JP5117012B2 - Direction detection system and signal extraction method - Google Patents

Description

  The present invention relates to a direction detection system and a signal extraction method for separating a signal included in an observation signal, obtaining an arrival direction of the separated signal, and extracting a desired signal.

  In the conventional direction detection system, the blind signal separation method based on the ICA (Independent Component Analysis) method is used, and the observation signals obtained from a plurality of sensors are independent components at the sampling period. The number of signals is estimated by analyzing the signal, the estimated number of signals are separated from the observed signals, and the separated signals obtained at the sampling period are connected.

  However, the blind signal separation method based on the ICA method has a problem that the order of the separated signals obtained for each separation process is indefinite, and it is difficult to connect the separated signals. As a solution to this problem, a method has been reported in which a correlation between separated waveforms is determined, and a desired signal is continuously extracted from the determination result (see, for example, Patent Document 1). It was necessary to stabilize the operation.

In addition, in the blind signal separation method, an error is likely to occur in the signal number estimation, so that only a waveform of noise is output, or an output with incomplete separation is treated as one independent signal. was there. Furthermore, since the arrival direction of the signal cannot be obtained only by the separation processing, the arrival direction of each separated signal must be detected individually, and there is a problem that the configuration of the entire system becomes complicated.
JP 2006-178314 A

  As described above, in the conventional direction detection system, the problem of unstable operation and determination error when analyzing and separating the components periodically separated from the observed signal and pointing them together has been pointed out. Since it is necessary to detect a separate direction of arrival for each separated signal, the complexity of the system is regarded as a problem.

  The present invention has been made paying attention to the above circumstances, and its object is to extract the separated signals with high accuracy while obtaining the arrival directions of the signals periodically separated from the observation signals. It is an object of the present invention to provide a direction detection system and a signal extraction method capable of performing the above.

In order to achieve the above object, a direction detection system according to the present invention includes a plurality of sensors that observe incoming signals independently from each other, sampling means that periodically samples the observation signals of the plurality of sensors, Separating means for separating a sampled signal based on a blind signal separation method and outputting a separation waveform and a separation vector corresponding to the number of sensors, and signals from the plurality of sensors based on the output separation vectors Each of the mixing vectors representing the mixing characteristics of each of the above, and determining the arrival directions of the separated waveforms based on the azimuth characteristics of each mixing vector, and the arrival directions determined based on the azimuth characteristics of the mixing vectors overlap, and azimuth calculation means for calculating each arrival direction of the isolation waveform based on the orientation characteristics of the separation vector, before Based on the arrival direction obtained characterized by comprising a connecting means for connecting the separation waveform output for each of the periods.

The signal extraction method according to the present invention is a signal extraction method used in a direction detection system for detecting a desired signal and an arrival direction from signals observed independently from each other by a plurality of sensors. Each of the observed signals is periodically sampled, the sampled signal is separated based on a blind signal separation method, a separation waveform and a separation vector corresponding to the number of sensors are output, and the output separation vector is Based on the azimuth characteristics of the mixed vectors , the mixing vectors representing the mixing characteristics of the signals from the plurality of sensors are obtained, the arrival directions of the separated waveforms are obtained based on the azimuth characteristics of the mixed vectors, respectively. If the arrival direction has overlap, respectively arrival direction of the isolation waveform based on the orientation characteristics of the separation vector Because, characterized by connecting a separate waveform output for each of the period based on the arrival direction obtained.

  In the above configuration, the arrival directions of the separation waveforms are obtained for each separation result, the separation waveforms are connected based on the obtained arrival directions, and output as continuous signals. That is, it is possible to accurately extract continuous signals by connecting signals in the same direction.

  Therefore, according to the present invention, it is possible to provide a direction detection system and a signal extraction method capable of accurately connecting and extracting separated signals while obtaining arrival directions of signals periodically separated from observation signals. Can do.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a block diagram showing an embodiment of a direction finding system according to the present invention.
The direction detection system includes n sensors 1-1 to 1-n, a sampling unit 10, a blind signal separation unit 20, an azimuth calculation unit 30, a connection processing unit 40, a terminal selection switch unit 50, And a separated signal output unit 60.

  The sensors 1-1 to 1-n are arranged in an array and observe incoming signals independently of each other. A plurality of signals (for example, S1 to S1 in the figure) are simultaneously received by the sensors 1-1 to 1-n. S3) is received. The observation signals received by the sensors 1-1 to 1-n are mixed signals including signals S1 to S3, for example, and are supplied to the sampling unit 10, respectively. The sampling unit 10 divides the observation signals input from the sensors 1-1 to 1-n by the sampling period, and samples each continuous sampling period. The sampled observation signals of the sensors 1-1 to 1-n are supplied to the blind signal separation unit 20.

  The blind signal separation unit 20 analyzes each observation signal sampled by the sampling unit 10 using an algorithm such as ICA, and separates signals up to the same number (n) as the number of sensors from each observation signal. A separation vector is obtained and the waveform of the separated signal is obtained by multiplying the waveform of the observation signal by this vector. The n separation signals (separation vector and waveform data) obtained here are supplied to the azimuth calculation unit 30 and also to the terminal selection switch unit 50.

  The azimuth calculation unit 30 receives the separation vector and waveform data of each of the n separation signals, and obtains the orientation of each separation signal (waveform data) based on the separation vector. The direction data of the separation signal is supplied to the connection processing unit 40 together with the waveform data.

Based on the azimuth data obtained by the azimuth calculation unit 30, the connection processing unit 40 selects the waveform data of the separation signal having the same direction for each sampling period, and the separation signal having the same direction is continuous. Switching control is performed between input and output terminals of the terminal selection switch unit 50. The n separated signals output from the terminal selection switch unit 50 are supplied to the separated signal output unit 60.
The separated signal output unit 60 shapes the separated signal of the designated orientation from the n separated signals obtained by the terminal selection switch unit 50, and outputs it as a continuous signal.

In the direction detection system having the above-described configuration, a processing operation that is a feature thereof will be described below.
Now, ignoring the presence of noise, if the input signal is X and the mixing matrix representing the mixing characteristics by the arrangement of the sensors 1-1 to 1-n is A, the signal Y obtained as the sensor output is
Y = A ・ X
It is represented by

In contrast, in the blind signal separation method based on the ICA method, a separation matrix V corresponding to the inverse matrix of A is obtained,
V · Y = A ⁻¹ · Y = A ⁻¹ (A · X) = X
X (individual waveform before mixing) is obtained from the relational expression.

  The problem here is the dimension of A, which corresponds to the number of incoming signals. Usually, the estimation of the number of signals is performed by performing eigenvalue analysis and the distribution of the eigenvalues. However, there are many errors in the actual signal propagation environment. If it is incorrect, the dimension of A is estimated to be more or less than the actual dimension. If it is large, noise components other than the signal are output, and if it is small, the desired signal is lost. In general, in the signal number estimation, a large number of signals are often selected so that a desired signal is not lost. In addition, since the separation waveform output from the blind signal separation unit 20 has an indefinite order for each separation process, it is necessary to accurately search and connect consecutive signals.

  Therefore, in the present embodiment, the azimuth calculation unit 30 obtains the azimuths of the separated waveforms, and based on this azimuth information, the connection processing unit 40 accurately determines consecutive signals while eliminating excess noise components, Based on the determination result, the terminal selection switch unit 50 connects continuous signals. In this way, even when the number of signals is estimated to be large, the desired signal and the noise component are accurately discriminated, and separated signals having continuity are accurately discriminated and connected. Below, each process is demonstrated concretely.

FIG. 2 is a flowchart showing the processing procedure of the azimuth calculation unit 30 and its contents.
In FIG. 2, the azimuth calculation unit 30 first generates a mixture vector from the separation vector output from the blind signal separation unit 20 (step S2a). This process is a process of reproducing A as an inverse matrix of the separation matrix obtained as V, and uses A as a mixed vector. Next, the separation vector is multiplied by the steering vector to obtain a multiplication waveform representing the azimuth characteristic of the separation vector (step S2b). The steering vector indicates response characteristics of the sensor with respect to the target arrival direction range and elevation angle range. FIG. 3 shows an example of a multiplication waveform of the separation vector and the steering vector.

  Next, a peak is searched from this multiplication waveform (step S2c), and the peak value is compared with a preset threshold value (step S2d). FIG. 3 shows an example of the threshold value. This threshold is determined in the actual signal environment. In this comparison, if all the peak values for all the separation vectors V exceed the threshold value, the process proceeds to step S2e, where the mixture vector A and the steering vector are multiplied, and a multiplication waveform representing the azimuth characteristic of the mixture vector is obtained. The peak of this multiplication waveform is detected. FIG. 4 shows an example of a multiplication waveform of the mixed vector and the steering vector.

  Finally, the azimuth calculation unit 30 adopts the azimuth corresponding to the detected peak as the arrival azimuth of the signal (step S2f). For example, as shown in FIG. 5, when all peak values in the azimuth characteristic of the separation vector exceed a threshold value, the peak obtained from the azimuth characteristic of the mixed vector is adopted as the arrival azimuth of the signal. In the case of the figure, the arrival directions of the signals are estimated to be 200 degrees, 100 degrees, and 300 degrees, respectively.

  On the other hand, in the comparison in step S2d, when there is a peak value of the multiplication waveform of all the separated vectors and the steering vector that is less than the threshold value, the process proceeds to step S2g and first the multiplication of the mixed vector and the steering vector. Detect peaks from the waveform. And it is determined whether the direction corresponding to the detected peak overlaps (step S2h). In this determination, as shown in FIG. 6, when there is no azimuth overlap in the azimuth characteristics of the mixed vector, the process proceeds to step S2f, and the azimuth corresponding to the peak of the multiplication waveform of the mixed vector is used as the arrival direction of the signal. adopt. In the case of FIG. 6, the arrival directions of the signals are estimated as 200 degrees, 300 degrees, and 100 degrees, respectively.

  Further, in the determination of step S2h, as shown in FIG. 7, when the orientations obtained from the azimuth characteristics of the mixed vector overlap, the process proceeds to step S2i, where the peaks of the azimuth characteristics of the separation vectors overlap. Determine whether or not. In FIG. 7, the azimuth characteristics of the mixed vector overlap the azimuths of the second and third signals at 100 degrees, but according to the azimuth characteristics of the separation vector, the second signal detects the peak at the azimuth of 200 degrees. it can. For this reason, the azimuth of the second signal is 200 degrees, and the third signal is azimuth 100 degrees (step S2j).

  Furthermore, in the determination of step S2i, when the directions overlap in the multiplication waveform of the separation vector and the steering vector, in step S2k, the time axis waveform corresponding to the separation waveform is converted into a frequency spectrum, and the dispersion of the frequency spectrum is obtained. Compare. A larger dispersion value represents a true signal, and a smaller dispersion value is regarded as a noise component. For example, as shown in FIG. 8, when it is difficult to obtain the azimuth from the azimuth characteristics of the separation vector, the third signal becomes unnecessary by obtaining the variance of the spectrum of the separation waveform (a larger number is estimated by estimating the number of signals). Signal). In addition, it is effective to use a method for determining a correlation between time-axis waveforms for a signal whose azimuth is very close.

  The azimuth calculation unit 30 executes the above process for each separation process, and outputs azimuth information representing the obtained arrival azimuth of each signal to the connection processing unit 40 (step S2l). The connection processing unit 40 generates a switching signal for connecting signals having the same direction based on the direction information and sends the switching signal to the terminal selection switch unit 50. In accordance with this switching signal, the connection between the input / output terminals of the terminal selection switch unit 50 is switched, the separation signal for each separation processing is connected, and the separation signal output unit 60 outputs the continuous signal.

  As described above, in the above-described embodiment, the blind signal separation unit 20 separates up to the same number of incident signals as the number of sensors based on the input signal from the sampling unit 10 using an algorithm such as independent component analysis. Extract. The azimuth calculation unit 30 obtains a mixture vector representing the mixing characteristics of the signals from the sensor based on the separation vector output from the blind signal separation unit 20, and multiplies the mixture vector and the steering vector indicating the response characteristics of the sensor. Based on the waveform, the arrival direction of the separated signal is obtained. The connection processing unit 40 performs switching control between the input / output terminals of the terminal selection switch unit 50 so as to connect signals of the same direction based on this direction information. As a result, the separation signal for each separation process is connected in the terminal selection switch unit 50 and is output from the separation signal output unit 60 as a continuous signal.

  Therefore, according to the above-described embodiment, the direction of the separation waveform is obtained for each separation result, and continuous signals are accurately connected based on this direction information. Furthermore, by obtaining the variance of the frequency spectrum of the separated waveform, even when the number of signals is estimated to be large, the desired signal and the noise component are accurately discriminated, and the separated signals having continuity are accurately discriminated and connected. Is possible.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, 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.

The block block diagram which shows one Embodiment of the direction detection system concerning this invention. The flowchart which shows the process sequence of the azimuth | direction calculating part shown in FIG. 1, and its processing content. The figure which shows the multiplication waveform of an isolation | separation vector and a steering vector. The figure which shows the multiplication waveform of a mixing vector and a steering vector. The figure explaining the direction calculation processing method. The figure explaining the direction calculation processing method. The figure explaining the direction calculation processing method. The figure explaining the direction calculation processing method.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1-1-1-n ... Sensor, 10 ... Sampling part, 20 ... Blind signal separation part, 30 ... Direction calculating part, 40 ... Connection process part, 50 ... Terminal selection switch part, 60 ... Separation signal output part.

Claims (4)

Multiple sensors that observe incoming signals independently of each other;
Sampling means for periodically sampling the observation signals of the plurality of sensors;
Separating means for separating the sampled signal based on a blind signal separation method, and outputting a separation waveform and a separation vector corresponding to the number of sensors;
Based on the output separation vector, each of the mixing vectors representing the mixing characteristics of the signals from the plurality of sensors is obtained, and the arrival direction of the separation waveform is obtained based on the azimuth characteristics of each mixing vector. When the arrival directions obtained based on the azimuth characteristics overlap, the azimuth calculation means for obtaining the arrival directions of the separated waveforms based on the azimuth characteristics of the separation vector,
A direction detecting system comprising: a connecting means for connecting separated waveforms output for each period based on the obtained arrival direction. Multiple sensors that observe incoming signals independently of each other;
Sampling means for periodically sampling the observation signals of the plurality of sensors;
Separating means for separating the sampled signal based on a blind signal separation method, and outputting a separation waveform and a separation vector corresponding to the number of sensors;
Based on the output separation vector, each of the mixing vectors representing the mixing characteristics of the signals from the plurality of sensors is obtained, and the arrival direction of the separation waveform is obtained based on the azimuth characteristics of each mixing vector. When the arrival directions determined based on the azimuth characteristics overlap, the time-axis waveform of the separated waveform is converted into a frequency spectrum, and the variance of the frequency spectrum is calculated. Based on the calculated variance value, Azimuth calculating means for obtaining the arrival directions of the separated waveforms,
A direction detecting system comprising: a connecting means for connecting separated waveforms output for each period based on the obtained arrival direction. A signal extraction method used in a direction detection system that detects a desired signal and an arrival direction from signals observed independently by a plurality of sensors,
Periodically sampling the observation signals of the plurality of sensors,
Separating the sampled signal based on a blind signal separation method, and outputting a separation waveform and a separation vector corresponding to the number of sensors;
Based on the output separation vector, each of the mixing vectors representing the mixing characteristics of the signals from the plurality of sensors is obtained, and the arrival direction of the separation waveform is obtained based on the azimuth characteristics of each mixing vector. When the arrival directions determined based on the azimuth characteristics overlap, the respective arrival directions of the separated waveforms are determined based on the directional characteristics of the separation vector,
A signal extraction method comprising: connecting separated waveforms output for each of the periods based on the obtained arrival direction. A signal extraction method used in a direction detection system that detects a desired signal and an arrival direction from signals observed independently by a plurality of sensors,
Periodically sampling the observation signals of the plurality of sensors,
Separating the sampled signal based on a blind signal separation method, and outputting a separation waveform and a separation vector corresponding to the number of sensors;
Based on the output separation vector, each of the mixing vectors representing the mixing characteristics of the signals from the plurality of sensors is obtained, and the arrival direction of the separation waveform is obtained based on the azimuth characteristics of each mixing vector. When the arrival directions determined based on the azimuth characteristics overlap, the time-axis waveform of the separated waveform is converted into a frequency spectrum, and the variance of the frequency spectrum is calculated. Based on the calculated variance value, Find the arrival directions of the separated waveforms,
A signal extraction method comprising: connecting separated waveforms output for each of the periods based on the obtained arrival direction.

Images