JP2763821B2 - Sound source feature extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial application field) The present invention relates to a feature amount of a sound source signal from a sound source arbitrarily present in space in signal processing of a sonar for performing position measurement (acoustic positioning) in water and the like. The present invention relates to a sound source feature extraction method for extracting (amplitude, frequency, direction).

(Prior art) Conventional techniques in such a field include, for example, Oki Electric Research and Development, 53 [4] (Showa 61-10), by Nitori Igarashi, "Digital Signal Processing Technology in Underwater Sound," p.53- There was one described in 58.

As described in the above document, the sonar performs search, position measurement, and classification of a target moving in a three-dimensional space (underwater) using underwater sound. Sonars are classified into passive sonars and active sonars according to the operation mode. The passive sonar uses sound waves emitted by the target, and the active sonar applies sound waves toward the target, and uses the reflected waves (echoes) to search for, measure, and classify the target. The signal processing used in sonar is as follows:
Temporal processing used to extract the temporal characteristics (waveform, spectrum, etc.) of the signal is divided into spatial processing used to extract the spatial characteristics (position, shape, moving speed, etc.) of the signal.

Of the temporal processing, the matched filter and the Wiener filter obtain the maximum signal-to-noise ratio (S / N ratio) when a signal having a defined waveform and spectrum is buried in noise having a known spectrum. Filter. Spectral estimation estimates the strength of a signal with respect to frequency, and is used to detect a periodic signal (line spectrum) buried in noise and classify targets that emit the signal.

Among the spatial processing, beamforming (BF) uses the output signals of a large number of receivers that constitute a receiving array, and uses the directionality of the wave propagating in space to improve the S / N of the signal, The estimation of the incident direction and the intensity (spatial spectrum) is performed. Delay and phase estimation is a problem of estimating a delay or phase difference occurring between signals received by a small number of receivers, and is mainly used for position measurement of a target. This is thought to be a simplification of beamforming.

Beamforming is basically a delay-addition BF in which a delay for compensating for the difference in propagation delay is added to output signals of a large number of receivers and then added, but when the signal has a narrow band, Can use phase compensation instead of delay compensation. Spatial resolution is particularly important as a beam forming characteristic, and it is desired that the main beam width be narrow and the side lobe level be low. As a method of controlling the beam pattern, shading is used in which the output signal of each receiver is multiplied by a predetermined weight, which is effective mainly for suppressing the side lobe level. Next, a side lobe canceling technology was developed to remove the interfering wave incident from a specific direction.
Eventually, adaptive beamforming (ABF) was devised to remove interference in arbitrary directions. More recently, a signal subspace method that dramatically improves the resolution of the main beam by applying the latest spectrum estimation method to azimuth estimation has been studied.

(Problems to be Solved by the Invention) However, in any of the above sound source feature extraction methods, in order to obtain high spatial resolution, the number of receivers must be increased in order to increase the number of input data. . Moreover, in order to obtain high resolution, the receiver must be arranged in a state suitable for a sound source signal such as a linear shape or a circumferential shape. However, a window function is applied depending on the state of the receiver. There is a problem that the resolution is degraded, and there is a limit in setting an accurate arrangement state. for that reason,
Not only does the calculation become complicated due to the increase in the number of receivers, but there is a problem that the size of the apparatus is increased and the arrangement of the receivers is difficult, which has been difficult to solve.

The present invention has a problem that the prior art has had to increase the number of receivers in order to obtain high spatial resolution, thereby simplifying the calculation, simplifying the device structure,
It is another object of the present invention to provide a sound source feature extraction method that solves the problem that miniaturization is difficult.

(Means for Solving the Problems) In order to solve the above problems, a first invention receives a sound source signal from a sound source by a plurality of receivers, and receives the received signal based on a sampling theorem of a time / space signal. In a sound source feature extraction method of sampling and obtaining a discrete time series signal and extracting a feature amount of the sound source signal from the discrete time series signal, based on the discrete time series signal, a feature amount of the sound source signal is subjected to nonlinear coupling. A mathematical expression is formed using parameters, and the characteristic amount of the sound source signal is directly calculated according to the mathematical expression.

In a second aspect based on the first aspect, the parameters of the non-linear combination are calculated using a least squares linear Taylor differential correction method.

In a third aspect based on the first aspect, the receiver comprises:
The arrangement is such that the sampling theorem is satisfied.

(Operation) According to the first aspect, since the sound source feature extraction method is configured as follows, a sound source signal coming from a sound source is received by a plurality of receivers and is sampled at a predetermined frequency based on a sampling theorem. It is converted to a discrete time series signal. Then, based on the discrete time-series signal, a mathematical model (formulaization) of the sound source signal is constructed using the characteristic amount of the sound source signal. Then, according to the constructed mathematical expression, the characteristic amount of the sound source signal is directly calculated by calculation. As a result, feature extraction with high spatial resolution can be performed in a short analysis time without increasing the number of receivers.

In the second invention, since the parameters of the non-linear combination at the time of the mathematical expression are calculated using the least squares linear Taylor differential correction method, the mathematical expression with a small amount of calculation can be set accurately.

In the third aspect, the receiver is arranged so as to satisfy the sampling theorem, so that the degree of freedom of the spatial arrangement position of the receiver is increased.

 Therefore, the above problem can be solved.

Embodiment FIG. 1 is a functional block diagram of a sound source feature extraction device using a sound source feature extraction method according to an embodiment of the present invention.

This sound source feature extraction apparatus is capable of extracting feature amounts (amplitude, frequency, direction) of a sound source signal S1 from, for example, D sound sources 1 in water.
It has a signal receiving unit 2 for receiving a sound source signal S1 from a sound source 1, and a non-linear feature amount estimating unit 10 is connected to an output side thereof. The signal receiving section 2 is a sound source signal
It has a plurality of receivers for converting S1 into an electric signal, and a function of sampling the output of the receiver at a predetermined sampling frequency based on the sampling theorem and outputting a discrete time series signal s (n, i). . Here, n in s (n, i) is the number of the sampled time-series data, and i is the number of the receiver.

The nonlinear feature amount estimating unit 10 calculates the discrete time series signal s (n, i)
Has a function of estimating the feature amount of the sound source signal S1,
Estimated signal generator 11 for calculating an estimated amount of a received signal received by the receiver using a non-linear combination function of the estimated feature amount of sound source 1
The output side is connected to an estimation error calculation unit 12 that calculates an estimation error of the estimated value with respect to the received signal. On the output side of the estimation error calculation unit 12, a normal equation determination unit
13, a correction amount calculation unit 14 and an estimation error determination unit 15 are connected. Further, an output side of the estimation error determination unit 15 is connected to an estimation signal generation unit 11 via an estimation feature amount update unit 16.

The normal equation determination unit 13 has a function of determining a normal equation based on the least squares linear Taylor differential correction from the estimation error calculated by the estimation error calculation unit 12 and the estimated feature amount.
The correction amount calculation unit 14 has a function of calculating a correction amount for the estimated feature amount from the normal equation determined by the normal equation determination unit 13. The estimation error determination unit 15 includes an estimation error calculation unit
Using the estimation error obtained in step 12, it is determined whether or not the estimation error criterion for the estimation error is satisfied. If the estimation error criterion is not satisfied, the estimated feature updating unit 16 is controlled to update the estimated feature. That is, in the estimated feature amount updating unit 16, a result obtained by adding the correction amount obtained by the correction amount calculating unit 14 to the estimated feature amount is input to the estimated signal generating unit 11 as a new estimated feature amount,
It has a function of repeating the updating process until the estimation error satisfies the estimation error criterion. When the estimation error calculated by the estimation error calculation unit 12 satisfies the estimation error criterion, the estimation feature amount is used as the final estimation feature amount of the sound source 1.
Output from 15.

Next, a sound source feature extraction method according to the present embodiment using the above-described sound source feature extraction device will be described.

First, it is assumed that there are D sound sources 1 and that each sound source 1 is received by the receiver as a plane wave, and the feature vector P of the sound source 1 to be extracted is expressed by the following equation (1). Amplitude a _j , angular frequency ω _j , and azimuth θ _j (j = 1,
2, ..., D). Further, as shown in the following equation (2), the estimated feature amount vector for the feature amount vector P is represented by _j , the estimated amplitude of each sound source 1, and the estimated angular frequency by Estimated bearing And

P = (p ₁ , p ₂ , ..., p _3D ) = (a ₁ , a ₂ , ..., a _D , ω ₁ , ω ₂ , ..., ω _D , θ ₁ , θ ₂ ,…, θ _D )… … (1) The number of receivers of the signal receiving unit 2 is M, and the position vector of each receiver is r _i (i = 1, 2,..., M). The signal received by the i-th receiver is converted into a sampling frequency in the signal receiving unit 2.
The sampled discrete time-series signal is sampled at f _s , and the sampled discrete time series signal is s (n, i), and is considered as N block data per frame.

The sound source signals S1 from the D sound sources 1 are received by the signal receiving unit 2, and the discrete time series signals s are received from the signal receiving unit 2.
When (n, i) is output to the non-linear feature amount estimating unit 10, the estimated signal generating unit 11 in the non-linear feature amount estimating unit 10 performs a discrete time series signal s (n, Calculate the estimated signal (n, i) for i).

That is, when each sound source 1 enters the receiver as a plane wave, the estimated signal (n, i) for the received signal s (n, i) of the i-th receiver has an estimated amplitude _j and an estimated angular frequency Equation (3), which is a nonlinear coupling function of

Here, n = 1,2, ..., N i = 1,2, ..., M k _j : wave number vector (direction), The estimation error calculation unit 12 receives the estimation signal (n, i), calculates the estimation error power of the estimation signal (n, i) with respect to the reception signal s (n, i) according to the following equation (4), and calculates The calculation result is output to the normal equation determination unit 13.

The normal equation determining unit 13 obtains an estimated feature vector that minimizes the estimated error power by the least square method. Here, the estimated signal (n, i) is the estimated value feature Since this is a function of the nonlinear combination of, the least squares linear Taylor differential correction method is applied.

In this Taylor differential correction method, the estimated signal (n, i) is Taylor-expanded around each estimated feature to the first order, and this is substituted into equation (4). (5) is obtained. This is a normal equation that determines the correction vector δP for the estimated feature vector that reduces the estimated error power.

E = F · δP (5) where correction amount vector E = (e ₁ , e ₂ ,..., E _3D ) matrix F = (f _kl ) (k, l = 1, 2,..., 3D) Matrix element g _k (n, i) = δ (n, i) / δp _k (k = 1,2,..., 3D) Next, the correction amount calculation unit 14 inputs the estimated feature amount vector, and ), That is, a correction vector δP is obtained from the following equation (5-1), and the obtained correction vector δP is provided to the estimation error determination unit 15.

δP = F ⁻¹ · E (5-1) The estimation error determination unit 15 sets the estimation error power as, for example, the minimization of the estimation error power. If the minimum value is not attained, the estimated feature value updating unit 16 sets a value obtained by adding the correction vector δP to the estimated feature value vector as a new estimated feature value vector as shown in the following equation (6).

← + δP (6) The above processing is repeated, and an estimated feature vector that minimizes the estimated error power is output as an estimated feature vector. Thereby, the amplitude a _i , the angular frequency ω _j , and the azimuth θ _{j which} are the feature amounts of the sound source signal are obtained.

FIG. 2 is a diagram showing a comparison of analysis results between the sound source feature extraction method of the present embodiment and a conventional beamformer, for example.
In FIG. 2, the azimuth θ (゜) is plotted on the horizontal axis and the power is plotted on the vertical axis, and the directional characteristic curve of the conventional beamformer is shown.

In order to compare the sound source feature extraction method of the present embodiment with a conventional beamformer, 37 receivers are arranged at 5 ° intervals in a semicircle having a radius of 1.5 m, and two sound sources arrive at the receiver as plane waves. And amplitude a ₁ = 1v, a ₂ = 1v, frequency f ₁ = 2.0k, respectively.
Hz, f ₂ = 2.0 kHz, azimuth θ1 = 81.0 °, θ ₂ = 80.0 °.

In the present embodiment, among the 37 receivers, only five adjacent (M = 5) arranged at 80 to 100 degrees were used, and the analysis time was 20 msec, and each feature amount was estimated. On the other hand, in the beamformer, all the 37 receivers (M = 37) are used, and the analysis time is 40 msec, and each feature is estimated (sampling frequency f _s ; 8 kHz).

As is clear from the directivity characteristics of the beamformer shown in FIG. 2, in the conventional beamformer, θ ₁ = 80.5
゜, θ ₂ = 80.3 °, and the peaks corresponding to the two sound sources overlap, making it difficult to distinguish them. On the other hand, the estimated values of the present embodiment are θ ₁ = 81.0 ° and θ ₂ = 79.9 °, so that the two values can be clearly divided and a more accurate estimated value can be calculated.

As described above, in the present embodiment, the sound source signal S1 arriving at the signal receiving unit 2 is converted into the sound source signal S1 using the nonlinear feature amount estimating unit 10.
Modeled by the nonlinear combination of the features of S1, the sound source signal S1
Since the characteristic amount of the data is directly calculated, it is possible to extract the characteristic amount with high resolution with a small number of receivers and with a short analysis time. Furthermore, since the number of receivers may be small, the amount of calculation is also small, whereby the structure of the apparatus can be simplified and the apparatus can be downsized. Also, the least-squares linear Taylor
Since the parameters of the non-linear combination are calculated using the differential correction method, the normal equation determination unit 13 can accurately determine the equation. Furthermore, if the spatial arrangement of the receiver is set so as to satisfy only the sampling theorem of the time / space signal, the degree of freedom of the arrangement position of the receiver is increased, and the above-described analysis is performed independently of the arrangement position of the receiver. This makes it possible to optimally design a sound source feature extraction device having the effects of shortening the time, increasing the spatial resolution, and reducing the size of the device.

Note that the present invention is not limited to the above embodiment, and the nonlinear feature estimating unit 10 shown in FIG. 1 may be constituted by a digital signal processor (DSP), a program control using a microcomputer, or the like. Various modifications are possible, such as extracting a feature amount of a sound source in the air using the present invention.

(Effects of the Invention) As described above in detail, according to the first aspect, a sound source signal arriving at a receiver is modeled by a non-linear combination of features of the sound source signal, and the features of the sound source signal are modeled. Is directly calculated, the effects of reducing the number of receivers used, shortening the analysis time, increasing the spatial resolution, and simplifying and miniaturizing the device structure can be obtained.

In the second invention, the parameter of the nonlinear coupling is a minimum of 2
Since the calculation is performed using the multiplicative linear Taylor differential correction method, the mathematical formulas to be modeled are accurately grouped, thereby improving the feature extraction accuracy. Note that the least-squares linear Tayl
The parameters of the non-linear coupling can also be obtained by using a method other than the differential correction method.

In the third invention, the receivers are arranged so as to satisfy the sampling theorem. Therefore, the degree of freedom in spatial arrangement of the receivers is increased, and the arrangement is substantially the same as that of the first invention regardless of the arrangement of the receivers. Similar effects can be obtained, and the optimal design of the sound source feature extraction device can be performed.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a sound source feature extracting apparatus using a sound source feature extracting method according to an embodiment of the present invention, and FIG. 2 is a diagram showing a comparison between the present embodiment and a conventional analysis result. 1 ... sound source, 2 ... signal receiving unit, 10 ... nonlinear feature amount estimating unit, 11 ... estimated signal generating unit, 12 ... estimated error calculating unit,
13: normal equation determination unit, 14: correction amount calculation unit, 15 ...
Estimating unit error determining unit, 16 ... Estimated feature amount updating unit.

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Satoshi Shimizu 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP 4-64075 (JP, A) JP JP-A-4-66890 (JP, A) JP-A-4-66889 (JP, A) JP-A-4-64076 (JP, A) JP-A-2-280074 (JP, A) (58) Fields investigated (Int) .Cl. ⁶ , DB name) G01S 3/80-3/86 G01S 7/52-7/66 G01S 15/00-15/96

Claims (3)

(57) [Claims] A sound source signal from a sound source is received by a plurality of receivers, and the received signal is sampled based on a sampling theorem of a time / space signal to obtain a discrete time series signal. In the sound source feature extraction method of extracting a feature amount of a signal, based on the discrete time-series signal, a feature amount of the sound source signal is formed into a mathematical expression by a parameter of a non-linear combination, and the feature amount of the sound source signal is directly calculated according to the formula. A sound source feature extraction method characterized by calculating. 2. The sound source feature extraction method according to claim 1, wherein the parameter of the non-linear combination is calculated using a least squares linear tailored differential correction method. 3. The sound source feature extraction method according to claim 1, wherein the receiver is arranged so as to satisfy the sampling theorem.