JP2731792B2 - Narrowband signal tracking method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention performs a frequency analysis on the output of a received beam directed in a plurality of directions, and obtains a directional characteristic based on the distribution of the input signal strength obtained in the analysis result in the direction.times.frequency space. The present invention relates to a narrow-band signal tracking method for measuring an unknown narrow-band signal with time continuously.

[0002]

2. Description of the Related Art In a conventional narrow-band signal tracking method, an input signal is received using a plurality of reception beams phased toward slightly different directions θ _i (i = 1, 2,...). . By performing a discrete Fourier transform on the output signal of each received beam, the signal intensity X (θ _i , f
_j ) get Next, a local maximum point is detected from the distribution of the signal intensity. Here, the maximum point refers to a point that has a signal strength that is larger than the azimuths on both sides and higher than the frequencies on both sides, such as a point (θ _i , f _j ) shown in the following equation (1). _{X (θ i, f j)} > X (θ i-1, f j), X (θ i + 1, f j), X (θ i, f j-1), X (θ i, f j + ₁ ) (1) If the point (θ _i0 , f _i0 ) is a maximum point and the signal strength X (θ _i0 , f _i0 ) exceeds a predetermined threshold, a narrow band Judge that the signal has arrived. This point (θ _i0 ,
f _i0 ) is called an event, θ _i0 is the direction of the event, f _i0
Is called the frequency of the event. The above processing is performed at predetermined time intervals T, and a time series of events is detected to be a track. The tracks are formed as follows.

If an existing track exists, an event having an azimuth whose difference from the azimuth of the track is equal to or less than a predetermined threshold and an event having a frequency whose difference from the frequency of the track is equal to or less than a predetermined threshold is generated. Extract at each time interval T and integrate the event into existing tracks. When integrated, the direction of the integrated event is newly set as the track direction, and the frequency of the event is set as the new track frequency. If there is no event that can be integrated in the track, the track is held as it is. This state in which there is no event that can be integrated is referred to as temporary search. If the temporary missing has continued for a predetermined number of times, it is considered that the detection of the signal has been lost, and the track is deleted. If there is an event that is not integrated in any track, a new track is formed based on the event. That is, the azimuth and frequency of the event are used as the initial values of the azimuth and frequency of the new track. Note that there is also a method of smoothing the azimuth of the event using a tracking filter or the like, and using this as the azimuth of the track. In this case, the track holds the smoothed direction and the direction change rate, and estimates the predicted direction at each time interval T. If there is an event having an azimuth whose difference from the predicted azimuth is equal to or smaller than a predetermined threshold, the event is integrated into the track. When merging, the smoothed direction and the direction change rate of the track are updated using the direction of the event as a new measured value. This can be realized by, for example, a Kalman filter. Similarly, there is a method in which the frequency of the event is smoothed and used as the frequency of the track.

[0004]

However, the conventional narrow band signal tracking method has the following problems. In some cases, a plurality of events that can be integrated are detected simultaneously for one existing track. Conversely, there is a case where one extracted event can be integrated into a plurality of tracks. These are called conflicts. FIG. 2 is a diagram for explaining a problem of a conventional narrow-band signal tracking method, and shows an example of a conflict. For simplicity, this figure shows the azimuth limited to a certain one position θ _i1 . A narrowband signal frequencies are becoming drop narrowband signal and frequency is rising is coming from around the azimuth theta _i1, its frequency at time t ₀ is assumed to intersect. Event E ₀ detected at time t ₀ includes track L 1 corresponding to the narrow-band signal whose frequency is increasing,
It can be integrated into both the track L2 corresponding to the narrow band signal whose frequency is decreasing. In such a case, it is necessary to integrate a track and an event in a one-to-one correspondence by some method. To achieve this, for example, there is a method of selecting a combination that minimizes the azimuth difference between the track and the event. However, when conflicts occur frequently, a large amount of calculation is required to select a combination. Therefore, a tracking method in which the occurrence frequency of the conflict is as low as possible is desired.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a narrow-band signal for continuously measuring the frequency of a narrow-band signal in an input signal and the arrival direction of the narrow-band signal. In the signal tracking method, the following reception processing, storage processing, pattern detection processing, event detection processing, and track selection processing are performed. In the receiving process, the receiving beam having directivity in a plurality of different directions monitors the direction in a continuous range, and the frequency analysis of the input signal is performed for each receiving beam. In the storage processing, the signal intensity distribution in the time × frequency space for each direction is obtained by accumulating the respective frequency analysis results over a certain period of time. In the pattern detection process, a rectangular portion limited to the vicinity of the sample frequency of each of the signal intensity distributions is input to a plurality of neural networks having different ranges of responsibility with respect to the gradient defined by the temporal change in frequency, and And when the narrow band signal is present in the vicinity of the sample frequency and there is a linear pattern that rises up in the rectangular portion, a large value is output to each of the plurality of neural networks. The event detection processing includes, among the output values of the plurality of neural networks obtained as a result of repeatedly performing the pattern detection processing while changing the sample frequency and the azimuth, a local maximum in a three-dimensional space of azimuth × frequency × gradient. Those that are points and exceed a predetermined threshold are detected as events indicating the arrival of the narrowband signal. In the track selection process, a track having an azimuth, a frequency, and a gradient close to the azimuth, frequency, and gradient corresponding to the newly detected event is detected from among the tracks indicating the temporal sequence of events detected in the past. By making the track of the narrow band signal, the frequency and the arrival direction of the narrow band signal are measured continuously over time.

[0006]

According to the present invention, since the narrow-band signal tracking method is configured as described above, the receiving process can be applied to the azimuth in a range continuous with the receiving beam having directivity for a plurality of different azimuths. Monitoring is performed, and frequency analysis of the input signal is performed for each receiving beam. By amnestics,
Each frequency analysis result is accumulated for a certain period of time, and a signal intensity distribution in a time × frequency space for each direction is obtained. In the pattern detection processing, a rectangular portion limited to the vicinity of the sample frequency of each of the signal intensity distributions is input to a plurality of neural networks having a range of responsibility with respect to the temporal gradient of the frequency. Then, when a narrow band signal exists in the vicinity of each sample range and a sample frequency, and there is a linear pattern that emerges in the rectangular portion,
A large value is output from each neural network. Among the output values of a plurality of neural networks obtained as a result of repeating the pattern detection processing by changing the sample frequency and the azimuth by the event detection processing, the maximum value in the azimuth × frequency × gradient three-dimensional space, and An event exceeding a predetermined threshold is detected as an event indicating the arrival of a narrowband signal. In the track selection process, a track having an azimuth, a frequency, and a gradient close to the azimuth, frequency, and gradient corresponding to the newly detected event is detected from the tracks of the events detected in the past. By setting the detected track as the track of the narrowband signal, the frequency and the arrival direction of the narrowband signal are measured continuously in time. Therefore, the above problem can be solved.

[0007]

FIG. 3 is a diagram for explaining the principle of the present invention.
Elements common to FIG. 2 for describing the conventional problem are denoted by common reference numerals. In the narrow-band signal tracking method, a large amount of calculation is required when conflicts occur frequently. Therefore, a tracking method in which the frequency of occurrence of conflicts is as low as possible is desired. The present embodiment is made to solve such a problem. When detecting an event, a frequency gradient of the event is obtained in addition to the direction and frequency of the event. The fact that the difference between the frequency gradient of the event and the frequency gradient of the track is equal to or smaller than a predetermined value is added to the integration condition. The frequency gradient is a rate of change in frequency with time, and is hereinafter simply referred to as a gradient. Also,
The magnitude of the gradient is defined by the rate of increase of the frequency per unit time (a negative value when the frequency decreases). As shown in FIG. 3, a narrow-band signal whose frequency is increasing and a narrow-band signal whose frequency is decreasing come from around the azimuth θ _i1 ,
It is assumed that the frequencies cross at time t ₀ . Those two
L1 and L2 respectively corresponding to the two narrowband signals
And each event has its own gradient. Track L1 to a positive slope, negative slope track L2 is, and the time t event E ₀ detected in ₀ has a positive slope. At this time, since the event E ₀ is not within the range in which the track L2 can be integrated (the integration window W in FIG. 3), the event E ₀ is integrated with the track L1. That is, no conflict occurs. FIG. 1 is a configuration block diagram of a narrow-band signal tracking device according to an embodiment of the present invention. The narrow-band signal tracking device includes an acoustic sensor array 1.
A plurality of acoustic sensors (not shown) in the acoustic sensor array 1 have directivity in slightly different directions.
The output side of the acoustic sensor array is connected to the phasing device 2, and the output side of the phasing device 2 is connected to the frequency analysis device 3. The output side of the frequency analysis device 3 is connected to the storage device 4. The storage device 4 is connected to a plurality of neural networks 5-1 to 5-K (K is an arbitrary number of 2 or more). The output side of each of the neural networks 5-1 to 5-K is commonly connected to the event detection device 6, and the output side of the event detection device 6 is connected to the tracking device 7. The track information file 8 is also connected to the tracking device 7.

Next, a narrow-band signal tracking method for tracking a narrow-band signal with the narrow-band signal tracking apparatus of FIG. 1 will be described. (1) Reception processing The acoustic sensor array 1 receives sound waves of an input signal by acoustic sensors directed to a plurality of different directions θ _i , and the phasing device 2 generates a plurality of reception beams S corresponding to a plurality of directions θ _i. form _i . The frequency analyzer 3 performs a frequency analysis on each received beam. That is, discrete Fourier transform and square detection are performed at predetermined time intervals T. As a result, a signal intensity X (θ _i , f _j ) for each direction θ _i and each frequency f _j is obtained. (2) Storage Processing The storage device 4 receives the signal intensity X (θ _i , f _j ) and stores the signal intensity X (θ _i , f _j ) for a fixed time in the past (that is, for N times of frequency analysis). As a result, the distribution X (θ _i , f _j , t _n ) of the signal intensity in the time × frequency space is obtained for each direction θ _i . (3) Pattern detection processing First, from the storage device 4, for example, from the signal intensity distribution in the time × frequency space in the first azimuth θ _i1 , the vicinity of the sample frequency f _j1 (a number J is determined in advance and f _{j1 − J to} f
A rectangular portion limited to the range of _{j1 + J} ) is extracted and input to the neural networks 5-1 to 5-K. That is, ｛X (θ _i , f _j , t _n ) | j = j1-J, ..., j1
+ J; n = 1, 2, 3,... N} are input to the neural networks 5-1 to 5-k.

Each of the neural networks 5-1 to 5-K
Is previously learned so as to output a large value if there is a pattern on a straight line during input. That is, by using an error back propagation (error back propagation) algorithm, an ideal output of 1 is output for learning data in which the linear pattern exists, and random learning data in which the pattern does not exist is output. Outputs an ideal output of 0. This error back-propagation algorithm is described in the following literature and the like. Document; Author Hideki Aso "Neural Network Information Processing", first edition (1988-6-20) Sangyo Tosho, in receive beam corresponding to P.50-54 orientation theta _i1, frequency frequency f _j1
When there is a narrow band signal in the vicinity, a linear pattern emerges in the extracted rectangular portion. The neural networks 5-1 to 5-K share signals having different gradients, respectively. For example, K gradients df ₁ ,..., Df _K (df _k = df _k−1 + δ) are determined for every predetermined width δ, and each neural network 5-k (1 ≦ k ≦ K)
Has been learned to respond to a pattern with a signal having a gradient in the section [df _k −δ / 2, df _k + δ / 2]. Each neural network 5-1 to 5-
K inputs a rectangular portion of the signal intensity distribution input from the storage device 4 and outputs a large value with 1 as an ideal output when a linear pattern exists and its gradient is within the assigned range. I do. The above processing is performed for frequencies other than the frequency f _j1 , and the same processing is performed for the second and subsequent directions θ.
Perform for all _i .

[0010] (3) of the signal intensity distribution of the event detection process orientation theta _i, the output value when the input rectangular portion in the vicinity of the frequency f _j in the neural network 5-k (1 ≦ k ≦ K) Y (θ _i , f _j , df _k ). The event detection device 6 calculates Y (θ _i , f _j , df _k )
Is detected, and the maximum point is defined as an event. For example, if Y (θ _i0 , f _j0 , df _k0 ) exceeds a predetermined threshold value and is a local maximum point in a distribution in a three-dimensional space of azimuth × frequency × gradient, and a narrowband signal arrives , And that point is event E. Then, the event detection device 6 calculates the azimuth θ _i0 , frequency f _j0, and gradient df of the event.
Output _k0 . (4) Track Selection Process The tracking device 7 receives the information of the event E from the event detection device 6 and forms a track that is a time series of events. When searching for an event to be integrated into an existing track, a track of a past event having a gradient in which the azimuth difference and the frequency difference are equal to or less than a predetermined threshold value and the difference from the track gradient is equal to or less than a predetermined threshold value. Look for and integrate. At this time, if the frequency of the existing track is φ and the frequency of the event E is f, and the event E is detected after T ₁ second has elapsed from the previous integration of the track, the following (2) The gradient df of the track is obtained by the equation (3). df = (f−φ) / T ₁ (2) The frequency and direction of the event E are stored in the track information file 8 together with the obtained gradient df as a new frequency and direction of the track. When two or more events E ₁ and E ₂ can be integrated into one track, the event with the larger output Y of the neural network is integrated with priority. When two or more tracks select one event E, the track having the closest azimuth difference from the event E is integrated with priority, and the other tracks are temporarily lost. If there is an event E that cannot be integrated with any track, a new track is formed, and the azimuth, frequency, and gradient of the event are recorded in the track information file 8 as initial values of the azimuth, frequency, and gradient of the new track. Store. By repeating these processes, it is possible to track and measure a narrow band signal whose frequency and arrival direction change every moment.

As described above, in the present embodiment, the reception processing,
Since a storage process, a pattern detection process, an event detection process, and a track selection process are performed to track an incoming narrowband signal, a temporal gradient of a frequency is added to an integration condition for an existing track. Therefore, occurrence of a conflict is suppressed. The present invention is not limited to the above-described embodiment, and various modifications are possible. For example, each block in the narrowband signal tracking device shown in FIG. 1 may be configured by an individual electric circuit, or a part or all of them may be realized by a computer program. In the above embodiment, the input signal is a sound wave, but the input signal is not limited to a sound wave. For example, the present invention is applicable even when the input signal is a radio wave. However, in this case, a receiving antenna is used instead of the acoustic sensor array 1.

[0012]

As described above in detail, according to the present invention, reception processing, storage processing, pattern detection processing, event detection processing, and track selection processing are performed, and frequency time The target gradient is added. Therefore, the frequency of occurrence of a conflict in tracking the incoming narrowband signal can be suppressed as compared with the related art.

[Brief description of the drawings]

FIG. 1 is a configuration block diagram of a narrowband signal tracking device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a problem of a conventional narrow-band signal tracking method.

FIG. 3 is a diagram illustrating the principle of the present invention.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 acoustic sensor array 2 phasing device 3 frequency analysis device 4 storage device 5-1 to 5-K neural network 6 event detection device 7 tracking device 8 track information file Si received beam X (θ _i , f _j ) signal strength

 ──────────────────────────────────────────────────続 き Continuation of front page (72) Inventor Toru Mizota 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-9-43330 (JP, A) JP-A-9-15032 (JP, A) JP-A-8-313337 (JP, A) JP-A-4-216162 (JP, A) JP-A-4-216161 (JP, A)

Claims (1)

(57) [Claims] 1. A narrow-band signal tracking method for measuring a frequency of a narrow-band signal in an input signal and an arrival direction of the narrow-band signal continuously in time, wherein directivity is provided for a plurality of different directions. A receiving process for monitoring the direction of a continuous range with the received beam having the same, and performing a frequency analysis of the input signal for each received beam, and accumulating the respective frequency analysis results for a certain period of time to obtain the respective direction. A storage process for obtaining a signal intensity distribution on each time × frequency space; and a plurality of neural networks having different ranges of responsibility with respect to a gradient defined by a temporal change in frequency. Is input, and when the narrow-band signal exists in the range of each of the areas and near the sample frequency, the rectangular part floats in the rectangular part. When there is a linear pattern that rises, a pattern detection process for outputting a large value to each of the plurality of neural networks, and the pattern detection process obtained by repeatedly performing the pattern detection process while changing the sample frequency and the azimuth. Among the output values of a plurality of neural networks, azimuth × frequency ×
An event detection process for detecting a maximum point on the three-dimensional space of the gradient and exceeding a predetermined threshold value as an event indicating the arrival of the narrowband signal; and a process for detecting a time series of events detected in the past. Of these, by detecting a track having an azimuth, frequency, and gradient close to the azimuth, frequency, and gradient corresponding to the newly detected event, and by detecting the track of the narrowband signal, the frequency of the narrowband signal and A narrow band signal tracking method, comprising: performing track selection processing for continuously measuring an arrival direction with time.