JP2010258891A - Signal quantizing apparatus, method and program, recording medium therefor, and signal quantizing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve quantization performance by adding noise of a suitable strength to an analog signal to be input to a signal quantizing apparatus whose quantization parameter is not optimal. <P>SOLUTION: A noise adding section 1 adds noise of a high strength σ sequentially to an analog signal to be input. A quantizing section 2 quantizes each analog signal to which the noise of the strength σ has been added. An evaluation amount calculating section 3 calculates the average value of absolute values of errors between sampled values of samples and output values as an evaluation amount for each analog signal to which the noise of the strength σ has been added. An optimal noise strength determining section 5 compares the evaluation amounts and determines the strength of noise corresponding to the smallest evaluation amount as the optimal noise strength. The noise adding section 1 adds noise of the determined optimal strength to the analog signal to be input. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for quantizing an analog signal. In particular, it relates to a technique for optimizing the quantization of an analog signal.

Quantization of an analog signal refers to converting an analog signal having a continuous value into a digital signal having a discrete value.
An I and any positive integer, threshold theta _i is _{θ 1 <θ 2 <... <} θ i <θ i + 1 <... <θ is predetermined so that _I, in each section, separated by the threshold, Assume that a representative value, which is a discrete value representing a section, is determined in advance. For example, in the sections (−∞, θ ₁ ], (θ ₁ , θ ₂ ],..., (Θ _i , θ _{i + 1} ],..., (Θ _I , ∞), the representative values x ₀ ∈ (−∞, θ ₁ ], x ₁ ∈ (θ ₁ , θ ₂ ],..., X _i ∈ (θ _i , θ _{i + 1} ],..., X _I ∈ (θ _I , ∞) are defined.

Sample value x obtained by sampling an input analog signal, the threshold theta _{i <a} x ≦ threshold theta _{i + 1,} the threshold theta _i and the threshold theta _{i + 1} separated by intervals _{(θ i, θ i + 1} ] Is included, the representative value x _i of the section (θ _i , θ _{i + 1} ) is output. Specifically, the analog signal is quantized and converted into a digital signal in this way.
In Non-Patent Document 1, a technique for improving the quantization performance by optimizing the quantization parameters such as the threshold value θ _i and the representative value x _i is known.

Allen Gersho (Author), Robert M.Gray (Author), Furui Sadaaki (Translation), Hiroshi Kodera (Translation), Saburo Tazaki (Translation), Hiroshi Watanabe (Translation), "Vector quantization and information compression", Corona October 1998

However, in the real world, there are signal quantizers that cannot be changed once the threshold value θ _i and the representative value x _i are determined. In this signal quantization apparatus, there is a problem that the quantization performance cannot be increased using the technique described in Non-Patent Document 1.

When the threshold value θ _i and the representative value x _i are not optimized, it is known that the quantization performance can be improved by adding noise having an appropriate intensity to the input analog signal (for example, See Reference 1.). Here, the high quantization performance is expressed by an evaluation amount such as an error between the sample value and the output value, a mutual information amount to be described later, and when the noise intensity is changed, the evaluation amount is only one minimum value or The inventors have found that the quantization performance is maximized by adding noise having a maximum value and intensity corresponding to the minimum value or the maximum value.
[Reference 1] S. Mizutani, J. Muramatsu, K. Arai, P. Davis, “Noise-assisted quantization”, International Symposium on Nonlinear Theory and Its Applications (NOLTA 2006), pp.843-846, 2006.

  The present invention uses this property to increase the quantization performance by adding noise of appropriate strength to the input analog signal.

  The quantization performance can be improved by adding noise of appropriate strength.

The functional block diagram of the example of a signal quantization apparatus. The functional block diagram of the example of an optimal noise strength determination part. The functional block diagram of the example of a signal quantization system. The flowchart of the example of a signal quantization method. The flowchart of the example of the method of narrowing down the range of the optimal noise intensity. The figure which illustrates an experimental result. The figure which illustrates the mode of change of mutual information amount I when standard deviation (sigma) is changed.

Hereinafter, embodiments of the present invention will be described in detail.
FIG. 1 illustrates a functional block of a signal quantization apparatus according to the present invention, and FIG. 4 illustrates a flowchart of a signal quantization method according to the present invention. As illustrated in FIG. 1, the signal quantization apparatus includes a noise addition unit 1, a quantization unit 2, an evaluation amount calculation unit 3, a storage unit 4, and an optimum noise intensity determination unit 5.

<Step S1>
The noise adding unit 1 adds at least two types of noises having an intensity to the input analog signal (step S1). The analog signal to which noise has been added is sent to the quantization unit 2.
Here, the intensity of noise refers to the magnitude of the spread of the probability distribution, assuming that noise occurs according to a certain probability distribution. For example, when variance σ ² and standard deviation σ are defined in the probability distribution, variance σ ² and standard deviation σ are noise intensities.

Adding noise of a predetermined intensity to an input analog signal means adding noise generated according to a probability distribution based on the predetermined intensity to the analog signal. Accordingly, even when noise of the same predetermined intensity is added, the magnitude of noise that is normally added changes.
In this embodiment, the noise is generated according to the normal distribution N (0, σ ² ), and the standard deviation σ is used as the noise intensity. The noise adding unit 1 first selects a standard deviation σ ₁ close to a predetermined 0, generates noise according to the normal distribution N (0, σ ₁ ² ), and generates noise generated in the input analog signal. Add.

Thereafter, the noise adding unit 1 inputs noise generated according to the normal distribution N (0, σ ² ) based on the standard deviation σ ₂ = σ ₁ + Δ obtained by increasing the standard deviation σ ₁ by a predetermined change width Δ. Added to the analog signal. Δ is an arbitrary positive real number. In this way, the noise adding unit 1 uses the normal distribution N (0, σ based on the standard deviation σ _{i + 1} = σ _i + Δ obtained by increasing the standard deviation σ _i by a predetermined width Δ, where i is a positive integer. _The noise generated according to _{i + 1} ² ) is sequentially added to the analog signal.
In the sampling of the quantization unit 2 to be described later, each normal distribution is obtained for the length of time in which about 10 to 100,000 samples to which noise generated according to each normal distribution N (0, σ _i ² ) is added are obtained. Each noise generated according to N (0, σ _i ² ) is added to the analog signal.

  The input analog signal usually changes every moment. A predetermined intensity of noise is added to the analog signal that changes every moment. Therefore, the sample obtained by sampling of the quantizing unit 2 is obtained by adding noise of a predetermined intensity to a sample of an analog signal having a normally changing value. That is, the sample obtained by sampling of the quantizing unit 2 is not obtained by adding noise of a predetermined intensity to a sample of an analog signal having the same value.

<Step S2>
The I and any positive integer, threshold theta _i is _{θ 1 <θ 2 <... <} θ i <θ i + 1 <... <θ is predetermined so that _I, in each section, separated by the threshold, Assume that a representative value, which is a discrete value representing a section, is determined in advance. For example, in the sections (−∞, θ ₁ ], (θ ₁ , θ ₂ ],..., (Θ _i , θ _{i + 1} ],..., (Θ _I , ∞), the representative values x ₀ ∈ (−∞, θ ₁ ], x ₁ ∈ (θ ₁ , θ ₂ ],..., X _i ∈ (θ _i , θ _{i + 1} ],..., X _I ∈ (θ _I , ∞) are defined.

The quantization unit 2 samples the analog signal to which noise has been added at a predetermined sampling frequency to obtain a sample value x of each sample. When the sample value x of a certain sample satisfies the threshold θ _i <x ≦ threshold θ _{i + 1} and is included in the section (θ _i , θ _{i + 1} ) delimited by the threshold θ _i and the threshold θ _{i + 1.} Outputs the representative value x _i of the section (θ _i , θ _{i + 1} ) as the output value of the sample (step S2).

The output value x _i is sent to the evaluation amount calculator 3.
<Step S3>
The evaluation amount calculation unit 3 calculates an evaluation amount representing the high level of quantization performance in the quantization unit 2 for an analog signal to which noise of a predetermined intensity is added (step S3). The calculated evaluation amount is stored in the storage unit 4.
The storage unit 4 stores an evaluation amount for an analog signal to which noise of each intensity is added.

As the evaluation amount, for example, an error between the sample value and the output value can be used. Specifically, as the evaluation amount, it is possible to use an average value of the absolute value of the error between the output value x _i and the sample values x of each sample of a given intensity analog signal from which noise has been added. The average value of the absolute values of the errors is referred to as “average error”. The total number of samples s _j (j = 1,..., J, J is an integer of 10 ≦ J ≦ 100000. For example, J = 10000) of the analog signal to which noise of a predetermined intensity is added is J, and the sample s _j If the sample value of x is expressed as x (s _j ) and the output value is expressed as x _i (s _j ), the average error can be described by the following equation, for example. Thus, it can be said that the smaller the value of the evaluation amount defined by the average error, the higher the quantization performance.
(1 / J) · Σ _{j = 1} ^J | x (s _j ) −x _i (s _j ) | (1)

<Step S4>
The optimum noise intensity determination unit 5 compares the evaluation amounts read from the storage unit 4 and determines the noise intensity that enhances the quantization performance (step S4). The optimum noise intensity determination unit 5 includes a comparison unit 51 and a determination unit 52 as illustrated in FIG.

  The comparison unit 51 (FIG. 2) of the optimum noise intensity determination unit 5 is one level lower than the evaluation amount (hereinafter referred to as the first evaluation amount) for the analog signal to which noise of a certain intensity is added. An evaluation amount (hereinafter referred to as a second evaluation amount) for the analog signal to which strong noise is added is compared (step S41). If the first evaluation amount is higher than the second evaluation amount, the determination unit 52 determines that the noise intensity corresponding to the second evaluation amount is lower than the noise intensity corresponding to the second evaluation amount. The noise intensity between the noise intensities one step higher than the intensity is determined as the optimum noise intensity (step S42). The noise intensity that is one step higher than the noise intensity corresponding to the second evaluation amount is the noise intensity corresponding to the first evaluation amount. For example, the determination unit 52 selects the noise intensity corresponding to the second evaluation amount as the optimum noise intensity.

  In step S41, if the first evaluation amount is smaller than the second evaluation amount, the process returns to step S1, and the noise adding unit 1 adds noise having a higher level to the analog signal. Thereafter, the processing from step S1 to step S41 is repeated until the first evaluation amount becomes higher than the second evaluation amount. If the first evaluation amount is smaller than the second evaluation amount in step S41, the noise intensity comparison unit 54 of the optimum noise intensity determination unit 5 determines the predetermined sufficiently high threshold T and the first evaluation amount. You may compare with the intensity | strength of corresponding noise (step S5). If the noise intensity is greater than the threshold value, it can be determined that noise having a higher intensity than the noise intensity corresponding to the minimum value has already been added to the analog signal input from the outside. In this case, the noise adding unit 1 does not add noise to the input analog signal. If the noise intensity is not greater than the threshold value, the process returns to step S1, and the noise adding unit 1 adds noise having a higher level to the analog signal.

  It has been found that an evaluation amount based on an error such as an average error has only one minimum value when the noise intensity is changed. That is, when gradually adding high-intensity noise from low-intensity noise, the evaluation amount based on the error usually decreases and rises at the minimum value. For this reason, when it changes from the state of the first evaluation amount <the second evaluation amount to the state of the first evaluation amount> the second evaluation amount, the second evaluation satisfies the relationship of the first evaluation amount> the second evaluation amount. The noise intensity that gives the highest quantization performance is between the noise intensity one level lower than the noise intensity corresponding to the quantity and the noise intensity one level higher than the noise intensity corresponding to the second evaluation quantity. It can be determined that it exists.

In this example, the standard deviation σ, which is the noise intensity, is sequentially changed by the change width Δ. Therefore, when the noise intensity corresponding to the second evaluation quantity is σ _i , the noise intensity one step lower than the noise intensity σ _i corresponding to the second evaluation quantity is σ _i −Δ, and the second evaluation The noise intensity one step higher than the noise intensity σ _i corresponding to the quantity is σ _i + Δ. Therefore, the determination unit 52 selects an optimal noise intensity that satisfies the relationship of σ _i −Δ <optimal noise intensity <σ _i + Δ.

Information about the determined optimum noise intensity is sent to the noise adding unit 1. The noise adding unit 1 adds the determined optimum intensity noise to the input analog signal.
Thus, even in a signal quantizer that cannot change the threshold value θ _i and the representative value x _i , if the threshold value θ _i and the representative value x _i are not optimized, it is appropriate for the input analog signal. By adding noise of a high intensity, the quantization performance can be enhanced.

[Narrowing down the optimal noise intensity range]
When the comparison unit 51 (FIG. 2) determines that the first evaluation amount is higher than the second evaluation amount, the noise adding unit 1 is more than the noise intensity corresponding to the second evaluation amount. A change that is finer than the change width Δ when the noise intensity is increased first, within a range that includes the noise intensity that is one step lower than the noise intensity corresponding to the second evaluation quantity. Optimal noise is obtained by repeating the processing from step A1 (FIG. F2) to step A4 in which the same processing as step S1 to step S4 is performed with the addition of noise having a width Δ ′ (Δ ′ <Δ) in sequence. You may narrow down the range of the intensity. Since step A1 to step A4 are the same as step S1 to step S4, respectively, description thereof will be omitted.

  In Step A41, when it is determined that the newly obtained first evaluation amount is higher than the newly obtained second evaluation amount, the determining unit 52 of the optimum noise intensity determining unit 5 determines the newly obtained first evaluation amount. The noise intensity between the noise intensity one level lower than the noise intensity corresponding to the second evaluation quantity and the noise intensity one level higher than the noise intensity corresponding to the newly obtained second evaluation quantity Is determined as the optimum noise intensity. For example, the noise intensity corresponding to the newly obtained second evaluation amount is determined as the optimum noise intensity.

The noise intensity corresponding to the local minimum value, that is, the range in which the optimum noise intensity exists is narrowed down from, for example, the section (σ _i −Δ, σ _i + Δ) to the section (σ _i −Δ ′, σ _i + Δ ′). Therefore, it is possible to select a noise intensity closer to the noise intensity corresponding to the minimum value. Further, by roughly estimating the range in which the optimum noise intensity exists with the rough change width Δ, and then narrowing down the optimum noise intensity range with the fine change width Δ ′ in that range, suddenly the fine change width Δ It can reduce the amount of calculation than finding the optimal noise intensity.

  In Step A41, if it is determined that the newly obtained first evaluation amount is higher than the newly obtained second evaluation amount, a further finer change width Δ ″ (Δ ″ <Δ ′). ), Noise with high intensity may be added sequentially, and the processing from step A1 to step A4 may be repeated. Thereby, it is possible to further narrow down the range of the optimum noise intensity. In this manner, by repeating the processing from step A1 to step A4 with the change width being made finer in sequence, the optimum noise intensity range can be narrowed down as necessary.

[Random search]
The noise adding unit 1 randomly selects a predetermined number K (K is an arbitrary natural number) in order of the noise intensity in a predetermined noise intensity range, and selects each of the selected intensity for the input analog signal. Noise may be added. Since the processes of the quantization unit 2 and the evaluation amount calculation unit 3 in this case are the same as described above, the description thereof is omitted.

  In this case, the optimum noise intensity determination unit 5 compares the K evaluation quantities respectively corresponding to the selected K noise intensities, and determines the noise intensity that enhances the quantization performance. For this purpose, for example, the intensity of noise that enhances the quantization performance is determined as follows.

First, the noise adding unit 1 randomly selects a noise intensity within a predetermined noise intensity range, and adds the noise having the selected intensity r ₁ to the input analog signal. The quantization unit 2 and the evaluation amount calculation unit 3 perform the same processing as described above, calculate the evaluation amount corresponding to the noise of the selected intensity r ₁ , and store it in the storage unit 4.

After that, the noise adding unit 1 randomly selects the noise intensity again within a predetermined noise intensity range, and adds the noise having the selected intensity r ₂ to the input analog signal. The quantization unit 2 and the evaluation amount calculation unit 3 perform the same processing as described above, calculate the evaluation amount corresponding to the noise of the selected intensity r ₂ , and store it in the storage unit 4.

The comparison / selection unit 53 of the optimum noise intensity determination unit 5 compares the evaluation amount corresponding to the noise of the intensity r ₁ with the evaluation amount corresponding to the noise of the intensity r ₂ , and determines an evaluation amount that improves the quantization performance. select. In the case where the evaluation amount indicates that the quantization performance is higher as the evaluation amount is smaller as the error-based evaluation amount, the evaluation amount corresponding to the noise of intensity r _{1 and} the noise of intensity r ₂ are supported. An evaluation amount having a small value is selected from the evaluation amounts to be performed.

After that, the noise adding unit 1 randomly selects a noise intensity again in a predetermined noise intensity range, and adds the noise having the selected intensity r ₃ to the input analog signal. The quantization unit 2 and the evaluation amount calculation unit 3 perform the same processing as described above, calculate the evaluation amount corresponding to the noise of the selected intensity r ₃ , and store it in the storage unit 4.

The comparison / selection unit 53 of the optimum noise intensity determination unit 5 compares the evaluation amount corresponding to the noise of the intensity r ₃ with the evaluation amount selected by the previous comparison / selection unit 53 to evaluate the quantization performance. Select the amount.

Thus, the k as an integer 2 or more K, random selection of strength r _k of noise in the noise adding unit 1, and, in the comparison selection unit 53, estimated amount and the previous selection corresponding to the intensity r _k of the noise The comparison selection with the evaluated amount is repeated. As a result, the last selected evaluation amount is the evaluation amount that maximizes the quantization performance among the evaluation amounts corresponding to the selected intensity of noise. The optimum noise strength determination unit 5 determines the strength corresponding to the last selected evaluation amount as the optimum noise strength.

[Signal quantization system]
As illustrated in FIG. 3, the noise adding unit 1 may be disposed outside the signal quantizing device as the noise adding device 10. The signal quantizing device 11 is obtained by removing the noise adding unit 1 from the signal quantizing device. The signal quantizing device 11 is the same as the signal quantizing device described above except that the noise adding unit 1 is arranged outside as the noise adding device 10. The signal quantization system includes a noise adding device 10 and a signal quantization device 11.

  Noise having a predetermined intensity is added to the analog signal input to the signal quantizing device 11 by the noise adding device 10. Since the method of adding noise is the same as that of the noise adding unit 1 described above, description thereof is omitted. The analog signal to which noise has been added is input to the signal quantizing device 11 via the communication network.

  Similarly to the above, the quantization unit 2, the evaluation amount calculation unit 3, and the optimum noise intensity determination unit 5 quantize the analog signal to which noise has been added, calculate the evaluation amount, and compare the calculated evaluation amounts to make the optimum The intensity of noise. Information on the determined optimum noise intensity is sent to the noise adding device 10 via the communication network. The noise adding device 10 adds the determined optimum intensity noise to the analog signal input to the signal quantizing device 11.

  Thus, by providing the noise adding device 10 outside the signal quantizing device 11, an analog signal input to the signal quantizing device 11 even if the signal quantizing device 11 does not have the noise adding unit 1. Noise with the optimum intensity can be added.

[Experimental result]
Assume that the input analog signal follows a uniform distribution U [−1/2, 1/2], and the added noise follows a normal distribution N (0, σ ² ). Further, the threshold value θ ₁ = 0.25, the threshold value θ ₂ = 0.5, the threshold value θ ₃ = 0.75, the representative value of the section (−∞, 0.25) is 0.125, and the section (0.25, 0 The representative value of .5 is set to 0.375, the representative value of the section (0.5, 0.75) is 0.625, and the representative value of the section (0.75, ∞) is 0.875. To do.

The standard deviation σ is used as the noise intensity, the initial value of the noise intensity σ is 0.1, the change width Δ of the noise intensity σ is 0.1, and the evaluation amount is n = 2 in the following equation (2). The average value of the square of the absolute value of the error between the sample value x (s _j ) and the output value x _i (s _j ) of each sample s _j is used (referred to as mean square error). J = 10000.
FIG. 6 shows the relationship between the noise intensity σ simulated using a computer and the mean square error as an evaluation amount in this case.

The relationship between the noise intensity σ and the evaluation amount {σ, evaluation amount} changes as follows.
{0.1, 0.1399969} → {0.2,0.139215} → {0.3,0.136665} → {0.4,0.136316} → {0.5,0.141758}
When σ = 0.5, the first evaluation amount = 0.141758> second evaluation amount = 0.136316. In this case, according to the present invention, the noise intensity corresponding to the second evaluation amount is less than 0.4. Between the noise intensity 0.3 lower in level and the noise intensity 0.5 higher than the noise intensity 0.4 corresponding to the second evaluation quantity, that is, the section (0.3, 0.5). It can be determined that there is an optimum noise intensity. Here, it can be confirmed from FIG. 6 that there is an optimum noise intensity in the section (0.3, 0.5).

[Modifications, etc.]
In the above embodiment, the noise adding unit 1 generates noise according to the normal distribution N (0, σ ² ) whose average value is 0, but the normal distribution N (a, σ ² ) whose average value is other than 0. The noise may be generated according to the above and added to the input analog signal. a is an arbitrary real number. In the above example, the normal distribution N (0, σ ²⁾ is caused noise according to normal distribution N (0, σ ²⁾ Noise may be generated according to the probability distribution other than. For example, a distribution such as a uniform distribution and a Cauchy distribution can be used as the noise probability distribution.

  In the above embodiment, the strength is gradually increased from a value close to 0, but the strength may be gradually decreased from a predetermined high value. The predetermined high value is set to a value higher than the noise intensity corresponding to the maximum value or the minimum value.

  In this case, the comparison unit 51 of the optimum noise intensity determination unit 5 determines in step S41 the evaluation amount (hereinafter referred to as the first evaluation amount) for the analog signal to which noise of a certain intensity is added, and the intensity thereof. An evaluation amount (hereinafter, referred to as a second evaluation amount) for an analog signal to which noise of one step higher intensity is added is compared. Then, in step S42, when the first evaluation amount is higher than the second evaluation amount, the determination unit 52 sets the noise intensity one step lower than the noise intensity corresponding to the second evaluation amount and the second evaluation amount. The noise intensity between the noise intensity levels one step higher than the corresponding noise intensity is determined as the optimum noise intensity. For example, the noise intensity corresponding to the second evaluation amount is determined as the optimum noise intensity.

  In the above-described embodiment, the noise adding unit 1 sequentially generates high-intensity noise with a predetermined change width Δ, but the change width Δ may be fixed or variable. In other words, as long as a higher or lower intensity is selected, the magnitude of the fluctuation range of the intensity does not matter. The same applies to the change widths Δ ′, Δ ″, and the like.

In the above embodiment, the average error is used as the evaluation quantity. However, the absolute value of the error between the sample value x (s _j ) and the output value x _i (s _j ) of each sample s _j is assumed, where n is an integer of 2 or more. An average value of the nth power of the value (referred to as an average nth power error) may be used. Further, as an evaluation value, a value obtained by setting n ′ to an integer of 2 or more and an average n-th power error to 1 / n ′ may be used. For example, n ′ = n.
(1 / J) · Σ _{j = 1} ^J | x (s _j ) −x _i (s _j ) | ⁿ (2)

Moreover, you may use the mutual information amount defined by a following formula as an evaluation amount. The following formula is a definitional formula for mutual information when N is an integer of 2 or more and quantization is performed by N division (that is, the number of thresholds is N−1). p _x (x) is the probability distribution of the input analog signal, p (i | x) is the probability of i being output when the input is x, and p (i) is the probability of output i.

  The mutual information amount means that the higher the value, the higher the quantization performance. Further, when the mutual information amount is used as the evaluation amount, it is known that the evaluation amount has one maximum value as illustrated in FIG. 7 when the noise intensity is changed.

In FIG. 7, it is assumed that the input analog signal follows a uniform distribution U [−1/2, 1/2], and the added noise follows a normal distribution N (0, σ ² ). _{When 1} = 0.5, threshold value θ ₂ = 0.75, threshold value θ ₃ = 1.0 and quantization is performed in four divisions, and the standard deviation σ is used as the noise intensity, the standard deviation that is the noise intensity It is a figure which illustrates the mode of the change of the mutual information amount when (sigma) is changed. In the example of FIG. 7, the mutual information amount has a maximum value when the standard deviation σ = 0.225. Thus, it is known that the mutual information has one maximum value when the noise intensity is changed.

  Therefore, the optimum noise intensity determination unit 5 selects the noise intensity near the maximum value as the optimum noise intensity in step S41. Therefore, the comparison unit 51 of the optimum noise intensity determination unit 5 includes an evaluation amount (hereinafter referred to as a first evaluation amount) for an analog signal to which noise of a certain intensity is added, and an intensity that is one step lower than the intensity. The evaluation amount (hereinafter referred to as the second evaluation amount) for the analog signal to which the noise is added is compared. Then, in step S42, when the first evaluation amount is lower than the second evaluation amount, the determination unit 52 sets the noise intensity one step lower than the noise intensity corresponding to the second evaluation amount and the second evaluation amount. The noise intensity between the noise intensity levels one step higher than the corresponding noise intensity is determined as the optimum noise intensity. For example, the noise intensity corresponding to the second evaluation amount is determined as the optimum noise intensity.

For example, the optimum noise intensity determination unit 5 obtains the noise intensity that makes the evaluation quantity close to the minimum value or the maximum value by using the first and second derivatives of the evaluation quantity or by probabilistically searching the parameter space. (See, for example, Reference 2).
[Reference 2] Masatoshi Sakawa (Author), “Optimization of Nonlinear Systems”, Morikita Publishing, May 1986

  Further, the evaluation amount calculation unit 3 may calculate the evaluation amount based on data configured by output values output from the signal quantization device. For example, in the case of image data, the degree of deterioration is determined from image data composed of output values by using physical constraints such as the luminance of each pixel changing smoothly except for edges on the two-dimensional plane of the image. It may be obtained as an evaluation amount.

The comparison unit 51 of the optimum noise intensity determination unit 5 compares the evaluation amounts to select the evaluation amount having the lowest evaluation amount, and the determination unit 52 determines the noise intensity corresponding to the lowest evaluation amount as the optimum noise. It may be determined as the intensity of. When using an evaluation quantity that has a higher quantization performance as the value of mutual information or the like is larger as the evaluation quantity, the comparison unit 51 of the optimum noise strength determination unit 5 compares the evaluation quantities and has the highest evaluation quantity. The high evaluation amount is selected, and the determination unit 52 determines the noise intensity corresponding to the highest evaluation amount as the optimum noise intensity.
These modifications can be combined as appropriate.

  When the above configuration is realized by a computer, the processing contents of the functions of each unit of the signal quantization apparatus are described by a program. By executing this program on a computer, the functions of the above-described units are realized on the computer.

  That is, when the CPU sequentially reads and executes the program, the noise adding unit 1, the quantizing unit 2, the evaluation amount calculating unit 3, the optimum noise intensity determining unit 5, the comparing unit 51, the determining unit 52, the comparison selecting unit 53, and Each of the noise intensity comparison units 54 is realized. An auxiliary storage device or memory functions as the storage unit 4.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  As an execution form different from the above-described embodiment, the computer may read the program directly from the portable recording medium and execute processing according to the program. Each time is transferred, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to a computer but has a property that is based on computer processing).

In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.
In addition, the various processes described above are not only executed in time series according to the description, but may be executed in parallel or individually according to the processing capability of the apparatus that executes the processes or as necessary.

  The present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 1 Noise addition part 2 Quantization part 3 Evaluation amount calculation part 4 Storage part 5 Optimal noise intensity determination part 10 Noise addition apparatus 11 Signal quantization apparatus 51 Comparison part 52 Determination part 53 Comparison selection part

Claims (9)

A noise adding unit for adding at least two kinds of noises to the input analog signal;
A quantizing unit that quantizes the analog signal to which each of the noises is added;
An evaluation amount calculation unit for calculating an evaluation amount representing the high performance of the quantization for the analog signal to which each of the noises is added;
An optimum noise intensity determination unit that compares the evaluation quantities and determines the intensity of noise that enhances the quantization performance;
A signal quantizer comprising: The signal quantization apparatus according to claim 1,
The noise adding unit is a unit that sequentially adds high or low noise,
The optimum noise intensity determination unit includes an evaluation amount for an analog signal to which noise of a certain intensity is added (referred to as a first evaluation amount), and an analog to which noise having a level one step lower or higher than that intensity is added. When the first evaluation amount is higher or lower than the evaluation amount (the second evaluation amount) for the signal, it is one step lower than the noise intensity corresponding to the second evaluation amount. Or, it is a unit that determines the noise intensity between the high noise intensity and the noise intensity one step higher or lower than the noise intensity corresponding to the second evaluation amount as the optimum noise intensity.
A signal quantizer characterized by the above. The signal quantization apparatus according to claim 2, wherein
When the optimum noise intensity determination unit determines that the first evaluation amount is higher or lower, the noise adding unit is a noise that is one step lower or higher than the noise intensity corresponding to the second evaluation amount. In the range including the intensity of noise and the intensity of noise that is one step higher or lower than the noise intensity corresponding to the second evaluation amount, noise with successively higher or lower intensity is added in a narrower width than previously applied. ,
The optimum noise intensity determination unit determines the noise intensity corresponding to the newly obtained second evaluation quantity if the newly obtained first evaluation quantity is higher or lower than the newly obtained second evaluation quantity. The noise intensity between the noise intensity one level lower or higher than the noise intensity corresponding to the newly determined second evaluation amount and the noise intensity one level higher or lower than the noise intensity corresponding to the newly determined second evaluation amount is set as the optimum noise intensity. The part to be determined
A signal quantizer characterized by the above. The signal quantization apparatus according to claim 1,
The noise adding unit randomly selects a predetermined number K (K is an arbitrary natural number) at random within a predetermined range, and adds noise of each selected intensity to the input analog signal. Is a part to do,
A signal quantizer characterized by the above. The signal quantization apparatus according to claim 4, wherein
The optimum noise intensity determination unit sets k as an integer equal to or greater than 2, and evaluates an analog signal to which noise having the intensity selected at the kth time is added (hereinafter referred to as a kth evaluation value) and the first time. The evaluation amount for the analog signal to which noise of the selected intensity is added is compared with the evaluation amount selected last time or the evaluation amount selected last time, and the evaluation amount that enhances the quantization performance is obtained from either of the evaluation amounts compared above. The operation of selecting is repeated K-1 times, and the intensity corresponding to the last selected evaluation amount is determined as the optimum noise intensity.
A signal quantizer characterized by the above. A noise adding device for adding each of at least two kinds of intensity noise to an analog signal;
A quantization unit that quantizes the analog signal to which each of the noises has been added; an evaluation amount calculation unit that calculates an evaluation amount that represents the high performance of the quantization for the analog signal to which each of the noises has been added; A signal quantization device including an optimum noise intensity determination unit that compares the evaluation amount and determines the intensity of noise that enhances the quantization performance;
Including signal quantization system. A noise adding step for adding each of at least two types of noise to the input analog signal;
A quantization step for quantizing each of the above-mentioned noise-added analog signals;
An evaluation amount calculating step for calculating an evaluation amount representing the high performance of the quantization for the analog signal to which each of the noises is added;
An optimum noise intensity determination step for comparing the evaluation quantities and determining the intensity of the noise that enhances the quantization performance;
A signal quantization method including:   A signal quantization program for causing a computer to function as each unit of the signal quantization apparatus according to claim 1.   A computer-readable recording medium on which the signal quantization program according to claim 8 is recorded.

Images