JP2013038592A - Signal separation device and signal separation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To separate plural interfering signals mixed in a received signal although the period of an incident signal may not be known in advance.SOLUTION: A signal separation device includes a peak sequence extraction unit 3 which detects a peak sum of squares evaluation function value J(τ) among the sum of squares evaluation function values J(τ) of each time difference τ to determine a time difference τ corresponding to that sum of squares evaluation function value J(τ), and a period estimation unit 4 which classifies time differences τ appearing in the same period among a plurality of time differences τ into one and the same group and estimates the period τ(p) of a time difference τ belonging to each group. Assuming that the period of an incident signal s matches the period τ(p) estimated by the period estimation unit 4, the incident signal s is separated from the whitening received signal z thereof.

Description

  The present invention relates to a signal separation device and a signal separation method for separating a plurality of signals (a plurality of incident signals having different signal sources) interfering with a reception signal of an antenna.

In a signal separation device that separates a plurality of signals (a plurality of incident signals having different signal sources) interfering with a reception signal of an antenna, for example, a signal separation algorithm called independent component analysis (ICA) is used. Sometimes.
This ICA performs blind separation and extraction of a plurality of signals from a reception signal in which a plurality of signals interfere with each other on the assumption that signals transmitted from different signal sources are independent from each other. Application to various fields including sensors is possible (see Non-Patent Document 1).

ただし、ｓ（ｋ）は入射信号、ｘ（ｋ）は受信信号、Ａは入射信号の混信を表す混合行列であり、混合行列Ａの各要素は混信を意味する係数である。
また、ｋ＝１，２，・・・，Ｋはサンプル時間のインデックス、Ｋは総サンプル数である。 FIG. 12 is an explanatory diagram showing signal separation by ICA.
The outline of ICA will be described below.
ICA is an algorithm that separates each incident signal from a reception signal in which incident signals that are independent from each other interfere in the interference reception model shown in the following equation (1).

Here, s (k) is an incident signal, x (k) is a received signal, A is a mixing matrix representing interference of the incident signal, and each element of the mixing matrix A is a coefficient meaning interference.
In addition, k = 1, 2,..., K is an index of sample time, and K is the total number of samples.

ｓ_ｊ（ｋ）（ｊ＝１，２，・・・，Ｊ）は第ｊ番目の入射信号であり、ｓ（ｋ）は第ｊ番目の入射信号ｓ_ｊ（ｋ）をベクトル状に格納している入射信号ベクトルである。Ｊは入射信号数である。

s _j (k) (j = 1, 2,..., J) is the jth incident signal, and s (k) stores the jth incident signal s _j (k) in a vector form. Incident signal vector. J is the number of incident signals.

ｘ_ｉ（ｋ）（ｉ＝１，２，・・・，Ｉ）は第ｉ番目のアンテナの受信信号であり、ｘ（ｋ）は第ｉ番目のアンテナの受信信号ｘ_ｉ（ｋ）をベクトル状に格納している受信信号ベクトル）である。Ｉは受信信号数である。

x _i (k) (i = 1, 2,..., I) is a received signal of the i-th antenna, and x (k) is a vector of the received signal x _i (k) of the i-th antenna. Received signal vector). I is the number of received signals.

Hereinafter, in order to simplify the notation, the index k of the sample time is omitted when it is clear.
Generally speaking, the ICA estimates the mixing matrix A so that a plurality of separated signals that are signals after separation processing are independent of each other, and the estimated mixing matrix A hat ( In the text of the specification, the symbol “^” cannot be added on the character because of the electronic application, so the received signal x is represented in the (pseudo) inverse matrix of “A”. To obtain separated signal vectors y independent of each other.

ただし、Ａハットの右上の「＋」の記号はＡハットの擬似逆行列を意味し、Ｍは白色化用の行列である。 However, here, when obtaining the separated signal vectors y independent from each other, as a preprocessing for the received signal x, the received signal x is whitened, and the whitened received signal z (the following equation (5) However, the separation matrix W is estimated so as to be independent from each other, and signal separation is performed by the following equation (4).

However, the symbol “+” in the upper right of the A hat means a pseudo inverse matrix of the A hat, and M is a matrix for whitening.

Hereinafter, a method for estimating the separation matrix W will be described.
In ICA, the separation matrix W is estimated so that some independent index is optimum (maximum).
In ICA, various independent indices are known. For example, there are those using time difference cross-correlation values as independent indices. Such an ICA is called TDSEP (Temporal Decoration Source Separation) (see Non-Patent Document 2).

式（７）の（１，２）要素と（２，１）要素に時間差相互相関が現れているため、これらを最小化するような分離行列Ｗを計算する。これは式（７）の対角化を意味している。
なお、Ｅは期待値（または平均値）を意味する数学記号であり、Ｈは行列の転置を意味し、＊は共役複素数を意味する。 TDSEP performs signal separation (estimates the separation matrix W) by assuming periodicity in the incident signal and minimizing the time difference cross-correlation value of the whitened reception signal z.
For example, when the number of incident signals is 2, the time difference cross-correlation matrix Rz (τ) of the whitened reception signal z at the time difference τ is expressed as the following equation (7).

Since the time difference cross-correlation appears in the (1,2) element and the (2,1) element of the equation (7), a separation matrix W that minimizes these is calculated. This means diagonalization of equation (7).
Note that E is a mathematical symbol that means an expected value (or an average value), H means transposition of a matrix, and * means a conjugate complex number.

In general, for each time difference τ within a predetermined range, a time difference cross-correlation matrix Rz (τ) is calculated and diagonalized at the same time. For example, a separation matrix W is obtained by a simultaneous diagonalization method such as an extended Jacobian method. Minimize.
Actually, since the time difference cross-correlation matrix Rz (τ) is not a symmetric matrix, as shown in the following equation (8), the time difference cross-correlation matrix Rz (τ) is converted to a symmetric matrix, and the converted matrix Rz (Τ) tilde (in the text of the description, the symbol “˜” cannot be added on the letter because of electronic application, so it is expressed as Rz (τ) tilde). Keratinize.

In TDSEP, as described above, signal separation is performed by minimizing the time difference cross-correlation value of the whitened reception signal z. At this time, an appropriate time difference is selected from many time differences τ within a predetermined range. It is necessary to select τ.
Generally, how to select an appropriate time difference τ is roughly divided into the following two types.
(1) Method of selecting a plurality of time differences τ at random Although this method does not select the optimal time difference τ, a plurality of time differences τ are used, so that stable separation performance can be obtained.
(2) Method of selecting a time difference τ other than 0 that maximizes the time difference autocorrelation value of the incident signal This method needs to know the period of the incident signal in advance, so it is not blind but captures the temporal correlation Best in meaning.

  In the conventional signal separating apparatus, as shown in FIG. 13, a plurality of incident signals (separated signals) that interfere with the received signal are set by manually setting the time difference τ selected by the above two methods to TDSEP. Signal separation is performed.

Bingham E. and Hyvarinen A., “A fast fixed-point algorithm for independent component analysis of complex valued signals,” International Journal of Neural Systems, Vol. 10, No.1, pp.1-8, Feb. 2000. Andreas Ziehe and Klause-Reber Muller, “TDSEP-an efficient algorithm for blind separation using time structure,” ICANN98, pp.675-680, 1998.

  Since the conventional signal separation device is configured as described above, the optimal time difference τ is not necessarily selected in the method of selecting a plurality of time differences τ at random. In the method of selecting a time difference τ other than 0 that maximizes the time difference autocorrelation value of the incident signal, it is necessary to know the period of the incident signal in advance, and if the period of the incident signal is not known in advance, the time difference τ is calculated. There were issues such as being unable to select.

  The present invention has been made to solve the above-described problems. Even if the period of the incident signal is not known in advance, an appropriate time difference τ is automatically selected to interfere with the received signal. It is an object of the present invention to provide a signal separation device and a signal separation method capable of separating a plurality of signals (a plurality of incident signals having different signal sources).

  The signal separation device according to the present invention provides a whitening means for whitening a reception signal in which a plurality of incident signals from different signal sources interferes, and a time difference between peaks in the reception signal whitened by the whitening means is a variable. Generating a time difference cross-correlation matrix of the received signal, and sequentially substituting an arbitrary time difference within a predetermined range into the time difference cross-correlation matrix to calculate a cross-correlation value of each time difference; and Among the cross-correlation values of each time difference calculated by the cross-correlation value calculating means, a cross-correlation value that is larger than a cross-correlation value of adjacent time differences and larger than a predetermined threshold is detected, and the detected cross-correlation value is detected. Classify the time difference that appears in the same period among the plurality of time differences identified by the peak correlation value detection means and the peak correlation value detection means that identify the time difference corresponding to the correlation value, Period estimation means for estimating the period of the time difference belonging to the group, and the signal separation means is whitened by the whitening means on the assumption that the period of the incident signal coincides with the period estimated by the period estimation means. Signal separation of a plurality of incident signals interfering with the received signal.

  According to the present invention, the whitening means for whitening a reception signal in which a plurality of incident signals with different signal sources interferes, and the time difference between peaks in the reception signal whitened by the whitening means is a variable. A cross-correlation value calculating means for generating a time-difference cross-correlation matrix of a received signal, sequentially substituting an arbitrary time difference within a predetermined range into the time-difference cross-correlation matrix, and calculating a cross-correlation value of each time difference; and a cross-correlation value Among the cross-correlation values of each time difference calculated by the calculation means, a cross-correlation value that is larger than a cross-correlation value of adjacent time differences and larger than a predetermined threshold is detected, and the detected cross-correlation value Among the plurality of time differences identified by the peak correlation value detection means for identifying the corresponding time difference and the peak correlation value detection means, classify the time differences that appear in the same period into the same group, A period estimating means for estimating the period of the time difference being received, and the signal separating means assumes that the period of the incident signal coincides with the period estimated by the period estimating means. Since it is configured to separate multiple incident signals interfering with the signal, even if the period of the incident signal is not known in advance, an appropriate time difference is automatically selected to interfere with the received signal. There is an effect that a plurality of incident signals can be separated.

It is a block diagram which shows the signal separation apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content of the signal separation apparatus by Embodiment 1 of this invention. It is a block diagram which shows the frequency estimation part 4 of the signal separation apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the example of calculation of the square sum evaluation function value J ((tau)) by the time difference cross correlation calculation part 2. FIG. It is explanatory drawing which shows the time difference (tau) corresponding to the square sum evaluation function value J (tau) detected by the peak row extraction part. It is explanatory drawing which shows the outline | summary of the period estimation method by the period estimation part 4. FIG. It is explanatory drawing which shows the time difference autocorrelation of the incident signal of 3 waves. It is explanatory drawing which shows the time-difference cross correlation of a whitening reception signal. It is explanatory drawing which shows time difference (tau) ((tau) = 30,74,77) set as a time difference parameter in TDSEP. It is explanatory drawing which shows the simulation result of the separation performance of the incident signal in the signal separation apparatus by Embodiment 1 of this invention and the conventional signal separation apparatus. It is a block diagram which shows the frequency estimation part 4 of the signal separation apparatus by Embodiment 2 of this invention. It is explanatory drawing which shows the signal separation by ICA. It is a block diagram which shows the conventional signal separation apparatus.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a signal separation device according to Embodiment 1 of the present invention.
In FIG. 1, a whitening unit 1 whitens a reception signal x (a reception signal x in which a plurality of incident signals s having different signal sources are mixed) received by a plurality of antennas, and is a reception signal after whitening. A process of outputting the whitened reception signal z to the time difference cross-correlation calculation unit 2 and the signal separation unit 5 is performed. The whitening unit 1 constitutes whitening means.

The time difference cross-correlation calculation unit 2 generates a time difference cross-correlation matrix Rz (τ) of the whitened reception signal z using the time difference τ between the peaks in the whitened reception signal z output from the whitening unit 1 as a variable. Arbitrary time difference τ within the range is sequentially substituted into the time difference cross-correlation matrix Rz (τ), and the process of calculating the cross-correlation value of each time difference τ (the square sum evaluation function value J (τ) of the time difference cross-correlation) is performed. To do. The time difference cross-correlation calculation unit 2 constitutes a cross-correlation value calculation unit.
The peak string extraction unit 3 includes, among the square sum evaluation function values J (τ) of each time difference τ calculated by the time difference cross-correlation calculation unit 2, the square sum evaluation function values J of the adjacent time differences τ−1 and τ + 1. A sum of squares evaluation function value J (τ) larger than (τ-1), J (τ + 1) and larger than a predetermined threshold value threshold is detected, and a time difference τ corresponding to the square sum evaluation function value J (τ) is detected. Implement the process to identify. The peak string extraction unit 3 constitutes a peak correlation value detection unit.

The period estimation unit 4 classifies the time differences τ appearing in the same period among the plurality of time differences τ specified by the peak string extraction unit 3 into the same group, and the period τ (p) of the time differences τ belonging to each group. The process which estimates is carried out. p = 1, 2,..., P, where P is the number of classified groups. The cycle estimation unit 4 constitutes cycle estimation means.
The signal separating unit 5 assumes that the periods of the plurality of incident signals s interfering with the received signal x coincide with the period τ (p) estimated by the period estimating unit 4, and outputs the white color output from the whitening unit 1. Signal separation is performed on a plurality of incident signals s interfering with the synthesized reception signal z.
That is, the signal separation unit 5 performs TDSEP which is an independent component analysis using a time difference cross-correlation value as an independent index, and the time difference of the whitened reception signal z in the period τ (p) estimated by the period estimation unit 4. By minimizing the cross-correlation value, signal separation of a plurality of incident signals s interfering with the whitened reception signal z is performed.
The signal separation unit 5 constitutes signal separation means.

In the example of FIG. 1, each of the whitening unit 1, the time difference cross-correlation calculation unit 2, the peak sequence extraction unit 3, the period estimation unit 4, and the signal separation unit 5 which are components of the signal separation device is dedicated hardware (for example, , A semiconductor integrated circuit on which a CPU is mounted, or a one-chip microcomputer or the like is assumed, but the signal separation device may be configured with a computer.
When the signal separation device is configured by a computer, a program describing the processing contents of the whitening unit 1, the time difference cross-correlation calculation unit 2, the peak sequence extraction unit 3, the period estimation unit 4 and the signal separation unit 5 is stored in the computer. What is necessary is just to make it memorize | store in memory and to run the program stored in the said memory of CPU of the said computer.
FIG. 2 is a flowchart showing the processing contents of the signal separation apparatus according to Embodiment 1 of the present invention.

FIG. 3 is a block diagram showing the frequency estimation unit 4 of the signal separation device according to Embodiment 1 of the present invention.
In FIG. 3, the first period setting unit 11 performs a process of setting the first time difference τ among the plurality of time differences τ specified by the peak string extraction unit 3 to the first period τ (1). .
The cycle storage unit 12 is a recording medium that stores a set cycle τ (p) (p = 1, 2,..., P).

The next peak time difference selection unit 13 performs a process of selecting the nth time difference τ among the plurality of time differences τ specified by the peak string extraction unit 3.
The peak time difference / cycle division unit 14 divides the nth time difference τ selected by the next peak time difference selection unit 13 by the set cycle τ (p) stored by the cycle storage unit 12 (nth). Time difference τ / τ (1), nth time difference τ / τ (2),..., Nth time difference τ / τ (P)), and the remainder r (p) (p) = 1, 2,..., P) is output to the same column determination unit 15.

If the remainder r (p) output from the peak time difference / period division section 14 is zero or the remainder r (p) is sufficiently small (the remainder is smaller than the predetermined value ε), the same column determination unit 15 is nth. Are included in the set period τ (p) column, and the nth time difference τ is classified into a group of time differences that appear in the set period τ (p). On the other hand, when all the remainders r (p) output from the peak time difference / period division unit 14 are sufficiently large (the remainder is larger than the predetermined value ε), the nth time difference τ is equal to any period τ (p). It is determined that it does not belong to any column, and a process of setting the nth time difference τ to the (P + 1) th period τ (p + 1) is performed.
The process end determination unit 16 determines whether or not the determination process of the same column determination unit 15 has been completed for all the time differences τ specified by the peak string extraction unit 3, and if the determination process has not been completed, The process of instructing the next peak time difference selection unit 13 to select the time difference τ that has not yet been selected is performed.

Next, the operation will be described.
1 schematically selects a time difference τ having a large time difference cross-correlation value of the whitened reception signal z that is a non-diagonal term of the time difference cross-correlation matrix Rz (τ) of the time difference τ, The time difference τ is set to TDSEP.
However, in the case of an ideal periodic signal, the time difference cross-correlation matrix Rz (τ) has a peak at the period of each incident signal s, so that the time difference τ is selected one by one from each peak row.
That is, for each time difference τ within the predetermined range, the time difference cross-correlation value of the whitened reception signal z is calculated, and the time difference τ corresponding to the time difference cross-correlation value having a peak is selected from the calculation results. To do.
When a plurality of time differences τ are selected, the period τ (p) corresponding to each peak sequence is estimated from the sequence of these time differences τ.

ただし、Ｍは白色化用の行列である。 Hereinafter, the processing content of the signal separation device of FIG. 1 will be described in detail.
When a plurality of antennas receive a mixed signal (a signal in which a plurality of incident signals s of different signal sources are mixed) (see FIG. 12), the signal receiving unit (not shown) receives an A / D for the mixed signal. A reception process such as a conversion process is performed, and a received signal x, which is a digital mixed signal, is output to the whitening unit 1.
When receiving the digital reception signal x, the whitening unit 1 whitens the reception signal x as shown in the following equation (11), and converts the whitened reception signal z, which is a whitened reception signal, into a time difference. It outputs to the correlation calculation part 2 and the signal separation part 5 (step ST1). Since the process of whitening the received signal x is a known technique, detailed description thereof is omitted.

Here, M is a whitening matrix.

  In addition, since the amplitude of the incident signal s separated by the signal separation unit 5 is indefinite, the whitening unit 1 whitens the reception signal x in advance, and the whitened reception signal which is a reception signal after whitening When TDSEP, which is independent component analysis described later, is performed on z, learning in TDSEP is expected to be faster than when TDSEP is performed on reception signal x that has not been whitened.

式（１２）の（１，２）要素と（２，１）要素に時間差相互相関が現れているため、これらを最小化するような分離行列Ｗを計算する。これは式（１２）の対角化を意味している。
なお、Ｅは期待値（または平均値）を意味する数学記号であり、Ｈは行列の転置を意味し、＊は共役複素数を意味する。 When the time difference cross-correlation calculation unit 2 receives the whitened reception signal z from the whitening unit 1, the time difference cross-correlation matrix Rz ( (τ) is generated (step ST2).
For example, when the number of incident signals is 2, the time difference cross-correlation matrix Rz (τ) of the whitened reception signal z at the time difference τ is expressed as the following Expression (12).

Since a time difference cross-correlation appears in the (1,2) element and the (2,1) element of the equation (12), a separation matrix W that minimizes these is calculated. This means diagonalization of equation (12).
Note that E is a mathematical symbol that means an expected value (or an average value), H means transposition of a matrix, and * means a conjugate complex number.

ただし、ｏｆｆ（．）は行列の非対角項を意味する。
また、Ｒｚ_ｊ１ｊ２チルダは、行列Ｒｚチルダの第（ｊ１，ｊ２）要素であることを意味し、ｊ１≠ｊ２は非対角項を意味している。 When the time difference cross-correlation calculation unit 2 generates the time difference cross-correlation matrix Rz (τ) of the whitened reception signal z, as shown in the following equation (13), an arbitrary time difference τ within a predetermined range is converted to the time difference cross-correlation. By sequentially substituting in the matrix Rz (τ), the square sum evaluation function value J (τ) of the time difference cross-correlation is calculated as the cross-correlation value of each time difference τ (step ST3).

However, off (.) Means an off-diagonal term of the matrix.
Rz _j1j2 tilde means the (j1, j2) element of the matrix Rz tilde, and j1 ≠ j2 means a non-diagonal term.

Here, an arbitrary time difference τ within a predetermined range is substituted into the time difference cross-correlation matrix Rz (τ). For example, the predetermined range may be about half the length of the received signal x.
The time difference τ to be substituted into the time difference cross-correlation matrix Rz (τ) is a discrete value. For example, if the length of the received signal x is 2000, it is a discrete value between 0 and 1000.
FIG. 4 is an explanatory diagram showing a calculation example of the square sum evaluation function value J (τ) by the time difference cross-correlation calculation unit 2, and the square sum evaluation function value J (τ) of each time difference τ is calculated.
The square sum evaluation function value J (τ) of the time difference τ corresponding to the period of the incident signal s has a large peak, but the square sum evaluation function value J (τ) of the time difference τ that does not correspond to the period of the incident signal s. May have small peaks, and it is necessary to distinguish the small peak from the original large peak.

Therefore, when the time difference cross-correlation calculation unit 2 calculates the square sum evaluation function value J (τ) of each time difference τ, the peak string extraction unit 3 calculates the sum of squares of each time difference τ as shown in the following equation (14). Among the evaluation function values J (τ), the squares larger than the square sum evaluation function values J (τ−1) and J (τ + 1) of the adjacent time differences τ−1 and τ + 1 and larger than a predetermined threshold value threshold. The sum evaluation function value J (τ) is detected, and the time difference τ corresponding to the square sum evaluation function value J (τ) is specified (step ST4).

ただし、ｍｅａｎ（．）は平均値の計算を意味している。 Various methods are conceivable for setting the threshold value threshold. Here, as shown in the following equation (15), the square sum evaluation function value J (τ) of the adjacent time differences τ _i −1 and τ _i +1 is used. _i −1) and an average value of all square sum evaluation function values J (τ _i ) larger than J (τ _i +1).

However, mean (.) Means calculation of the average value.

FIG. 5 is an explanatory diagram showing the time difference τ corresponding to the square sum evaluation function value J (τ) detected by the peak string extraction unit 3.
In the example of FIG. 5, 14 square sum evaluation function values J (τ) are detected, and 14 time differences τ are specified.

  When the peak sequence extraction unit 3 detects a plurality of square sum evaluation function values J (τ) and specifies a time difference τ corresponding to the plurality of square sum evaluation function values J (τ), the period estimation unit 4 Among the time differences τ, the time differences τ appearing in the same period are classified into the same group, and the period τ (p) of the time difference τ belonging to each group is estimated (step ST5). p = 1, 2,..., P, where P is the number of classified groups.

Hereinafter, the processing content of the period estimation part 4 is demonstrated concretely.
FIG. 6 is an explanatory diagram showing an outline of the period estimation method by the period estimation unit 4.
In the period estimation unit 4, the time difference between the periodic peaks uses a property existing in a time difference that is an integral multiple of the period, and the time difference is divided in order from the time difference of the first peak. If it exists in the peak row and the remainder is large, it is considered that it belongs to the peak row of a different cycle, and a method of detecting each cycle is adopted.

The first cycle setting unit 11 of the cycle estimation unit 4 detects a plurality of square sum evaluation function values J (τ) by the peak string extraction unit 3 and time differences corresponding to the plurality of square sum evaluation function values J (τ). When τ is specified, the first time difference τ is set to the first period τ (1) among the plurality of time differences τ.
The first cycle τ (1) set by the first cycle setting unit 11 is stored by the cycle storage unit 12.

The next peak time difference selection unit 13 selects the nth time difference τ among the plurality of time differences τ specified by the peak string extraction unit 3, and outputs the time difference τ to the peak time difference / period division unit 14. n = 1, 2,..., N, where N is the number of time differences τ specified by the peak string extraction unit 3.
When the peak time difference / period division unit 14 receives the nth time difference τ from the next peak time difference selection unit 13, the nth time difference τ is set as a period candidate τc, and the period candidate τc is input by the period storage unit 12. The division is performed by the stored set period τ (p), and the remainder r (p) obtained by the division is output to the same column determination unit 15.
That is, the peak time difference / period division unit 14 calculates mod (τc, τ (p)), and outputs the remainder r (p) at the time of division to the same column determination unit 15. Mod (a, b) is a function that returns a remainder obtained by dividing a by b.

  At the initial stage, only the first period τ (1) is stored in the period storage unit 12, so only mod (τc, τ (1)) is calculated and the remainder r (1) is output. Then, P−1 periods τ (2), τ (3),..., Τ (P) are set, and P periods τ (1), τ (2),. If τ (P) is stored in the period storage unit 12, P mod (τc, τ (1)), mod (τc, τ (2)), ..., mod (τc, τ (P )) And P remainders r (1), r (2),..., R (P) are output.

When the same column determination unit 15 receives the remainder r (p) from the peak time difference / period division unit 14, it compares the remainder r (p) with a predetermined value ε. The predetermined value ε is a value close to zero smaller than the cycle τ (p), and a value such as 2 or 3 is set, for example.
If P remainders r (1), r (2),..., R (P) are received from the peak time difference / period division unit 14, P remainders r (1), r (2 ,..., R (P) and a predetermined value ε are respectively compared.

When the remainder r (p) is zero or when the remainder r (p) is smaller than the predetermined value ε (0 <r (p) <ε), the same column determination unit 15 determines that the nth time difference τ is The nth time difference τ is determined to belong to the column of the set cycle τ (p), and is classified into a group of time differences that appear in the set cycle τ (p).
For example, when P remainders r (1), r (2),..., R (P) are received from the peak time difference / period division unit 14, only the remainder r (2) is smaller than the predetermined value ε. For example, the nth time difference τ is classified into a group of time differences that appear in the set period τ (2).

The same column determination unit 15 determines that all the remainders r (p) output from the peak time difference / period division unit 14 are larger than a predetermined value ε (P remainders r (1), r (2),... , R (P), the remainders r (1), r (2),..., R (P) are all greater than a predetermined value ε), and the nth time difference τ is any It is determined that it does not belong to the column of the period τ (p), and the nth time difference τ is newly set to the (P + 1) th period τ (P + 1).
Note that the P + 1-th cycle τ (P + 1) set by the same column determination unit 15 is stored by the cycle storage unit 12.

The process end determination unit 16 determines whether or not the determination process of the same column determination unit 15 has ended for all the time differences τ specified by the peak column extraction unit 3, and if the determination process has not ended, The next peak time difference selection unit 13 is instructed to select a time difference τ that has not yet been selected at the present time.
On the other hand, if the determination process is completed, the cycle storage unit 12 is instructed to output the set cycle τ (p).
If M periods τ (p) are set, M periods τ (1), τ (2),..., Τ (M) are transmitted from the period storage unit 12 to the signal separation unit 5. Is output.

Here, the peak time difference / period division unit 14 has been shown to calculate mod (τc, τ (p)), but the remainder r (p) returned from the function of mod (τc, τ (p)) is Only positive values.
However, since the peak appearance error appears in both positive and negative directions, in addition to calculating mod (τc, τ (p)), τ (p) −mod (τc, τ (p)) is also calculated. You may make it perform the test | inspection which compares the remainder r (p) obtained with predetermined value (epsilon).
The test formula is as shown in the following formula (16).

  In addition, here, the processing end determination unit 16 determines whether or not the determination processing of the same column determination unit 15 has been completed for all the time differences τ specified by the peak column extraction unit 3. It is determined whether or not a desired number of periods τ (p) are set. If a desired number of periods τ (p) are set, all time differences τ specified by the peak string extraction unit 3 are determined. Even if the determination process is not completed, the cycle storage unit 12 may be instructed to output the set cycle τ (p).

When the signal separating unit 5 receives M periods τ (1), τ (2),..., Τ (M) from the period estimating unit 4, each of the incident signals s interfering with the received signal x is received. Assuming that the period coincides with one of M periods τ (1), τ (2),..., Τ (M), interference occurs with the whitened reception signal z output from the whitening unit 1. The signal separation of the plurality of incident signals s is performed (step ST6).
That is, the signal separation unit 5 sequentially sets M periods τ (1), τ (2),..., Τ (M) output from the period estimation unit 4 as time difference parameters in TDSEP, By minimizing the time difference cross-correlation value of the whitened reception signal z in the set period τ (p), the incident signal s having the period τ (p) is separated from the whitened reception signal z.

Here, the period estimation unit 4 outputs M periods τ (1), τ (2),..., Τ (M) to the signal separation unit 5, and the signal separation unit 5 uses the time difference parameter in TDSEP as Although M cycles τ (1), τ (2),..., Τ (M) are sequentially set, in order to reduce the processing load of TDSEP, M cycles τ (1 ), Τ (2),..., Τ (M), when a certain period is an integer multiple of the other periods, the period estimation unit 4 determines only one of the periods as the signal separation unit 5. The other period may not be output to the signal separation unit 5.
For example, when the period τ (4) is an integral multiple of the period τ (1), only one of the period τ (4) or the period τ (1) is output to the signal separation unit 5 and the other period is signal-separated. Do not output to part 5. In this case, if the period τ (1) is set as the time difference parameter in TDSEP, the period τ (4) is not set, and if the period τ (4) is set, the period τ (1) is not set.

Hereinafter, the simulation result of the signal separation of the incident signal by the signal separation device of FIG. 1 will be described.
Here, a received signal (periodic Gaussian signal) in which three waves of an incident signal with a period of “30”, an incident signal with a period of “74”, and an incident signal with a period of “77” are mixed is received. A simulation result in the case of being performed will be described.
FIG. 7 is an explanatory diagram showing time difference autocorrelation of incident signals of three waves.
As is apparent from FIG. 7, each incident signal has a peak periodically. However, the period of the incident signal is unknown because it is not known in advance.

FIG. 8 is an explanatory diagram showing a time difference cross-correlation of the whitened reception signal, and FIG. 9 is an explanatory diagram showing a time difference τ (τ = 30, 74, 77) set as a time difference parameter in TDSEP.
FIG. 10 is an explanatory diagram showing simulation results of incident signal separation performance in the signal separation device according to the first embodiment and the conventional signal separation device.
As is apparent from FIG. 8, a peak appears in the time difference cross-correlation of the whitened reception signal z at the period of each incident signal s.
In this simulation, as shown in FIG. 9, the time difference τ (τ = 30, 74, 77) is set as the time difference parameter in TDSEP, and the periods “30”, “74”, “ 77 ″.
According to the signal separation device according to the first embodiment, as shown in FIG. 10, the separation performance of incident signals is approximately higher than that of a conventional signal separation device (a signal separation device using a method of randomly selecting a time difference τ). 20 dB improvement.

  As is apparent from the above, according to the first embodiment, the whitening unit 1 that whitens the received signal x mixed with a plurality of incident signals s from different signal sources, and the whitening unit 1 performs whitening. Generating a time difference cross-correlation matrix Rz (τ) of the whitened received signal z using the time difference τ between the peaks in the whitened received signal z, which is the received signal, as a variable, and changing an arbitrary time difference τ within a predetermined range A time difference cross-correlation calculation unit 2 that calculates a sum of squares evaluation function value J (τ) that is a cross-correlation value of each time difference τ by sequentially substituting into the cross-correlation matrix Rz (τ) Among the square sum evaluation function values J (τ) of the respective time differences τ that are larger than the square sum evaluation function values J (τ−1) and J (τ + 1) of the adjacent time differences τ−1 and τ + 1, and , A sum-of-squares evaluation function value J greater than a predetermined threshold value threshold τ) is detected, and the peak sequence extraction unit 3 that identifies the time difference τ corresponding to the square sum evaluation function value J (τ) is the same among the plurality of time differences τ specified by the peak sequence extraction unit 3 The time difference τ appearing in the period is classified into the same group, and a period estimation unit 4 for estimating the period τ (p) of the time difference τ belonging to each group is provided, and the signal separation unit 5 determines the period of the incident signal s. Since it is configured to perform signal separation of a plurality of incident signals s interfering with the whitened reception signal z on the assumption that it coincides with the period τ (p) estimated by the period estimation unit 4, the incident is performed in advance. Even if the period of the signal s is not known, it is possible to automatically select an appropriate time difference τ and to separate a plurality of incident signals s interfering with the received signal x.

  Further, according to the first embodiment, when the period of the time difference τ belonging to a certain group is an integral multiple of the period of the time difference τ belonging to another group, the period estimation unit 4 is either Since only the period of the time difference τ belonging to the group is output to the signal separation unit 5 as an estimation result, useless TDSEP can be omitted, resulting in degradation of signal separation performance. The TDSEP processing load in the signal separation unit 5 can be reduced without incurring the above.

  Further, according to the first embodiment, the period estimation unit 4 selects an arbitrary time difference τ from the plurality of time differences τ specified by the peak string extraction unit 3 and sets the time difference τ to the set period τ. If the remainder r (p) when divided by (p) is smaller than the predetermined value ε, the time difference τ is classified into a group of time differences τ appearing in the period τ (p), and the remainder r (p) is the predetermined value ε. If it is larger, the selected time difference τ is classified into a group of time differences τ appearing in a period other than the period τ (p), so that the time difference τ appearing in the same period can be classified into the same group. Play.

Embodiment 2. FIG.
In the first embodiment, the period estimation unit 4 selects an arbitrary time difference τ from a plurality of time differences τ specified by the peak string extraction unit 3, and the time difference τ is set with a set period τ (p). If the remainder r (p) in the case of division is smaller than the predetermined value ε, the time difference τ is classified into a group of time differences τ appearing in the period τ (p), and if the remainder r (p) is larger than the predetermined value ε, Although the selected time difference τ is classified into a group of time differences τ appearing in a period other than the period τ (p), as shown in FIG. 11, the period estimation unit 4 includes a PRI conversion unit 4a, and the PRI The conversion unit 4a performs PRI conversion on the plurality of time differences τ identified by the peak string extraction unit 3, thereby identifying the time difference τ that appears in the same period among the plurality of time differences τ, and grouping the time differences τ (The time difference τ that appears in the same period And the same effects as those of the first embodiment can be obtained.
Since PRI conversion is a known technique, detailed description thereof is omitted (see, for example, Japanese Examined Patent Publication No. Sho 62-26603), but from peak sequences in which peak sequences having different periods are mixed, This is an algorithm for detecting a peak train and detecting the period of each peak train.

  In the present invention, within the scope of the invention, any combination of the embodiments, or any modification of any component in each embodiment, or omission of any component in each embodiment is possible. .

  DESCRIPTION OF SYMBOLS 1 Whitening part (whitening means) 2 Time difference cross-correlation calculation part (cross-correlation value calculation means) 3 Peak sequence extraction part (peak correlation value detection means) 4 Period estimation part (period estimation means) 4a PRI conversion , 5 signal separation unit (signal separation unit), 11 first period setting unit, 12 period storage unit, 13th order peak time difference selection unit, 14 peak time difference / period division unit, 15 identical column determination unit, 16 processing end determination unit .

Claims (6)

  A whitening means for whitening a reception signal in which a plurality of incident signals with different signal sources interfere with each other, and a time difference between the reception signals using a time difference between peaks in the reception signal whitened by the whitening means as a variable. A correlation matrix is generated, and an arbitrary time difference within a predetermined range is sequentially substituted into the time difference cross-correlation matrix to calculate a cross-correlation value of each time difference, and calculated by the cross-correlation value calculation means Among the cross-correlation values of each time difference, a cross-correlation value that is larger than a cross-correlation value of adjacent time differences and larger than a predetermined threshold is detected, and a time difference corresponding to the detected cross-correlation value is determined. Among the plurality of time differences identified by the peak correlation value detection means to be identified and the peak correlation value detection means, the time differences that appear in the same period are classified into the same group and belong to each group. Assuming that the period of the incident signal coincides with the period estimated by the period estimating means, and interfering with the reception signal whitened by the whitening means. A signal separation device comprising signal separation means for separating a plurality of incident signals.   The signal separation means performs TDSEP, which is an independent component analysis using a time difference cross-correlation value as an independent index, and minimizes the time difference cross-correlation value of the received signal in the period estimated by the period estimation means. 2. The signal separation device according to claim 1, wherein signal separation is performed on a plurality of incident signals interfering with the received signal.   When the period of time difference belonging to a group is an integer multiple of the period of time difference belonging to another group, the period estimation means uses only the period of time difference belonging to one of the groups as the estimation result. 3. The signal separation device according to claim 1, wherein the signal separation device outputs the signal to signal separation means.   The period estimating means selects an arbitrary time difference from a plurality of time differences specified by the peak correlation value detecting means, and if the remainder when the time difference is divided by the estimated period is smaller than a predetermined value, the time difference is calculated. 3. The time difference group that appears in the period is classified, and if the remainder is larger than a predetermined value, the selected time difference is classified into a time difference group that appears in a period other than the period. The signal separation device as described.   The period estimation unit performs PRI conversion on the plurality of time differences identified by the peak correlation value detection unit, thereby identifying a time difference that appears in the same period among the plurality of time differences, and sets the same time difference that appears in the same period. 3. The signal separation device according to claim 1, wherein the signal separation device is classified into groups.   The whitening means whitens a received signal in which a plurality of incident signals with different signal sources are interfering, and the cross-correlation value calculating means includes a whitening process in the received signal whitened in the whitening process step. A cross-correlation matrix of the received signal using the time difference between peaks as a variable is generated, and an arbitrary time difference within a predetermined range is sequentially substituted into the time difference cross-correlation matrix to calculate a cross-correlation value of each time difference. The correlation value calculation processing step and the peak correlation value detection means are larger than the cross-correlation values of adjacent time differences among the cross-correlation values of the time differences calculated in the cross-correlation value calculation processing step, and are predetermined. A peak correlation value detection processing step for detecting a cross-correlation value greater than a threshold value of the two and identifying a time difference corresponding to the detected cross-correlation value; Among the plurality of time differences identified in the processing step, the time difference appearing in the same period is classified into the same group, the period estimation processing step for estimating the period of the time difference belonging to each group, and the signal separation means, Assuming that the period of the incident signal coincides with the period estimated in the period estimation processing step, signal separation that performs signal separation of a plurality of incident signals interfering with the reception signal whitened in the whitening processing step And a signal separation method.

Images