JPH08307281A - Nonlinear quantization method and nonlinear inverse quantization method - Google Patents

Abstract

PURPOSE: To provide the nonlinear quantization method by reducing a quantization error more than a conventional nonlinear quantization method. CONSTITUTION: A nonlinear function f() is fed to a quantization value calculation device 31 from a nonlinear function block 33 and an input signal representing a quantization level number is fed together with quantization data via an input terminal 11. A quantization value calculation device 31 calculates a quantization value as to each quantization level and the obtained quantization value is fed to a quantization device 32. Quantized data are fed to the quantization device 32 and each data is quantized to a quantization level of the quantization value closest among sets of quantization values at an equal interval received from the quantization value calculation device 31.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a non-linear quantization method and a non-linear dequantization method, and more particularly to signal compression for expressing an information signal with a small number of bits and signal expansion for reproducing the information signal from the number of bits.

[0002]

2. Description of the Related Art When a digital information signal is compressed for recording and / or transmission at a low bit rate, a sample value (hereinafter referred to as a sample) obtained by discretely sampling the information signal has a short word length. It is requantized using (word length). By shortening the word length representing each sample, a quantization error (hereinafter referred to as a quantization error) increases, but the number of bits required to represent information is reduced. Usually, this requantization is performed linearly by equally allocating the quantized value corresponding to the quantized level over the range of the level of the information signal. Then, each sample of the information signal is quantized to the nearest quantized value. Similar techniques are used for requantizing transform coefficients that represent information signals.

On the other hand, some applications, such as layer III of the so-called ISO / MPEG audio coding standard.
Then, the allocation of the quantized value corresponding to the quantized level is unequal. By allocating the quantized values unevenly, it is possible to quantize a certain value of the information signal with higher accuracy than other values. For example, if a value quantized with high precision occurs more often than a value quantized with low precision, or if the error of the value quantized with high precision is significant, the nonlinear quantization is a linear quantization. Produces better results. Both of these cases are possible for audio compression at low bit rates.

Most audio compression schemes orthogonally transform the audio input signal into the frequency domain prior to quantization.
In this case, the small-amplitude spectrum coefficient occurs more than the large-amplitude spectrum coefficient in a practical input signal. Furthermore, because of the psychoacoustic phenomenon of simultaneous masking that occurs between groups of adjacent spectral coefficients,
Noise due to quantization errors is less audible at large amplitude spectral coefficients than at small amplitude spectral coefficients. Therefore, by concentrating the quantization level on the small-amplitude spectrum coefficient and performing non-linear quantization,
The sound quality can be greatly improved.

Non-linear quantization is usually performed by applying linear quantization to the non-linearized data. The data to be quantized is usually a data block or data set consisting of a set of spectral coefficients,
It is quantized using the same number of quantization levels or the same word length. In the non-linear quantization, the data to be quantized is pre-processed using a non-linear function, and the pre-processed data is quantized using quantized values at equal intervals. That is, Q
If [x] represents a linear quantization of x and f (x) represents a non-linear function of x, then the non-linear quantization can be represented by Q [f (x)]. This method is specified in Layer III of the ISO / MPEG audio coding standard.

FIG. 19 shows a conventional linear quantization method Q [x] using five quantization levels. FIG. 19 shows the relationship between the data value x of the quantized data and the obtained quantization level. The five quantization levels are represented by circled integers on the horizontal axis. Each quantization level has a quantization value. For example, in FIG. 19, quantization level 2 has a quantization value of 1.0, and quantization level-1 has a quantization value of-
With 0.5. The data value x of the data to be quantized and the quantization level of the quantized value are -1.
It is represented by 0, -0.5, 0, 0.5, 1.0.

In the linear quantization shown in FIG. 19, the data value x
Is quantized to the quantization level with the quantization value closest to x. In FIG. 19, a broken line indicates a decision value between adjacent quantized values. For example, the dashed line D ₀₁ indicates a decision value between a quantized value of 0 and a quantized value of 0.5. Data values x between adjacent decision values are all quantized to the same quantization level,
Therefore, they are quantized to the same quantized value.

The data value of the data to be quantized is -1.
It may be limited to the range of 0 to 1.0. Also, the data may have data values outside this range, in which case
Data values greater than the maximum decision value D ₁₂ are quantized to the highest quantization level, and data values less than the minimum decision value D _-1-2 are quantized to the lowest quantization level. In the linear quantization shown in FIG. 19, since the data value is quantized to the nearest quantized value, the decision value is in the middle of the quantized values.

As described above, the non-linear quantization can be performed according to the prior art by first pre-processing or transforming the data using a non-linear function and then linearly quantizing the pre-processed data. it can. If f () is a non-linear function and the linear quantization of x is represented by Q [x], then
The nonlinear quantization of x can be represented by Q [f (x)].

FIG. 20 is a block diagram showing the structure of a conventional nonlinear quantizer. In FIG. 20, the data to be quantized is input to the preprocessor 21 via the input terminal 10.
The non-linear function f () is supplied from the non-linear function block 23 to the preprocessor 21. The preprocessor 21 performs preprocessing on each data based on the nonlinear function f () from the nonlinear function block 23. The obtained preprocessed data is supplied from the preprocessor 21 to the linear quantizer 22. The linear quantizer 22
Non-linear quantized data obtained by quantizing the preprocessed data to the nearest quantized value using the quantized values at equal intervals set according to the data indicating the number of quantized levels from the input terminal 11. (Hereinafter, referred to as non-linear quantized data.)
Is output from the output terminal 14.

Note that linear quantization is a special case of the above-mentioned nonlinear quantization. If the non-linear function f () is defined by f (x) = x, the data is linearly quantized, as shown in FIG. In FIG. 21, the data value a is a linear function f
The data a ′ obtained by preprocessing with (x) = x and obtained on the vertical axis is linearly quantized by Q []. Function f (x) has a slope of 1.0
Therefore, the result of Q [f (x)] is equal to Q [x]. In the example of FIG. 21, the preprocessed data value is greater than the decision value D ₁₂ between quantization level 1 and quantization level 2, so the data value a is quantized to quantization level 2.

If f (x) is a non-linear function, the quantization equation Q [f (x)] is non-linear quantization according to the prior art. FIG. 22 shows an example of this method in nonlinear quantization.
Similar to the linear quantization example described in FIG. 21, the data is preprocessed by the function f (x), and this preprocessed data is linearly quantized in the non-linearized region shown on the vertical axis of FIG. In this case, the function f (x) is a non-linear function.

[0013]

The preprocessing using the nonlinear function f (x) has the effect of changing the quantized value in the input region (horizontal axis in FIG. 22) of each quantized level. FIG. 23
Shows this effect. In FIG. 23, each quantized value q is transformed by using the inverse function g (q) of the nonlinear function f (x) (FIG. 23).
22 shows the same quantization method as that of FIG. 22 except that the input area (horizontal axis) is inversely transformed. Therefore, FIG. 23 shows the level in the input domain of the quantized value of each quantization level in the transform domain. The quantized value is
The conversion areas have equal intervals, but the input areas have unequal intervals. In this example, the slope of the non-linear function f (x) is larger when the data value corresponds to near 0 and is larger than that when the data value corresponds to near 1.0 and -1.0. Quantization performs quantization with higher accuracy on small data values.

In FIG. 24, only the input area is used, and FIG.
FIG. 23 shows the quantized values of FIG. 23 and the corresponding decision values in the same format as. The decision value is calculated in the same way as the quantized value is calculated in FIG. 23, that is, the decision value in the conversion area (vertical axis in FIG. 23) is inversely converted into the input area (horizontal axis in FIG. 23). . In this case, the decision value does not exist in the middle of the quantized value in the input area. Therefore, some data values have a non-minimized quantization error when quantized.

Further, the above-mentioned non-linear quantizer conforming to the layer III of the ISO / MPEG audio coding standard has the following problems. First, the quantizing effect of a given non-linear function f () changes according to the number of quantizing levels used to quantize the data value after being pre-processed by this non-linear function f (). However, the nonlinear quantizer conforming to the above standard uses the fixed nonlinear function f (), but the number of quantization levels changes. In a non-linear quantizer in which the number of quantization levels changes greatly, as with many high-efficiency compression systems, preprocessing using a fixed non-linear function f () gives optimum results for all values of the quantizing level. I can't.

Also, computing f (x) for each datum in the non-linear function f () requires significant time and / or signal processing capability. In order to save time and processing power, the data values are not optimal but simple non-linear function f ()
There is a problem that it must be preprocessed by using.

[0017]

SUMMARY OF THE INVENTION A nonlinear quantization method according to the present invention uses a small number of bits to generate a quantized data representing an information signal by nonlinearly quantizing the data representing the information signal. A non-linear quantization method for performing non-linear quantization of data according to a non-linear function, using a quantization level selected from a plurality of quantization levels when non-linear quantizing each data. ,
The step of receiving each data having a data value, the step of receiving word length information indicating the number of quantization levels, and the step of determining the quantization value for each quantization level of the number of quantization levels from the word length information and the non-linear function Then, the determination step of making the quantization values determined for all the quantization levels of the number of quantization levels uneven intervals, and the quantization level having the quantization value closest to the data value of each data, And a selection step of selecting the converted data.

The quantizing method according to the present invention is a quantizing method for quantizing data representing an information signal in order to generate quantized data representing an information signal using a small number of bits. In the quantization method for quantizing data by using a quantization level selected from a plurality of quantization levels when quantizing each data, a step of receiving each data having a data value, and a quantization level Receiving word length information indicating a number, and providing a quantization table set consisting of a set of quantization tables for each quantization level having a table entry for each quantization level for each number of quantization levels. , A step of selecting a quantization table for the number of quantization levels indicated by word length information from the quantization table set, and a table from the selected quantization table. Entry quantization level corresponding to the data value of each data, characterized by a step of selecting a corresponding respective quantized data.

The nonlinear quantizing method according to the present invention is a nonlinear quantizing method for nonlinearly quantizing data representing an information signal to generate quantized data representing an information signal using a small number of bits. Therefore, when nonlinearly quantizing each data, a quantization level selected from a plurality of quantization levels is used, and in the nonlinear quantization method of nonlinearly quantizing data according to the nonlinear function, each nonlinear quantization characteristic To supply a non-linear function set that determines the number of data, a step of receiving data having data values, a step of receiving word length information indicating the number of quantization levels, and a step of receiving 1 from the non-linear function set according to the word length information. There is a step of selecting two nonlinear functions and a step of preprocessing the data by applying the selected nonlinear function to each data and generating preprocessed data. It is characterized in.

The non-linear inverse quantization method according to the present invention is obtained by non-linearly quantizing the original data of the information signal according to one non-linear function selected from a plurality of non-linear functions. A non-linear inverse quantization method for generating an inverse quantized data by dequantizing non-linear quantized data having a quantization level selected from a large number of quantization levels, each of which represents an information signal with fewer bits. In, a step of supplying a set of non-linear inverse functions corresponding to each of the plurality of non-linear functions, a step of receiving non-linear quantized data, a step of receiving word length information indicating the number of quantization levels, word length information Linearly inverse-quantizing the non-linear quantized data according to to generate non-linear data,
The method includes the steps of selecting one non-linear inverse function from the non-linear inverse function set, and post-processing the non-linear data by applying the selected non-linear inverse function to each non-linear data to generate inverse quantized data. It is characterized by

[0021]

In the non-linear quantization method according to the present invention, each data having a data value is received, word length information indicating the number of quantization levels is received, and each of the quantization level numbers is received from the word length information and the non-linear function. The quantization value is determined for the quantization level, the quantization values determined for all the quantization levels of the number of quantization levels are made unevenly spaced, and the quantization level having the closest quantization value to the data value of each data is determined. , When each data is nonlinearly quantized by selecting the corresponding quantized data, the quantization level selected from a plurality of quantization levels is used, and the data is nonlinearly quantized according to the nonlinear function.

In the quantization method according to the present invention, each data having a data value is received, word length information indicating the number of quantization levels is received, and a table entry for each quantization level of each number of quantization levels is received. A quantization table set consisting of a set of quantization tables for each quantization level having is supplied, and a quantization table for the number of quantization levels indicated by the word length information is selected from the quantization table set and selected. From the quantization table, by selecting the quantization level whose table entry corresponds to the data value of each data as the corresponding quantization data, when quantizing each data, Quantize the data using the selected quantization level.

In the non-linear quantization method according to the present invention, a non-linear function set for determining each non-linear quantization characteristic is supplied,
Receiving data each having a data value, receiving word length information indicating the number of quantization levels, selecting one nonlinear function from the nonlinear function set according to the word length information, and selecting the selected nonlinear function for each data By applying pre-processing to the data and generating pre-processed data, the quantization level selected from a plurality of quantization levels is used when nonlinearly quantizing each data, and according to the nonlinear function. Non-linear quantization of data.

In the non-linear inverse quantization method according to the present invention, a set of non-linear inverse functions corresponding to each of a plurality of non-linear functions is supplied, non-linear quantized data is received, and word length information indicating the number of quantization levels is provided. Receive, linearly inverse quantize the non-linear quantized data according to the word length information, generate non-linear data, select one non-linear inverse function from the non-linear inverse function set, and select the selected non-linear inverse function to each non-linear data Non-linear quantization of the original data of the information signal according to one non-linear function selected from a plurality of non-linear functions by applying post-processing to the non-linear data by applying and generating inverse quantized data The non-linear quantization data, which is obtained by the above method, represents the information signal with fewer bits than the original data, and has a quantization level selected from a large number of quantization levels. The inversely quantized, and generates the inverse quantized data.

[0025]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the non-linear quantization method and non-linear inverse quantization method according to the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a non-linear quantization device 30 for carrying out the first non-linear quantization method according to the present invention. The configurations of the non-linear quantizer 30, other non-linear quantizers described later, and non-linear inverse quantizer are shown in block diagrams. For example, the quantized value calculator 31, the quantizer 32, shown in these block diagrams,
The functional elements forming the nonlinear quantizer such as the nonlinear function block 33 can be realized by using an appropriate large scale integrated circuit, or by using a small scale integrated circuit or individual components. Further, by programming a microcomputer or a digital signal processor and using an auxiliary circuit as needed, the functional elements forming the nonlinear quantizer or the nonlinear inverse quantizer may be realized.

In the non-linear quantizer 30, the quantizer 32 is supplied with the data to be quantized via the input terminal 10 and is supplied with a set of quantized values from the quantized value calculator 31. Each data is quantized to the quantization level of the nearest quantization value. The quantizer 32 outputs the obtained quantized data via the output terminal 14. The data supplied via the input terminal 10 is usually a data block or data set, such as a set of spectral coefficients, which are quantized using the same number of quantization levels or the same word length.

The non-linear function f () is supplied from the non-linear function block 33 to the quantized value calculator 31, and the quantized value calculator 31 receives an input indicating the number of quantization levels together with the data to be quantized. Supplied. Quantization value calculator 31
Calculates a quantization value for each quantization level and supplies the obtained quantization value to the quantizer 32. The number of calculated quantization values is equal to the number of quantization levels indicated by the input.
Instead of the input indicating the number of quantization levels, an input indicating the word length together with the data to be quantized may be supplied to the quantized value calculator 31 via the input terminal 11.

FIG. 2 is a flow chart for explaining in detail the calculation of the quantized value of each quantized level by the quantized value calculator 31. In the process shown in FIG. 2, first,
The linear quantization value x in the transform domain is equal to each quantization level i.
Is calculated. Then, the quantized value q [i] in the input area is calculated as q [i] = g (x). here,
g () is the inverse function of the nonlinear function f (), that is, g (f (x))
= X.

In step S0, the processing is started, and in step S1, the input indicating the non-linear function f () is supplied from the non-linear function block 33. In step S2,
The inverse function g () of the nonlinear function f () is calculated from this nonlinear function f (). In step S3, the number of quantization levels l
An input indicating ev is supplied via the input terminal 11.
In step S4, the quantization level i is set to the lowest quantization level using the following equation (1).

I =-(lev-1) / 2 (1) Quantized value q [i] in the input region for each quantization level i
Is calculated by a loop including steps S5, S6, S7, and S8. This is performed until the quantized values for all the quantized levels are calculated in step S7.

That is, in step S5, the linear quantized value x in the transform domain for the quantized level i is
It is calculated using equation (2).

X = (2 × i) / (lev-1) (2) In step S6, the quantized value q [i] in the input region
Is the quantized value in the transform domain using Equation (3),
It is calculated from the inverse function g ().

Q [i] = g (x) (3) In step S7, it is determined whether or not the value i corresponds to the highest quantization level obtained by the equation (4). It is determined whether or not the quantized value for the quantization level has been calculated.

I = (lev-1) / 2 (4) In step S7, when it is determined that the value i does not correspond to the highest quantization level, the process proceeds to step S8, the value i is incremented by 1, In step S5, the loop is repeated.

If it is determined in step S7 that the value i corresponds to the highest quantization level, it indicates that the quantization values have been determined for all the quantization levels, and the process proceeds to step S9. In step S9, the quantized value calculator 31
The quantized value calculated by is supplied to the quantizer 32 and is used for data quantization. Next in step S10
Go to and return to the main routine.

The quantizer 32 selects a quantum selected from the quantization levels indicated by the input indicating the number of quantization levels (or the number of quantization levels corresponding to the word length input) supplied via the input terminal 11. Each data is quantized by outputting the quantization level from the output terminal 14. The quantizer 32 is supplied with the quantized value for each quantized level from the quantized level calculator 31 and is used to quantize the data to the nearest quantized value. For example, the quantizer 32 determines whether any of the quantized values supplied from the quantized value calculator 31 is closest to the value of each data to be quantized, and the quantized value corresponding to the quantized value is determined. The quantization level is output from the output terminal 14 as one of the quantized data. Each quantization level output from the output terminal 14 has the number of bits indicated by the word length input supplied to the quantization value calculator 31 from the input terminal 11 (or corresponds to the input indicating the number of quantization levels). With the number of bits). The quantizer 32, for example,
By determining the difference between one of the data and each quantized value, it is determined which of the quantized values is closest to each data value, and the quantization level corresponding to the quantized value with the smallest difference is output. It can be output from the terminal 14.

The quantized value which is the result of the processing for the quantized level obtained by the quantized value calculator 31 is the quantized value in the conventional non-linear quantizer using the same non-linear function f () (see FIG. 23). (Shown in). However, in the non-linear quantization by the first non-linear quantization method according to the present invention, the quantizer 32 processes in the input region, and each data corresponds to the nearest one of the equally-quantized quantized values. Since it is quantized to a quantization level which is set, the quantization is different from the quantization shown in FIG. As a result, unlike the decision values of the conventional non-linear quantizers described with reference to FIGS. 23 and 24, decision values in the middle of adjacent quantized values are obtained. As a result, in the first nonlinear quantization method according to the present invention, the quantization error (hereinafter referred to as the quantization error) is less than or equal to the quantization error generated by the conventional nonlinear quantization.

In FIG. 3, an example of a non-linear function is shown by a solid line, and a linear function is shown by a dotted line for comparison. 4 and 5
Shows the quantization error generated by the nonlinear quantization method according to the present invention (solid line) in comparison with the quantization error generated by conventional nonlinear quantization (dotted line). Here, in each case, the example of the nonlinear quantization function shown by the solid line in FIG. 3 is used. FIG. 4 shows the quantization error that occurs when data values in the range -1.0 to +1.0 are quantized with three quantization levels. According to the non-linear quantization method according to the present invention, it can be seen that the quantization error at data values near −0.4 and +0.4 is extremely small. FIG. 5 shows the quantization error that occurs when data values in the same range are quantized with seven quantization levels. In any case, the quantization error generated by the nonlinear quantization method according to the present invention is less than or equal to the quantization error generated by the conventional nonlinear quantization method.

The quantized value calculator 31 does not supply the quantized value for each quantized level to the quantizer 32, but instead the input range (input range) of the number of quantized levels for each quantized level.
May be calculated and supplied to the quantizer 32. Thereby, the configuration of the quantizer 32 can be simplified. The quantized value calculator 31 sets a determined value in the middle of adjacent quantized values from the quantized values calculated as described above. The quantization error is minimized by setting the decision value in the middle of the adjacent quantization values. The quantization value calculator 31 can derive the input range of each quantization level from the calculated determination values of the adjacent quantization levels. The quantized value calculator 31 supplies the input range of each quantized level to the quantizer 32 instead of each quantized value. The quantizer 32 simply determines which of the input ranges supplied from the quantization value calculator 31 contains the data value of each piece of data to be quantized, and outputs each quantization level to the output terminal 14. The data is quantized by outputting from.

Next, an embodiment in which the first nonlinear quantization method according to the present invention is applied to an audio signal transmission system or recording system will be described with reference to FIGS. 6 to 8. The nonlinear quantization method according to the present invention is not limited to this embodiment, and can be applied to other applications of the nonlinear quantization device, and the above-mentioned effects can be obtained in these applications. You can For example, in the case of performing non-linear quantization of a coefficient obtained by orthogonally transforming a square block or a rectangular block of a moving image signal, or a square block or a rectangular block of a prediction error generated by motion compensation for a moving image signal, the present invention The non-linear quantization method according to the above can be applied.

FIG. 6 shows the configuration of the audio signal transmission / recording system 110. In this system 110, in order to reduce the recording capacity required for the recording medium and / or the bit rate required for the transmission medium, the audio signal is
It is compressed before it is recorded or transmitted. The audio input signal is compressed in the audio compression device 111, and the resulting compressed signal is supplied to the transmission or recording medium 113. The compressed signal reproduced from the transmission or recording medium 113 is decompressed in the audio decompression device 112, and the audio signal is output. Transmission / recording system 110
Normally operates according to the psychoacoustic characteristics of humans to minimize the audibility of errors caused by compression and decompression of audio signals.

FIG. 7 is a block diagram showing the configuration of the audio compression apparatus 111 which operates according to the psychoacoustic characteristics of humans. An audio input signal is input to the time-frequency analyzer 121, and the time-frequency analyzer 121
The audio input signal is divided into a set of spectral components (hereinafter referred to as a spectral component set) in the time domain and the frequency domain. For example, the time-frequency analyzer 121 divides the audio input signal into blocks in the time domain, and orthogonally transforms the samples of the audio input signal of each block from the time domain to the frequency domain to generate spectral components. Also, the time-frequency analyzer 121
The audio input signal may be divided into a plurality of frequency bands by using, for example, a multi-filter, and the signal component of each frequency band may be divided into spectral component sets in the time domain. The time-frequency analyzer 121 supplies the spectral component set to the bit allocator 122 and the spectral quantizer 123.

In the audio compression apparatus 111 shown in FIG. 7, each spectrum component set represented by using a predetermined number of bits is subjected to block floating processing before each spectrum component is quantized. Increase the accuracy of the spectral components of. The bit allocator 122 determines the scale factor for each spectral component set, and the spectral quantizer 123 divides each spectral component of the spectral component set by the scale factor. Then, the obtained mantissa is quantized. In fact,
As will be described later, the spectral components of each spectral component set are band-divided by frequency, and the bit allocator 122
Determines the scale factor for each frequency band,
The spectrum quantizer 123 performs block floating processing on the spectrum component of each frequency band using each scale factor.

The bit allocator 122 determines a word length and a scale factor for quantizing the spectral coefficient of each spectral component set, and supplies this word length and the scale factor to the spectral quantizer 123. The spectrum quantizer 123 uses the scale factor and the word length,
For example, the block floating process and the non-linear quantization are used to non-linearly quantize the spectrum components of each spectrum component set, and a set of non-linear quantized spectrum components (hereinafter referred to as non-linear quantized spectrum component set) is generated. Then, the bit allocator 122 supplies the bit allocation parameter to the multiplexer 125,
The spectrum quantizer 123 supplies the nonlinear quantized spectrum component set to the multiplexer 125. The multiplexer 125 multiplexes the bit allocation parameter and the non-linear quantized spectrum component to generate a compressed signal, and supplies this compressed signal to the transmission / recording medium 113.

The bit allocation by the bit allocator 122 is schematically represented by the psychoacoustic model 124 so that the listener hardly hears the quantization noise generated by the non-linear quantization by the spectrum quantizer 123. Based on the psychoacoustic model. The psychoacoustic model used for this bit allocation usually consists of data such as a detection threshold curve, an equal loudness curve, a simultaneous masking characteristic, a temporal masking characteristic, and the like.

In the audio compression apparatus 111 for carrying out the non-linear quantization method according to the present invention, the spectrum quantizer 123 shown in FIG. 7 has the first non-linear quantization method according to the present invention described with reference to FIGS. May be used. Further, instead of the first non-linear quantization method in the spectrum quantizer 123, another embodiment of the non-linear quantization method described below may be used. The spectrum quantizer 123 will be described with reference to FIG. 7.

In the spectrum quantizer 123, the block floating processor 132 is supplied with the spectrum component set from the time-frequency analyzer 121,
The bit allocator 122 supplies the scale factor for each spectral component set. The block floating processor 132 performs a block floating process on each spectral component set by dividing the spectral component of the spectral component set by each scale factor. The set of spectral components subjected to the block floating process (hereinafter referred to as the block floating spectral component set) obtained by this is the input terminal 1.
It is supplied to the non-linear quantizer 133 via 0. Also,
The non-linear quantizer 133 is supplied with the word length for each spectrum component set from the bit allocator 122 via the input terminal 11, and the non-linear quantizer 133 supplies each block floating spectrum according to the word length. Non-linear quantization of the component set. The non-linear quantizer 133 supplies the obtained set of quantized spectral components (hereinafter referred to as a quantized spectral component set) to the multiplexer 125 via the output terminal 14.

A non-linear quantizer 13 is used as the non-linear quantizer for executing the first non-linear quantizing method described in FIG.
When used as 3, the quantizer 32 quantizes each block floating spectral component of the block floating spectral component set supplied from the block floating processor 132 via the input terminal 10. The quantizer 32 quantizes the block floating spectrum component to the closest quantized value among the quantized value numbers calculated by the quantized value calculator 31. The quantized value calculator 31 uses the non-linear function from the non-linear function block 33 to calculate the number of quantized values from the input indicating each word length supplied via the input terminal 11. The quantizer 32 supplies the obtained quantized spectral component set to the multiplexer 125 via the output terminal 14.

In audio compressor 111, time-frequency analyzer 121 is not limited to this,
It may be a hybrid structure of transform and subband coding. For example, FIG. 8 shows an example of a time-frequency analyzer 121 to which an audio adaptive conversion coding (ATRAC) technique used for generating a compressed signal for compressing a PCM audio signal and recording it on a magneto-optical disk is applied. The input terminal 151 has, for example, a frequency band of 0 to 20 kHz.
PCM audio input signal is supplied. This audio input signal is set to 0H by the band division filter 153.
It is divided into a band signal of z to 10 kHz and a high band signal of 10 to 20 kHz. Further, the band signal of 0 Hz to 10 kHz is converted to 0 Hz to 5 kHz by the band division filter 155.
Is divided into a low frequency signal and a medium frequency signal of 5 kHz to 10 kHz. It is preferable to use a quadrature mirror filter (QMF) as the band division filters 153 and 155.

Further, the PCM audio input signal is 0 Hz.
When it has a band of ~ 22 kHz, the high frequency signal is 11 kHz.
It has a band of z to 22 kHz, and the mid-range signal is 5.5 kHz.
It has a band of up to 11 kHz and a low frequency signal is from 0 Hz to 5.5.
It has a band of kHz.

The high band signal is supplied from the band division filter 153 to the high band orthogonal transformation circuit 157, the middle band signal is supplied from the band division filter 155 to the middle band orthogonal transformation circuit 159, and the low band signal is band divided. It is supplied from the filter 155 to the low frequency orthogonal transform circuit 161.

These orthogonal transformation circuits 157, 159, 1
61 is a modified discrete cosine transform (MDCT)
A circuit is preferable, and the frames of each band signal are orthogonally transformed to collectively generate respective spectral component sets, which are supplied to the bit allocator 122 and the spectral quantizer 123. That is, these orthogonal transformation circuits 1
The spectral component sets collectively generated by 57, 159, and 161 are supplied to the bit allocator 122 and the spectral quantizer 123 via the output terminal 165.

The audio input signal supplied to the input terminal 151 is conceptually divided into frames of a predetermined number of samples. The orthogonal transformation circuits 157, 159, 161 orthogonally transform the frames of the three band signals obtained from the respective frames of the audio input signal, and collectively generate a spectrum component set. To further increase flexibility and eliminate audible time domain effects such as noise modulation, orthogonal transform circuits 157, 159, 161 orthogonally transform each frame of a band signal of blocks having variable block lengths. The block length in which each band signal is orthogonally transformed is equal to the frame length divided by 2 ⁿ . Here, n is a positive integer of 0 or more. The block length by which each frame of the audio input signal is orthogonally transformed in the three frequency bands is determined by the block length determination block 163 based on the dynamic spectrum characteristic of the digital input signal. Generally, a high frequency signal is orthogonally transformed into a block having a block length shorter than that of a low frequency signal. Further, the block length determination block 163 shortens the block length of the frame of the digital input signal whose level fluctuates as compared with the frame of the digital input signal whose level is relatively constant.

As another configuration example of the time-frequency analyzer 121 of the audio compression device 111, when time-frequency conversion conforming to layer III of the MPEG audio coding standard is used, the audio input signal is processed by a plurality of filters. After being divided into 32 sub-bands having the same bandwidth, MDCT conversion is performed.

In the above-described audio compression device 111, the spectral components of the spectral component set may be frequency band divided before being quantized by the spectral quantizer 123. In this case, the bit allocator 122 allocates a scale factor and a word length for quantizing the spectrum component of each frequency band. The block floating processor 132 performs block floating processing on the spectrum component of each frequency band, and the nonlinear quantizer 13
3 nonlinearly quantizes the spectrum component of each frequency band according to the word length of each frequency band. When the non-linear quantizer 133 implements the first non-linear quantization method according to the present invention, the quantized value calculator 31 determines that the quantized level of each quantized level indicated by the word length of each frequency band. Compute a quantized value for. Then, the quantizer 32 calculates the block floating spectrum component of each frequency band as
Quantize to the closest quantized value obtained by the quantized value calculator 31.

In the transmission / recording system 110, the audio decompressor 112 is the audio decompressor 11.
As long as the second non-linear dequantizer operates by using the inverse function of the non-linear function by the non-linear quantizer 133 of the audio compression device 111, the compressed signal reproduced from the transmission / recording medium 113 can be decompressed accurately. . For the decompressor, it is irrelevant whether the quantized spectral components of the compressed signal are quantized with the non-linear quantization according to the invention or with the conventional non-linear quantization.
This is because the quantized values of the quantized levels are equal in both cases. However, when the quantized spectral component of the compressed signal is non-linearly quantized by the non-linear quantization method according to the present invention, the quantization error generated when the audio decompression device 112 decompresses the compressed signal is the conventional non-linear quantization. It is small as compared with the case of using the method. This is because the decision value used in the non-linear quantization method according to the present invention is different from the decision value in the conventional non-linear quantization using the same non-linear function.

The non-linear quantizer described with reference to FIGS. 1 and 2 for carrying out the first non-linear quantizing method according to the present invention uses a fixed non-linear function f () and changes the word length or the number of quantizing levels. Each time, the quantized value is newly calculated. This includes
It requires considerable computing power. Moreover, the fixed nonlinear function f
The quantization effect of () changes depending on the number of quantization levels.

Therefore, the non-linear quantizing device for carrying out the second non-linear quantizing method according to the present invention has a lookup table for outputting the quantized output value T [x] by designating the input range x. It is used according to the quantization level. Therefore, it is not necessary to perform preprocessing using the non-linear function f () when performing the quantization processing. Then, the non-linearity of the quantization is determined by the distribution of the input range defined by T []. These input ranges can be set arbitrarily without changing the complexity of the quantization processing. An independent look-up table T [] must be specified according to the number of quantization levels, but the input range and the quantized value in each look-up table are based on a single nonlinear function f (). In this case, the input range and the quantization value are the same as those of the conventional nonlinear quantizer which uses the same nonlinear function as the quantization error generated by the nonlinear quantizer which implements the second nonlinear quantization method according to the present invention. It can be selected to be small in comparison.

FIG. 9 is a block diagram showing the configuration of a non-linear quantization device 40 for carrying out the second non-linear quantization method according to the present invention. In the second non-linear quantization method, the quantized data is non-linear quantized by using a look-up table selected from a look-up table set including a plurality of look-up tables, and the non-linear quantized data (hereinafter , Non-linear quantized data) is output.

In the non-linear quantizer 40 shown in FIG. 9, the table quantizer 42 is supplied with the data to be quantized through the input terminal 10 and uses the look-up table supplied from the table selector 41. The data is quantized to generate non-linear quantized data, and the obtained quantized data is output via the output terminal 14. The data supplied via the input terminal 10 is usually a data block or a data set such as a spectral component set, which is quantized using the same number of quantization levels or the same word length.

The table selector 41 stores the look-up table set used in the table quantizer 42. An input indicating the number of quantization levels used for data quantization is supplied to the table selector 41 via the input terminal 11. The table selector 41 selects one look-up table from the stored look-up table sets according to the input indicating the number of quantization levels, and supplies the selected look-up table to the table quantizer 42. To do. Instead of the input indicating the number of quantization levels, an input indicating the word length used for data quantization may be supplied to the table selector 41.

Normally, the lookup table of the lookup table set stored in the table selector 41 is obtained from the nonlinear function f () supplied from the nonlinear function block 43, as shown in FIG. The look-up table set stored in the table selector 41 corresponding to a predetermined non-linear function is composed of a look-up table for each value of the quantization level number (or word length). Each lookup table is composed of table entries for each quantization level in the number of quantization levels indicated by the quantization level data supplied via the input terminal 11. Each table entry consists of an input range for the quantization level and a quantized output value.

The set of input ranges for all table entries in each lookup table must include the range of data values of the quantized data. This allows a table entry to exist for any data value in the range of data values. Moreover, the input ranges of the table entries in each lookup table must not overlap so that there is no more than one table entry for any data value in the range of data values. That is, in each lookup table, there is only one table entry having an input range corresponding to each data value in the data value range.

FIG. 10 shows an example of a look-up table for five quantization levels. In order to simplify the table, the input data range will be represented by a decimal number having a precision of 0.01 instead of a binary number having a precision represented by the least significant bit. In fact, the input range is
It is represented by a binary number having a bit number equal to or larger than the bit number of each data to be quantized. In order for the input range to cover the entire range of data values without duplication, the maximum of each data range may differ from the minimum of the next highest input range by the least significant bit.

Returning to FIG. 9, the table selector 41 selects an appropriate one from the stored look-up table set in accordance with the input indicating the number of quantization levels used for the quantization of the data supplied through the input terminal 14. Select a look-up table and supply the selected look-up table to the table quantizer 42. And the table quantizer 4
2 determines, for each piece of data to be quantized, one input range in the selected look-up table in which that data value resides, and the corresponding quantized output value from that look-up table is set to the non-linear quantized data. Output as one of the above.

The nonlinear quantizer 40 shown in FIG. 9 is highly flexible because it can quantize using any nonlinear function f (). In addition, the nonlinear quantizer 4
With 0, even if the nonlinear function f () becomes complicated, the structure of itself does not become complicated.

When generating the look-up table stored in the table selector 41, the limitation of the input range is determined by a desired method such as conversion from the non-linear function f () or direct calculation. can do.

When the input range is directly calculated, the quantization error generated by the nonlinear quantizer 40 is similar to the quantization error generated by the first nonlinear quantization method according to the present invention using the same nonlinear function. There are few quantization errors.
This quantization error is smaller than the quantization error generated by the conventional nonlinear quantizer shown in FIG. 20, which uses the same nonlinear function. As shown in FIG. 24, the decision value provided between the quantization levels of the conventional nonlinear quantizer shown in FIG. 20 is not located at the midpoint of the quantized value. As a result, there are data values that are not quantized to the closest quantized value. This is a problem of the conventional non-linear quantizer. On the other hand, in the non-linear quantization apparatus 40 that implements the second non-linear quantization method according to the present invention, the look-up table can be selected so that the determined value is located exactly at the midpoint of the quantized value. . Here, in FIG. 4 and FIG.
Regarding the quantization using one quantization level and the quantization using seven quantization levels, the quantization error generated by the second non-linear quantization method according to the present invention is shown by a solid line. Further, in the same figure, for comparison, a quantization error due to conventional non-linear quantization is shown by a dotted line.

As described above, instead of the non-linear quantizer 40 for implementing the second non-linear quantizing method according to the present invention, instead of the non-linear quantizing device 30 for implementing the first non-linear quantizing method according to the present invention. In addition, it may be used as the non-linear quantizer 133 in the audio compression apparatus 111 of the transmission / recording system 110 shown in FIGS. The second non-linear quantization method according to the present invention is not limited to this embodiment and can be applied to other non-linear quantization applications, and the above-mentioned effects can be obtained in these applications as well. Obtainable.

When the non-linear quantizer for carrying out the second non-linear quantizing method according to the present invention is used as the non-linear quantizer 133, the table selector 41 includes the bit allocator 122 via the input terminal 11 for each. The word length for each spectral component set is supplied, and the block quantizer 42 is supplied with the block floating spectral component set from the block floating processor 132 via the input terminal 10. The table quantizer 42 quantizes the block floating spectral components of each block floating spectral component set using the lookup table selected by the table selector 41 according to the word length of each spectral component set from the bit allocator 122. Turn into. The table quantizer 42 supplies the obtained quantized spectral component set to the multiplexer 125 via the output terminal 14.

When the spectral component set is divided into frequency bands before quantization, the bit allocator 122 allocates the scale factor and word length used for quantization of the spectral components of each frequency band, and the block floating processor 132 Block floating processing is performed on the spectral components of each frequency band. In this case, the table selector 41 selects a lookup table for each frequency band and supplies it to the table quantizer 42, and the table quantizer 42 uses the lookup table for each frequency band from the table selector 41. Then, the block floating spectrum component of each frequency band is quantized.
The table selector 41 selects a look-up table according to each different word length when the spectrum component set is quantized, and a mutual relation indicating which look-up table is used for the quantization of each frequency band. These look-up tables may be supplied to the table quantizer 42 along with the data. The table quantizer 42 quantizes the spectral component of each frequency band using a look-up table indicated by the mutual relationship data.

In the above-described embodiment, the quantized value of each quantized level may be stored in the table of the table selector 41 instead of the input range of each quantized value. In this case, the table quantizer 42 quantizes each data to the nearest quantized value.

Next, a non-linear quantizer 50 for carrying out the third non-linear quantizing method according to the present invention will be described with reference to FIG.
This will be described with reference to 1, 12, and 13. This non-linear quantizer 50 is based on the non-linear quantizer shown in FIG. 20, pre-processes the quantized data according to a non-linear function, and then performs linear quantization. In this non-linear quantizer 50, the non-linear quantizer shown in FIG. 20 is used by using the selected polygonal line approximation value (piecewise linear approximation) of the non-linear function instead of the non-linear function used in the pre-processing of the data. There are fewer quantization errors than the ones that occur. The selected polygonal line approximation value corresponds to the number of quantization levels (or word length) in the quantization after the preprocessing of the data. The preprocessed data obtained by the preprocessing using the selected polygonal line approximation value is linearly quantized corresponding to the number of quantization levels having equally-spaced quantized values, and nonlinear quantized data is generated. The third non-linear quantization method according to the present invention has less quantization error than the conventional non-linear quantization using the same prototype non-linear function.

To perform quantization using different numbers of quantization levels (or different word lengths), a set of polygonal line approximations of the prototype non-linear function is provided. Each polygonal line approximation value of the nonlinear function is a different polygonal line approximation value of the prototype nonlinear function corresponding to each quantization level number (or each word length). Each linear approximation of the non-linear function is equal to the original non-linear function at each quantized value and is linear between adjacent quantized values.
11 and 12, the prototype non-linear function is shown by a dotted line, and the polygonal line approximation values of the corresponding five quantization levels and seven quantization levels are shown by a solid line.

FIG. 13 is a block diagram showing the configuration of a non-linear quantization device 50 for implementing the third non-linear quantization method according to the present invention. In this nonlinear quantizer 50,
The quantized data is supplied to the preprocessor 21 via the input terminal 10. The quantized data is usually a data block or a data set such as a spectrum coefficient set, and is quantized using the same number of quantization levels or the same word length. The preprocessor 21 performs preprocessing on each data according to the polygonal line approximation value of the nonlinear function from the polygonal line approximation value calculator 51, and supplies the obtained preprocessed data to the linear quantizer 22. The linear quantizer 22 uses equally-quantized quantization values set according to an input indicating the number of quantization levels supplied via the input terminal 11. The linear quantizer 22 quantizes the preprocessed data to the quantization level corresponding to the closest quantization value among the equally-quantized quantization values, and outputs the obtained nonlinear quantization data via the output terminal 14.

The polygonal line approximation value calculator 51 is supplied with the original nonlinear function from the nonlinear function block 53, and
An input indicating the number of quantization levels used to quantize data is supplied via an input terminal 11. The polygonal line approximation value calculator 51 generates a polygonal line approximation value of the non-linear function corresponding to the number of quantization levels (or word length) according to the input indicating the number of quantization levels and the prototype non-linear function, and causes the preprocessor 21 to generate the polygonal line approximation value. Supply. Instead of the input indicating the number of quantization levels, an input indicating the word length in the data quantization may be alternately supplied to the polygonal line approximation value calculator 51 and the linear quantizer 22.

Compared with the conventional non-linear quantization using the same prototype non-linear function, the third non-linear quantizing method according to the present invention has less quantizing error. It can be understood by comparing the determined values between the converted values. When the non-linear function of FIG. 22 is used as the prototype non-linear function, the quantized value is the same as the quantized value shown in FIGS. 23 and 24. That is, the decision value shown by the dotted line in FIG. 24 is not located at the midpoint of the adjacent quantized values. However, the polygonal line approximation value of the nonlinear function used in the third nonlinear quantization method according to the present invention is linear between adjacent quantized values, and as shown in FIG. It exists at the midpoint of the value. This allows the quantization error to be reduced without changing the non-linearity of the quantized data points.

The quantized data generated by the third non-linear quantization method according to the present invention can be dequantized by using a conventional non-linear dequantization device based on the inverse function of the original non-linear function. Such inverse quantization can reduce the quantization error. For example, when the nonlinear function shown by the solid line in FIG. 3 is used as the prototype nonlinear function, the quantized data obtained by the quantization by the third nonlinear quantization method according to the present invention is the inverse of the prototype nonlinear function. When dequantized by a conventional dequantizer using a function, the quantization error is 3 quantization levels, 7
The solid line is shown in FIGS. 4 and 5 for one quantization level. For comparison, FIGS. 4 and 5 show a quantization error obtained by quantization of data using conventional non-linear quantization by a dotted line. In any case, the quantization error by the nonlinear quantization method according to the present invention is equal to or less than the quantization error by the conventional nonlinear quantization.

As described above, instead of the non-linear quantizer 30 implementing the first non-linear quantizing method according to the present invention, a non-linear quantizing device implementing the third non-linear quantizing method according to the present invention is used. It may be used as the non-linear quantizer 133 in the audio compression apparatus 111 of the transmission / recording system shown in FIGS. The third aspect of the present invention
The non-linear quantization method of (1) is not limited to this application, and the above-mentioned effect can be obtained even when applied to other non-linear quantization.

When the non-linear quantizer for carrying out the third non-linear quantizing method according to the present invention is used as the non-linear quantizer 133, the polygonal line approximation value calculator 51 and the linear quantizer 22 are assigned with bit allocation. The word length of each spectral component set is supplied from the converter 122 via the input terminal 11, and the pre-processor 21 is supplied with the block floating spectral component set from the block floating processor 132 via the input terminal 10. The preprocessor 21 uses the polygonal line approximation value of the non-linear function output by the polygonal line approximation value calculator 51 according to the word length of each spectrum component set from the bit allocator 122 and uses the block floating of each block floating spectrum component set. Preprocess the spectral components. Linear quantizer 2
2 linearly quantizes the resulting preprocessed spectral components,
Generate a set of non-linear quantized spectral components. The linear quantizer 22 supplies this nonlinear quantized spectral component set to the multiplexer 125 via the output terminal 14.

As described above, if the spectral component set is frequency band divided before quantization, the bit allocator 1
22 assigns a scale factor and word length for quantizing the spectral component of each frequency band. The block floating processor 132 performs block floating processing on the spectrum component of each frequency band. In this case, the polygonal line approximation value calculator 51 generates the polygonal line approximation value of the non-linear function for each frequency band, and the preprocessor 2
Feed to 1. The preprocessor 21 uses the polygonal line approximation value from the polygonal line approximation value calculator 51 to pre-process the block floating spectrum component of each frequency band.

The above-mentioned nonlinear quantizer and the conventional nonlinear quantizer use the same nonlinear function f () regardless of the number of quantization levels (or word length). This limits the performance of the quantizer. This is because the quantization effect of the predetermined function changes according to the number of quantization levels (or word length) used for data quantization. Here, in the non-linear quantization apparatus for implementing the fourth non-linear quantization method according to the present invention, the non-linear function f () is calculated according to the number of quantization levels (or word length) assigned for data quantization. It is selected from a non-linear function set consisting of multiple non-linear functions. The non-linear function set consists of a set of polygonal line approximations of the non-linear function of each prototype that differ according to the number of quantization levels assigned for quantizing the data.

FIG. 15 shows an example of a nonlinear function set used in the fourth nonlinear quantization method according to the present invention. Since this non-linear function set consists of seven types of functions, it is suitable for a quantization method in which the number of quantization levels has seven allowable values. In addition, this non-linear function set can be used in a non-linear quantization method having an allowable value of eight or more quantization levels when two or more of the values of the number of quantization levels use the same non-linear function.

In the non-linear function set shown in FIG.
The function closest to the linear function is preferably used for the smallest number of quantization levels, ie the smallest quantized word length. A non-linear function having high non-linearity is selected as the number of quantization levels increases or as the quantization word length increases. This makes it possible to perform quantization with high non-linearity using a long quantized word length, and prevent distortion that occurs when a function with high non-linearity is used with a short quantized word length.

For example, a nonlinear function such as y = x ^0.55 can be used with a word length of about 2 bits or 3 bits, and a nonlinear function such as y = x ^0.4 can be used with a word length of 4 bits. You can

FIG. 16 is a lock diagram showing the configuration of a non-linear quantization apparatus 60 for implementing the fourth non-linear quantization method according to the present invention. In the non-linear quantizer 60, the quantized data is pre-processed by the pre-processor 21 according to the non-linear function obtained by the non-linear function selector 62. The obtained preprocessed data is supplied from the preprocessor 21 to the linear quantizer 22. The linear quantizer 22 linearly quantizes this data to generate nonlinear quantized data, and outputs it through the output terminal 14.

The non-linear function block 63 stores a different set of non-linear functions. Further, the non-linear function block 63 may store the polygonal line approximation value of each non-linear function instead of the non-linear function. By using a polygonal line approximation value of each non-linear function instead of the non-linear function, it is possible to reduce the quantization error by changing the decision value so that it exists at the midpoint of the adjacent quantized values. For simplicity, the following “non-linear function” indicates a non-linear function and its polygonal line approximation value, unless otherwise specified.

The non-linear function selector 62 selects one of the non-linear functions used for non-linear quantization of data from the non-linear function set stored in the non-linear function block 63 using a predetermined non-linear function selection method. The non-linear function selector 62 outputs the selected non-linear function to the preprocessor 2
Feed to 1. The preprocessor 21 is supplied with data via the input terminal 10, performs preprocessing on each data according to the selected nonlinear function, and supplies the obtained preprocessed data to the linear quantizer 22. The linear quantizer 22 linearly quantizes the preprocessed data and outputs the non-linear quantized data via the output terminal 14.

In the fourth embodiment using the non-linear quantizer 60, generally, the number of permissible quantization levels (or each permissible word length) is included in the non-linear function set stored in the non-linear function block 63. Includes a nonlinear function for. An input indicating the number of quantization levels used for data quantization is supplied to the linear quantizer 22 and the non-linear function selector 62 via the input terminal 11. The non-linear function selector 62 selects one of the non-linear functions stored in the non-linear function block 63 according to the input indicating the number of quantization levels (or word length) used for data quantization, and the preprocessor 21.

In another example, the non-linear function selector 62 may be constructed so that two or more of the values of the number of quantization levels select the same non-linear function. But,
When the polygonal line approximation value of the non-linear function is used instead of the non-linear function, different polygonal line approximation values are required even if two or more of the quantization level values have the same prototype non-linear function.

As described above, in place of the non-linear quantizing device implementing the first non-linear quantizing method according to the present invention, a non-linear quantizing device implementing the fourth non-linear quantizing method according to the present invention is The transmission / recording system 1 shown in FIGS.
10 may be used as the non-linear quantizer 133 in the audio compression apparatus 111. The fourth non-linear quantization method according to the present invention is not limited to this application and can be applied to other non-linear quantization, and in this case, the above-described effect can be obtained. .

When the non-linear quantizer for carrying out the fourth non-linear quantizing method according to the present invention is used as the non-linear quantizer 133, the non-linear function selector 62 is connected to the non-linear function selector 62 from the bit allocator 122 via the input terminal 11. Thus, the word length for each spectral component set is provided. Preprocessor 2
1 is supplied with the block floating spectral component set from the block floating processor 132 via the input terminal 10. The preprocessor 21
According to the word length from the bit allocator 122, the non-linear function selected by the non-linear function selector 62 is used to pre-process the block floating spectral components of each block floating spectral component set. The linear quantizer 22 linearly quantizes the obtained preprocessed spectral components to generate a non-linear quantized spectral component set,
It is supplied to the multiplexer 125 via the output terminal 14.

Assuming that the spectral component set is divided into frequency bands before quantization, as described above, the bit allocator 122 determines the scale factor and word length for quantizing the spectral components of each frequency band. Assign The block floating processor 132 performs block floating processing on the spectrum component of each frequency band. In this case, the nonlinear function selector 62 selects a nonlinear function from the nonlinear function set stored in the nonlinear function block 63 for each frequency band. The non-linear function selector 62 supplies the non-linear function selected for each frequency band to the preprocessor 21, and the preprocessor 21
Uses the non-linear function from the non-linear function selector 62 to pre-process the block floating spectrum component in each frequency band.

The first and second nonlinear quantization methods according to the present invention can use different nonlinear functions depending on the word length or the number of quantization levels. In the non-linear quantization apparatus for implementing the first non-linear quantization method according to the present invention, as shown in FIG.
The non-linear function block 33 can be supplied to the non-linear function block 33. The non-linear function block 33 can supply the non-linear function that changes depending on the number of quantization levels or the word length to the quantization value calculator 31.

In the non-linear quantization apparatus for carrying out the second non-linear quantization method according to the present invention, as shown in FIG.
The data entry in the table of the table set stored in the table selector 41 can be calculated by using a non-linear function that differs depending on the number of quantization levels. The present invention automatically selects different nonlinear functions according to the number of quantization levels by selecting a lookup table according to an input indicating the number of quantization levels (or word length) via the input terminal 11. The use of different nonlinear functions in the third nonlinear quantization method is as described with reference to FIG.

When the quantized data is quantized using a non-linear function selected according to the number of quantization levels in the data quantization, the quantized data cannot be inverse quantized by the conventional inverse quantizer. This is because the conventional dequantizer is based on only a single non-linear function. Figure 17
It is a block diagram which shows the nonlinear dequantization apparatus 70 which implements the nonlinear dequantization method which concerns on this invention. In the nonlinear inverse quantizer 70, the linear inverse quantizer 71 receives the nonlinear quantized data via the input terminal 15, and
Each input indicating the number of quantization levels (or word length) at the time of quantization of the non-linear quantized data is received via the input terminal 16. The input indicating the number of quantization levels is supplied to the non-linear inverse function selector 74. The linear dequantizer 71 dequantizes the non-linear quantized data according to an input indicating the number of quantization levels, and supplies the obtained non-linear data to the post processor 72. The post processor 72 post-processes the selected nonlinear inverse function and supplies the obtained inverse quantized data to the output terminal 17.

Since the non-linear inverse quantizer 70 has to dequantize the quantized quantized data using different non-linear functions selected according to the number of quantization levels, a set of plural non-linear inverse functions is set. Has a non-linear inverse function block 73 in which is stored. Each non-linear inverse function corresponds to one of the non-linear functions used in audio compressor 111 (FIG. 6). For example, each non-linear inverse function may correspond to one of the non-linear functions stored in the non-linear function block 63 in the non-linear quantizer 60 (FIG. 16).

Further, the non-linear inverse quantization device 70 has a non-linear inverse function selector 74, and the non-linear inverse function selector 74 uses the same non-linear method as the selection method used for the non-linear data selection in the quantizer. Inverse function block 73
Select one of the non-linear inverse functions stored in. The selected nonlinear inverse function is supplied to the post processor 72. The post processor 72 uses the selected nonlinear inverse function to post-process the nonlinear data from the linear inverse quantizer 71. Therefore, in the illustrated example, the post-processor 72 performs post-processing on the non-linear data using the inverse function of the non-linear function used by the quantizer 60 to pre-process the data. The post processor 72 supplies the obtained dequantized data to the output terminal 17.

In a general example of the non-linear inverse quantizer 70, the non-linear inverse function set stored in the non-linear inverse function block 73 includes a non-linear inverse function for each allowable value of the number of quantization levels (or word length). Is included. Number of quantization levels (or word length) used for quantization of non-linear quantized data supplied to the linear quantizer 71 via the input terminal 11.
The input indicating is also supplied to the non-linear inverse function selector 74. The non-linear inverse function selector 74 selects one of the non-linear inverse functions stored in the non-linear inverse function block 73 according to the input indicating the number of quantization levels used for quantizing the non-linear quantized data, and the post-processor Supply to 72.

Even in the example in which the number of non-linear inverse functions of the non-linear inverse function set is small as compared with the number of allowable values of the number of quantization levels (or word length), the non-linear inverse function selector 74 uses the quantization level (or word length). ) Is required to be input from the input terminal 11.

FIG. 18 is a block diagram showing an example in which an inverse quantizer for implementing the nonlinear inverse quantization method according to the present invention is applied to the audio decompression device 112 of the transmission / recording system 110 shown in FIG. It should be noted that the nonlinear dequantization method according to the present invention is not limited to this usage, and can be applied to other nonlinear dequantization to provide the above-mentioned advantages.

In the audio decompression device 112, the compressed signal from the transmission / recording medium is sent to the demultiplexer 22.
5 is supplied. The demultiplexer 225 extracts a quantized spectrum component set and each bit allocation parameter from the compressed signal, and also extracts a word length and a scale factor for each quantized spectrum component set from the bit allocation parameter. The demultiplexer 225 supplies the spectral component set to the input terminal 15 of the non-linear inverse quantizer 233 and supplies each word length to the input terminal 16 of the non-linear inverse quantizer 233. Also, the demultiplexer 2
25 supplies the scale factor to the block floating release processor 232.

The non-linear inverse quantizer 233 non-linearly dequantizes each non-linear quantized spectrum component set from the demultiplexer 225 using each word length, and the quantized spectrum component by the linear de-quantizer 71 (FIG. 17). In addition to controlling the non-linear dequantization of (1), the non-linear inverse function used by the post-processor 72 is selected, and the non-linear spectrum component obtained by the linear dequantization of the linear dequantizer 71 is post-processed.

The block floating spectral component set output by the non-linear inverse quantizer 233 is supplied to the block floating release processor 232. The block floating release processor 232 performs the reverse process of the block floating performed for each spectral component set in the audio compression device. Block floating processor 2
32 multiplies each set of block floating spectral components by the scale factor from the demultiplexer 225 to provide a reconstructed spectral component set.

The block floating release processor 232 supplies the reconstructed spectral component set to the time-frequency synthesizer 221. Time-frequency synthesizer 221
Transforms each set of reproduced spectral components back into the time domain and reproduces the audio output signal block.

When the spectral component set is frequency band-divided in the audio compression device 111 before quantization, the demultiplexer 225 sets each frequency band of the quantized spectral component and its corresponding scale factor and word length as follows. Extract from the compressed signal. The non-linear dequantization device 233 non-linearly dequantizes the quantized spectral component of each frequency band using the word length of each frequency band. The block floating release processor 232 releases the block floating processing applied to each of the obtained block floating spectral components, and outputs the reproduced spectral component band. The time-frequency synthesizer 221
The reproduction spectrum components of all frequency bands are inversely orthogonally transformed into the time domain, the obtained band signal blocks are combined, and the audio output signal block is reproduced.

Although the embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various modifications can be made without departing from the spirit of the present invention defined by the claims. Is.

[0109]

According to the nonlinear quantization method of the present invention, each data having a data value is received, word length information indicating the number of quantization levels is received, and the number of quantization levels is calculated from the word length information and the nonlinear function. Quantized values are determined for each of the quantized levels of, and the quantized values determined for all the quantized levels of the number of quantized levels are unevenly spaced, and the quantized value having the quantized value closest to the data value of each By selecting a level as each corresponding quantized data, when quantizing each data non-linearly, the quantizing level selected from a plurality of quantizing levels is used, and the data is non-linear quantizing according to the non-linear function. To do. As a result, the nonlinear quantization method according to the present invention can reduce the quantization error as compared with the conventional nonlinear quantization method.

In the quantization method according to the present invention, each data having a data value is received, word length information indicating the number of quantization levels is received, and a table entry for each quantization level of each number of quantization levels is received. A quantization table set consisting of a set of quantization tables for each quantization level having is supplied, and a quantization table for the number of quantization levels indicated by the word length information is selected from the quantization table set and selected. From the quantization table, by selecting the quantization level whose table entry corresponds to the data value of each data as the corresponding quantization data, when quantizing each data, Quantize the data using the selected quantization level. As a result, the quantization method according to the present invention can reduce the quantization error as compared with the conventional quantization method.

In the non-linear quantization method according to the present invention, a non-linear function set for determining each non-linear quantization characteristic is supplied,
Receiving data each having a data value, receiving word length information indicating the number of quantization levels, selecting one nonlinear function from the nonlinear function set according to the word length information, and selecting the selected nonlinear function for each data By applying pre-processing to the data and generating pre-processed data, the quantization level selected from a plurality of quantization levels is used when nonlinearly quantizing each data, and according to the nonlinear function. Non-linear quantization of data. As a result, the nonlinear quantization method according to the present invention can reduce the quantization error as compared with the conventional nonlinear quantization method.

In the non-linear inverse quantization method according to the present invention, a set of non-linear inverse functions corresponding to each of a plurality of non-linear functions is supplied, non-linear quantized data is received, and word length information indicating the number of quantization levels is obtained. Receive, linearly inverse quantize the non-linear quantized data according to the word length information, generate non-linear data, select one non-linear inverse function from the non-linear inverse function set, and select the selected non-linear inverse function to each non-linear data Non-linear quantization of the original data of the information signal according to one non-linear function selected from a plurality of non-linear functions by applying post-processing to the non-linear data by applying and generating inverse quantized data The non-linear quantization data, which is obtained by the above method, represents the information signal with fewer bits than the original data, and has a quantization level selected from a large number of quantization levels. The inversely quantized, and generates the inverse quantized data. As a result, the non-linear inverse quantization method according to the present invention can obtain data with less quantization error than the conventional non-linear quantization method.

[Brief description of drawings]

FIG. 1 is a block diagram showing a specific configuration of a non-linear quantization device for implementing a first non-linear quantization method according to the present invention.

FIG. 2 is a flowchart illustrating calculation of a quantized value at each quantization level in the first non-linear quantization method according to the present invention.

FIG. 3 is a characteristic diagram showing an example of a non-linear function used in the first non-linear quantization method according to the present invention, and also shows a linear function for comparison.

FIG. 4 is a characteristic diagram showing an absolute value of a quantization error caused by non-linear quantization at three quantization levels according to the present invention, and for comparison, caused by non-linear quantization at three conventional quantization levels. The absolute value of the quantization error is also shown.

FIG. 5 is a characteristic diagram showing an absolute value of a quantization error caused by non-linear quantization at 7 quantization levels according to the present invention, and for comparison, caused by non-linear quantization at conventional 7 quantization levels. The absolute value of the quantization error is also shown.

FIG. 6 is a block diagram showing a specific configuration of an audio transmission / recording system that compresses an audio signal using the nonlinear quantization method according to the present invention before transmission / recording.

FIG. 7 is a block diagram showing a specific configuration of an audio compression apparatus of the transmission / recording system shown in FIG. 6, the audio compression apparatus being a non-linear quantization apparatus for implementing the non-linear quantization method according to the present invention. Have.

8 is a block diagram showing a specific configuration of a time-frequency analyzer in the audio compression device of the transmission / recording system shown in FIG.

FIG. 9 is a block diagram showing a specific configuration of a non-linear quantization device that implements a second non-linear quantization method according to the present invention.

FIG. 10 is a diagram showing an example of a quantization table for five quantization levels used in the second nonlinear quantization method according to the present invention.

FIG. 11 is a characteristic diagram showing polygonal line approximation values of a nonlinear function for five quantization levels used in the third nonlinear quantization method according to the present invention.

FIG. 12 is a characteristic diagram showing polygonal line approximation values of a non-linear function for seven quantization levels used in the third non-linear quantization method according to the present invention.

FIG. 13 is a block diagram showing a specific configuration of a non-linear quantization apparatus that implements a third non-linear quantization method according to the present invention.

FIG. 14 is a diagram of an input area for explaining a quantized value and a decision value obtained by the non-linear quantization method according to the present invention.

FIG. 15 is a characteristic diagram showing an example of a nonlinear function set used in a fourth nonlinear quantization method according to the present invention.

FIG. 16 is a block diagram showing a specific configuration of a non-linear quantization device that implements a fourth non-linear quantization method according to the present invention.

FIG. 17 is a block diagram showing a specific configuration of a non-linear inverse quantization device that implements the non-linear inverse quantization method according to the present invention.

18 is a block diagram showing a specific configuration of an audio decompression device of the transmission / recording system shown in FIG. 6, the audio decompression device performing a nonlinear dequantization method according to the present invention. Have a device.

FIG. 19 is a diagram showing an input region of conventional linear quantization.

FIG. 20 is a block diagram showing a configuration of a conventional non-linear quantization device in which data is non-linearly transformed before performing linear quantization.

FIG. 21 is an input / output diagram illustrating linear quantization based on a nonlinear quantization model.

FIG. 22 is an input / output diagram illustrating conventional non-linear quantization.

FIG. 23 is an input / output diagram showing a quantized value of each quantization level generated by conventional non-linear quantization.

FIG. 24 is a diagram of an input area for explaining the quantized value of each quantization level generated by the conventional nonlinear quantization method shown in FIG. 23.

[Explanation of symbols]

 30 Non-Linear Quantizer 31 Quantized Value Calculator 32 Quantizer 33 Non-Linear Function Block 110 Transmission / Recording System 111 Audio Compressor 112 Audio Decompressor 113 Transmission / Recording Medium 121 Time-Frequency Analyzer 122 Bit Allocation 123 Spectral Quantum Quantifier 124 psychoacoustic model 125 multiplexer 132 block floating processor 133 non-linear quantizer

Claims (58)

[Claims] 1. A non-linear quantization method for non-linearly quantizing data representing an information signal in order to generate quantized data representing an information signal using a small number of bits, wherein each data is non-linear quantized. In the non-linear quantization method of non-linearly quantizing the data according to the non-linear function, using a quantization level selected from a plurality of quantization levels, the step of receiving each data having a data value, Receiving word length information indicating the number of quantization levels, determining a quantization value for each quantization level of the number of quantization levels from the word length information and a non-linear function, and performing total quantization of the number of quantization levels. Determining the quantized values determined for the levels to be unequal intervals, and selecting the quantization level having the closest quantized value to each data as the corresponding quantized data Nonlinear quantization method characterized by having a that selection process. 2. The non-linear function has an output range, and in the determining step, a step of obtaining a non-linear inverse function which is the inverse of the non-linear function from the non-linear function, the quantization level being the output of the non-linear function. The method further comprises the steps of equally allocating over the range and generating a corresponding number of equally spaced values, and processing each equally spaced value using the nonlinear inverse function to produce each corresponding quantized value. The nonlinear quantization method according to claim 1. 3. The nonlinear quantization method according to claim 1, further comprising a step of performing time-frequency analysis on the information signal to generate data. 4. The information signal is composed of a sequence of a plurality of samples, and the step of performing the time-frequency analysis includes the step of performing time-frequency analysis on a sample block of the information signal to generate data. 4. The non-linear quantization method according to claim 3, further comprising a step of obtaining word length information received in the step of receiving the word length information from a sample block of an information signal. 5. In the step of performing the time-frequency analysis, obtaining a spectrum component set from the sample block of the information signal, obtaining a scale factor from the spectrum component set, and using the scale factor, a spectrum is obtained. The block floating is performed on the component set to generate data, and the step of obtaining the word length information includes a step of obtaining word length information from the spectral component set.
Non-linear quantization method described. 6. The information signal comprises a sequence of a plurality of samples, and the step of performing the time-frequency analysis includes a step of orthogonally transforming a sample block of the information signal to generate data. Item 5. The non-linear quantization method according to item 3. 7. The step of orthogonally transforming the sample block to generate data includes the step of orthogonally transforming the sample block to generate frequency band divided spectral components, and scaling the spectral components of each frequency band. A step of obtaining a factor, and a step of performing block floating on the spectrum component of each frequency band according to the scale factor, and supplying the obtained block floating spectrum component as data, from the spectrum component of each frequency band 7. The non-linear quantization method according to claim 6, further comprising a step of obtaining word length information received in the step of receiving the word length information. 8. The step of performing the time-frequency analysis comprises filtering the information signal to generate a sequence of time-frequency spectral components in one of a plurality of subbands as data. The non-linear quantization method according to claim 3. 9. The filtering step comprises obtaining a scale factor for a sequence of time-frequency spectral components in one of the plurality of subbands, and time-frequency in one of the plurality of subbands depending on the scale factor. Block-floating a sequence of spectral components and supplying the resulting block-floating spectral components as data, receiving the word length information from a sequence of time-frequency spectral components in one of the plurality of subbands. 9. The non-linear quantization method according to claim 8, further comprising a step of obtaining word length information received in the step of performing. 10. A step of performing block floating on data to generate block floating data having a block floating data value, wherein in the selecting step, a quantization value closest to the block floating data value of each block floating data. The non-linear quantization method according to claim 1, characterized in that a quantization level having is selected as each corresponding quantized data. 11. A step of calculating a decision value between adjacent quantization levels at each quantization level of the number of quantization levels from the quantization value of the quantization level and the quantization value of the adjacent quantization level, For each quantization level of the number of quantization levels, there is a step of determining the input range from the adjacent decision value including the quantization value of each quantization level, in the quantization level selection step, 2. The non-linear quantization method according to claim 1, wherein a quantization level having an input range including data values is selected as each corresponding quantized data. 12. A step of storing each quantization level and its input range in a quantization table, wherein the step of selecting the quantization level compares the data value with the input range stored in the quantization table. , A step of identifying an input range including the data value of each data, and a step of reading the quantization level of the input range including the data value of each data as the corresponding quantized data from the quantization table. The non-linear quantization method according to claim 11, which is characterized in that 13. A step of providing an allowable range of the number of quantization levels, a step of determining a quantization value of the quantization level for each quantization level, and a quantization level and a quantization level adjacent thereto. By repeating the steps of calculating the decision value of γ, the step of determining the input range of the quantization level, and the step of storing the quantization level and its input range in the quantization table. There is a step of constructing a quantization table for each number of quantization levels within the range, and a step of selecting a quantization table for the number of quantization levels indicated by the tone information from the constructed quantization table. 13. The non-linear method according to claim 12, wherein the step of comparing the data values and the step of reading from the quantization table are performed using the selected quantization table. Quantization method. 14. Data is non-linearly quantized using one of a plurality of non-linear functions, and a plurality of non-linear functions corresponding to different numbers of quantization levels are provided, and a plurality of non-linear functions are provided according to word length information. In the step of specifying one of the nonlinear functions and the step of determining the quantization value for each quantization level of the number of quantization levels, the quantization value is determined from the specified nonlinear function and word length information. The nonlinear quantization method according to claim 13. 15. A step of storing each quantization level and its quantization value in a quantization table, wherein the step of selecting the quantization level compares the data value with an input range stored in the quantization table. Then, the method includes the step of identifying an input range including the data value of each data, and the step of reading the quantization level of the input range including the data value of each data from the quantization table as the corresponding quantized data. The nonlinear quantization method according to claim 1, wherein 16. A step of providing an allowable range of the number of quantization levels, a step of determining a quantization value for each quantization level,
Constructing a quantization table for each quantization level number within the allowable range of the quantization level number by repeating the quantization level and the step of storing the quantization value in the quantization table; A step of selecting a quantization table for the number of quantization levels indicated by the word length information from the constructed quantization table, the step of comparing the data values and the step of reading from the quantization table 16. The non-linear quantization method according to claim 15, wherein the non-linear quantization method is performed by using the generated quantization table. 17. A step of nonlinearly quantizing data using one of a plurality of nonlinear functions to supply a plurality of nonlinear functions corresponding to different numbers of quantization levels, and In the step of determining the quantization value for each quantization level of the number of quantization levels, the step of designating one of them is used to determine the quantization value from the specified nonlinear function and word length information. The non-linear quantization method according to claim 16, wherein 18. Data is nonlinearly quantized using one of a plurality of non-linear functions to provide a plurality of non-linear functions corresponding to different numbers of quantization levels; and a plurality of non-linear functions according to word length information. A step of designating one of the functions, and in the step of determining a quantization value for each quantization level of the number of quantization levels, the quantization is determined from the designated nonlinear function and word length information. The non-linear quantization method according to claim 1, wherein 19. The specified non-linear function has an output range, and the determining step includes: obtaining from the non-linear function a non-linear inverse function that is the inverse of the non-linear function; and quantizing the number of quantization levels. Assigning levels evenly over the output range of the quantizing function to generate a corresponding number of equally spaced values, and processing each equally spaced value using the nonlinear inverse function to produce each quantized value 19. The non-linear quantization method according to claim 18, further comprising: 20. A quantizing method for quantizing data representing an information signal in order to generate quantized data representing an information signal using a small number of bits, wherein each data is quantized. In a quantization method for quantizing data using a quantization level selected from a plurality of quantization levels, a step of receiving each data having a data value, and word length information indicating the number of quantization levels are included. A step of receiving, a step of supplying a quantization table set consisting of a set of quantization tables for each quantization level having a table entry for each quantization level for each number of quantization levels, and the quantization table set From the step of selecting a quantization table for the number of quantization levels indicated by the word length information, and The quantization levels but which correspond to the data value of each data,
And a step of selecting each corresponding quantized data. 21. In the step of supplying the quantization table set, the table entry of the quantization table is
3. A non-linear quantization characteristic is supplied.
The quantization method described in 0. 22. In the step of supplying the quantization table set, the table entry of the quantization table is
22. Quantization method according to claim 21, characterized in that it is obtained from a single non-linear function which determines the non-linear quantization characteristic. 23. In the step of supplying the quantization table set, the table entry of the quantization table is
21. Quantization method according to claim 20, characterized in that different non-linear quantization characteristics are provided in different quantization tables. 24. In the step of providing the quantization table set, the table entries of the quantization table for a large number of quantization levels are compared with the table entries of the quantization table for a small number of quantization levels,
24. Quantization method according to claim 23, characterized in that it provides a higher non-linear quantization characteristic. 25. In the step of providing the quantizer table set, each quantizer level has a respective quantizer value, and each table entry of the quantizer table is adjacent to the quantizer value of the quantizer level. There is an input range for each quantization level that includes a value between the quantized value and the quantized value that is in the middle of the quantized level, and the data value of each data is included in the input range by the above selection process. 21. The quantizing method according to claim 20, wherein the quantizing level of the table entry is selected. 26. In the step of supplying the quantization table set, the input range in the table entry of the quantization table is set so as to supply a non-linear quantization characteristic. Quantization method. 27. In the step of providing the quantization table set, each quantization level has each quantization value, each table entry of the quantization table has a quantization value of each quantization level, 21. The quantization method according to claim 20, wherein in the selection step, the quantization level of the table entry whose quantization value is the closest to the data value of each data is selected. 28. In the step of supplying the quantization table set, the quantization value in the table entry of the quantization table is set so as to supply a non-linear quantization characteristic. Quantization method. 29. A time-frequency analysis is performed on the information signal,
21. The quantization method according to claim 20, further comprising the step of generating data. 30. The information signal comprises a sequence of a plurality of samples, and in the step of performing the time-frequency analysis, a time-frequency analysis is performed on a sample block of the information signal to generate data, and 30. The quantization method according to claim 29, further comprising the step of obtaining word length information received in the step of receiving the word length information from a block. 31. The step of performing the time-frequency analysis includes: obtaining a spectral component set from a sample block of an information signal; obtaining a scale factor from the spectral component set; and using the scale factor to obtain a spectral component. The step of performing block floating on a set to generate data, wherein the step of obtaining word length information includes the step of obtaining word length information from a spectral component set.
The quantization method described in 0. 32. The information signal comprises a sequence of a plurality of samples, and the step of performing the time-frequency analysis includes a step of orthogonally transforming a sample block of the information signal to generate data. 29. The quantization method described in 29. 33. The step of orthogonally transforming the sample block to generate data includes the step of orthogonally transforming the sample block to generate frequency band divided spectral components, and a scale for spectral components of each frequency band. A step of obtaining a factor, and performing block floating on the spectrum component of each frequency band according to the scale factor, and supplying the obtained block floating spectrum component as data, from the spectrum component of each frequency band, 33. The quantization method according to claim 32, further comprising a step of obtaining word length information received in the step of receiving word length information. 34. The step of performing the time-frequency analysis comprises:
The information signal is filtered to obtain a sequence of time domain spectral components in one of the subbands,
30. The quantization method according to claim 29, further comprising a step of generating it as data. 35. The step of filtering comprises obtaining a scale factor for a sequence of time domain spectral components in one of the plurality of sub-bands, and in one of the plurality of sub-bands depending on the scale factor. Performing block floating on the sequence of time domain spectral components and generating the obtained block floating spectral components as data, from the sequence of time domain spectral components in one of the plurality of subbands, the word length information The quantization method according to claim 34, further comprising the step of obtaining word length information received in the step of receiving. 36. A step of performing block floating on data to generate block floating data having a block floating data value, wherein the table entry corresponds to the block floating data value of each block floating data in the selecting step. 21. The quantization method according to claim 20, wherein the quantization level to be selected is selected as each corresponding quantization data. 37. A non-linear quantization method for non-linearly quantizing data representing an information signal in order to generate quantized data representing an information signal using a small number of bits, the non-linear quantization method comprising: A non-linear quantization method that uses a quantization level selected from multiple quantization levels to quantize data in accordance with the non-linear function, and supplies a set of non-linear functions that determine each non-linear quantization characteristic. , A step of receiving data each having a data value, a step of receiving word length information indicating the number of quantization levels, and a step of selecting one nonlinear function from the nonlinear function set according to the word length information. And pre-processing the data by applying the selected non-linear function to each data, and generating pre-processed data. Linear quantization method. 38. A step of determining an allowable range of the number of quantization levels, wherein in the step of supplying the set of nonlinear functions, a different nonlinear function is set for each number of quantization levels within the allowable range of the number of quantization levels. The non-linear quantization method according to claim 37, characterized in that the non-linear quantization method is provided. 39. In the step of supplying the non-linear function set, the non-linearity of each non-linear function is different, and in the selecting step, the non-linear function selected when the word length information indicates a large number of quantization levels is 38. The non-linear quantization method according to claim 37, characterized in that the non-linear function is higher than the non-linear function selected when the long information indicates a small number of quantization levels. 40. A step of determining a permissible range for the number of quantization levels, wherein in the step of supplying the non-linear function set, each quantum within the permissible range of the number of quantization levels is set as each non-linear function of the non-linear function set. The polygonal line approximation value of the supplied non-linear function for the number of quantization levels is supplied, and the polygonal line approximation value has a coincidence point with the same number of non-linear functions as the number of quantization levels and a linear part connecting these consecutive points. 38. The non-linear quantization method according to claim 37, wherein: 41. In the step of supplying the set of non-linear functions, the non-linear function estimated by the polygonal line approximation is
The nonlinearity is different, and the nonlinear function estimated by the polygonal approximation for a large number of quantization levels has a nonlinearity compared to the nonlinear function estimated by a polygonal approximation for a small number of quantization levels. The non-linear quantization method according to claim 40, which is low. 42. The non-linear quantization method according to claim 40, wherein in the step of supplying the non-linear function set, all of the polygonal line approximation values approximate the same non-linear function. 43. The non-linear quantization method according to claim 37, further comprising the step of performing time-frequency analysis on the information signal to generate data. 44. The information signal comprises a sequence of a plurality of samples, and in the step of performing the time-frequency analysis, a sample block of the information signal is subjected to time-frequency analysis to generate data, and the data signal is extracted from the sample block of the information signal. 44. The nonlinear quantization method according to claim 43, further comprising a step of obtaining word length information received in the step of receiving the word length information. 45. The step of performing the time-frequency analysis includes: obtaining a spectral component set from a sample block of an information signal; obtaining a scale factor from the spectral component set; and using the scale factor to obtain a spectral component set. Applying block floating and generating data, wherein the step of obtaining word length information comprises the step of obtaining word length information from a spectral component set.
4. The non-linear quantization method described in 4. 46. The information signal comprises a sequence of a plurality of samples, and the step of performing the time-frequency analysis includes a step of orthogonally transforming a sample block of the information signal to generate data. 43. The non-linear quantization method described in 43. 47. The step of orthogonally transforming the sample block to generate data includes the step of orthogonally transforming the sample block to generate frequency band divided spectral components, and a scale factor for the spectral components of each frequency band. And a step of performing block floating on the spectrum component of each frequency band according to the scale factor, and supplying the obtained block floating spectrum component as data. The non-linear quantization method according to claim 46, further comprising a step of obtaining word length information received in the step of receiving length information. 48. The step of performing the time-frequency analysis comprises:
The non-linear quantization method according to claim 43, further comprising the step of filtering the information signal and generating a sequence of time domain spectral components in one of the plurality of subbands as data. 49. The step of filtering comprises obtaining a scale factor for a sequence of time domain spectral components in one of a plurality of subbands, and one of the plurality of subbands depending on the scale factor. Block-floating the sequence of time-domain spectral components in, and supplying the resulting block-floating spectral components as data, from the sequence of time-domain spectral components in one of the plurality of subbands, the word length 49. The non-linear quantization method according to claim 48, further comprising the step of obtaining word length information received in the step of receiving information. 50. A step of performing block floating on data to generate block floating data, wherein in the preprocessing step, the preprocessing data is generated by applying a selected nonlinear function to each block floating data. 38. The non-linear quantization method according to claim 37, wherein: 51. The information signal is obtained by nonlinearly quantizing the original data of the information signal according to one nonlinear function selected from a plurality of nonlinear functions, and the information signal is obtained with fewer bits compared to the original data. In a nonlinear dequantization method that dequantizes nonlinear quantized data having a quantization level selected from a large number of quantization levels to generate dequantized data. Supplying a set of corresponding non-linear inverse functions, receiving non-linear quantized data, receiving word length information indicating the number of quantization levels, and non-linear quantized data according to the word length information. Linearly inverse quantizing to generate nonlinear data, selecting one nonlinear inverse function from the nonlinear inverse function set, and selecting each of the selected nonlinear inverse functions. Applying non-linear data to post-process the non-linear data to generate dequantized data. 52. The non-linear inverse quantization method according to claim 51, wherein in the selecting step, one of the non-linear inverse functions is selected according to the word length information. 53. A step of determining the allowable range of the number of quantization levels, wherein in the step of supplying the set of nonlinear inverse functions, a different nonlinearity is obtained for each number of quantization levels within the allowable range of the number of quantization levels. The non-linear inverse quantization method according to claim 52, wherein an inverse function is supplied. 54. In the step of supplying the set of non-linear inverse functions, the non-linear inverse functions have different non-linearities, and in the selecting step, the non-linear inverse function selected when the word length information indicates a large number of quantization levels is 6. The non-linearity is higher than the non-linear inverse function selected when the word length information indicates a small number of quantization levels.
2. The non-linear inverse quantization method described in 2. 55. The non-linear inverse quantization method according to claim 51, further comprising the step of performing time-frequency synthesis on the data to reproduce the information signal. 56. Before the step of receiving a scale factor and the step of performing the time-frequency combination, the block floating release data is released by canceling the block floating performed on the dequantized data according to the scale factor. 56. The non-linear inverse quantization method according to claim 55, wherein the block-floating release data is subjected to time-frequency synthesis in the time-frequency synthesis step. 57. The information signal comprises a sequence of a plurality of samples, and in the time-frequency synthesis step, time-frequency synthesis is performed on the block floating release data to reproduce a sample block of the information signal. The non-linear inverse quantization method according to claim 56. 58. The non-linear inverse quantization method according to claim 51, further comprising the step of releasing block floating from the inversely quantized data after the post-processing.