WO2008001439A1

WO2008001439A1 - Mimo reception device and mimo reception method

Info

Publication number: WO2008001439A1
Application number: PCT/JP2006/312919
Authority: WO
Inventors: Kazunori Inogai
Original assignee: Panasonic Corporation
Priority date: 2006-06-28
Filing date: 2006-06-28
Publication date: 2008-01-03

Abstract

It is possible to provide a MIMO reception device and a MIMO reception method capable of perform highly accurate signal separation with a small number of known symbols and a small calculation amount. The MIMO reception device includes a Fast-ICA processing unit (210) capable of performing signal separation not requiring a known symbol and a channel estimation unit (202) for acquiring a channel estimation value H by using a known symbol. An S, Λ matrix calculation unit (221) obtains a correlation matrix (S, Λ) by using the channel estimation value H. Thus, it is possible to instantaneously obtain the correlation matrix (S, Λ) usedin the Sphering process, which eliminates the need of the draw-in time in the Fast-ICA process, and to realize a MIMO reception device (200) capable of coping with transmission of a small number of data such as packet transmission.

Description

MIMO receiving apparatus and MIMO receiving method

Technical field

The present invention relates to a MIMO receiving apparatus and a MIMO receiving method used in a wireless communication system such as a wireless LAN (Local Area Network) or a cellular system.

Background art

In recent years, a multi-input-multiple-output (MIMO) transmission system has attracted attention as a wireless transmission system that does not increase the bandwidth even if the number of channels increases, and its research has been actively conducted. For example, Patent Document 1 discloses a MIMO transmission system.

FIG. 1 shows the concept of a conventional MIMO transmission system. As shown in FIG. 1, the MIMO transmitter 10 transmits M transmit antennas TxANT—0 to TxANT—M—1 power every one frame time, and k-channel TDM multiplexed signals. These signals are combined and received by N receiving antennas RxANT—0 to RxANT —N—l connected to the MIMO receiver 20. Here, when the service area is narrow or when divided into subcarriers when combined with OFDM transmission, frequency selective fading due to propagation path delay difference of multipath can be ignored. Each element is a complex number, and the following equation holds between the transmitted symbol and the received symbol. The following equation is expressed as X = Hs in matrix representation.

[Number 1]

N-1 Roh ^h

x _;, i = 0,1, · '·, Ν -1 is the received symbol at the i-th receiving antenna s-, j = 0,1, ···, Μ-1 is the j-th transmission Transmitter / receiver from receiver antenna> symbol h .. is the transmission coefficient y between jth and ith receiver antennas. [0004] MIMO receiver 20 performs signal separation to obtain M signals transmitted from M antennas on the transmission side from N received symbols. Signal separation is performed in two stages: channel estimation and separation.

[0005] First, as shown in FIG. 1! /, Because the transmitting antenna TxANT—0 sends out a known symbol! /, The other transmitting antennas TxANT—l to TxANT—M—1 do not send out signals. At this time, the distortion of the known symbols received by each receiving antenna RxANT—0 to RxANT—N—1 is detected by complex division, and the quotient becomes h 1, h 2,. .

00 10 N- l 0

Similarly, when only the transmitting antenna TxANT 1 transmits a known symbol, h and h 1 01 1

,..., H t, all elements can be measured to obtain an N X M transmission line matrix H.

1 N-l 1

This is channel estimation, which is performed by the known symbol distortion measurement units 21-0 to 21-M-1. Using this channel estimation value, the original transmission symbol vector s can be extracted and separated from the reception symbol vector X, and the processing is performed by the signal separation unit 22. The signal separation algorithm performed by the signal separation unit 22 is roughly divided into the following three.

[0006] (1) ZF: Ignores the presence of transmission line noise and obtains an inverse matrix ^H_1 of the channel estimation value, and performs inverse matrix operation to separate each signal.

(2) MMSE: Correct the estimate as the channel estimation error power is minimized, the reverse transfer sending passage matrix H ^{_ 1} determined, to separate the signals using the inverse matrix thereof.

(3) MLD: The channel estimation value H is used to generate a received symbol replica for all transmitted symbol patterns, and the most reliable transmission symbol is selected while evaluating each error component.

[0007] Note that the separated signal separated as described above includes residual interference components and receiver noise from other separated signals, and these magnitudes are related to the state of the MIMO transmission path and the obtained channel. Depends on the accuracy of the estimation and separation algorithm used.

Patent Document 1: Japanese Patent Laid-Open No. 2001-237751

Disclosure of the invention

Problems to be solved by the invention

However, the conventional MIMO reception method has the following problems.

(1) Signal separation by ZF and MMSE is a low amount of computation, but the transmission characteristics are degraded (i.e. Accompanied by degradation of separation accuracy).

(2) Signal separation by MLD can theoretically provide the best transmission characteristics, but the amount of computation is enormous. For example, when 16QAM transmission is performed using four antennas each for transmission and reception, 1.9 x 10 ^1G multiplications are required. A simple algorithm that reduces the number of multiplications to 1 X 10 ⁷ times that is the same as MMSE by introducing a bold simple key to this original MLD has also been proposed, but its transmission characteristics have been fully evaluated. That said, it is the current situation.

(3) The channel estimation values used in the above (1) and (2) are obtained by detecting distortions experienced by known symbols. However, the number of known symbols that can be transmitted is often limited in order to avoid a drastic decrease in transmission speed, so that the number of known symbols that can be used on the receiving side is limited and sufficient separation accuracy cannot be obtained.

An object of the present invention is to provide a MIMO receiving apparatus and a MIMO receiving method capable of performing high-accuracy signal separation with a relatively small number of known symbols and an operation amount. Means for solving the problem

[0010] The MIMO receiver of the present invention performs independent blind signal separation processing using an independent component analysis method on a plurality of received signals received by a plurality of antennas, thereby obtaining a plurality of independent separated signals. Component analysis means, channel estimation means for obtaining a channel estimation value between each transmitting / receiving antenna using the received known symbols, and a correlation matrix of received signals used in Sphering processing performed as preprocessing for performing the independent component analysis A correlation matrix calculating means for obtaining the value using the channel estimation value is employed. The invention's effect

[0011] According to the present invention, since signal separation processing is performed using an independent component analysis method, basically, a known symbol is not required and high-precision signal separation can be performed. Become. In addition, channel estimation using a known symbol is performed supplementarily, and the correlation matrix of the received signal used in the Sphering process is obtained using the channel estimation value from which the known symbol power is also obtained. Therefore, it is possible to realize a MIMO-ICA receiver that does not require the pull-in time in the independent component analysis processing and can cope with transmission with a small number of data such as packet transmission. Brief Description of Drawings

[0012] [FIG. 1] A diagram showing the concept of a conventional MIMO transmission system

[Figure 2] Figure showing a typical example of an independent component analysis algorithm

[Fig. 3] Qualitatively explaining the meaning of separation so that “non-Gaussianity” is maximized. FIG. 3A shows sub-Gauss, FIG. 3B shows Gauss, and FIG. Illustration showing

[Fig.4] Diagram showing an example of configuration using Fast-ICA for MIMO reception

[Fig.5] Fast— Flowchart for explaining 1-signal separation processing by ICA

[Figure 6] Fast—Timing chart showing the concept of MIMO reception processing using the ICA algorithm

FIG. 7 is a block diagram showing the configuration of the MIMO receiving apparatus according to the first embodiment.

[FIG. 8] Timing chart showing the concept of MIMO reception processing using the Fast—ICA algorithm of Embodiment 1.

FIG. 9 is a block diagram showing the configuration of the MIMO receiving apparatus according to the second embodiment.

FIG. 10 is a timing chart showing the concept of MIMO reception processing using the Fast-ICA algorithm of Embodiment 2.

FIG. 11 is a block diagram showing the configuration of the MIMO receiving apparatus according to the third embodiment.

FIG. 12 is a timing chart showing the concept of MIMO reception processing using the Fast-ICA algorithm of Embodiment 3.

BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[Principle of Embodiment]

First, the inventor of the present invention can use an independent component analysis algorithm (hereinafter also referred to as an ICA (Independent Component Analysis) algorithm) studied in the acoustic and medical fields without using a known symbol. We thought that signal separation could be performed. Figure 2 shows a typical independent component analysis algorithm. The inventor of the present invention considered that the Fast-ICA algorithm with the fastest convergence among the independent component analysis algorithms in FIG. 2 is optimal for application to a MIMO transmission system. In an embodiment, independent component analysis The case where the Fast-IC A algorithm is applied as an algorithm will be mainly described.

[0015] Furthermore, the inventors of the present invention have considered problems in applying the Fast-ICA algorithm to a MIMO receiver, and have devised the present invention to solve the problems.

[0016] (i) Fast—ICA anoligo rhythm

First, the Fast-ICA algorithm will be described. Here, in radio transmission, complex signals and complex coefficients are often described, but the ICA algorithm can be applied to complex signals. However, even when applied to complex signals, the basic concept is the same as that described in real numbers, so all are described as real numbers below for simplicity. Furthermore, the case where the number of transmitting antennas and receiving antennas is the same (M = N) will be described as an example.

[0017] Fast—ICA is an algorithm that extracts and separates signals one by one in adaptive signal processing so as to maximize the “non-Gaussianity” of the separated signal. The quantity representing the non-Gaussian principle, which is the guiding principle of the adaptive algorithm, uses the statistic of the absolute value of Kurtosis (kurtosis) or negentropy. Especially, the latter uses numerical stability and computational complexity. This is a practical algorithm. In MIMO transmission, each transmission signal has a non-Gaussian distribution and is considered to be independent of each other, and therefore, signal separation by Fast-ICA is possible. In the case of transmission signals with strong non-Gaussian characteristics, there is a possibility that superior transmission characteristics may be obtained in a low SINR environment.

[0018] The meaning of separation so that "non-Gaussianity" is maximized will be qualitatively explained with reference to FIG. In general, the shape of the probability density distribution of a signal is largely divided into three types: sub-Gauss (Fig. 3A), Gauss (Fig. 3B), and Super Gauss (Fig. 3C), depending on whether the force cusps that are dented from the Gaussian distribution are present. Separated. The two-dimensional distribution of the two independent transmission signals according to each distribution is as shown in the second stage. When rotational distortion is applied to the transmission line, the received signal has a two-dimensional distribution at the third stage. If these are simply received and separated, the distribution of the separated signal becomes the lowest level.

[0019] (1) The distribution after mixing is closer to the Gaussian distribution than the distribution before mixing (central limit theorem)

(2) The Gaussian distribution does not change before and after mixing! /. [0020] Fast—The ICA's guiding principle “maximize the“ non-Gaussianity ”of the separated signal” is appropriate so that a separated signal conforming to the non-Gaussian distribution can be obtained in the lowermost received signal in FIG. Give the original signal. However, as can be seen from Fig. 3, if the original signal follows a Gaussian distribution! /, The two-dimensional distribution becomes circular, so an appropriate rotation angle for returning to the original cannot be determined and cannot be separated. In an actual wireless transmission line, the force of non-Gaussianity that is subject to amplitude fluctuations in addition to phase rotation depends only on the shape of the distribution, so even if the level is indefinite, it does not become an essential problem. However, with Fast-ICA, preprocessing called Sphering is performed to convert the mixed signals into signals that are uncorrelated with each other and have the same power, thereby reducing the required amount of computation and achieving high-speed convergence characteristics. It has come to realize.

FIG. 4 shows a configuration when Fast-ICA that maximizes negentropy is applied to MIMO receiver 100. In FIG. 4, the configuration other than the Fast-ICA processing unit 110 in the MIMO receiver 100 is omitted in order to simplify the drawing.

[0022] Each symbol in the figure has the following meaning.

s (t): N-dimensional transmission signal vector (each element signal has a mean of 0 and follows a non-Gaussian distribution that is independent and even from each other)

A: N X N transfer coefficient matrix (a constant execution matrix because multipath and Rayleigh fluctuations are not considered)

x (t): N-dimensional received signal vector (expressed as x (t) = A 's (t))

Λ: Ν Χ ΝDecorrelated matrix

S: N X N power equalization matrix

z (t): N-dimensional Sphering output signal vector (expressed as z (t) = S A 'x (t), each element signal is uncorrelated with mean 0 and variance 1)

R: N X N rotation matrix (orthogonal matrix for obtaining separated signals y (t) independent of decorrelated signals z (t))

y (t): N-dimensional separated signal vector (represented by y (t) = R'z (t))

G (y): an observation function that takes one of the separated signals as input (necessary to determine the matrix R, but in practice it can be selected from several function candidates described later) The Fast-ICA processing unit 110 includes a Sphering processing unit 120, which is a preprocessing unit, and an ICA processing unit 130 that executes approximate Newton method processing. The Sphering processing unit 120 first obtains eigenvalues ~ λ and eigenvectors e ~ e by performing eigenvalue decomposition on the correlation matrix E [xx ^T ] of the received signal as shown in the following equation.

[Equation 2]

X ^λ 0- ^Λ ι ½-1 ' ^Λ 0 ^e l-

-~ '~ —, Eh 0, 8 1, is Ε | ΧΧ ^Τ ]) eigenvalue

Τ

· ·, E _m lt The corresponding Elxx ¹ specific vector

(2)

The Sphering processing unit 120 obtains the eigenvalues ~ λ and eigenvectors e thus obtained.

0 N-1

The following equation is executed using ~ e.

0 N-1

[Equation 3]

(3) In the following description, E [·] means the set average value (expected value) of '.

The ICA processing unit 130 detects the rotation angle that maximizes the separation signal's negentropy with respect to the output z of the Sphering processing unit 120, forms a rotation matrix R, and performs signal separation. Therefore, first, the R matrix calculation unit 131 determines a line r ^tau of the matrix R to separate the first signal using the Newton method. This is performed as follows:

0

[Equation 4]

( Four )

[0026] Large This process utilizes the fact and that each element signal output z of Sphering processor 120 are uncorrelated and the power to each other, the r ^tau is the norm 1 because row vector of rotation matrix

0

This is a reduction in the amount of computation, and is called the approximate Newton method. In equation (4), j represents the sampling processing time. In addition, under g '(y), g' in equation (4)

j and (y)

j

A function like the table is used.

[table 1]

[0027] It is known that the convergence of the approximate Newton method is a third-order convergence in which the estimation error decreases in the order of the third power due to the effect of a very fast Sphering process.

Summarizing the above, the 1-signal separation process is represented by the flowchart shown in FIG. That is, the estimation vector is initialized in step ST1, and the received signal vector is input to the Sphering processing unit 120 in step ST2. In step ST3, the DC component of each element signal is removed, and in step ST4, the S, Λ matrix calculation unit 121 calculates the decorrelation matrix Aj and the level equalization matrix Sj. Then, in step ST5, the whitening unit 122 performs Sphering processing. Execute. That is, the processing of steps ST3-ST4-ST5 is performed by the Sphering processing unit 120. Note that in FIG. 4, the force that omits the part that removes the DC component of each element signal is omitted for the sake of simplicity. Actually, the preceding stage of the Whitenig part 122 and the S, Λ matrix calculation part 121 Is provided with a DC component removal circuit. [0029] Next, signal separation processing is performed in step ST6, and estimation vector updating and normalization are performed in step ST7. Steps ST6 to ST7 correspond to the approximate Newton method and are executed by the ICA processing unit 130.

[0030] The other separated signals are distributed in the vertical direction with respect to the already separated signals. Therefore, the vector orthogonal to the obtained row vector r ^τ is gram = Schmidt orthonormalization

0

It can be obtained sequentially by the method and used as another row vector of matrix R. That is, M mutually independent row vectors r ^T , q ^T , ..., q ^Τ including row vector r ^T of matrix R obtained by one signal separation

Rotation matrix R can be obtained by preparing 0 0 1 M-l appropriately and executing the following equation.

[Equation 5]

rM-l ^P M-l / || ^P M -l || ' ^P M-1 — ^q Ml

^r M-2 / ^r M-2

[0031] In addition to the above, the Fast-ICA execution procedure includes, in addition to the above, executing the approximate Newton method of Eq. (4) in parallel to obtain M row vectors and calculating the force so that they are orthogonal to each other. There is also a way to adjust. In addition, there are a method in which the approximate Newton method of Eq. (4) is sequentially processed online, and a method in which batch processing is performed in which data is accumulated and batch processed.

[0032] Fig. 6 shows a conceptual diagram (timing chart) of MIMO reception processing using the Fast-ICA algorithm. MIMO receiving apparatus 100 performs channel estimation using known symbols transmitted by each transmission antenna power time division in a known symbol transmission period. Further, MIMO receiving apparatus 100 obtains a separated signal by performing signal separation processing using a Fast-ICA algorithm on data symbols transmitted simultaneously from the respective transmission antennas during the data transmission period.

[0033] Fast-ICA is the correlation matrix of the received signal shown in without using a channel estimate (2) E ^[χχ τ], by performing eigenvalue decomposition, to separate the signals. Since a large amount of received signal data is required to calculate the correlation matrix Ε [χχ ^τ ], the received data is stored and stored for each frame, and the correlation matrix is calculated and the eigenvalues are decomposed. And output Become so.

[0034] Here, as a supplement, a brief description of negentropy, kth moment, and kth cumulant will be given.

[0035] Negentropy is entropy corrected by the entropy of a Gaussian signal, and can be expressed by the following equation.

[Equation 6]] = ^H l ^y gauss) ^-H ], ^H ( ^y j) =-,] ^dy j where y _gauss is a Gaussian signal whose mean and variance are equal to y;

(6)

[0036] The entropy H (y) in Eq. (6) is the maximum for a signal with Gaussian distribution among signals of the same power. This is precisely called differential entropy, and relative comparison is possible, but its absolute value is meaningless. On the other hand, negentropy is always greater than 0 and its absolute value is also meaningful.

[0037] The statistic of the signal X, and the k-th moment (product moment) is expressed by the following equation. For example, the first moment is the average = ίΧρ (X) dX.

[Equation 7]

[0038] In addition, a k-th moment around an average represented by the following equation is often used.

[Equation 8]

[0039] Incidentally, the second moment around the mean is the variance σ. The kth moment is given by

It appears in the coefficient when the product generating function M (t) is expanded by Tiller like k.

X

[Equation 9] M _x (t)

(9)

[0040] On the other hand, the k-th order cumulant κ is the logarithm of the product moment generating function Mx (t) as

k

This is a statistic obtained as a coefficient when the cumulant generating function K (t)

X

The

[Equation 10]

(0 = log [M (t)] = 1 + t

(1 o) where the 4th order cumulant is Kurtosis.

[0041] Kurtosis, which is a fourth-order cumulant, is a statistic whose value changes according to the sharpness of the probability density distribution, as shown in FIG. In other words, it is 0 for the Gaussian distribution (Fig. 3B), a positive value for the super Gaussian (Fig. 3C) with a larger degree of sharpness, and a negative value for a sub-Gaussian with a smaller degree of sharpness (Fig. 3A). Therefore, the absolute value of Kurtosis includes the fourth power of the force signal, which is a measure of non-Gaussianity. Therefore, even if one sample contains an abnormal value due to noise or other factors, the value changes sensitively. Lack of robustness. However, Kurtosis is often used because it has convenient properties as shown in the following equation.

[Equation 11]

Kurt (x + y) = Kurt (x) + Kurt (y)

Kurt (jS x) = TKurt (x)

[0042] (ii) Fast—Improved points of ICA algorithm

When the Fast ICA algorithm is applied to MIMO transmission, signals can be separated without using known symbols. Therefore, since the data transmission period can be extended by an amount that does not require the known symbol transmission period, the substantial data transmission rate can be increased. For example, in Fig. 6, the Fast-ICA algorithm is used in the case where a known symbol transmission period is provided on the transmission side and channel estimation is performed using known symbols on the reception side. In principle, the known symbol transmission period and channel estimation process are not required.

[0043] In addition, the required amount of computation of ICA is difficult to evaluate because it includes numerical calculations of nonlinear functions, but it is considered to be relatively easy to implement, with sufficiently less than the original MLD.

However, since ICA operates based on statistics, it has the following problems. (A) Since the set average value included in the algorithm is replaced with the time average value and executed, a sufficient number of data until the value stabilizes, that is, a pull-in time is required, and the number of data such as packets , There is a risk that it will not be applicable to transmission. (B) Because it is an adaptive algorithm that uses a non-linear function V, it is close to the convergence point to some extent! If the initial value force estimation is not started, the convergence speed drops significantly, so there is a possibility that it will not be able to follow high-speed transmission path fluctuations. is there.

[Iii] Principle of the present invention

The inventor of the present invention has reached the present invention by considering the above points to be improved when applying the independent component analysis algorithm (Fast-ICA algorithm) to MIMO transmission.

[0046] A feature of the present invention is that an independent component analysis algorithm that does not require a known symbol and that can perform signal separation is used, and an independent component analysis algorithm that performs channel estimation using a known symbol in an auxiliary manner. In other words, signal separation processing using channel estimation values is performed. As a result, highly accurate signal separation can be performed with a relatively small number of known symbols and a large amount of computation.

[0047] Specifically, Fast-ICA signal separation algorithm, which is one of independent component analysis algorithms, and channel estimation are used together as follows.

[0048] 1. Fast-ICA pull-in time is reduced by performing pre-processing using channel estimation values.

[0049] 2. Use channel estimates to speed up convergence by finding an appropriate initial Fast—ICA estimate.

[0050] 3. Fast—By using the channel matrix at each time point, which also calculated the ICA estimated power, the channel estimation accuracy is improved without increasing the number of known symbols (or the frame period is lengthened). Improve feed speed).

[0051] By performing any one of steps 1 to 3, high-precision signal separation can be performed with a relatively small number of known symbols and a lower computational complexity than MLD.

[0052] The detailed configuration of 1. above will be described in Embodiment 1, the detailed configuration of 2. above will be described in Embodiment 2, and the detailed configuration of 3. above will be described in Embodiment 3.

[0053] (Embodiment 1)

FIG. 7 shows the configuration of the MIMO receiver of this embodiment. In FIG. 7, the configuration other than the Fast-ICA processing unit 210 in the MIMO receiver 200 is omitted in order to simplify the drawing. In FIG. 7, parts corresponding to those in FIG. 4 are assigned the same reference numerals as in FIG. 4, and descriptions thereof are omitted.

FIG. 7 differs from FIG. 4 in that switching switch 201 and channel estimation section 202 are provided, and that the channel estimation value obtained by channel estimation section 202 in S, Λ matrix calculation section 221 is provided. H is input.

MIMO receiving apparatus 200 selectively inputs N received signals x (t) received by N antennas to Fast-ICA processing section 210 or channel estimation section 202 via switching switch 201. To do. Specifically, the switching switch 201 outputs the received signal x (t) to the channel estimation unit 202 during the period in which the known symbol is received, and the received signal x ( t) is output to the Whitening unit 122 of the Fast-ICA processing unit 210.

[0056] Channel estimation section 202 obtains channel estimation value H using a known symbol. In practice, the channel estimation value H is also obtained for the relational power in Eq. (1). Channel estimation section 202 transmits the obtained channel estimation value H to S, Λ matrix calculation section 221 of Sphering processing section 220.

[0057] The S, Λ matrix calculator 221 calculates a level uniformization matrix (eigenvalue matrix) S and a decorrelation matrix (inherent vector) Λ. At this time, the S, Λ matrix calculation unit 221 instantly obtains the matrices S and Λ using the channel estimation value H that does not use the received signal X, and eliminates the pull-in time. This will be described below.

[0058] First, the S, Λ matrix calculation unit 221 performs an eigenvalue decomposition operation as shown in Equation (2). Here, the i and j elements of the correlation matrix on the right-hand side of equation (2) are set average values. Is generally calculated as a time average as follows: However,

In the following formula, n represents time.

[Equation 12]

L-1 L-l N-1 N-1

^ 1 ∑ x. (Nk) "x. (N-kj == — ∑ <] ∑ h. -S (nk) t {∑ h. -S (nk) ^

^L k = o ^l ''', -k = o [ _P = o' ppj [ _q = oq

(1 2)

[0059] Here, since the transmission signals s and s are independent of each other, the term of p ≠ q on the right side of equation (12) becomes 0 when calculated with the original aggregate average, but when calculated with the time average, the calculated data If the number L is small, it will not approach 0. In other words, when calculating the correlation matrix with high accuracy by time averaging, it is necessary to increase L, and signal separation processing by Fast-ICA can only be started after the L data time pull-in time. This means that it is difficult to apply when receiving small chunks of data such as packet transmission.

[0060] On the other hand, the channel estimation value H is a value measured in a good transmission environment without interference obtained by a known symbol transmitted while switching the transmitting antenna, and the time when the above-mentioned L is small. It is considered more reliable than the correlation value.

[0061] The S, Λ matrix calculation unit 221 of the present embodiment uses the channel estimation value H to calculate the correlation matrix based on the original set average instead of calculating the correlation matrix by time average using a large amount of data. By calculating as in the equation, the pull-in time is almost unnecessary.

[Equation 13]

E XX T = ElHss ^T H ^A = Η · Ε bs TTT

H HH (1 3) FIG. 8 shows a conceptual diagram (timing chart) of the soot reception process using the Fast-ICA algorithm of this embodiment. As is clear from the comparison with Fig. 6, when the channel estimation value H is obtained from the known symbols received at the beginning of the received frame, the correlation matrix is obtained instantaneously according to Eq. (13). You can proceed to. In the case of packet transmission, frames are not received continuously for a long time as shown in the figure. As a result, the frame signal separation operation can be completed within a problem-free time.

[0063] Note that the Fast-ICA process! As described above, the approximate Newton method has a very fast convergence, so there is no problem with online operation, but in the case of packet transmission, data is accumulated. If batch processing is performed, good transmission characteristics can be obtained without depending on the convergence process.

[0064] As described above, according to the present embodiment, a plurality of received signals received by a plurality of antennas are subjected to blind signal separation processing using an independent component analysis method, whereby a plurality of separated signals that are independent of each other are obtained. ICA processing unit 130 to be obtained, channel estimation unit 202 for obtaining a channel estimation value H between each transmitting and receiving antennas using the received known symbols, and Sphering processing performed as a preprocessing for performing independent component analysis. S, Λ matrix calculation unit 221 that obtains the correlation matrix of the received signal using the channel estimation value eliminates the pull-in time in the independent component analysis process and reduces the number of data as in packet transmission It is possible to realize the MIMO receiver 200 that can also support transmission.

[Embodiment 2]

FIG. 9 in which the same reference numerals are assigned to the parts corresponding to FIG. 7 shows the configuration of MIMO receiving apparatus 300 according to the present embodiment. MIMO receiving apparatus 300 has initial estimated value calculating section 301 for calculating initial estimated value R of R matrix calculating section 331 in addition to the configuration of FIG. Initial estimate calculation

0

The unit 301 inputs the channel estimation value H and the matrix S, Λ obtained by the S, Λ matrix calculation unit 221 using the channel estimation value H, and uses these to obtain an initial estimation value R of the approximate Newton method R The initial estimated value R is sent to the R matrix calculation unit 331.

0 0

[0066] Fast— The ICA algorithm is a combination of Sphering processing and approximate Newton's method that achieves third-order convergence and extremely fast convergence speed. When the convergence point is reached after about 10 iterations It is said. However, if it does not start with an appropriate initial estimation vector, it may stay at a stop point other than the convergence point, and the convergence may be significantly delayed or may diverge in some cases.

[0067] In the example of Fig. 5, a random number is generated and used as an initial value (step ST1). However, in this embodiment, an initial estimated value R that is close to the channel estimated value H force convergence point is calculated and used.

0

By doing so, it is always possible to obtain Fast—the original convergence speed of ICA. This This will be described below.

[0068] First, as shown in FIG. 4, y ^ R'SA and x ^ R'SA'H's are satisfied in an example in which Fast-ICA is applied to MIMO transmission. Here, is used because the receiver noise is ignored and the channel estimation value is used for H. If the matrix R is chosen as follows so that y = s, that is, R'SA'H = I (unit matrix), the initial estimated value R close to the convergence point can be obtained.

0

wear.

[Equation 14]

—1 —1 — 1

_{^{^{R 0 = H Χ Λ ^ 1}}} (14)

[0069] In spite of this, the above equation includes an inverse matrix calculation with a large required calculation amount. The inverse matrix of Λ, which is a rotation matrix, is transposed, and the inverse matrix of S, which is a diagonal matrix, is directly obtained by replacing the diagonal element with its inverse to obtain the inverse matrix of the force channel estimate Η directly. And the amount of computation increases.

[0070] Therefore, in the first embodiment, using the fact that the correlation matrix is also calculated as the channel estimation value repulsive force, substituting equation (14) into equation (2) yields equation (15), and (16) An expression can be derived.

[Equation 15]

(1 5)

[0071] [Equation 16]

H: ¹ = Η ^Τ (Λ ^Τ ΣΛ 厂

(1 6)

[0072] Therefore, by substituting equation (16) into equation (15), we can eliminate 、 ^_1 and The channel estimate by the quantity H force can also calculate the initial estimate! ^.

[Equation 17]

^{^¾} = Η Τ Λ Τ Σ one ^{^{^{^{1 AA _1 S _1 = Η Τ}}}} Λ Τ (Σ - ½ _1) = H T A T S o,

R _Q = H ^T A ^T S

(1 7)

[0073] Note that the initial estimation vector actually required to separate one signal is

0 Since only one line is selected appropriately, the amount of computation can be further reduced from equation (17).

FIG. 10 shows a conceptual diagram (timing chart) of MIMO reception processing using the Fast-ICA algorithm of the present embodiment. Matrix eigenvalue decomposition of the channel estimation value H force further correlation matrix HH ^T by known symbol received by the head of the received frame S, lambda is that Motomema. Therefore, the initial value estimated value calculation unit 301 calculates the initial estimated value R according to Equation (17), and the ICA process.

0

By starting the adaptive process using the approximate Newton method from the initial estimate R

0

Always, Fast-ICA's original third-order convergence characteristics can be obtained.

As described above, according to the present embodiment, in addition to the configuration of the first embodiment, the initial estimated value calculation unit 301 for estimating the initial estimated value R of the independent component analysis using the channel estimated value H is provided. Setting

0

As a result, a stable and high convergence speed can be realized, so that it is possible to realize a MIMO receiver 300 that can follow high-speed transmission path fluctuations.

[Embodiment 3]

FIG. 11 in which parts corresponding to those in FIG. 9 are assigned the same reference numerals shows the configuration of the MIMO receiving apparatus of the present embodiment. In this embodiment, in addition to the configuration of FIG. 9, R matrix calculation unit 331 Approximate Newton method estimated value obtained from rotation (rotation matrix) R force also reversely calculates transmission line matrix H ′ Transmission line matrix reverse calculation part 401 and an averaging unit 402 that averages the transmission path matrix H ′ and the channel estimation value H.

[0077] As described above, in MIMO-ICA reception, the channel estimation value H is the ICA algorithm. This is a very important amount to bring out the performance. However, since the number of known symbols is limited in order to suppress a decrease in transmission speed, the channel estimation value H is reduced in estimation accuracy due to receiver noise or is updated at the frame period, so high-speed channel fluctuation You may not be able to follow. On the other hand, the Fast-ICA estimated value R is updated in the symbol period by online operation. Therefore, if the channel matrix H 'at that time is calculated backward from the Fast-ICA estimated value R at the frame end time and this is reflected in the channel estimated value H, it will follow even under high-speed channel fluctuations. As a result, better transmission characteristics can be obtained. This will be described below.

The rotation matrix R force, which is the Fast-ICA estimate, can also be calculated as follows by substituting R for matrix R in Eq. (17) and H for matrix H in Eq. (17). That is,

0

The transmission path matrix inverse calculation unit 401 calculates the transmission path matrix H ′ by performing the following calculation.

[Equation 18]

R Η ' ^Τ Λ S

, Τ 1

Η R ^T SI = RS RS ^{_ 1} A

H 'A ^T S-]

(18) The averaging unit 402 reflects the channel matrix H ′ obtained in this way on the channel estimation value H as a weighted average as shown in the following equation, thereby averaging the channel estimation value. Find H.

[Equation 19]

H-(l-α) Η + α Η ', 0≤ α≤ 1 (1 9) Figure 12 shows a schematic diagram of a MIMO reception process using the Fast-ICA algorithm of this embodiment (timing chart). ). Known symbols received at the beginning of each received frame Force The force that the channel estimation value H can be obtained. The Fast—IC A estimation value R force obtained in the previous frame is also obtained by the equation (18), and the channel matrix H 'at that time is obtained and averaged by the equation (19). Is the channel estimation value H again. Subsequent matrix by eigenvalue decomposition of the correlation matrix HH ^T S, seeking lambda. Therefore, the initial estimated value R is calculated by equation (16) and approximated from here.

0

Performs adaptive processing using the Newton method. In this way, channel estimation and Fast-ICA operate in a complementary manner, improving the estimation accuracy without increasing the number of known symbols for channel estimation and reducing the transmission speed, and providing good transmission characteristics. It will be obtained.

As described above, according to the present embodiment, in addition to the configuration of the second embodiment, the transmission path matrix H ′ is calculated backward from the estimated value R of the approximate Newton method obtained by the R matrix calculation unit 331. In addition to the effects of the first embodiment, the channel estimation unit 401 and the averaging unit 402 which averages the channel matrix H ′ and the channel estimation value H are provided. Therefore, it is possible to improve the channel estimation accuracy without increasing the number of known symbols to reduce the transmission rate, and to perform independent component analysis processing with better error rate characteristics. In addition, since the followability to transmission path fluctuation is improved, it is possible to increase the transmission rate by extending the frame period.

Industrial applicability

The present invention is particularly suitable for application to an Ml MO receiver adapted to perform signal separation processing by independent component analysis (ICA).

Claims

The scope of the claims

[1] Independent component analysis means for obtaining a plurality of independent separated signals by subjecting a plurality of received signals received by a plurality of antennas to blind signal separation processing using an independent component analysis method;

Channel estimation means for obtaining a channel estimation value between each transmitting and receiving antenna using the received known symbol;

A MIMO receiving apparatus comprising: a correlation matrix calculating unit that obtains a correlation matrix of a received signal used for Sphering processing performed as preprocessing for performing the independent component analysis using the channel estimation value.

[2] An initial estimated value calculating means for obtaining an initial value of the independent component analysis using the channel estimated value is further provided.

The MIMO receiver according to claim 1.

[3] Transmission path matrix inverse calculation means for calculating back a transmission path estimation matrix from the rotation matrix obtained by the independent component analysis means to obtain the separated signals independent of each other;

An averaging process is performed using the transmission path estimation matrix obtained by the transmission path matrix reverse calculation means and the channel estimation value obtained by the channel estimation means, and the value of the reverse calculation of the transmission path estimation matrix is obtained. Averaging means for calculating a reflected channel estimate, and

The correlation matrix calculation means uses the channel estimation value reflecting the back-calculated transmission path estimation matrix value obtained by the averaging means, and uses the correlation matrix of the received signal used for the Sphering process. Ask for

The MIMO receiver according to claim 1.

[4] A signal separation step of performing signal separation processing on a plurality of received signals received by a plurality of antennas using a Fast-ICA algorithm,

A channel estimation step for obtaining a channel estimation value between the transmission and reception antennas using known symbols received by the plurality of antennas;

Including

Received signal used for Sphering processing in the Fast-ICA algorithm! MIMO reception method for obtaining a correlation matrix of a signal using the channel estimation value.