JPS59107630A - Adpcm decoding circuit - Google Patents

Abstract

PURPOSE:To prevent deterioration of characteristics from being accumulated even with the titled circuits connected in cascade by outputting a representative residual value signal and a high/low both-limit residual value signal to an output of an adaptive inverse quantizing circuit and forming a nonlinear PCM decoding signal by these values. CONSTITUTION:When an ADPCM code nj is inputted to a terminal 8, an inverse adaptive quantizing circuit 91 outputs a representative residual signal corresponding to a residual signal ej at a coder side and ThU and ThL representing both limits of a section. A representative decoding signal, a lower-limit decoding signal XLj and a higher-limit decoding signal XUj are obtained by passing these values through adders 105, 106, an adaptive filter 111 and a fixed filter 112. The representative decoding signal is converted into a PCM quantizing signal X by a linear-nonlinear PCM converter 120, and the X is converted into a PCM quantizing signal XRj by a nonlinear-linear PCM converter 121. When the XRj exists in sections XLj, XUj, the XRj is used for a linear input to an ADPCM code circuit of the next stage. If the XR exists at the outside of the said sections, an output adding 1 to the XRj or subtracting 1 therefrom is obtained at a selecting circuit 124.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an ADPCM circuit that alternately repeats PCM encoding and ADPCM encoding, and particularly to an ADPCM decoding circuit that does not accumulate quantization noise.

Regarding ADPCM, see 'Proceedings of IEEE' 4 published by IEEE in April 1980.
For details on pages 88 to 525, and regarding the improved ADPCM with characteristics that are strong against transmission line bit errors, see "Proceedin" published by IEEE in May 1982.
gs of ICASSP'82' pages 960-96
Details on page 3. Hereinafter, ADPCM will be explained based on the second document to the extent necessary for explaining the present invention.

FIG. 1 shows a conventional ADPCM code and decoding circuit, including an input signal terminal 1, a subtracter 2, a quantization circuit 3, an inverse adaptive quantization circuit 4, an adder 5, an adaptive prediction circuit 6, and a code output terminal. 7, and an ADPCM decoding circuit including a code input terminal 8, an inverse adaptive quantization circuit 9, an adder 10, an adaptive prediction circuit 11, and an output terminal 12.

The adaptive quantization circuit 3 is a circuit that obtains an output signal with a length of N bits smaller than M when the input signal is expressed with a length of M bits. Then, the redundancy result is expressed as N bits of the input signal X at this time.
, then the output signal is n, and the next sample time (j+
The quantization width Δ, +1 in 1) is companded using the following equation according to the quantizer input signal level.

△1+, -be・M(nρ (2) Here, M(n) is a multiplier determined by the intention of n, and the audio signal sampled at 8kHz is converted into 4 bits (N=
If β is set as a positive constant smaller than 1 in Equation (2) of Table 1, then as long as the adaptive prediction circuit is a time-invariant filter, the calculation of It is known that it is resistant to bit errors on the transmission line because it has the effect of leaking the quantization width. When the output signal of the N-pit adaptive quantization circuit is input, it outputs the reproduced input signal of M bits corresponding to the dark value, and △ +jr (3)x=n
The transmission signal is dequantized by △105△1. This X is called a representative value. The quantization characteristics shown by equations (1) and (3) are such that the width between the dark values is always constant, so the width between the representative values is also constant, and is called the linear quantization characteristic. ing. In general, the width between the dark values and the width between the representative values are not constant, and are usually set to have widths depending on the statistical distribution function of the signal to be quantized, which will be described in detail later. The transfer functions of the adaptive prediction circuits 6 and 11 are the same and are expressed as p (
z), it becomes . Here, (a) mouth=1. ..., k) is called the prediction coefficient at time j, and if the predictor input △ signal at time j is xl and the inverse quantizer output signal is e, then
Change each coefficient so as to minimize. In other words, each coefficient is a171 == (1-δ) "I+g"I・Xl-1
It is known that it is good to change (5) from time to time61. Here, δ and g are positive constants smaller than 1.

A conventional ADPCM encoding circuit/decoding circuit will be described below with reference to FIG. Input signal sample value X at time j,
is input to the ADPCM encoding circuit from terminal 1, the subtracter 2 converts the input signal At the same time, the signal is input to the inverse adaptive quantization circuit 4.

The inverse adaptive quantum △ conversion circuit 4 △ reproduces the M-bit error signal e. The reproduced error signal e and the output of the adaptive prediction circuit 6 are added to the adder 5 to reproduce the quantized input signal X. Thereafter, the quantization widths of the adaptive quantization circuit 3 and the inverse adaptive quantization circuit 4 and the coefficients of the adaptive prediction circuit 6 are used to encode the next input signal according to equations (2) and (5) as described above. amended for. As mentioned above, the coefficient correction of the adaptive prediction circuit is performed by minimizing the power of the error signal e, that is, ◇2 +
Since the e signal is modified to become 1, the dynamic range of the e signal is smaller than that of the This means that the code is small and can be encoded with high precision.

On the other hand, in the conventional ADPCM decoding circuit, the received quantization code n is input from the terminal 8, and the inverse adaptive quantization circuit 9
A reproduction error signal eI is generated.

This e and the output of the adaptive prediction circuit 1 are added by the adder △ 10, and the sum of Add. On the ADPCM decoding circuit side, the quantization width of the inverse adaptive quantization circuit is changed moment by moment according to equation (2) using the adaptive quantization code n or the error signal hole, and the difference between Xl and X, that is, e, is The coefficients of the adaptive prediction circuit 11 are changed according to equation (5) so as to minimize the power.

In the ADPCM encoder and decoder, an inverse adaptive quantization circuit 4,
If the internal states of 9 and the adaptive prediction circuit 6.11 match, the values of e + + X+ * Therefore, even if the ADPCM encoding circuit and decoding circuit are provided at a distance, the input signal X applied to terminal 1 and terminal 12
The Xl output from will take almost the same 1st shift. By the way, from the encoder terminal 7 to the decoder terminal 8
Until then, it becomes a transmission path, but there is a possibility that bit errors may occur in the transmission path due to thermal noise, etc.

In this case, the ADPCM decoding circuit often falls into an unstable state and cannot recover. This can be explained as follows.

If the transfer function D (Z) from the output △ e of the inverse adaptive quantization circuit 9 of the ADPCM decoding circuit to the multi-output terminal 12 is determined as the transfer function of the adaptive prediction circuit 11 using equation (4), the following is obtained. . 5 is a value calculated from e as described above, and a transmission path bit error occurs. ! : The correction value of the prediction coefficient of the adaptive prediction circuit of the ADPCM decoding circuit is ADPCM
This value is different from the prediction coefficient of the adaptive prediction circuit of the encoding circuit. Equation (6) has 9-poles determined by the prediction coefficients, and as a result of the above transmission line bit error, the pole position on the ADPCM decoding circuit side moves outside the unit circle on the Z plane. Sometimes it happens. In this situation, AD
The PCMYJ circuit enters an oscillation state and cannot return to normal operation again.
) is expanded as follows to realize an ADPCM encoding circuit and decoding circuit having a transfer function excluding poles that move adaptively.

Here, the coefficient (all) is a fixed constant (b: J is applied to it.If the fixed coefficient (al) is selected to a value that matches the average nature of the voice, the above-mentioned cutting error will be small, There is almost no deterioration in the encoding quality.Here, the method of determining the fixed coefficient (al) suitable for the average nature of speech is detailed on page 498 of the above-mentioned first document.

10- FIG. 2 shows a conventional ADPCM encoding circuit and decoding circuit based on equation (7). FIG. 2 shows input terminal 1, subtractor 21. ．． 22, an ADPCM code circuit consisting of an adaptive quantization circuit 3, an inverse adaptive quantization circuit 4, a W adder 51, 52, an adaptive filter 61, a fixed filter 62, and an output terminal 7;
Input terminal 8, inverse adaptive quantization circuit 9, adders 101, 10
2, an ADPCM decoding circuit consisting of an adaptive filter 111, a fixed filter 112, and an output terminal 12. Fixed filters 62 and 112 have the following transfer functions with the fixed prediction coefficients (al) used in equation (4).

P2(Z)=ha,z-'(s)1=1 Furthermore, the adaptive filters 61 and 111 have the following transmission function.

PI(Z)=Σb', Z-'
(9) Direct=Correct, but the adaptation coefficients are modified as follows, which is e
, the second thing is to be modified in the direction of minimizing the signal power.
It is stated in the literature.

bj+'=(1-δ)b:+ge,-,e,
(+o) Now, input signal X from terminal 1,
When is input, the subtracter 21 takes the difference from the output ``;Cr'' of the fixed filter 62 and inputs y to the subtracter 22.The subtracter 22 converts y to the output y of the adaptive filter.
, and is added to the adaptive quantization circuit 3. The adaptive quantization circuit 3 quantizes e, and outputs the code n from the output terminal 7, which is also applied to the inverse adaptive quantization circuit 4 to obtain a quantized error signal e.

Δe is input to the adaptive filter 61 and used for filter calculation at the next sample time, and the adaptive filter 61
The output y, is added to the adder 51 and transmitted to the adder 52 as y. The adder 52 adds y and Therefore, if the output of the fixed filter 62 is suitable for the average behavior of the input signal, the first
The amplitude level of the error signal y, is decreased, and the output of the adaptive filter 61 is subtracted from this signal, resulting in a second error signal e,
becomes an even lower level signal. Generally speaking, the first
The adaptive prediction circuit 6 shown in the figure predicts the next input signal value from the reproduced quantized input value, while the adaptive filter 61 shown in FIG.
, does not predict the next input signal from the error 1N. Although the adaptive filter 61 shown in FIG. Therefore, the encoder shown in FIG. 2 can perform encoding comparable to that of the encoder shown in FIG. 1 as a whole.

Next, the operation of the ADPCM decoding circuit shown in FIG. 2 will be explained. When the quantization code is input from the input terminal 8, the inverse adaptive quantization circuit 9 generates a quantized error signal e, inputs it to the adaptive filter 111, and uses it for the adaptive filter calculation at the next sample time. And the adaptive filter 111 for the adder 101
The output y is added to reproduce the sheath. y reproduces the output X of the identification filter fi 112 and the encoder side input signal X which has been added and quantized by the adder 102, and is supplied to the output terminal 12 and the fixed filter 112. The transfer functions PI(Z) and P2(Z) of the adaptive filter 111 and the fixed filter 112 are as shown in equations (8) and (9), and the transfer functions from the output of the inverse adaptive quantization circuit 9 to the output terminal 13- Since the transfer function D (Z) up to 12 is, it matches Equation (7) and does not have an adaptively moving pole on the Z plane, so stable operation is expected even if transmission line pitch hpb occurs. can.

In addition to the above-mentioned ADPCM, there is also an ADPCM circuit having an adaptive zero/adaptive pole type prediction filter that uses the fixed filters 62, 112 shown in Fig. 2 by limiting the range in which the poles can move. However, since it can be explained in the same way, the details will be omitted.

We have looked at the ADPCM encoding/decoding circuit above, but
Considering the introduction of an ADPCM circuit with an existing PCM network into an existing PCM network, a signal encoded by PCM will be ADPCM encoded, and then PCM encoded again and transmitted. As a result, a situation occurs in which PCM encoding and ADPCM encoding are performed alternately.

In general, operations inside the ADPCM encoding/decoding circuit are performed using linear codes of 14 bits or more because 8-bit nonlinear PCM becomes equivalent to 14 bits when linearized, so it is necessary to perform encoding comparable to PCM. There is. In this case, ADPCM encoding/decoding circuit and other ADP
If the CM code/decoding circuit can be connected with linear code bits equal to but greater than the number of ADPCM internal calculation bits, there will be no deterioration due to the connection itself even if the ADPCM code/decoder circuits are connected in cascade. For this reason, the first ADP
If the internal states of the CM code/decoding circuit and the subsequent ADPCM code/decoding circuit all match, the ADPCM code/
Even if the decoding circuits are connected in cascade, the internal state changes in the same way in each ADPCM code/decoding circuit, and no matter how many stages are connected in cascade, the deterioration of the ADPCM circuit for one stage remains.

However, as mentioned above, the ADPCM code/decoder circuit and its wires (ADPCM code/decoder circuits are connected by a non-linear 8-bit PCM code), so if they are connected in cascade, the number of usable bits will decrease and the number of bits used will decrease. This is accompanied by deterioration of the connection itself due to the non-linear weighting of each bit of the possible number of bits.The deterioration due to the connection itself by the PCM signal is due to the fact that it is different from the initial ADPCM encoding/decoding circuit. When the selected ADPCM codes are different, the multipliers shown in Table 1 given by the adaptive quantization equation (2) will be different, and the adaptation of equations (5), 2, and 4 will be different. As the prediction coefficients become different, the coincidence of the internal states breaks down.For this reason, the deterioration caused by cascade connection is due to the deterioration due to the ADPCM encoding/decoding circuit in addition to the deterioration due to the PCM connection described above. This will accumulate by the number of cascade-connected stages, resulting in very large deterioration as a whole.

Regarding the mechanism in which the coincidence of the internal states described above collapses,
IEEE 1 as long as the relationship between the threshold value and the representative value of the quantization circuit used in the ADPCM encoding/decoding circuit has the linear quantization characteristics shown by equations (1) and (3).
``Proceedings of 19'' published in 1979
79I 5CAS', pages 969 to 970, and if the states of the -direction parts match,
A method of maintaining the coincidence of internal states by utilizing the property of the linear quantization characteristic that the threshold interval and the representative value interval are coincident is also described (see Table 2 of the same document).

However, the conventional internal state maintenance method cannot be applied to an ADPCM encoder/decoder circuit having a nonlinear Doji conversion characteristic, which is generally used to improve the quantization ability. This nonlinear quantization characteristic examines the statistical distribution of the signal input to the quantization circuit and determines the threshold and representative value suitable for this distribution. For example, when the distribution function is Gaussian distribution, quantization is performed. If the number of code bits is 4, it is defined as 2, according to IRE 'Tra' published in May 1960.
nsactionsonInformationT
theory”, pages 7 to 12.As is clear from Table 2, the interval between dark values and the interval between representative values are the linear quantization characteristics given by equations (1) and (3). The difference is that it is no longer a constant width.

For this reason, the conventional method of maintaining coincidence of internal states using the fact that the threshold interval and the representative value interval are constant cannot be applied, and the encoding/decoding circuit having such a quantization circuit can be used as a nonlinear PCM code. When cascade connections were made through 17- Table 2 Gaussian distribution quantization percentage (positive sign side only), accumulation of characteristic deterioration occurred.

An object of the present invention is to provide an ADPCM encoding circuit in which characteristic deterioration does not accumulate even when ADPCM encoding/decoding circuits having nonlinear quantization characteristics are connected in cascade via nonlinear PCM encoding.

Another object of the present invention is to provide a method in which characteristic deterioration during cascade connection does not accumulate by simply adding an auxiliary circuit to the ADPCM decoding circuit without changing the characteristics of the conventional ADPCM code circuit.

18- The configuration of the ADPCM decoding circuit of the present invention is such that the residual signal, which is the difference between the signal obtained by linearizing the input nonlinear encoded PCM signal and the prediction signal predicted in the I-adaptive manner, is An ADPC that receives a code signal output from an ADPCM encoder and outputs a nonlinear PCM decoded signal.
In the M decoding circuit, from the quantization code signal from the ADPCM encoder, a representative residual value signal, a low limit residual signal, and a high limit residual signal are generated in accordance with the Ail residual signal on the encoder side. an inverse adaptive quantization circuit that determines the quantization characteristic at the next sampling time based on the quantization code signal; the representative residual value signal from the inverse adaptive quantization circuit; and the low limit residual. addition means for adding an adaptive prediction signal to be described later to each of the signal and the high limit residual signal to generate a representative decoded class signal, a low limit decoded value signal, and a high limit decoded class signal; Convert to encoded PCM,
a nonlinear PCM conversion circuit that generates a quasi-output nonlinear PCM signal, a linear PCM conversion circuit that linearizes the quasi-output nonlinear PCM signal and generates a console output linear PGM signal, the quasi-output linear PCM signal, and the low limit decoded value signal. By comparing the maximum decoded value signals, the quasi-output nonlinear PCM signal can be used as it is or +1. /-1 addition and nonlinear P
means for generating a CM decoded signal, and inputting the representative decoded type signal, or the representative decoded type signal and the representative residual value signal, generates an adaptive prediction signal at the current time, and generates an adaptive prediction signal at the next sample time. It is composed of at least an adaptive prediction circuit that determines prediction characteristics, and outputs not only a representative residual value signal but also high and low extreme residual value signals as the output of the adaptive inverse quantization circuit. The invention will now be described in detail with reference to the following figures of decoded value signals. FIG. 3 is an embodiment of the ADPCM circuit shown in FIG. 2 of the present invention, and the input terminal 8
, inverse metering circuit 91, adders 101 to 109, adaptive filter 111, fixed filter 112, linear-nonlinear PCM
conversion circuit 120, nonlinear-linear PCM conversion circuit 121,
It consists of a comparison circuit 123, a selection circuit 124, an output terminal 126, an adaptive filter 111, a fixed filter 11.
21. IJII calculators 101 and 102 are ADPCs in FIG.
This is the same as the M decoding circuit. Also, linear-nonlinear P
CM conversion circuit 120, nonlinear-linear PCIV [Details of conversion circuit 121 are provided in September 1970 Be1l System
"BSTJ" published by Laboratories 1
Those described on pages 555 to 1588 are available. When the inverse adaptive quantizer 91 receives the input ADPCM code n, it outputs a value obtained by multiplying each of the representative value, threshold, and n+1 threshold corresponding to n shown in Table 2 by the quantization width Δ. In this way, the representative value takes a form that represents the interval indicated by the Ryoya value. When n is 8, a sufficiently large value (for example, 10000) is virtually set and used as the threshold for n+1.

Assuming that an ADPCM code n is now input to the terminal 8, the inverse adaptive quantization circuit 91 sets the current quantization width Δ, Outputs the value multiplied by . This output signal includes a representative residual signal e1 corresponding to the residual signal e on the encoder side,
Let this representative residual signal 21- be ThL. Adaptive filter 11 and fixed filter 1
12, while the predicted value at the current time is being output, adders 105 and 106 add the output predicted values of the adaptive filter 111 and the fixed filter 12, respectively.
A representative decoded signal X, a low limit decoded signal xL1, and a high limit decoded signal xr are obtained, respectively.

Here again, the representative decoded signal X is in the section CX"r' + X
+ ) is a representative value.

The representative decoded signal X is converted into a normal 8-bit PCM code
Childification signal x? is converted to

Now, consider the case where xR exists within the interval Cx";+'x':). In this situation, the comparator 123 is the selection circuit 124
If we assume that the internal state of the ADPCM code circuit in the next stage is the same as the current internal state of the PCM code X, xl is used as a linear force in the next stage ADPCM code circuit, and [x7. xT) is assigned the same ADPCM code as the active one. Unfortunately, the internal states of the active and next-stage ADPCM encoding/decoding circuits are the same.

Next, consider the case where XR is not in the interval (x'',

X〒) should contain at least one representative value of PCM. (If the PCM representative value is not in this section, the ADPCM code that generated this section should not be selected.

) Furthermore, the representative value X, which is within the interval [XF,
Is the M quantized value x? Therefore, the quantization threshold of PCM exists at (x''; +41), and since xR>Xu, the quantization width of PCM is 2 (XR-II

1△
It will be understood that this occurs when the quantization width of ``M'' becomes from one to at most twice the quantization width of ADPCM.

In this case, X? Nonlinear PCM code X that generated
The difference between the input nonlinear PCM code of the current ADPCM code circuit and the input nonlinear PCM code of the current ADPCM code circuit is that be. Since the nonlinear PCM code is a special polarity amplitude expression, the comparison circuit 123 activates the selection circuit 124 in this situation, and when X7 is positive, the adder 107 adds + to X.
1. When the value is negative, the input PC of the active ADPCM code circuit is used to select the value obtained by adding 1 to X by the adder 108 as the output.
It will be understood that the M signal and the input PCM signal of the next stage ADPCM code circuit are completely equal, and the coincidence of internal states is maintained.

Furthermore, xR is not in the interval (xL; , xT ), and x
7 Consider the case of (x”;. In this case as well, at least one PCM representative value should be in the interval [xL, X+, +).
Is the CM quantized value x? Therefore, the U threshold for PCM quantization exists at (X,,!,). For this reason, t of PCM
It child width C2 (/;C-xR) ~2 (xu-x''
) Tonari, II II It will be understood that this also occurs when the quantization width of PCM becomes from one to at most twice the quantization width of ADPCM. In such a case, the nonlinear PCM code X that generated X and the input nonlinear PC of the active AI) PCM code circuit
It is obvious that the difference from the M code is that X is a PCM code that is smaller by 1. Therefore, the comparison circuit 12
3 operates the selection circuit 124 in this situation and selects X? When is positive, adder 108 adds -1 to X, and when it is negative, adder 107
In order to select the +1 value as the output, the active AD
The operation of the filter in which the input PCM signal of the PCM code circuit and the input PCM signal of the next-stage ADPCM code circuit are completely equal is as explained using FIG. 2 as an explanation of the conventional ADPCM.

As described above, according to the present invention, even if ADPCM code/decoding circuits are connected in multiple stages via PCM code/decoding circuits, if the internal states of the ADPCM code/decoding circuits match, the ADPCM code/Enoki code It is possible to realize an ADPCM decoding circuit which has a property that the characteristics deteriorate only by one stage of the circuit.

25- Also, although the third factor has been explained with respect to the ADPCM circuit of FIG. 2, the present invention can be easily applied to the ADPCM circuit of FIG. Furthermore, the ADPCM in Figure 2
Although the prediction filter 112 in the circuit is a fixed filter, the essence of the present invention does not change even if it is an adaptive filter.

Furthermore, as can be easily inferred, the output of the inverse adaptive multipletization h' Δ unit 91 is expressed as e, , (ThL−e, ), (
The present invention also includes a method of obtaining by adding Th)-e, ) and (ThLer), respectively.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a conventional ADPCM encoding/decoding circuit, FIG. 2 is a block diagram showing another conventional ADPCM encoding/decoding circuit, and FIG. 3 is a block diagram showing an ADPCM decoding circuit of the present invention. . In the figure, 91・−inverse adaptive multiplierization+i, l, 111・
...Adaptive filter, 112--Fixed filter, 101-
10826-...Adder, 120...Linear-nonlinear PCM converter, 121...Nonlinear-linear PCM converter, 123...
. . ratio I bite circuit, 124 . . . selection circuit. 11- Rokuko

Claims (1)

[Claims] An ADPCM encoder that adaptively quantizes a residual signal that is the difference between a signal obtained by linearizing an input nonlinearly encoded PCM signal and/or an adaptively predicted prediction signal at each sample time. In an ADPCM decoding circuit that receives a code signal output from the ADPCM encoder and outputs a nonlinear PCM decoded signal, a representative an inverse adaptive quantization circuit that generates a residual value signal, a low limit residual signal, and a high limit residual signal, and determines a quantization characteristic at a next sampling time based on the quantization code signal; An adaptive prediction signal, which will be described later, is added to each of the representative residual value signal, the low limit residual signal, and the high limit residual signal from the inverse adaptive quantization circuit to obtain a representative decoded value signal,
Low limit decoded value signal. and an addition means for generating a high limit decoded value signal; a nonlinear PCM conversion circuit for converting the representative decoded signal into a nonlinear encoded PCM and generating a quasi-output nonlinear PCM signal; and a nonlinear PCM conversion circuit for linearizing the quasi-output nonlinear PCM signal and a linear PCM conversion circuit for outputting a linear PCM signal; a means for comparing the desk output linear PCM signal with the low limit decoded value signal and the high limit decoded value signal; inputting the decoded value signal and the representative residual value signal;
It is composed of at least an adaptive prediction circuit that generates an adaptive prediction signal at the current time and determines prediction characteristics at the next sample time, and performs PCM encoding/decoding and ADPCM encoding/
In a tandem connection state where decoding is repeated alternately, once the internal states of each ADPCM coder/decoder circuit completely match, from then on, it is assumed that the characteristic deterioration due to the tandem connection is only for one ADPCM stage. ADPCM decoding circuit that can do the same.