JPH01221021A - Digital signal processing unit - Google Patents

Abstract

PURPOSE:To approach the shape of spectrum of quantized noise to the spectrum shape of an audio signal by weighting an output signal of an inverse requantizer and an input signal to a requantizer respectively and outputting the result to a prediction filter. CONSTITUTION:A subtractor 51 is used to obtain a quantization error signal SE1 being a difference between an input signal to a requantizer 21 and an output signal of an inverse requantizer 22. Moreover, after the quantization error signal SE1 is weighted by gamma(0<gamma<1) via a multiplier 52, the result is added to the input signal of the requantizer 21 via the multiplier 53 and fed back to the adder 23 and the prediction filter 24. Thus, the spectrum shape LNS of the requantizer noise is made close to the spectrum shape LSS of the audio signal in response to the weight coefficient gamma to obtain a function of noise shaping, then the signal versus quantization noise ratio (SNR) in the listening sense is improved by utilizing the masking effect.

Description

[Detailed description of the invention] The present invention will be explained in the following order.

A. Industrial field of application B. Overview of the invention C. Conventional technology (Fig. 7, Fig. 14 to Fig. 17) D. Problems to be solved by the invention (Fig. 7, Fig. 14 to Fig. 17) E Means to solve the problem (Fig. 1) F action (Fig. 1)
Figure) G Example (Figures 1 to 13) (G1) First Example (Figures 1 to 7) (G2) Second
Examples (Figures 4 to 6, Figures 8 to 12) (G3) Third example (Figure 13) (G4) Fourth example H Effect of the invention A Industrial application FIELD The present invention relates to a digital signal processing device, and is suitable for application to, for example, a digital signal processing device designed to record, reproduce, and transmit audio signals and the like with high quality.

B. Summary of the Invention The present invention provides a digital signal processing device in which the input signal of a requantizer and the output signal of an inverse quantizer are each weighted by a predetermined amount and fed back to a prediction filter, thereby improving overall performance. Simple configuration and short processing time.
A feedback type digital signal processing device having a noise shaving function can be obtained.

C. Prior Art Conventionally, in this type of digital signal processing device, an adaptive predictive coding method has been used.
vecoding: By encoding and transmitting the audio signal using the APC method, the S/N
There is a method that prevents deterioration of ratio, clarity, etc. and transmits data with high transmission efficiency (Japanese Patent Laid-Open No. 59-223
033, JP 60-223034, JP 61-158217, JP 61-15821
Publication No. 8).

In FIGS. 14 and 15, 1 and 2 respectively indicate feedforward type and feedback type digital signal processing devices configured to transmit an input digital signal SI using an adaptive predictive coding method. S, is subjected to linear predictive analysis (linear pred
Active coding: requantize using the LPC method.

That is, in the feedforward type digital signal processing device 1, by providing the input digital signal SI to the adder 4 via the prediction filter 3, the difference signal between the input digital signal SI and the output signal of the prediction filter 3 is obtained. A residual signal SzI is obtained.

The prediction filter 3 takes advantage of the fact that the audio signal, which is a time function, has a correlation not only between adjacent sampling points but also between sampling points that are a certain distance apart.
The input digital signal SI is divided into predetermined periods, the characteristics of the input digital signal S in each period are detected (that is, linear predictive analysis is performed), and the filter characteristics are switched based on the characteristics.

Therefore, via the adder 4, a residual signal SKI consisting of the linear prediction residual of the input digital signal S1 for the feature can be obtained.

Further, in the digital signal processing device 1, the residual signal Sol is passed through the subtracter 5 and the requantizer 7 to the transmission line Ll.
At the same time, the output signal is given to the subtracter 8 along with the input signal of the requantizer 7 via the inverse quantizer 9 having the inverse characteristics of the requantizer 7, and the resulting difference is The signal S2□ is input to the subtracter 5 via a prediction filter 10 having the same characteristics as the prediction filter 3.

Therefore, in the transmission line Ll, the input digital signal S1
The residual signal Sz+, which is the linear prediction residual of be able to.

Therefore, on the receiving side, by decoding the transmission signal SLI using a prediction filter 14 having the same characteristics as the prediction filter 3, an inverse quantizer 15 and an adder 16 having the same characteristics as the inverse quantizer 9, Input digital signal S with high transmission efficiency
1 can be transmitted.

On the other hand, in the feedback type digital signal processing device 2, the input digital signal S1 is
0 and is outputted via the requantizer 21, and the output signal SLI is given to the prediction filter 24 via the inverse quantizer 22 and adder 23.

Furthermore, the output signal of the prediction filter 24 is added to the subtracter 2
0 and to the adder 23, a feedback loop is formed in the prediction filter 24, and the transmission signal SL is fed back through the prediction filter 24, thereby converting the input digital signal Sl into the adaptive predictive coding method. It is encoded using the following method.

(Thus, the feedback type digital signal processing device 2 can also transmit the input digital signal S1 with high transmission efficiency like the feedforward type digital signal processing device 1.

Specifically, as shown in FIG. 16 using a feedback type as an example, in the digital signal processing device 30,
Applying the input digital signal Sl to the linear predictive analyzer 31,
The spectrum shape of the input digital signal S is detected at predetermined intervals.

The linear predictive analyzer 31 outputs a predictive filter parameter signal S, which is a switching signal for the coefficients of the predictive filters 32 and 33, based on the detection result, so that the linear predictive analyzer 31 can perform The encoding of the transmission signal St+ is, for example, straight P CM (p
ulse code a+odulati.

n) The encoding system is designed to switch between summation PCM and difference PCM to a coding system with high compression efficiency.

On the other hand, the maximum value detector 34 outputs the output signal of the prediction filter 32 obtained via the subtracter 35 and the residual signal S2. of the input digital signal S1. is received, its maximum value is detected and output to the floating coefficient detector 36.

The floating coefficient detector 36 outputs a floating coefficient signal S based on the output signal of the maximum value detector 34 to a multiplier 37 provided between the subtracter 20 and the requantizer 21.
The input signal thus corrected to a predetermined dynamic range is input to the requantizer 21.

Further, a multiplier 38 having characteristics opposite to those of the multiplier 37 is provided between the requantizer 21 and the adder 23, and the input signal of the requantizer 21 is floating in the multiplier 37, and the input signal of the requantizer 21 is re-re-quantized. The output signal of the quantizer 21 is floated.

In this way, the prediction filter parameter signal SP and the floating coefficient signal S are transmitted together with the transmission signal SLI, and the prediction filter 33 and the multiplier 38 are respectively transmitted on the receiving side.
By decoding using the prediction filter 40 and multiplier 39 having the same characteristics as the above, the input digital signal S1 can be compressed and transmitted.

By the way, in this type of digital signal processing device, it is inevitable to generate quantization noise when requantizing on the transmitting side (hereinafter referred to as requantization noise), so noise shaving techniques are used to avoid the occurrence of quantization noise. Auditory signal to quantized sound ratio (S
IEEE TRANSACTIONS ON ACO
USTICS, 5PEECH, AND 5IGNAL
PROCESSING, VOL, ASSP-27, NO
, 3, JIINE 1979, Journal of the Institute of Electronics, Information and Communication Engineers 4
/'87 VOL, 70. NO, 4 pages 392-400,
JP-A-59-223032, JP-A-60-103
746, JP-A-61-158220).

That is, as shown in FIG. 17 corresponding to FIG. 14,
A noise filter 45 is used in place of the prediction filter 10 to generate a difference signal between the input signal of the requantizer 7 output from the subtracter 8 and the output signal of the inverse quantizer 9 (i.e., during requantization). By feeding back Szz, which is a requantization error signal of , thereby approximating the spectral shape of the requantization noise to the spectral shape of the audio signal.

That is, let the frequency characteristics of the prediction filter 3 and the noise filter 45 be P (z) and F (z), respectively, and let the flat frequency characteristics be Δ, and 2 can be expressed as follows: z - exp (jωt) . ...(
1), the frequency characteristic S3 of the audio signal consisting of the input digital signal S can be expressed by the following relational expression, and the frequency characteristic N of the quantization noise when requantized is as follows. It can be expressed by the relational expression:

Therefore, the frequency characteristic F (z) of the noise filter 45 is
For the frequency characteristic P (z) of the prediction filter 3, using an arbitrary constant α, the following formula, F (z) = P (z / α) ...... (4
), it can be substituted into equation (3) and expressed as the following equation.

That is, as shown in FIG. 7, the spectral shape LNS of the requantization noise can be brought closer to the spectral shape L S s of the audio signal according to the value of the constant α,
This makes it possible to improve the signal-to-quantization-noise ratio (SNR) by utilizing the auditory masking effect.

Therefore, since the signal to quantization noise ratio is improved, the input digital signal SI can be further compressed and transmitted.

D Problems to be Solved by the Invention Incidentally, since the feedback type digital signal processing device 2 can be constructed using only one prediction filter, it has problems compared to the feedforward type digital signal processing device 1. This feature has the advantage of simplifying the overall configuration.

Therefore, if the noise shaving method is applied to the feedback type digital signal processing device 2, just as the noise shaving method is applied to the feedforward type digital signal processing device 1, the overall configuration of the digital signal processing device can be simplified. It is thought that it is possible to obtain

However, when applying the noise shaving method to the feedback type digital signal processing device 2, it becomes necessary to switch the characteristics of the noise filter in a complicated manner according to the input digital signal S1 at the same time as the prediction filter. There was a problem that the configuration became complicated and the processing time became long.

The present invention has been made in consideration of the above points, and is an attempt to propose a feedback type digital signal processing device that has an overall simple configuration, short processing time, and has a noise shaving function. be.

Means for Solving Problem E In order to solve this problem, in the present invention, a prediction filter 24 equipped with a feedback loop 23 and an input signal SI
and the output signal of the prediction filter 24, and the requantization means 21 requantizes and outputs the difference signal of the output signal of the prediction filter 24, and the requantization means 21 requantizes and predicts the output signal of the requantization means 21 with the inverse characteristic of the requantization means 21. In a digital signal processing device 50 equipped with an inverse quantization means 22 that outputs to a quantization filter 24, the output signal output from the inverse quantization means 22 to the prediction filter 24 is converted into an input signal of the requantization means 21. In addition, each of the signals is weighted by a predetermined amount and output to the prediction filter 24.

If the output signal outputted from the F-action inverse quantization means 22 to the prediction filter 24 is weighted by a predetermined amount together with the input signal of the requantization means 21, and then output to the prediction filter 24, the overall process can be simplified. With this configuration, the spectral shape of the quantization noise can be brought close to the spectral shape of the input signal SI.

G Embodiment (G1) In FIG. 1, in which parts corresponding to those in the first embodiment shown in FIG. The subtracter 51 is used to obtain a quantization error signal Set consisting of a difference signal between the input signal of the requantizer 21 and the output signal of the inverse quantizer 22.

Further, in the digital signal processing device 50, the quantization error signal si+ is converted to the value r (0<r
After weighting by <1), the signal is added to the input signal of the requantizer 21 via the adder 53 and fed back to the adder 23 and the prediction filter 24.

Therefore, in the adder 23, the quantization error signal S! is weighted by the value (1-γ) in addition to the output signal of the inverse quantizer 22! I is input, and the input signal of the requantizer 21 and the output signal of the inverse quantizer 22 are each weighted by a predetermined amount and fed back to the requantizer 21 via the prediction filter 24. .

Therefore, as shown in FIG. 2, the prediction filter 24 and adder 23 with the frequency characteristic P (
z) / (1-P(z)) equivalent filter 54
It can be expressed as an equivalent circuit by replacing it with .

Further, as shown in FIG. 3, if the input signals of the adder 53 are separated and represented in an equivalent circuit, the filter 54 and the adder 23 have a frequency characteristic P (z) /'(
1-P (z) ) filter 55 and frequency characteristics 1/
It can be replaced with a filter 56 and an adder 57 of (1-P (z)).

Therefore, as shown in FIG.
2, the frequency characteristic rP (z) / (1-P(2))
It can be expressed by replacing the adder 57 with the filter 59 and the subtracter 58.

Furthermore, as shown in FIGS. 5 and 6, a filter 56
If the subtracter 20 and 20 are combined into one filter and represented in an equivalent circuit, it can be replaced with a filter 60 with frequency characteristics (1-P(z)), and the filter 60 can be replaced with the subtracter 58.
When moved to the input side of and expressed in an equivalent circuit, the frequency characteristic γ
P (z) / (1-P (z)) filter 59
can be represented by an equivalent circuit having the same configuration as the configuration shown in FIG. 17 by replacing the filter 61 with the frequency characteristic γP (z).

Therefore, in equation (3), the frequency characteristic F of the noise filter 45 is expressed as
(z), the frequency characteristic γP (
By substituting z), it is possible to obtain the frequency characteristic N of the requantization noise expressed by the following relational expression.

Therefore, as shown in FIG. 7, the weighting is applied to the spectral shape LN of the requantization noise and the spectral shape LS of the audio signal according to the value of the coefficient γ.

The noise shaving function can be obtained by approaching the quantization noise ratio (SNR) by utilizing the masking effect.

In this way, a simple configuration is possible in which the quantization error signal si+ is weighted and fed back to the prediction filter 24 together with the input signal of the requantizer 21, without using a noise filter whose operation is complicatedly switched together with the prediction filter 24. It is possible to obtain a noise shaving function with this, simplifying the overall configuration and shortening the processing time accordingly.

Incidentally, when the weighting coefficient γ is set to the value 1, only the output signal of the inverse quantizer 22 can be fed back to the prediction filter 24, which is different from the conventional feedback type that does not have a noise shaving function. A digital signal processing device can be obtained.

In the configuration shown in FIG. 1, the quantization error signal S□ obtained via the subtracter 51 is weighted by the value γ via the multiplier 52, and then input to the requantizer 21 via the adder 53. The signal is output to the prediction filter 24 together with the signal, and the output signal of the inverse quantizer 22 and the input signal of the requantizer 21 are weighted by a predetermined amount via the prediction filter 24 and then output to the requantizer 21. will be returned to.

According to the above configuration, by feeding back the input signal of the requantizer 21 and the output signal of the inverse quantizer 22, each weighted by a predetermined amount, to the prediction filter 24, the spectral shape of the requantization noise is LN can be brought close to the spectral shape LSs of the audio signal, and the input digital signal S1 can be compressed and transmitted.

In this way, it is possible to obtain a feedback type digital signal processing device which has a simple configuration as a whole, has a short processing time, and is equipped with a noise shaving function.

(G2) Second Embodiment In FIG. 8, 65 indicates a feedback type digital signal processing device as a whole, which brings the spectral shape of the requantization noise close to the spectral shape of the audio signal and on the high frequency side. Equipped with noise shaving function to enhance noise.

In fact, in the frequency band above 9 (kHz),
It is known that the auditory sensation becomes duller than in the frequency band below this, and this can be used to emphasize requantization noise on the high frequency side and suppress the requantization noise on the lower frequency side. In other words, the perceptual signal-to-quantization noise ratio can be further improved.

However, in a feedback type digital signal processing device, simply making the spectrum shape of the requantization noise close to the spectrum shape of the audio signal complicates the overall configuration. If the spectrum shape is emphasized on the higher frequency side, there is a problem in that the configuration becomes significantly more complicated.

Therefore, in this embodiment, the quantization error signal S□ obtained via the subtracter 51 is converted to the value γt via the multiplier 66.
After weighting by (0<rt=1), the subtractor 6
The value γ2 obtained through the noise filter 68 using
A difference signal is obtained from the quantization error signal S□ weighted by the quantization error signal S□.

The multiplier 69 converts the difference signal output from the subtracter 67 into a value γ
The quantization error signal S□, which is weighted by l (0<Tt = 1) and output to the adder 70, and thereby weighted by a predetermined amount via the adder 70, and its output signal from the noise filter 68; The input signal of the requantizer 21 is outputted to an adder 23 and a prediction filter 24.

Further, an adder 71 is provided between the subtracter 20 and the prediction filter 24, so that the output signal of the noise filter 68 is output.

Therefore, the input signal of the requantizer 21 and the inverse quantizer 2
The output signal of No. 2 is weighted by a predetermined amount and input to the prediction filter 24, and the quantization error signal is weighted by a predetermined amount and output to the requantizer 21 and the prediction filter 24 via the noise filter 68. It is designed so that

Therefore, as shown in FIG. 9, the weighting of the multipliers 66 and 69 is expressed in an equivalent circuit by setting the coefficients γ2 and γ to values 1 and β, respectively. The frequency characteristic P (z) / (
1-P(z)).

Furthermore, as shown in FIG. 10, if the input signal input to the adder 70 via the noise filter 68 is separated and represented in an equivalent circuit, the filter 72 and the adder 70 have a frequency characteristic P (2) / Filter 7 of (1-P(2))
3. Filter 7 with frequency characteristic 1/(1-P(z))
4 and adders 75 and 76.

Furthermore, as shown in FIG.
7 input signals are separated and represented in an equivalent circuit, the subtracter 67, noise filter 68, multiplier 69, and adder 76 have the same frequency characteristics F (z
), and multipliers 69A and 69B with the same weighting as the multiplier 69 have a coefficient β.
1 filter 73 and the same frequency characteristic P (Z) / (
1-P (z)) filters 73A, 73B and 7
It can be replaced with a 3C1 adder 77 and a subtracter 78.

Therefore, as shown in FIG. 4, if the quantization error signal SZ+ is summarized and expressed in an equivalent circuit, the filters 68A, 68
B, 73B, 73C and 74, multipliers 69A and 69B
, the adders 75 and 77, the subtracter 78, the subtracter 58 and the frequency characteristic F+(z) can be replaced with a filter 80 expressed by the following relational expression: (7) .

Therefore, as in the case of the first embodiment, as shown in FIGS. 5 and 6, the filter 73A and the subtracter 20 are combined into one filter, and the frequency characteristic (1-P (z
)) is replaced with the filter 60 of the subtractor 5.
8 to the input side and represent it in an equivalent circuit, the frequency characteristic F
+(z) filter 80 is expressed by the following formula:
...(8) Filter 8 with frequency characteristic Fz(z) expressed by the relational expression
It can be replaced with 1.

Therefore, by substituting the frequency characteristic Fz(z) of equation (8) into the frequency characteristic F (z) of equation (3), the following equation N! The frequency characteristic N of the requantization noise can be expressed by the relational expression 1-P(ZJ (9).

Here, as shown in Fig. 12, while the value Δ represents a flat frequency characteristic represented by the straight line L7, the frequency represented by the equation (1-F (z) ) on the right side of equation (9) By selecting the frequency characteristic F (z) to be a predetermined frequency characteristic, the characteristic can be expressed as a frequency characteristic that emphasizes the high frequency side, as represented by a curve.

On the other hand, the remaining equation (1-β・P
(z) ) / (1-P (z) ) represents a frequency characteristic that approximates the spectrum shape of an audio signal, similar to equation (3).

Therefore, the frequency characteristic N of the requantization noise expressed by equation (9) approximates the spectral shape of the requantization noise to the spectral shape of the audio signal, and has a high frequency The frequency characteristics are emphasized on the side.

Thus, by using a noise filter with a fixed frequency characteristic, the spectral shape of the requantization noise can be adjusted to the audio signal 1.
It is possible to obtain a noise shaving function that approximates the spectrum shape of the signal and emphasizes it on the high frequency side, and it is possible to further improve the auditory signal-to-quantization noise ratio with a simple overall configuration.

Incidentally, the weighting of the multipliers 66 and 69 is the coefficient 12 and T,
Letting them be the values β and 1, respectively, the frequency characteristic N of the requantization noise can be expressed by the following relational expression: can do.

In the configuration of FIG. 8, the quantization error signal Stl obtained through the subtracter 51 is passed through the multipliers 66 and 69 to the value T! and is weighted by T1 and input to the prediction filter 24, and is also weighted by a predetermined amount via the noise filter 68, and is then input to the prediction filter 24 and the requantizer 21.
is input.

Furthermore, the input signal of the requantizer 21 is weighted by a predetermined amount and input to the prediction filter 24, thereby allowing replacement with an equivalent circuit as shown in FIG.

Thus, the weighting of the multipliers 66 and 69 changes the spectral shape of the requantization noise to the spectral shape of the audio signal by setting the coefficients γ2 and γ to values 1 and β, respectively, as shown in equation (9). The input digital signal S1 can be compressed and transmitted while being approximated and emphasized on the high frequency side.

On the other hand, the weighting of the multipliers 66 and 69 is by the coefficient γ2
If and T1 are set to values β and 1, respectively, as shown in equation (10), it is possible to compress the information of the input digital signal S1 and transmit it with the -stage and noise shaving characteristics freely selected. can.

According to the configuration shown in FIG. 8, the input signal of the requantizer 21 and the output signal of the inverse quantizer 22, each weighted by a predetermined amount, are output to the prediction filter 24, and the quantization error signal S□ is output to the prediction filter 24. By outputting it to the prediction filter 24 and the requantizer 21 via the noise filter 68, the spectral shape of the requantization noise is approximated to the spectral shape of the audio signal and is emphasized on the high frequency side. The digital signal S1 can be compressed and transmitted.

In this way, by using a noise filter with a fixed frequency characteristic, it is possible to obtain a noise shaving function with a frequency characteristic that is extremely complex. A digital signal processing device can be obtained.

(G3) Third Embodiment In FIG. 13, 85 indicates a feedback type digital signal processing device as a whole, and a noise filter 68
(FIG. 8), a filter 86 with more complicated frequency characteristics is used to obtain a noise shaving function with even more complicated frequency characteristics.

That is, in the filter 86, the output signal of the correction filter 87 and the weighted quantization error signal S□ are applied to the adder 88, and the output signal is applied to the correction filters 87 and 9.
Give to 0.

Furthermore, the output signals of the correction filters 87 and 90 are applied to the subtracter 67 and the adder 70 via a subtracter 91.

Therefore, the frequency characteristic Ft (Z) of the filter 86 is the frequency characteristic of the correction filters 87 and 90, respectively A (Z
) and B (Z), it can be expressed by the following relational expression.

Therefore, as in the second embodiment, multipliers 66 and 6
For the weighting of 9, if the coefficients γ8 and T1 are set to values 1 and β, respectively, the frequency characteristic N3 of the requantization noise is calculated by adding the value F (z) of equation (9) to the value p'5 ( z) can be expressed by the following relational expression (12), and a noise shaving function with an even more complex frequency characteristic can be obtained.

On the other hand, the weighting of the multipliers 66 and 67 is by the coefficient γ2
and TI are set to values β and 1, respectively, the frequency characteristic N of the requantization noise becomes (1
1) By substituting the value Fg(z) in the equation, it can be expressed by the following relational expression.

Therefore, it is possible to obtain a noise shaving function with a frequency characteristic that is much more complex than that of the second embodiment.

According to the configuration shown in FIG. 13, the input signal of the requantizer 21 and the output signal of the inverse quantizer 22, each weighted by a predetermined amount, are output to the prediction filter 24, and the quantizer The effect of the second embodiment is achieved by outputting the quantization error signal Stl to the prediction filter 24 and the requantizer 21 through the filter 86, which is composed of correction filters 87 and 90 and has complex frequency characteristics. In addition to this, it is possible to obtain a noise shaving function with even more complex frequency characteristics.

(G4) Fourth Embodiment In the second and third embodiments, the weighting of the multipliers 66 and 69 is such that the coefficients 12 and γ are changed to the values β and 1, respectively.
In the above description, the weighting coefficients 12 and y are not limited to these values, but can be set to a wide variety of values as necessary.

Further, in the above-described embodiment, the case where an audio signal is transmitted has been described, but the present invention is not limited to this. It can be widely applied to recording and reproducing audio signals with high quality.

Effects of the Invention As described above, according to the present invention, the input signal of the requantizer and the output signal of the inverse quantizer are weighted by a predetermined amount and output to the prediction filter, thereby reducing quantization noise. Feedback-type digital signal processing that can bring the spectrum shape of the input digital signal close to that of the audio signal and compress the information of the input digital signal before transmitting it, thus having an overall simple configuration, short processing time, and a noise shaving function. You can get the equipment.

[Brief explanation of the drawing]

FIG. 1 shows the first part of the digital signal processing device according to the present invention.
FIG. 2 to FIG. 6 are block diagrams showing the equivalent circuit, FIG. 7 is a characteristic curve diagram showing the spectrum shape of the requantization noise, and FIG. 8 is the second one. A block diagram showing the embodiment, FIGS. 9 to 11 are block diagrams showing the equivalent circuit, FIG. 12 is a characteristic curve diagram showing the spectrum shape of the requantization noise, and FIG. 13 is the third implementation. A block diagram showing an example, FIG. 14 is a block diagram showing a conventional feedforward type digital signal processing device, FIG. 15 is a block diagram showing a conventional feedback type digital signal processing device, and FIG. 16 is a specific example thereof. FIG. 17 is a block diagram showing the configuration of a feedforward type digital signal processing device having a noise shaving function. l, 2.30, 50.65.85... Digital signal processing device, 3.10, 14.24.32.33.
40... Prediction filter, 7.21...
- Requantizer, 45.68... Noise filter.

Claims (3)

[Claims] (1) A prediction filter equipped with a feedback loop, a requantization means for requantizing and outputting a difference signal between an input signal and an output signal of the prediction filter, and an inverse characteristic of the requantization means, and inverse requantization means for requantizing the output signal of the requantization means and outputting it to the prediction filter, wherein the output signal from the inverse requantization means is output to the prediction filter. A digital signal processing device characterized in that the output signal is weighted by a predetermined amount together with the input signal of the requantization means and outputted to the prediction filter. (2) A prediction filter equipped with a feedback loop, a requantization means for requantizing and outputting a difference signal between an input signal and an output signal of the prediction filter, and an inverse characteristic of the requantization means, and inverse requantization means for requantizing the output signal of the requantization means and outputting it to the prediction filter, wherein the output signal from the inverse requantization means is output to the prediction filter. The output signal is weighted by a predetermined amount together with the input signal of the requantization means and outputted to the prediction filter, and the output signal output from the inverse requantization means and the above of the requantization means are weighted by a predetermined amount. A digital signal processing device characterized in that an input signal is weighted by a predetermined amount and fed back to the prediction filter and the requantization means via a filter with predetermined frequency characteristics. (3) The digital signal processing device according to claim 2, wherein the filter is composed of a plurality of filters.