JPS6243223A - Method for adapting adpcm coder to quantizer - Google Patents

Abstract

PURPOSE:To prevent a wrong shift to the high-speed adaptation and to attain the adaptive quantization with high quality by detecting the change of an estimating signal as well as the change of a digital input signal supplied to a difference circuit which subtracts the estimating signal from a digitized input signal and suppressing the follow-up of the adapting speed against the increase of the difference signal. CONSTITUTION:For an ADPCM coder, the adapting speed is increased in response to the increase of the estimated difference to produce an unstable answer when the low-speed adaptation is applied to the voice band data signal. Thus, the adaptation is discontinued if the increase of the estimated difference is due to the wrong follow-up. then, the state of a quantizer is held at that time point to prevent the shift to the high-speed adaptation. For instance, an input signal S(j) is supplied to a difference circuit 2 from a terminal 1 at a time point (j) and the difference is obtained between the signal S(j) and an estimated signal Se(j). Thus, a difference signal d(j) is supplied to a quantizer 3. Here, the signal d(j) is scaled by a scale factor Y(j) and controlled to be set at the level of the quantizer 3. Then, the differential PCM quantization value I(j) is delivered to an output terminal 8 and an adverse quantizer 4.

Description

Detailed Description of the Invention [Field of Application of the Invention] The present invention relates to an adaptive differential PCM encoder (hereinafter referred to as an ADPCM encoder) used for data compression of voice, voice band data signals, tone signals, etc. ) relates to a quantizer adaptation method.

[Background of the invention]

Conventional ADPCM encoding is described in, for example, Japanese Patent Publication No. 59-500.
As stated in Publication No. 077 and CCITT Recommendation G721, when the input signal is audio, high-speed adaptation is required.
In the case of voice band data signals or tone signals, a slow adaptation is applied to the quantizer to follow the input signal.

In this conventional ADPCM encoding method, when the error between the input signal and the predicted signal of the input signal becomes large, the quantizer shifts from the slow adaptation state to the fast adaptation state. This may cause inconveniences such as being damaged.

[Purpose of the invention]

An object of the present invention is to prevent erroneous adaptation to changes in prediction residuals and improve quantization characteristics through stable adaptation when a voice band data signal is input to an ADPCM encoder. The purpose is to provide a good quantizer adaptation method.

[Summary of the invention]

In an ADPCM encoder, when a voice band data signal is adapted at a low speed, the adaptation speed shifts to a high speed as the prediction residual increases, resulting in an unstable response.

Therefore, in the present invention, it is possible to determine whether the increase in the prediction residual is due to a rapid amplitude change in the voice band data signal, insufficient tracking of the predictor, or incorrect tracking, and in the latter case, adaptive By stopping the quantizer and maintaining the quantizer state at that point, the transition to high-speed adaptation is prevented. This results in a stable adaptive quantizer.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram of an ADPCM encoder, which includes an input terminal 1, a difference circuit 2, a quantizer 6, an inverse quantizer 4, an adder 5, a predictor 6, an adaptation circuit 7, and an output terminal 8.

The input signal SO1 at time 0 is input to the difference circuit 2 from the terminal 1, and the difference between it and the predicted signal 5-0) is taken, and the difference signal d(
71 is input to the quantizer 3. There, d(71 is scaled and adjusted to fall into the level of quantizer 3, 2-ru) is quantized and encoded by a scale factor Y, and the differential PCM quantized value IO) is output to the output terminal 8 and the inverse quantizer 4. The inverse quantizer 4 performs an inverse operation on 7[1, outputs the value dgIjl obtained by quantizing d(71), and the adder 5 outputs the sum with the output signal S*(71) of the predictor 6. input signal 59'l
A reproduced signal S r (j) is given and input to the predictor 6. The signal Sr9°] is used to generate the prediction signal Sg(7+1) used at the next sampling time +1.

On the other hand, to the adaptation circuit 7, the above 1sl and 5rljl
.

Souo) is given and used to generate the scale factor Y()+1) at the next time. In FIG. 2, the adaptation circuit 7 when the number of quantizer bits is 4 bits will be explained in detail.

The adaptation circuit in Fig. 2 is a function F(・) that amplifies the fluctuation of the output signal 1 (71). ), long time constant LPF 1
3 (e.g. 1fmzac), these two L P
F10.

A comparator 12 that compares the difference in the output of [l] with a predetermined threshold and outputs 1 when the threshold is exceeded and 0 otherwise, and a short time constant L P 113 that smoothes the output of the comparator 12.
(e.g. 2m5aa), output signal 11j) to generate a number for updating the scale factor Y91 114, LP115 to introduce finite memory into the adaptation process to reduce the influence of transmission path performance 7. for smoothing the output of the filter 15; PFlB.

Multipliers 17 and 18, subtractor 19, and adder 20 for taking the arithmetic average of the respective outputs of these two L P F 15 and 16 using the output value of L P F s3 as a weighting coefficient.
, the output signal I(7) is the overload value (II(71+=7)
a comparator 21 that detects whether the change in the reproduced signal 5rljl is in an increasing direction or a decreasing direction;
Comparator 21, differential circuit 22, 23.1. It consists of a determiner U which determines from the signal given from PFlB whether the overload has arisen from a rapid change in the input signal SQl or from an insufficiency of the predictor.

First, the comparator 21 detects whether or not l l1j)1=7, and the difference of the υ order is taken between the reproduced signal 5r51 and the predicted signal 5sQl by the difference circuits 22 and 23;
In OhSf(>-1), 5s(jhsacj-1>determiner S, the overload occurs when "α, sl
< 1/2, then Hn(Srljl-5rCj-1))"qzln(So
If ul 5g(j-1)), it is determined that the cause is insufficient ability of the predictor or a prediction error, and outputs that to LPFlo, u. Here, snx is 1 (J≧O) or -1 (
It takes a value of 0). Note that α, 9) is the output signal of L P 713, and as described later, it approaches 1 if the input signal is voice, and approaches 0 if it is a voice band data signal or tone signal, so ap(7'l < 1 /2 indicates the latter input signal.

The fluctuation of the function F (S9 is according to the following table, I I)') is amplified and output.

It should be noted that Fσ)=7 gives a large weight to fluctuations during overload, which greatly affects the outputs of L P F 10 and 11 at the subsequent stage. L P F 10 ,
Fl is equivalent to taking a short-time average and a long-time average, respectively, and these are expressed as dnhzlj) and dnhaljl. Also, when the output of the determiner indicates an overload due to poor tracking of the predictor 6, drLzlj)=ttmz(j-1), drpsa
(j)=cLmacj-1) and stop adaptation. The comparator 12 has a certain threshold value T and outputs 1 or 0 depending on whether the difference between dnszlj) and dmmJ) is large or small. (0<T×1) For example, (1-T)dFIKg(7)<dm#(r+<
(1 + T) Mu 10) → 0 Otherwise →
1, then for a stationary signal with certain statistical properties such as a voice band data signal, dmslj) and dma9'
l are close to each other and the comparator outputs O.

L P 113 smoothes the output and gives ip (7+ 1 ) which varies slowly between 0.1.

The function S.14 is for outputting the number for updating the scale factor Y (7), and it is used to output the number for updating the scale factor Y (7).
Output v(7). Jin's robust fast scale factor Yu(j) is smoothed with L P F 16 to give a scale factor Ya(jl, which is convenient for signals such as voice band data. These YμIj), Ya(
Calculate the arithmetic mean of jl with the weight of αp (,)' + 1) to obtain Y()+1); 1'(7+1) -4pO+1)Yu(71+ (
1-αpO"1)Ya (71According to this embodiment, this is due to the tracking defect of the predictor 6 that occurs when audio band data with large amplitude changes such as 9.6kApz of C1TT and V29 recommendations is input. Even during overload, drp
sa5), dnha(j) is not updated, therefore α,
0) does not move toward 1, Y (7'l remains the slow scale 7 actor Y-0), which has the effect of not resulting in unstable adaptation.

〔Effect of the invention〕

According to the present invention, when a stationary signal with almost constant statistical properties is input, such as a voice band data signal or a tone signal, the increase in the residual signal due to the tracking defect of the predictor is prevented. , it is possible to prevent erroneous transition to high-speed adaptation by considering rapid amplitude changes of the input signal, so high-quality adaptive quantization can be expected for the above-mentioned signals.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an ADPCM encoder according to an embodiment of the present invention, and FIG. 2 is a detailed block diagram of the adaptation circuit in FIG. 1.

Claims (1)

[Claims] ADPCM in which the output signal of a difference circuit that subtracts the predicted signal of the input signal from the digitized input signal input at each sampling time is quantized by an adaptive quantizer and converted into a differential PCM signal.
In the quantizer adaptation method of controlling the adaptation speed of the adaptive quantizer of the encoder according to the input signal, detecting a change in the digital input signal input to the difference circuit and a change in the prediction signal, When the adaptation speed is slow, the above two changes are observed, and when the difference signal increases due to the opposite direction of change, the adaptation speed is suppressed from following, and the state of the adaptive quantizer is maintained. A quantizer adaptation method for an ADPCM encoder, characterized in that: