JPH0943330A - Narrow band signal tracking signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve accuracy of tracking of a narrow band signal. SOLUTION: A frequency analyzer 2 performs analysis of frequency of a sound signal received by a sound sensor 1 on the basis of the given frequency analyzing width W for each the specified time. An integrator 3 integrates signal power for each frequency in time on the basis of the given integral time T. A degree of freedom calculating device 11 calculates the degree of freedom υof X<2> square distribution. A normalizing device 12 normalizes the signal power. An event detector 5 detects the event. A multiplier 13 finds a product of the power (x) of the event multiplied by the smoothing power A of the track, a table reference device 15 referrers to an intermediate tolerance rate table 14, and calculates the intermediate tolerance rate. A power tolerance rate calculator 16 finds the power tolerance rate Λa from the intermediate tolerance rate. A tracking device 6 selects an event having the largest power tolerance rate Λa, and unifies it to the track.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a narrow band signal for continuously measuring a narrow band signal whose characteristics are unknown based on a signal power distribution obtained by performing frequency analysis on an input signal such as a sound wave. The present invention relates to a band signal tracking method.

[0002]

2. Description of the Related Art FIG. 2 is a functional block diagram showing a narrow band signal tracking device for carrying out a conventional narrow band signal tracking method for tracking a narrow band signal. A conventional narrowband signal tracking method will be described below with reference to FIG. Here, the input signal is a sound wave. The acoustic sensor 1 receives an acoustic signal. The frequency analysis device 2 is
The frequency of the acoustic signal received by the acoustic sensor 1 is analyzed according to the given frequency analysis width W. This processing is usually realized by FFT (Fast Fourier Transform) processing and square detection processing. The integrator 3 temporally integrates the signal power for each frequency according to the given integration time T. In order to normalize the influence of noise, the normalization device 4 obtains the average value and standard deviation of the signal power of all frequencies, subtracts the average value from the signal power of all frequencies, and divides by the standard deviation. . The event detection device 5 detects an event from the normalized signal power. Here, an event is defined as a point at which the signal power is higher than the frequency bins on both sides and the signal power exceeds a predetermined threshold value. The tracking device 6 forms a temporal integration of events (hereinafter referred to as a track). If an existing track exists, the predicted frequency of the track is calculated, and if the difference between the frequency of the newly detected event and the predicted frequency is less than or equal to a threshold value w _f , the event is recorded in the track. To integrate. However, when there are two or more corresponding events, the event having the largest power likelihood ratio calculated by the power likelihood ratio calculation device 8 described later is selected and integrated.

If there is no applicable event,
The signal is considered lost and the track is deleted. If there is an event that has not been integrated into any track, a new track is generated based on the event. When the event is integrated into the track, the smoothing power A of the track is updated based on the power x of the event. For this, for example, a suitable positive number η (0
<Η <1) may be determined in advance, and ηx + (1-η) A may be newly set as A. Further, the smoothed frequency φ of the track is updated based on the frequency f of the event, and the predicted frequency is calculated. For this, for example, a Kalman filter can be used. Track information is recorded in the track information file 7. The power likelihood ratio calculation device 8 calculates the power likelihood ratio for the pair of the track being tracked and the newly detected event. When the smoothing power of the track is A and the instantaneous power of the event is x, the power likelihood ratio is calculated by the following equation (1). Λ _g (x) = A ² -2Ax (1) Here, the basis of the equation (1) is shown. First, the hypothesis "the event is caused by noise" is H ₀ , and the hypothesis "the event is caused by the narrowband signal represented by the track" is H ₁ . Next, regarding the probability distribution that the integrated and normalized input signal follows, the probability distribution when the hypothesis H ₀ is true is represented by the probability density function P _0a (x), and similarly, when the hypothesis H ₁ is true Probability distribution function P _1a
It is represented by (x).

At this time, the power likelihood ratio when an event having power x is detected is P _0a (x) of P _1a (x).
It is defined to be the numerical value of the ratio to.

(Equation 5) It is usually assumed that the power of the input signal in the presence of only noise follows a standard normal distribution. The standard deviation may be 1 because the input signal is normalized.

(Equation 6) On the other hand, if a narrowband signal is present and the power after integration and normalization is A, then the input signal is assumed to follow a normal distribution with mean A and standard deviation 1.

(Equation 7) Assuming Equations (3) and (4) above, the power likelihood ratio Λ _g (x) is obtained as follows, and Equation (1a) is obtained.

(Equation 8) The power likelihood ratio is used to select events to integrate into the track as described above. Further, it is output to the subsequent processing as a certainty factor indicating the certainty that a true narrowband signal exists, not due to noise in the track.

[0005]

However, the conventional narrow band signal tracking method has the following problems. The calculation of the power likelihood ratio by the above method is based on a model in which the power of the input signal follows a normal distribution, which is a rough approximation. According to a more sophisticated model, in the case of noise only, chi ² square distribution, when the narrow band signal of the noise to the power A is applied follows the non-central chi ² square distribution ^{(non-central χ 2 distribution)} . Each probability density distribution function is expressed by the following equation.

[Equation 9] However, υ is the degree of freedom and is equal to 2TW. T is the integration time, and W is the frequency analysis width. I _n () is a modified Bessel function of the nth order of the first kind. The χ ² distribution and the non-central χ ² distribution are described in the following documents. Reference: William S. Burdic, UNDERWATER ACOUSTIC
ANALYSIS ", PENTICE-HALL, P.P. 387
When the distributions represented by the equations (5) and (6) are approximated by a normal distribution, a large error occurs especially when the degree of freedom is small.

FIGS. 3 (a) and 3 (b) are explanatory views of the problem when the degree of freedom is 4, and in particular, FIG. 3 (a) shows the non-center χ.
^{This is} a probability density distribution calculated by a model based on a square distribution, and FIG. 6B is a probability density distribution calculated by a model based on a normal distribution. In FIGS. 3A and 3B, the horizontal axis represents power and the vertical axis represents probability density. FIG. 3 shows the case where the average value = 4 and the standard deviation σ = 8 ^1/2 . It can be seen that there is a considerable difference between the two models, as shown in FIG. Therefore, in order to accurately obtain the power likelihood ratio under the condition that the degree of freedom is small, it is necessary to obtain the likelihood ratio based on the noncentral χ ² distribution as follows.

(Equation 10) However, while the power likelihood ratio in the normal distribution model can be calculated only by addition and subtraction multiplication as shown in the equation (1), the equation (7) includes the Bessel function and the like, A large amount of calculation is required to obtain high calculation accuracy. For this reason, models based on the non-central χ ² distribution are not often used in applications where speed performance is required. An object of the present invention is to enable more accurate narrowband signal tracking by calculating the likelihood ratio based on the non-central χ ² distribution without significantly increasing the amount of calculation.

[0007]

The narrow band signal tracking method of the first invention executes the following processing in order to solve the above problems. That is, for every predetermined time for the input signal, the frequency analysis is performed according to the predetermined frequency analysis width, and the frequency analysis processing for obtaining the distribution of the signal power on the frequency,
An integration process of temporally integrating the signal power according to a predetermined integration time, and χ ² which the integrated signal power follows
A degree of freedom calculation process for calculating a degree of freedom ν of a power distribution, and assuming that the integrated signal power follows a χ ² distribution, based on the standard deviation of the integrated signal power and the degree of freedom, the integration is performed. A normalization process for normalizing the signal power thus obtained is performed. Then, a track is formed by an event detection process of detecting an event by detecting a point having the maximum value of the integrated signal power distribution and having a value exceeding a predetermined threshold value, and a time series of events having frequencies close to each other. If the difference between the frequency of the newly detected event and the smoothed frequency of the existing track is less than or equal to the threshold value w _f and there are a plurality of events that satisfy the difference, the event is tracked based on the power likelihood ratio. And the tracking process for calculating the smoothing power A of the track.

Further, a multiplication process for calculating a product Ax from the power x of the normalized event and the smoothing power A of the track, and a plurality of numerical values υ determined at predetermined intervals
_{For i} (integer of i ≧ 2) and α _j (integer of j ≧ 2),
Intermediate likelihood ratio

[Equation 11] With reference to the intermediate likelihood ratio table giving, (upsilon, alpha) intermediate likelihood ratio A _a1 (υ, Ax) of the intermediate likelihood ratio calculation processing for obtaining the intermediate likelihood ratio A _a1 (υ, Ax ), The power likelihood ratio calculation process for obtaining the power likelihood ratio Λ _a = Λ _a1 (υ, Ax) -A / 2 is performed.

According to the first aspect of the invention, since the narrow band signal tracking method is configured as described above, the degree of freedom calculation process
The degree of freedom ν of the χ ² distribution that the integrated signal power follows is calculated, and the integrated signal power is χ ² by the normalization process.
Assuming that the power distribution is followed, the integrated signal power is normalized based on the standard deviation and the degree of freedom of the integrated signal power. The product of the normalized event power x and the track smoothing power A by the multiplication process A
Calculate x. The intermediate likelihood ratio calculation process is performed by smoothing power x
And Ax function, refer to the intermediate likelihood ratio table that gives the intermediate likelihood ratio for ν _i (integer of i ≧ 2) and α _j (integer of j ≧ 2) sampled in advance. Then, the intermediate likelihood ratio A _a1 (υ, Ax) of (υ, α) is obtained. By the power likelihood ratio calculation process, the intermediate likelihood ratio A
_A power likelihood ratio is calculated from _a1 (υ, Ax), and based on the power likelihood ratio (for example, the maximum likelihood ratio),
Integrate events into tracks. It is a precise model,
For only noise, chi ₂ square distribution, when the narrow band signal of the noise to the power A is applied, on the assumption that follows a non-central chi ₂ square distribution, so called for power likelihood ratio, high Enables accurate narrowband signal tracking. In addition, the amount of calculation is not significantly increased. Therefore, the above problem can be solved.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION First Embodiment FIG. 1 is a functional block diagram of a narrowband signal tracking device for carrying out the narrowband signal tracking method of the first embodiment of the present invention. The elements common to the elements of No. 2 are denoted by the same reference numerals. The narrow band signal tracking method of the first embodiment differs from the conventional narrow band signal tracking method in that the power of the input signal follows a χ ² distribution model, and the degree of freedom ν _i ( _i
≧ 2), the power α of the event, and the sample α _j (j ≧ 2) of the product α of the smoothing power A of the track, the intermediate likelihood ratio Λ is stored in a table, and the intermediate likelihood ratio Λ is stored. Referring to the table, the intermediate power-likelihood ratio Λ _a1 (x, Ax) is calculated, power likelihood ratio from the middle likelihood ratio Λ _a (x, _a
x) is calculated, and the events are integrated based on the power likelihood ratio. As shown in FIG. 1, the narrowband signal tracking apparatus according to the first embodiment has a degree of freedom for calculating a signal power degree of freedom ν from an acoustic sensor 1 that receives an input signal, a frequency analysis width W, and an integration time T. Calculation device 11,
The intermediate likelihood ratio table 14 stores the intermediate likelihood ratios of a plurality of degrees of freedom υ _i (i ≧ 2) and a plurality of pairs of α _j (j ≧ 2) (υ _i , α _j ).

A frequency analyzer 2 for analyzing the frequency of a sound wave is connected to the output side of the acoustic sensor 1, and an integrator 3 is connected to the output side of the frequency analyzer 2. To the output side of the integrator 3 and the degree-of-freedom calculation device 11, a normalization device 12 that normalizes the signal power from the degree of freedom υ and the standard deviation σ of the power is connected, and the output side of the normalization device 12 is further connected. The event detection device 5 is connected to. To the output side of the event detection device 5, one terminal of the tracking device 6 and the multiplier 13 is connected. A track information file 7 is connected to the input / output side of the tracking device 6. The output side of the track information file 7 is connected to the other terminal of the multiplier 13 and the power likelihood ratio calculation device 16. A table search device 15 is connected to the output sides of the multiplier 13 and the intermediate likelihood ratio table 14, and a power likelihood ratio calculation device 16 is connected to the output side of the table search device 15. The tracking device 6 is connected to the output side of the power likelihood ratio calculation device 16. Hereinafter, with reference to FIG. 1, description of the narrowband signal tracking method of the first embodiment (a) to
Do (j).

(A) Frequency Analysis Processing The acoustic sensor 1 receives an acoustic signal and outputs the acoustic signal to the frequency analysis device 2. The frequency analysis device 2 provides the acoustic signal received by the acoustic sensor 1 at a predetermined frequency analysis width W (frequency bins f _i and f _{i + 1} (i =
The frequency of the acoustic signal received by the acoustic sensor 1 is analyzed according to the interval (1, 2, ...,). This processing is usually realized by FFT processing and square detection processing. (B) Integration Processing The integrator 3 temporally integrates the signal power for each frequency f _i (i = 1, 2, ...) According to the given integration time T. (C) Degree-of-Freedom Calculation Processing The degree-of-freedom calculation device 11 calculates the degree of freedom ν of the χ ² distribution that the signal power follows. This is calculated by the following equation (8) from the integration time T and the frequency analysis width W. υ = 2WT (8) (d) Normalization processing The normalization device 4 obtains the standard deviation σ of the signal power x _i at all frequencies f _i , and then the signal power is expressed by the following equation (9). Multiply the coefficient S. ^{S = (2υ) 1/2 / σ} ··· (9) where the power of the input signal because it shall follow the chi ² distribution model is for performing a normalization process in the chi ² distribution model. (E) Event detection processing The event detection device 5 detects an event from the signal power after normalization. Here, an event is defined as a point at which the signal power is higher than the frequency bins on both sides and the signal power exceeds a predetermined threshold value. (F) Multiplication Processing The multiplier 13 inputs the event power x from the event detection device 5, the track smoothing power A from the tracking device 6 described later, and the product Ax (= α) to the table search device 15 Output to.

(G) Table Retrieval Processing The table retrieval device 15 inputs the degree of freedom ν and the product α, and retrieves the intermediate likelihood ratio table 14 using these as arguments.
Here, the intermediate likelihood ratio is a portion obtained by removing −A / 2 from the power likelihood ratio represented by the equation (7), and is represented by the following equation (10). Λ _a1 = lnI _{2 / υ-1} − ((υ-2) / 4) lnAx + lnΓ (υ / 2) + (υ / 2-1) ln2 (10) As can be seen from the equation (10), The likelihood ratio can be regarded as a function of two variables (υ, Ax = α). Therefore,
The intermediate likelihood ratio table 14 has intermediate likelihoods calculated in advance by the following equation (11) for υ _i (≧ 2) determined at appropriate intervals and α _j (≧ 0) determined at appropriate intervals. Record the ratio. Λ _a1 (υ _i , α _j ) = lnI _{υi / 2 -1} (α _j ^1/2 ) − ((υ _i -2) / 4)) ln α _j + ln Γ (υ _i / 2) + (υ _i / 2 -1) ln2 (11) FIG. 4 shows, as a specific example, υ _i = {2,4,6,8}, 0 ≤α _j ≤20
It is a figure which shows the value of the intermediate likelihood ratio of. For example, the value of Λ _a1 (υ _i , α _j ) where υ _i and α _j shown in FIG.
It is stored in the intermediate likelihood ratio table 14 together with the values of _i and α _j . In the above table search, for example, after performing linear interpolation on α _j , linear interpolation is performed on υ _i . That is, to obtain Λ _al (υ, Ax), _{i0 such} that υ _i0 ≤ υ <υ _{i0 + 1} and α _j0 ≤ Ax <α
_Using j0 which is _{j0 + 1} , Λ _i0 = Λ _al (υ _i0 , α _j0 ) + [(Ax−α _j0 ) / (α _{j0 + 1} −α _j0 )] (Λ _a1 (υ _i0 , α _{j0 + 1} ) −Λ _a1 (υ _i0 , α _j0 )) (12) Λ _{i0 + 1} = Λ _al (υ _{i0 + 1} , α _j0 ) + [(Ax−α _j0 ) / (α _{j0 + 1} − α _j0 )] (Λ _a1 (υ _{i0 + 1} , α _{j0 + 1} ) −Λ _a1 (υ _{i0 + 1} , α _j0 )) ・ ・ ・} (13) Λ _a1 (υ, Ax) = Λ _i0 + [ (υ−υ _i0 ) / (υ _{i0 + 1} −υ _i0 )] (Λ _{i0 + 1} −Λ _i0 ) ... (14)

(H) Power Likelihood Ratio Calculation Processing The power likelihood ratio calculation device 16 uses the intermediate likelihood ratio Λ _a1 (υ, A
x), the power likelihood ratio Λ _{a given} by the following equation (15) is obtained. Λ _a = Λ _a1 (υ, Ax) -A / 2 (15) (i) Tracking process The tracking device 6 forms a track. If an existing track exists, the predicted frequency of the track is calculated, and if the difference between the frequency of the newly detected event and the predicted frequency is less than or equal to a threshold value w _f , the event is recorded in the track. To integrate. However, when there are two or more corresponding events, the event having the largest power likelihood ratio Λ _a calculated by the power likelihood ratio calculation device 16 is selected and integrated. If there is no corresponding event, the signal detection is considered lost and the track is deleted. If there is an event that has not been integrated into any track, a new track is generated based on the event. When the event is integrated into the track, the smoothing power A of the track is updated based on the power x of the event. For this purpose, for example, an appropriate positive number η (0 <η <1) is predetermined, and ηx + (1-
η) A may be newly set as A. Further, the smoothed frequency φ of the track is updated based on the frequency f of the event, and the predicted frequency is calculated. For this, for example, a Kalman filter can be used. Track information is recorded in the track information file 7. The tracking device 6 can also use the frequency likelihood ratio shown below together with the power likelihood ratio.

(J) Frequency Likelihood Ratio Calculation Processing If there is no signal and an event appears due to noise, the frequency is uniformly distributed in the section of width 2w _f . On the other hand, if there is a signal with frequency φ and frequency deviation σ _f , the event frequency follows a normal distribution.

(Equation 12) Therefore, when there is an event of frequency f, a smoothed frequency (obtained in the same manner as the smoothed power) φ, and a track of frequency deviation σ _f , the hypothesis “The event is due to the signal represented by the track. The likelihood ratio of
Represented by 8).

(Equation 13) Furthermore, to be equal to a constant multiple of the previously sigma _f threshold w _f in selecting an event, if the threshold value w _f variable, method of calculating the frequency likelihood ratio, as the following equation (19) It will be easy.

Λ _f =-(f-φ) ² / 2σ _f ² + C _f (19) where C _f is a constant and C _f = ln (w _f / σ _f ) +1/2
ln (2 / π). In addition, in order to obtain the likelihood ratio using both the power and the frequency, the likelihood ratio is calculated using the simultaneous distributions P _0af (a, f) and P _1af (a, f) of the signal power and the frequency. To do. If this integrated likelihood ratio is Λ,

[Equation 14] That is, the power likelihood ratio and the frequency likelihood ratio are calculated independently and then added, and the tracking device 6 selects the event that gives the maximum addition result of the power likelihood ratio and the frequency likelihood ratio. , Integrate the event into a track. As described above, according to the first embodiment, the model based on the non-centered χ ² distribution, which is closer to the reality than the conventional model based on the normal distribution, is adopted as the signal model. When it is small, it is possible to perform highly accurate tracking based on a precise power likelihood ratio. Moreover, since the added operations are only table search and addition / subtraction multiplication as compared with the conventional technique, the amount of operation is only slightly increased.

Second Embodiment FIG. 5 is a functional block diagram of a narrow band signal tracking device for carrying out a narrow band signal tracking method according to a second embodiment of the present invention. Common elements are given common reference numerals. The narrow band signal tracking method of the second embodiment is different from that of the first embodiment in that input signals for each beam directed in different directions are tracked. As shown in FIG. 5, the narrowband signal tracking device of the first embodiment has an acoustic sensor 1 that receives an input signal, a degree-of-freedom calculation device 11, and an intermediate likelihood ratio table 14. The output side of the acoustic sensor 1 is connected to a phasing device 21 that forms beams directed in different directions. Further, the output side of the phasing device 21 is a frequency for analyzing the frequency of the input signal for each beam. The analyzer 22 is connected. An integration device 23 is connected to the output side of the frequency analysis device 22. On the output side of the integrator 23 and the degree-of-freedom calculator 11, the degrees of freedom ν and the standard deviation σ of power are provided.
Therefore, the normalization device 24 for normalizing the power is connected, and the event detection device 25 is connected to the output side of the normalization device 24. The output side of the event detection device 25 is connected to one terminal of the tracking device 26 and the multiplier 13. A track information file 27 is connected to the input / output side of the tracking device 26. To the output side of the track information file 27, the other terminal of the multiplier 13 and the power likelihood ratio calculation device 16 are connected. Multiplier 13 and intermediate likelihood ratio table 1
A table search device 15 is connected to the output side of the table 4. A power likelihood ratio calculation device 16 is connected to the output side of the table search device 15, and a tracking device 26 is connected to the output side of the power likelihood ratio calculation device 16.
The narrowband signal tracking method (a) to (j) of the second embodiment will be described below with reference to FIG.

(A) Phasing process The acoustic sensor 1 receives an acoustic signal and outputs the acoustic signal to the phasing device 21. The phasing device 21 forms beams directed in different directions and sends an input signal for each beam to the frequency analysis device 22. (B) Frequency Analysis Processing The frequency analysis device 22 performs frequency analysis of the acoustic signal for each beam at a predetermined time and in accordance with the given frequency analysis width W, as in the first embodiment. , Signal power in a two-dimensional space of azimuth × frequency is obtained and sent to the integrator 23. (C) Integral Processing The integrator 23 outputs a signal for each frequency for each azimuth according to the integration time T for each azimuth with respect to the signal power in the secondary space of azimuth × frequency. Integrate power over time. (D) Degree-of-Freedom Calculation Processing As in the first embodiment, the degree-of-freedom calculation device 11 calculates the degree of freedom ν of the distribution that the signal power follows from the frequency analysis width W and the integration time T.

(E) Normalization processing The normalization device 24 is assumed to follow the non-central χ ² distribution model of the signal power, as in the first embodiment, and the standard deviation σ of the signal power and the degree of freedom υ are used. , Normalize the power. (F) Event detection processing The event detection device 25 detects an event from the signal power after normalization. The signal power is given as a distribution in the frequency space in the first embodiment, but here,
It is input in the direction x distribution in the frequency space. Therefore, an event is defined as a point having an azimuth and a frequency that are larger than the frequency bins on both sides of the beam in the same beam, are larger than the signal power of the same frequency bin on the beams on both sides, and have a signal power exceeding a predetermined threshold. (G) Multiplication processing The multiplier 13 multiplies the event power by the smoothing power of the track corresponding to the event for each event and sends it to the table search device 15, as in the first embodiment.

(H) Table Retrieval Processing The table retrieval device 15 calculates the intermediate likelihood ratio for each event as in the first embodiment. (I) Power Likelihood Ratio Calculation Processing The power likelihood ratio calculation device 16 calculates the power likelihood ratio for each event, as in the first embodiment. (J) Tracking process If an existing track exists, the tracking device 26 calculates the predicted frequency and predicted direction of the track, and sets a threshold value having a difference between the frequency of the newly detected event and the predicted direction. If w _θ or less, integrate the event into the track. When there are two or more events to be integrated, the event that maximizes the power likelihood ratio is selected as in the first embodiment. At the time of integration, the smoothed power, the smoothed frequency, and the predicted frequency are calculated in the same manner as in the first embodiment, and the direction θ of the event is calculated.
The smoothed azimuth B and the predicted azimuth of the track are calculated based on This can also be realized by the Kalman filter. In addition to the power likelihood ratio, the frequency likelihood ratio shown in the first embodiment can be used.

Further, as described below, the azimuth likelihood ratio can be calculated by the same method as the calculation of the frequency likelihood ratio. That is, when there is a track with an azimuth φ and an azimuth deviation σ _θ and an event with an azimuth θ, the likelihood ratio Λ _θ based on the narrowband signal represented by the track is Λ _{θ θ = ln (1 / ((} 2π) 1/2 σ θ) -ln (1 / 2w θ) = - (θ-B) 2 / 2σ θ 2 -lnσ θ + lnw θ +1/2 ln2 / π ··· (21) Furthermore, if the threshold value w _θ for selecting an event is changed in advance so as to be equal to a constant multiple of σ _θ , the frequency likelihood ratio calculation method is simplified as shown in the following expression (22). Λ _θ =-(θ-B) ² / 2σ _θ ² + C _θ (22) where C _θ is a constant and C _θ = lnw _θ / σ _θ +1/2 ln
It is 2 / π. As described above, according to the second embodiment, even when the signal is received by a plurality of beams and the azimuth tracking is performed with respect to the arrival direction of the signal, the same as in the first embodiment. It is possible to perform highly accurate tracking based on a precise power likelihood ratio. Note that the present invention is not limited to the above embodiment, and various modifications are possible. For example, there are the following modifications.

(1) Each block in FIGS. 1 and 5 may be realized by an electric circuit, or a part or all of these may be realized by a computer program. (2) In the embodiment, the input signal is a sound wave, but the present invention is not limited to this. For example, when the input signal is a radio wave, in FIG. 1 and FIG.
If a receiving antenna is used instead of, the same operation is performed. (3) The power likelihood ratio was obtained by linear interpolation, but many intermediate likelihoods are stored in the intermediate likelihood table, the closest combination of ν _i0 and α _j0 is obtained, and Λ _a1 (υ _i0 ，
α _j0 ) may be the intermediate likelihood.

[0023]

As described in detail above, the first to fifth embodiments
According to the invention, since a model based on a non-centered χ ² distribution that is closer to reality than a model based on a conventional normal distribution is adopted as a signal model, when the degree of freedom is small,
It is possible to perform highly accurate tracking based on a precise power likelihood ratio. Moreover, since the added operations are only table search and addition / subtraction multiplication as compared with the conventional technique, the amount of operation is only slightly increased.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a narrowband signal tracking device for carrying out a narrowband signal tracking method according to a first embodiment of the present invention.

FIG. 2 is a functional block diagram of a narrowband signal tracking device for implementing a conventional narrowband signal tracking method.

FIG. 3 is an explanatory diagram of a conventional problem.

FIG. 4 is a diagram showing a value of an intermediate likelihood ratio.

FIG. 5 is a functional block diagram of a narrowband signal tracking device for carrying out the narrowband signal tracking method according to the second embodiment of the present invention.

[Explanation of symbols]

 1 Acoustic Sensor 2,22 Frequency Analyzer 3,23 Integrator 5,25 Event Detector 6,26 Tracking Device 7,27 Track File 11 Freedom Calculator 13 Multiplier 14 Intermediate Likelihood Ratio Table 15 Table Search Device 16 Power Likelihood ratio calculation device 21 Phase adjusting device

Claims (6)

[Claims] 1. A frequency analysis process for performing a frequency analysis on an input signal at a predetermined time interval according to a predetermined frequency analysis width to obtain a distribution of signal power on a frequency, and the signal power according to a predetermined integration time. Integral processing that integrates in time, and the degree of freedom ν of the χ ² distribution that the integrated signal power follows
A degree of freedom calculation process for calculating, and assuming that the integrated signal power follows a χ ² distribution, based on the standard deviation of the integrated signal power and the degree of freedom, with respect to the integrated signal power Normalization processing for normalization, event detection processing for detecting a point having a maximum value of the distribution of the integrated signal power and having a value exceeding a predetermined threshold value as an event, and an event having a frequency close to each other. A track is formed from the temporal continuity of, and the difference between the frequency of the newly detected event and the smoothing frequency of the existing track is less than or equal to the threshold value w _f ,
When there are a plurality of events that satisfy the conditions, the event is integrated into the track based on the power likelihood ratio, and the tracking process for calculating the smoothed power A of the track, and the normalized event power x And a multiplication process for calculating the product Ax from the smoothing power A of the track, and a plurality of numerical values υ _i (integer of i ≧ 2) and α _j (integer of j ≧ 2) defined at predetermined intervals, Intermediate likelihood ratio [Equation 1] With reference to the intermediate likelihood ratio table giving, (upsilon, alpha) intermediate likelihood ratio A _a1 (υ, Ax) of the intermediate likelihood ratio calculation processing for obtaining the intermediate likelihood ratio A _a1 (υ, Ax And a power likelihood ratio calculating process for obtaining the power likelihood ratio Λ _a = Λ _a1 (υ, Ax) -A / 2 from the above), and a narrowband signal is tracked. Tracking method. 2. A phasing process for forming a beam directed to a plurality of azimuths with respect to an input signal, and a predetermined frequency for every predetermined time with respect to the input signal received by the beam of each azimuth. Frequency analysis is performed according to the analysis width to obtain a distribution of signal power in a two-dimensional space of azimuth × frequency; integration processing for temporally integrating the signal power according to a predetermined integration time; and the integration Degree of freedom χ ² distribution of signal power υ
A degree of freedom calculation process for calculating, and assuming that the integrated signal power follows a χ ² distribution, based on the standard deviation of the integrated signal power and the degree of freedom, with respect to the integrated signal power The normalization processing for normalization, the event detection processing for detecting a point having a maximum value of the integrated signal power distribution and having a value exceeding a predetermined threshold as an event, and the azimuth and frequency are both A track is formed from a time series of adjacent events, and the difference between the frequency of the newly detected event and the smoothing frequency of the existing track is a threshold value w.
_If it is less than or equal to _f , and the difference between the azimuth of the event and the smoothed azimuth of the track is less than or equal to the threshold value w _θ and there are multiple events that satisfy them, the event is tracked based on the power likelihood ratio. Tracking processing for calculating the smoothed power A and smoothed frequency of the track, and multiplication processing for calculating the product Ax from the normalized event power x and the smoothed power A of the track. For a plurality of numerical values υ _i (integers of i ≧ 2) and α _j (integers of j ≧ 2) defined by intervals, the intermediate likelihood ratio With reference to the intermediate likelihood ratio table giving, (upsilon, alpha) intermediate likelihood ratio A _a1 (υ, Ax) of the intermediate likelihood ratio calculation processing for obtaining the intermediate likelihood ratio A _a1 (υ, Ax And a power likelihood ratio calculating process for obtaining the power likelihood ratio Λ _a = Λ _a1 (υ, Ax) -A / 2 from the above), and a narrowband signal is tracked. Tracking method. 3. From the smoothed frequency φ of the track and the frequency deviation σ _{f of the} track, the frequency likelihood ratio And a likelihood ratio adding process for calculating a sum of the power likelihood ratio and the frequency likelihood ratio, the tracking process including: the power likelihood ratio and the frequency likelihood ratio. Based on the sum of the ratios,
The narrowband signal tracking method according to claim 1 or 2, wherein the event is integrated into a track. 4. The azimuth likelihood ratio of the event from the smoothing azimuth B of the track, the azimuth standard deviation σ _{θ of} the track, and the azimuth θ of the event. And a likelihood ratio addition process for calculating the sum of the power likelihood ratio and the azimuth likelihood ratio, and the tracking process performs the power likelihood ratio and the azimuth likelihood. The narrow band signal tracking method according to claim 2, wherein the event is integrated into the track based on a sum of the ratios. 5. A frequency likelihood ratio calculation process according to claim 3, an azimuth likelihood ratio calculation process according to claim 4, and a sum of the power likelihood ratio, the frequency likelihood ratio, and the azimuth likelihood ratio. And a tracking process for integrating the event into a track based on the sum of the power likelihood ratio, the frequency likelihood ratio, and the azimuth likelihood ratio. The narrowband signal tracking method according to claim 2, wherein 6. The intermediate likelihood ratio calculation process is performed by A _a1 (υ, which is given from the power likelihood ratio table, for ν _i0 ≦ υ <υ _{i0 + 1} and α _j0 ≦ α <α _{j0 + 1} . _i0 , α
_j0 ), A _a1 (υ _i0 , α _{j0 + 1} ), A _a1 (υ _{i0 + 1} , α _j0 ),
The narrow band signal according to claim 1 or 2, wherein the intermediate likelihood ratio A _a1 (υ, α) is obtained by interpolation processing from A _a1 (υ _{i0 + 1} , α _{j0 + 1} ). Tracking method.