JPH03139923A - Highly efficient encoder for digital data - Google Patents

Abstract

PURPOSE:To attain efficient coding corresponding to human's hearing characteristics by setting up the number of data in respective blocks almost to the same value, and at the time of quantizing floating coefficients, reducing the number of bits used for quantizing the floating coefficients in the ascending direction of frequency bands. CONSTITUTION:Input digital data are transmitted to a blocking circuit 11 and blocked in each prescribed unit time and then the data in each unit time block are converted into data on a frequency axis by a fast Fourier transformation circuit 12. The data of each band are blocked to compute the floating efficient of each block, the floating processing of each block is executed by the floating coefficient and then the processed data are quantized. In this case, the number of data in respective blocks is adjusted to almost the same value and quantizing means 181 to 1825.k for quantizing the floating coefficients of respective blocks quantize the coefficients by reducing the number of bits used for the quantization in the ascending direction of the frequency bands. Consequently, efficient bit allocation considering human's hearing characteristics can be attained.

Description

Detailed Description of the Invention [Industrial Field of Application] The present invention performs block floating processing [1a of the invention] The present invention converts input digital data into data on the frequency axis and converts the data of each band into data on the frequency axis. In a high-efficiency encoding device for digital data that performs a floating process for each block using a floating coefficient for each block and then quantizes the data, the number of data in each block is made approximately the same, and When quantizing floating coefficients, the higher the frequency of floating coefficients, the smaller the number of focus points.This enables efficient encoding of digital data in accordance with human auditory characteristics. It provides equipment.

[Conventional technology]

For example, as a high-efficiency encoding technology for digital data based on audio signals, etc., there is band division encoding (sub-band coding) in which a signal on the time axis, such as an audio signal, is divided into multiple frequency bands and encoded. :5B
C), so-called adaptive transform coding (ATC), which converts a time-axis signal into a frequency-axis signal (orthogonal transformation), divides it into multiple frequency bands, and adaptively encodes each band, or Combining the above SBC and so-called adaptive predictive coding (APC), the time axis signal is divided into bands, each band signal is converted to basepan) (low frequency), and then multi-order linear predictive analysis is performed to make predictions. Examples include so-called adaptive focus assignment (APC-AB) for encoding.

Among these various encoding techniques, for example, in the above-mentioned adaptive transform encoding, audio signals on the time axis are processed using fast Fourier transform (FFT) Ii32 or M-cosine transform (DC).
T), etc., to the axis (frequency axis) orthogonal to the time axis, and then divide it into multiple bands, and convert the FFT41f, number, OCT coefficient, etc. of each divided band into adaptive bits. Quantization (requantization) is performed by assignment. As an example of requantization in the adaptive transform encoding of the fast Fourier transform, as shown in FIG.
Block the FT amplitude value Am, etc. (blocks B1, B2...
), a floating coefficient is calculated for each block, each block data is normalized using this floating coefficient, and then quantized, thereby performing a so-called block floating process.

In this case, the floating coefficient used is the peak value or average value of each block multiplied by a coefficient, and the data of each block is divided by the floating coefficient corresponding to the block (or the floating coefficient is When set as the reciprocal of a value, etc., normalization is performed by multiplying it. Note that the floating coefficient itself is also quantized and sent.

Furthermore, in the above-mentioned high-efficiency coding, for example, band division coding, as shown in FIG.
...), and the signals S for each of these multiple frequency bands are quantized. It is also possible to bronchize each predetermined sample of these signals S in the time axis direction (blocks BL1, BL2, . . . ) and perform floating processing similar to the above for each of these blocks. In the case of block floating processing in this band division coding, as in the case of block floating processing in adaptive transform coding described above, the data of each block is normalized with a floating coefficient and then quantized, and the floating Quantization of coefficients is performed9 [Problem to be solved by the invention] By the way, when quantizing the floating coefficients mentioned above,
Generally, the allocation and number of nodes for each block are fixed. At this time, if the size of the block is not constant, for example, when the audio signal band is divided into so-called critical bands taking into account human auditory characteristics (frequency analysis ability), the high frequency band has a different bandwidth ( If the block $g) is wide, the number of samples in the block is larger in the high range than in the low range, so the number of floating coefficient bits per sample in the block in the high range is low. extremely small compared to the area (
Become. In such a case, the effect of block floating processing in high frequencies will be reduced. Conversely, if the above block size is constant, the number of bits allocated to the floating coefficient will be the same for both the low-frequency block and the high-frequency block, and the number of bits allocated to the floating coefficient will be the same for the high-frequency block, which does not originally require a large number of bits due to human auditory characteristics. This also gives a large number of bits, which prevents effective use of high frequencies.

Therefore, the present invention was proposed in view of the above-mentioned actual situation, and it is an efficient bit allocation method that takes into consideration the characteristics of human hearing while maintaining the efficiency of high-frequency profloating processing. An object of the present invention is to provide a highly efficient digital data encoding device that can perform the following steps.

[Means for Solving the Problems] The high-efficiency encoding device for digital data of the present invention has been proposed to achieve the above-mentioned object, and converts input digital data into data on the P1 wavenumber axis ( For example, the transformation by the fast Fourier transform circuit 12 shown in FIG.
The data of each band is divided into blocks (the critical banding circuit 131 and the subbanding circuit 21 shown in FIG. 1).
=s1. ~21zs, *) to calculate the floating coefficient for each block (floating coefficient calculation circuit 17.~17zs, * shown in Figure 1) After performing the normalization circuits 14. to 14zs,b) shown in FIG.
h) In the high-efficiency encoding device for digital data, the number of data in each block is approximately the same (N pieces).
and quantization means for quantizing the floating coefficients of each block (FC quantization circuit in Fig. 1, ~1
8□, +1) is configured to quantize with the number of bits that does not exceed the floating coefficient frame of the high frequency range.

[Effect]

According to the present invention, since the data of each band is formed into a block by several approximately identical samples, the number of floating coefficient bits per sample in a block in the high frequency range is lf! It does not become less than @.

Furthermore, when floating coefficients are quantized, the number of bits is not multiplied by the floating coefficient frame in the high frequency range, so that the number of bits used for quantization in the high frequency range is less wasted.

〔Example〕

Embodiments to which the present invention is applied will be described below with reference to the drawings.

FIG. 1 is a block IiJ diagram showing a schematic configuration of a high-efficiency digital data encoding device according to an embodiment of the present invention. The embodiment shown in FIG. 1 is a high-efficiency encoding apparatus to which the above-mentioned adaptive transform encoding is applied, as an example of high-efficiency encoding.

That is, in FIG. 1, the input digital data is
The signal is transmitted to the blocking circuit 11 via the input terminal 1, where it is blocked every predetermined unit of time (unit time blocking), and then transmitted to the fast Fourier transform (FFT) circuit 12. In the fast Fourier transform w112, data for each unit time block is transformed into data on the frequency axis,
For example, if we perform fast Fourier transform processing on 2048 samples, we will find FFT coefficient data expressed by 1023 phase angle points and 1025 amplitude terms (or 1023 imaginary parts and 1025 real parts). Become. These FFT coefficient data are processed by the critical banding circuit 13.
~13□, and is divided into blocks into, for example, 25 critical bands.

At this time, in the block for each band, the higher frequency band has a wider band width (block width), so the number of samples in the block is larger in the higher frequency band than in the lower frequency band. In such a case, the effect of block floating processing, which will be described later, in high frequencies is reduced.

Therefore, in this embodiment, the data of each band is divided into blocks by combining several substantially identical samples. In other words, the number of data within a block is made substantially the same. For example, N samples (FFT coefficient data) are collected (as viewed in the frequency axis direction) to form one block. Here, to explain the path after the critical banding circuit 131 as an example,
The critical banding circuit 13. N samples (1 block) are output from. The block is a normalization circuit]4. At the same time, the floating coefficient calculation circuit 17. It is also transmitted to In the floating coefficient calculation circuit 171, a floating coefficient is calculated for each block, and the floating coefficient is calculated in the normalization circuit 1.
4. transmitted to. The normalization circuit 14. Then, normalization is performed by performing floating processing on the block using this floating coefficient, and then the quantization circuit 15. The normalized block is sent to quantization. During quantization at this time, bit number information is sent from an allocated bit number determination circuit I9 that determines the allocated bit number for each band priority of the critical band based on the FFT coefficient data from the fast Fourier transform circuit 12. Quantization is performed based on This quantized output is transmitted to the combining circuit 16. Further, the floating coefficients are stored in a floating coefficient quantization (PC quantization) circuit 18, which is a quantization means for quantizing the floating coefficients of the blocks. After being quantized using a predetermined number C of vias in block units, the signal is transmitted to the synthesis circuit 16. The quantization output of the block and the quantization output of the floating coefficients are synthesized from the synthesis circuit 16 and outputted from the output terminal 2.

Here, as mentioned above, in order to maintain the efficiency of block floating processing in the high frequency range and to perform efficient bit allocation taking into account human auditory characteristics, a quantization means for quantizing the floating coefficient of each block is used. (FCit conversion circuit high frequency floating coefficient frame does not apply)
・I try to quantize using numbers. In other words, in the high-efficiency encoding device of this embodiment, in the high frequency range where the bandwidth is wide and the number of samples is large, blocks of N samples (k is a natural number and a different number for each band) are processed from one band. I'm trying to get it. For example, taking the output of the high-frequency critical banding circuit 13□ as an example,
The circuit output is divided into subbanding circuits 21zs, t~
21□51. These sub-banding circuits 21
The above block of N samples is obtained from z-, + to 21□'Ark. Each of these blocks operates in the same manner as each circuit after the critical banding circuit 13I described above. S+ 1~14 xs, h,
Each floating coefficient calculation circuit 17! s, +~17□3
9. , each quantization circuit 15□5° to 15□2. The signal is processed by the FC quantization circuits 18zs, 1 to 18, S, and then transmitted to the synthesis circuit 16.

At this time, the FC quantization circuits 18zs1 to 18□3. .

Here, the floating coefficients of each block are quantized using a bit number r (c>r) smaller than the predetermined bit number C in the FC quantization circuit 181 . Note that the number N of samples in each block in each band is set to be uniform to some extent.

For example, as shown in FIG. 2, in each of the critical bands 81 to B25, the low frequency bands B1. B2.
. . ., the high frequency band, for example, 2 subbands in band B25 "!!+I ~ 5bxs, or a predetermined number of bits smaller than the above predetermined number of bits C of each low frequency band. r is given, and quantization is performed using the number of bits r.As this predetermined number of bits, for example, as shown in parentheses in the figure, bands B1, B2, etc.
and 4 bits for band B25 (4 bits for each subband 3btS+ to 5bzs+w).Although not disclosed at this time, 6 bits are used for bands B1-85. , 5 bits for bands 86 to B15, and 4 bits for bands 816 to B25. Furthermore, when determining the number of quantization bits of this floating coefficient, the number of bits can be adjusted by taking into account the distribution of data within the block.

In this case, the number of bits allocated for quantization of floating coefficients is reduced for blocks with large variance.

In this way, in this embodiment, when quantizing floating coefficients, a predetermined number of bits is given to each high frequency subband. Even if the width (block width) is wide, the number of floating coefficient bits per sample in the band in the high band may be extremely small compared to the number of bits in the low band. Therefore, deterioration of the effect of block floating processing in high frequencies can be prevented. In addition, since quantization is performed on a pitch-1-lose basis that does not take into account the floating coefficient frame in the high frequency range, efficient bit use is possible in the high frequency range, which does not require a large number of bits, taking into consideration the characteristics of human hearing. It becomes possible.

FIG. 3 shows a high-efficiency encoding device that is compatible with the above-mentioned band division encoding.

In FIG. 3, the input digital data on the time axis supplied to the input terminal l is filtered by a band pass filter (BPF) 5.
11-512. For example, it is divided into critical bands. The bandpass filter 51. ~51□, the signals of each divided band are further passed through a low-pass filter and frequency-shifted (low-pass conversion) downward by the center frequency of the pass band. Next, the data of each band is divided into blocking circuits 521-52! Blocked with S. When creating blocks at this time, the number of data in the block (the number of blocks) is made to be approximately the same in the time axis direction (1 for N samples).
)゛lock). Therefore, this 8-blocking circuit 5
2. ~52t, the output from the blocking circuit 11 in FIG.
Unlike the output of , the time intervals are different from each other. After that, the data of each block is processed by the normalization circuit 531
~53! Floating coefficient calculation circuit 55. ~55
t, and is further normalized by a floating coefficient from quantization circuit 54.t. ~542. quantized by . The quantization at this time is also performed using the same number of bits allocated for each band as in FIG. In addition, the PC quantization circuits 56, ~5
6. ．． Then, the floating coefficients are quantized. Each of these quantized outputs is outputted from the output terminal 2 via the multiplexer 4j16.

Here, the FC5i childization circuit 56. ~568. When quantizing the floating coefficients for each block, quantization is performed using a number of bits that does not exceed the floating coefficient frame of the high frequency band. In the example of FIG. 3, the number of bits for quantization is shown in parentheses in the figure, for example, the low frequency FCi quantization circuit 56. Then, 6 bits, F(J quantization circuit 56□
Then, 6 bits..., high frequency FCi quantization circuit 56
□, it is set to 4 bits.

In this embodiment as well, deterioration of the effect of block floating processing in high frequencies can be prevented. Furthermore, it is possible to use bits efficiently in consideration of human auditory characteristics.

In addition, when forming the allocated bit information for each band in the allocated bit number determination circuit 19 of FIG. 1, for example, the allowable noise level of the signal is set, Considering the masking effect,
By setting a high allowable noise level for the energy between the high-frequency band frames of the critical band, the number of bits to be allocated to each band can be determined. Here, the above-mentioned masking effect refers to a phenomenon in which one signal masks another signal and becomes inaudible due to the characteristics of human hearing.This masking effect includes the masking effect on signals on the time axis. and a masking effect on signals on the frequency axis. In other words, due to the frequency axis masking effect,
Even if there is noise in the masked area, this noise will not be heard. Therefore, in the actual audio signal, the noise in the portion masked on the frequency axis is considered to be tolerable noise.

Therefore, when quantizing audio data or the like, it becomes possible to reduce the number of allocated bits by the allowable noise level.

FIG. 4 shows a specific example of the allocated bit number determination circuit I9 that sets the permissible noise level.

That is, data from the fast Fourier transform circuit 12 in FIG. 1 is transmitted to the sum detection circuit 32 via the input terminal 31 of the bit number determination circuit 19.

In the summation detection circuit 32, the energy of each critical band (spectral intensity in each band) is determined by taking the summation (peak, average, or energy summation) of each FFT coefficient data in each band. . The output of the sum detection circuit 32, that is, the sum spectrum of each band is generally called a park spectrum, and the park spectrum SB of each band is as shown in FIG. 5, for example. In addition, in FIG. 5, in order to simplify the illustration, 12 park spectra SB are used to represent the 25 park spectra SB corresponding to the 25 critical bands.

2. Next, in order to consider the influence of the park spectrum SB on masking, the park spectrum SB is convolved with a predetermined weighting function (conpolusion);
Therefore, each of the park spectra SB is transmitted to the filter circuit 33. As shown in FIG. 6, the filter circuit 33 includes a delay element (z-') that sequentially delays input data, that is, data of the park spectrum SB.
) 101a-t~101,. 3 and each delay element 10
1-z to 101□, multipliers 102-3 to 102, which multiply outputs from 1-z to 101□ by filter coefficients (weighting functions), and a summation adder 104.

At this time, multipliers 102-s to 102. . .

For example, in multiplier 102. , the filter coefficient is 0
．． 0000086 to multiplier 102. A filter coefficient of 0.0019 is applied to I-z, and a multiplier 102. ~1 with a filter coefficient of 0.15, multiplier 102. , the filter coefficient 1 is obtained by multiplication H102, , and the filter coefficient 0.4 is obtained by multiplier 102-g.
The park spectrum SB is subjected to convolution processing by multiplying the output of each delay element by a filter coefficient of 0.007 in step 2--3. Through this acceptance process, the above-mentioned fifth
The sum of the portions indicated by dotted lines in the figure is taken and output from terminal 105.

By the way, when calculating the masking spectrum (acceptable noise spectrum) of the park spectrum SB, the level α corresponding to an acceptable noise level, which will be described later, is
In the quantization circuits 15 to 15□3. of FIG. It becomes necessary to increase the number of focuses allocated when quantizing wrinkles. Conversely, if the level α is large, the masking spectrum will rise, and as a result, the number of bits allocated during quantization can be reduced. Note that the level α corresponding to the allowable noise level is a level that becomes the allowable noise level for each critical band by performing inverse composition processing. Additionally, in general, audio signals have low spectral intensity (energy) in the high frequency range, so
In this specific example, taking these things into consideration, the level α is increased as the energy goes to a higher frequency range, and the number of bits and positions allocated to the higher frequency range is reduced. For this reason, when setting the allowable noise level above, consider the above level α for the energy between high frequency frames.
is set high.

That is, in this specific example, a level α corresponding to the above-mentioned allowable noise level is calculated, and the level α is controlled so that it becomes higher as the frequency range increases. Therefore, the output of the filter circuit 33 is sent to the subtracter 34.

The subtracter 34 calculates the level α in the convolved area. Here, the subtracter 34 is supplied with a tolerance function (a function expressing the masking level) for determining the level α. The level α is controlled by increasing or decreasing the tolerance function. The tolerance function is supplied from a function generation circuit 36.

That is, the /bell α corresponding to the allowable noise level is
Assuming that i is a number given sequentially from the low band of the critical band, it can be determined by equation (1).

α-3-(n-ai)...(1) This (1)
In the equation, n and a are constants, a>0, S is the intensity of the park spectrum after convolution processing, and in equation (1), (n
-ai) is the tolerance function. Here, as mentioned above, since it is more advantageous to reduce the number of bits from the high frequency range with less energy in order to reduce the overall number of bits, in this specific example, n=38.
With a=1, there was no deterioration in sound quality at this time, and good encoding was achieved.

As described above, the level α is determined and this data is transmitted to the divider 35. In the divider 35,
This is for inverse convolution of one novel α in the area subjected to the convolution process. Therefore, by performing this inverse convolution process, a masking spectrum can be obtained from the level α.

That is, this masking spectrum becomes the allowable noise spectrum. Although the above deconvolution processing requires complicated calculations, in this specific example, the deconvolution is performed using a simplified divider 35.

Thereafter, the masking spectrum is transmitted to a subtracter 37. Here, the park spectrum SB is supplied to the subtracter 37 via a delay circuit 38. Therefore, by performing a subtraction operation between the masking spectrum and the park spectrum SB in the subtracter 37, the park spectrum SS becomes as shown in FIG.
The level below the level indicated by each level of the masking spectrum MS is masked. The output of the subtracter 37 is transmitted to each quantization circuit 15. of FIG. ~ILs, is output from the output terminal 39 as allocated bit number information during quantization.

As mentioned above, in this specific example, the allowable noise level is increased as the energy goes to the higher frequency range, and the number of bits allocated to the higher frequency range is reduced. Efficient bit allocation can be performed.

〔Effect of the invention〕

In the high-efficiency encoding device for digital data of the present invention, the number of data in each block is made approximately the same during bronchization, and the number of floating coefficients in focus per sample in a block in the high frequency range is extremely small. Therefore, it is possible to prevent the effect of block floating processing from deteriorating in high frequencies. In addition, when quantizing the floating coefficients of each block, the number of bits is not multiplied by the floating coefficient frame in the high frequency range, so the number of bits for quantization in the high frequency range can be used efficiently. .

Therefore, it becomes possible to perform efficient encoding according to human auditory characteristics.

[Brief explanation of the drawing]

FIG. 1 is a block circuit diagram showing a schematic configuration of a high-efficiency coding device for digital data according to an embodiment to which adaptive transform coding is applied; the second leap is a diagram schematically showing bands (blocks) of this embodiment; FIG. 3 is a block circuit diagram showing a schematic configuration of a high-efficiency encoding device for digital data according to an embodiment to which band division encoding is applied; FIG. 4 is a block circuit diagram showing a configuration for determining the number of allocated bits; FIG. 5 is a diagram showing a park spectrum, FIG. 6 is a circuit diagram showing a filter circuit, FIG. 7 is a diagram showing a masking spectrum, and FIG. 8 is a diagram schematically showing blocks of adaptive transform coding. FIG. 9 is a diagram showing blocks of band division encoding. 11.52t~52. , . . . Pro-link conversion circuit 12 . . . Fast Fourier transform circuit 131 ~], 3 = s . . .
...Critical banding circuit +7t ~17zs
, -, 551 to 5515', floating coefficient calculation circuit

Claims (1)

[Claims] Input digital data is converted to data on the frequency axis, data of each band is divided into blocks, a floating coefficient is calculated for each block, and floating processing is performed for each block using this floating coefficient. Then, in a high-efficiency encoding device for digital data that performs quantization, the number of data in each block is made approximately the same, and the quantization means for quantizing the floating coefficients of each block is A high-efficiency encoding device for digital data, characterized in that the coefficient is quantized using a smaller number of bits.