that this model structure is handled.

The approximate model of service routine 42, the MMSE algorithm produces the data of assessment, as program 43.

$\hat{d}=R_{\mathrm{cir}}^{-1}{H}_{\mathrm{cir}}^{H}r$

Wherein $R_{\mathrm{cir}}=H_{\mathrm{cir}}^H{H}_{\mathrm{cir}}+{\mathrm{\&sigma;}}^{2}I$ Program 43

H _Cir ^HAnd R _CirThe two is circulation and R _CirForm for program 44.

Program 44

Use data such as the program 45 of the characteristic ， Ping Huan of circular matrix.

$\hat{d}=F_M^{-1}{\mathrm{\&Lambda;}}_M^{-1}\left(R_{\mathrm{cir}}[:,1]\right){\mathrm{\&Lambda;}}_M\left(H_{\mathrm{cir}}^H[:,1]\right)F_{M}r$ Program 45

Figure 11 is graphic according to the circuit of program 45 elimination data.The circuit of Figure 11 may be implemented on a single IC for both (IC), and for example Application Specific Integrated Circuit (ASICs) is implemented on the multiple ICs, as discrete assembly, or the combination of ICs and discrete component.

Evaluated channel response

Be by one

Determination device 80 is handled with decision several complex variable matrix

Circulation approximation apparatus 82 is handled

To produce circular matrix A H _Cir, H _Cir ^HUse H _Cir, H _Cir ^HAnd noise variable σ ², R _CirBy a R _CirDetermination device 86 decisions.Use H _Cir ^HFirst the row, by Λ _M(H _Cir ^H[:, 1]) a pair of angular moment battle array of determination device 88 decisions.Use R _CirFirst the row, by Λ _M ^-1(R _Cir[:, 1]) determination device 90 determines a contrary diagonal matrix.Discrete Fourier conversion equipment 92 is carried out conversion on the vectorial r that receives.The diagonal angle, contrary diagonal angle and fourier transform result are multiplied each other together by multiplier 96.Contrary fourier transform is put 94 and is got the inverse conversion of multiplied result to produce data vector

This sliding window method is to be the constant basis that is assumed to be with channel in each sliding window.The channel pulse wave response that begins near sliding window can be used to each slip step.

Decision window step size N _SsAnd the method for window size M is according to program 46, though can use other method.

N _Ss=2N _Symbol* SF and M=4N _Symbol* SF program 46

N _Symbol∈ 1,2 ... } be the quantity of Fu Yuan and for should selecteed design parameter, so M＞L.Because the parameter that M also is the DFT that can use fft algorithm to implement to be used.M can reach greatly, therefore can use radix 2FFT (radix-2FFT) or main factor algorithm (prime factoralgorithm (PFA)) FFT.After data are evaluated, 2N _Symbol* SF sample is from N _Symbol* SF ^ThBe carried out and separate diffusion.Figure 11 is the explanation that obtains the sample of separating diffusion usefulness.

Changes such as multiple receive antenna

Below be to use the embodiment of multiple receive antenna, for example the K reception antenna.The assessment of the sample that is received vector of each antenna of independently getting and the response of channel pulse wave.Follow the program identical, each antenna input r with single antenna _kBe similar to according to program 47.

r _k=H _{Cir, k}D+n _k, k=1 ..., K program 47

Or according to the block matrices form of program 48.

Program 48

Program 49 and 50 is characteristics of active association of noise item purpose and cross correlation.

$E\left\{n_k{n}_k^H\right\}={\mathrm{\&sigma;}}^{2}I,\mathrm{for\; k}=1,...,K$ Program 49

And

$E\left\{n_k{n}_j^H\right\}=0,\mathrm{for\; k}\&NotEqual;j$ Program 50

Use the MMSE algorithm, evaluated data can be expressed according to program 51.

$\hat{d}=R_{\mathrm{cir}}^{-1}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{H}_{\mathrm{cir},k}^H{r}_k$

Wherein $R_{\mathrm{cir}}=\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{H}_{\mathrm{cir},k}^H{H}_{\mathrm{cir},k}+{\mathrm{\&sigma;}}^{2}I$ Program 51

R _CirStill be that circular matrix and evaluated data can be according to program 52 decisions.

$\hat{d}=F_M^{-1}{\mathrm{\&Lambda;}}_M^{-1}\left(R_{\mathrm{cir}}[:,1]\right)\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{\mathrm{\&Lambda;}}_M\left(H_{\mathrm{cir},k}^H[:,1]\right)F_M{r}_k$ Program 52

If reception antenna is closely arranged, the noise project can be carried out association in time and space.Therefore, may produce degeneration on some performance.

Multi-chip rate sampling changes such as (excessively samplings)

Use with the multi-chip rate sampling embodiment based on the equalization method of sliding window is below described.Multi-chip rate sampling be when channel when a specific assignment sampling speed is sampled, it is the integral multiple of spreading rate.For example 2 times, 3 times or the like.Though hereinafter concentrate on 2 times of every spreading rate, applicable other multiple of these methods.

Use N chip sliding window width and 2 times of spreading rate samplings, our reception vector is r=[r ₀, r ₁..., r _2N-1] ^TThis vector can be rearranged and be separated into an idol and receive vectorial r _e=[r ₀, r ₂..., r _2N-2] ^TAnd the vectorial r of a strange reception _o=[r ₁, r ₃..., r _2N-1] ^TDo not have most loss, data transfer model is according to program 53.

$\left[\begin{array}{c}{r}_e\\ r_o\end{array}\right]=\left[\begin{array}{c}{H}_e\\ H_o\end{array}\right]d+\left[\begin{array}{c}{n}_e\\ n_o\end{array}\right]$ Program 53

Program 53 is 2 spreading rate discrete time channels with effective every chip 2 sample discrete time channel separation.

Matrix H in the program 53 _eAnd H _oCorresponding idol and strange channel response matrix.These matrixes are from idol and strange channel response vector h _eWith h _o, it is to be divided into vectorial acquisition of even and strange channel response by every chip 2 samples to the channel response sampling and with it.

This channel noise is configured as has a variable σ ²White model, as program 54.

$E\left[n_e{n_e}^H\right]=E\left[n_o{n_o}^H\right]={\mathrm{\&sigma;}}^{2}I$ Program 54

If this channel is the white Gauss noise (white Gaussian noise (AWGN)) of addition, the data of channel and reception directly provide from the channel of sampling, and generating routine 55 then.

$E\left[n_e{n_o}^H\right]=0$ Program 55

Therefore, this problem is similar situation with spreading rate equalizer that 2 reception antennas of correlated noise not use on mathematics, as previously mentioned.Yet the aerial signal that is received among many embodiment is handled by a receiving terminal root raised cosine filter (root-raisedcosine (RRC) filter) before doing further processing being provided for digit receiver.After this kind processing, the noise vector of reception no longer is white, but has initiatively correlation function of raised-cosine (RC).RC is the frequency domain square of RRC response.Because the RC pulse wave is Nai Kuisi (Nyquist) pulse wave, program 54 is kept, but 55 of programs deny.Matrix ${\mathrm{\&Lambda;}}_{\mathrm{cross}}\stackrel{\mathrm{<figure-callout\; id="1"\; label="def"\; filenames="H04815584420070530E000011.png,H04815584420070530E000021.png"\; state="\{\{state\}\}">def</figure-callout>}}{=}\frac{1}{{\mathrm{\&sigma;}}^2}E\left[n_e{n_o}^H\right]$ (i, j) element is according to program 56.

$\frac{1}{{\mathrm{\&sigma;}}^2}E{\left[n_e{n_o}^H\right]}_{(i,j)}=x_{\mathrm{RC}}\left(\right|i-j|+0.5)$ Program 56

x _RCIt is the symbol elementary time normalization RC of unit pulse shape.

Λ _CrossCharacteristic be that it is real number (real), symmetry and several complex variable (Toeplitz); It is not band shape and does not have 0 project, and its project diminishes and near 0 when they when main diagonal is more and more far away.

∑ _nThe cross correlation matrix of expression full Bu miscellany news vector and according to program 57.

${\mathrm{\&Sigma;}}_n={\mathrm{\&sigma;}}^2\left[\begin{array}{cc}I& {\mathrm{\&Sigma;}}_{\mathrm{cross}}\\ {\mathrm{\&Sigma;}}_{\mathrm{cross}}& I\end{array}\right]$ Program 57

Certain solution

From certain solution of the problem of the linear least mean-square assessment of the d that observes r according to program 58.

${\hat{d}}_{\mathrm{MMSE}}={(H^H{\mathrm{\&Sigma;}}_n^{-1}H+I)}^{-1}{H}^H{\mathrm{\&Sigma;}}_n^{-1}r$

Wherein $y=H^H{\mathrm{\&Sigma;}}_n^{-1}r$ Be albefaction matched filtering (whitening matched filtering (WMF))

${\hat{d}}_{\mathrm{MMSE}}={(H^H{\mathrm{\&Sigma;}}_n^{-1}H+I)}^{-1}y$ Be change programs 58 such as linear MMSE

H ^H∑ _n ^-1And H ^H∑ _n ^-1H+I is neither to be several complex variable and also can not to be formed several complex variable via element unit-distance code [for example row/row rearranges], because ∑ _nStructure.Therefore, based on the approximate method of the circulation of several complex variable matrix based on DFT can not be applicable to this and certain separate very complicated.

Describe to derive effective algorithm that this problem of answer uses with two embodiment.First embodiment uses simple approximate, and second embodiment uses almost certain solution.

Simple approximate

The simple approximate n that ignores _eWith n _o, between association, ∑ _Cross=0.Therefore, use identical method with the multi-chip rate reception antenna.

The complicated property of simple approximation method is as described below.Consider N chip block.With rough approximation, the complicated property of N point DFT is supposed per second NlogN computing (operations per second (ops)).In addition, suppose that N point vector multiplication is to carry out N ops and to ignore vectorial addition.

DFT can be divided into 2 partly roughly for the complicated property of method on basis: must receive the flow process carried out on the data set and the flow process when channel assessment is updated at each, its frequency that is performed is usually than the former size of little one to two grade of computing.

For receiving the flow process of carrying out on the data set at each, carry out following running: 2N point DFTs is so that the vector that will receive is converted to frequency domain; 2N point vector multiplication (is multiplied by suitable " state (state) " vector with each vector that connects; And many DFT return time domain (time domain) to change this product.Therefore, the complicated property of Shi Heing is shown in program 59.

C _{1, r}=3N log N+2N program 59

Performed flow process when being updated about carrying out channel response is carried out following running: 2DFT computing, 6 N point vector multiplications and a vectorial division, the computing that its needs one vector multiplication is 10 times.Therefore, the complicated property of this program is approximately shown in program 60.

C _{1, r}=2N log N+16N program 60

Almost certain separates

For almost certain separating of using the fast several complex variable solution in district, vector and matrix are rearranged for its natural order, and therefore vectorial r is by r=[r ₀, r ₁..., r _2N-1] ^TObtain.Program 61 is nature Order Model.

r＝H _bTd+n

H wherein _BTBe defined as $H_{\mathrm{bT}}=\left[\begin{array}{c}{h}_{e,1}\\ h_{o,1}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ h_{o,N}\end{array}\right]=\left[\begin{array}{c}{G}_1\\ {\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="H04815584420070530E000061.png"\; state="\{\{state\}\}">G</figure-callout>}}_2\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ G_N\end{array}\right]$ Program 61

h _{E, i}Be H _eI row and h _{O, i}Be H _oI row.G _iBe the 2xN matrix, its 1st row are h _{E, i}And its secondary series is h _{O, i}. use G _i[x, y] is as G _iRow x, row y element, H _BTIt is the block several complex variable shown in program 62.

G _i[x，y]＝G _j[x，y+(i-j)]

Suppose 1≤y+ (i-j)≤N program 62

H _BTBlock several complex variable structure immediately from H _eThe H that reaches _oSeveral complex variable and rearranging of row and produce.Several complex variable structure and ∑ from I _Cross, the active incidence matrices that redefines in the problem also is the block several complex variable.Because matrix also is symmetrical, can be written as program 63 again.

∑ _bT＝[∑ _i，j] _{1≤i，j≤N}

∑ wherein _{I, j}Be the 2x2 matrix, have the characteristic ∑ _{I, j}=∑ _{| i-j|}

Program 63

Then produce the block circulation of block several complex variable matrix approximate.Because H _BTMatrix also is banded, directly obtains H subsequently _BTBlock circulation approximate.But, ∑ _BTNot banded, be similar to so can not directly produce the block circulation from it.Because Λ _BTElement away from main diagonal the time, tend to 0, to ∑ _BTBand shape approximate according to program 64.

${\mathrm{\&Sigma;}}_{\mathrm{bT}}\&ap;{\widetilde{\mathrm{\&Sigma;}}}_{\mathrm{bT}}={\left[{\widetilde{\mathrm{\&Sigma;}}}_{i,j}\right]}_{1\&le;i,j\&le;N}$

Wherein

Be the 2x2 matrix and have following characteristic

${\widetilde{\mathrm{\&Sigma;}}}_{i,j}={\mathrm{\&Sigma;}}_{|i-j|}$ If | i-j|≤B _nAnd ${\widetilde{\mathrm{\&Sigma;}}}_{i,j}=0$ Otherwise program 64

This noise co-variation alien frequencies wide (noise-covariance-bandwidth) B _nIt is selecteed design parameter.Because the fade characteristics of RC pulse shape, tending to only has several chips.Now

It is banded block several complex variable and approximate to its generation circulation.

H _BTCirculation approximate and

Be respectively H _BCWith ∑ _BCW _nExpression n point DFT matrix, if xis is a n-tuple exactly, x then _f=W _nX is the DFT of x.The block circular matrix is the form of program 65.

C wherein _iBe the NxN matrix and therefore C be the MNxMN matrix

Program 65

C also can be written as program 66.

$C=W_{M\&times;N}^{-1}{\mathrm{\&Lambda;}}_{M\&times;N}\left(C\right)W_{M\&times;N}$

W wherein _{M * N}Is is a block N-DFT matrix, is defined as $W_{M\&times;N}=W_M\&CircleTimes;I_N$

Program 66

Λ _{M * N}(C) be the block diagonal matrix fixed and represented as program 67 according to C.

Program 67

Λ _i(C) be the NxN matrix.For specifying Λ fully _i(C), λ _{I, (k, l)}Expression Λ _i(C) (k, l) element and being defined as ${\mathrm{\&lambda;}}_{(k,l)}\stackrel{\mathrm{def}}{=}{[{\mathrm{\&lambda;}}_{1,(k,l)},{\mathrm{\&lambda;}}_{2,(k,l)},...,{\mathrm{\&lambda;}}_{M,(k,l)}]}^T.$ c _{I (k, l)}The (k, the l) element, and being defined as of expression C $c_{(k,l)}\stackrel{\mathrm{def}}{=}{[{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="H04815584420070530E000011.png,H04815584420070530E000021.png"\; state="\{\{state\}\}">c</figure-callout>}}_{1,(k,l)},{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="H04815584420070530E000061.png"\; state="\{\{state\}\}">c</figure-callout>}}_{2,(k,l)},...,c_{M,(k,l)}]}^T.$ λ _{(k, l)}Be c _{(k, l)}M point DFT and be defined as program 68.

λ _{(k, l)}=W _Mc _{(k, l)}Program 68

Program 66-68 specifies the expression of the block DFT of square block circular matrix.Calculate Λ _{M * N}(C) need N ²DFTs.

The MMSE evaluator is write as program 69 again.

${\hat{d}}_{\mathrm{MMSE}}=H^H{({\mathrm{\&Sigma;}}_n+{\mathrm{HH}}^H)}^{-1}r$ Program 69

Form according to the MMSE evaluator of program 68 has several advantages.It only needs a single inverse matrix to calculate and therefore only needs a single vector to cut apart in the DFT territory.This provides potential important saving, is the complexity of height because cut apart.

This almost certain solution has two steps in preferred embodiment, though also can use other method.What each acquisition was new comments when estimating, channel filter is updated, (decision H ^H(∑ _n+ HH ^H) ^-1).To each block, this filter is applicable to the block of reception.Using this to cut apart is because the frequency of channel update is more not frequent in comparison with the processing that is received block, and therefore two steps can reduce complicated property greatly by dividing overall flow for this reason.

∑ _nDFT be that the DFT of pulse shape filter is multiplied by noise variable σ ²Because pulse shape filter normally fixing its DFT of feature of system can be calculated and is stored in the internal memory in advance and therefore σ is only arranged ²Be updated.Because pulse shape filter is approaching probably " desirable " (IRR) pulse shape, the DFT of ideal pulse shape can be ∑ _nUsed, reduce complicated property, and away from carrier.

Be the channel update step, carry out following flow process:

1. need to calculate H's " block DFT ".Because the width of block is 2, it needs 2 DFT.The result who is produced is a Nx2 matrix, and this matrix column is h _eAnd h _oDFTs.

2.HH ^∥" block DFT " be to seek h by element ground of an element _eAnd h _oActive relevance and cross correlation and being calculated.This needs many multiplication of 6N and many additions of 2N: N 2x2 matrix is calculated with the conspicuous transposition (Hermitian transposes) of itself.

3. ∑ _nBlock DFT be added, it needs the 3N multiplication (with σ ²The size of the block DFT of the RRC filter that is stored of decision) and 3N mutually in addition with the block DFT addition of two matrixes.

4. ∑ _n+ HH ^HReverse be put into block DFT field.For this reason, the reverse of each of N 2x2 matrix is put in the block DFT field.For assessing the quantity of whole computings, consider a hermetian matrix $M=\left[\begin{array}{cc}a& b\\ b^{*}& a\end{array}\right].$ This inverse of a matrix changes the program 70 that is illustrated in.

$M^{-1}=\frac{1}{a^2-{\left|b\right|}^2}\left[\begin{array}{cc}a& -b\\ -b^{*}& a\end{array}\right]$ Program 70

Therefore, the complicated property of calculating each reverse comprises 3 real multiplications and 1 real number subtraction (approximately being a plurality of multiplication) and a real number division.

5. this result and H ^∥Block DFT carry out block and multiply each other, it uses 8N multiplication+4N addition (because H altogether ^∥Not He Mei).

Generally speaking, need following calculating: 2N point DFS; Many multiplication of 18N (the 17N point vector multiplication+independent multiplication of N standard); Many additions of 11N (11N point vectorial addition); And 11 real number division.

The block r that handles a 2N numerical value (N chip lengths) comprises: 2N point DFTs; The product (filter and data) of N point block DFTs, it needs many multiplication of 8N and many additions of 4N; And the contrary DFTs of 1N point.

Comprehensive speech, need following calculating: 3N point DFTs; Many multiplication of 8N (8N point vector multiplication); And many additions of 4N (4N point vectorial addition).

Changes such as multiple chip sample at rates and multiple receive antennas

The embodiment that below is to use multiple chip sample at rates and multiple receive antennas etc. to change.With the L reception antenna, of 2L channel matrix-each antenna product " idol " and one " very " matrix.The channel matrix of l antenna is denoted as H _{L, e}And H _{L, o}And h _{L, e, n}And h _{L, o, n}The n row of representing this kind matrix.Each channel matrix is a several complex variable, and with the rearranging of the row that are fit to, the joint channel matrix is the many complex matrix of block, as program 71.

$H_{\mathrm{bT}}=\left[\begin{array}{c}{h}_{l,e,1}\\ h_{l,o,1}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ h_{L,o,N}\end{array}\right]=\left[\begin{array}{c}{G}_1\\ {\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="H04815584420070530E000061.png"\; state="\{\{state\}\}">G</figure-callout>}}_2\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ G_N\end{array}\right]$

Program 71

G _iThe matrix of ar is H _BTThe several complex variable block.Each G _iIt is the 2LxNr matrix.

Vectorial d from the observation r that is received can be formed model from program 72.

R=H _BTD+n program 72

The MMSE assessment is according to program 73.

${\hat{d}}_{\mathrm{MMSE}}={H_{\mathrm{bT}}}^H{({\mathrm{\&Sigma;}}_n+H_{\mathrm{bT}}{H_{\mathrm{bT}}}^H)}^{-1}r$ Program 73

∑ _nIs is that the co-variation of noise vector n is different.The form of separating of program 73 is based on and is used hypothesis.The importing of multiple antenna is guided out an additional space territory.Though the interactive relevance of the time and space is extremely complicated, can suppose the space correlation characteristic discord property association in time characteristic reciprocation of noise, except the direct product of the two, shown in program 74.

${\mathrm{\&Sigma;}}_n={\mathrm{\&Sigma;}}_{n,1\mathrm{ant}}\&CircleTimes;{\mathrm{\&Sigma;}}_{\mathrm{sp}}$ Program 74

∑ _{N, 1 ant}Be according to the co-variation different matrix of program 57 at the noise of single antenna observation.∑ _{N, 1 ant}Dimension be 2Nx2N.∑ _SpBe the normalization different matrix of space co-variation synchronously, that is it is for having 1 matrix at leading diagonal between the simultaneously observed L noise sample of L antenna and by regular turning to. Expression Kroenecker product.

∑ _nBe that 2LNx2LN He Mei positive semidefinite just partly limits matrix (Hermitian positivesemi-definite matrix), it is the block several complex variable with 2Lx2L block.For assessing this data, 4 preferred embodiments are described: certain solution; The simplification that has incoherent noise by hypothesis L reception antenna; By ignoring the strange simplification that reaches property association in time of even serial from same antenna; And by supposing that the chip serial is incoherent simplification.

Use the approximate complicated property of circulation can be split into two partly: the processing of the data that need carry out for the processing of the channel assessment of each new block execution and for each block itself based on DFT.In all 4 embodiment, the complicated property of deal with data comprises: 2L is N point DFTs forward; Many multiplication of 2LN; And 1 reverse N-point DFT.The complicated property of handling channel assessment changes because of each embodiment.

In the situation of certain MMSE solution, calculating is as follows from the complicated property of " the MMSE filter " of channel assessment: 2L N point DFT ' s; N2Lx2L matrix product+N2Lx2L addition of matrices is to calculate (∑ _n+ H _BTH _BT ^H); The N2Lx2L matrix reverses to calculate (∑ _n+ H _BTH _BT ^H) reverse; And the N2Lx2L matrix product is to produce real filtering.

Main contribution is that the matrix that must carry out the 2Lx2L matrix reverses step to the whole complicated property of this flow process.The complicated property that can be lowered by the incoherent nature of noise is as described below:

Main contribution is that the matrix that must carry out the 2Lx2L matrix reverses step to the whole complicated property of this flow process.The complicated property that can be lowered by the incoherent nature of noise is as described below:

1. if the hypothesis noise is unconnected at the two noise of time (strange/even sample) and space (crossing antenna), then ∑ _nBe reduced to a diagonal matrix and this problem and equal the sampling of the single sample of every chip that has the 2L antenna and have the space uncorrelated noise.Therefore, the computing that matrix reverses is reduced to a division simply, all is several complex variable because all are included in interior matrix.

2. if the hypothesis noise is incoherent in the space, it is that the 2X2 inverse of a matrix changes that the matrix that is then comprised reverses.

3. if suppose that the time of strange/even serial is uncorrelated, but keep the relevance of spatial noise, the matrix that is comprised is LxL.

Claims (10)

1. the data assessment method of a wireless telecommunication system comprises:

Receive vector by produce one in a multiple data signal chip sample at rates;

Using one, to handle this receptions based on the method for sliding window vectorial, and this ignores a relevance between the noise relevant with each multiple chip rate samples based on the method for sliding window, and comprises, to each window:

One non-several complex variable channel response matrix is converted to a several complex variable matrix;

With this several complex variable matrix conversion is a loop channel response matrix; And

In one with discrete Fourier be converted in the basic methods use this loop channel response matrix with assessment mutually should window a data vector; And

Be combined in the data vector assessed in each window to form the data vector of a combination.

2. method according to claim 1 is characterized in that this reception vector is to use this to handle this reception vector based on the method for sliding window in this to handle by a root raised cosine filter before.

3. method according to claim 2, it is characterized in that this uses this to handle this reception vector based on the method for sliding window and comprises the channel response matrix using a reception vector and arranged with a natural order, wherein the channel response matrix of this arrangement is a block several complex variable matrix, and this nature order is an order of this reception vector and the certain received element of this channel response matrix.

4. method according to claim 1, it is characterized in that this uses this to handle based on the method for sliding window and comprises a cross correlation that uses a noise vector in this reception vector, and a pulse shape filter is the signal in order to handle to receive, and being predetermined the discrete Fourier conversion and being multiplied by a noise variable of measuring to determine the discrete Fourier conversion of this noise vector cross correlation of this pulse shape filter.

5. method according to claim 1 is carried out by a wireless transmission receiving element.

6. wireless transmission/receive unit comprises:

In order to receive vectorial device by produce one in a multiple data signal chip sample at rates;

In order to use one to handle the device of this reception vector, should ignore the relevance between the noise relevant and make each window pass through based on the method for sliding window with each multiple chip rate samples based on the method for sliding window:

One non-several complex variable channel response matrix is converted to a several complex variable matrix;

With this several complex variable matrix conversion is a loop channel response matrix; And

Be converted in the basic methods with discrete Fourier in one and use this loop channel response matrix processed with assessment a data vector that mutually should window; And

In order to be combined in the data vector assessed in each window device with the data vector that forms a combination.

7. wireless transmission/receive unit according to claim 6 comprises that also a root raised cosine filter is to be configured to handle this reception vector before this is based on the processing of sliding window.

8. wireless transmission/receive unit according to claim 7, it is characterized in that this is configured to a channel response matrix of using a reception vector and a natural order to be arranged in order to use this device of handling this reception vector based on the method for sliding window, wherein the channel response matrix of this arrangement is a block several complex variable matrix, and this nature order is an order of this reception vector and the certain received element of this channel response matrix.

9. wireless transmission/receive unit according to claim 6 also comprises plural reception antenna and receives the device of vector signal in order to take a sample from this plural number reception antenna at a multiple spreading rate.

10. wireless transmission/receive unit according to claim 6, it is characterized in that this in order to the cross correlation that uses this device of handling this reception vector based on the method for sliding window to be configured to use a noise with using a pulse shape filter handling the signal that receives, and be multiplied by a noise variable of measuring to determine the discrete Fourier conversion of this noise vector cross correlation with the discrete Fourier conversion that one of this pulse shape filter is predetermined.

Images