JP5094281B2 - Signal separation device - Google Patents

Description

  The present invention relates to a signal separation device that separates and extracts a source signal from an observation signal in which multiple source signals interfere.

  First, before explaining the prior art, an independent component analysis (ICA) and its uncertainty will be briefly explained.

  The ICA separates and extracts a source signal from the received observation signal by using a stochastic statistical property that a plurality of reception sensors receive signals mixed with a plurality of source signals and each source signal is independent of each other. It is a technique. This technique itself is known and widely known. FIG. 15 is a conceptual diagram of ICA processing.

In ICA, an observation signal having a certain time length N is statistically processed batchwise to calculate a separation signal. X _i (i = 1,..., I) is a signal obtained by interfering with the J-wave source signal s _j (j = 1,..., J). And Each is a signal arranged in a time series transmitted and received from time to time, that is, s _j = {s _j (k)} = {s _j (1), s _j (2),..., S _j (N )}, X _i = {x _i (k)} = {x _i (1), x _i (2),..., X _i (N)}. Here, I, J, and N are each a positive integer of 2 or more. At this time, the mathematical model of the interference observation signal is given by the following formula (1).

Here, s = [s ₁ ... S _J ] ^T is a source signal vector in which source signals s _j (j = 1,..., J) are arranged in a vector shape, and A means interference. X = [x ₁ ... X _I ] ^T is an observation in which observation signals x _i (i = 1,..., I) are arranged in a vector shape. It is a signal vector. ICA assumes stochastic independence between source signals and estimates A (^) of mixing matrix A (where the notation (^) is the character preceding the parenthesis so that the separated signals are independent). The source signal is separated by multiplying the observed signal vector x by the following equation (2).

y (^) = [y ₁ (^) ... y _J (^)] ^T is a separated signal vector, and y _j (^) = {y _j (^) (k)} = {y _j ( ^) (1), y _j (^) (2)..., Y _j (^) (N)]. If ICA can correctly estimate the mixing matrix A as shown in the above equation (2), A (^) ⁻¹ A becomes a value close to the unit matrix I, so that the source signal s can be separated and extracted.

  However, the ICA observation model represented by the above equation (1) can be modified as the following equation (3) or (4). In order to simplify the description, a case where the number of source signals is two will be described.

Here, A _j (j = 1,..., J) is a column vector of the mixing matrix A, and α _j (j = 1,..., J) is an arbitrary coefficient. The above equation (3) indicates that the observed signal vector x does not change even if an arbitrary coefficient α _j is present. In other words, no matter how the observed signal vector x is processed, it means that the value of α _j cannot be known only from the information of x. That is, a certain unknown coefficient is applied to the separated signal obtained by ICA.

On the other hand, the above equation (4) means that the observed signal vector x does not change even if the order of s _j is changed and the columns of the mixing matrix A are changed. In other words, no matter how the observation signal vector x is processed, it means that the order of the source signal s _j cannot be known only from the information of the observation signal vector x. This is the uncertainty of the order of the separated signals, that is, the obtained separated signal vector y (^) has the source signal estimated values s ₁ (^) and s ₂ (^) as elements. It means that it is not known whether it is the formula (5) or (6).

  The above is an explanation of ICA uncertainty.

  Due to the above property, if the observation vector obtained from time to time by ICA is divided and processed for every certain length of time (each block), the uncertainties remain in the correspondence between the separated signals. is there. If the uncertain signal is connected as it is, the source signal cannot be correctly restored. For this reason, it is necessary to determine the association by some method.

There is a conventional technique for determining the correspondence to the above problem (see, for example, Non-Patent Document 1). FIG. 16 is a block diagram of a conventional signal separation device that associates separation signals. FIG. 17 is a diagram for explaining the outline of processing by a conventional signal separation device. In this conventional method, the correspondence is determined by the similarity between the blocks of the column vector A _j (（) of the mixed matrix estimation value A (（) obtained as the ICA output.

Now, it is assumed that the signal separation device 10 finishes a series of processing up to the p-1th block (corresponding to the previous block) and starts processing of the pth block (corresponding to the current block). In this conventional method, first, an interference signal received from time to time is divided into blocks each having a certain time length by the block dividing unit 20 and output to the independent component analyzing unit 30. Here, the p-th observation signal vector is referred to as x ^(p) . The independent component analysis unit 30 performs a separation process on the observation signal vector x ^(p) .

As a result, the separated signal y ^(p) and the mixing matrix estimation value A (^) ^(p) are obtained. Next, as described in the above equation (4), the order of the column vector A _j (^) ^(p) (j = 1,..., J) of the mixed matrix estimation value A (^) ^(p) is , Corresponding to the order of the separated signals s _j (^) ^(p) , the correspondence of the separated signals between the blocks is determined by the similarity of the column vector A _j (^) ^(p) .

However, here, the normalized column vector D _j ^(p) obtained by normalizing each column vector A _j (^) ^(p) of the mixed matrix estimation value A (^) ^(p) by the normalized column vector calculation unit 50 is used. Use (refer to Non-Patent Document 1 for details). For the determination of similarity, the tracking processing by the tracking processing unit 60 is used by using various information such as the normalized column vector D _j ^(p) and the phase value obtained therefrom as a tracking filter input vector. More specifically, the normalized column vectors {D _j ⁽¹⁾ ,..., D _j ^(p−1) from the first block to the p−1 block that have already been associated by correlation processing. } And the normalized column vector D _j ^{(p) of} the p-th block are determined.

  FIG. 18 is a block diagram of a correlation processing unit 70 included in a conventional tracking processing unit, and only the correlation processing unit is briefly touched (for details of the tracking processing unit, refer to Non-Patent Document 1). . For example, when a GNN (Global Nearest Neighbor) tracking filter is used, the correlation processing unit 70 first calculates a residual secondary format as shown in the following equation (7) by the residual secondary format calculation unit 71.

Here, j1 is the number of the tracking track, j2 is the number of the tracking filter input vector, z _j1 (−) ^(p) is the track prediction vector (where the notation (−) is the character preceding the parenthesis) Z _j2 ^(p) is the tracking filter input vector, d _{i1 — i2} is the track and tracking normalized by the residual quadratic form (s _j1 ^(p)) (Distance of filter input vector), d is a gate radius (residual quadratic threshold), and s _j1 ^(p) is a residual covariance matrix.

  Next, the residual secondary form threshold determination unit 72 and the optimum combination calculation unit 73 based on the residual secondary form perform correspondence determination based on the correlation processing formula as shown in the following formula (8).

  The above equation (8) means that the optimum correspondence is selected so that one tracking filter input vector corresponds to each track without overlapping and the sum of the residual quadratic form is minimized. is there. This determination result is a determination result of correspondence between the standardized column vectors from the first block to the (p-1) th block and the standardized column vector of the pth block.

  Here, d is a threshold value for the residual quadratic form, and by imposing an upper limit threshold value, the distance is shorter than a certain length, that is, the correlation is considered only for the tracking filter input vector having a high possibility of correspondence. It means to do. Note that if this value is set to infinity, there is no case of disqualification in the threshold determination, so that correlation can be established without fail.

Finally, the separated signal s _j (^) that has been associated and concatenated from y _j (^) ⁽¹⁾ to y _j (^) ^(p−1 ), and the separated signal y _j (^) ^(p) Are connected corresponding to the correlation result of the normalized column vector and output. FIG. 19 is a block diagram of a conventional connection processing unit 40. In connecting them, first, the unknown coefficient removal unit 41 needs to remove the unknown coefficient α _j ^(p) described in the above equation (3). If they are connected without removing them, separated signals scaled with different α _j ^(p) for each block will be connected, and will not be connected correctly. Assume that the output results of ICA are correctly separated. That is, the following expressions (9) and (10) are established.

Then, by selecting an arbitrary row i ′ of the mixing matrix estimation value A (^) and multiplying the corresponding element by the separation signal, the uncertainty α _j ^(p) is expressed by the following equation (11). Can be canceled.

As in the above equation (11), with unknown coefficients removal unit 41, _{A I'i} ^{(^) (p)} in scaled separated signal ^{_{A i'j (^) (p)}} y j (^) (p) _Is newly replaced with y _j (^) ^(p) . Then, the separation signal y of the p-th block corresponding to the separation signal s _j (＾) that has been associated and concatenated from y _j (＾) ⁽¹⁾ to y _j (＾) ^(p−1) in the combining unit 42. _j (^) ^(p) is connected.

By repeating the above processing for the observation signals that are sequentially input, the separation signals are sequentially associated, and the concatenation result s _j (^) is sequentially output.

"Separated signal ID estimation method using mixing matrix of independent component analysis", 2006 IEICE National Convention, A-4-22, March 2006

However, the prior art has the following problems.
As described above, in the prior art, the separation signal correspondence is determined by the tracking processing of the (normalized) column vector of the mixing matrix estimation value. However, in a poor radio wave environment such as low S / N, the mixing matrix estimation value varies between blocks due to deterioration of the separation performance of ICA. That is, since the variation of the column vector between the blocks is large, the values of the column vectors between the blocks are remarkably different, and the probability that a correlation error or correlation is not established increases. As a result, there is a problem that the accuracy rate of the association method is lowered.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to obtain a signal separation device that can improve the accuracy rate of separation signal correspondence.

The signal separation device according to the present invention receives a signal in which a plurality of source signals interfere with each other as a plurality of time series observation signals via a plurality of reception sensors, and each of the plurality of time series observation signals has a certain length of time. A block division unit that divides each block, and an independent component analysis unit that extracts a plurality of separated signals and mixed matrix estimation values by performing separation processing by applying independent component analysis to a plurality of observation signals divided by the block division unit And a correlation processing unit that associates a plurality of separated signals extracted from the current block with a plurality of separated signals extracted from the previous block based on the extraction result by the independent component analyzing unit, and a correlation processing unit In the signal separation device including a concatenation processing unit that concatenates each associated separated signal and generates a concatenated signal corresponding to a plurality of source signals, the block dividing unit includes: When dividing the observation signals of the column for each block is divided by partially overlap the preceding and succeeding blocks in time, the correlation processing unit, a plurality of separation signals the extracted from the current block, from the previous block Correlation processing is performed in the overlapping portion with the plurality of extracted separation signals, and a plurality of separation signals extracted from the current block and a plurality of separation signals extracted from the previous block are obtained based on the correlation processing result in the overlapping portion. The association is performed .

  According to the present invention, it is possible to reduce the variation between the blocks of the column vector by partially overlapping the blocks, and further to use the correlation coefficient value of the overlapping portion of the separation signal between the blocks together with the tracking correlation processing. Thus, it is possible to improve the correlation determination accuracy rate, and to obtain a signal separation device that can improve the correlation accuracy rate.

  Hereinafter, preferred embodiments of a signal separation device of the present invention will be described with reference to the drawings.

Embodiment 1 FIG.
The main points of the problem solving method in the signal separation device of the present invention are broadly divided into the following three points.
1) Points where processing blocks are duplicated 2) Points where correlation coefficient values between separated signals at overlapping locations are used in combination with tracking correlation processing 3) Points where connection processing is performed in consideration of overlapping locations. This will be described in detail.

  FIG. 1 is a block diagram of a signal separation device according to Embodiment 1 of the present invention. Compared with the signal processing apparatus of FIG. 16 corresponding to the prior art, the signal processing apparatus of FIG. 1 uses the overlapping block division unit 20a instead of the block division unit 20, and the function of the correlation processing unit 70a. Is different.

Moreover, FIG. 2 is a figure explaining the outline | summary of the process of the signal separation apparatus in Embodiment 1 of this invention. Now, it is assumed that the signal separation device 10 finishes the association and separation signal connection up to the (p-1) th block and starts the processing of the pth block. The overlapping block dividing unit 20a obtains an observation signal of the p-th block by partially overlapping the separation result s _j (^) up to the (p-1) _-th block from the signal received from time to time. (See FIG. 2).

It is assumed that the observation signal x ^(p) of the p-th block takes time k of K _a ^(p) ≦ k <K _b ^(p) . At this time, the p-th block length is N ^(p) = K _b ^(p) −K _a ^(p) , and the overlap width is W ^(p) = K _b ^(p−1) −K _a ^(p). . Here, the overlap width and block length can be set / changed to arbitrary values for each block, but in order to overlap, the condition of K _b ^(p−1) ≧ K _a ^(p) is necessary. . In addition, it is not limited to the (p-1) th, but may be overlapped retroactively.

Next, the independent component analysis unit 30 performs the separation process of the observation signal block acquired by the overlapping block division unit 20a, and obtains the mixed matrix estimation value A (^) ^(p) and the separated signal y (^) ^(p) . . Then, as described in the related art, the correlation between the blocks in the column vector is associated by using the correlation processing in the subsequent stage, thereby associating the separated signals between the blocks.

  Here, since the independent component analysis unit 30 separates the signals by the probability statistical processing, there is a common part in the observation signal between the blocks as the p-th block overlaps the p-1 block. For this reason, the result of the probability statistical process becomes similar, and as a result, the variation of the mixed matrix estimation value can be suppressed to be small (in the extreme case, if the overlap is 100%, the p-1 block and the pth block). Since the blocks handle the same observed signal, the same mixed matrix estimate is obtained (ie, no variation occurs).

However, even if it is excessively overlapped, the processing in the time direction does not proceed. Therefore, the degree of overlap W ^(p) is reduced according to the radio wave environment so that the variation is small enough to enable tracking. An appropriate redundancy W ^(p) may be set.

  The second point of the present invention is to use both the correlation processing based on the correlation coefficient and the tracking correlation processing. However, it is possible to associate the separated signals only by the correlation processing using the correlation coefficient, and this method will be briefly described. FIG. 3 is a block diagram of the signal separation device according to Embodiment 1 of the present invention. In the case of the signal separation device 10 shown in FIG. 3, the tracking process is not used, and only the correlation coefficient is used as the determination criterion.

  The independent component analysis unit 30 performs a separation process on the p-th observation signal vector obtained by the overlapping block division unit 20a. Therefore, the correlation processing unit 70b calculates a correlation coefficient in the overlapping part of the p-th block separation signal and the separation signal that has been processed so far.

  Further, the correlation processing unit 70b determines a correspondence having a high correlation coefficient value in the overlapping part of the p-th block separation signal and the separation signal processed so far. The associated separation signals are output as final separation results after being connected by the connection processing unit 40.

  Next, the correlation processing unit 70b will be described in more detail. FIG. 4 is a block diagram of correlation processing unit 70b according to Embodiment 1 of the present invention. The correlation coefficient calculation unit 74 calculates the correlation coefficient with brute force at a place where time overlaps according to the following equation (12).

Here, j1 is the number of the signal that has been processed up to the (p−1) _-th block, and s _j1 (＾) (k) is the separated signal result. J2 is the number of the separated signal of the p-th block, y _j2 (^) ^(p) (k) is the separated signal of the p-th block, and corr is a correlation coefficient operator. The higher the correlation coefficient, the higher the similarity between separated signals, so there is a higher possibility of corresponding. Therefore, a combination that maximizes the sum of the correlation coefficients may be determined based on each correlation coefficient value according to the following equation (13) (based on the correlation coefficient threshold value determination unit 75 and the correlation coefficient value). Optimal combination calculator 76).

  Here, C is a threshold for the correlation coefficient, and means that a correlation is determined only for a separated signal that has a certain degree of correspondence possibility by imposing a lower limit constraint on the correlation coefficient. If this value is set to 0, there is no correspondence that is disqualified in the threshold determination, so that the correlation can be established without omission.

  Next, returning to FIG. 1, a method of using the correlation coefficient value in the tracking correlation process will be described. FIG. 5 is a block diagram of correlation processing unit 70a according to Embodiment 1 of the present invention. As shown in FIG. 5, from both the combination determination result 1 that is the tracking correlation processing result of FIG. 18 and the combination determination result 2 that is the correlation coefficient correlation processing result of FIG. Use it to make a final decision and determine the final combination.

Finally, the connection processing unit 40 shown in FIG. 1 will be described. The point here is that, unlike FIG. 19 which is the prior art, the separated signal s _j (^) up to the p−1 block and the signal y _j (^) ^{(p) of} the p-th block are overlapped. Since it exists, a process for selecting either one is required.

FIG. 6 is an explanatory diagram of a method for selecting s _j (＾) at the overlapping portion in the concatenation processing unit 40 according to the first embodiment of the present invention. FIG. 7 is a block diagram of the connection processing unit 40 according to Embodiment 1 of the present invention. As shown in FIGS. 6 and 7, when s _j (^) is selected in the overlapping portion of the separated signal, the signal is concatenated to the signal pair determined to be associated as in the following equation (14).

Of course, it is also conceivable to select y _j (）) ^(p) in the reverse direction, that is, in the overlapping portion. Since the method is clear, description is omitted.

  As described above, according to the first embodiment, by overlapping the processing blocks, the variation of the standardized column vector between the processing blocks is reduced, so that tracking becomes easy and the correspondence accuracy rate of the separated signal is increased. improves. Further, by using the correlation coefficient value between the separated signals at overlapping portions together in the association determination, the correlation accuracy is improved as the amount of information increases, and the association accuracy rate is improved.

  Note that, as described above, it is not always necessary to use the correlation coefficient value between the separated signals at overlapping portions in combination with the tracking correlation process. It is also possible to improve the correspondence correct rate by using only the correlation coefficient values between the separated signals at the overlapping portions.

Embodiment 2. FIG.
In the second embodiment, a correlation processing unit 70c having a new form different from the correlation processing unit 70a of FIGS. 1 and 5 described in the first embodiment will be described. In this correlation processing unit 70c, a method of determining an optimization combination based on a value obtained by weighting the correlation coefficient value and the residual quadratic form will be described.

  FIG. 8 is a block diagram of correlation processing unit 70c in the second embodiment of the present invention. The weighting threshold value determination unit 78 and the optimum weighting combination calculation unit 79 shown in FIG.

  Here, α is a weighting coefficient having a value from 0 to 1, and h is a scaling coefficient that adjusts the value of the reciprocal of the residual quadratic form and the correlation coefficient to the same level.

  As described above, according to the second embodiment, the correlation coefficient value and the residual quadratic form are weighted based on the same scaling, the correlation is determined by the weighting optimization problem, and the combined signal combination determination result is obtained. Can do. Also with such a method, the correlation coefficient value between the separated signals at the overlapping portion can be used together with the tracking correlation processing, and the correspondence accuracy rate can be improved.

Embodiment 3 FIG.
In the third embodiment, a new form different from the correlation processing unit 70a of FIGS. 1 and 5 described in the first embodiment or the correlation processing unit 70c of FIG. 8 described in the second embodiment. The correlation processing unit 70d having In this correlation processing unit 70d, a method of performing association using correlation coefficients again for those for which correlation has not been established after tracking correlation processing will be described.

  FIG. 9 is a block diagram of the correlation processing unit 70d according to Embodiment 3 of the present invention. The correlation processing unit 70d illustrated in FIG. 9 first performs tracking correlation processing. Next, assuming that the set of the number of the track and the tracking filter input vector in which the correlation is established is Φ, the correlation processing based on the correlation coefficient is performed on the remaining signals not included in the set Φ according to the following equation (16). .

  As described above, according to the third embodiment, the tracking correlation processing in the residual quadratic format is performed with priority, and the correlation processing based on the correlation coefficient is performed on the remaining signals for which correlation is not correctly obtained. ing. Also by such a method, the correlation coefficient value between the separated signals at the overlapping portion can be used together with the tracking correlation process, and the correspondence accuracy rate can be improved while reducing the amount of calculation processing.

  The reverse is also true. That is, the correlation may be determined by performing tracking correlation processing on the remaining input values that cannot be correlated by the correlation processing based on the correlation coefficient. Since the method is clear, the description is omitted.

Embodiment 4 FIG.
In the fourth embodiment, a correlation processing unit 70e having a new form will be further described. In this correlation processing unit 70e, a method of using two threshold determination criteria simultaneously will be described.

  FIG. 10 is a block diagram of correlation processing unit 70e in the fourth embodiment of the present invention. The correlation processing unit 70e illustrated in FIG. 10 performs association processing based on the residual quadratic format only for input values that exceed the threshold values of both the residual quadratic format and the correlation coefficient. For example, when both threshold determination units 72a are used and a combination that satisfies both thresholds is found such that the sum of the residual quadratic form is minimized, the following equation (17) is obtained.

In this _case , both conditions of d _{j1 — j2} and C _{j1 — j2} are satisfied, so that a more reliable signal is associated based on the residual quadratic form. If the set of signal numbers correlated in the above equation (17) is Φ, the remaining input values not included in the set Φ because the correlation is not established are expressed by the following equation (18): The association may be performed based on the correlation coefficient.

  As described above, according to the fourth embodiment, after performing the tracking correlation process in the residual quadratic format and the correlation process based on the correlation coefficient, the residual quadratic is obtained for those satisfying both threshold conditions. Corresponding by the format is performed, and correlation processing based on the correlation coefficient can be performed on the remaining signals in which both threshold conditions cannot be satisfied. Even with such a method, the correlation coefficient value between the separated signals at the overlapping portion can be used together with the tracking correlation processing, and it becomes possible to perform the association based on the residual quadratic form for the reliable signal. Therefore, it is possible to improve the correspondence correct answer rate.

  Note that the tracking correlation process and the correlation coefficient correlation process may be included. In other words, those that satisfy both threshold conditions are associated by correlation processing based on the correlation coefficient, and based on the residual quadratic form for the remaining signals that fail to satisfy both threshold conditions. Correlation processing can be performed. Since the method is clear, the description is omitted.

  In addition, a method of finding a combination for one satisfying only one condition is also conceivable. Since the AND process of the above equation (17) is only the or process, the details are omitted.

Embodiment 5 FIG.
In the fifth embodiment, a connection processing unit 40a having a new form different from the configuration of FIG. 7 of the first embodiment will be described. First, as in the conventional method, the unknown coefficient removal unit 41 removes the unknown coefficient α _j ^(p) described using the above equation (3). Next, in order to connect the corresponding signals, it is necessary to perform some processing on the signal duplication part. In the first embodiment, the case has been described in which one of the separation signal up to the (p−1) -th block or the separation signal of the p-th block is selected. On the other hand, in the fifth embodiment, a method of taking the average of overlapping portion signals will be described.

FIG. 11 is an explanatory diagram of a method of averaging at overlapping portions in the concatenation processing unit 40a according to the fifth embodiment of the present invention. FIG. 12 is a block diagram of the connection processing unit 40a according to the fifth embodiment of the present invention. As shown in FIGS. 11 and 12, when the average of overlapping portions is taken, the overlapping portion average signal calculation unit 43 uses signal pairs determined to be associated as shown in the following equation (19), By using the average value, the final separated signal s _j (＾) after the concatenation is obtained.

  As described above, according to the fifth embodiment, the average value of the separated signal up to the (p−1) -th block and the separated signal of the p-th block can be used as the overlapping portion signal. As a result, both the values up to the past p-1 block and the current value of the pth block can be taken into account as the overlapping signal.

Embodiment 6 FIG.
In the sixth embodiment, a connection processing unit 40b having a new form different from the connection processing unit 40a of FIGS. 11 and 12 described in the fifth embodiment will be described. First, as in the conventional method, the unknown coefficient removal unit 41 removes the unknown coefficient α _j ^(p) described using the above equation (3). Next, in order to connect the corresponding signals, it is necessary to perform some processing on the signal duplication part.

  In the first embodiment, the case has been described in which one of the separation signal up to the (p−1) -th block or the separation signal of the p-th block is selected. Further, in the above-described fifth embodiment, the case where the average of the signals of overlapping portions is taken has been described. On the other hand, in the sixth embodiment, a method for calculating the weighted sum of overlapping signals will be described.

  FIG. 13 is an explanatory diagram of a method for calculating a weighted sum at overlapping portions in the concatenation processing unit 40b according to the sixth embodiment of the present invention. FIG. 14 is a block diagram of the connection processing unit 40b according to Embodiment 6 of the present invention. The overlapped part weighted sum signal calculation unit 44 has a weight of the separated signal up to the (p−1) -th block of 0 in the first sample of the overlapped part, a weight of the separated signal of the p-th block is 1, and 1 / 2. In the final sample, the weighted sum signal is gradually changed according to the sample time so that the weight of the separated signal up to the p-th block is 1 and the weight of the separated signal of the p-th block is 0. Calculate

When taking the weighted sum of the overlapping parts, the overlapping part weighted sum signal calculating unit 44 uses the signal pair determined to be associated as shown in the following equation (20), and uses the value of the weighted sum for the overlapping part. Thus, the final separated signal s _j (^) after the connection is obtained.

As shown in the overlapping part (K _a ^(p) ≦ k <K _b ^(p−1) ) in the above equation (20), β = 1 at k = K _a ^(p) −1, so s _j (^) (K) = s _j (^) (k), and since k = 0 in k = K _b ^(p−1) , s _j (^) (k) = y _j (^) ^(p) (k The sum of signals weighted by β is used for k in the meantime.

  As described above, according to the sixth embodiment, the value of the weighted sum of the separated signal up to the (p−1) -th block and the separated signal of the p-th block can be used as the overlapping portion signal. As a result, both the values up to the past p-1 block and the current value of the pth block can be taken into account as the overlapping signal.

It is a block diagram of the signal separation apparatus in Embodiment 1 of the present invention. It is a figure explaining the outline | summary of the process of the signal separation apparatus in Embodiment 1 of this invention. It is a block diagram of the signal separation apparatus in Embodiment 1 of the present invention. It is a block diagram of the correlation process part in Embodiment 1 of this invention. It is a block diagram of the correlation process part in Embodiment 1 of this invention. It is explanatory drawing of the method of selecting _sj (^) in an overlap part in the connection process part of Embodiment 1 of this invention. It is a block diagram of the connection process part in Embodiment 1 of this invention. It is a block diagram of the correlation process part in Embodiment 2 of this invention. It is a block diagram of the correlation process part in Embodiment 3 of this invention. It is a block diagram of the correlation process part in Embodiment 4 of this invention. It is explanatory drawing of the method of averaging in an overlap part in the connection process part of Embodiment 5 of this invention. It is a block diagram of the connection process part in Embodiment 5 of this invention. It is explanatory drawing of the method of taking the weighting sum in an overlap part in the connection process part of Embodiment 6 of this invention. It is a block diagram of the connection process part in Embodiment 6 of this invention. It is a conceptual diagram of ICA processing. It is a block diagram of the conventional signal separation apparatus which performs matching between separation signals. It is a figure explaining the outline | summary of the process by the conventional signal separation apparatus. It is a block diagram of the correlation process part contained in the conventional tracking process part. It is a block diagram of the conventional connection process part.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Signal separator, 20 block division part, 20a Overlapping block division part (block division part), 30 Independent component analysis part, 40, 40a, 40b Concatenation processing part, 41 Unknown coefficient removal part, 42 Combining part, 43 Overlapping partial average Signal calculation unit, 44 overlapped partial weighted sum signal calculation unit, 50 normalized column vector calculation unit, 60 tracking processing unit, 70, 70a, 70b, 70c, 70d, 70e correlation processing unit, 71 secondary form calculation unit, 72 2 Next format threshold value determination unit, 72a Both threshold value determination unit, 73 Optimal combination calculation unit based on residual quadratic format, 74 Correlation coefficient calculation unit, 75 Correlation coefficient threshold value determination unit, 76 Optimal combination calculation unit based on correlation count value 77 Final combination determining unit, 78 Weighted threshold determining unit, 79 Weighted optimum combination calculating unit.

Claims (15)

A block dividing unit that receives signals mixed with a plurality of source signals as a plurality of time-series observation signals via a plurality of reception sensors, and divides each of the plurality of time-series observation signals into blocks each having a certain time length; ,
An independent component analyzer that extracts a plurality of separated signals and a mixed matrix estimation value by applying an independent component analysis to the plurality of observation signals divided by the block divider, and performing a separation process;
A correlation processing unit that associates a plurality of separated signals extracted from the current block with a plurality of separated signals extracted from the previous block based on the extraction result by the independent component analyzing unit;
A signal separation device comprising: a concatenation processing unit that concatenates the separated signals associated by the correlation processing unit and generates a concatenated signal corresponding to the plurality of source signals;
The block division unit, when dividing the time-series observation signal for each block, to divide the preceding and following blocks partially in time ,
The correlation processing unit performs a correlation process in an overlapping portion between a plurality of separated signals extracted from the current block and a plurality of separated signals extracted from the previous block, and based on a correlation processing result in the overlapping portion, A signal separation device that associates a plurality of separated signals extracted from a block with a plurality of separated signals extracted from a previous block . The signal separation device according to claim 1, wherein
The block dividing unit can arbitrarily set / change the overlap width and block length of each divided block when the time-series observation signal is divided for each block. . The signal separation device according to claim 1 , wherein
The correlation processing unit includes a plurality of separated signals extracted from a current block and a plurality of separated signals extracted from a previous block based on a tracking processing result based on the mixing matrix estimation value and a correlation processing result in the overlapping portion. A signal separation device characterized by performing association. The signal separation device according to claim 3 .
The correlation processing unit associates the plurality of separated signals extracted from the current block with the plurality of separated signals extracted from the previous block, based on the tracking processing result based on the mixing matrix estimation value. Based on the first combination determination result and the second combination determination result, the first combination determination result performed and the second combination determination result associated with the correlation processing result in the overlapping portion are obtained. And determining a final correspondence. The signal separation device according to claim 3 .
The correlation processing unit weights and adds each of the tracking processing result based on the mixing matrix estimation value and the correlation processing result in the overlapping portion, determines a correlation by a weighted optimization problem, and extracts a plurality of extracted from the current block A signal separation device that associates the separated signal with a plurality of separated signals extracted from the previous block. The signal separation device according to claim 3 .
The correlation processing unit associates a plurality of separated signals extracted from a current block with a plurality of separated signals extracted from a previous block based on a tracking processing result based on the mixing matrix estimation value, and performs the tracking processing A signal separation device characterized in that, with respect to a separated signal that has not been matched as a result, matching is performed based on a correlation processing result in the overlapping portion. The signal separation device according to claim 3 .
The correlation processing unit associates a plurality of separated signals extracted from the current block with a plurality of separated signals extracted from the previous block based on a correlation processing result in the overlapping portion, and responds based on the correlation processing result. A signal separation apparatus characterized in that a separation signal for which attachment is not established is associated with a result of tracking processing based on the mixing matrix estimation value. The signal separation device according to claim 3 .
The correlation processing unit includes a plurality of separated signals extracted from a current block and a plurality of separated signals extracted from a previous block based on a correlation processing result in the overlapping portion and a tracking processing result based on the mixing matrix estimation value. A signal separation apparatus characterized by performing association and determining association by tracking correlation processing for those satisfying both residual quadratic threshold determination and correlation coefficient threshold determination. The signal separation device according to claim 3 .
The correlation processing unit includes a plurality of separated signals extracted from a current block and a plurality of separated signals extracted from a previous block based on a correlation processing result in the overlapping portion and a tracking processing result based on the mixing matrix estimation value. A signal separation apparatus characterized by performing association and determining association by correlation coefficient correlation processing for those satisfying both residual quadratic form threshold determination and correlation coefficient threshold determination. The signal separation device according to claim 3 .
The correlation processing unit includes a plurality of separated signals extracted from a current block and a plurality of separated signals extracted from a previous block based on a correlation processing result in the overlapping portion and a tracking processing result based on the mixing matrix estimation value. A signal separation device characterized by performing association and determining association by tracking correlation processing for at least one of residual quadratic threshold determination and correlation coefficient threshold determination. The signal separation device according to claim 3 .
The correlation processing unit includes a plurality of separated signals extracted from a current block and a plurality of separated signals extracted from a previous block based on a correlation processing result in the overlapping portion and a tracking processing result based on the mixing matrix estimation value. A signal separation device characterized by performing association and determining association by correlation coefficient correlation processing for those satisfying at least one of residual quadratic threshold determination and correlation coefficient threshold determination. The signal separation device according to any one of claims 1 to 11 ,
The said connection process part employ | adopts the isolation | separation signal of a previous block as a signal of an overlap part, when connecting each isolation | separation signal matched by the said correlation process part. The signal separation device according to any one of claims 1 to 11 ,
The said connection process part employ | adopts the separation signal of the present block as a signal of an overlap part, when connecting each separation signal matched by the said correlation process part. The signal separation device according to any one of claims 1 to 11 ,
The concatenation processing unit employs an average value of the separation signal of the previous block and the separation signal of the current block as a signal of the overlapping portion when concatenating each separation signal associated by the correlation processing unit. A signal separator characterized by the above. The signal separation device according to any one of claims 1 to 11 ,
The concatenation processing unit adopts a weighted sum value of the separation signal of the previous block and the separation signal of the current block as the overlapping portion signal when concatenating each separation signal associated by the correlation processing unit A signal separating device.

Images