JP2021135422A - Signal processing device and signal processing method - Google Patents

Abstract

To provide a signal processing device and signal processing method that reduce a calculation amount.SOLUTION: A signal processing device that estimates noise from data including a narrow-band signal acquired by a sensor, comprises: a storage unit that stores an estimated value of a frequency characteristic of noise from a plurality of samples including samples around a frequency in correspondence with the frequency; and an arithmetic logical unit that, when estimating noise at each frequency, reads, from the storage unit, an estimated value of a frequency characteristic of noise at a frequency other than a frequency targeted for estimation, and adds a frequency spectrum of a newly added sample to estimate noise.SELECTED DRAWING: Figure 2

Description

The present invention relates to a signal processing device and a signal processing method for detecting a desired signal from a signal containing noise.

Conventionally, when a narrow band signal is detected by using an acoustic sensor such as a microphone or a hydrophone for the purpose of searching for a target and monitoring an abnormality of an object, a frequency analysis process such as FFT (Fast Fourier Transform) is used. After forming the frequency spectrum by the frequency analysis process, the frequency spectrum is normalized in the frequency direction for the purpose of emphasizing the narrow band signal. In order to normalize the frequency spectrum in the frequency direction, it is necessary to estimate the frequency characteristics of noise such as background noise and wideband signals. A method is known in which a plurality of signals are sampled and a median of the sampled signals is used for noise estimation.

As another noise estimation method, a method using OTA (Order Truncate Average) has been proposed (see, for example, Non-Patent Document 1).

William A. Struzinski and Edward D. low, "The effect of September normalization on the performance of an acoustic vector", The Journal of the Acoustic 1985 (The Journal of the Acoustic 1985).

A conventional noise estimation method, for example, a method using a median, has a large processing amount as compared with an FFT, and in a situation where real-time processing is required, a method of reducing the calculation amount of signal processing has been desired.

The signal processing device according to the present invention is a signal processing device that estimates noise from data including a narrow band signal acquired by a sensor, and noise due to a plurality of samples including samples around the frequency corresponding to the frequency. When estimating the noise at each frequency and the storage unit that stores the estimated value of the frequency characteristic of the above, the estimated value of the frequency characteristic of the noise at a frequency other than the frequency to be estimated is read from the storage unit and newly added. It has a calculation unit for estimating noise by adding a frequency spectrum of a sample.

The signal processing method according to the present invention is a signal processing method by a signal processing device that estimates noise from data including a narrow band signal acquired by a sensor, and includes a plurality of samples around the frequency corresponding to the frequency. The step of storing the estimated value of the frequency characteristic of the noise by the sample of the above, and when estimating the noise at each frequency, the estimated value of the noise related to the frequency other than the frequency to be estimated is read from the stored information and newly added. It has a step of estimating the noise by adding the frequency spectrum of the sample.

According to the present invention, when estimating noise at a certain frequency, the estimated value of noise already calculated for other frequencies can be used, so that the amount of calculation required for noise estimation can be reduced.

It is a figure which shows an example of the hardware composition of the signal processing apparatus which concerns on Embodiment 1. FIG. It is a functional block diagram which shows one structural example of the arithmetic unit shown in FIG. It is a block diagram which shows one configuration example of the signal processing apparatus of the comparative example. It is a schematic diagram which shows the flow of the normalization processing in the frequency direction by the signal processing apparatus of the comparative example. It is a schematic diagram for demonstrating the process of estimating the frequency characteristic of the background noise by the signal processing apparatus of the comparative example. It is a schematic diagram for demonstrating the process of step S201 shown in FIG. It is a schematic diagram for demonstrating the process of estimating the frequency characteristic of the background noise by the signal processing apparatus which concerns on Embodiment 1. FIG.

Embodiment 1.
The configuration of the signal processing device of the first embodiment will be described. FIG. 1 is a diagram showing an example of a hardware configuration of the signal processing device according to the first embodiment. The signal processing device 1 of the first embodiment is, for example, an information processing device such as a computer. As shown in FIG. 1, the signal processing device 1 includes an input unit 2, a storage unit 3, a calculation unit 4, and an output unit 5. The storage unit 3 is, for example, an HDD (Hard Disk Drive) device. The output unit 5 is, for example, a display device. The calculation unit 4 has a memory 11 for storing a program and a CPU (Central Processing Unit) 12 for executing processing according to the program. The memory 11 is, for example, a non-volatile memory such as a flash memory.

The input unit 2 is communicably connected to an acoustic sensor 7 such as a microphone and a hydrophone. A signal detected by the acoustic sensor 7 is input to the input unit 2. The acoustic sensor 7 is, for example, an array sensor composed of a plurality of receivers. The input unit 2 includes an amplifier (not shown) that amplifies the signal, a filter (not shown) that selects the signal, and an A / D (Analog to Digital) converter (not shown) that converts an analog signal into a digital signal. It is composed of electronic circuits including.

FIG. 2 is a functional block diagram showing a configuration example of the calculation unit shown in FIG. The calculation unit 4 displays the FFT processing means 21 that Fourier transforms the signal input from the input unit 2, the normalization processing means 22 that normalizes the frequency spectrum after the Fourier transform in the frequency direction, and the detection processing means 23. It has a processing means 24. The detection processing means 23 detects a narrow band signal to be detected based on the frequency spectrum normalized by the normalization processing means 22. The display processing means 24 causes the output unit 5 to output the detection processing result by the detection processing means 23.

Before explaining the configuration of the signal processing device 1 of the first embodiment in detail, the configuration and operation of the signal processing device of the comparative example will be described. FIG. 3 is a block diagram showing a configuration example of a signal processing device of a comparative example. The signal processing device 100 is a device that uses a median for noise estimation.

The signal processing device 100 is, for example, an information processing device such as a computer. The signal processing device 100 of the comparative example shown in FIG. 3 is different from the signal processing device 1 shown in FIG. 1 in the normalization processing means 122, and other configurations are the same as those of the signal processing device 1. Although omitted in FIG. 3, the acoustic sensor 7 shown in FIG. 1 is communicably connected to the input unit 2.

The operation of the signal processing device 100 of the comparative example will be described. A narrow band signal arrives at the acoustic sensor 7 shown in FIG. Here, the narrowband signal received at the center of the array sensor is expressed by the following equation (1).

In equation (1), t is time, A is the amplitude of the narrowband signal, f is the frequency of the narrowband signal, and φ ₀ is the initial phase.

In general, the signal received by the acoustic sensor 7 or the vibration sensor (not shown) includes not only the desired signal but also noise (underwater noise and a signal generated from other than the target). The data including the narrow band signal of the equation (1) and the noise is expressed by the following equation (2).

E (t) in the equation (2) is time series data of noise. The output (time waveform) of each receiver of the acoustic sensor 7 passes through an electronic circuit such as an amplifier (not shown) and a filter (not shown) of the input unit 2, and then an A / D converter (not shown). Is converted to a digital signal.

In equation (3), [・] represents discrete data. n is an index indicating time. Next, the FFT processing means 21 applies the FFT to the discrete data of the equation (3). Equation (4) shows the case of DFT (Discrete Fourier Transform).

In the formula (4), k is an index indicating the frequency, and N is the number of samples when FFT is applied. Let K be the number of discrete frequencies, and 1 ≦ k ≦ K. K corresponds to the frequency range to be analyzed with respect to k, and the frequency when k = 1 corresponds to the minimum frequency.

Next, the normalization processing performed by the normalization processing means 122 will be described. FIG. 4 is a schematic diagram showing the flow of frequency direction normalization processing by the signal processing device of the comparative example. The normalization processing means 122 normalizes the output (frequency spectrum) of the FFT according to the processing flow shown in FIG.

In order to normalize the frequency spectrum in the frequency direction, as shown in FIG. 4, the frequency characteristics of the background noise and the broadband signal are estimated, and the original frequency spectrum is divided by the estimated frequency characteristics of the background noise and the broadband signal. By doing so, the original frequency spectrum is normalized.

FIG. 5 is a schematic diagram for explaining a process of estimating the frequency characteristic of background noise by the signal processing device of the comparative example. FIG. 6 is a schematic diagram for explaining the process of step S201 shown in FIG. The normalization processing means 122 estimates the frequency characteristics of the background noise by estimating the noise according to the procedure shown in FIG.

Specifically, in step S201 shown in FIG. 5, when estimating noise with respect to a certain frequency k, the normalization processing means 122 samples L values around the frequency k as shown in FIG. Here, L is an odd number. The normalization processing means 122 uses L samples and uses a median as shown in the equation (5) so that the background is not included in the influence of the narrow band signal for each frequency X [k]. The frequency characteristic Y [k] of noise is estimated.

In equation (5), X'is a sorted sample for L samples around the frequency k. The normalization processing means 122 normalizes the FFT output of the equation (4) by using the estimated value of the frequency characteristic of the background noise of the equation (5) as shown in the following equation (6).

In the signal processing device 100 of the comparative example, when estimating the frequency characteristic of the background noise, it is necessary to calculate the equation (6) for each frequency, and as a result, the processing amount becomes larger than that of the FFT. Therefore, in a situation where real-time processing is required, it is indispensable to reduce the processing amount. Therefore, the signal processing device 1 of the first embodiment aims to reduce the amount of calculation in the process of estimating the frequency characteristic of the background noise.

The configuration of the signal processing device 1 of the first embodiment will be described as being different from the signal processing device 100 of the comparative example. The normalization processing means 22 shown in FIG. 2 uses a part of the noise estimation output of step S202 shown in FIG. 5 for the frequency k for the processing of step S201 shown in FIG. 5 for the frequency (k + 1). ..

The operation of the signal processing device 1 of the first embodiment will be described. FIG. 7 is a schematic diagram for explaining a process of estimating the frequency characteristic of background noise by the signal processing device according to the first embodiment.

In step S102, the normalization processing means 22 sorts L samples and outputs a median value. Further, in step S102, the normalization processing means 22 outputs not only the noise estimation value (median value) Y (k) for a certain frequency k but also the result of sorting the samples X'(k) = [X ( The set of l _k ) and l _k ] is also output at the same time. Storage unit _{3, X '(k) = [} X (l k), l k] stores. Normalization processing means _{22, X '(k) = [} X (l k), l k] the set of return to the process of step S101 (step S103), the set, noise estimate for the frequency (k + 1) (Median value) Used when calculating Y (k + 1). l _k is a vector of indices of the surrounding L samples of the frequency k, X (l _k) is an L number of values sampled. X '(k) is a vector sampled X (l _k) are sorted in power.

The operation of step S101 regarding the frequency (k + 1) will be described. Normalization processing means 22, when sampling the L pieces from the frequency around the frequency (k + 1), X obtained when calculated with respect to the frequency _{k '(k) = [X} (l k), l k] from , Delete the sample corresponding to the frequency m _{k + 1} and add the sample corresponding to _{m k + 1'.} Here, m _{k + 1} is a frequency excluded from "L samples from the surrounding frequency" with respect to the frequency (k + 1), and m _{k +} 1'is a frequency excluded from "L samples from the surrounding frequency" with respect to the frequency (k + 1). The frequency added to the "sample".

The number M of m _{k + 1} and m _{k + 1'may} be plural, but the number of samples to be deleted and the number of samples to be added are equal, and the total number of samples is L. Further, the normalization processing unit 22 compares 'when adding the sample X which has already been sorted _{X (m k + 1)'} (k) = [X (l k), l k] and the magnitude of the value By doing so, the rearrangement is performed as X'(k + 1) = [X (l _{k + 1} ), l _{k + 1].}

Also in the first embodiment, the normalization processing means 22 obtains an estimated value Y [k] of the frequency characteristic of the background noise for each frequency k, and uses the obtained estimated value Y [k] to obtain the equation (6). As shown in, the FFT output of Eq. (4) is normalized. In the first embodiment, a storage unit 3, the sort result _{X '(k) = [X} (l k), l k] have been described in the case of storing, the memory 11 is sort result X' (k _{) = [X (l k)} , l k] may be stored. In this case, the memory 11 functions as a storage unit for storing the sort result of each frequency.

The effect of the first embodiment will be described in comparison with the signal processing method by the signal processing device 100 of the comparative example. In the signal processing method of the comparative example, L samples were sorted for each frequency k. Assuming that the worst value of the processing amount related to sorting is L ² , the number of discrete frequencies is K, so the processing amount required to calculate the noise of all frequencies is K (L ² ).

On the other hand, in the signal processing method of the first embodiment, it is not necessary to sort the overlapping samples again among the L samples used for estimating the noise for each frequency k. Only the process of comparing which of the (LM) samples newly added to estimate the noise with respect to k is already sorted (LM) is performed. Therefore, for k = 1, it is necessary to sort L samples, which results in ^{the processing amount of L 2.} However, regarding the subsequent (K-1) times of processing, for example, when M = 1, the processing is performed. The amount is (L-1). Therefore, in the signal processing method of the first embodiment, the amount of processing required to calculate the noise of all frequencies without lowering the noise estimation ability is reduced as shown in the following equation (7). can do.

The signal processing device 1 of the first embodiment has a storage unit 3 and a calculation unit 4. The storage unit 3 stores an estimated value of the frequency characteristic of noise by a plurality of samples including a sample around the frequency k corresponding to the frequency k. When estimating the noise at each frequency, the calculation unit 4 reads out the estimated value of the frequency characteristic of the noise of the frequency other than the frequency to be estimated from the storage unit 3, adds the frequency spectrum of the newly added sample, and adds the noise. To estimate.

According to the first embodiment, when estimating the noise at a certain frequency, the estimated value of the noise already calculated for the other frequency can be used, so that the amount of calculation required for the noise estimation can be reduced. ..

In the first embodiment, a method of frequency direction normalization processing has been described as a process for detecting and analyzing a narrowband signal, but the process is not limited to frequency direction normalization processing. Further, in the first embodiment, the case of using the median having the highest noise estimation ability has been described, but the noise estimation method using OTA disclosed in Patent Document 1 also has the present embodiment 1. It is possible to apply the method of.

1 Signal processing device 2 Input unit 3 Storage unit 4 Calculation unit 5 Output unit 7 Acoustic sensor 11 Memory 12 CPU
21 FFT processing means 22 Normalization processing means 23 Detection processing means 24 Display processing means 100 Signal processing device 122 Normalization processing means

Claims (4)

A signal processing device that estimates noise from data including narrowband signals acquired by sensors.
A storage unit that stores an estimated value of the frequency characteristic of noise by a plurality of samples including a sample around the frequency corresponding to the frequency, and a storage unit.
When estimating noise at each frequency, an operation of reading out an estimated value of the frequency characteristic of the noise at a frequency other than the frequency to be estimated from the storage unit and adding the frequency spectrum of a newly added sample to estimate the noise. Department and
A signal processing device having. The calculation unit estimates noise from the frequency spectra of the plurality of samples by sequentially changing the frequency from the minimum frequency for a determined frequency range, and obtains an estimated value of the frequency characteristic of the noise corresponding to the frequency. To memorize in the memory part,
The signal processing device according to claim 1. The storage unit stores the result of sorting the frequency spectra of the plurality of samples corresponding to the frequency as an estimated value of the frequency characteristic of the noise.
The calculation unit compares the result of sorting the frequency spectra of the plurality of samples at frequencies other than the frequency to be estimated with the frequency spectra of the newly added sample.
The signal processing device according to claim 1 or 2. It is a signal processing method by a signal processing device that estimates noise from data including narrowband signals acquired by a sensor.
A step of storing an estimate of the frequency characteristics of noise by a plurality of samples including samples around the frequency corresponding to the frequency, and
When estimating the noise at each frequency, the step of reading the estimated value of the noise related to the frequency other than the frequency to be estimated from the stored information and adding the frequency spectrum of the newly added sample to estimate the noise, and the step of estimating the noise.
Signal processing method having.

Images