JPS6251542B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an adaptive differential PCM (ADPCM) method that is applied to, for example, band compression transmission of voice band signals and is equipped with an adaptive quantizer and an adaptive predictor, and an apparatus used therefor. It is.

As a method for encoding and transmitting an audio signal waveform with a low transmission bit capacity, it utilizes the fact that there is a correlation between the sample values when the audio signal waveform is sampled at the Nyquist rate. In other words, the signal at the current time is predicted using a signal sequence transmitted at least one sample time before the current time, and the residual after subtracting the predicted value from the signal at the current time, that is, the prediction residual, is quantized, encoded, and transmitted. Differential PCM
method, so-called DPCM (Differential Pulse Code)
Modulation method is known. In addition, as a method to further reduce the transmission bit capacity, we use an adaptive quantizer in which the quantization step size in the quantizer is adaptively changed according to the quantizer input signal, and a prediction coefficient value in the predictor as a prediction residual. ADPCM (Adaptive
DPCM) method is known. The adaptive predictor in this ADPCM method can be divided into two types depending on the prediction coefficient correction method.

One type is called a forward adaptive predictor, which calculates an autocorrelation function using an input signal of a certain time length, and uses this autocorrelation function to minimize the prediction residual at a certain time interval. It calculates the predicted coefficient value. For more information,
For example, in the US journal Bell Systems Technical Journal, October 1970 issue (BSTJ,
“Adaptive Predictive Coding of Speech Signals” (October 1970, PP.1973-1986)
Predictive Coding of Speech Signals.''

The other type is called a backward adaptive predictor, which uses the quantized prediction residual and the input value of the predictor to sequentially modify the prediction coefficient values to minimize the prediction residual. It is something to do.
For details, see the American magazine “IEE Transactions on Communications September 1975” (IEEE
Transactions on Communications;
September 1975, pp.935-941) “The Residential Encoder
Speech Digestification” (“The
Residual Encoder−An Improved ADPCM
System for Speech Digitization.'').

Adaptive quantizers are also divided into two types depending on the method of modifying the quantization step size. One is called a forward adaptive quantizer.
The quantization step size is calculated at fixed time intervals based on an input signal of a fixed time length. For details, see e.g. P. Noll.
"Bell Systems Technical Journal November 1975 issue" (BSTJ,
November 1975, pp. 1594-1614)
Quantization Schemes for Speech Encoding (“A Comparative
Study of Various Quantization Schemes for
The other type is called a backward adaptive quantizer, which uses the current quantization step size and adaptive quantizer output level. The quantization step size at the next sample time is determined based on .For details, see, for example, the American magazine Bell Systems Technical Journal by N.S. No. (BSTJ, September 1973, PP, 1119-1144)
“Adaptive Quantization”
With a One-Word Memory” (“Adaptive
Quantization With a One-Word Memory.”)
This is described in the document 4 entitled, etc.

Of the ADPCM methods described above, this invention particularly relates to the ADPCM method using a backward adaptive predictor and a backward adaptive quantizer, and therefore it is necessary to understand the basic configuration of this method and the conventional method. This aspect will be explained in more detail using the drawings.

In the following description, in order to make the explanation more detailed and easier to understand, the backward type adaptive predictor is divided into an adaptive predictor and a backward prediction coefficient value correction circuit, and the backward type adaptive quantizer is divided into an adaptive predictor and a backward prediction coefficient value correction circuit. The adaptive quantizer and the backward quantization step size correction circuit will be explained separately.
FIG. 1 shows a conventional A using a backward adaptive predictor and a backward adaptive quantizer.
This shows an example of the DPCM method.
In FIG. 1, on the transmitting side 11, a subtracter 12 subtracts the predicted value x^ _j , which is the output of the adaptive predictor 14, from the input signal _xj from the input terminal 13 to obtain a prediction residual _ej . The prediction residual e _j is the adaptive quantizer 15
is quantized to obtain a quantized output level l _j . The quantized output level l _j of the encoder 16
is converted into a code q _j and transmitted to the transmission line 17. Further, the quantized output level l _j is dequantized by an adaptive dequantizer 18 to obtain a signal e^ _j . The signal e^ _j is added to the output predicted value x^ _j of the adaptive predictor 14 by an adder 19 to obtain a locally decoded signal _x〓j . The local decoded signal x〓 _j is input to the adaptive predictor 14 to obtain the predictor output value, that is, the predicted value x^
get _j .

Here, a backward type quantization step size correction circuit 21 is provided, which is described in, for example, the previous document 4.
The correction method shown in is applied. In that case, the quantization output level l _j of the adaptive quantizer 15 is used to modify the quantization step size according to the following equation.

△ _j+1 = △ _j・M(l _j )+C (1) In equation (1), △ _j corresponds to the quantization step size at time j, and M(l _j ) corresponds to the quantization output level l _j . C represents a constant that determines the lower limit of the quantization step size.

A backward prediction coefficient value modification circuit 22 is provided, and in this example, a learning identification method is used as a method for modifying the prediction coefficient value by this circuit 22. For details, see, for example, J. Nagumo (J.
IEEE Transactions Automation Control June 1967 issue (IEEE
Transactions Automatic Control
“A Learning Method for System Identification” published in “June 1976PP.282-287”)
System Identification."). In other words, the prediction coefficient values are corrected according to the following equation.

In equation (2), a ^j _i is i at time j (i=
1, 2, . . . , N)-th prediction coefficient values, where N represents the prediction order in the adaptive predictor 14. Also, g is selected to be a minute amount of about 0.1.

However, for the purpose of temporally attenuating the influence of transmission code errors caused by noise generally generated in a line, for example, as shown in the above-mentioned document 2,
The following equation is used instead of equation (2) Here, δ is a minute amount less than 1, and serves to attenuate the influence of transmission code errors over time with a time constant determined by 1/δ. Other signals are the same as those in equation (2) above.

On the receiving side 20, the code q _j received from the transmission line 17 is input to a decoder 23 to obtain a signal l _j . The signal l _j is input to the adaptive inverse quantizer 24, and the signal e^ _j
get. The signal e^ _j is added to the output value x^ _j of the adaptive predictor 26 by the adder 25, and is supplied to the output terminal 27 from which the reproduced signal _x〓j is obtained. Here, the backward prediction coefficient value correction circuit 28 on the receiving side
and backward type quantization step size correction circuit 29 respectively perform the same operations as the backward type prediction coefficient value correction circuit 22 and backward type quantization step size correction circuit 21 on the transmitting side, and the former performs the same operation as, for example, according to the above equation (3). The prediction coefficient value is modified, and the latter modifies the quantization step size, for example, according to equation (1) above.

It has the conventional backward adaptive quantizer and backward adaptive predictor described above.
The ADPCM method (hereinafter referred to as the backward type ADPCM method) has a simpler device configuration than the ADPCM method (hereinafter referred to as the forward type ADPCM method) which has a forward type adaptive quantizer and a backward type adaptive predictor. There is an advantage that the amount of transmitted bits does not increase because there is no need to transmit the quantization step size and the predictor prediction coefficient values. However, as will be explained below, it has the disadvantage of being sensitive to transmission code errors.

In other words, in a backward adaptive quantizer, when a transmission code error occurs, the quantization step sizes on the transmitting and receiving sides become inconsistent, and the quantization step size becomes the predetermined upper limit of the step size or the above (1). The disadvantage is that the quantization step sizes on the transmitting and receiving sides do not match until the lower limit of the step size shown in equation (2) is reached, and it takes time to recover from the effects of transmission coding errors.

In addition, in the backward type adaptive predictor, when a transmission code error occurs, the prediction coefficient values on the transmitting and receiving sides become inconsistent, and the influence of the transmission code error changes over time with a time constant determined by 1/δ. The disadvantage is that it takes a considerable amount of time for the predicted coefficient values to completely match, although the attenuation is achieved.

In recent years, systems have appeared that improve the above-mentioned drawbacks and have a certain degree of robustness against transmission code errors.
An example of a backward adaptive quantizer is DJGOODMAN.
American magazine “I.E.E. Transactions on Comiunications” by Mr.
November 1975 issue” (IEEE Transactions on
Communications.November1975.PP.1362−
1365) “A Robust Adaptive Quantizer”
In addition, an example of a backward type adaptive predictor is Japanese Patent Application No. 53-123255 (Reference 7). A backward ADPCM method using a backward adaptive quantifier and a backward adaptive predictor will be explained with reference to the drawings.

FIG. 2 is an example of a block diagram of a backward type ADPCM system that uses a backward type adaptive quantizer according to Reference 6 and a backward type adaptive predictor according to Reference 7 and is less susceptible to transmission code errors. In the figure, the backward adaptive quantizer is shown as an adaptive quantizer 15 and a backward quantization step size correction circuit 21, and the backward adaptive predictor is shown as an adaptive predictor 14 and a backward quantization step size correction circuit 21. It is described separately from the numerical value correction circuit 22. In addition, in the figure, a subtracter 12 and an encoder 1
6, decoder 23, and adder 25 perform the same operations as the components with the same numbers in FIG.

In FIG. 2, the backward quantization step size correction circuit is on the transmitting side (component number 21),
The receiving side (component number 29) also performs the same operation. As a method for correcting this quantization step size, the algorithm described in the above-mentioned document 6 is used.
The algorithm is shown in the following equation.

△ _j+1 = (M(l _j )・△ _j ) (4) In equation (4), △ _j is the quantization step size at time j, and M (l _j ) is the quantization output level l _j
The coefficient β determined corresponding to is a positive constant smaller than 1. The fact that this algorithm is not easily affected by transmission code errors is described in detail in the above-mentioned document 6, so the explanation will be omitted here.

Regarding the adaptive predictor, compared to the adaptive predictor in FIG. 1 which configures a recursive filter on the receiving side, _the adaptive predictor in FIG. Since the output of the adaptive inverse quantizer 24 is supplied to the predictor 26 without being fed back to the predictor 26 to form a nonrecursive filter, it is less susceptible to transmission code errors. In addition, the backward prediction coefficient value correction circuit in FIG. 2 performs the same operation on both the transmitting side (component number 22) and the receiving side (component number 28), and uses the same type of equation as described in Document 7. Correct the prediction coefficient value according to the following formula.

In the above equation, a ^j _i represents the i-th predictor coefficient value at time j. N is the number of prediction taps of the adaptive predictor. δ and g are the same as in equation (3). In the second term on the right side of equation (6), the portion other than g is the normalized input signal e^ _j of the predictor 14. According to equation (6), even if a transmission code error occurs at time j and the values of the output e _^ (According to equation (6), N time)
If the time period elapses, the influence of the transmission code error on the second term on the right side of equation (6) becomes zero.

The adaptive predictor performs the operation shown in the following equation on both the transmitter side 14 and the receiver side 20.

In the above equation, x^ _j is the output value of the adaptive predictor, and N represents the number of prediction taps.

Therefore, compared to the ADPCM method shown in FIG. 1, the backward ADPCM method with the configuration shown in FIG. 2 has the advantage of greatly improved resilience against transmission code errors and extremely short recovery time. Ta.

However, the backward type described in Figure 2
The use of the ADPCM method has the drawback of deteriorating the symbol error rate for data modem signals. This is the prediction coefficient value correction algorithm (formula (6) above) in the backward prediction coefficient value correction circuit.
This is thought to be due to.

Generally speaking, short-term deterioration of the input signal waveform due to encoding and decoding hardly causes quality deterioration in voice communication, which is communication between humans. However, data communication, which is machine-to-machine communication, is extremely sensitive to the short-term deterioration and is easily susceptible to quality deterioration. Generally, the symbol error rate is used as a measure to evaluate the degree to which a data signal deteriorates when encoding and decoding is performed.If this value is very bad, the receiving side will It becomes impossible to demodulate.

The purpose of this invention is to enable high-quality transmission of audio signals with a device configuration comparable to that of the conventional ADPCM method.
A backward type that can obtain an extremely low symbol error rate for data modem signals.
Our objective is to provide an ADPCM method and equipment for use with it.

According to the present invention, in an ADPCM method using an adaptive predictor configured with an acyclic filter on the transmitting side, the input value of the adaptive predictor is normalized, but the output value of the adaptive predictor is taken into account. Let's do it. The prediction system numerical values are corrected based on this normalized value. Compared to the backward ADPCM method with the configuration shown in FIG. 2, the backward ADPCM method according to the configuration of the present invention can obtain an extremely low symbol error rate even for data modem signals, and is capable of high-quality transmission. There is an effect.

FIG. 3 shows an embodiment of the ADPCM method according to the present invention. Components in FIG. 3 labeled with the same numbers as in FIG. 2 operate in the same way as in FIG. 2. Therefore, the backward quantization step size correction circuit (transmitting side 21, receiving side 29) performs the above-mentioned (4).
Modify the quantization step size according to the formula.

Next, components related to the present invention that are different from the components shown in FIG. 2 in this embodiment will be explained. On the transmitting side 11, an adder 31 is used to calculate the input value e^ _j and the output value x^ _j of the adaptive predictor 14.
and obtain the adder output x〓 _j . The output value e^ _j of the adaptive inverse quantizer 18 and the output value x〓 _j of the adder 31
is input to the backward prediction coefficient value correction circuit 22. Backward type prediction coefficient value correction circuit 22
corrects the prediction coefficient value according to the following equation using the above e^ _j and x〓 _j .

In this equation, a ^j _i represents the i-th prediction coefficient value at j sample time. g is a sufficiently small positive constant less than 1, and δ has the same function as the same symbol in equation (3) above. N is the adaptive predictor 15
represents the predicted order of As can be understood by comparing equations (8) and (6), in this invention, the normalization operation for e^ _j is performed by adding the output value x^ _j , that is, by adding x〓 _j . It's on.

Now, if the prediction coefficient value expressed by the above equation (8) is very close to the value suitable for the input signal, the output value x^ _j of the adaptive predictor 14 will be very close to the input sample value x _j . Therefore, the input sample value x _j
e _j which is the difference signal between the output value x _j of the adaptive predictor and the adaptive predictor output value x j becomes small, and therefore the value of the output value e _j of the adaptive inverse quantizer 18 also becomes small. In other words, the x _j
The ratio between and e^ _j becomes considerably large. Since the output value x〓 _j of the adder 31 is the sum of the above-mentioned e^ _j and x^ _j , it can be considered to be close to the above-mentioned input sample value _xj . Therefore, the ratio between x〓 _j and e^ _j becomes considerably large. Therefore, if we consider the above equation (8), it is clear that the value of the second term on the right side is considerably small. This means that the prediction coefficient value in the backward prediction coefficient value correction circuit 22 is
It is shown that by modifying according to the formula, when the predictive coefficient value approaches a value suitable for the input signal, the modification of the predictive coefficient value can be made almost impossible.

On the receiving side 20, the backward prediction coefficient value modification circuit 28 performs the same operation as the prediction coefficient value modification circuit 22 on the transmission side 11. That is, the prediction coefficient value is corrected according to the above equation (8).

As explained above, by adopting the configuration of this embodiment, it is possible to converge the prediction coefficient value in the backward prediction coefficient value correction circuit to a value suitable for the input signal.
Therefore, the deterioration of the data signal symbol error rate when there is no transmission code error, which was a drawback in the conventional backward ADPCM method shown in FIG. 2, is eliminated, and an extremely low symbol error rate can be obtained, resulting in high quality. This has the advantage that transmission can be performed. Of course, high quality transmission is also possible for audio signals.

Furthermore, the configuration of this embodiment has the effect of shortening the recovery time from the influence of transmission code errors.

In addition, prediction coefficient value correction formula (8) in this example
The formula may be as shown below. That is, In this formula, M is a positive constant of 1 or more.

Furthermore, the sum of M pieces of (x〓 _j-1 ) ² in equation (10) may be expressed as a leakage integral of (x〓 _ji ) ² . By doing so, it is believed that the recovery time from the effects of transmission code errors will be further shortened. Furthermore, the calculation formula for the denominator in the second term on the right side of equation (10) is x〓 _ji
It is also possible to use only the sum of the squares of . Also,
Furthermore, |x〓 _ji | may be used instead of the square of x〓 _ji . By doing so, it becomes possible to significantly reduce the amount of calculations. In the above description, a part or all of each section may be operated by program processing using a microcomputer.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing an example of a conventional backward ADPCM system having a backward adaptive quantizer and a backward adaptive predictor.
The figure is a block diagram showing an example of a backward type ADPCM method that is less susceptible to transmission code errors and has a backward type adaptive quantizer and a backward type adaptive predictor that are less susceptible to transmission code errors. FIG. 2 is a block diagram showing an embodiment of the ADPCM method according to the present invention. 12: Subtractor, 13: Input terminal, 15: Adaptive quantizer, 17: Transmission line, 18, 24: Adaptive inverse quantizer, 25: Adder, 14, 26: Adaptive predictor, 21, 29: Backward type quantization step size correction circuit, 22, 28: Backward type prediction coefficient value correction circuit, 16: Encoder, 23: Decoder.

Claims (1)

[Claims] 1. On the transmitting side, an input signal is adaptively predicted using an acyclic filter type adaptive prediction means, and the difference between the input signal and the output value of the adaptive prediction means is adaptively predicted. The quantization means adaptively quantizes, encodes and transmits the code, and the adaptive inverse quantization means dequantizes the code and inputs it to the adaptive prediction means, and the receiving side receives the code and performs the adaptive quantization. The received code is dequantized by an adaptive dequantization means having a characteristic opposite to that of the transmission means, and the original signal is reproduced using an adaptive prediction means that performs the same operation as the transmission side adaptive prediction means in the form of an acyclic filter. adaptive differential
In the PCM method, the result of the inverse quantization and the output value of the adaptive prediction means are added, the square of the added value and the square of the result of the inverse quantization are added, and the added value is inversely An adaptive differential PCM method characterized by correcting prediction coefficient values using a proportional value as a correction term. 2. On the transmitting side, the input signal is adaptively predicted using an adaptive prediction means in the form of an acyclic filter, and the difference between the input signal and the output value of the adaptive prediction means is calculated using an adaptive quantization means. and converting the output value of the adaptive quantization means into a code sequence, transmitting the obtained code sequence at a predetermined sampling time, and converting the output value of the adaptive quantization means into a code sequence. An adaptive differential PCM transmitter that dequantizes the resultant data by an adaptive dequantization means and inputs the resultant to the adaptive prediction means, the output value of the adaptive prediction means and the output of the adaptive dequantization means on the transmitting side. , and add the square of the added value and the square of the result of dequantization by the adaptive dequantization means on the transmission side, and calculate the prediction coefficient value using a value inversely proportional to the added value as a correction term. an adaptive differential PCM transmitting device having predictive coefficient value modification means for modifying the prediction coefficient value, the receiving side receives the code sequence, decodes the code sequence using a decoder, and converts the output value of the decoder into the Inverse quantization is performed using adaptive inverse quantization means having characteristics opposite to those of the adaptive quantization means on the transmitting side, and based on the result of the inverse quantization, an acyclic filter is configured, which is the same as the adaptive predicting means on the transmitting side. An adaptive differential PCM receiving device that reproduces an original signal using an adaptive prediction means that operates, the output value of the adaptive prediction means on the reception side and the result of the adaptive dequantization means on the reception side. , the square of the added value and the square of the result of inverse quantization on the receiving side are added, and a value inversely proportional to the added value is used as a correction term to calculate the prediction coefficient value of the adaptive prediction means on the receiving side. Adaptive difference with prediction coefficient value correction means to correct
An adaptive differential PCM transmitter/receiver including a PCM receiver.