KR100768737B1 - Reduced complexity sliding window based equalizer - Google Patents

Abstract

The present invention has many aspects. One aspect of the present invention is to perform equalization using a sliding window. The second side reuses the derived information for each window for use by subsequent windows. The third aspect uses an approach based on Discrete Fourier Transform for equalization. A fourth aspect relates to processing oversampling of received signals and a fourth response. The fifth aspect relates to processing multiple receive antennas. The sixth embodiment relates to processing both oversampling and multiple receive antennas.

Data estimation, cutting, sliding windows, error compensation, equalizers.

Description

REDUCED COMPLEXITY SLIDING WINDOW BASED EQUALIZER}

The present invention relates generally to wireless communication systems, and more particularly to data detection in such systems.

Due to the increased demand for improved receiver performance, many advanced receivers use a zero forcing (ZF) block linear equalizer and a Minimum Mean Square Error (MMSE) equalizer.

In both of these approaches, the received signal is typically modeled by equation (1).

(수학식 1)

(Equation 1)

r is a received vector and contains samples of the received signal.

H is the channel response matrix and d is the data vector to be estimated. In spread spectrum systems, such as code division multiple access (CDMA) systems, d represents a data symbol or a complex spread data vector. In the case of a complex spread data vector, the data symbols for each individual code are generated by spreading the estimated data vector d into that code. n is a noise vector.

In the ZF block linear equalizer, the data vector is estimated by equation (2).

(수학식 2)

(Equation 2)

(·) ^H is a complex conjugated transpose (or Hermetian) operation. In the MMSE block linear equalizer, the data vector is estimated according to equation (3).

(수학식 3)

(Equation 3)

In wireless channels undergoing multipath propagation, to accurately detect data using these approaches, an infinite number of received samples are required, which is not practical. Therefore, it is preferable to use an approximation technique. One such approach is the sliding window approach. In the sliding window approach, predetermined window and channel responses of the received samples are used in data detection. After the initial detection, this window slides to the next window of samples. This process continues until communication is interrupted.

By not using an infinite number of samples, an error flows into the symbol model shown in Equation 1, causing incorrect data detection. This error is most pronounced at the beginning and end of the window where the effectively truncated portions of an infinite sequence will have the greatest impact. One approach to reducing these errors is to use large window sizes and truncate the results at the beginning and end of the window. The cuts of the window are determined in the previous and subsequent windows. This approach has considerable complexity, especially when the channel delay spread is large. Large window sizes increase the dimensions of the matrices and vectors used in data estimation. Also, this approach is not computationally efficient due to the detection of data at the beginning and end of the window and the discarding of that data.

Thus, there is a need for an alternative approach for data detection.

The present invention has many aspects. One aspect of the present invention is to perform equalization using a sliding window approach. The second side reuses the derived information for each window for use by subsequent windows. The third aspect uses an approach based on Discrete Fourier Transform for equalization. The fourth aspect relates to processing the received signals and oversampling and channel response. The fifth aspect relates to processing multiple receive antennas. The sixth embodiment relates to processing both oversampling and multiple receive antennas.

1 is a diagram illustrating a banded channel response matrix.

2 is a diagram illustrating the middle portion of the band-shaped channel response matrix.

3 is a diagram illustrating a divided data vector window as an example.

4 is a diagram illustrating a divided signal model.

5 is a flowchart of sliding window data detection using past correction coefficients.

6 is a receiver using sliding window data detection using past correction coefficients.

7 is a flowchart of sliding window data detection using noise autocorrelation correction coefficients.

8 is a receiver using sliding window data detection using noise autocorrelation correction coefficients.

9 is a graphical representation of a sliding window process.

10 is a graphical representation of a sliding window process using cyclic approximation.

11 is a circuit of an embodiment for data detection using Discrete Fourier Transform (DFT).

Although the features and elements of the present invention are described in the preferred embodiments of a particular combination, each feature and element may be used alone (without other features and elements of the preferred embodiments) or other features and elements of the present invention. Can be used in various combinations with or without them.

Hereinafter, a wireless transmission / unit (WTRU) includes, but is not limited to, user equipment, mobile stations, fixed or mobile subscriber units, pagers, or any other type of device capable of operating in a wireless environment. As mentioned below, a base station includes, but is not limited to, a Node-B, a site controller, an access point, or any other type of interfacing device in a wireless environment.

Although sliding window equalizers with reduced complexity, good wireless code division multiple access communication systems such as CDMA2000 and UMTS Universal Mobile Terrestrial System Frequency Division Duplex (FDD), time division duplex (TDD) mode and time division synchronous CDMA (TD-SCDMA) Although described in connection with, it may be applied to various communication systems, in particular, to various wireless communication systems. In a wireless communication system, this equalizer may be applied to transmissions from base stations received by the WTRUs, such as in transmissions from one or more WTRUs received by the base stations, or in ad hoc mode operation. The same may apply to transmissions from another WTRU received from one WTRU.

The implementation of a sliding window based equalizer with reduced complexity using a good MMSE algorithm is described below. However, other algorithms may be used, such as a zero forcing algorithm. h (·) is the impulse response of the channel. d (k) is the k th transmitted sample generated by spreading the symbol using a spreading code. This may be the sum of the chips generated by spreading a set of symbols using a set of codes, such as orthogonal code. r (·) is the received signal. The model of this system can be represented by equation (4).

(수학식 4)

(Equation 4)

n (t) is the sum of additional noise and interference (intracell interference and intercell interference). Although other sampling rates may be used, such as multiples of the chip rate, for simplicity, we will assume that chip rate sampling is used at the receiver. The sampled received signal can be represented by equation (5).

(Equation 5)

T _c has been omitted for brevity.

h (·) has finite support and is time invariant. This means that in the discrete-time domain, the index L is present such that h (i) = 0 (for i <0 and i> L). As a result, Equation 5 may be rewritten as Equation 6.

(수학식 6)

(Equation 6)

Considering that the received signal has M received signals r (0), ..., r (M-1) , Equation 7 results in the following.

here,

(Equation 7)

In Equation 7, CM represents a space of all complex vectors having the dimension M. The portion of the vector d can be determined using an approximation. Assuming M > L and defining N = M - L + 1, the vector d is (8).

(Equation 8)

In Equation 7, the H matrix is a banded matrix and may be expressed as shown in FIG. 1. In FIG. 1, each row of the shaded region is a vector [ h ( L −1) , h ( L −2) ,..., H (1) , h (0) , as shown in equation (7). ].

는 수학식 9에 따른 중간 N개 요소들이다.Instead of estimating all of the elements in d, it is estimated that only the middle N elements of d.

Is the middle N elements according to equation (9).

(수학식 9)

(Equation 9)

간의 근사적 선형 관계는 수학식 10에 의한 것과 같다. and for r using the same rule, and r

The approximate linear relationship between is as in Equation 10.

(수학식 10)

(Equation 10)

는 도 2의 도면이나 수학식 11에 도시된 바와 같이 표시될 수 있다.procession

May be displayed as shown in FIG. 2 or Equation 11.

(수학식 11)

(Equation 11)

의 2개 끝에서의 요소들은 중심에 가까운 요소들보다 덜 정확하게 추정된다. 이러한 속성으로 인해, 슬라이딩 윈도우 접근법은, 칩들과 같은, 전송된 샘플들의 추정에 양호하게 사용된다.As shown, the first L −1 and last L −1 elements of r are not equal on the right side of equation (10). As a result, vector

The elements at the two ends of are estimated less accurately than the elements near the center. Due to this property, the sliding window approach is well used for the estimation of transmitted samples, such as chips.

를 추정하는데 사용된다. 벡터

가 추정된 후에, 추정된 벡터

의 "중간(middle)" 부분만이 역확산(despreading)과 같은 추가의 데이터 처리에 사용된다.

의 "하위" 부분(시간적으로 나중 부분)은 슬라이딩 윈도우 프로세스의 다음 단계에서 다시 추정된다. 여기서, r[k+1]은 요소들 r[k] 중 일부와 새로이 수신된 샘플들 일부를 가진다. 즉, r[k]의 시프트(슬라이드) 버전이다.In each k th step of the sliding window approach, a predetermined number of received samples are maintained in r [ k ] of dimension N + L -1. They are a set of transmitted data in dimension N using Equation 10

Used to estimate vector

After is estimated, the estimated vector

Only the "middle" part of is used for further data processing, such as despreading.

The "lower" part of (the later part in time) is estimated again in the next step of the sliding window process. Here, r [ k +1] has some of the elements r [ k ] and some of the newly received samples. That is, it is a shift (slide) version of r [ k ].

Although preferably, the window size N and the sliding step size are design parameters (based on the delay spread (L) of the channel, the accuracy requirements of the data estimation, and the complexity limitations in the implementation), but for purposes of exemplification purposes below: For example, the window size of Equation 12 is used.

(수학식 12)

(Equation 12)

SF is the diffusion coefficient. Typical window sizes are 5 to 20 times larger than the channel impulse response, although other sizes are available.

의 중간에 있는 2N_s × SF개의 요소들이다. 이 프로시져는 도 3에 도시되어 있다. 이와 같은 슬라이딩 윈도우 모델에서, 시스템 모델은 모델 내의 몇개 항들을 버림으로써 근사화되고 있다. 이하에서는, 이전의 슬라이딩 단계에서 추정된 정보를 이용하거나 항들을 모델 내에서 노이즈로서 특화시킴으로써 항들을 유지하는 기술이 설명될 것이다.The sliding step size based on the window size in (12) is preferably 2 N _s × SF . N _s ∈ {1, 2, ...} is preferably left as a design parameter. Also, in each sliding step, the estimated chips sent to the despreader are estimated

Are 2 N _s × SF elements in the middle of. This procedure is shown in FIG. In such a sliding window model, the system model is approximated by discarding several terms in the model. In the following, a technique for holding the terms by using the information estimated in the previous sliding step or by specifying the terms as noise in the model will be described.

One algorithm of data detection uses the MMSE algorithm, and its model error correction uses the system model of equation (10) and a sliding window based approach.

Due to the approximation, the estimation of data such as chips has errors, in particular errors at both ends (first part and end part) of the data vector in each sliding step. To correct this error, the H matrix of equation (7) is divided into a block row matrix according to equation (13) (step 50).

(수학식 13)

(Equation 13)

는 수학식 10에 따른다. H _p 는 수학식 14에 따른다.The subscript " p " represents "past" and " f " represents "future".

Is based on Equation 10. H _p is according to equation (14).

(수학식 14)

(Equation 14)

H _f is according to Equation 15.

(수학식 15)

(Equation 15)

The vector d is divided into blocks according to equation (16).

(수학식 16)

(Equation 16)

는 수학식 8에 따르고, d _p 는 수학식 17에 따른다.

Is according to Equation 8 and d _p is according to Equation 17.

(수학식 17)

(Equation 17)

d _f is according to Equation 18.

(수학식 18)

(Equation 18)

The original system model is in accordance with equation (19) and is shown in FIG.

(수학식 19)

(Equation 19)

One approach to modeling equation (19) follows equation (20).

이고,

이다. (수학식 20)here,

ego,

to be. (Equation 20)

는 수학식 21에 따른다.Estimated data vector using MMSE algorithm

Is based on Equation 21.

(수학식 21)

(Equation 21)

는 수학식 22에 따른 칩 에너지이다.In Equation 21,

Is the chip energy according to (22).

(수학식 22)

(22)

은 수학식 23에 따른다.

Is according to equation (23).

(수학식 23)

(Equation 23)

는 이전의 슬라이딩 윈도우 단계에서

의 추정의 부분이다.

은

의 자기상관 행렬이다. 즉,

이다. 만일 H _f d _f 및 n이 상관되지 않는다고 가정하면, 수학식 24의 결과가 나온다.

In the previous sliding window stage

Is part of the estimate.

silver

Is an autocorrelation matrix of. In other words,

to be. If we assume that H _f d _f and n are not correlated, the result of equation (24) is given.

(수학식 24)

(Equation 24)

의 신뢰성은 (채널 지연 스팬 L 에 관한) 슬라이딩 윈도우 크기와 슬라이딩 단계 크기에 의존한다.

The reliability of P depends on the sliding window size and the sliding step size (relative to the channel delay span L ).

This approach is also described in conjunction with the flowchart of FIG. 5 and the preferred receiver components of FIG. 6, which may be implemented with a WTRU or base station. The circuit of FIG. 6 may be implemented on one integrated circuit (IC), such as an application specific integrated circuit (ASIC), or may be implemented on a plurality of ICs, such as individual components or a combination of ICs and individual components.

, 및 H _f 를 생성한다(단계 50). 미래 노이즈 자기상관 장치(24)는 미래 노이즈 자기상관 계수,

를 결정한다(단계 52). 노이즈 자기상관 장치(22)는 노이즈 자기상관 계수, E{nn ^H}를 결정한다(단계 54). 합산기(26)는 2개의 계수들을 합산하여

를 생성한다(단계 56).The channel estimating apparatus 20 processes the received vector r to perform channel estimation matrix portions,

, And H _f are generated (step 50). The future noise autocorrelation apparatus 24 includes a future noise autocorrelation coefficient,

(Step 52). The noise autocorrelation apparatus 22 determines the noise autocorrelation coefficient, E { nn ^H } (step 54). Summer 26 adds the two coefficients

Create (step 56).

의 과거의 사전설정된 부분을 취하여, 과거 보정 계수

를 생성한다(단계 58). 감산기(30)는 수신 벡터로부터 과거 보정 계수를 감산하여, 수정된 수신 벡터

를 생성한다(단계 60). MMSE 장치(34)는 수학식 21에 따른 수신 데이터 벡터 중심부

를 결정하기 위해

,및

를 사용한다(단계 62).

의 부분을 다음 윈도우 결정에서의

로서 사용하여 동일한 방식으로 다음 윈도우가 결정된다(단계 64). 이 접근법에서 설명된 바와 같이, 관심 부분에 대한 데이터

만이 결정되어, 데이터 검출 및 데이터 벡터의 원치 않는 부분의 절삭에 연관된 복잡도를 줄이게 된다.The past input correction device 28 is a past portion of the channel response matrix H _p and a data vector.

A past correction factor, taking the past preset portion of

Create (step 58). The subtractor 30 subtracts the past correction coefficients from the received vector to correct the received vector.

Create (step 60). MMSE device 34 is the center of the received data vector according to Equation 21

To determine

, And

(Step 62).

The part of the next window determined

The next window is determined in the same manner using as (step 64). As described in this approach, data for the portion of interest

Only the determination is made, which reduces the complexity associated with data detection and cutting of unwanted portions of the data vector.

In another approach to data detection, only noise terms are corrected. In this approach, the system model is in accordance with (25).

(수학식 25)

(Equation 25)

는 수학식 26에 따른다.Estimated Data Vectors Using MMSE Algorithm

Is based on Equation 26.

(수학식 26)

(Equation 26)

Equation 27 results as assuming that H _p d _p , H _f d _f and n are not correlated.

(수학식 27)

(Equation 27)

및

에 대한 전체 행렬 곱셈은 필요하지 않다. 이는 대개 H _p 및 H _f 의 상위 및 하위 코너만이 0이 아니기 때문이다.To reduce the complexity of solving Equation 26 using Equation 27,

And

Full matrix multiplication for is not necessary. This is usually because only the upper and lower corners of H _p and H _f are not zero.

This approach is also described in conjunction with the flowchart of FIG. 7 and the components of the preferred receiver of FIG. 8, which may be implemented in a WTRU or base station. The circuit of FIG. 8 may be implemented on one integrated circuit (IC), such as an application specific integrated circuit (ASIC), or may be implemented on a plurality of ICs, such as individual components or a combination of ICs and individual components.

, 및 H _f 를 생성한다(단계 70). 노이즈 자기상관 장치(38)는 채널 응답 행렬의 미래부 및 과거부를 사용하여 노이즈 자기상관 보정 계수

를 결정한다(단계 72). 노이즈 자기상관 장치(40)는 노이즈 자기상관 계수 E{nn ^H}을 결정한다(단계 74). 합산기(42)는 노이즈 자기상관 계수에 노이즈 자기상관 보정 계수를 가산하여

를 생성한다(단계 76). MMSE 장치(44)는 데이터 벡터

의 중심 부분을 추정하기 위해 채널 응답 행렬

의 중심 부분, 수신 벡터 r, 및

를 사용한다(단계 78). 이러한 접근법의 한 잇점은, 검출된 데이터를 사용하는 피드백 루프가 필요하지 않다는 것이다. 그 결과, 상이한 슬라이드 윈도우 버전이, 순차적 방식이 아니라 병렬로 결정될 수 있다.The channel estimating apparatus 36 processes the received vector so that the channel estimation matrix portion

, And H _f are generated (step 70). The noise autocorrelation apparatus 38 uses the future and past portions of the channel response matrix to determine the noise autocorrelation correction coefficients.

(Step 72). The noise autocorrelation apparatus 40 determines the noise autocorrelation coefficient E { nn ^H } (step 74). Summer 42 adds the noise autocorrelation correction coefficient to the noise autocorrelation coefficient

Create (step 76). MMSE device 44 is a data vector

Channel response matrix to estimate the central part of the

The central part of, the reception vector r , and

(Step 78). One advantage of this approach is that no feedback loop using the detected data is needed. As a result, different slide window versions can be determined in parallel rather than in a sequential manner.

Equalization Based on Discrete Fourier Transform

The sliding window approach described above requires a matrix inverse transformation, which is a complex process. One embodiment for implementing a sliding window uses a Discrete Fourier Transform (DFT) as follows. Although a good implementation of the DFT-based approach is done with MMSE, it can be applied to other algorithms such as zero forcing (ZF) based algorithms.

는, 수학식 28에 따른 이하의 형태를 가진다면, 순환 행렬(circulant matrix)을 사용하여 표현된다.For any integer N, the matrix

Is expressed using a circulant matrix if it has the following form according to equation (28).

(수학식 28)

(Equation 28)

This kind of matrix is represented using the DFT and IDFT operators, as shown in equation (29).

(수학식 29)

(Equation 29)

, 즉, 행렬

의 제1 컬럼. 적절히 퍼뮤테이션(permutation)된다면, 제1 컬럼이 아닌 컬럼들이 사용될 수 있다. F _N 은 임의의 x ∈ C^N에 대해 수학식 (30)에 따라 정의되는 N-포인트 DFT 행렬이다.here,

, I.e. matrix

First column of. If properly permutated, columns other than the first column may be used. F _N is an N-point DFT matrix defined according to equation (30) for any x ∈ C ^N.

(수학식 30)

(30)

F _N ⁻¹ is an N-point inverse DFT matrix defined according to equation (31), for any x ∈ C ^N.

(수학식 31)

(Equation 31)

는, 임의의 x ∈ C^N에 대해 수학식 (32)에 따라 정의되는 대각 행렬이다.

Is a diagonal matrix defined according to equation (32) for any x ∈ C ^N.

(수학식 32)

(Equation 32)

의 역은 수학식(33)에 따라 표현된다.procession

The inverse of is expressed according to equation (33).

(수학식 33)

(Equation 33)

The following is an example of applying a DFT-based approach to the data estimation process using a sliding window based chip level equalizer. The first embodiment uses one receive antenna. Subsequent embodiments use a plurality of receive antennas.

The receiver system is modeled according to equation (34).

(수학식 34)

(Equation 34)

은 채널의 임펄스 응답이다.

는 확산 코드를 사용하여 심볼들을 확산시킴으로써 발생되는 k번째 전송된 칩 샘플들이다.

은 수신된 신호이다.

은 부가적 잡음 및 (셀간 및 셀내) 간섭의 합이다.

Is the impulse response of the channel.

Is the k th transmitted chip samples generated by spreading the symbols using a spreading code.

Is the received signal.

Is the sum of the additional noise and the intercell and intracell interference.

을 이용하면, 이것은 이산-시간 도메인에서, i < 0 이고 i ≥ L에 대한 h(i) = 0이 되는 정수 L이 있음을 의미하고, 샘플링된 수신된 신호는, 수학식(35)에 따라 표현될 수 있다.(T_c는 표기의 간략화를 위해 생략되었다)With chip rate sampling and finite support

This means that in the discrete-time domain, there is an integer L where i <0 and h (i) = 0 for i ≥ L, and the sampled received signal is given by Can be expressed (T _c is omitted for simplification of the notation).

(수학식 35)

(Equation 35)

Based on the M (M> L) received signals r (0), ..., r (M-1), equation (36) is obtained.

here,

(수학식 36)

(Equation 36)

As shown in equation (36), the H matrix is Toeplitz. As described subsequently in applications for multiple chip rate sampling and / or multiple receive antennas, the H matrix is block topless. Discrete Fourier transform techniques can be applied using the block topless attribute. The topless / block topless attribute is generated as a result of convolution of one channel or convolution of an input signal with a finite number of valid parallel channels. Effective parallel channels appear as a result of either oversampling or multiple receive antennas. For one channel, one row essentially slides down, creating a topless matrix on the right side.

The statistics of the noise vector are treated as having autocorrelation according to equation (37).

(수학식 37)

(Equation 37)

를 형성한다. 근사화된 모델은 수학식 38에 따라 명시적으로 표현될 수 있다.The left side of equation (5) can be viewed as the "window" of the continuous input signal stream. To estimate the data, an approximated model is used. In this approximated model, the first L-1 elements and the last L-1 elements of vector d are assumed to be zero prior to applying the MMSE algorithm, and the reset M-L + 1 elements of d are new vectors.

To form. The approximated model can be expressed explicitly according to equation (38).

here,

(수학식 38)

(38)

가 추정된 후에, 그 중간 부분만이 역확산용으로 취해진다. 후속해서, 관측의 윈도우(즉, 수신된 신호)는 (M - L + 1)/2 요소들만큼 슬라이드되고, 프로세스는 반복된다. 도 9는 상술한 슬라이딩 윈도우 프로세스의 그래픽 표현이다.vector

After is estimated, only the middle part thereof is taken for despreading. Subsequently, the window of observation (ie the received signal) is slid by (M − L + 1) / 2 elements, and the process is repeated. 9 is a graphical representation of the sliding window process described above.

Using the MMSE algorithm, the estimated data is represented using equation (39).

, 여기서,

, here,

(수학식 39)

(Equation 39)

중 어느 것도 순환 행렬(circulant)이 아니다. DFT 구현을 용이하게 하기 위해, 각각의 슬라이딩 단계에 대해, 수학식(40)에 따른 근사화된 시스템 모델이 사용된다.In equation (39), matrix R and matrix to facilitate DFT implementation

Neither of them is a circulant. To facilitate the DFT implementation, for each sliding step, an approximated system model according to equation (40) is used.

(수학식 40)

(Equation 40)

In Equation 40, only the first L-1 elements (Equations) are approximations to the elements of Equation 36.

는 수학식 41에 따라, 순환 행렬로 치환된다.procession

Is replaced with a circulant matrix according to Equation 41.

(Equation 41)

The system model for each sliding step is according to equation (42).

(수학식 42)here,

(Equation 42)

The vector d in (42) is different from the vector d in (36) because of the new model. (42) adds additional distortion to the first L-1 element of (39). This distortion makes the two ends of the estimated vector d inaccurate. 10 is a graphical representation of a model construction process.

Using the approximated model according to (42), the MMSE algorithm yields the estimated data according to (43).

이다.here,

to be.

및

은 순환 행렬이고,

은 수학식 44에 따른 형태이다.

And

Is a cyclic matrix,

Is a form according to equation (44).

(44)

Applying the properties of the circulant matrix, the estimated data is in accordance with equation (45).

(수학식 45)

(Equation 45)

11 is a circuit diagram for estimating data according to Equation 45. FIG. The circuit of FIG. 11 may be implemented on one integrated circuit (IC), such as an application specific integrated circuit (ASIC), or may be implemented on a plurality of ICs, as a discrete component or a combination of an IC and a discrete component.

는 토플리츠 행렬

를 결정하기 위해

결정 장치(80)에 의해 처리된다. 순환 근사화 장치(82)는 순환 행렬

을 생성하기 위해

를 처리한다. 헤르메시안 장치(84)는

의 헤르메시안,

을 생성한다.

,

, 및 노이즈 편차

을 이용하여,

이

결정 장치(86)에 의해 결정된다.

의 첫번째 컬럼을 사용하여,

결정 장치(90)에 의해 역 대각 행렬이 결정된다. 이산 푸리에 변환 장치(92)는 수신 벡터 r 상에 변환을 수행한다. 대각, 역대각, 및 푸리에 변환 결과는 곱셈기(96)에 의해 서로 곱해진다. 역 푸리에 변환 장치(94)는 데이터 벡터

를 생성하기 위해 곱셈 결과의 역변환을 취한다.Estimated Channel Response

Topless procession

To determine

It is processed by the determination device 80. Cyclic approximation device 82 is a circular matrix

To generate

To process Hermesian device 84

Hermesian,

Create

,

, And noise deviation

Using

this

It is determined by the determination device 86.

Using the first column of,

The inverse diagonal matrix is determined by the determining device 90. The discrete Fourier transform device 92 performs the transform on the reception vector r. The diagonal, inverse diagonal, and Fourier transform results are multiplied by each other by the multiplier 96. Inverse Fourier Transform 94

Take the inverse of the multiplication result to produce.

The sliding window approach is based on the assumption that the channel is invariant within each sliding window. A channel impulse response near the beginning of the sliding window can be used for each sliding step.

이고

이다. (수학식 46)

ego

to be. (46)

는 심볼의 갯수이며, M > L이 되도록 선택되어야 하는 설계 파라미터이다. M은 FFT 알고리즘을 이용하여 구현될 수 있는 DFT에 대한 파라미터이기도 하기 때문에, M은 Radix-2 FFT 또는 프라임 팩터 알고리즘(PFA) FFT가 적용될 수 있도록 충분히 크게 만들어질 것이다. 데이터가 추정된 후에,

샘플로부터 시작하는 역확산을 처리하기 위해

샘플들이 취해진다. 도 11은 역확산을 위해 샘플들을 취하는 도면이다.

Is the number of symbols and is a design parameter that should be chosen such that M> L. Since M is also a parameter for a DFT that can be implemented using the FFT algorithm, M will be made large enough so that the Radix-2 FFT or Prime Factor Algorithm (PFA) FFT can be applied. After the data is estimated,

To handle despreading starting from the sample

Samples are taken. 11 is a diagram that takes samples for despreading.

Multiple receive antenna equalization

가 수학식 47에 따라 근사화된다.The following is an embodiment using multiple receive antennas, such as K receive antennas. Estimates of the sample of the received vector and the channel impulse response are taken independently for each antenna. Each antenna input, following the same process as for the single antenna embodiment

Is approximated according to equation (47).

(수학식 47)

(Equation 47)

Or in the form of a block matrix according to equation (48).

(수학식 48)

(48)

Equations 49 and 50 are estimates of autocorrelation and cross correlation attributes of noise terms.

(수학식 49)

(Equation 49)

And

(수학식 50)

(Equation 50)

Applying the MMSE algorithm, the estimated data can be represented according to equation (51).

(수학식 51)here,

(Equation 51)

은 여전히 순환 행렬이며, 추정된 데이터는 수학식 52에 따라 결정될 수 있다.

Is still the circulant matrix, and the estimated data can be determined according to Eq.

(수학식 52)

(Equation 52)

If the receive antennas are located near each other, the noise terms can be correlated temporally and spatially. As a result, some performance deterioration may result.

Multiple chip rate sampling (oversampling) equalization

In the following, description will be made of embodiments using a sliding window based equalization approach using multiple chip rate sampling. Multiple chip rate sampling is when a channel is sampled at a sampling rate that is an integer multiple of the chip rate, such as twice, three times, and the like. Although the following focuses on sampling twice the chip rate, these approaches can be applied to other multiples as well.

이다. 이 벡터는 재정렬되어, 짝수 수신 벡터

이고, 홀수 수신 벡터

로 분리될 수 있다. 여기서,

이다. 일반성을 잃지 않고, 데이터 전송 모델은 수학식 53에 따른다.With a sliding window of N chips wide and 2x chip rate sampling, the receive vector

to be. This vector is rearranged so that even reception vector

, Odd-receive vector

Can be separated. here,

to be. Without losing generality, the data transfer model follows Equation 53.

(수학식 53)

(Equation 53)

Equation 53 separates the effective two-sample discrete-time channel per chip into two chiprate discrete-time channels.

및

는 짝수 및 홀수 채널 응답 행렬에 대응한다. 이들 행렬들은 짝수 및 홀수 응답 벡터

및

로부터 구성되며, 이들은, 칩당 2 샘플로 채널 응답을 샘플링하고 이를 짝수 및 홀수 채널 응답 벡터들로 분리함으로써 얻어진다.Matrix in Equation 53

And

Correspond to even and odd channel response matrices. These matrices are even and odd response vectors

And

Are obtained by sampling the channel response at two samples per chip and separating it into even and odd channel response vectors.

인 화이트 노이즈로서 모델링된다.Channel noise is distributed according to Equation 54

It is modeled as in white noise.

(수학식 54)

(Equation 54)

If the channel is an additional white Gaussian noise (AWGN) channel and the received data is provided directly from the sampled channel, Equation 55 is as follows.

의

번째 요소는 수학식 56을 따른다.As a result, the problem is mathematically similar to that of the chip-rate equalizer for two receive antennas with uncorrelated noise, as described above. In many implementations, however, received antenna signals are processed by receive-side root-raised cosine (RRC) before being provided to digital receiver logic for further processing. According to this process, the received noise vector is no longer white, but has a raised-cosine (RC) autocorrelation function. RC is the frequency domain square of the RRC response. Since the RC pulse is a Nyquist pulse, Equation 54 is valid but Equation 55 is not valid. procession

of

The first element follows equation (56).

(수학식 56)

(Equation 56)

는 유니티-심볼-타임 정규화된 RC 펄스 형상이다.

Is the unity-symbol-time normalized RC pulse shape.

의 속성은 실수(real), 대칭, 및 토플리츠이며, 띠형이 아니고, 제로 엔트리를 갖지 않으며, 그 엔트리들은, 주 대각선으로부터 떨어질수록 더 작아지며 0이 되는 경향이 있다.

The attributes of are real, symmetrical, and topless, are not band-like, have zero entries, and the entries tend to become smaller and zero away from the main diagonal.

은 전체 노이즈 벡터의 상호상관을 나타내며, 수학식 57에 따른다.

Denotes the cross-correlation of the entire noise vector, and follows Equation 57.

(수학식 57)

(Equation 57)

Exact year

The exact solution to the linear least mean square estimating problem of d from the observation of r is given by equation (58).

은 화이트닝 정합된 필터링(WMF)이고,here,

Is whitening matched filtering (WMF),

은 선형 MMSE 등화이다. (수학식 58)

Is a linear MMSE equalization. (Equation 58)

와

는 어느 것도 토플리츠가 아니며,

의 구조에 기인하여, 기본 단위 연산(elemental unitary operation; 예를 들어, 행/열 재배열)을 통해 토플리츠로 만들어질 수 없다. 따라서, 토플리츠 행렬의 순환 근사화에 기초한 DFT-기반의 방법들은 여기서는 적용될 수 없으며 정확한 해는 고도로 복잡하다.

Wow

None of them is topless,

Due to the structure of, it cannot be made topless via elementary unitary operation (e.g., row / column rearrangement). Thus, DFT-based methods based on the cyclic approximation of the Topless matrix cannot be applied here and the exact solution is highly complex.

Two embodiments are described for deriving an efficient algorithm to solve this problem. The first embodiment uses simple approximation and the second embodiment uses a nearly accurate solution.

Simple approximation

와

간의 상관(correlation)을 무시한다.

. 그 결과, 다수의칩-레이트 수신 안테나와 동일한 접근법이 사용된다.Simple approximation

Wow

Ignore the correlation between

. As a result, the same approach as with many chip-rate receive antennas is used.

The complexity of this simple approximation approach is N-chip data blocks are considered. For a rough approximation, the N-point DFT complexity, given by NlogN operations (ops) per second, is assumed. In addition, N-point vector multiplication is assumed to take N ops, and vector additions are ignored.

The DFT-based complexity can be divided into roughly two components: processing that must be performed on every received data set and processing performed when the channel estimate is updated. The latter processing is done one to two orders of magnitude less frequently than the former processing operation.

For the processing performed on each received data set, the following operations are performed: a 2N-point DFT that transforms the received vector into the frequency domain; 2N-point vector multiplication (multiply each received vector by the appropriate "state" vector); And one or more DFTs for converting the results back to the time domain. Thus, the appropriate complexity is in accordance with (59).

(수학식 59)

(Equation 59)

For the processing performed when the channel response is updated, the following operations are performed: two DFT operations, six N-point vector multiplications, and a vector division that is about 10 times the vector multiplication operation. Thus, the complexity of this step is given roughly according to equation (60).

(수학식 60)

(Equation 60)

Almost accurate solution

이 되도록, 그들의 자연순(natural order)에 따라 재정렬된다. 수학식 61은 자연순 모델이다.For an almost accurate solution using the block-toplit solution, vectors and matrices

To be reordered according to their natural order. (61) is a natural order model.

는,

인 것으로 정의된다.here,

Is,

Is defined as

(Equation 61)

는

의 i번째 행(row)이고,

는

의 i번째 행이다.

는 그 첫번째 행이

이고 그 두번째 행이

인 2×N 행렬이다. 행-x, 열-y 요소

로서

를 사용하는 것이 수학식 62에 도시된 블럭-토플리츠이다.

Is

I row of,

Is

Is the i th row of.

Has that first row

And the second row

Is a 2 × N matrix. Row-x, column-y elements

as

Is a block-toplit shown in (62).

이라 가정. (수학식 62)here,

I suppose. (62)

의 블럭-토플리츠 구조는

및

및 행-재배열의 토플리츠 구조로부터 직접적으로 따른다. I 및

의 토플리츠 구조로부터, 재정의된 문제에서 노이즈의 자기상관 행렬도 역시 토플리츠이다. 이 행렬은 역시 대칭이기 때문에, 수학식 63에 따라 다시 표현할 수 있다.

Block-toplit structure

And

And directly from the topless structure of the row-rearrangement. I and

From the toplit structure of, the autocorrelation matrix of the noise in the redefined problem is also toplit. Since this matrix is also symmetric, it can be represented again according to Equation 63.

는 그 속성이

인 2×2 행렬이다. (수학식 63)here,

Has that property

Is a 2x2 matrix. (Equation 63)

행렬도 역시 띠형이기 때문에,

의 블럭 순환 근사화는 직접 얻어진다. 그러나,

는 띠형이 아니고, 그에 따라 직접적으로 블럭-순환 근사화를 생성할 수 없다.

의 요소들은 주 대각선으로부터 멀어지는 경향이 있으므로,

의 띠형 근사화는 수학식 64에 따른다.Subsequently, a block-cyclic approximation is generated for the Topless matrix.

Since the matrix is also a band,

The block cyclic approximation of is obtained directly. But,

Is not a band and therefore cannot directly generate block-cyclic approximation.

Elements of tend to deviate from the main diagonal,

The band approximation of is in accordance with (64).

는 그 속성이 이하와 같은 2×2 행렬이다.here,

Is a 2x2 matrix whose attributes are

이면,

이고, 그 외에는

.if

If,

, Otherwise

.

(Equation 64)

은 선택되는 설계 파라미터이다. RC 펄스 형상의 감쇄 속성으로 인해, 단지 몇개의 칩일 가능성이 높다. 이제

는 띠형 블럭-토플리츠이고, 이에 대한 순환 근사화가 생성된다.Noise-covariance-bandwidth,

Is the design parameter selected. Due to the attenuation nature of the RC pulse shape, it is likely to be only a few chips. now

Is a band-shaped block-toplit, with which a cyclic approximation is generated.

와

의 순환 근사화는 각각

와

이다.

은 n-포인트 DFT 행렬을 가리킨다. 즉, x가 n-벡터이면,

는 x의 DFT이다. 블럭-순환 행렬 C는 수학식 65의 형태이다.

Wow

Circular approximation of each

Wow

to be.

Denotes an n-point DFT matrix. That is, if x is an n-vector,

Is the DFT of x. The block-circulation matrix C is in the form of equation (65).

는

행렬이고, 따라서

는

이다.here,

Is

Matrix, so

Is

to be.

(Equation 65)

는 또한 수학식 66으로 표현될 수도 있다.

May also be represented by Equation 66.

은,

으로 정의되는 블럭-N-DFT 행렬이다.here,

silver,

Is a block-N-DFT matrix defined by.

(Equation 66)

은

에 의존하는 블럭 대각 행렬이며 수학식 67에 의해 주어진다.

silver

It is a block diagonal matrix that depends on and is given by Eq.

(수학식 67)

(Equation 67)

는

행렬이다.

를 완벽하게 명시하기 위해,

는

의

번째 요소를 가리키며,

로서 정의된다.

는

의

번째 요소이고,

로서 정의된다.

는 M-포인트 DFT이며 수학식 68에 따른다.

Is

It is a matrix.

To make it completely clear,

Is

of

Point to the second element,

It is defined as

Is

of

Element,

It is defined as

Is an M-point DFT and is according to equation (68).

(수학식 68)

(Equation 68)

는

을 계산할 것을 요구받는다.Equations 66-68 specify block-DFT representations of square block circular matrices.

Is

You are asked to calculate

The MMSE estimator can be rewritten according to equation (69).

(수학식 69)

(Equation 69)

The MMSE estimator form according to (68) has several advantages. This requires only one inverse computation and therefore only one vector division in the DFT domain. This potentially offers significant savings, because division is highly complex.

이 결정된다). 매 데이터 블럭에 대해, 이 필터는 수신된 데이터 블럭에 적용된다. 이러한 분할은, 수신된 데이터 블럭 프로세싱에 비해 채널이 덜 빈번하게 갱신되기 때문에 이용된다. 따라서 전체 프로세스를 이들 2개 단계들로 분리함으로써 상당한 복잡도 감소가 얻어질 수 있다.The near-exact solution has two steps in the best embodiment, although other approaches may be used. Each time a new channel estimate is obtained. The channel filter is updated (

Is determined). For every data block, this filter is applied to the received data block. This partitioning is used because the channel is updated less frequently than the received data block processing. Thus a significant complexity reduction can be obtained by separating the entire process into these two steps.

의 DFT는, 노이즈 분산

에 의해 곱해지는 펄스 성형 필터의 DFT이다. 펄스 성형 필터는 전형적으로 시스템의 고정된 특징이기 때문에, 그 DFT는 미리 계산되어 메모리에 저장될 수 있다. 따라서, 값

만이 갱신된다. 펄스-성형 필터는 "이상적인" (IIR) 펄스 형상에 가까우므로, 이상적 펄스 형상의 DFT가

에 대해 사용될 수 있어, 복잡도를 감소시키고, 캐리어로부터 멀리 떨어질 수도 있다.

DFT, noise variance

It is the DFT of the pulse shaping filter multiplied by. Because pulse shaping filters are typically a fixed feature of the system, their DFTs can be precomputed and stored in memory. Thus, the value

Only is updated. The pulse-shaping filter is close to the "ideal" (IIR) pulse shape, so the ideal pulse shape DFT

Can be used to reduce complexity and be far from the carrier.

For the channel update phase, the following operations are performed:

의 "블럭-DFT"가 계산될 필요가 있다. 블럭은 그 폭이 2이기 때문에, 2회의 DFT를 요구한다. 그 결과는, 그 행들이

와

의 DFT인 N×2 행렬이다.One.

The "block-DFT" needs to be calculated. Since the block is 2 in width, it requires two DFTs. The result is that the rows

Wow

Is an N × 2 matrix that is the DFT of.

의 "블럭-DFT"는 요소별 자기상관을 발견한 다음

와

의 상호상관에 의해 계산된다. 이것은 6N회의 복소수 곱셈과 2N회의 복소수 덧셈을 요구했다: N개의 2×2 행렬들의 곱들이 그들 자신의 헤르메시안 트랜스포즈를 이용하여 계산된다.2.

"Block-DFT" finds elemental autocorrelation

Wow

Calculated by cross-correlation of This required 6N complex multiplications and 2N complex additions: the products of N 2 × 2 matrices are computed using their own Hermesian transpose.

의 블럭-DFT가 가산된다. 이것은 3N회의 곱셈(RRC 필터의 저장된 블럭-DFT를

으로 스케일링)과 2개 행렬의 블럭-DFT를 더하는 3N회의 덧셈을 요구한다.3.

The block-DFT of is added. This is a 3N multiplication (stored block-DFT of RRC filter

3N additions are required to add the scaling and the two matrixes of block-DFT.

의 역이 블럭-DFT 도메인에서 취해진다. 이를 위해, N개의 2×2 행렬들 각각의 역이 블럭-DFT 도메인에서 취해진다. 연산의 총 갯수를 추정하기 위해, 헤르메시안 행렬

을 고려해 보자. 이 행렬의 역은 수학식 70에 따라 주어진다.4.

The inverse of is taken in the block-DFT domain. For this purpose, the inverse of each of the N 2x2 matrices is taken in the block-DFT domain. Hermesian matrix to estimate the total number of operations

Consider. The inverse of this matrix is given according to equation (70).

(수학식 70)

(Equation 70)

Thus, the complexity of calculating each inverse includes three real multiplications, one real subtraction (about one complex multiplication), and one real division.

의 블럭-DFT에 의해 블럭-곱셈되고, (

는 헤르메시안이 아니기 때문에) 이것은 총 8N회의 곱셈 + 4N회의 덧셈을 요구한다.5. The result is

Block-multiplied by the block-DFT of (

Is not a Hermesian), this requires a total of 8N multiplications + 4N additions.

In summary, the following calculation is required: two N-point DFTs; 18 N complex calculations (17 N-point vector multiplication + N singular multiplication); And 1N real divisions.

The complexity of processing a data block r of 2N values (N chip lengths) includes two N-point DFTs; One-time multiplication of the N-point block DFT (filter and data), requiring 8N complex multiplications and 4N complex additions; And one N-point inverse DFT.

In summary, the following operation is required: three N-point DFTs; 8N complex multiplications (8 N-point vector multiplications); And 4N complex additions (4 N-point vector additions).

Multiple chip rate sampling and multiple receive antenna equalization

번째 안테나에 대한 채널 행렬은

및

로 표기되고,

및

는 이와 같은 행렬의 n번째 행을 나타낸다. 각각의 채널 행렬은 토플리츠이고 적절한 행 재배열을 통해 조인트 채널 행렬은 수학식 71에 따른 블럭-토플리츠 행렬이다.The following are embodiments using a plurality of chip rate sampling and a plurality of receive antennas. For L receive antennas, a 2L channel matrix—one “even” and one “odd” matrix—for each antenna, results.

The channel matrix for the first antenna

And

Denoted by,

And

Represents the nth row of such a matrix. Each channel matrix is topless and with appropriate row rearrangement the joint channel matrix is a block-toplit matrix according to equation (71).

(수학식 71)

(71)

는

의 토플리츠 블럭이다. 각각의

는

행렬이다.procession

Is

This is a topless block of. Each

Is

It is a matrix.

The estimation of the vector d from the received observation r is modeled according to (72).

(수학식 72)

(72)

The MMSE estimation formula follows Equation 73.

(수학식 73)

(73)

은 노이즈 벡터 n의 공분산이다. 수학식 73의 해의 형태는

에 대해 이루어진 가정에 따라 달라진다. 다수의 수신 안테나의 도입은 추가적인 공간 차원(spatial dimension)을 도입한다. 비록 시간적 및 공간적 상관의 상호작용은 지극히 복잡하지만, 노이즈의 공간적 상관 속성은, 수학식 74에 따라 2개의 직접적 곱으로 표시되는 경우를 제외하고는, 시간적 상관 속성과 상호작용하지 않는다고 가정할 수 있다.

Is the covariance of the noise vector n. The form of the solution of equation (73) is

It depends on the assumptions made about. The introduction of multiple receive antennas introduces additional spatial dimensions. Although the interaction of temporal and spatial correlation is extremely complex, it can be assumed that the spatial correlation property of noise does not interact with the temporal correlation property, except when represented by two direct products according to (74). .

(수학식 74)

(74)

는 수학식 57에 따라 신호 안테나에서 관측되는 노이즈의 노이즈 공분산 행렬이다.

의 차원은

이다.

는 정규화된 동기 공간 공분산 행렬이다. 즉, L개 안테나들에서 동시에 관측되며 주 대각선 상에서 1을 갖도록 정규화된 L개 노이즈 샘플들간의 공분산 행렬이다.

는 크로넥커 곱(Kroenecker product)을 나타낸다.

Is a noise covariance matrix of noise observed at the signal antenna according to Equation 57.

The dimension of

to be.

Is a normalized sync space covariance matrix. That is, it is a covariance matrix between L noise samples observed simultaneously at L antennas and normalized to have 1 on the main diagonal.

Denotes the Kroenecker product.

은

헤르메시안 포지티브 반유한(Hermetian positive semi-definite) 행렬로서,

블럭들을 갖는 블럭-토플리츠이다. 데이터를 추정하기 위해, 4개의 양호한 실시예들이 기술된다: 정확한 해; L개 수신 안테나는 상관되지 않은 노이즈를 갖는다는 가정에 의한 단순화; 동일한 안테나로부터의 짝수 및 홀수 스트림의 시간적 상관의 무시에 의한 단순화; 및 모든 2L개 칩-레이트 노이즈 스트림은 상관되지 않는다는 가정에 의한 단순화.

silver

Hermetian positive semi-definite matrix,

Block-toplit with blocks. To estimate the data, four preferred embodiments are described: exact solution; Simplification by the assumption that L receive antennas have uncorrelated noise; Simplification by ignoring the temporal correlation of even and odd streams from the same antenna; And simplification by the assumption that all 2L chip-rate noise streams are uncorrelated.

The complexity of DFT-based processing using cyclic approximation can be divided into two components: channel estimation processing that does not need to be done for every new data block, and processing of the data itself that is performed for every data block. In all four embodiments, the complexity of data processing is 2L forward N-point DFTs; 2LN complex multiplications; And one N-point DFT. The complexity of channel estimation processing is different for each embodiment.

를 계산하기 위한, N회의 2L×2L 행렬 곱셈 + N회의 2L×2L 행렬 덧셈;

의 역변환을 계산하기 위한, N회의 2L×2L 행렬 역변환; 및 실제 필터를 생성하기 위한 N회의 2L×2L 행렬 곱셈.For the correct MMSE solution, the complexity of calculating the “MMSE filter” from the channel estimate is as follows: 2L N-point DFTs;

N 2L × 2L matrix multiplications plus N 2L × 2L matrix additions to calculate;

N inverse 2L × 2L matrix inverse transforms for calculating the inverse transform of? And N 2L × 2L matrix multiplications to produce the actual filter.

The main contributor to the overall complexity of this process is the matrix inverse transform step, where an inverse transform of a 2L × 2L matrix must be taken. This complexity can be reduced by various assumptions about the uncorrelated nature of noise, as listed below.

은 대각 행렬로 축소되고, 문제는 공간적으로 상관되지 않는 노이즈를 갖는 2L개의 안테나에서의 칩-샘플링당-단일-샘플링과 동일해진다. 그 결과, 행렬 역변환의 연산은 나눗셈으로 전환되는데, 이것은 모든 포함된 행렬들이 토플리츠이기 때문이다.1. Assuming that noise is not correlated both temporally (even / odd samples) and spatially (over antennas),

Is reduced to a diagonal matrix, and the problem is the same as per chip-sampling-single-sampling at 2L antennas with spatially uncorrelated noise. As a result, the operation of the matrix inverse transform is converted to division, because all the included matrices are topless.

2. Assuming that noise is not spatially correlated, the inverse of the matrix involved is the same as for a 2x2 matrix.

3. Assuming that spatial noise correlation is maintained but odd / even streams are not temporally correlated, the inverse of the included matrices is L × L.

Claims (70)

A method for data estimation in a wireless communication system, Generating a received vector; Processing the received vector using a sliding window based approach in which a plurality of windows are processed, wherein for each of the plurality of windows: Convert the non-Toeplitz channel response matrix to a Toeplitz matrix, Convert the topless matrix into a cyclic channel response matrix, Processing the received vector using the cyclic channel response matrix in a Discrete Fourier Transform based approach to estimate a data vector corresponding to the window; And Combining the data vectors estimated in each window to form a combined data vector Data estimation method comprising a. The method of claim 1, wherein the received vector is generated by sampling at a multiple of a chip rate. 3. The method of claim 2, wherein the received vector is processed by a root-raised cosine filter prior to processing the plurality of windows. 4. The method of claim 3, wherein the processing of the plurality of windows ignores correlation between noise associated with samples of each of the multiples of the chip rate. 4. The sliding window based approach uses a channel response matrix and a receive vector arranged in natural order, wherein the arranged channel response matrix is a block topless matrix, the natural order being And the elements of the received vector and the channel response matrix are in the order actually received. The method of claim 1, wherein the method is applied to a series of receive vectors more frequently than channel filtering. 2. The method of claim 1, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And based on the assumption that it is not correlated across the multiple antennas. 2. The method of claim 1, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Not correlated, but based on the assumption that they are correlated across the multiple antennas. 2. The method of claim 1, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Is correlated and is not correlated across the multiple antennas. 2. The method of claim 1, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And the assumption of being correlated across the plurality of antennas. The method of claim 1, wherein a discrete Fourier transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and a discrete Fourier transform of the cross-correlation of the noise vector is used. To determine, the discrete Fourier transform of the pulse shaping filter is predetermined and multiplied by the measured noise variance. The method of claim 1, wherein an approximation of the Discrete Fourier Transform of the cross-correlation of the noise vectors in the processing of the plurality of windows is used, a pulse shaping filter is used to process the received signals, and the Discrete Fourier of the cross-correlation of the noise vector. The discrete Fourier transform of the ideal pulse shape is multiplied by the measured noise variance to determine an approximation of the transform. In a wireless transmit / receive unit (WTRU), An input for receiving a reception vector; A channel equalizer for processing the received vector using a sliding window-based approach in which a plurality of windows are processed, for each of the plurality of windows, converting a non-Toplits channel response matrix into a Topless matrix, Use the cyclic channel response matrix in a Discrete Fourier Transform-based approach to transform the topless matrix into a cyclic channel response matrix, estimate the data vector corresponding to the window, and form each window to form a combined data vector. Combining the data vectors estimated at Wireless transmit and receive unit (WTRU) comprising a. 14. The WTRU of claim 13 further comprising a sampling device for sampling the multiples of the chip rate to generate the received vector. 15. The WTRU of claim 14 wherein the received vector is processed by a square root-raised cosine filter prior to processing the plurality of windows. 16. The WTRU of claim 15 wherein the processing of the plurality of windows ignores correlation between noise associated with samples of each of the multiple rates of the chip rate. The method of claim 15, wherein the sliding window based approach uses a channel response matrix and a receive vector arranged in natural order, wherein the arranged channel response matrix is a block topless matrix, and wherein the natural order is the receive vector and channel. WTRU, wherein the elements of the response matrix are in the order actually received. 14. The WTRU of claim 13 wherein the channel equalizer is configured to process the received vector more frequently than channel filtering. 14. The method of claim 13, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And a hypothesis that the data is not correlated across the multiple antennas. 14. The method of claim 13, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And is based on the assumption that they are correlated across the multiple antennas. 14. The method of claim 13, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Is correlated and is not correlated across the multiple antennas. 14. The method of claim 13, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And a hypothesis that the information is correlated across the multiple antennas. 14. The method of claim 13, wherein a discrete Fourier transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and a discrete Fourier transform of the cross-correlation of the noise vector is used. To determine, the discrete Fourier transform of the pulse shaping filter is predetermined and multiplied by the measured noise variance. 14. The method of claim 13, wherein an approximation of the Discrete Fourier Transform of the cross correlation of noise vectors in the processing of the plurality of windows is used, and a pulse shaping filter is used to process the received signals, and the Discrete Fourier of the cross correlation of the noise vectors is used. A wireless transmit / receive unit (WTRU), wherein the discrete Fourier transform of the ideal pulse shape is multiplied by the measured noise variance to determine an approximation of the transform. In a wireless transmit / receive unit (WTRU), An input for receiving a reception vector; Means for processing the received vector using a sliding window based approach in which a plurality of windows are processed, for each of the plurality of windows, Means for converting a non-Toplice channel response matrix into a Topless matrix, Means for converting the topless matrix into a cyclic channel response matrix; Means for using the cyclic channel response matrix in a Discrete Fourier Transform based approach to estimate a data vector corresponding to the window. Means for processing the received vector; And Means for combining the estimated data vector in each window to form a combined data vector Wireless transmit and receive unit (WTRU) comprising. 27. The WTRU of claim 25 further comprising means for generating the receive vector by sampling at a multiple of a chip rate. 27. The WTRU of claim 26 wherein the received vector is processed by a square root-raised cosine filter prior to processing of the plurality of windows. 28. The WTRU of claim 27 wherein the processing of the plurality of windows ignores correlation between noise associated with samples of each of a multiple of the chip rate. 28. The method of claim 27, wherein the sliding window based approach uses a channel response matrix and a receive vector arranged in natural order, wherein the arranged channel response matrix is a block topless matrix, and wherein the natural order is the receive vector and channel. WTRU, wherein the elements of the response matrix are in the order actually received. 27. The WTRU of claim 25 wherein the means for processing the received vector is configured to process the received vector more frequently than channel filtering. 27. The method of claim 25, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of chip rate, the sliding window based approach wherein noise is for each sample of the multiple rate of chip rate. And an assumption that it is not correlated across the plurality of antennas. 27. The method of claim 25, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample of the multiple rate of a chip rate. And is based on the assumption that they are correlated across the multiple antennas. 27. The method of claim 25, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample of the multiple rate of a chip rate. Is correlated and is not correlated across the multiple antennas. 27. The method of claim 25, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample of the multiple rate of a chip rate. And a hypothesis that the information is correlated across the multiple antennas. 26. The discrete Fourier transform of claim 25, wherein a discrete Fourier transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and determine a discrete Fourier transform of the noise vector cross-correlation. To determine a discrete Fourier transform of the pulse shaping filter in advance and multiply the measured noise variance. 26. The method of claim 25, wherein an approximation of the Discrete Fourier Transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and the Discrete Fourier of the cross-correlation of the noise vector. A wireless transmit / receive unit (WTRU), wherein the discrete Fourier transform of the ideal pulse shape is multiplied by the measured noise variance to determine an approximation of the transform. In the base station, An input for receiving a reception vector; A channel equalizer for processing the received vector using a sliding window-based approach in which a plurality of windows are processed, for each of the plurality of windows, converting a non-Toplits channel response matrix into a Topless matrix, Use the cyclic channel response matrix in a Discrete Fourier Transform-based approach to transform the topless matrix into a cyclic channel response matrix, estimate the data vector corresponding to the window, and form each window to form a combined data vector. Combining the data vectors estimated at Base station comprising a. 38. The base station of claim 37, further comprising a sampling device for sampling the multiples of the chip rate to generate the received vector. 39. The base station of claim 38 wherein the received vector is processed by a square root-raised cosine filter prior to processing the plurality of windows. 40. The base station of claim 39 wherein the processing of the plurality of windows ignores correlation between noise associated with samples of each of the multiples of the chip rate. 40. The method of claim 39, wherein the sliding window based approach uses a channel response matrix and a receive vector arranged in natural order, wherein the arranged channel response matrix is a block topless matrix, wherein the natural order is the receive vector and channel. Wherein the elements of the response matrix are in the order actually received. 38. The base station of claim 37 wherein the channel equalizer is configured to process the receive vector more frequently than channel filtering. 38. The method of claim 37, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And baseline that is not correlated across the multiple antennas. 38. The method of claim 37, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Relative to the plurality of antennas. 38. The method of claim 37, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Is correlated and is not correlated across the multiple antennas. 38. The method of claim 37, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And base the assumption that they are correlated across the multiple antennas. 38. The method of claim 37, wherein a discrete Fourier transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and a discrete Fourier transform of the cross-correlation of the noise vector is used. To determine, the discrete Fourier transform of the pulse shaping filter is predetermined and multiplied by the measured noise variance. 38. The method of claim 37, wherein an approximation of the discrete Fourier transform of the cross-correlation of noise vectors in the processing of the plurality of windows is used, and a pulse shaping filter is used to process the received signals, and the discrete Fourier of the cross-correlation of the noise vector is used. The base station, wherein the discrete Fourier transform of the ideal pulse shape is multiplied by the measured noise variance to determine an approximation of the transform. In the base station, An input for receiving a reception vector; Means for processing the received vector using a sliding window based approach in which a plurality of windows are processed, for each of the plurality of windows, Means for converting a non-Toplice channel response matrix into a Topless matrix, Means for converting the topless matrix into a cyclic channel response matrix; Means for processing the received vector, comprising means for using the cyclic channel response matrix in a Discrete Fourier Transform based approach to estimate a data vector corresponding to the window; And Means for combining the estimated data vector in each window to form a combined data vector Base station comprising a. 50. The base station of claim 49 further comprising means for sampling at a multiple of a chip rate to generate the receive vector. 51. The base station of claim 50 wherein the received vector is processed by a square root-raised cosine filter prior to processing the plurality of windows. 53. The base station of claim 51 wherein the processing of the plurality of windows ignores correlation between noise associated with samples of each of the multiples of the chip rate. 53. The method of claim 52, wherein the sliding window based approach uses a channel response matrix and a receive vector arranged in natural order, wherein the arranged channel response matrix is a block topless matrix, wherein the natural order is the receive vector and channel. Wherein the elements of the response matrix are in the order actually received. 50. The base station of claim 49 wherein the means for processing the received vector is configured to process the received vector more frequently than channel filtering. 50. The method of claim 49, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach provides noise for each sample of the multiple rate of the chip rate. And an assumption that it is not correlated across the plurality of antennas. 50. The method of claim 49, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Base station is based on the assumption that it is not correlated, but is correlated across the multiple antennas. 50. The method of claim 49, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. Is correlated and is not correlated across the multiple antennas. 50. The method of claim 49, wherein the receive vector comprises samples received from multiple receive antennas at a multiple rate of a chip rate, wherein the sliding window based approach allows noise to be applied to each sample at multiple rates of the chip rate. And base the assumption that they are correlated across the multiple antennas. 50. The method of claim 49, wherein a discrete Fourier transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and a discrete Fourier transform of the cross-correlation of the noise vector is used. To determine, the discrete Fourier transform of the pulse shaping filter is predetermined and multiplied by the measured noise variance. 50. The method of claim 49, wherein an approximation of the Discrete Fourier Transform of the cross-correlation of noise vectors is used in the processing of the plurality of windows, a pulse shaping filter is used to process the received signals, and the Discrete Fourier of the cross-correlation of the noise vector. The base station, wherein the discrete Fourier transform of the ideal pulse shape is multiplied by the measured noise variance to determine an approximation of the transform. An integrated circuit for estimating data from a received vector, comprising: A first input configured to receive a receive vector; A Fourier transform device for determining a Fourier transform of the received vector; A second input configured to receive a channel response matrix; A topless approximation device for determining a topless approximation of the channel response matrix; A cyclic approximation device for determining a cyclic approximation to a topless approximation of the channel response matrix; A Hermesian apparatus for determining a Hermetian of the cyclic approximation; A cross-correlation matrix determination device for determining a cross correlation matrix using the cyclic approximation and the Hermesian of the cyclic approximation; A first diagonal determination device that uses a column of the Hermesian of the cyclical approximation to generate a first diagonal matrix; A second diagonal determination device that uses a column of the cross-correlation matrix to produce a second diagonal matrix; A multiplier for multiplying the first diagonal matrix, the second diagonal matrix, and a Fourier transform of the received vector; And Inverse Fourier transform apparatus for determining an inverse Fourier transform of the result of the multiplier to produce an estimate of the data vector Integrated circuit comprising a. A data estimation method in a wireless communication system for a receiver having a plurality of receive antennas, For each antenna, generating a receive vector and a channel response matrix; Processing the received vector using a sliding window based approach in which a plurality of windows are processed, wherein for each of the plurality of windows: Convert each non-Toplice channel response matrix into a Topless matrix, Convert each topless matrix into a cyclic channel response matrix, Combine the cyclic channel response matrices to form a combined cyclic channel response matrix, Processing the received vector using a combined received vector comprising each received vector and the combined cyclic channel response matrix in a discrete Fourier transform based approach to estimate a data vector corresponding to the window. ; And Combining the data vectors estimated in each window to form a combined data vector Data estimation method comprising a. A data estimation method in a wireless communication system for a receiver using multiple chip rate sampling, For each multiple of the chip rate, generating a receive vector and a channel response matrix; Processing the received vector using a sliding window based approach in which a plurality of windows are processed, wherein for each of the plurality of windows: Convert each non-Toplice channel response matrix into a Topless matrix, Convert each topless matrix into a cyclic channel response matrix, Combine the cyclic channel response matrices to form a combined cyclic channel response matrix, Processing the received vector using the combined cyclic channel response matrix and the combined received vector including each received vector in a discrete Fourier transform based approach to estimate a data vector corresponding to the window; And Combining the data vectors estimated in each window to form a combined data vector Data estimation method comprising a. 64. The method of claim 63, wherein a receive square root-raised cosine (RRC) filter is used and the sliding window based approach ignores the correlation of noise between each sample of multiple rates of the chip rate. . A data estimation method in a wireless communication system for a receiver using multiple chip rate sampling and a receive square root-raised cosine filter, For each multiple of the chip rate, generating a receive vector and a channel response matrix; Processing the received vector using a sliding window-based approach in which a plurality of windows are processed after a receive square root-raised cosine filter process, wherein for each of the plurality of windows: Provide a combined received vector with the elements of the received vectors in the order in which each element is actually received, Providing a combined channel response matrix of a block topless structure having the rows and columns of the channel response matrices, in an order such that the same rows and columns of the channel response matrices are grouped together in a combined channel response matrix, Transforming the combined channel response matrix to block block combined channel response matrix, Processing the received vector using the combined block cyclic channel response matrix and a combined received vector containing each received vector in a discrete Fourier transform based approach to estimate a data vector corresponding to the window; And Combining the data vectors estimated in each window to form a combined data vector Data estimation method comprising a. A data estimation method in a wireless communication system for a receiver using multiple chip rate sampling and multiple receive antennas, For each combination of a receive antenna and a multiple of the chip rate, providing a receive vector and a channel response matrix; Processing the received vector using a sliding window based approach in which a plurality of windows are processed, wherein for each of the plurality of windows: Generating a combined cyclic channel response matrix using the channel response matrix; Processing the received vector using the combined cyclic channel response matrix and the combined received vector including each received vector in a discrete Fourier transform based approach to estimate a data vector corresponding to the window; And Combining the data vectors estimated in each window to form a combined data vector Data estimation method comprising a. 67. The method of claim 66, wherein the sliding window based approach uses an exact solution. 67. The method of claim 66, wherein the sliding window based approach assumes that noise between each receive antenna is not correlated. 67. The method of claim 66, wherein the sliding window based approach assumes that noise between multiples of the chip rate does not have a temporal correlation. 67. The method of claim 66, wherein the sliding window based approach assumes that there is no correlation of noise between any combination of a receive antenna and a multiple of the chip rate.

Images