JP2893691B2 - Digital signal processor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention will be described in the following order.

A Industrial application field B Outline of the invention C Conventional technology (Fig. 3) D Problems to be solved by the invention (Fig. 3) E Means for solving the problems (Fig. 1) F function ( (Fig. 1) G embodiment (Figs. 1 and 2) (G1) Principle of optimal quantization (G2) First embodiment (Figs. 1 and 2) (G3) Other embodiments H invention BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital signal processing device, and is suitably applied to, for example, a digital signal processing device configured to record, reproduce, and transmit an audio signal or the like with high quality.

B. Summary of the Invention The present invention can reduce a quantization error in a digital signal processing device by quantizing an input signal by selecting an optimal quantization value and a threshold value at predetermined intervals. .

C Conventional technology Conventionally, in this kind of digital signal processing device,
Adaptive predictive coding (APC)
There is a method in which an audio signal is encoded and transmitted by using the method of (1), thereby preventing deterioration of the S / N ratio, intelligibility and the like beforehand and transmitting the signal with high transmission efficiency (Japanese Patent Laid-Open No. 59-1984). JP-A-223033, JP-A-60-223034, JP-A
No. 61-158217, JP-A-61-158218, Japanese Patent Application No. 63
-46595, Japanese Patent Application No. 63-61724, Japanese Patent Application No. 63-61725,
Japanese Patent Application No. 63-65192).

That is, a predetermined prediction filter receives an input digital signal, and obtains a residual signal which is a difference signal between the output signal of the prediction filter and the input digital signal.

At this time, the input digital signal is blocked at predetermined intervals, and a linear prediction analysis (linear pre-
The residual signal is reduced by switching the frequency characteristic of the prediction filter by applying the technique of dictive coding (LPC).

Further, as shown in FIG. 3, each block BL1, BL2,.
..., in response to the maximum value of the residual signal S _Z1 (FIG. 3 (A)), after Bitsutoshifuto the residual signal S _Z1, to transmit data from the most significant bit for example 2 bits.

Accordingly, the quantized values y ₀ , y ₁ , y ₂ , y ₃ (FIG. 3B) are switched for each block, whereby the dynamic range of the data to be transmitted is switched and the residual signal S _Z1 is changed. Requantize.

Thus, by requantizing the residual signal and transmitting the requantized residual signal together with the coefficients of the predictive filter (ie, representing the frequency characteristics of the predictive filter) and the amount of bit shift, the input digital signal can be directly converted. The input digital signal can be transmitted with higher transmission efficiency than in the case of transmission, and the deterioration of the S / N ratio, clarity, and the like can be prevented beforehand.

D Problems to be Solved by the Invention By the way, in this kind of digital signal processing device, an error cannot be avoided at the time of requantization, and therefore, when the signal is decoded on the transmission side, the error component becomes a noise component. Cannot be avoided.

Therefore, if this noise component (hereinafter referred to as requantization error) can be suppressed, it is considered that a desired input digital signal can be transmitted with higher quality.

The present invention has been made in view of the above points, and has as its object to propose a digital signal processing device capable of suppressing a requantization error as compared with the related art.

In the present invention for solving means above problems to solve E problems, a first prediction of filter output data obtained by applying the input signal data S _I to the first prediction of the filter 2 input signal data S _I and prediction error filter means for obtaining a residual data S _Z1 (2,3) of, with the re-quantized data S _L and re-quantizes the residual data S _Z1, the re-quantized data Re-quantizing means 7 for outputting S _L as first processing data, and re-quantizing the difference data between the re-quantized data _SL and the input data of the re-quantizing means 7 via a second predictive filter 11 a requantization error signal generating means (8, 10, 11) which fed back as a requantization error data S _Z2 to the input terminal of the means 7, and blocks of each first period based on the input signal data S _I A first and a second predictive filter 2 for each block, 11 and the predicted filter parameter P _PZ is given.
A linear prediction analysis means 4 for outputting the predicted filter parameter P _PZ as second processing data, and residual data S _Z1
, The residual data S _Z1 is blocked every second period to generate an optimal quantization parameter P _Q for the requantizing means 7 for each block, and to generate the optimal quantization parameter P _Q Optimal quantization analysis means 6 for outputting as third processing data.
Is And the following equation Threshold value _xj and requantization value expressed by the relationship
y _j is generated as an optimal quantization parameter P _Q (where x is a sample value of the residual data S _Z1 to be requantized, p
(X) is the probability distribution of the sample value x, M is the number of requantized values y _j ), so that the requantized error data S _Z2 is minimized, and the first, second, and third The processing result data _SL0 can be decoded based on the processing data.

From the initial threshold value x _j , And sequentially obtaining re-quantized values y _j and threshold values x _j ,
Next formula The convergence condition is examined, and when the convergence condition is satisfied, the following formula y ₁ = y ₁ + Δ... (2D) is repeated, and the operations of the formulas (2A) and (2B) are repeated, so that the optimized By obtaining the quantized value y _j and the threshold value x _j , the re-quantization error data is minimized, and the data that generates the optimum quantization parameter as the third processing data is used.

F action: requantized data S _{L obtained} by requantizing residual data S _Z1 ,
Along with the predicted filter parameter P _PZ given to the first and second predicted filters 2 and 11, a re-quantized value optimized for each predetermined period and an optimal quantization parameter P _Q consisting of a threshold value are obtained. Digital signal processing that can reduce quantization errors that occur at the time of quantization as compared with the prior art, and thus quantize desired input signals with high quality. A device can be obtained.

F action Based on the input signal S _Z1 , the quantization value y _j
And the threshold value x _j , and the quantization value y _j and the threshold value
If the input signal S _Z1 is quantized based on x _j , the quantization error can be suppressed.

G Example (G1) Principle of Optimal Quantization The requantization error ε is x for each sample value of the input signal to be requantized, P (x) for the probability distribution of the sample value x, and a threshold value for requantization. Let x _j be the requantized value and y _j be It is represented by

Here, M represents the number of requantization values y _j .

Therefore, if the equation (1) is differentiated by y _j and x _j and set to 0, the following equation is obtained. Is obtained.

Therefore, if the re-quantization threshold x _j and the re-quantization value y _j are selected so as to satisfy the relations of equations (2) and (3) for each block, the re-quantization error ε Can be re-quantized so as to be the minimum value (hereinafter referred to as optimal quantization).

(G2) in FIG. 1 a first embodiment, 1 indicates a digital signal processing device as a whole, based on the principle of the optimum quantization described above, and re-threshold x _j requantization for each block Select the quantization value y _j .

That is, the digital signal processing device 1 supplies a 16-bit input digital audio signal S _I to the prediction filter 2, and outputs a difference signal between the output signal of the prediction filter 2 and the input digital audio signal S _I via the subtractor 3. _Is obtained.

This LPC analysis circuit 4 with respect is to block the input digital audio signal S _I for each section of 20 [msec],
By applying a linear prediction analysis technique to each block, a prediction filter parameter P _PZ for switching the frequency characteristics of the prediction filter 2 is output.

Thus, as the maximum value of the residual signal S _Z1 for each block is minimized, are made as a frequency characteristic of the predictive of filter 2 is switched.

Optimal quantizer analyzer 6 on the other hand, for each block of the residual signal S _Z1, Lloyd shown in FIG. 2 Matsukusu (LIOY
(D-MAX) method, thereby selecting a threshold value x _j and a re-quantized value y _j for re-quantizing the 16-bit residual signal S _Z1 to 2 bits. .

That is, the optimal quantization analyzer 6 calculates the threshold value of the requantization.
After initializing x _j (j = 0, 1, 2, 3, 4), the process moves from step SP1 to step SP2, sets the first internal counter k to a value of 1, and then moves to step SP3. Set the second internal counter j to the value one.

Here, the first internal counter k is used to count the number of repetitions of the processing procedure, and the second internal counter j is used to count the number of re-quantization thresholds x _j. Can be

Subsequently, the optimal quantization analyzer 6 proceeds to step SP4 to set the first internal register Y to a value of 0, and sets the third internal counter c to a value of 0. 4 is set to a value 0.

Here fourth internal counter i of is used to count the number of samples in one block constituting the residual signal S _Z1.

Subsequently, the optimal quantization analyzer 6 proceeds to step SP6, where the sample value x _(i) (in this case, i = 0 _{) of} the input residual signal _SZ1 is changed from the threshold value _xj-1. during the threshold x _j (in this case becomes the range of the threshold x ₁ from the threshold x _0, threshold x ₀ = -∞, consisting of x ₄ = ∞) it is determined whether or not located in the If a negative result is obtained here, the process proceeds to step SP7, where the fourth internal counter i is incremented by one.

On the other hand, if a positive result is obtained in step SP6, the process proceeds to step SP8, where the first internal register Y is set.
In addition to adding the sample values x _(i), increments the third internal counter c by the value 1, the following step SP7
Move on to

Thus the third internal count c counts the number of sample values x _(i) located between the threshold x _j-1 of the threshold x _j.

Subsequently, the optimal quantization analyzer 6 proceeds to step SP9, and determines whether or not the value of the fourth internal counter i matches the number N of samples in the block. In this case, a negative result is obtained, and step SP6 is performed. Return to

Accordingly, the optimal quantization analyzer 6 determines in step SP6 that the sample value x _(i) (i = 0)
For _{(i) (i = 1)} , it is determined whether or not the position of the threshold x ₀ in a range of threshold x _1.

Here, if a positive result is obtained, the result is added to the first internal register Y and the third internal counter c is incremented in step SP8, whereas if a negative result is obtained, the process proceeds to step SP9 via step SP7. .

Thus, by repeating the sequential step SP6-SP7-SP9-SP6 or step SP6-SP8-SP7-SP9- SP6 processing loop LOOP1, the sample value x is located in the range of the threshold x ₁ from the threshold x _{0 (i)} are sequentially added to the first internal register Y, and the number of sample values x _(i) added is stored in the third internal counter c.

Further, every time the processing loop LOOP1 is repeated,
Since the value of the fourth internal counter i is incremented by 1 each time, if the process is repeated N times in total, step SP7 is executed.
Sets the value of the fourth internal counter i to the value N.

Therefore, since the number of samples in the block is the value N, when the processing loop LOOP1 is repeated for all the sample values x _(i) in the block, a positive result is obtained in step SP9, and the optimal quantization analysis is performed. Instrument 6 is step SP10
Move on to

Thus, by repeating the processing loop LOOP1, the following equation is obtained. Is executed.

Subsequently, in step SP10, the optimal quantization analyzer 6
Based on the value of the first internal register Y and the value of the third internal counter c, In selecting the first re-quantized values y ₁ represented.

Thus, after obtaining the sum of the sample values x _(i) is located in the range of the threshold x ₁ from the threshold x _0, can set the average value to the re-quantized value y _1, thereby in the range from the threshold x ₀ of the threshold x _1, the re-quantization error ε can be selected requantization values y ₁ to a minimum.

After storing the _first requantized value y1 in a predetermined register circuit, the optimum quantization analyzer 6 subsequently proceeds to step SP11 and increments the value of the second internal counter j by the value 1.

Subsequently, moves connexion to step SP 12, whether the value of the second internal counter j matches the number of requantization values y _j M (made by M = 4 since the re-quantization in this case 2 bits) In this case, a negative result is obtained and the process returns to step SP4.

Here, the optimal quantization analyzer 6 subsequently sets the first internal register Y and the third internal counter c to a value of 0,
Step SP5 to move connexion, a fourth of the internal counter i is set to 0, followed by transfer to Tetsupu SP6 in connexion, sample values x of the input residual signal _{S Z1 (i) (i =} 0), the teeth from threshold x ₁ determines whether located within the range of threshold x _2.

Thus, the processing loop LPPO1 is repeated again, thereby similarly to the case of selecting the first re-quantized values y ₁ between the threshold x ₁ from the threshold x _0, threshold x ₁ mustard Threshold
Sample values are summed with sample values x between x _{2 (i)} is stored in the first internal register Y is sequentially added x _(i)
Is stored in the third internal counter c.

When the processing loop LOOP1 is repeated for all the sample values x _(i) in the block, the optimal quantization analyzer 6 proceeds to step SP10 and It represented the second selected requantization values y ₂ is in, thereby selecting the re-quantized value y ₂ for the re-quantization error ε to the minimum between the threshold x ₁ threshold x ₂ I do.

Subsequently, the optimum quantization analyzer 6 increments the second internal counter j in step SP11, and then, proceeds to step SP12.
Then, it is determined whether or not the value of the second internal counter j matches the number M of the requantized values y _j .

Thus, steps SP12-SP4-SP5-LOOP1-SP10-S
By repeating the processing loop LOOP2 of P11-SP 12, sequentially threshold x ₁ from the threshold x _0, the threshold from the threshold x ₁
x _2, threshold x ₂ from the threshold x _3, threshold x ₃ from the threshold x requantization values requantization error ε to the minimum between ₄ y _j (j
= 1, 2, 3, 4), whereby the requantized value y _j represented by the general expression of the expression (3) can be selected.

On the other hand, when the requantized value y _j is selected between all the threshold values x _j , the value of the second internal counter j is incremented to the value M in step SP11.

Therefore, in this case, a positive result is obtained in step SP12, and the optimum quantization analyzer 6 proceeds to step SP13.

Here, the optimal quantization analyzer 6 sets the second internal counter j to the value 1 again, and then proceeds to step SP14, where the requantized values y _j and y _{j + 1} (in this case, j = 1) Therefore, for the requantized values y ₁ and y ₂ ), Selecting a second threshold x ₁ represented by the relationship.

Thus, it is possible to select the second threshold value x ₁ which is represented by the general formula (2) between the re-quantized value y ₁ and y _2,
The requantization error ε can be selected second threshold x ₁ to minimize between requantization values y ₁ and y _2.

Subsequently, the optimal quantization analyzer 6 proceeds to step SP15 and increments the value of the second internal counter j by the value 1, and then proceeds to step SP16 to change the value of the second internal counter j to the requantized value y. _{It is} determined whether or not the number matches _j (M = 4).

In this case, a negative result is obtained in step SP16, and the optimal quantization analyzer 6 returns to step SP14.

Thus, after the second internal counter j is incremented, by repeating the step SP14, the second
Of As with the threshold value x _1, third threshold x ₂ for the re-quantization error ε to the minimum between requantization values y ₂ and y ₃ are selected.

Accordance connexion, step SP14-SP15-SP16-SP14 by repeating the processing loop LOOP3 of requantization values from the re-quantized value y ₁ y _2, requantization values y ₂ from the re-quantized values y _3, requantization Value
from y ₃ requantization values y ₄ between the requantization error ε respectively can be selected threshold x _{j (j} = 1,2,3) be minimized.

On the other hand, when the threshold value x _j is selected between all the requantized values y _j , the value of the second internal counter j is incremented to the value M in step SP15, whereby the positive result is obtained in step SP16. Are obtained, and the optimal quantization analyzer 6 proceeds to step SP17.

Here, the optimal quantization analyzer 6 has a first internal counter k
Is incremented by 1 and the process proceeds to step SP18 to determine whether the value of the first internal counter k matches the predetermined value 5.

In this case, a negative result is obtained in step SP18, and the optimal quantization analyzer 6 returns to step SP3.

Therefore, the optimal quantization analyzer 6 repeats the processing procedure of step SP3 and thereafter repeats the processing loop LOOP2, thereby replacing the initialized threshold with the processing loop LOOP.
OP3 between repeatedly selected threshold x _j a, selecting a re-quantized values y _j that minimizes the re-quantization error ε again.

Subsequently, when the re-quantized value y _j is re-selected, the optimal quantization analyzer 6 executes the processing procedure of step SP13 and then repeats the processing loop LOOP3, thereby re-selecting the re-quantized value y _j during the re-quantization error ε reconfigure the threshold x _j that minimizes, proceeds to step SP17.

Thus, steps SP18-SP3-LOOP2-SP13-LOOP3 are performed until the value of the first internal counter k matches the predetermined value 5.
Processing loop LOOP4 of -SP17-SP18 is repeated, thereby as to converge the requantization error of the entire re-quantization process, re-quantized value y _j and threshold x _j are selected alternately.

On the other hand, if the processing loop LOOP4 is repeated five times, a positive result is obtained in step SP18, and the optimal quantization analyzer 6 proceeds to step SP19 and ends the processing procedure, thereby optimizing the requantization. A threshold value y _j and a threshold value x _j are selected.

Actually, if the processing loop LOOP4 is repeated five times, the requantization error of the entire requantization process can be optimized within a necessary and sufficient range. Can be reduced.

Optimal quantizer analyzer 6 further includes, in place of the initialized threshold in subsequent block, using the optimized threshold x _j in the previous block, it has been made to repeat the processing procedure .

In this case, in the audio signal, there is a correlation between each block, and the threshold value x _j optimized for all the blocks is used.
Is used as an initial value, an optimized requantized value y _j and threshold value x _j can be obtained even when the number of repetitions of the processing loop LOOP4 is as small as about five.

Accordance connexion that amount, since it is possible to obtain an optimized re-quantized values y _j and threshold x _j in a short time as a whole, requantization process can be simplified.

Further optimization quantizer analyzer 6, for each block, optimized re-quantized values y _j and the quantizer 7 and the inverse requantizer threshold x _j as the optimal quantization parameter P _Q 8 And to the inverse requantizer 9 to be transmitted.

The re-quantizer 7 re-quantizes the residual signal S _Z1 with the optimized re-quantized value y _j and threshold value x _j based on the optimal quantization parameter P _Q , and obtains the resulting two bits. and outputs the re-quantized signal S _L of the transmission target.

In contrast, the inverse requantizer 8 decodes the re-quantized signal S _L based on the optimal quantization parameter P _Q.

The subtracter 10 outputs the output signal of the inverse requantizer 8 and the residual signal.
By obtaining the difference signal of S _Z1 , the re-quantization error signal S _Z2 is obtained, and the re-quantization error signal S _Z2 is output to the prediction filter 11.

The prediction filter 11 has a prediction filter parameter P _PZ
Depending on the switches to the same frequency characteristics and prediction of filter 2, after the frequency characteristic of the requantization換誤difference signal S _Z2 corrected, adapted to return to the re-quantizer 7 via the subtractor 12 I have.

As a result, the spectrum shape of the requantization error signal _SZ2 is corrected to a flat spectrum shape with the minimum energy.

In transmission subject contrast, and decoding the re-quantized signal S _L at opposite requantizer 9, thereby the two bits of the re-quantized signal S _L requantization previous residual signal S _Z1 After decoding, the signal is output to the prediction filter 15 via the subtractor 14.

The predictive filter 15 has a predictive filter parameter P _PZ
Is switched to the same frequency characteristic as that of the prediction filter 2 and the output signal of the inverse requantizer 9 is fed back to the subtractor 14 so that the decoded residual signal S _Z1 is converted to the digital audio signal S Decode to _L0 .

Thus, for the transmission target, a 2-bit requantized signal
By transmitting the S _L sequentially and transmitting the prediction filter parameter P _PZ and the optimum quantization parameter P _Q for each block, the 16-bit input digital audio signal
It can transmit S _I.

In this embodiment, the requantizer 7, the inverse quantizer 8, the subtracters 10 and 12, and the prediction filter 11 convert the input residual signal S _Z1 into an optimized requantized value y _j. And threshold x
Construct a quantizer that quantizes with _j .

According to the above configuration, to obtain a re-quantized value y _j and threshold x _j optimized by applying the technique of Lloyd Matsukusu, the optimized re-quantized values y _j and the threshold x _j is the residual signal S
By re-quantizing the _Z1, resulting suppresses the requantization errors for each block, thus it is possible to transmit the input digital audio signals S ₁ at a higher than conventional quality.

(G3) Other Embodiments In the embodiments described above, the case where the requantized value y _j and the threshold value x _j optimized by applying the Lloyd-Mux method are described. Not limited to this, for example, the method of Matsukusu (MAX),
Various techniques such as the technique of R-DITE and the technique of Smith (SMITH) can be widely applied.

In this case, for example, when the Matsukus method is applied, the following expression is obtained from the initial threshold value x _j And sequentially obtaining re-quantized values y _j and threshold values x _j ,
Next formula The convergence condition is examined, and when the convergence condition is satisfied, the following expression y ₁ = y ₁ + Δ (11) is performed, and the expressions (8) and (9) are repeated.
An optimized requantization value y _j and threshold value x _j can be obtained.

 Here, δ and Δ are appropriate minute amounts.

Further, in the aforementioned embodiments, 16 has described the case where the input digital audio signal S _I of the bit and transmits the re-quantized to requantization signal 2 bits, the present invention is not limited thereto, various bit amount The present invention can be widely applied to a case where various input digital signals are requantized into a requantized signal having a desired bit.

Further, in the above-described embodiment, a case has been described in which the requantization value y _j and the threshold value x _j are sequentially selected so as to converge the error. However, the present invention is not limited to this. leave selected requantization values y _j and the threshold x _j of, be selected requantization values y _j and the threshold x _j requantization error re is also reduced from this Good.

Further, in the above embodiment, LPC
The case where the requantized value y _j and the threshold value x _j are selected for each block with respect to the analyzed residual signal S _Z1 has been described, but the present invention is not limited thereto, and the residual signal is independent of the LPC analysis. to block the signal may be selected to re-quantized value y _j and the threshold x _j.

Further, in the above-described embodiment, the case of transmitting a digital signal has been described. However, the present invention is not limited to the case of transmitting a digital signal, but can be widely applied to the case of recording and reproducing a desired digital signal.

H Advantageous Effects of the Invention As described above, according to the present invention, the re-quantized data obtained by re-quantizing the residual data, the predicted filter parameters given to the first and second predicted filters, By obtaining, as transmission data, a requantization value optimized in accordance with the above and an optimal quantization parameter consisting of a threshold value, it is possible to reduce a quantization error generated at the time of quantization as compared with the related art. Thus, a digital signal processing device capable of quantizing a desired input signal with high quality can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a digital signal processing apparatus according to the present invention, FIG. 2 is a flowchart for explaining its operation, and FIG. 3 is a signal waveform diagram for explaining conventional requantization. DESCRIPTION OF SYMBOLS 1 ... Digital signal processing apparatus, 2, 11, 15 ... Predictive filter, 4 ... LPC analysis circuit, 6 ... Optimal quantization analyzer, 7 ... Requantizer, 8, 9 ... Reverse Quantizer.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁶ , DB name) H03M 7/32-7/38

Claims (2)

(57) [Claims] 1. A prediction error filter means for obtaining residual data between output data of a first predictive filter obtained by providing input signal data to a first predictive filter and said input signal data; Requantize the difference data to get requantized data,
Re-quantizing means for outputting the re-quantized data as first processing data; and difference data between the re-quantized data and the input data of the re-quantizing means through the second predictive filter. A requantization error signal generating means for feeding back as requantization error data to an input terminal of the quantization means; and a block for each first period based on the input signal data, and the first and second blocks for each block. A linear prediction analysis means for providing a predicted filter parameter to the predicted filter and outputting the predicted filter parameter as second processing data; and converting the residual data to a second , And an optimal quantization parameter for the requantization means is generated for each block, and the optimal quantization parameter is Comprising an optimum quantization analyzing means for outputting a third process data, the optimum quantizing analysis means, the following equation And the following equation Threshold value _xj and requantization value expressed by the relationship
y _j is generated as the optimal quantization parameter (where x is a sample value of the residual data to be requantized, p
(X) is the probability distribution of the sample value x, M is the number of requantized values y _j ) so that the requantized error data is minimized, and the first, second, and third A digital signal processing device characterized in that processing result data can be decoded based on processing data. 2. A prediction error filter means for obtaining residual data between output data of the first prediction filter obtained by providing input signal data to the first prediction filter and said input signal data; Requantize the difference data to get requantized data,
Re-quantizing means for outputting the re-quantized data as first processing data; and difference data between the re-quantized data and the input data of the re-quantizing means through the second predictive filter. A requantization error signal generating means for feeding back as requantization error data to an input terminal of the quantization means; and a block for each first period based on the input signal data, and the first and second blocks for each block. A linear prediction analysis means for providing a predicted filter parameter to the predicted filter and outputting the predicted filter parameter as second processing data; and converting the residual data to a second , And an optimal quantization parameter for the requantization means is generated for each block, and the optimal quantization parameter is Comprising an optimum quantization analyzing means for outputting a third process data, the optimum quantizing analysis means, threshold initial value of
From x _j , the following equation And sequentially obtaining re-quantized values y _j and threshold values x _j ,
Next formula The convergence condition is examined, and when the convergence condition is satisfied, the following formula y ₁ = y ₁ + Δ... (2D) is repeated, and the operations of the formulas (2A) and (2B) are repeated, so that the optimized By obtaining the quantized value y _j and the threshold value x _j , the re-quantization error data is minimized, and the processing result data is calculated based on the first, second, and third processing data. A digital signal processing device characterized in that decoding is possible.