JPH0466888A - Extraction of feature for sound source - Google Patents

Abstract

PURPOSE:To achieve a reduction in the number of receivers used and a shortening of analysis time by calculating directly a sound signal by modeling it with a nonlinear combination of the feature quantities possessed by the sound source signal. CONSTITUTION:Sound source signals S1 from D pieces of sound source 1 are received with a signal receiving section 2 and when discrete time-series signals S are outputted from a nonlinear feature quantity estimating section 10, an estimation signal generating part 11 in the estimating section 10 calculates estimation signals for the signals S. An estimation error calculating section 12 inputs the estimation signals to calculate an estimation error power and the results of the calculation are outputted to a normal equation deciding section 13, which 13 determines such a estimation feature quantity vector as to minimize the estimation error power by the least squares method. Then, a correction value calculating section 14 inputs the estimation feature quantity vector to determine a correction vector, which is applied to an estimation error judging section 15. For example, the judging section 15 performs a minimization processing of the estimation error power to be used as estimation error reference. Then, an estimation feature vector corresponding to the minimum of the estimation error power obtained is outputted as reference estimation feature vector thereby enabling the determination of feature quantities of the sound source signal.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention is applicable to sound sources that exist arbitrarily in space in signal processing of sonar that performs underwater L1 position measurement (acoustic positioning), etc. Signal features (S width, frequency, direction)
The present invention relates to a sound source feature extraction method for extracting sound source features.

(Prior art) Conventionally, as a technology in this field, for example, Oki Electric Research and Development, l1 [4] (1986-10>, "Digital signal processing technology in underwater acoustics" by Nigarasu and Igarashi, jp. 5)
There was one described in B-58.

As described in the above document, SONA uses underwater acoustics to search for targets moving in a three-dimensional space (underwater);
It performs position measurement and classification.

Sonar is divided into passive sonar and active sonar depending on the mode of operation. Passive sonar uses sound waves emitted by the target, while active sonar emits sound waves toward the target and uses the reflected waves (echoes) to search for the target.
Performs position measurement and classification.

The signal processing used in SONA consists of temporal processing used to extract the temporal characteristics of the signal (waveform, spectrum, etc.) and spatial processing used to extract the spatial characteristics of the signal (position, shape, moving speed, etc.). It is divided into spatial processing used for extraction.

Among temporal processing, matched filters and Wiener filters obtain the maximum signal-to-noise ratio (S/N ratio) when a signal with a defined waveform and spectrum is buried in noise with a known spectrum. It's a filter. Spectrum estimation is used to estimate the intensity of a signal relative to its frequency, and is used to detect a periodic signal (line spectrum) buried in noise and to classify targets that radiate the signal.

Among spatial processing, beamforming (BF) uses the output signals of a large number of passers that make up a receiving array, and uses the directionality of waves propagating in space to improve the S/H of the signal and to improve the signal S/H. The incident direction and intensity (spatial spectrum) are estimated. Delay and phase estimation is a problem of estimating the delay or phase difference that occurs between signals received by a small number of passers, and is mainly used for target position measurement. This can be considered a simplification of beamforming.

Beamforming is basically a delay-addition BP that adds a delay to compensate for the difference in propagation delay to the output signals of multiple passers and then adds them. However, when the signal is a narrow band, , phase compensation can also be used instead of delay compensation. Spatial resolution is particularly important as a beamforming characteristic, and it is desirable that the main beam width be narrow and the sidelobe level be low. As a beam pattern control method, shading is used in which the output signal of each transfer device is multiplied by a predetermined weight, which is mainly effective in suppressing the sidelobe level. Next, sidelobe cancellation technology was developed to remove interference waves incident from a specific direction, and eventually adaptive beam forming (ABF) was devised to remove interference waves in any direction. Furthermore, recently, signal subspace methods have been studied that apply the latest spectrum estimation methods to azimuth estimation and dramatically improve the resolution of the main beam.

(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 passing devices must be increased in order to increase the number of input data. Furthermore, in order to obtain high resolution, the placement of the transfer device must be in a linear or circumferential arrangement that is suitable for the sound source signal, but depending on the placement of the transfer device, the window function may be too strong and the resolution may be affected. There is a problem of deterioration, and there are restrictions on setting an appropriate arrangement state. Therefore, not only does the calculation become complicated due to the increase in the number of books to be delivered, but also the device becomes larger and the arrangement of the delivery device becomes difficult, which has been difficult to solve.

The present invention solves a problem that the prior art had, in order to obtain high spatial resolution, it is necessary to increase the number of transfer devices, which makes it difficult to simplify calculations, simplify the device structure, and make it compact. The present invention provides a sound source feature extraction method that solves this problem.

(Means for Solving the Problems) In order to solve the above problems, the first invention receives a sound source signal from a sound source with a plurality of receivers, and processes the received signals based on the temporal and spatial signal sampling theorem. In a sound source feature extraction method that obtains a discrete time series signal by sampling and extracts a feature quantity of the sound source signal from the discrete time series signal,
Based on the discrete time series signal, the feature quantity of the sound source signal is expressed as a formula using nonlinear combination parameters, and the feature quantity of the sound source signal is directly calculated according to the formula.

A second invention is based on the first invention, wherein the parameters of the nonlinear combination are calculated using a minimum two-east linear Taylor differential correction method.

In a third aspect of the invention, in the first aspect, the receiver is arranged so as to satisfy the sampling theorem.

(Operation) According to the first invention, since the sound source feature extraction method is configured as described above, the sound source signal arriving from the sound source is received by a plurality of receivers, and is sampled at a predetermined frequency based on the sampling theorem to be discrete. Converted to a time series signal. Then, based on this discrete time-series signal, a mathematical model (formula) of the sound source signal is constructed using the feature quantities of the sound source signal.

Next, the characteristic @I of the sound source signal is directly calculated according to the constructed formula. This allows feature extraction with high spatial resolution in a short analysis time without increasing the number of receivers.

In the second aspect of the invention, since the least squares linear ray+or differential correction method is used to calculate the parameters of the non-linear combination in formulating the formula, it is possible to accurately set a formula with a small weight.

In the third invention, since the receivers are arranged so as to satisfy the sampling theorem, the degree of freedom in the spatial arrangement of the receivers 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 showing an embodiment of the present invention.

For example, this sound source feature extraction device extracts features (amplitude, frequency, direction) of sound source signals S1 from D sound sources 1 underwater.
This device extracts a sound source signal S1 from a sound source 1, and has a signal receiving section 2 that receives a sound source signal S1 from a sound source 1, and a nonlinear feature estimating section 10 is connected to the output side of the signal receiving section 2. The signal receiving unit 2 includes a plurality of receivers that convert the sound source signal S1 into electrical signals, and samples the outputs of the receivers at a predetermined sampling frequency based on the sampling theorem to obtain a discrete time series signal S(n,
It has a function of outputting i>. Here, s(n, i
), n is the number of sampled time series data, and 1 is the number of the receiver.

The nonlinear feature estimation unit 10 calculates a discrete time series signal S(n, i
) and estimates the special @ quantity of the sound source signal S1, and the estimated quantity for the received signal received by the receiver is input to the sound source 1.
It has an estimated signal generation section 11 that calculates using a non-linear combination function of estimable minute amounts, and an estimation error 1 output section 12 that calculates an estimation error of the estimated value with respect to the received signal is connected to its output side. A normal equation determination unit 13, a correction amount calculation unit 14, and an estimation error determination unit 15 are connected to the output side of the estimation error calculation unit 12, and an estimated feature amount update unit is connected to the output side of the estimation error determination unit 15. The estimated signal generator 11 is connected via 16.

The normal equation determination unit 13 calculates a least squares linear Tay I from the estimation error calculated by the estimation error calculation unit 12 and the estimated trace amount.
It has a function to determine a normal equation based on or differential correction. The correction amount calculation unit 14 includes the normal equation determination unit 1
It has a function of calculating the correction amount for the estimated trace amount from the normal equation determined in step 3. Estimation error determination unit 15
uses the estimation error obtained by the estimation error calculation unit 12 to determine whether the estimation error standard for the estimation error is satisfied, and if the estimation error criterion is not satisfied, controls the estimated feature amount updating unit 16 to update the estimated trace amount. let That is, the estimated feature amount updating unit 16
In , the result of adding the correction amount obtained by the correction amount calculation unit 14 to the estimated trace amount is inputted to the estimation signal generation unit 11 as a new estimation time @ amount, and the above update process is performed so that the estimation error is the estimation error. It has a function to iterate until the criteria is met. When the estimation error calculated by the estimation error calculation unit 12 satisfies the estimation error criterion, the estimation error determination unit 15 outputs the estimated small amount as the final estimated feature quantity of the sound source 1.

Next, a sound source feature extraction method of this embodiment will be explained using the sound source feature extraction device as described above.

First, considering that there are D sound sources 1 and that each sound source 1 is received by the receiver as a plane wave, the feature vector P of the sound source 1 to be extracted is calculated using the following equation (1). The feature quantities are amplitude a, angular frequency j゛ω0, and orientation θ-(j=1.2..., D> to J
J. In addition, as shown in the following equation (2), the estimated feature vector for the feature vector P is p, the estimated amplitude of each sound source 1 is 58, the estimated angular frequency is 7, J
, J estimated direction as θ−(j=1.
2. ..., D).

P:(Pl, P2++++, p3D>=(a 1 ,
a2. ”・, aD, (t) 1. ω2.”
・ωD・θ1・θ2・°゛°・aD) ......(1) p=(p engineering, l'2...., +53D)" (al
, a2. −, aD, ω1. ω2. ++δ, θ1. θ2. ..., aD) ...... (2) The number of receivers in the signal receiving section 2 is M, and the position vector of each receiver is r-(i=1.2...,M) and work it out. The received signal of the i-th receiver is sampled at the sampling frequency f5 in the signal receiving unit 2, and the sampled discrete time series signal is defined as s(n, i),
Consider one frame as N block data.

Sound source signals S1 from D sound sources 1 are received by the signal receiving unit 2, and the signal receiving unit 2 receives the discrete time series signal s(
n, i) is output to the nonlinear feature estimation unit 10, the estimated signal generation unit 11 in the nonlinear time @quantity estimation unit 10,
The discrete time series signal s(n,i) of the received signal is
Calculate the estimated signal &(n, i) for .

That is, when each sound source 1 is incident on the receiver as a plane wave, the i-th
The estimated signal i(
n, i) are the estimated amplitude 5·, the estimated angular frequency object, and the estimated orientation θ3 (j=1.2...).

D) is mathematically expressed (modeled) using the following equation (3), which is a nonlinear combination function.

s (n, 1)=IajCO3((Thjn+J−r
i>J=1 (3) However, n=1.2. ..., N 1=1 2. ...9M k, two-wave number vector (azimuth).

The estimation error calculation unit 12 inputs the estimated signal s(n, i) and calculates the received signal s(n, i) of the estimated signal s(n, i>
The estimated error power ♀ is calculated according to the following equation (4), and the calculation result is output to the normal equation determination unit 13.

The normal equation determination unit 13 determines an estimated feature vector p that minimizes the estimated error power ♀ by the least squares method. Here, the estimated signal 9(n.

i) is the estimated value feature a j, ship, θj (j=1゜2, ・
Since it is a function of a nonlinear combination of ..., D>, the minimum 2 east linear T
Apply the aylor differential correction method.

In this day+or differential correction method, the estimated signal 5(nj)
By expanding the equation (1) to the first-order term around each estimated feature quantity and substituting this into equation (4), the following equation (5) is obtained from the principle of the least squares method. This is the correction vector δ for the estimated feature vector p that reduces the estimation error power 9.
Normal equation to determine P % Formula % However, correction amount vector E = (e e ..., e3D) 1. 2. Matrix F = (fkρ) (k, , Q = 1.2..., 3D ) vector element ek = 8M Σ Σ(s (n, 1) - s (n, 1)) fk (n
, t) n=1 i=1 matrix elements fk,l! ” Matrix element gk (n, i) = δs (n, i>/δpk (k=1.2..., 3D) Next, the correction amount calculation unit 14 inputs the estimated trace vector p, A correction vector δP is obtained from equation (5), that is, the following equation (5-1>), and is provided to the estimation error determination unit 15.

δP=F-1·E (5-1) The estimation error determination unit 15 uses, for example, the minimization of the estimation error power ♀ as the estimation error standard. If not the minimum,
In the estimated feature amount updating unit 16, as shown in the following equation (6), the sum of the estimated trace amount vector p and the correction vector δP is set as a new estimated feature amount vector p.

p-+5+δP (6) The above procedure is repeated, and the estimable trace vector p that minimizes the estimated error power 9 is output as the estimable trace vector. As a result, the amplitude a, which is the special quantity of the sound source signal, the angular frequency 1, and the orientation θ are J J
J is found.

FIG. 2 is a diagram showing a comparison of analysis results between the sound source feature extraction method of this embodiment and a conventional beamformer, for example. In FIG. 2, the horizontal axis represents the azimuth θ (°), and the vertical axis represents the power, and shows the directivity characteristic curve of the conventional beamformer.

In order to compare the sound source feature extraction method of this embodiment with a conventional beamformer, 37 receivers were arranged at 5° intervals in a semicircle with a radius of 1.5 m, and two sound sources were sent to the receiver as plane waves. and the amplitude a1=1v, a2=IV, respectively.
Frequency f1=2.0kHz, f2=2-○kHz, direction θ1=8i.

0° and θ2=80.0°.

In this example, out of 37 receivers, 80° to 100°
Each feature was estimated using only 5 adjacent receivers (M=5>) arranged at 37°, and the analysis time was 20m5ec.On the other hand, with the beamformer, all 37 receivers (
M = 37>, and the analysis time was 40 m5ec to estimate each feature (sampling frequency f,
8kH2).

As is clear from the directivity characteristics of the beamformer in FIG. 2, the conventional beamformer has a maximum value of θ1=80.
5", θ2=80.3", and the peaks corresponding to the two sound sources overlap, making it difficult to distinguish them. On the other hand, the estimated value of this example is θ1=81. o°, θ2=79.
9°, which allows the two to be clearly separated and a more accurate estimated value to be calculated.

As described above, in this embodiment, the sound source signal S1 arriving at the signal receiving section 2 is modeled by a nonlinear combination of the feature quantities of the sound source signal S1 using the nonlinearity trace estimating section 10, and the features of the sound source signal S1 are Since the quantities are directly calculated, it is possible to extract high-resolution feature quantities with a small number of receivers and a short analysis time. Furthermore, since the number of receivers is small, the total amount of space required is also small, thereby simplifying the structure of the device and making the device more compact. Furthermore, since the parameters of the nonlinear combination are calculated using the minimum 2-east linear ray+or differential correction method, the equation can be determined accurately in the normal equation determination unit 13.

Furthermore, if the spatial arrangement of the receivers is set so as to satisfy only the sampling theorem for temporal and spatial signals, the degree of freedom in the receiver arrangement increases, and the above-mentioned analysis can be carried out regardless of the receiver arrangement. It becomes possible to optimally design a sound source feature extraction device that has effects such as time reduction, high spatial resolution, and miniaturization of the device.

Note that the present invention is not limited to the above-mentioned embodiments, and the nonlinear feature estimation unit 10 shown in FIG. Various modifications are possible, such as extracting feature amounts for sound sources in the air using the invention.

(Effects of the Invention) As described in detail above, according to the first invention, the sound source signal arriving at the receiver is modeled by a nonlinear combination of the feature quantities of the sound source signal, and the feature 1 of the sound source signal is is calculated directly, reducing the number of receivers used.
The effects of shortening analysis time, increasing spatial resolution, and simplifying and downsizing the device structure can be obtained.

In the second aspect of the invention, the parameters of the nonlinear combination are calculated using the minimum two-east linear Taylor differential correction method, so that the modeling formula can be accurately compiled, thereby improving the accuracy of feature extraction.

Note that it is also possible to obtain the parameters of the nonlinear combination using a method other than the minimum 2 east linear ray+or differential correction method.

In the third invention, since the receivers are arranged so as to satisfy the sampling theorem, the degree of freedom in the spatial arrangement of the receivers is increased, and it is almost the same as the first invention, regardless of the arrangement of the receivers. A similar effect can be obtained, and an optimal design of the sound source feature extraction device can be achieved.

[Brief explanation of the drawing]

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, and FIG. 2 is a diagram showing a comparison between the present embodiment and conventional analysis results. 1...Sound source, 2...Signal receiving section, 10
......Nonlinear feature estimation section, 11... Estimated signal generation section, 12... Estimation error calculation section, 13
...... Normal equation determination section, 14... Correction amount calculation section, 15... Estimation section, difference judgment section, 16.
...Estimated feature update unit.

Claims (1)

[Claims] 1. A sound source signal from a sound source is received by a plurality of receivers, the received signal is sampled based on the sampling theorem of temporal and spatial signals to obtain a discrete time series signal, and the discrete time series signal is obtained. In the sound source feature extraction method of extracting the feature amount of the sound source signal from A sound source feature extraction method characterized by directly calculating. 2. The sound source feature extraction method according to claim 1, wherein the parameters of the nonlinear combination are calculated using a least squares linear Taylor differential correction method. 3. The sound source feature extraction method according to claim 1, wherein the receivers are arranged so as to satisfy the sampling theorem.