JPS59127440A - Predictive encoding system by band-division adaptive bit allocation - Google Patents

Abstract

PURPOSE:To improve the quality of a code by allotting adaptively the number of quantized bits to every frame according to the deviation in the electric power of a band-divided signal. CONSTITUTION:An input signal converted into a digital signal is divided into two frequency bands to be converted into low-frequency signals as sub-band signals, but they are stored in temporary buffer parts 6 and 7 before being encoded. The buffer parts 6 and 7 calculate pseudo predictive residual signals, which are inputted to a calculator 8 for the assigned number of bits to calculate the numbers of quantized bits of those bands, thereby sending them to encoders 9 and 10 and a code multiplexer 11 until the 1st sampled value of the next frame is inputted. Prediction coefficients for obtaining the pseudo predictive residual signal use prediction coefficients used by the encoders 9 and 10, which use the encoding outputs to vary quantization width adaptively according to amplitude. Thus, only the encoding output of each band and information on the number of bits are outputted to make decoding on a decoding side possible.

Description

[Detailed Description of the Invention] This invention divides an input signal such as voice or music into multiple frequency bands, and uses the properties of the input signal to adaptively allocate bits for predictive coding. The present invention relates to a band division adaptive bit allocation predictive coding method.

<Prior art> Conventional high-efficiency waveform encoding methods that utilize the redundancy of speech include: (1) Adaptation that quantizes the difference signal between the predicted signal calculated from past input signals and the actual input signal. Predictive coding method, (2) Coding method that divides the input signal into multiple frequency bands, fixes the number of bits for quantizing each band signal in advance, and then applies the method in (1) above, (3) Above (2
) is realized by varying the number of quantization bits, and there is also an encoding method that periodically divides the time axis and adaptively changes the number of quantization bits in the h-direction. The method (1) above has a simple configuration, and if a method of calculating prediction coefficients from a decoded signal is used, no delay time due to encoding occurs. but,
It is somewhat difficult to achieve the same level of encoding quality as conventional companding (μm28w) PcM with half the amount of information using this method, and even higher encoding quality is required when targeting music signals. becomes. In the method (2) above, the allocation of the number of quantization bits is determined based on the statistical properties of the amount of information contained in the band-divided signal, so on average, compared to the method (1) above, The redundancy suppression effect is improved. Although the preferable number of bits to be allocated to each band changes over time, there is a problem in that the quality deteriorates in a specific section because the bit allocation is fixed. Above (3
) is characterized by the ability to transmit high-quality audio signals even when the amount of transmitted information is relatively low, but due to the complexity of the encoding algorithm, it would be large-scale if realized as hardware. Another disadvantage is that the delay time caused by encoding is large.

<Summary of the Invention> The purpose of the present invention is to make it possible to obtain high encoding quality for voice and music signals having a wider band than the telephone voice band, and to provide an adaptive system with small hardware scale and delay time. This paper proposes a band division coding method that performs prediction and adaptive bit allocation.

In this invention, a pseudo-prediction residual signal is extracted for each band by linear prediction, and the number of quantization bits for each band is calculated using the pseudo-prediction residual signal. The prediction coefficient used in the encoding means is used as the coefficient, and the quantization width is adaptively iodinated according to the amplitude using the oxidation output in the encoding means. In this way, only the encoded output of each band and the information indicating the number of bits are output, so that the decoding side can perform all-purpose decoding.

<Example> FIG. 1 shows an example of the present invention. Input signals such as audio and music signals converted into digital signals are input to the input terminal 1. This input signal is divided into two frequency bands by a low pass filter 2 and a high pass filter 3. These band-divided signals are converted into low-band signals by sampling frequency converters 4 and 5, and become so-called subband signals. The subband signals are temporarily stored in buffer units 6 and 7 before being encoded.

The buffer sections 6 and 7 have a structure as shown in FIG. 2, for example. The subband signal is input to the input terminal 23 and sequentially delayed by a sampling time by a plurality of delay devices 24 connected in series. On the other hand, the buffer units 6 and 7 have an adaptive prediction function, and the output of each delay device 24 is branched and supplied to the prediction coefficient multiplier 26 of each column.
The sum of the prediction coefficient multiplication outputs for past neighboring samples obtained from each prediction coefficient multiplier 26 is taken by an adder 27, and a difference signal between this adder 27 and the currently input signal from the terminal 23 is obtained. , that is, the pseudo prediction residual signal is sent to the subtracter 2
8 and is inputted to the bit allocation number calculator 8 in FIG. 1 through the output terminal 29.

The bit allocation number calculator 8 takes in each of the pseudo prediction residual signals from the buffer sections 6 and 7, and generates a signal between the input terminal 23 and the output terminal 25 of the entire delay device 24 in FIG. The total sum of pseudo-prediction residual signal power is calculated within the same time as the delay time. This time unit is hereinafter referred to as a frame. The number of bits allocated to each band is calculated from the ratio of pseudo-prediction residual signal powers so as to minimize the sum of quantization noise powers occurring in each band. The bit allocation decision is made after the input of the last sample value of the current frame is completed, and the encoders 9 and 10 are determined before the input of the first sample value of the next frame.
and sent to the code multiplexer 11.

The encoder 9.10 has the structure shown in FIG. 3, for example.

The output signals of the output terminals 25 of the buffer units 6 and 7 are input to the input terminal 31, and the difference with the output signal of the adaptive predictor 33 is calculated by the subtracter 32 to obtain a prediction residual signal. This prediction residual signal is divided by the step size output by the step size adaptor 38 in the divider 34, and the divided output is sent to the quantizer 36 by the quantizer input from the adaptive bit allocation number calculator 8 through the terminal 35. The signal is converted into a discrete quantized signal based on the number of quantized bits and sent to the output terminal 37.

A step size adaptor 38 receives the quantized value of the quantizer 36 and determines the step size of the next sampled value.

In this invention, in order to use a backward type method as an adaptive prediction method, which has the advantage that there is no need to transmit prediction coefficients to a decoder, encoders 9 and 10 have local decoders, and prediction coefficients are obtained from decoded signals. Analyzed and determined. Specifically, the local decoder is composed of an inverse quantizer 39, a multiplier 40, and an adaptive predictor 33, as shown in FIG. The inverse quantizer 39 is branch-supplied with the output of the quantizer 36, has a function opposite to that of the quantizer 36, and multiplies the product of this output and the output step size of the step size adaptor 38 to the multiplier 40. When calculated, the decoded signal of the prediction residual signal is obtained. The adaptive predictor 1UII 33 is a conventional linear/proximity predictor; A plurality of series-connected delay devices 42 are supplied, each delay device 4
a plurality of prediction coefficient multipliers 43 to which the outputs of 2 are branched;
Adder 44 that adds the outputs of these prediction coefficient multipliers 43
, a prediction coefficient calculator 45. The prediction coefficient calculator 45 calculates a prediction coefficient at the end of the frame using the decoded signal obtained as the output of the adder 41, and the prediction coefficient multiplier 43 calculates the prediction coefficient at the end of the frame.
In addition to updating the prediction coefficient of output terminal 4 in this invention,
6 to the buffer sections 6 and 7 of each band (Fig. 1).
Similarly, the prediction coefficients of the prediction coefficient multiplier 26 (FIG. 2) are updated. Since the prediction coefficient is thus obtained from the decoded signal of one frame before, the delay time required for adaptive bit allocation can be reduced to one frame.

The code multiplexer 11 multiplexes the allocated bit number information for either band and the codes of both band signals from the encoders 9 and 10, and sends it to the transmission path 12.8 At this time, the bit number is set at the beginning of the frame. If the information is attached, the decoder side can decode the low-frequency and high-frequency signals from the coded sequence that follows.

The decryption method is described below. The signal sent from the transmission path 12 is input to the code separator 13, and the bit allocation information is sent to the bit allocation number calculator 14.

If the number of bits in one band is known, the number of bits in the other band is uniquely fixed because the amount of transmitted information is fixed, and the number of bits allocated to both bands is determined by the code separator 13 and decoder 15, 1.
Sent to 6.

The decoders 15 and 16 have the configuration shown in FIG. 4, for example. The encoded signal of each band is input from an input terminal 47 and enters an inverse quantizer 49 . The number of quantization bits is input to the input terminal 48. Hereinafter, the process by which the decoded signal is outputted to the output terminal 58 of the post-coder is exactly the same as that described for the operation of the local decoder in the encoders 9 and 10, and the step-size adaptive fixture 50. a multiplier 51 that obtains a decoded signal of the prediction residual signal; an adaptive predictor 52; an adder 53 that obtains a decoded signal;
A delay device 54 and a prediction coefficient multiplier 55 connected in series. It includes an adder 56 and a prediction coefficient calculator 57 for obtaining a prediction signal. The explanation of its operation will be omitted.

In FIG. 1, the subband signals output from J decoders 15 and 16 are input to sampling frequency converters 17 and 18, respectively, and are returned to the sampling frequency of the original input signal. The folding of the spectrum accompanying this frequency conversion is converted into a P wave by a low pass filter 19 and a high pass filter 20, respectively, and an adder 21 adds the output signals of each filter 19 and 20 to produce a decoded signal. is the output terminal 22
obtained from.

<Effects> As explained above, the present invention is a coding method that adaptively allocates the number of quantization bits for each frame according to the polarity of the power of the band-divided signal. Encoding quality can be improved for a signal in which power is concentrated in a low frequency, such as a wideband signal having a band approximately twice as large. Specifically, for speech and music in the 7 KHz band, the S/N ratio is improved by about 3 dB compared to the adaptive predictive coding method that does not perform band division.1 When the coding quality is subjectively evaluated, Compared to the quality of a signal with the same band as above that is compressed and PCM encoded, the same level of audio quality can be obtained using half the amount of information, and the encoding quality is high despite the simple configuration. This is a method that can realize the following.

Furthermore, since the method of this invention calculates the prediction coefficients of the predictor necessary for adaptive bit allocation from the decoded signal of one frame before, it is possible to reduce the delay time due to encoding to within about 10 msec. This has the advantage of not causing any problems in practice. Further, the prediction coefficients at the encoder 9 and 10 are used as prediction coefficients to obtain a pseudo-prediction residual signal,
Furthermore, by quantizing the quantization width in accordance with the outputs of the encoders 9 and 10, it is only necessary to send out information indicating the encoded output of each band and the number of allocated bits.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a block diagram showing a specific example of the buffer sections 4 and 5 in FIG. 1, and FIG. 3 is a block diagram showing an example of the encoder 9 in FIG. 1. FIG. 4 is a block diagram showing a specific example of the decoders 15 and 16 in FIG. input terminal, 2: low-pass filter, 3: high-pass filter, 4, 5: sampling frequency converter, 6.7: buffer section, 8: bit allocation number calculator, 9.10: encoder, 1
1: Code multiplexer, 12: Transmission line, 13: Code separator,
14: bit allocation number calculator, 15, 16: decoder, 17, ts: sampling frequency converter, 19: low pass filter, 20: high pass filter, 21: adder, 22: output terminal, 23 : input terminal, 24: delay device, 2
5: Output terminal, 26: Prediction coefficient multiplier, 27: Adder,
28: Subtractor, 29: Output terminal, 30: Input terminal, 31
: input terminal, 32: subtractor, 33: adaptive predictor, 34:
Divider, 35: Input terminal, 36: Quantizer, 37: Output terminal, 38 Step size adaptor, 39: Inverse quantizer, 40: Multiplier, 41: Adder, 42: Delay unit, 43
: Prediction coefficient multiplier, 44: Adder, 45: Prediction coefficient calculator, 46: Output terminal, 47, 48: Input terminal, 49
: Inverse quantizer, 50 step size adaptor, 51:
Multiplier, 52: Adaptive predictor, 53: Adder, 514 = Delay device, 55: Prediction coefficient multiplier, 56: Adder, 57: Prediction coefficient calculator, 58: Output terminal. Patent applicant: Takashi Kusano, agent of Nippon Telegraph and Telephone Public Corporation

Claims (1)

[Claims] (1) means for dividing an input signal into multiple frequency bands;
Means for linearly predicting the signals of each of these divided bands and extracting pseudo-prediction residual signals; means for calculating the quantization focus number for each of the bands from the extracted pseudo-prediction residual signals; For each band signal,
An encoding means that performs linear prediction to obtain prediction residual signals, and quantizes these prediction residual signals using each of the calculated quantization bit numbers, and an encoding means that uses the output code of the encoding means means for changing the quantization width in the encoding means in accordance with the amplitude; and means for outputting information indicating the output code for each band from the encoding means and the calculated number of quantization bits. A band division adaptive focus allocation predictive coding method, wherein the prediction coefficients used in the encoding means are used as the prediction coefficients for extraction of the pseudo-prediction residual signal.