JP3282450B2 - Code decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a codec for transmitting digital signals with high quality.

[0002]

2. Description of the Related Art Music recording and reproduction using digital signals such as compact discs and DATs are widely performed. For example, a compact disc has a sampling frequency of 44.1 kHz.
z, recorded by a 16-bit linear code. This method cannot reproduce sound exceeding 22.05kHz, and 98d
Obtaining a dynamic range exceeding B is also impossible in principle. Acoustic signals from live musical instruments include 2
In spite of including a component exceeding 2.05 kHz, it was said that there was no need to reproduce this component because it was out of the audible band. However, in recent years, studies have been conducted on the possibility that ultra-treble activates α-wave, which is a human brain wave, and it has begun to be considered that ultra-treble has some effect on brain wave. Whether or not it can be heard by humans can not be said unconditionally due to individual differences, but it has some physical and physiological influences and effects, and in order to leave higher sound quality as a cultural heritage in the future, it is extremely high in the reproduced signal It has been pointed out that regional components are important. In addition, while there is an actual sound whose dynamic range exceeds 100 dB and reaches 130 dB, clip distortion is likely to occur when this is expressed by a linear code of 16 bits. It has been pointed out that the dynamic range is insufficient because distortions due to quantization errors make the sound muddy.

[0003] Therefore, as shown in the radio and experiment magazine, February 1995, pages 100 to 101, published by Seibundo Shinkosha, LSB of 16-bit data is used, and 2 bits are used for this bit.
A method of recording music signal information of 2.05 kHz or more using ADPCM (method 1), and a radio technology magazine, April 1991, pp. 147 to 150, published by IAI Publishing Co., Ltd.
As shown on the page, there has been proposed a method (method 2) for improving the perceived dynamic range by driving the quantization noise to 15 kHz to 22.05 kHz using noise shaping.

[0004]

However, in the above configuration, although the reproduction band is wide in the scheme 1, there is a problem that the dynamic range in the audible band is reduced by 6 dB. Within 15
There was a problem that the dynamic range was remarkably reduced between kHz and 20 kHz.

An object of the present invention is to solve the above problem and to provide a code decoding apparatus having a wide dynamic range and a high dynamic range.

[0006]

According to the present invention, a non-uniform quantization for non-uniformly quantizing an input code having a non-uniform quantization characteristic and an inverse transform of the non-uniform quantization are performed. Inverse quantization, the difference between before and after the process of performing both the non-uniform quantization and the inverse quantization is added to the input signal by multiplying by a predetermined transfer function H1 (z), and the result of the addition is A coding method is used in which a coding result is obtained by a system that feeds back to a non-uniform quantization process. In order to achieve this object, the encoding / decoding apparatus according to the present invention employs, at the time of encoding, a requantization system comprising a non-uniform quantizer having non-uniform quantization characteristics and an inverse quantizer having the inverse characteristics thereof. An error signal or quantization noise between the signal input to the output device and the output signal is combined with the input signal via a feedback circuit having a predetermined transfer characteristic, and at the time of decoding, the coded signal is dequantized. The signal which was input to the device and dequantized was reproduced.

Also, at the time of encoding, the output of the inverse quantizer is taken out to obtain a code-decoded output. At the time of decoding, a gate for opening the ΔΣ modulation loop is provided, and a fixed value at that point is set.

[0008]

As described above, the inversely quantized signal changes the magnitude of the rounding error, that is, the quantization error, depending on the signal strength, by the nonlinear transformation by the nonuniform quantization and the inverse transformation by the inverse quantization, depending on the magnitude of the code. Has the effect of causing. For example, when the input code is small, it is finely quantized by 24 bits, and gradually increases to 23 bits and 22 bits as the size of the input code increases, and when the input code is maximum, it is quantized by 15 bits most coarsely. I can do it. That is, within the dynamic range of 146 dB, the instantaneous S / N ratio of the coding can be changed from 92 dB to 146 dB depending on the size of the input code. By feeding back the quantization error and adding the quantization error to the input code, an effect of performing spectrum conversion occurs, so that the dynamic range in the low frequency band can be expanded even in the region of the input code of coarse quantization.

The requantized signal is subjected to non-linear transformation by a non-uniform quantizer and inverse transformation by an inverse quantizer, without changing the magnitude of the signal. Is changed by the signal strength. For example, if the input signal is small, it is finely quantized by 24 bits, and gradually increases to 23 bits and 22 bits as the input signal strength increases, and if the input signal strength is maximum, it is quantized by 15 bits most coarsely. Can be. That is, within the dynamic range of 146 dB, the instantaneous S / N ratio is changed from 92 dB to 146 depending on the input signal strength.
dB. By combining this quantization noise with the input signal via the feedback circuit, an effect of performing spectrum conversion occurs, and even in the region of the input signal strength of coarse quantization, the dynamic range in the audible band can be expanded.

Also, by receiving the above code as a signal and dequantizing it, if the signal is small, it is finely quantized with 24 bits, and gradually becomes coarser with the increase of the input signal strength. When the input signal strength is the maximum, the signal roughly quantized with 15 bits, that is, within the dynamic range of 146 dB, the instantaneous S / N ratio is changed from 92 dB to 146 depending on the input signal strength.
By combining the quantization noise of the signal changed to dB to the input signal via the feedback circuit, an effect of decoding the signal subjected to the spectrum conversion occurs, and even in the region of the input signal strength of the coarse quantization. , The dynamic range within the audible band can be expanded.

Also, a code decoding output is obtained at the time of coding,
It can be used as a code monitor signal. At the time of decoding, Δ
初期 By opening the modulation loop, the initial operation when returning to the code operation can be stabilized in a short time.

[0012]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a code decoding apparatus according to a first embodiment of the present invention. In the figure, 1
10 is an adder, 120 is a non-uniform quantizer, 130 is an inverse quantizer, 140 is a subtractor, 150 is a feedback circuit, 160 is a switch 160, 220 is a digital filter, and 230 is D
A converter 240 is a low-pass filter. Note that the number entered without a leader line beside the signal line indicates the number of bits. First, the description at the time of sign will be described. At the time of encoding, the switch 160 is controlled by the mode control signal 204 to output the output signal of the non-uniform quantizer 120 to the inverse quantizer 13.
The signal path is switched so as to supply 0.

An input signal X input from an input terminal 101
(Here, the sampling frequency is 192 kHz and the word length is 24 bits) is supplied to the non-uniform quantizer 120 through the adder 110. The other input signal of the adder 110 is the quantization noise Q supplied from the feedback circuit 150. The non-uniform quantizer 120 compresses the input 24-bit signal into a 16-bit signal. The compression method will be described later. The compressed 16-bit signal is output from the output terminal 102 as an output signal W1 and transmitted to the recording device, and is input to the inverse quantizer 130 via the switch 160. The inverse quantizer 130 inversely quantizes the data so as to have a characteristic opposite to the compression characteristic of the non-uniform quantizer 120 and expands the signal into a 24-bit signal. Non-uniform quantizer 12
0 and the requantized signal Y1 requantized by the inverse quantizer 130 are input to the subtractor 140. The subtractor 140 outputs a difference signal between the requantization input signal and the requantization output signal, that is, a quantization noise Q. This quantization noise Q is converted in frequency and / or phase characteristics by a feedback circuit 150 having a transfer characteristic H1 (z), and is returned to the adder 110. When the requantized signal Y1 is expressed by an equation, Y1 = X + (1−H1 (z)) * Q, and the requantized signal Y1 is represented by a component of the input signal X and a transfer characteristic H1 (z) of the quantization noise Q. It is the sum of the returned components. Therefore, the spectrum conversion of the quantization noise Q can be performed by the transfer characteristic H1 (z). These loops make Δ
Σ A 16-bit signal that has been modulated and subjected to a predetermined spectrum conversion and compressed by the non-uniform quantizer 120 is output to the output terminal 10.
2 as an output signal W1.

The output signal W1 is output to a transmission device, here, a writing device (not shown) for a high-density optical disk, forms a recording format, and is recorded on a medium (not shown) for the high-density optical disk. The medium is reproduced by reproducing means (not shown) to extract a decoded signal. Here, the requantized signal Y1
Is expressed by using the output signal W1. Y1 = W1 * P ^{-1 When the} signal W1 is inversely quantized by the inverse quantizer 130, a requantized signal Y1 is obtained, and the equivalent input signal W2 is inversely quantized. The inversely quantized signal Y2
Can be taken out as a reproduction signal.

The operation at the time of decoding will be described based on this.
At the time of decoding, the switch 160 is controlled by the mode control signal 204 to switch the signal path so that the decoded input signal W2 input from the input terminal 201 is supplied to the inverse quantizer 130. The input signal W2 is input from the input terminal 201 and supplied to the inverse quantizer 210. The 16-bit input signal W2 is inversely quantized by the inverse quantizer 210 so as to have an inverse characteristic to the compression characteristic of the non-uniform quantizer 120, and expanded to a 24-bit inversely quantized signal Y2. The 24-bit inversely quantized signal Y2 corresponds to the requantized signal Y1 at the time of encoding. That is, it becomes a signal containing quantization noise Q whose spectrum has been converted by ΔΣ modulation. The 192 kHz, 24-bit inverse quantized signal Y2 is subjected to double oversampling processing in the digital filter 220, converted into a 384-kHz, 24-bit signal, and input to the DA converter 230.
The D / A converter 230 converts this signal into an analog signal, then removes aliasing distortion by a low-pass filter 240 and outputs the analog signal from the output terminal 203 as an analog signal output A. It is sufficient if the combined characteristics of the digital filter 220 and the low-pass filter 240 are such as to reduce components of 96 kHz or more, which is the Nyquist frequency.
2 can be directly D / A converted by the DA converter 230. In this case, it is preferable that the low-pass filter 240 sharply cuts off a component of 96 kHz or more.

In the embodiment constructed as above, various characteristics of the quantization noise will be described in detail. Since the cause of the quantization noise is an error signal generated by the roughness of the quantization step, first, the relationship between the control of the roughness of the quantization step by the input signal strength and the quantization noise, and then the quantization noise by ΔΣ modulation Will be described. Since the conventional ΔΣ modulation uses a uniform quantizer having a fixed word length, the rounding error becomes substantially constant irrespective of the signal strength, and a 16-bit signal has a dynamic range of 98 dB.
On the other hand, in the embodiment, a method is used in which the magnitude of the rounding error, that is, the quantization noise is changed depending on the signal intensity. The magnitude of the rounding error is changed for the requantized signal Y1 by the signal strength while the magnitude of the signal remains unchanged by the non-linear transformation by the non-uniform quantizer 120 and the inverse transformation by the inverse quantizer 130. To change the magnitude of the quantization noise. When the input signal is small, it is finely quantized by 24 bits,
23 bits and 22 bits are gradually coarsened as the input signal strength increases, and when the input signal strength is the maximum, quantization is performed at the coarsest 15 bits. Therefore, the quantization noise is minimized when the input signal strength is small, and a 24-bit accuracy, that is, an instantaneous S / N ratio and a dynamic range of 146 dB can be obtained. When the input signal strength is large, the instantaneous S / N ratio becomes 15 bits, that is, 92 dB. 146
Within the dynamic range of dB, the instantaneous S / N ratio changes from 92 dB to 146 dB depending on the input signal strength.

FIG. 3 shows a non-uniform quantizer 12 according to the embodiment.
0 and the instantaneous S / N when the inverse quantizer 130 is combined
FIG. 9 is a diagram illustrating characteristics of a ratio versus input signal strength. The vertical axis in the figure is the instantaneous S / N ratio, and the instantaneous S / N ratio is a signal-to-noise distortion rate in a signal band from 0 Hz to 96 kHz which is the Nyquist frequency. As can be seen from FIG. 3, the instantaneous S / N ratio can be greatly improved over almost the entire input level as compared with the conventional linear code (16 bits). As a specific compression method of the non-uniform quantizer 120, a run-length 1 / n compressed floating code was used. A run-length 1 / n compression floating code method will be described.

The continuous data Q0 in which bits of a predetermined logic continue above the linear code as the original data, the inverted bit T0 for breaking the continuity of the continuous data Q0, and the lower data D0 after the inverted bit T0. A compressed continuous data Q1 obtained by inputting a linear code of L bits configured and compressing the run length of the continuous data Q0, an inverted bit T1 for breaking the continuity of the compressed continuous data Q1, and the run length Remainder F generated when compressing
The data is converted into M-bit compressed data composed of compressed remainder data C1 representing 1 and mantissa data D1 obtained by rounding the lower-order data D0 and output. However, the continuous data Q
When the run length of 0 is L0, the run length of the compressed continuous data Q1 is L1, and n is an integer of 2 or more, L1 = int (L0 / n) and F1 = L0 mod n.

In the inverse quantization of the inverse quantizer 130, the following method was used to perform the inverse transform of the run-length 1 / n compressed floating code. Compressed continuous data Q1 in which bits of a predetermined logic continue at a higher order;
Bit T1 that breaks the continuity of the data, compressed remainder data C representing the remainder F1 generated when compressing the run length
Based on the compressed data composed of 1 and the mantissa data D1, the run length of Q1 is expanded by n times, and continuous data having a length corresponding to the value of C1 is added to break the continuity of Q0. The inversion bit T0 is added, the mantissa data D1 is subsequently added, the continuous data Q0, the inversion bit T0 and the mantissa data D0 are read out, and the expanded data is output.

Here, the run length of the continuous data Q0 is L0, and the run length of the compressed continuous data Q1 is L.
1. The remainder obtained from the compressed remainder data C1 is F1, and n is 2
Assuming the above integer, L0 = L1 * n + F1 D0 = D1.

The above-mentioned method for compressing a run-length 1 / n compressed floating code, a compression apparatus and a decoding apparatus are disclosed in Japanese Patent Application Laid-Open Nos. 4-286421 and 5-1.
Specific descriptions are given in JP-A-83445 and JP-A-5-284039. Here, the outline and characteristics of the compression method will be described. FIG. 4A is a conceptual diagram showing the configuration of a run-length 1/2 compressed floating code used in the first embodiment of the present invention, and FIG. It is a figure for explaining code conversion which compresses to / 2 compression floating code (16 bits). Hereinafter, the procedure of compression (non-uniform quantization) will be described first with reference to the drawings.

The compressed continuous data Q1 is continuous data having a length of the run length L1 obtained by dividing the run length L0 of the continuous data Q0 by 2 and converting it into an integer. That is, L1 = int (L0 / 2). If the remainder term of the integer division is a compression remainder F1, then F1 = L0 mod 2.

The inversion bit T1 is an inversion bit for breaking the run of the compressed continuous data Q1. Compressed remainder data C
1 is a complement representation of the compression remainder F1. Further, the mantissa data D1 is upper partial data of the data D0.
The run-length 1/2 compressed floating code is arranged in the following order: polarity bit P, compressed continuous data Q1, inverted bit T1, compressed remainder data C1, and mantissa data D1.

Hereinafter, the case where the run length L0 is "0" to "24" will be described with reference to FIG.
When L0 = 0, L1 and F1 are as follows: L1 = int (0/2) = 0 F1 = 0 mod 2 = 0.

Since the run length L1 is "0", there is no compressed continuous data Q1. The compression remainder data C1 is “1”. The data D0 is 24 bits, and the upper 13 bits “ABCDEFGHIJKLM” are mantissa data D1.
It is. In the run-length 1/2 compressed floating code, the polarity bit P, the inverted bit T1, the compression remainder data C1, and the mantissa data D1 are arranged in this order, and "P11AB
CDEFGHHIJKLM ".

Similarly, the following is obtained. When L0 = 1, L1 = int (1/2) = 0 F1 = 1 mod 2 = 1, there is no compressed continuous data Q1, and there is no compressed residual data C1.
Is “0”. Upper 13 bits of data D0 “ABC
"DEFGHIJKLM" is the mantissa data D1. The run-length 1/2 compression floating code is "P10AB
CDEFGHHIJKLM ".

When L0 = 2, since L1 = int (2/2) = 1 F1 = 2 mod 2 = 0, the continuous data Q1 is "0" and the compression remainder data C1 is "1". The upper 12 bits “A” of data D0
"BCDEFGHIJKL" is the mantissa data D1. The run length 1/2 compression floating code is “P011
ABCDEFGHIJKL ".

When L0 = 3, L1 = int (3/2) = 1 F1 = 3 mod 2 = 1, the continuous data Q1 is "0", and the compression remainder data C1 is "0". The upper 12 bits “A” of data D0
"BCDEFGHIJKL" is the mantissa data D1. The run-length 1/2 compression floating code is "P010
ABCDEFGHIJKL ".

When L0 = 17, L1 = int (17/2) = 8 F1 = 17 mod 2 = 1, and the continuous data Q1 is "00000000".
The compression remainder data C1 is “0”. The upper five bits “ABCDE” of the data D0 are the mantissa data D1.

Therefore, the run length 1/2 compression floating code is "P0000001010 ABCDE". When L0 = 18 to 24, the compressed remainder data C
1 is omitted, L1 = int (18/2) = 9, the continuous data Q1 is set to “00000000”, and the upper 5 bits “ABCDE” of the data D0 are set to the mantissa data D
Set to 1.

Therefore, the run length 1/2 compression floating code is "P000000010001ABCDE". In this way, the linear code (24 bits) is converted into a run-length 1/2 compressed floating code (16 bits)
Compress to Next, the procedure of decompression (inverse quantization) will be described.

The decompression procedure removes the polarity bit P from the compressed data and obtains a run length L1 from the compressed continuous data Q1 and the inverted bit T1. Also, the compression remainder data C1 (1 bit) and the mantissa data D1 immediately after the inversion bit T1
Get. A compression remainder F1 obtained by inverting the compression remainder data C1 is obtained. From these, the run length L0 of the continuous data Q0 of the original data is obtained from L0 = 2 * L1 + F1.

The continuous data Q0 is restored by connecting the length "0" of the run length L0. The inversion bit T0 is added after the continuous data Q0, and the mantissa data D1 is added after that. The polarity bit P is added to the head to be the original data. If the code length is less than W0, a fixed value is applied to the lower part of the mantissa data D1 to make the code length W0. The decoding process is performed according to this procedure.

For example, when L1 = 0 and F1 = 0, since L0 = 2 * 0 + 0 = 0, there is no continuous data Q0, and the mantissa data D1 is 13-bit "ABCDEFGHIJKLM".
At this time, the original data includes a polarity bit P and an inverted bit T0.
And the mantissa data D1 to form “P1ABCDEF
GHIJKLM **** ". In order to minimize the offset distortion at the time of expansion, “***
"" Is a fixed value "011 ..." or "100 ..."
Allot.

In this manner, the run length 1/2 compressed floating code (16 bits) is converted to the linear code (24 bits).
) Is decompressed and dequantized. In the same manner as above, the results summarized for all cases are shown in (Table 1).

[0036]

[Table 1]

In Table 1, the linear code (24 bits) is a folded binary code, and the floating code is a folded run-length 1/2 compressed floating code. The columns of run length L0, run length L1, and resolution are in decimal notation. The expression accuracy, that is, the resolution, of the restored code (inverse quantized signal) obtained by decoding (inverse quantizing) the compressed code (non-uniform quantized signal) and expanding it is determined by rounding the linear code, and varies depending on the run length L0. As is apparent from Table 1, a precision of up to 24 to 15 bits can be obtained.

Tables 2 and 3 show the results summarized so as to be suitable for mathematical expression conversion and table conversion by the DSP.

[0039]

[Table 2]

[0040]

[Table 3]

Table 2 is a conversion table for non-uniform quantization, where X is an input code for non-uniform quantization, and W is an output code for non-uniform quantization. 16 if the code length of W exceeds 16
Round. If the code length of X is insufficient, "0" is inserted at the lower position. (Table 3) is a conversion table of inverse quantization, that is, requantization, where W is a compression code, and Y1 is an output code of requantization, that is, a requantized signal Y1. The code length of Y1 is 24
, Rounded to 24, and when the code length of Y1 is insufficient, lower-order “0, 1, 1, 1,...” Or “1,
0, 0, 0,... "Are inserted.

Here, a linear code is input, compression encoding (non-uniform quantization) is performed using a run-length 1/2 compression floating code, and this code is decoded and decompressed (dequantized). The instantaneous signal-to-noise ratio when outputting a digitized signal will be described. However, in the following description, for the sake of simplicity, the instantaneous S / N ratio does not include the improvement of the S / N ratio of the rectangular wave (about 2 dB). The input level is based on the amplitude of the maximum sine wave that can be represented by a code (0 dB). The linear code "P1111
... "corresponds to this.

The range of input levels from 0 dB to -6 dB is "P1ABC..." By a linear code, as shown in Table 1.
More resolution is 15 bits. The quantization noise of the 15-bit data is -90 dB as -6 dB per bit. Therefore, the instantaneous S / N ratio becomes 90 dB to 84 dB in the range of the input level from 0 dB to -6 dB. In the input level range of -6 dB to -18 dB, "P01ABC ..." or "P001ABC ..."
. "And the resolution is 16 bits from (Table 1).
Since the quantization noise of 16-bit data is -96 dB, the instantaneous S / N ratio is 90 dB to 78 dB.

In the input level range of -18 dB to -30 dB, a linear code is "P0001ABC ..." or "P00001ABC ...", and from Table 1, the resolution is 17 bits. Since the quantization noise of 17-bit data is -102 dB, the instantaneous S / N ratio is 84 dB.
B to 72 dB. Similarly, the instantaneous S / N ratio is obtained in the input level region up to -144 dB.

FIG. 3 referred to above also shows run length 1 /
FIG. 4 is a characteristic diagram illustrating an instantaneous S / N ratio of a requantized signal obtained by decoding a compressed code (a non-uniform quantized signal) using a two-compression floating code and expanding (dequantizing). The horizontal axis of the figure shows the input level, and the vertical axis shows the instantaneous S / N ratio. It can be seen that the improvement of the instantaneous S / N ratio works over a wide range of input levels from -18 dB to -144 dB. Although the instantaneous S / N ratio is degraded in the range of the input level from 0 dB to -6 dB, it is at most 6 dB
B is a deterioration of the instantaneous S of 90 dB to 84 dB.
/ N ratio.

In the embodiment described above, the linear code is a folded binary code using a run-length 1 / n compressed floating code. However, other linear codes such as a 2'S complementary code and an offset binary code are used. Can be applied in exactly the same way, only by converting each other or changing a predetermined logical value. Also, the case where n is “2” has been described, but n may be any integer as long as it is an integer of “2 or more”. In this case, since the number in the case of the compression remainder changes according to the value of n, it goes without saying that the word length of the compression remainder data may be changed. In addition to the circuit configuration, the device may be configured with DSP (digital signal processor) hardware and software for performing table conversion and data conversion.

As described above, when the run length of the original data is small, the exponent part, that is, the range is represented by a small number of bits, and when the run length is large, the bit number is allocated, and the exponent part, that is, the range, is represented by a large number of bits. Since the word length of the entire code is fixed, the number of bits of the mantissa changes according to the run length. By these actions, the expression space of the range of the compression code output from the output unit is expanded, and at the same time, the expression accuracy can be improved.

The human hearing limit of the instantaneous S / N ratio is 90 dB.
Although this is almost enough because it is below, realizing perfect audio recording transmission with one point advocating ultimate audio as a cultural heritage to be preserved in the future, or securing and editing headroom in the case of studio recording Considering the necessity of a margin required for mixing and sound effect processing at the stage, the instantaneous S /
It is desirable to secure an N ratio of about 120 dB or more.

Then, the frequency spectrum of the quantization noise will be described. As described above, the 24-bit inverse quantized signal Y2 output from the code decoding device corresponds to the requantized signal Y1. That is, it becomes a signal having quantization noise Q subjected to spectrum conversion by Δ ス ペ ク ト ル modulation. First of the present invention
In the embodiment, the fifth-order transfer characteristic H1 (z) of the feedback circuit 150 is used. Therefore, the instantaneous S / N ratio and dynamic range in the audible band up to 20 kHz are lower than those in the case where no feedback is performed (0th order). An improvement of about 28 dB was achieved. In the audible band up to 10 kHz, an improvement of 58 dB can be achieved. Table 4 shows the relationship between the number of bits, the order of the transfer characteristic H1 (z), and the instantaneous S / N ratio at a sampling frequency of 192 kHz.

[0050]

[Table 4]

FIG. 5 shows the characteristics of (Table 4). The numerical values in (Table 4) and FIG. 5 are instantaneous S / N ratio characteristics without waiting and without auditory correction. In the embodiment, the fifth-order transfer characteristic H1 (z) is used. However, a good result can be obtained by using a higher-order transfer characteristic H1 (z) or a characteristic that becomes a fractional polynomial. As described above, the frequency spectrum of the quantization noise can be replaced by ΔΣ modulation, and the instantaneous S / N ratio and dynamic range of the audible band can be improved. This improvement in the instantaneous S / N ratio and the dynamic range is most effective when the quantization noise is large, that is, when the input signal strength is large.
In other words, the maximum effect is achieved by the synergistic and complementary interaction between the action of expanding the dynamic range by the non-uniform quantization and the action of replacing the quantization noise spectrum by the ΔΣ modulation. Thus, the frequency spectrum of the quantization noise can be replaced by the ΔΣ modulation, and both the dynamic range and the frequency band can be efficiently expanded at a low bit rate.

In the case of only a super-high frequency component or a sound source containing a large amount of the component, the level is generally small.
Transmission can be performed with 4 dB and 24-bit accuracy, the instantaneous S / N ratio is sufficiently ensured, and the noise floor in the ultrahigh frequency range can be suppressed low. Furthermore, when the signal level is large to some extent, the quantization noise slightly increases. However, since this spectrum is replaced by a high frequency band and the noise floor of the ultrahigh frequency band rises, the human brain wave (α wave)
There is a possibility that the effect of promoting the activation and relaxation of is obtained.

At the time of encoding, a quantization error signal or quantization noise between a signal input to the requantizer and an output signal of the requantizer is transmitted via a feedback circuit having a predetermined transfer characteristic H1 (z). If an output signal of the inverse quantizer is taken out as a code decoding output at the same time as obtaining a code output by synthesizing it with an input signal, the output can be effectively used as a code monitor signal. This signal may be converted to an analog signal A via the digital filter 220, the DA converter 230, and the low-pass filter 240 to obtain an output.

By opening the ΔΣ modulation loop at the time of decoding, the initial operation when returning to the code operation can be stabilized in a short time. FIG. 2 is a block diagram showing a second embodiment of the present invention in which a gate is provided. In the figure, reference numeral 170 denotes a gate, which opens a ΔΣ modulation loop during decoding by a mode control signal input from the input terminal 204 and gives a fixed value “0”. Subtractor 14
The requantized signal Y1 input to 0 becomes “0”, and the output of the subtractor 140 becomes the same as that of the adder 110. Since the requantized signal Y1 is "0", that is, the center value, the probability that the system is quickly stabilized when the code mode is re-entered is the highest. In the second embodiment of the present invention, the gate 170 is provided in front of the subtractor 140. However, the gate 170 may be provided at any point where the loop is opened. This has the effect of reducing time.

Further, it is also possible to apply emphasis by the code decoding device shown in FIG. In this case, for example, an audio input is input to the adder 110 via an emphasis circuit (not shown), and the codec decoding device 100 uses the converted audio signal and an emphasis identification signal indicating on / off of emphasis as a subcode. Output from At the time of decoding, a de-emphasis characteristic having a frequency characteristic opposite to that of the emphasis circuit is converted into an inverse quantizer 210, a digital filter 220, and a DA converter 23.
0, convolution with one of the low-pass filters 240, or a de-emphasis circuit (not shown) inserted between them, and turned on / off based on an emphasis identification signal (not shown) extracted as a subcode. Thus, for example, it is possible to suppress a slight increase in quantization noise in an ultrahigh frequency range due to ΔΣ modulation, and it is possible to suppress the maximum output level during reproduction to a certain value or less by using this characteristic curve. In particular, since the ultra-high frequency speaker generally has low input power resistance, such a protection characteristic of the maximum output level is effective.

The transfer function H1 (z) of the feedback circuit 150
Needless to say, it is not necessary to use the one shown in the embodiment, and a higher-order one or one that becomes a fractional polynomial may be used. Sampling frequency 192
The kHz is not limited to this, and the same effect can be obtained if the sampling frequency is higher than the conventional sampling frequency of 48 kHz. In addition to the run-length 1 / n compressed floating code, any non-uniform quantization that expands the dynamic range may be used. For example, the same effect can be obtained with a compressed code using the μ-law A-law. . Further, the number of input and output bits of the code decoding apparatus is set to 24 and 16, but is not limited to this.

Regarding de-emphasis, an identification signal may be transmitted to switch on / off.
Since the signal level in the band exceeding 20 kHz is not so large, only the ON mode may be used. The input signal X may be an analog signal, in which case the adder 11
The same operation and effect can be obtained by substituting 0, the subtractor 140, and the feedback circuit 150 for analog.

[0058]

As described above, according to the present invention, a signal input to a requantizer constituted by a non-uniform quantizer having non-uniform quantization characteristics and an inverse quantizer having the inverse characteristics thereof is provided. The quantization error signal or the quantization noise between the input signal and the output signal is combined with the input signal via a feedback circuit having a predetermined transfer characteristic. Further, since the coded signal is configured to reproduce a signal obtained by inversely quantizing the signal by the inverse quantizer, the requantized signal is subjected to nonlinear transformation by the non-uniform quantizer and inverse transformation by the inverse quantizer. The conversion has an effect that the rounding error, that is, the magnitude of the quantization noise changes according to the signal strength. For example, if the input signal is small, it is finely quantized by 24 bits, and gradually increases to 23 bits and 22 bits as the input signal strength increases, and if the input signal strength is maximum, it is quantized by 15 bits most coarsely. Can be. That is, within the dynamic range of 146 dB, the instantaneous S / N ratio can be changed from 92 dB to 146 dB depending on the input signal strength. By combining this quantization noise with the input signal via the feedback circuit, the effect of performing spectrum conversion occurs.
The effect of expanding the dynamic range of the audible band is obtained even in the region of the input signal strength of the coarse quantization. In addition to the above effects, there are the following specific effects.

(A) 16 bits per channel 192
At a low bit rate of kHz, both a high dynamic range of 146 dB or more and a wide band of Nyquist frequency of 96 kHz can be realized simultaneously. Eliminating sound turbidity caused by deterioration of distortion rate at a minute level, from 20 kHz to 96 kHz
To reproduce the original sound of the ultra-high frequency signal up to 44.1 kHz 16 bits, which is limited to the conventional limited reproduction space, and to record this signal as naturally as possible to reproduce as high as possible without any limitation. A medium and a reproduction sound field can be realized.

(B) 16 bits per channel 192
At a low bit rate of kHz, the signal-to-noise distortion ratio in the audible band up to 20 kHz can be improved to 123 dB to 177 dB. (C) Since the energy of the ultrasonic band in the music signal is small, lossless encoding can be performed with 24-bit accuracy and an instantaneous S / N ratio of 144 dB when the signal is only in the ultra-high band or mainly in the ultra-high frequency component. As a result, even when the Δ か け modulation is performed, the rise of the noise floor in the ultrahigh frequency range can be suppressed.

(D) When the audible band level is sufficiently large, the ultra-high frequency noise floor increases in accordance with the level, so that an ultra-high frequency noise component which may activate the α wave is automatically generated. Is done. (E) In general, the maximum withstand power of the super-high frequency range is 10
Maximum output of super high frequency range is 10-20dB
Although it is preferably low, the emphasis / de-emphasis characteristics can suppress the rise of the noise floor in the ultra-high frequency range and the maximum output in the ultra-high frequency range by 10 to 20 dB. Burn out due to output of power noise can be prevented.
Moreover, this characteristic is close to nature, and can reproduce a sound having a natural spectral distribution.

A code-decoded output is also obtained at the time of (f) coding,
It can be used as a code monitor signal. (G) By opening the ΔΣ modulation loop at the time of decoding, the initial operation when returning to the code operation can be stabilized in a short time.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a code decoding apparatus according to a first embodiment of the present invention.

FIG. 2 is a block diagram illustrating a code decoding apparatus according to a second embodiment of the present invention.

FIG. 3 Instantaneous S / N of non-uniform quantizer and inverse quantizer
Characteristic diagram showing specific characteristics

4A is a conceptual diagram showing a configuration of a run length 1/2 compression floating code, and FIG. 4B is a view showing a linear code (24 bits) converted to a run length 1/2.
Explanatory diagram for explaining code conversion for compressing to a compressed floating code (16 bits)

FIG. 5 is a characteristic diagram showing a relationship between the number of bits of the code decoding apparatus according to the embodiment, the order of the transfer characteristic H1 (z), and the instantaneous S / N ratio.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 code decoding device 110 adder 120 non-uniform quantizer 130 inverse quantizer 140 subtractor 150 feedback circuit 160 switch 170 gate 220 digital filter 230 DA converter 240 low-pass filter

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H03M 7/32 H03M 7/46

Claims (10)

(57) [Claims] When encoding, a non-uniform quantizer having a non-uniform quantization characteristic and an inverse quantizer having an inverse characteristic of the non-uniform quantization characteristic are connected via a switch to perform requantization. And a quantization error signal or quantization noise between a signal input to the requantizer and an output signal of the requantizer via a feedback circuit having a predetermined transfer characteristic H1 (z). In addition to synthesizing the input signal to obtain a code output, at the time of decoding, the connection is changed by the switch to switch to supply a decoded input signal to the inverse quantizer, and decode the output signal of the inverse quantizer. A code decoding device configured to take out as an output. 2. The code decoding apparatus according to claim 1, wherein the input signal is a signal of a linear code quantized substantially uniformly. 3. An input signal is an analog signal.
The code decoding device according to claim 1. 4. A method according to claim 1, wherein the predetermined transfer characteristic H1 (z) performs ΔΣ modulation for replacing the quantization noise with a frequency band having low sensitivity of the auditory characteristic even outside the audible band or within the audible band. Item 4. The code decoding device according to any one of Items 1 to 3. 5. A non-uniform quantizer or a non-uniform quantizer performs fine quantization when an input signal is small, and coarsely quantizes when an input signal is large.
5. The code decoding device according to any one of claims 4 to 4. 6. The non-uniform quantizer or non-uniform quantizer is L
6. The code decoding apparatus according to claim 5, wherein a linear code of bits is input, and M-bit data whose run length is compressed to 1 / n is output. (L, M and n are positive integers of 2 or more) 7. The non-uniform quantizer or non-uniform quantizer comprises:
A linear code serving as original data, that is, continuous data Q0 in which bits of a predetermined logic continue in the upper order;
Of continuous data Q0 obtained by inputting an L-bit linear code composed of an inverted bit T0 for breaking the continuity of data and lower-order data D0 after the inverted bit T0. Q1 and an inverted bit T1 for breaking the continuity of the compressed continuous data Q1.
A compression remainder data C1 representing a remainder F1 generated at the time of compressing the run length, and mantissa data D1 obtained by rounding the lower-order data D0 into M-bit compressed data for output. Item 6. The code decoding device according to Item 5. Here, the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1,
When n is an integer of 2 or more, L1 = int (L0 / n) and F1 = L0 mod n. 8. An inverse quantizer or an inverse quantizer comprises: compressed data, that is, compressed continuous data Q1 in which bits of a predetermined logic continue at an upper level, an inverted bit T1 for breaking the continuity of the compressed continuous data Q1, a run length. Compressed data composed of compressed remainder data C1 and mantissa data D1 representing a remainder F1 generated when the compression is performed, a C1 memory for storing the compressed remainder data C1, and a run length of the Q1 expanded by n times, Decompressing means for adding continuous data having a length corresponding to the value of the C1 memory, adding an inversion bit T0 for breaking the continuity of Q0, and subsequently adding the mantissa data D1; 6. The code decoding apparatus according to claim 5, wherein said code decoder reads out the inverted bit T0 and the mantissa data D0 and outputs decompressed data. However, the run length of the continuous data Q0 is L
0, L1 is the run length of the compressed continuous data Q1, F1 is the remainder obtained from the compressed remainder data C1, and n is an integer of 2 or more. L0 = L1 * n + F1 D0 = D1. 9. At the time of encoding, requantization is performed by connecting a non-uniform quantizer having non-uniform quantization characteristics and an inverse quantizer having inverse characteristics of said non-uniform quantization characteristics via a switch. A quantization error signal or quantization noise between a signal input to the requantizer and an output signal of the requantizer via a feedback circuit having a predetermined transfer characteristic H1 (z). 2. The apparatus according to claim 1, wherein the output of the inverse quantizer is extracted as a code-decoded output at the same time as obtaining a code output by synthesizing the signal.
9. The code decoding apparatus according to any one of claims 8 to 8. 10. A decoding circuit, wherein a ΔΣ modulation loop for feeding back a quantization error to an input is opened and a gate for giving a fixed value is inserted at one or more positions on the feedback circuit or the ΔΣ modulation loop during decoding. Code decoding apparatus according to any one of the above.