JP3057258B2 - Apparatus and method for efficient encoding of digital data - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly efficient digital data encoding apparatus and method for compressing input digital data by performing a so-called block floating process.

[Summary of the Invention]

According to the present invention, input digital data is divided into blocks, parameters necessary for encoding are obtained for each block, data based on linear approximation of parameters of a plurality of blocks is obtained, and a plurality of parameter data and data based on linear approximation are obtained. High-efficiency digital data coding that reduces the parameter transmission bit rate and the amount of vector quantization operation by vector-quantizing the difference and performing block coding on each block data based on the parameters. The present invention provides a chemical conversion device and method.

[Conventional technology]

For example, as a high-efficiency encoding technique of digital data based on an audio signal or the like, input digital data is divided into blocks, a floating coefficient of each block is calculated, and data of each block is quantized after being normalized by the floating coefficient. Such a so-called block floating technique is known. As the input digital data to be blocked, PCM data obtained by, for example, so-called linear pulse encoding of an analog audio signal or the like, or data subjected to various other encodings can be considered. That is, examples of the other various coding techniques include, for example, band division coding (sub-band coding: SBC) for dividing a signal on the time axis such as an audio signal into a plurality of frequency bands and coding the divided signals. So-called adaptive conversion coding (ATC), which converts an axis signal into a signal on the frequency axis (orthogonal transformation), divides the signal into a plurality of frequency bands, and adaptively codes each band, or so-called adaptive coding The so-called adaptive bit allocation, which combines predictive coding (APC), divides the signal on the time axis into bands, converts each band signal to baseband (low band), and performs multi-order linear prediction analysis to predictively code. High-efficiency coding such as (APC-AB) can also be mentioned.

In the above-mentioned adaptive transform coding among these various coding techniques, for example, the audio signal on the time axis is orthogonal to the time axis by orthogonal transform such as fast Fourier transform (FFT) or discrete cosine transform (DCT). The frequency is converted to an axis (frequency axis), then divided into a plurality of bands, and the FFT coefficients, DCT coefficients, and the like of each of the divided bands are adaptively quantized (requantized). As an example of the requantization in the adaptive transform coding of the fast Fourier transform, as shown in FIG. 6, for example, the FFT amplitude value A
m and the like are divided into blocks (blocks B1 to B12 ‥‥), floating coefficients are calculated for each of these blocks, and each block data is normalized (normalized) using the floating coefficients and then quantized, thereby quantizing the block. Floating processing is performed. As the floating coefficient in this case, a value obtained by multiplying the peak value or the average value of each block by a coefficient or the like is used, and the data of each block is divided by the floating coefficient corresponding to the block (or the floating coefficient is set as described above). In the case of setting as a reciprocal of a peak value or the like, multiplication is performed) to perform normalization. The floating coefficient is also quantized and transmitted. When quantizing the floating coefficients, each floating coefficient is quantized using a so-called scalar quantization technique. In general, the scalar quantization is, for example, a method of quantizing individual independent sample values and the like, and so-called one-dimensional quantization is collectively called scalar quantization.

In the high efficiency coding, for example, in band division coding, as shown in FIG. 7, for example, audio signals on the time axis are divided into a plurality of frequency bands (bands b1 to b4 ‥).
‥), and the signals S for each band divided by the plurality of frequency bands are respectively sampled in the time axis direction and are divided into blocks at predetermined samples (blocks BL1, BL2, ‥).
Then, the data of the block is quantized (requantized). Also in this band division coding, a floating process similar to the above may be performed for each block.
That is, in the block floating process in the band division coding, as in the case of the block floating process in the adaptive transform coding, normalization of block data using floating coefficients and scalar quantization of floating coefficients are performed.

[Problems to be solved by the invention]

As described above, in both the adaptive transform coding and the band division coding, the floating coefficient of the block floating process as the parameter is quantized by the scalar quantization method. In the scalar quantization at this time, since the floating coefficient of each block is individually quantized,
For example, when transmitting this quantized output, the bit rate for transmitting the quantized floating coefficient becomes high. That is, in other words, a large number of bits are required to transmit the parameters.

Therefore, the present invention has been proposed in view of the above situation, and has as its object to provide a high-efficiency digital data encoding apparatus and method capable of greatly reducing the number of transmission bits for parameters. Is what you do.

[Means for solving the problem]

The high-efficiency encoding apparatus for digital data of the present invention has been proposed to achieve the above-mentioned object, and includes a blocking means for blocking digital data and a parameter required for encoding for each block. Parameter calculating means for calculating, linear approximation data calculating means for calculating data based on linear approximation of parameters of a plurality of blocks, difference data calculating means for calculating a difference between the plurality of parameter data and data based on the linear approximation, , Vector quantization means for performing vector quantization on the difference data, and coding means for performing block coding on each block data based on the parameters.

The high-efficiency encoding method for digital data according to the present invention has been proposed to achieve the above-mentioned object. Digital data is divided into blocks to generate block data, and each block is required for encoding. Calculating the parameter data, calculating data based on the linear approximation of the parameter data of the plurality of blocks, calculating the difference between the plurality of parameter data and the data based on the linear approximation,
The difference data is vector-quantized, and each block data is block-coded based on the parameter.

[Action]

According to the present invention, since the difference is vector-quantized and transmitted, the transmission bit rate is reduced. Further, since the difference is obtained from the difference between the data based on the linear approximation of the plurality of parameter data and the plurality of parameter data, a plurality of differences similar to each other are obtained, and it is easy to share a codebook in vector quantization of the difference. Become.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows a digital data high-efficiency encoding apparatus according to an embodiment of the present invention to which the above-mentioned adaptive conversion coding of the high-efficiency encoding is applied.

High-efficiency coding apparatus in a digital data of the present embodiment shown in FIG. 1, the digital audio, as well as blocked in a blocking circuit _{131-134 25} for example, the frequency dividing input digital data is audio data, calculates the parameters necessary for the so-called block coding for each block by the parameter processing unit 20, the normalization circuit 15 ₁ to 1
In block ₂₅ , each block data is block-encoded based on the above parameters and then transmitted.
Least square operation circuit 26 for parameter data of multiple blocks
And data based on the linear approximation by the difference obtained by subtracting the plurality of the parameter data subtracter 28 _{1-28 25} and so as to transmit to the vector quantization by the vector quantization circuit 29 Things.

Here, as the input digital data such as digital audio data, for example, audio PC
For example, FFT coefficient data or the like obtained by converting data of a predetermined unit time of sample data such as M into a frequency axis by fast Fourier transform (FFT) processing can be used. In the fast Fourier transform process, if the fast Fourier transform process of, for example, 2048 points (samples) is performed on the input signal on the time axis, for example, a phase angle of 1023 points, an amplitude term of 1025 points (or 1023 points) FFT coefficient data of the imaginary part and the real part of 1025 points are obtained.

The input digital data is made via the input terminal 1 to be sent to the blocking circuit _{131-134 25.} In the blocking circuit _{131-134 25,} and divides the input digital data into the frequency bands of a plurality of (N), and the data for each obtained band as the blocks in the present invention embodiment. In the present embodiment, as the above-described frequency band division, for example, division is performed using a so-called critical bandwidth (critical band). The critical band takes into account human auditory characteristics (frequency analysis ability), and divides an audio signal or the like into, for example, 25 bands with a bandwidth of 20 kHz, and increases the bandwidth as the frequency band increases. Therefore, the number of blocks by the critical band division in the apparatus of the present embodiment is N = 25, and the number of each component of the apparatus that handles data for each block is 25.

Data for each block from these blocking circuit _{131-134 25,} a parameter calculation processing unit 20, the normalization circuit 15 ₁
Sent to ~ 15 ₂₅ and. In the normalization circuit 15 ₁ to 15 _25, parameter data (floating coefficients or no Maraize Lee Deployment-supplied from the parameter processing unit 20
Based on the factor, block coding of the FFT coefficient data for each block (normalization called the so-called block floating processing described above) is performed.

The parameter data, that is, the floating coefficient, may be an average value, a peak value, an RMS value, or the like of data (FFT coefficient data) for each block (each band of the critical band). The average RMS value is used. Thus, the floating coefficient (average RMS value), as shown in FIG. 2, the 25 floating coefficient fl ₁ to FL _25, for example corresponding to the critical band.

Then the data block encoding is performed by the normalization circuit 15 ₁ to 15 ₂₅ are synthesized in the synthesizing circuit 17 after being quantized by the quantization circuit 16 ₁ to 16 _25, the output terminal 2 of this embodiment device Is output from

By the way, in a high-efficiency coding apparatus that performs general block coding, normalization of each block is performed using a floating coefficient (parameter data) for each block, and the normalized data of each block is output from the coding apparatus. In addition, the floating coefficient is also quantized and transmitted. However, conventionally, since each floating coefficient is scalar-quantized and transmitted as described above, the transmission bit rate for these floating coefficients has increased.

For this reason, the apparatus of this embodiment uses vector quantization (vector quantization in the vector quantization circuit 29) as a method for quantizing the floating coefficient (parameter data). The above-described vector quantization refers to comparing a code vector from a code book composed of a plurality of code vectors (representative vectors) with an input vector, and identifying a code vector most similar to the input vector and a corresponding identification code. This is a quantization technique for performing data compression by reading out a code (index) to obtain this identification code.

Here, the floating coefficient fl ₁ to FL ₂₅ vector quantization of each block as described above (i.e., 25-dimensional vector quantization) is undesirable from the multitude of calculation amount. Therefore, it is conceivable that one unit of a vector is reduced to some extent, one unit is constituted by a plurality of (M) blocks, and quantization (M-dimensional vector quantization) is performed for each vector of each unit. For example, assuming that five blocks (M = 5) are one unit of a vector, five floating coefficients fl _{1 to} fl ₅ , f ₅ are sequentially arranged from the low frequency band as shown in FIG.
l _{6 to} fl ₁₀ ,..., fl _{21 to} fl ₂₅ , one unit of vector V _{1 to} V
_Make up ₅ . However, in this case, if each floating coefficient in each unit vector is directly vector-quantized, for example, the vector V ₁ (fl _{1 to} f ₁
The vectors as energy of the signal increases as l _5), now a high resolution when vector quantization is required, therefore codebook vector quantization there is a possibility that becomes huge.

For this reason, at the time of vector quantization in the vector quantization circuit 29 of the present embodiment, data based on a linear approximation of floating coefficients (parameter data) of a plurality (M) blocks and the plurality of (M) blocks are used. The difference between the floating coefficient of the block and the floating coefficient is vector-quantized. That is, the vector quantization circuit 29 quantizes not the vector of the floating coefficient but the vector constituted by the difference.

Here, the apparatus according to the present embodiment shows an example in which linear approximation of five floating coefficients in the vectors V _{1 to} V ₅ is performed. Therefore, by the floating coefficient fl ₁ to FL ₂₅ for each parameter data i.e. each vector V ₁ ~V ₅ of each block in Figure 2 for linear approximation straight line S ₁ ~ for each vector V ₁ ~V ₅ S ₅ is obtained. Of course, the floating coefficients for a plurality of (M) blocks on which the above-mentioned linear approximation is performed are not limited to the above-mentioned five, and may be an integer of two or more floating coefficients. Further, as a method of the linear approximation, for example, a so-called least square method described later can be cited. Therefore, in the present embodiment, the components of the straight lines S _{1 to} S ₅ obtained by the straight line approximation by the least squares method, that is, the respective slopes and offset values (center values of the straight lines) of the straight lines S _{1 to} S ₅ are obtained. The difference between the base value and the floating coefficient is obtained, and the difference is vector-quantized and transmitted.

In order to perform the above-described operation, in the parameter operation processing unit 20 of the present embodiment, the supplied data of each block is first converted to the floating coefficient operation circuits 24 ₁ to 24 2.
4 Sent to ₂₅ . In the circuits 24 _{1 to} 24 ₂₅ , for example, the floating coefficient of the above-described average RMS value is obtained for each block, and the floating coefficient is subtracted by each of the subtracters 28 _{1 to} 28
₂₅ and the least-squares operation circuit 26. In the circuit 26, a straight line is obtained from the 25 floating coefficients by performing a linear approximation by, for example, least squares operation for each of the five floating coefficients as described above, and the component, that is, the slope and the offset value (linear Center value). The respective slope and offset value after being quantized by the quantization circuit 27 which performs scalar quantization or vector quantization, is sent to the respective subtracter 28 _{1-28 25.} That data and the offset value is quantized and-out inclined is floating coefficient fl ₁ to FL ₂₅ corresponding to each of the straight line is sent to each subtracter 28 _{1-28 25} being fed.
Therefore, in each subtracter 28 _{1-28 25,} a floating coefficient supplied value based on the linear approximation from (parameter data) (data slope and offset values are quantized)
Is subtracted to obtain difference data sf as shown in FIG.
Will be required. The output of the quantization circuit 27 is also output from the output terminal 4.
Further, the data sent to each subtractor may be data before passing through the quantization circuit 27, and difference data sf based on the data before quantization may be obtained.

This difference data sf is sent to the vector quantization circuit 29,
The vector quantization circuit 29 performs the above-described vector quantization of the difference data sf. That is, in the vector quantization processing by the vector quantization circuit 29,
For example, a code vector closest to (similar to) a vector constituted by five pieces of difference data sf is selected, and an identification code corresponding to the code vector is obtained as a vector quantization output. Accordingly, the identification code is output from the output terminal 3 of this embodiment. Further, in the vector quantization circuit 29, a floating coefficient corresponding to the selected code vector is also read, and these floating coefficients are transmitted to the corresponding normalization circuits, respectively, so that Normalization is performed. The output from each of the output terminals 2, 3, and 4 is output and transmitted via, for example, a multiplexer or the like.

Regarding the difference, for example, as shown in (a) and (b) of FIG.
Even if the straight lines S _H and S _{L have} different slopes, the differences obtained based on the straight lines S _H and S _L are similar to each other. At the time of vector quantization of the similar difference, the same codebook can be used, so that the codebook can be made smaller and can be turned to the center value (center value of a straight line). . In other words, the codebook can be made smaller because more same differences can correspond to different center values.

As a method for obtaining the difference, it is also conceivable to obtain an average value of the floating coefficients of the plurality of blocks and subtract the average value from the floating coefficients of the plurality of blocks. The vector quantization of the difference based on the average value is defined as the average value removal vector quantization (mean-subtract V
Q). However, in the mean value removal vector quantization, the codebook of the difference cannot be shared in the cases of (a) and (b) of FIG. A codebook larger than the codebook in the circuit 29 is required. In other words, the vector quantization circuit 29 can be considered to be a vector quantization in which the dimension on the average side (mean side, center value) in the average value removal vector quantization is increased by one.

As described above, in the high-efficiency encoding apparatus to which the adaptive transform encoding according to the present embodiment is applied, when quantizing parameter data (floating coefficients) required when normalizing each block, a plurality of The difference between the data based on the straight line component (slope, offset value) obtained by the linear approximation by the least square operation of the floating coefficient (parameter data) of the block and the plurality of floating coefficients is vector-quantized. Since the identification code for quantization is transmitted as the quantization output of the floating coefficient, the transmission bit rate is greatly reduced as compared with the case where scalar quantization or the like is performed. Also, the codebook can be made smaller and the amount of calculation can be reduced as compared with the case where each floating coefficient is directly vector-quantized. That is, since the vectors are expressed hierarchically, the complexity at the same resolution can be reduced as compared with the straight vector quantization. Further, for example, if the center value of the straight line is further vector-quantized, the performance can be improved as compared with the straight vector quantization because the correlation in a longer section can be used. That is, for example, all correlations in one frequency frame can be used. Furthermore, since the two components (the difference vector and the straight line) have a more independent relationship than the simple mean value removal vector quantization, performance degradation due to codebook structuring is reduced. In addition, since the distribution of the difference vector becomes more random, the adverse effect of sharing the codebook between different bands is reduced, and the difference codebook can be shared between different bands.

Here, in general, the above-mentioned least square method includes (1)
An operation is performed to find a and b that satisfy the expression. However, in each of the following equations, the waveform W as shown in FIG.
Points (x _i, y _i) above and, y = ax + straight line of points of b (x _i, ax
_i + b).

In this equation (1), partial differentiation is performed with respect to a and b. From equation (2), From equation (3), From equation (5), Substituting this into equation (4) and rearranging it, As a result, a linear component is obtained.

Here, (x _L + x _R ) / 2 is substituted for x of y = ax + b, and a center value y ₀ is obtained and set as an offset value. However x _L at the left end of the x-coordinate values of the straight line, x _R is x-coordinate value of the right end of the straight line. That is, instead of transmitting the linear y-intercept, the center of the straight line _{(x = (x L + x} R) / 2) values of at y is used instead of the y-intercept (the straight line slope and one point Defined).

At this time, the value of y ₀ is performing the so having a strong correlation between adjacent vector further vector quantization to collect multiple y ₀ (Yes in scalar quantization) is effective. The inclination is separately quantized (vector quantization, scalar quantization, or the like). The combination in which the 誤差 value of the square error (in the vector) is the smallest is selected from the combination of the inclination and the offset value.

The present invention can be applied to, for example, a device that performs band division coding, in addition to so-called adaptive transform coding in which an input signal is processed by fast Fourier transform as in the embodiment of FIG. 1 described above. . That is, the input digital data supplied to the high-efficiency coding apparatus that performs band division coding according to the present embodiment is subjected to band-pass and low-band conversion processing by, for example, a band-pass filter (BPF), so that the time-base The input digital data is band-divided in the pass band of each BPF, and each signal is frequency-shifted (low-frequency conversion) down by the center frequency of each pass band and blocked in the frequency axis direction. In the present embodiment, the input digital data can also be divided into a so-called critical bandwidth (critical band) at the time of division by the BPF as described above. Therefore, in the case of the present embodiment, parameters are calculated from the block data of each of these BPFs as described above,
The difference between the data based on the linear approximation of the plurality of parameter data and the plurality of parameter data is vector-quantized and transmitted. The same effect as described above can be obtained in a device to which the present invention is applied with the band division coding.

Although the parameter of the above-described embodiment is a floating coefficient, the parameter is not limited to the floating coefficient. For example, bit allocation information indicating the number of bits allocated to each block can be used as the parameter. Therefore, at the time of vector quantization, not only the above-described vector quantization of the floating coefficient difference data but also the above-described bit allocation information or the difference data may be configured to be vector-quantized. . Also in this case, the transmission bit rate can be reduced.

〔The invention's effect〕

In the high-efficiency digital data encoding apparatus and method of the present invention, data based on linear approximation of a plurality of parameter data and a difference between the plurality of parameter data are vector-quantized and transmitted, so that the vector The amount of operation for quantization can be reduced, and the number of transmission bits for parameters can be significantly reduced.

[Brief description of the drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of a digital data high-efficiency coding apparatus to which adaptive transform coding according to an embodiment of the present invention is applied, FIG. 2 is a diagram showing floating coefficients, and FIG. Diagram showing difference data, fourth
FIG. 5 is a diagram showing the similarity of the vector of the difference data, FIG. 5 is a diagram for explaining the least squares operation, FIG. 6 is a diagram for explaining the adaptive transform coding, and FIG. 7 is a diagram for explaining the band division coding. It is a figure for explaining. 13 _{1 to} 13 ₂₅ … Blocking circuit 15 ₁ to 15 ₂₅ … Normalization circuit 16 _{1 to} 16 ₂₅ , 27… Quantization circuit 17… Synthesis circuit 20… Parameter operation processing unit 24 _{1 to} 24 ₂₅ … … Floating coefficient operation circuit 26 …… Least square operation circuit 28 _{1 to} 28 ₂₅ …… Subtractor 29 …… Vector quantization circuit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H03M 7/30

Claims (2)

(57) [Claims] 1. Blocking means for blocking digital data, parameter calculating means for calculating parameters required for encoding for each block, and linear approximation for calculating data based on linear approximation of parameters of a plurality of blocks. Data calculation means, difference data calculation means for calculating a difference between the plurality of parameter data and data based on the linear approximation, vector quantization means for vector-quantizing the difference data, and block data based on the parameters And a coding means for performing block coding on the digital data. 2. A method for generating block data by dividing digital data into blocks, calculating parameter data required for encoding for each block, calculating data based on a linear approximation of parameter data of a plurality of blocks, A high-efficiency encoding method for digital data, wherein a difference between the parameter data of the digital data and the linear approximation is obtained, the differential data is vector-quantized, and each block data is block-coded based on the parameter. .