JP2006246128A - Filtering method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve performance for following a propagation path variation, without reducing communication capacity or increasing computational complexity, in filtering processing to a reception signal in a wireless communication system. <P>SOLUTION: In the filtering method, serial/parallel conversion and FFT units 101 and 102 perform serial/parallel conversion and discrete inverse Fourier transformation upon a reception complex signal and a reference signal, respectively. A power spectrum estimation unit 103 estimates power spectrum based on the output of the serial/parallel conversion and FFT unit 101. A transport function estimation unit 104 estimates a transport function based on outputs of both the serial/parallel conversion and FFT units. A weight generation unit 106 generates an equalization weight and suppresses estimation error thereof based on outputs of the power spectrum estimation unit 103 and the transport function estimation unit 104. Filtering units 105 and 107 perform filtering processing upon a reception complex signal based on an output of the weight generation unit. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a filtering method and apparatus for equalizing a received signal in a wireless communication system.

  In a wireless communication system, interference and noise caused by signal delay and reflection in a propagation path cause deterioration in communication quality. As a method for improving the quality of a signal that has passed through a propagation path that causes such interference, linear filtering (equalizer) of a received signal based on an MMSE (Minimum Mean Square Error) standard is conceivable (Patent Documents 1, 2, and 5). 3). FIG. 13 shows a communication model assumed by the present invention. The transmission data is encoded and modulated by the encoder / modulator at the transmitter 400, and reaches the receiver 420 via the propagation path 410. The receiver 420 performs signal equalization with the equalizer 421, and is demodulated / decoded with the demodulation / decoding unit 422 to generate reception data.

  Linear filtering (equalizer) based on the MMSE standard can be realized by a Wiener filter. However, the tap coefficient of filtering is used as a solution of a multi-dimensional simultaneous linear equation whose coefficient is the cross-correlation of the received signal and the desired signal. Need to ask. For this reason, a complex algorithm and a large amount of calculation are required, such as correlation calculation regarding a received signal and a desired signal, and inverse matrix calculation for calculating a filter coefficient.

  As a method for reducing the amount of calculation, a sequential algorithm represented by an LMS (Least Mean Square) algorithm and an RLS (Recursive Least Mean Square) algorithm is known (Non-Patent Document 1). However, even if these algorithms are used, there is a problem that still requires a large amount of calculation.

As a method for solving such problems and realizing an equalizer, signal processing in the frequency domain can be considered (Patent Document 4, Non-Patent Document 2). Since the convolution calculation in the time domain can be realized by multiplication in the frequency domain, it is possible to reduce the amount of calculation for obtaining the autocorrelation characteristics of the received signal and the cross correlation characteristics of the received signal and the desired signal in the frequency domain. Because of such advantages, an equalizer in the frequency domain has been studied in Patent Document 4 and the like.
JP-A-6-61893 JP-A-7-1581 JP 7-297733 A JP 2003-348049 A Yoji Iiguni “Adaptive Signal Processing Algorithm”, July 19, 2000 David Falconer, "Frequency Domain Equalization for Single-Carrier Broadband Wireless Systems", IEEE Communications Magazine, April 2002

  However, the techniques disclosed in Patent Document 4 and Non-Patent Document 2 are based on the assumption that a guard interval (GI) is inserted into a transmission signal as a communication method. During the GI transmission period, information that should be transmitted cannot be transmitted, and communication capacity is sacrificed. Further, since the autocorrelation of the received signal and the cross-correlation between the received signal and the desired signal are estimated from the received signal including noise and interference, an estimation error occurs. In order to suppress such an estimation error, it is necessary to lengthen the average time for estimating the correlation, and there is a problem that the followability to the conversion of the propagation path characteristics is lowered.

  The present invention has been made in such a background, and its purpose is to improve the follow-up performance to propagation path fluctuations without causing a reduction in communication capacity or increase in calculation amount, and even when propagation path fluctuations are fast. It is an object of the present invention to provide a filtering method and apparatus in which reception quality does not deteriorate.

  A filtering method for performing a filtering process on a received signal in a wireless communication system according to the present invention includes: a first serial / parallel conversion / discrete Fourier transform step for performing a serial / parallel conversion and a discrete Fourier transform on a received complex signal; Based on the output of the second serial / parallel conversion / discrete Fourier transform step for performing serial / parallel conversion and discrete Fourier transform on the reference signal, and the output of the first serial / parallel conversion / discrete Fourier transform step, A power spectrum estimation step for performing estimation, and a transfer for estimating a transfer function based on the output of the first series / parallel conversion / discrete Fourier transform step and the output of the second series / parallel conversion / discrete Fourier transform step. Function estimation step, the output of the power spectrum estimation step and the transfer function estimation step A weight generation step for generating an equalization weight based on the output of the loop, and a filtering step for performing a filtering process on the received complex signal based on the output of the weight generation step, wherein the weight generation step includes the equalization weight The estimation error is suppressed.

  Thus, by performing the power spectrum estimation and the transfer function estimation for weight generation in the frequency domain, the required amount of calculation is reduced. Further, by suppressing the estimation error of the equalization weight, it is not necessary to lengthen the average time for estimating the correlation between the received signal and the desired signal, and the followability to propagation path fluctuation is improved.

  The weight generating step generates an equalized weight vector in the frequency domain, and the filtering step multiplies the equalized weight vector in the frequency domain by the output of the first serial / parallel transform / discrete Fourier transform step. A multiplication step, and a discrete Fourier inverse transform step for performing discrete Fourier inverse transform on the output of the multiplication step. In the frequency domain, filtering can be performed with a smaller amount of computation by multiplying equalization weights.

  When performing such filtering in the frequency domain, the first and second serial / parallel transform / discrete Fourier transform steps overlap each other in temporally adjacent blocks when the input signal sequence is blocked. Preferably, the discrete Fourier transform is performed by providing a block and the inverse discrete Fourier transform step includes a step of deleting an overlapped portion from the discrete Fourier inverse transformed output after filtering in the frequency domain. As a result, signal quality can be prevented from deteriorating due to convolution caused by convolution of the first and last signals of the block.

  The weight generation step may generate a time domain equalization weight, and the filtering step may include a convolution step of convolving the received complex signal with a time domain equalization weight. In this case, filtering is performed in the time domain, and a delay due to blocking of the received signal sequence is avoided as compared with the case where it is performed in the frequency domain.

  The error suppression in the weight generation step is to convert a vector to be determined into a time domain component, perform a predetermined threshold determination on each element of the time domain component obtained by this conversion, and reduce the element below the threshold It is preferable to set the value of 0 to 0. As described above, the processing in the time domain makes it possible to suppress the estimation error without requiring a long averaging time.

  According to the filtering method and apparatus of the present invention, the necessary calculation amount is reduced by performing power spectrum estimation and transfer function estimation in the frequency domain. Therefore, the processing load on the processor can be reduced. Further, since the guard interval is not necessary, the communication capacity is not reduced. Furthermore, since it is not necessary to lengthen the average time for estimating the correlation, the followability to propagation path fluctuation is enhanced. Therefore, the reception quality does not deteriorate even when the propagation path fluctuation is fast.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings.

In the present embodiment, the equalizer has the following functions.
(1) Equalization weight estimation by transfer function and power spectrum.
(2) Equalization weight estimation error suppression.
(3) Overlapping FFT blocks in time and filtering in the frequency domain.
(4) A coefficient part used in the convolution process is cut out from the estimated equalization weight and filtered in the time domain.
Functions (3) and (4) have the same function contents and are due to differences in processing methods. Therefore, it can be adopted alternatively.

That is, the problem of the present invention can be solved by combining these functions as follows.
Equalizer configuration by combining functions (1), (2), and (3).
Equalizer configuration by combining functions (1), (2), and (4).

  First, the principle and effect of each of these functions (1) to (4) will be described.

[(1) Equalization weight estimation by transfer function and power spectrum]
Consider a basic equalizer configuration as shown in FIG. The equalizer of FIG. 10 includes a functional unit that estimates equalization weights used for signal equalization and an equalization filter unit that convolves signals with estimated weights. The weight is estimated based on the MMSE standard using the received signal and the desired signal. In the figure, x _k is a received signal, d _k is a desired signal, and y ′ _k is an equalizer output signal. The equalization weight estimation method in the frequency domain is confirmed by performing a discrete Fourier transform on the time domain MMSE weight estimation formula.

First, each variable used for explanation is defined and explained.
Let x _{k be the} received signal sequence.
Let d _{k be the} desired signal sequence.
The equalizer output signal and y _'k.
Let the estimated equalization weight vector be w.
The relationship between the vector w and each element is defined as follows.

また、等化器入力信号のベクトルを次のように定義する。

The vector of the equalizer input signal is defined as follows.

等化器入力信号の巡回行列を次のように定義する。

The cyclic matrix of the equalizer input signal is defined as follows.

所望信号のベクトルを次のように定義する。所望信号は、送受信で既知のパイロット信号等を用いる。

The desired signal vector is defined as follows. As the desired signal, a known pilot signal or the like is used for transmission and reception.

等化器入力信号のベクトルを離散フーリエ変換したベクトルを次のように定義する。

A vector obtained by performing a discrete Fourier transform on a vector of the equalizer input signal is defined as follows.

所望信号のベクトルを離散フーリエ変換したベクトルを次のように定義する。

A vector obtained by performing a discrete Fourier transform on the vector of the desired signal is defined as follows.

等化器フィルタ係数（ウェイト）の離散フーリエ変換後のベクトルを次のように定義する。

A vector after the discrete Fourier transform of the equalizer filter coefficient (weight) is defined as follows.

等化器入力信号ｘ_kの離散フーリエ変換後の系列ｘ_kによる対角行列を次のように定義する。

The diagonal matrix by the sequence x _k after the discrete Fourier transform of the equalizer input signal x _k is defined as follows.

離散フーリエ変換行列を次のように定義する。

A discrete Fourier transform matrix is defined as follows.

  The principle and effect of the equalization weight estimation by the transfer function and power spectrum are clarified. This will be described by confirming a frequency domain weight equivalent to the MMSE weight in the time domain.

  As a relational expression for obtaining the MMSE weight, the following Wiener-Hoff equation (WH expression) is well known. Equivalent weights in the frequency domain are examined by performing a discrete Fourier transform on both sides of this equation.

Ｒ_xx,ｒ_xyは、それぞれ、受信信号の自己相関行列と、受信信号と所望信号の相互相関ベクトルである。
Ｎが十分大きい場合、Ｒ_xx, ｒ_xyは、ｘ_kの巡回行列を用いて次のように推定できる。

R _xx and r _xy are an autocorrelation matrix of the received signal and a cross-correlation vector of the received signal and the desired signal, respectively.
When N is sufficiently large, R _xx and r _xy can be estimated as follows using a cyclic matrix of x _k .

Ｒ_xx, ｒ_xyの推定値をＷ-Ｈ式に代入すると、次のように表せる。

When the estimated values of R _xx and r _xy are substituted into the WH formula, they can be expressed as follows.

両辺にＺを乗じて、離散フーリエ変換する。

Multiply both sides by Z and perform discrete Fourier transform.

Ｘは巡回行列であり、離散フーリエ変換行列によって対角化されるため次の関係が成り立つ。

X is a circulant matrix and is diagonalized by a discrete Fourier transform matrix, so that the following relationship is established.

この関係を利用すれば、式（１４）は次のように書き換えられる。

Using this relationship, equation (14) can be rewritten as follows.

さらに、Ｘ^*＝Ｚ^−１diag（ｘ^*）Ｚの関係、およびｄ＝Ｚｄ^*の関係を代入する。

Further, the relationship of X ^* = Z ⁻¹ diag (x ^* ) Z and the relationship of d = Zd ^* are substituted.

Ｗ＝Ｚｗを代入する。

Substitute W = Zw.

各要素で表すと次のようになる。

Each element is as follows.

上式より、右辺は参照信号と受信信号の積による伝達関数の推定式であり、左辺は受信信号の電力スペクトルの推定式になっていることが確認できる。
さらにｗについて解く。

From the above equation, it can be confirmed that the right side is a transfer function estimation formula based on the product of the reference signal and the reception signal, and the left side is an estimation formula for the power spectrum of the reception signal.
Solve for w.

  The above equation w is an estimated weight in the frequency domain equivalent to the MMSE weight estimated in the time domain. In the frequency domain, it was confirmed that the MMSE weight can be obtained from the transfer function and the power spectrum of the input signal. In the frequency domain, the circulant matrix of the received signal is diagonalized by discrete Fourier transform, and the inverse matrix is obtained by the reciprocal of the power spectrum, so that weight estimation can be performed with a relatively small amount of computation by multiplication and division. It became clear.

[(2) Equalization weight estimation error suppression]
As described above, the equalization weight can be estimated from the transfer function and the power spectrum. However, since the equalization weight is estimated from the received signal, an estimation error occurs due to the noise / interference component included in the received signal. Such an estimation error is undesirable because it degrades the performance of the equalizer. Normally, the estimation accuracy is improved by increasing the averaging time of the estimated value. However, if the averaging time is increased, the follow-up performance with respect to propagation path fluctuations is degraded. If the fluctuation in the propagation path cannot be sufficiently followed, of course, a large characteristic deterioration will occur.

  In the present invention, it is possible to suppress the estimation error of the equalization weight without impairing the followability to propagation path fluctuation. FIG. 4 is an explanatory diagram of the estimation error suppression method in the present invention. In the present invention, estimation errors are suppressed in the time domain for vectors such as a transfer function, a power spectrum, and an equalization weight estimated from a received signal and a desired signal. As shown in FIG. 4, each element of the estimated vector is determined by an appropriate threshold value Th, and the value of an element that falls below the threshold value is forcibly set to 0. By performing such threshold determination, it is possible to suppress the estimation error without requiring a long averaging time.

[(3) Filtering in the frequency domain by overlapping FFT blocks in time]
In order to equalize the received signal, consider filtering in the frequency domain. The received signal is filtered based on the estimated equalization weight. In the time domain, filtering is performed by convolving an equalization weight on the received signal. In the frequency domain, filtering is possible by multiplying equalization weights. However, frequency domain multiplication corresponds to time domain cyclic convolution, and when the equalization weight is multiplied by the received signal in the frequency domain, as shown in FIG. Is convolved with equalization weights. As shown in this figure, the first and last signals of the FFT block are simultaneously convolved and mixed due to the temporal spread of the equalization weight coefficient. In such a portion where the beginning and end signals of the block are simultaneously folded and mixed, the quality of the signal deteriorates.

  In order to avoid such deterioration, in the present invention, overlapping portions are provided in temporally adjacent FFT blocks as shown in FIG. 12, and the overlapping portions are deleted after filtering in the frequency domain. In this way, by giving the FFT block an overlapping portion, the influence of interference at the FFT block end due to frequency domain multiplication is avoided.

[(4) Cut out the coefficient part used in the convolution process from the estimated equalization weight and filter it in the time domain]
In order to equalize the received signal, consider filtering in the time domain. In the time domain, filtering is performed by convolving an equalization weight on the received signal. In the present invention, the equalization weight is estimated in the frequency domain, the estimated frequency domain equalization weight is converted into the time domain, and the received signal is convolved. In this case, in order to make the number of elements in the time domain of the estimated equalization weight vector coincide with the number of coefficients used in convolution, a coefficient portion used in convolution processing is cut out as shown in FIG. In this way, by cutting out the portion of the estimated equalization weight vector that is used for convolution in the time domain, it is possible to reduce useless computation during convolution processing.

Next, a specific configuration example of the embodiment according to the present invention will be described. FIG. 1 shows an example of the configuration of an equalizer according to the first embodiment of the present invention. The equalizer shown in FIG. 1 includes a first serial / parallel conversion / discrete Fourier transform unit 101 that performs serial / parallel conversion and discrete Fourier transform on a received complex signal x _k , and a predetermined reference signal c _k . The second serial / parallel conversion / discrete Fourier transform unit 102 that performs serial / parallel conversion and discrete Fourier transform on the output, and the power spectrum estimation based on the output of the first serial / parallel conversion / discrete Fourier transform unit 101. A power spectrum estimation unit 103 to perform, a transfer function estimation unit 104 to estimate a transfer function based on both outputs of the first and second serial / parallel conversion / discrete Fourier transform units 101 and 102, and a power spectrum estimation unit 103 Weight generation unit 106 that generates equalization weights based on the output ^ r _m and the output ^ h _m of the transfer function estimation unit 104, and the received complex signal based on the output of the weight generation unit 106 filtering means for performing a filtering process in the frequency domain for x _k . In this case, the weight generation unit 106 generates an equalization weight vector W _m in the frequency domain, and the filtering means includes the equalization weight vector weight generation unit in the frequency domain from the serial / parallel conversion / discrete Fourier transform unit 101 and the first. A multiplication unit 105 that multiplies the output x _{m of} one serial / parallel conversion / discrete Fourier transform unit 101 and an IFFT / parallel / serial conversion unit 107 that performs inverse discrete Fourier transform on the output y ′ _m of the multiplication unit 105. Have. The weight generation unit 106 has a function of suppressing the estimation error of the equalization weight in the time domain.

  The processing content of each functional unit in FIG. 1 will be described.

[Series / Parallel Conversion / FFT Unit]
In order to process the reception complex signal x _k in the frequency domain, it is parallelized as an FFT block. Similarly, the reference signal _ck is parallelized as an FFT block. The size of the FFT block and N _B, the block of the received signal x _k and a reference signal c _k sets the overlapping portion (overlap), as shown in Figure 5 (a). Let Np be the size of the overlapping portion.

The blocked reception signal is expressed as a vector x _m as follows. m is the block number.

ここで、Ｔはベクトルの転置を示す。以降の説明においても転置について同様の表現を用いる。

Here, T indicates transposition of the vector. In the following description, the same expression is used for transposition.

Further, the blocked reference signal _cm is defined as follows. m is the block number.

ここで^*の記号は複素共役を示す。以降の説明においても、複素共役について同様の表現を用いる。

Here, the symbol ^* indicates a complex conjugate. In the following description, similar expressions are used for complex conjugates.

x _m and _cm are subjected to FFT (Fast Discrete Fourier Transform) by serial / parallel conversion / FFT units 101 and 102, respectively. Let X _m be a vector obtained by performing FFT on the vector x _m of the equalizer input signal.

Moreover, it was FFT reference signal vector c _m vectors and C _m.

[Transfer function estimator transfer function estimator 104 obtains a reception signal vector x _m after FFT, based on the reference signal vector c _m, the transfer function estimation vector ^ h _m of the propagation path as follows. (The symbol ^ should overlap the character immediately after, but in text notation, it is added immediately before the character for convenience.)

ここで、λは、平均化用の忘却係数で０＜λ＜１である。また、diag（）は、括弧内のベクトルの各要素を対角成分とする対角行列を示す。以降の説明においても対角行列について同様の表現を用いる。

Here, λ is a forgetting factor for averaging, and 0 <λ <1. Further, diag () indicates a diagonal matrix having diagonal elements as elements of vectors in parentheses. In the following description, the same expression is used for the diagonal matrix.

The transfer function estimation unit 104 operates in accordance with the timing at which the signal used for the reference signal _cm is received. As shown in FIGS. 14 (a) and 14 (b), when a pilot signal transmitted in time division is used as a reference signal, the pilot signal operates intermittently according to the transmission timing of the pilot signal. If the pilots are code-multiplexed, they are operated continuously.

[Power Spectrum Estimator]
Power spectrum estimation section 103, from the received signal vector X _m after FFT, obtain power spectrum estimation vector ^ r _m of the received signal.

ここで、λは、平均化用の忘却係数で０＜λ＜１である。Ｈはベクトルもしくは行列の共役転置を示す。以降の説明においても共役転置について同様の表現を用いる。

Here, λ is a forgetting factor for averaging, and 0 <λ <1. H indicates the conjugate transpose of a vector or matrix. In the following description, the same expression is used for conjugate transposition.

[Wait generator]
The weight generation unit 106 has a function of suppressing an estimation error and generating an equalization weight. As the weight generation function, the processing shown in weight generation (1) or weight generation (2) shown in FIG. 3 is performed. The difference between weight generation (1) and (2) is the difference in the estimation error suppression method.

The weight generation (1) shown in FIG.
Divider 201, IFFT unit 202, threshold decision unit 203 and the FFT unit 204, transfer function estimation vector ^ h _m, based on the power spectrum estimation vector ^ r _m, determine the equalization weight vector W _m before the estimated error suppression . This vector W _m is defined as follows for each element.

ベクトルＷ_mの要素は伝達関数推定ベクトル^ｈ_m、電力スペクトル推定ベクトル^ｒ_mの各要素毎に除算し、次のように計算される。

The elements of the vector W _m by dividing each element of the transfer function estimation vector ^ h _m, power spectrum estimation vector ^ r _m, is calculated as follows.

ここで、ｋはベクトルの要素を示す番号で、ｋ＝０〜Ｎ_B−１である。
またベクトル^ｈ_m，^ｒ_mの各要素は次のように定義される。

Here, k is a number indicating a vector element, and k = 0 to N _B −1.
The vector ^ h _m, ^ each element of r _m is defined as follows.

Further, in order to suppress the estimation error included in the vector W _m , IFFT (Fast Discrete Fourier Transform) is performed to obtain a time domain equalization weight vector w _m . The vector w _m is input to the threshold determination unit. The processing content of the threshold determination unit will be described later. The output vector of the threshold determination unit is subjected to FFT to obtain a frequency domain equalization weight vector W _{m in} which the estimation error is suppressed. The weight generation function outputs the vector W _m as a frequency domain equalization weight vector.

Next, weight generation (2) will be described.
In the weight generation (2) shown in FIG. 3B, the IFFT units 211 and 212, the threshold determination units 213 and 214, the FFT units 215 and 216, and the division unit 217 perform transfer function estimation vector ^ h _m and power spectrum estimation vector. individually perform estimation error suppression respect to ^ r _m. Transfer function estimation vector ^ h _m, and IFFT respectively power spectrum estimation vector ^ r _m, respectively inputted into the threshold determination unit. The vectors output from the threshold determination unit are respectively FFTed to obtain a transfer function estimation vector ^ h ' _m and a power spectrum estimation vector ^ r' _m after the estimation error is suppressed.

The elements of the frequency domain equalization weight vector W _m are divided for each element of the transfer function estimation vector ^ h ′ _m and the power spectrum estimation vector ^ r ′ _m after estimation error suppression, and are calculated as follows.

ここで、ｋはベクトルの要素を示す番号で、ｋ＝０〜Ｎ_B−１である。
またベクトル^ｈ'_m，^ｒ'_mの各要素は次のように定義される。

Here, k is a number indicating a vector element, and k = 0 to N _B −1.
Each element of the vectors ^ h ' _m and ^ r' _m is defined as follows.

The frequency domain equalization weight vector W _m calculated as described above is used as the output of the weight generation unit.

[Threshold determination unit]
The threshold determination unit is a function common to each equalizer configuration. The types of vectors input to the threshold determination unit are propagation time estimation vectors, autocorrelation estimation vectors, and time domain equalization weight vectors. Depending on the configuration of the equalizer, the types of vectors input to the threshold value determination unit are different, but the processing contents are the same. To describe the process of the threshold value determination unit generally illustrating the input vector to the threshold determination unit is replaced with u _m. Each element is defined as follows.

Only significant elements are extracted from each element of the vector u _m . As shown in FIG. 4, the vector u _m, as shown in the figure, the ideal component and the estimated error component E are mixed. By a suitable threshold value Th, to determine the amplitude of each element of the vector u _m, and 0 elements amplitude is less than the threshold. In this way, the vector u ′ _m after the estimation error suppression is obtained. This vector u ′ _m is used as the output of the threshold value determination unit. Here, the amplitude refers to the length of a complex number that is an element of a vector.

Threshold used for the threshold determination, the entire element of the input vector u _m, the average value of the amplitude, it is conceivable to calculate the like variance.

[Multiplier]
The frequency domain signal vector X _m and the corresponding elements of the frequency domain equalization weight vector W _m are multiplied to obtain the equalized frequency domain signal vector Y ′ _m as follows.

[IFFT / serial / parallel converter]
FIG. 6 shows the processing contents by the IFFT / serial / parallel converter.
The frequency domain signal vector Y ′ _m after equalization is IFFT to obtain the time domain signal vector y ′ _m after equalization. Each element of the vector y ′ _m is expressed as follows.

[IFFT / serial / parallel converter]
For the vector, as shown in FIG. 7, the Np / 2 elements at the beginning and the end of the vector element are deleted as overlapping parts. A parallel / serial conversion is performed on the (N _B −Np) elements at the center of the vector element to obtain a complex output ^ y _k of the equalizer.

  FIG. 2 shows a configuration example of an equalizer according to the second embodiment of the present invention. The function of each functional unit in the figure will be described.

  In this configuration, filtering of the received signal is realized by convolution processing in the time domain. With such a configuration, the received signal to be filtered does not need to be subjected to FFT processing, and a delay due to FFT blocking can be avoided. Therefore, the delay time of signal processing by the equalizer can be shortened.

[Series / Parallel Conversion / FFT Unit]
In order to process the reception complex signal x _k in the frequency domain, it is parallelized as an FFT block. Similarly, the reference signal _ck is parallelized as an FFT block. The size of the FFT block and N _B, the block of the received signal and the reference signal are overlapped portion (overlapped) as shown in FIG. 5 (b) not set.

The blocked reception signal is expressed as a vector x _m as follows. m is the block number.

The blocked reference signal _cm is defined as follows. m is the block number.

x _m and _cm are subjected to FFT (fast discrete Fourier transform).
Let X _m be a vector obtained by performing FFT on the vector x _m of the equalizer input signal.

It was FFT reference signal vector c _m vectors and C _m.

[Transfer function estimator]
A received signal vector X _m after FFT, the reference signal vector C _m, obtaining a transfer function estimation vector ^ h _m of the propagation path as follows.

λ is a forgetting factor for averaging, and 0 <λ <1. Further, the transfer function estimation unit operates in accordance with the timing at which the signal used for the reference signal _cm is received. As shown in FIGS. 14 (a) and 14 (b), when a pilot signal transmitted in time division is used, the pilot signal operates intermittently in accordance with the transmission timing of the pilot signal, and the pilot signal as in CDMA. Are code-multiplexed, they are operated continuously.

[Power Spectrum Estimator]
From the received signal vector X _m after the FFT, a power spectrum estimation vector ^ r _m of the received signal is obtained.

ここに、λは、平均化用の忘却係数で０＜λ＜１である。

Here, λ is a forgetting factor for averaging, and 0 <λ <1.

[Wait generator]
The weight generation unit has a function of generating equalization weights by suppressing estimation errors. The weight generation function performs processing shown in weight generation (3) or weight generation (4) shown in FIG.

  The difference between weight generation (3) and (4) is the difference in the estimation error suppression method.

The weight generation (3) shown in FIG.
Divider 201, IFFT unit 202, threshold decision unit 203 and the convolution coefficient clipping unit 205, transfer function estimation vector ^ h _m, based on the power spectrum estimation vector ^ r _m, of the previous estimation error suppression equalization weight vector W _{Find m} .
Specifically, the vector W _m is defined for each element as follows.

The elements of the vector W _m by dividing each element of the transfer function estimation vector ^ h _m, power spectrum estimation vector ^ r _m, is calculated as follows.

ここで、ｋはベクトルの要素を示す番号で、ｋ＝０〜Ｎ_B−１である。
またベクトル^ｈ_m，^ｒ_mの各要素は次のように定義される。

Here, k is a number indicating a vector element, and k = 0 to N _B −1.
The vector ^ h _m, ^ each element of r _m is defined as follows.

Further, in order to suppress the estimation error included in the vector W _m , IFFT (Fast Discrete Fourier Transform) is performed to obtain a time domain equalization weight vector w _m . The vector w _m is input to the threshold determination unit. The processing contents of the threshold value determination unit are as described above. An element part used as a coefficient of the transversal filter of the subsequent convolution unit is cut out from the vector output from the threshold determination unit, and is set as an output vector w _m of the weight generation unit. The number of elements of the vector w _m is the same as the number of taps of transversal filtering performed in the subsequent convolution process.

The weight generation (4) shown in FIG.
The weight generating (4), similar to FIG. 3 (b), the transfer function estimation vector ^ h _m, is performed individually estimated error suppression respectively power spectrum estimation vector ^ r _m, configuration shown in FIG. 3 (b) In addition, the IFFT unit 218 and the convolution coefficient cutout unit 219 are used in the subsequent stage.

Specifically, the transfer function estimation vector ^ h _m, the power spectrum estimation vector ^ r _m and IFFT respectively, are input to respective threshold decision unit 213 and 214. The vectors output from these threshold value determination units are respectively subjected to FFT to obtain transfer function estimation vectors ^ h ' _m and power spectrum estimation vectors ^ r' _m after estimation error suppression.

The elements of the frequency domain equalization weight vector W _m are divided for each element of the transfer function estimation vector ^ h ′ _m and the power spectrum estimation vector ^ r ′ _m after estimation error suppression, and are calculated as follows.

ここで、ｋはベクトルの要素を示す番号で、ｋ＝０〜Ｎ_B−１である。

Here, k is a number indicating a vector element, and k = 0 to N _B −1.

Each element of the vectors ^ h ' _m and ^ r' _m is defined as follows.

The frequency domain equalization weight vector W _m calculated as described above is IFFT and converted into the time domain. The element part used as the coefficient of the transversal filter of the convolution part of a back | latter stage is cut out from the vector after conversion. The reason for cutting out is that the size of the equalization weight (number of vector elements) estimated up to this stage does not match the number of coefficients used in the subsequent convolution process. The size of the equalization weight estimated up to this stage is the FFT block size, which is relatively large, and the FFT block size is large with respect to the temporal spread of the equalization weight in the time domain. If it is included, useless processing increases. Therefore, only the coefficient part useful in the time domain is cut out from the estimated equalization weight element and used for the subsequent convolution process. As shown in FIG. 8, after performing such a convolution coefficient cut-out process, a vector w _m is output as the output of the weight generation unit. The number of elements of the vector w _m is the same as the number of taps of transversal filtering performed in the subsequent convolution process.

[Convolution part]
FIG. 9 shows transversal filter processing performed in the convolution unit. The received complex signal Y ′ _m is _convolved with a time domain equalization weight y ′ _m . Let the convolution result be the complex output ^ y _k of the equalizer.

[IFFT / serial / parallel converter]
As shown in FIG. 6, the frequency domain signal vector Y ′ _m after equalization is IFFT to obtain a time domain signal vector y ′ _m after equalization.

Each element of the vector y ′ _m is expressed as follows.

For the vector, as shown in FIG. 7, the first and last Np / 2 elements of the vector element are deleted as overlapping parts. Parallel / serial conversion is performed on the (N _B −Np) elements at the center of the vector element to obtain a complex output ^ y _k of the equalizer.

  The preferred embodiments of the present invention have been described above, but various modifications and changes other than those mentioned above can be made. For example, the present invention does not require a guard interval in a transmission signal, but does not exclude application of the present invention to a method and apparatus that uses a guard interval.

It is a figure which shows the structural example of the equalizer in 1st Embodiment by this invention. It is a figure which shows the structural example of the equalizer in 2nd Embodiment by this invention. It is a figure which shows the internal structural example of the weight production | generation part in embodiment of this invention. It is explanatory drawing of the estimation error suppression method in embodiment of this invention. It is explanatory drawing of the serial / parallel conversion process in embodiment of this invention. It is a figure which shows the processing content by IFFT * serial / parallel conversion part in embodiment of this invention. It is explanatory drawing of the duplication part deletion process shown in FIG. It is explanatory drawing of the convolution coefficient cutting-out process in the 2nd Embodiment of this invention. It is explanatory drawing of the convolution process (transversal filter process) in the 2nd Embodiment of this invention. It is a figure which shows a basic equalizer structure. It is explanatory drawing of the interference of cyclic convolution and an FFT block end. It is explanatory drawing of the setting of the overlap part of the FFT block in the 1st Embodiment of this invention. It is a figure which shows the communication model on which this invention presupposes. It is explanatory drawing which shows the relationship between a pilot signal and data.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 ... Parallel transformation / discrete Fourier transform unit, 102 ... Parallel transformation / discrete Fourier transform unit, 103 ... Power spectrum estimation unit, 104 ... Transfer function estimation unit, 104 ... Transfer function estimation unit, 105 ... Multiplication unit, 106 ... Weight generation 107: serial conversion unit, 201 ... division unit, 202 ... IFFT unit, 203 ... threshold determination unit, 204 ... FFT unit, 205 ... coefficient extraction unit, 211, 212 ... IFFT unit, 213, 214 ... threshold determination unit, 215, 216 ... FFT unit, 217 ... division unit, 218 ... IFFT unit, 219 ... convolution coefficient extraction unit, 400 ... transmitter, 410 ... propagation path, 420 ... receiver, 421 ... equalizer, 422 ... demodulator Decryption unit

Claims (10)

A filtering method for performing a filtering process on a received signal in a wireless communication system,
A first serial / parallel conversion / discrete Fourier transform step for performing serial / parallel conversion and discrete Fourier transform on the received complex signal;
A second serial / parallel conversion / discrete Fourier transform step for performing serial / parallel conversion and discrete Fourier transform on a predetermined reference signal;
A power spectrum estimation step for estimating a power spectrum based on an output of the first serial / parallel conversion / discrete Fourier transform step;
A transfer function estimation step for estimating a transfer function based on the output of the first series / parallel transform / discrete Fourier transform step and the output of the second series / parallel transform / discrete Fourier transform step;
A weight generation step of generating an equalization weight based on the output of the power spectrum estimation step and the output of the transfer function estimation step;
A filtering step for performing a filtering process on the received complex signal based on an output of the weight generation step,
The filtering method, wherein the weight generation step suppresses an estimation error of the equalization weight.   The weight generating step generates an equalized weight vector in the frequency domain, and the filtering step multiplies the equalized weight vector in the frequency domain by the output of the first serial / parallel transform / discrete Fourier transform step. 2. The filtering method according to claim 1, further comprising: a multiplying step; and a discrete Fourier inverse transform step for performing discrete Fourier inverse transform on an output of the multiplication step.   In the first and second serial / parallel conversion / discrete Fourier transform steps, when the input signal sequence is blocked, the blocks adjacent to each other in time are provided with overlapping portions to perform discrete Fourier transform. 3. The filtering method according to claim 2, wherein the discrete Fourier inverse transform step includes a step of deleting an overlapping portion from the output subjected to the discrete Fourier inverse transform after filtering in the frequency domain.   The filtering method according to claim 1, wherein the weight generation step generates a time domain equalization weight, and the filtering step includes a convolution step of convolving the received complex signal with a time domain equalization weight.   The weight generation step temporarily converts a vector to be determined into a time domain component, performs a predetermined threshold determination on each element of the time domain component obtained by this conversion, and sets a value of an element below the threshold to 0 5. The filtering method according to claim 1, wherein error suppression is performed. A filtering device that performs a filtering process on a received signal in a wireless communication system,
First serial / parallel conversion / discrete Fourier transform means for performing serial / parallel conversion and discrete Fourier transform on a received complex signal;
Second serial / parallel conversion / discrete Fourier transform means for performing serial / parallel conversion and discrete Fourier transform on a predetermined reference signal;
Power spectrum estimation means for estimating a power spectrum based on the output of the first serial / parallel conversion / discrete Fourier transform means;
Transfer function estimating means for estimating a transfer function based on the output of the first serial / parallel transform / discrete Fourier transform means and the output of the second serial / parallel transform / discrete Fourier transform means;
Weight generation means for generating equalization weights based on the output of the power spectrum estimation means and the output of the transfer function estimation means;
Filtering means for performing filtering processing on the received complex signal based on the output of the weight generation means,
The filtering apparatus, wherein the weight generation means suppresses an estimation error of the equalization weight.   The weight generating means generates an equalized weight vector in the frequency domain, and the filtering means multiplies the equalized weight vector in the frequency domain by the output of the first serial / parallel transform / discrete Fourier transform means. 7. The filtering apparatus according to claim 6, further comprising: a multiplying unit; and a discrete Fourier inverse transform unit that performs an inverse discrete Fourier transform on an output of the multiplication unit.   The first and second serial / parallel transform / discrete Fourier transform means block the input signal sequence by performing a discrete Fourier transform by making the blocks adjacent to each other in a temporally adjacent block into blocks. 8. The filtering apparatus according to claim 7, wherein said discrete Fourier inverse transform means includes means for deleting an overlapping portion from the output subjected to the discrete Fourier inverse transform after filtering in a frequency domain.   7. The filtering apparatus according to claim 6, wherein said weight generation means generates time domain equalization weights, and said filtering means has convolution means for convolving said received complex signal with time domain equalization weights.   The weight generation unit temporarily converts a vector to be determined into a time domain component, performs a predetermined threshold determination on each element of the time domain component obtained by the conversion, and sets the value of an element below the threshold to 0. 10. The filtering apparatus according to claim 6, wherein error suppression is performed.

Images