JP4649437B2 - Signal separation and extraction device - Google Patents

Description

  The present invention relates to a signal separation and extraction device for blindly separating a signal received by mixing a plurality of original signals and extracting each of the original signals.

  Blind separation is known as a technique for separating and extracting an original signal from a signal received by mixing a plurality of signals from different signal sources. In blind separation, signals generated from a plurality of different signal sources and superimposed on each other are received by a plurality of sensors, and the received signals are analyzed based on the statistical independence of each original signal. In this way, the original signal is separated and extracted.

  In this blind separation method, a series of data sampled during a predetermined period is regarded as one block, and separation processing is sequentially performed in units of the block. Each time the separation processing in units of blocks is completed, the separated signals are output in parallel from a plurality of output ports. In order to extract the original signal, it is necessary to further convert the separated signals output in units of blocks into continuous signals that are sequentially connected to each other. However, in the blind separation, the separated signals and these signals are output. The relationship with the output port number is not continuously maintained between the blocks, and the same original signal is not always continuously output from a specific output port. For this reason, a technique for extracting an original signal while dealing with such a change in the order of appearance of signals on the output port side is disclosed (for example, see Patent Document 1).

In the example disclosed in Patent Document 1, the original signal is extracted as a continuous signal while identifying the connection destination of the signal immediately after separation based on the degree of correlation with the signal separated in the immediately preceding block. ing.
Japanese Patent Laying-Open No. 2006-178314 (9th page, FIG. 1)

  However, signals received in a real environment are not always observed in a stable state. That is, the signal is received without including a large amount of waveform distortion or obtaining a sufficient signal-to-noise ratio necessary for signal processing. Further, when the signal is temporarily lost or the original signal is not a continuous signal but an intermittent signal, the reception is interrupted in time.

  For this reason, when blind separation is applied to a signal received under such an environment and the signal is connected using the correlation with the waveform or mixed vector obtained in the previous separation as described above, Since there is no significant difference in the correlation value, the correlation cannot be made and connection may be difficult. In particular, when a plurality of original signals include signals that are temporally interrupted, connection often fails, and a good separation and extraction result may not be obtained.

  The present invention has been made in consideration of the above-mentioned circumstances, and adapts to changes in the reception environment, while original signals with good reliability from signals received by mixing a plurality of signals from different signal sources. An object of the present invention is to provide a signal separation and extraction device that continuously separates and extracts a signal.

  In order to achieve the above object, a signal separation and extraction apparatus according to the present invention includes a plurality of sensors that receive a signal obtained by mixing a plurality of original signals, and the signals received by the plurality of sensors at predetermined intervals. Sampling means for sampling and obtaining sample data, and subjecting the sample data within this period to blind separation every predetermined period, and continuously obtaining the separated signal waveform and mixed vector corresponding to the original signal and the number of separated signals Based on the correlation between the signal separation means that performs the above and the mixed vector acquired by the signal separation means for a plurality of the predetermined periods based on the steering vector representing the directivity pattern formed by the arrangement of the plurality of sensors, respectively. The accumulated arrival direction information is stored in association with each other, and the maximum number of separated signals is calculated based on the direction value. Direction clustering means for generating the orientation clusters grouped into numbers corresponding to the values and associating the mixed vector immediately after separation by the signal separation means with the corresponding orientation clusters, and a plurality of the predetermined periods until immediately before The separated signal waveforms corresponding to the existing signals are extracted from the separated signal waveforms obtained by the signal separating unit over a minute, grouped into waveform clusters based on the cross-correlation between the separated signal waveforms, and sequentially stored. Waveform clustering means for associating the separated signal waveform immediately after separation by the signal separation means with these waveform clusters, and each of the separated signal waveforms immediately after separation by the signal separation means based on the results of the orientation clustering means and the waveform clustering means. For any one of the separated signal waveforms already connected Determining means for determining coupling destination, wherein the judgment result to the separation signal waveform based successively coupled to each have an extraction unit for extracting the plurality of original signals.

  According to the present invention, a signal capable of continuously separating and extracting an original signal with good reliability from a signal received by mixing a plurality of signals from different signal sources while adapting to changes in the reception environment. A separation and extraction device can be obtained.

  The best mode for carrying out the signal separation and extraction apparatus according to the present invention will be described below with reference to FIGS.

  FIG. 1 is a block diagram showing an embodiment of a signal separation and extraction apparatus according to the present invention. As illustrated in FIG. 1, the signal separation and extraction apparatus includes a sensor unit 10, a sampling unit 20, a signal separation unit 30, an orientation clustering unit 40, a waveform clustering unit 50, a connection destination determination unit 60, and an original signal extraction unit 70. It is composed of

  The sensor unit 10 includes a plurality of n sensors 11 to 1n. Each of these sensors is an antenna that receives an incident signal, for example, and is arranged in a predetermined shape such as a circle or a straight line to constitute the sensor unit 10. Then, a plurality of incoming original signals are received and sent to the sampling unit 20 respectively. The sampling unit 20 obtains sample data by sampling the reception signals from the sensors 11 to 1n of the sensor unit 10 at predetermined sampling intervals, and sends the sample data to the signal separation unit 30. The sampling unit 20 may include signal processing such as filtering, frequency conversion, and amplification performed within a range that does not lose information held in the signals when the signals from the sensor unit 10 are sampled.

  The signal separation unit 30 performs blind separation on the sample data from the sampling unit 20. As a method of blind separation, for example, a method called independent component analysis can be applied. When executing the blind separation, each sample data block, that is, continuous sample data within a predetermined period is set as one data block, and the processing is sequentially performed by using this data block as a unit to be separated. Therefore, for example, when the number of sample data in a data block is fixed to a predetermined value, the blind separation process is repeated at a fixed period corresponding to the data block length. As a result, the number of separated signal waveforms and mixed vectors and the number of separated signals corresponding to each of a plurality of original signals and not exceeding the number of sensors n are obtained as a result of each processing.

  The azimuth clustering unit 40 accumulates each mixture vector sent from the signal separation unit 30 for each blind separation while associating it with the arrival direction information based on the correlation with the steering vector. Based on the value of the azimuth, the azimuth cluster sorted into the number of groups corresponding to the maximum number of separation signals from the signal separation unit 30 is generated, and each mixed vector sent from the signal separation unit 30 immediately after the separation corresponds Correspond to the orientation cluster to be. The arrival direction information can include arrival angle information in the elevation direction in addition to arrival angle information in the azimuth direction. An example of a detailed configuration of the orientation clustering unit 40 is shown in FIG.

  FIG. 2 is a block diagram showing an example of the configuration of the orientation clustering unit 40 in FIG. 1 in more detail. As illustrated in FIG. 2, the orientation clustering unit 40 includes an orientation information correlation unit 41, an orientation information statistical processing unit 42, and an orientation cluster identification unit 43.

  The azimuth information correlation unit 41 performs correlation processing between each mixed vector and the steering vector, and associates each mixed vector with the arrival direction information. That is, the mixed vector is sent from the signal separation unit 30 as a result of the blind separation corresponding to a plurality of separated original signals each time. On the other hand, the steering vector represents a directivity pattern of the sensor unit 10 formed by an array of a plurality of sensors 11 to 1n, and includes a unique pattern that includes amplitude and phase information and is uniquely determined according to the sensor array. It is. Therefore, the azimuth information correlator 41 performs these correlation processes to replace each mixed vector corresponding to a plurality of original signals with the arrival azimuth information of each original signal.

  The azimuth information statistical processing unit 42 accumulates the arrival azimuth information and generates an azimuth cluster in which the accumulated arrival azimuth information is sorted into a number of groups corresponding to the maximum number of separated signals. The azimuth cluster is a group in which the arrival azimuth information is sorted into groups corresponding to the maximum number of separated signals according to the frequency of occurrence, and an example is shown in FIG. In the example shown in FIG. 3, the maximum value of the number of separated signals is 3, and the arrival direction information accumulated as a histogram is further statistically processed to generate three direction clusters A, B, and C. is doing. Since these azimuth clusters are generated based on respective mixed vectors corresponding to a plurality of original signals, it can be seen that each corresponds to the arrival azimuth of one of the original signals. Further, the ranges θA, θB, and θC of the arrival direction information of the generated direction clusters may overlap like θA and θB in FIG.

  The azimuth cluster identifying unit 43 identifies which of the azimuth clusters generated by the azimuth information statistical processing unit 42 corresponds to the arrival azimuth information from the azimuth information correlation unit 41. As a result, the connection destination determination unit 60 is notified of the corresponding orientation cluster and the orientation clustering result including the deviation from the average value of the arrival orientation information in the orientation cluster.

  The waveform clustering unit 50 targets the separated signal waveform that has been sent from time to time immediately before the signal separation unit 30 every time blind separation is performed, and further separates the separated signal waveform as no signal. The separated signal waveforms that are separated as signals that actually exist in that time period are grouped into waveform clusters based on the cross-correlation between these waveforms, and sequentially stored, and the signal separation unit 30 Each of the separated signal waveforms sent immediately after the separation is associated with the corresponding waveform cluster. An example of a detailed configuration of the waveform clustering unit 50 is shown in FIG.

  FIG. 4 is a block diagram showing an example of the configuration of the waveform clustering unit 50 in FIG. 1 in more detail. As illustrated in FIG. 4, the waveform clustering unit 50 includes a separated signal waveform storage unit 51 and a waveform cluster identification unit 52.

  The separated signal waveform storage unit 51 groups separated signal waveforms having a significant cross-correlation between separated signal waveforms of existing signals acquired by a plurality of past blind separations into one waveform cluster. The separated signal waveforms grouped for each waveform cluster are stored in time series. FIG. 5 is a conceptual diagram illustrating how the separated signal waveforms are stored in respective waveform clusters in time series. In the example shown in FIG. 5, three waveform clusters WA, WB, and WC are generated as a result of cross-correlation, and n separated signal waveforms up to the latest are stored in time series for each waveform cluster. ing. Since these waveform clusters are generated based on the separated waveforms of a plurality of original signals, it can be seen that each corresponds to one of the plurality of original signals.

  The waveform cluster identifying unit 52 determines which of the waveform clusters generated and stored in the separated signal waveform storage unit 51 each of the separated signal waveforms immediately after separation from the signal separating unit 30 corresponds to in each waveform cluster. Are identified by cross-correlation with the separated signal waveform. As a result, a waveform clustering result including the corresponding waveform cluster and the maximum value of the cross-correlation for each waveform cluster is notified to the connection destination determination unit 60.

  Based on the azimuth clustering result from the azimuth clustering unit 40 and the waveform clustering result from the waveform clustering unit 50, the connection destination determination unit 60 converts each of the separated signal waveforms immediately after being separated by the signal separation unit 30 into a continuous source waveform. As a connection destination for extraction as a signal, it is determined which of the separated signal waveforms is connected immediately before, and the determination result is sent to the original signal extraction unit 70. The original signal extraction unit 70 is, for example, a matrix-like switch that can control the setting of a signal path between input and output, and is separated immediately after being separated by the signal separation unit 30 based on the determination result from the connection destination determination unit 60. The signal waveforms are sequentially connected to extract the original signal.

  Next, the operation of the signal separation and extraction apparatus of the present embodiment configured as described above will be described with reference to the flowcharts of FIGS. 1 to 5 and FIG. FIG. 6 is a flowchart for explaining the operation of the signal separation and extraction apparatus according to the present invention.

  When a plurality of original signals are mixed and received by the sensors 11 to 1n of the sensor unit 10 (ST301), the received n signals are sent to the sampling unit 20, and after necessary signal processing is performed, Each is sampled at a predetermined interval, and sequentially sent to the signal separation unit 30 (ST302). The signal separation unit 30 blindly separates a plurality of original signals from these received signals by, for example, an independent component analysis method. In the separation process, for example, several tens to several thousand continuous within a predetermined period. A data block composed of a predetermined number of sample data is a unit to be separated. Therefore, when a predetermined period elapses and the sample data reaches a predetermined number (Y in ST303), a blind separation process is performed on this data block, and the separation signal waveform and mixing vector corresponding to the original signal, and the separation are separated. The number of signals is acquired (ST304).

  In order to extract the original signal, a plurality of separated signal waveforms separated for each data block are sequentially connected. In blind separation processing, the order relationship between the signal output after separation and the original signal is the data block. Not the same. For this reason, the connection destination of the separated signal waveform immediately after separation is determined based on the results of the following orientation clustering and waveform clustering.

  First, the operation of orientation clustering will be described. Each time mixed vectors are acquired by the signal separation unit 30, these mixed vectors are respectively associated with the arrival direction information by the correlation processing with the steering vector in the direction information correlation unit 41, and in the direction information statistical processing unit 42. For example, as illustrated in FIG. 3, the horizontal axis is accumulated as an azimuth value of arrival direction information, and the vertical axis is a histogram of the value. The accumulated arrival direction information is sorted into a number of groups corresponding to the maximum number of separated signals, and direction clusters A, B, and C are generated.

  For sorting, for example, the following statistical method can be applied. That is, a frequency maximum corresponding to the maximum value of the number of separated signals is searched on the histogram. In the case of FIG. 3, Pa, Pb, and Pc are extracted as maximum values. Next, the value of the minimum frequency between the maximum values is set as a temporary boundary, and after calculating the standard deviation σ for the histogram up to the temporary boundary on both sides of each maximum point, for example, ± 2σ around the maximum point Set the range of degree. In the example of FIG. 3, the set ranges are θA, θB, and θC. Then, the accumulated arrival direction information is sorted into these set ranges. Note that θA, θB, and θC that are the ranges of each orientation cluster are not necessarily discrete from each other, and may overlap, for example, θA and θB in FIG. In this case, the arrival direction information of the object is sorted into two direction clusters.

  In this way, when a direction cluster is generated from the mixture vectors separated immediately before, when a new mixture vector immediately after separation is sent from the signal separation unit 30, these mixture vectors are described above. In this way, after being associated with the arrival direction information by the direction information correlator 41, the direction cluster identification unit 43 corresponds to which direction cluster based on the range of the direction cluster generated by the direction information statistical processing unit 42 Identified. Then, the direction clustering result including the corresponding direction cluster and the deviation from the average value of the arrival direction information in the direction cluster is sent to the connection destination determination unit 60 (ST305).

  Next, the operation of waveform clustering will be described. Each time the separated signal waveform is acquired by the signal separating unit 30, the separated signal waveform unit 51 extracts only the separated signal waveform separated as a signal existing in the time zone from these separated signal waveforms. The separated signal waveform is temporarily held for a plurality of past times. Then, the cross-correlation between the retained separated signal waveforms is calculated, and the separated signal waveforms having a significant cross-correlation are grouped into one to generate a waveform cluster, as illustrated in FIG. In addition, these separated signal waveforms are stored in time series for each waveform cluster.

  In this way, when the separated signal waveforms separated immediately before including a plurality of past times are stored as waveform clusters, when a new separated signal waveform immediately after separation is sent from the signal processing unit 30, In addition to the processing in the separation signal waveform storage unit 51 as described above, the separation signal waveform of each of the separated signal waveforms is correlated with the separation signal waveform in each waveform cluster generated in the separation signal waveform storage unit 51 in the waveform cluster identification unit 52. Based on the correlation, which waveform cluster corresponds is identified. Then, the waveform clustering result including the corresponding waveform cluster and the maximum value of the cross-correlation for each waveform cluster is sent to the connection destination determination unit 60 (ST306).

  The connection destination determination unit 60 determines each connection destination of the separated signal waveform immediately after the separation from the signal separation unit 30 based on the orientation clustering result and the waveform clustering result in the above-described steps. Each of the azimuth cluster and the waveform cluster is considered to correspond to one of a plurality of original signals. Therefore, in determining, for each separated signal waveform immediately after separation, a cluster having a high degree of coincidence is selected in both the azimuth clustering result and the waveform clustering result, and the original signal corresponding to this selection result, that is, the separation separated immediately before is selected. The signal waveform is the connection destination. The determination result of the connection destination is sent to the original signal extraction unit 70 (S307).

  In the original signal extraction unit 70, based on the determination result from the connection destination determination unit 60, the signal path between the input and output is set in conjunction with the timing at which the separated signal waveform is transmitted from the signal separation unit 30. Then, a plurality of separated signal waveforms immediately after separation from the signal separation unit 30 applied to the input side are sequentially connected by being connected to the output side terminals that are the target connection destinations, and as continuous original signals Extracted (S308). Thereafter, the above-described steps S301 to S308 are repeated until an operation end is instructed (S309).

  As described above, in the present embodiment, when the signals received by mixing a plurality of original signals are subjected to blind separation and each of the original signals is extracted, the result of blind separation until just before connection is obtained by orientation clustering. In addition to grouping and accumulating in association with arrival direction information, the result of blind separation until just before connection is grouped and stored by cross-correlation between waveforms by waveform clustering. When the signals immediately after separation are sequentially connected, the connection destination is determined based on the two results of orientation clustering and waveform clustering, so that the original signal can be extracted continuously with good reliability. it can.

  In addition, the azimuth cluster generated by azimuth clustering and the waveform cluster generated by waveform clustering store and store past blind separation results corresponding to the elapsed time immediately before the connection. When the original signal is received with repeated disappearance and reproduction due to fluctuations in the signal, or when the original signal is an intermittent signal rather than a continuous signal, no significant correlation can be found with the previous blind separation result. Even in this case, it is possible to sequentially determine the connection destination based on the past separation results accumulated and stored in each cluster. Therefore, it is possible to continuously separate and extract the original signal with good reliability from a signal received by mixing a plurality of signals from different signal sources while adapting to changes in various reception environments.

(Modification 1)
In Modification 1, when determining the connection destination of the separated signal waveform immediately after separation, priority is given to the arrival direction information of the signal, that is, the direction clustering result. In the above-described first embodiment, when the connection destination determination unit 60 determines the connection destination of the separated signal waveform immediately after the separation, the orientation clustering result and the waveform clustering are performed for each separated signal waveform immediately after the separation as in step S307. Using both of the results, a signal having a high degree of coincidence is selected as the connection destination.

  On the other hand, in the first modification, when the mixed vector immediately after separation is associated with a single orientation cluster as a result of orientation clustering, the separation signal waveform immediately after separation is converted to this orientation cluster in step S307. It is determined to connect to the associated signal. As described above, each of the orientation clusters can be regarded as corresponding to the arrival direction of one of the original signals. In addition, although there are always fluctuations in the reception environment and the like, it is considered that the arrival direction of the signal hardly changes in a short time corresponding to, for example, the blind separation execution cycle.

  Therefore, it is considered that there is little influence on the azimuth cluster generated immediately before, and when each signal separated by blinds is associated with a single azimuth cluster without overlapping in its arrival azimuth, In step S307 described above, by determining the connection destination based on this result, the original signal can be separated and extracted while adapting to the disappearance / reappearance of the original signal, the intermittent signal, and the like. In addition, it is possible to obtain an effect of reducing the load of signal processing for determining the connection destination that is repeated every time the blind separation is performed.

(Modification 2)
In the second modification, when determining the connection destination of the separated signal waveform immediately after separation, the direction clustering result is determined as the first priority, and the waveform clustering result is determined as the second priority. In the first modification described above, the mixed vector immediately after separation is associated with a single orientation cluster. In addition, in the second modification, the mixed vector is not associated with a single orientation cluster. In step S307, based on the waveform clustering result, it is determined to connect to the signal corresponding to the waveform cluster having the strongest cross correlation.

  That is, in various receiving environments such as multiple original signals with similar arrival directions and original signals just after newly appearing, the mixed vector immediately after separation may not be associated with a single direction cluster. is there. On the other hand, the separated signal waveforms for a plurality of times until immediately before are stored in time series in the waveform cluster. Therefore, in such a reception environment, by using the waveform clustering result together with the azimuth clustering result as foresight information, the signal processing load for determining the connection destination is reduced, and the change of the reception environment is also achieved. While adapting, the original signal can be separated and extracted with good reliability.

  Note that the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

1 is a block diagram showing an embodiment of a signal separation and extraction device according to the present invention. The block diagram which shows an example of a structure of the direction clustering part 40 in detail. Explanatory drawing which shows an example of the formed orientation cluster. The block diagram which shows an example of a structure of the waveform clustering part 50 in detail. The conceptual diagram which illustrates a mode that the separated signal waveform was memorize | stored in the waveform cluster in time series. The flowchart for demonstrating operation | movement of the signal separation extraction apparatus which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Sensor part 20 Sampling part 30 Signal separation part 40 Direction clustering part 41 Direction information correlation part 42 Direction information statistical processing part 43 Direction cluster identification part 50 Waveform clustering part 51 Separation signal waveform storage part 52 Waveform cluster identification part 60 Connection destination determination part 70 Original signal extraction unit

Claims (3)

A plurality of sensors for receiving a mixed signal of a plurality of original signals;
Sampling means for sampling each of the signals received by the plurality of sensors at a predetermined interval to obtain sample data;
Signal separation means for performing blind separation on the sample data within this period every predetermined period, and continuously obtaining the separation signal waveform and the mixing vector and the number of separation signals corresponding to the original signal;
A plurality of mixed vectors acquired by the signal separating means over a predetermined period of time are accumulated while being associated with arrival direction information based on a correlation with a steering vector representing a directivity pattern formed by an array of the plurality of sensors, The accumulated arrival direction information is grouped into a number corresponding to the maximum value of the number of separated signals based on the value of the direction, and the orientation vector corresponding to the mixed vector immediately after separation by the signal separation unit is generated. Orientation clustering means for associating with a cluster;
A separated signal waveform corresponding to an existing signal is extracted from the separated signal waveforms acquired by the signal separating means for a plurality of predetermined periods until immediately before, and is grouped into waveform clusters based on the cross-correlation between the separated signal waveforms. A waveform clustering means for storing the divided signal waveforms immediately after separation by the signal separation means, and for correlating them with these waveform clusters;
Based on the results of the azimuth clustering means and the waveform clustering means, a determination means for determining which of the separated signal waveforms connected immediately before by the signal separation means to be connected to each of the separated signal waveforms. When,
An apparatus for extracting and separating signals, comprising: extracting means for sequentially connecting the separated signal waveforms to each other based on the determination result to extract the plurality of original signals.   In the case where the mixture vector is associated with a single azimuth cluster in the azimuth clustering unit, the determination unit associates the connection destination of the separated signal waveform immediately after the separation with the same azimuth cluster. The signal separation / extraction apparatus according to claim 1, wherein the separation signal waveform is until just before.   Further, the determining means determines the connection destination of the separated signal waveform immediately after the separation in the waveform clustering means when the mixed vector is not associated with a single orientation cluster in the orientation clustering means. 3. The signal separation / extraction apparatus according to claim 2, wherein the separation signal waveform is the waveform immediately before being associated with the waveform cluster having the strongest cross-correlation among the clusters.

Images