JPS6042480B2 - Adaptive predictive differential coding method - Google Patents

Description

[Detailed Description of the Invention] Background of the Invention The present invention is based on differential encoding used for high-efficiency encoding of highly correlated signals such as audio signals, and in particular adaptive prediction coefficients of a predictor used in differential encoding. The present invention relates to an adaptive predictive differential encoding method that performs encoding using different methods.

For highly correlated signals such as audio signals, the sample value of the signal at a certain time can be predicted fairly well using past sample values, so the prediction error signal power is considerably smaller than the input signal power.

The differential encoding method uses such properties to realize highly efficient encoding by encoding only the prediction error signal with a relatively small number of bits. Prediction in this differential encoding is performed by a prediction filter. If the characteristics of the prediction filter are matched to the average spectrum of the input signal, for example, when encoding a voice signal, a relatively accurate predicted value can be obtained. However, as a way to obtain a more accurate predicted value,
A method of changing the characteristics of a prediction filter according to the signal, specifically a so-called adaptive predictive differential encoding method, in which the coefficients of a prediction filter are modified as appropriate to minimize the prediction error. be. By using this method, more accurate prediction can be made than when the coefficients of the prediction filter are fixed, and as a result, the prediction error signal can be further reduced, making it possible to perform encoding with higher efficiency. As mentioned above, if prediction in differential coding is adaptive, it is possible to achieve more efficient coding than when fixed prediction is used, but as a result of constantly changing the prediction filter coefficients, This creates a stability problem for the prediction filter that does not occur when shape prediction is used.

In other words, if the coefficients of the adaptive prediction filter are modified and the coefficients happen to be at a value that causes the prediction filter to oscillate, the oscillation of the prediction filter will result in a predicted value that is unrelated to the input signal being output. It becomes impossible to make accurate predictions. In other cases, the encoder itself, which includes a prediction filter in its feedback loop, may constitute a recursive filter and oscillate. In order to prevent such an unstable state from occurring, it is possible to modify the prediction coefficients while controlling them so that they do not become coefficients that would make the prediction filter or encoder unstable. In order to do this, it is necessary to know in advance all the prediction coefficients that will cause an indeterminate state.

The prediction coefficients that cause this unstable state can be found relatively easily when the order of the prediction filter is low, but in this case there are few, and in many cases the order is high, and as the order increases, the number of combinations increases and its discovery becomes difficult. This becomes extremely difficult and the coefficient correction algorithm also becomes complicated. Furthermore, if we take into account the effects of nonlinearity caused by the quantizer that quantizes the prediction error signal, it becomes impossible to find prediction coefficients that would cause the encoder itself to be in an unstable state as described above. You can say it's close. SUMMARY OF THE INVENTION In view of the above instability problem in adaptive predictive differential encoding, an object of the present invention is to provide an adaptive differential encoding method that is relatively simple and has an effective instability prevention function. There are things aimed at.

In this invention, when the adaptive predictive differential encoder is operating normally, the values of the input signal Focusing on the fact that the values of the input signal X and the locally decoded signal differ greatly when the stability is reached, the difference between the input signal X and the locally decoded signal X is determined, and the level of this difference signal is determined by When it is determined that the encoder has clearly become unstable, the encoder is returned to a stable state by resetting the prediction coefficient to a value that makes the encoder stable.

That is, according to the present invention, the difference between a digital input signal to be encoded and a predicted value for the input signal is quantized, the quantized signal is output, and the quantized difference signal is dequantized, and In an adaptive predictive differential encoding device that obtains a predicted value for the input signal based on adaptive prediction by an adaptive prediction means, a locally decoded signal obtained by adding the predicted value and the quantized reproduced difference signal; These error signals are obtained by subtracting the input signals from each other using a subtraction means, the error signals are input to a level detection means to detect the level of the error signal, and the error signal is outputted from the level detection means. Based on the level information of the signal, the determining means determines whether or not the operation of the adaptive predictor is normal, and only when it is determined that the operation is abnormal, the prediction coefficient of the adaptive predicting means, and if necessary, the level detecting means A control signal for resetting information is output from the determination means to return the operation of the adaptive prediction means to a normal state.

In yet another embodiment, the level of the input signal is set to a second level.
The detection level is detected by the level detection means, and this detected level is also input to the determination means, and based on each signal level information output from these level detection means, it is determined whether the operation of the adaptive prediction means is normal or not. Only when it is determined, the prediction coefficient of the adaptive prediction means and, if necessary, the level information of both the level detection means are reset.

Prior Art Before a detailed description of the present invention, a conventional adaptive predictive differential encoding device will be described with reference to the drawings.

FIG. 1 shows a conventional adaptive predictive differential encoding device. The digital input signal X at time j is input through the input terminal 100 and the adaptive filter 130 is input through the signal line 131.
The predicted value sentence outputted from the subtracter 110 is input to the subtracter 110. The prediction error signal E, which is output from the subtracter 110 via the signal line 111, is encoded by a quantizer 120 using one or more bits, and the encoding result U, is sent via the signal line 121 to the inverse quantizer 140. and the output terminal 160
Output via . The inverse quantizer 140 performs inverse quantization on the input encoding result U, and reproduces the reproduction error signal j
is output via the signal line 141. Since this reproduced error signal j contains quantization noise caused by quantization, it generally does not match e. The reproduced difference signal X,
is output via the signal line 151.

This local decoded signal ◇. is input to the adaptive filter 130, and after being subjected to adaptive filtering, the predicted value M is outputted via the signal line 131. The filter coefficients of the adaptive filter 130 are modified using the reproduced error signal j output from the inverse quantizer 14-0 and the local decoded signal Xj. There are two typical correction algorithms: A with this)
1 is the first filter coefficient of the adaptive filter 130 at time j.

g is a correction gain, and Sigrl (Xj) takes a value of +1 when X is positive and -1 when it is negative. Alternatively, the following equation may be used as a simplified correction algorithm. In this), Δ is a normalization coefficient close to the effective value of the prediction error signal. (
The algorithms of formulas 1), (2), and (3) all add the correction term in the second term on the right side to the filter coefficient A1 at time j to obtain a new filter coefficient at time j+1. In this configuration, the path from the reproduced error signal to the prediction value statement forms one recursive adaptive filter, so when the filter coefficients are modified as described above, the new filter coefficients happen to be If the shape adaptive filter becomes a coefficient that oscillates, or if the filter coefficient when the encoder is powered on becomes a value that causes oscillation, the predicted value X3 becomes an oscillating signal that is unrelated to the input signal. As a result, normal prediction cannot be performed, and high-quality encoding cannot be performed. FIG. 2 shows another example of a conventional adaptive predictive differential encoder.

The basic operation is the same as that explained with reference to FIG. 1, so a description thereof will be omitted, but the difference is that the input to the adaptive filter 230 is the reproduced error signal j. With such a configuration, the process from the reproduced error signal , to the predicted value ma is a non-recursive adaptive filter, and oscillations as explained in Fig. 1 do not occur, but the encoder itself is a recursive filter. Therefore, the encoder itself may become unstable depending on the value of the prediction coefficient. In FIG. 3, a portion 400 of the conventional adaptive predictive differential encoder further surrounded by a line has the same configuration as the adaptive predictive differential encoder shown in FIG.
Since the operation is almost the same, the explanation will be omitted.

Input signals G and locally decoded signal X are input to the subtracter 470 via the signal line 401 and the signal line 451, respectively.
The - difference is input to the determination circuit 490 via the signal line 481-. - Normally, when adaptive predictive differential encoding is performed normally, the input signal ◇ and the locally decoded signal X have very close values. However, as described above, as a result of modifying the coefficients of the adaptive filter 130, the adaptive filter 130
If the encoder becomes unstable due to oscillation, the input signal
, and the locally decoded signal component 4 will have a significantly different value. Therefore, by monitoring the difference signal level output through the signal line 471, it is possible to determine whether the encoder is stable or unstable.
It is possible to determine whether The determination circuit 490 performs this determination function. That is, the determination circuit 490 determines that the encoder has become unstable when the level of the difference signal input via the signal line 471 exceeds a certain threshold, and determines the filter coefficient of the adaptive filter 130 via the signal line 491. A control signal for resetting the encoder to a value that stabilizes the encoder and resetting the output of the level detector 480 to zero is output via the signal line 491, thereby returning the encoder to a stable state. In this embodiment, the encoder shown in FIG. 1 is used for the portion 400 surrounded by a broken line in the figure, but the encoder shown in FIG. 2 or 3, for example, can also be used.

That is, the input signal X in the conventional adaptive predictive differential encoder is input through the signal line 401, the locally decoded signal G4 is input through the signal line 451, and the determination circuit 4
The encoder can be easily restored from an unstable state to a stable state by simply supplying the reset signal outputted from the adaptive filter 90 via the signal line 491. When using the encoder shown in Fig. 2, it is necessary to create a local decoded signal as shown by the dotted line in Fig. 2.Second Embodiment Fig. 5 shows another embodiment of the present invention. .

A portion 400 surrounded by broken lines in the figure has the same configuration as the adaptive predictive differential encoder shown in FIG. 1, and its operation is also almost the same, so a description thereof will be omitted. The arithmetic unit 470 receives input signals X,
and local decoded signal gj are input, and their difference signal is input to level detector 480 via signal line 471. Fourth
Level detector 480 as described in the illustrated embodiment
It is possible to judge whether the encoder is stable or unstable based on the difference signal level output from the encoder, but when the dynamic range of the input signal As the frequency increases, the level of the difference signal obtained via the signal line 471 also increases, which causes the difference signal level to reach a value at which the encoder is determined to be unstable. If the value becomes too high, it becomes impossible to accurately determine whether the encoder is stable or unstable. This is why we decided to use not only the difference between the input signal X and the locally decoded signal, but also the signal level of the input signal X, as an element for determining whether the encoder is stable or unstable. This is a feature of the second embodiment.

That is, the level of input signal X, is detected by level detector 580 and inputted to determination circuit 490 through signal line 581, and even though the level of input signal It is determined that the encoder is unstable only when the difference signal level from j is high, and the determination circuit 490 resets the filter coefficients of the adaptive filter 130, and the level detector 580 and the level detector 480
A control signal that also resets the respective outputs of and is outputted via signal line 491. In this way, even if the encoder becomes unstable when the input signal level is large, the determination circuit 490 will not output the aforementioned reset signal and the encoder will remain unstable. For audio signals, it is approximately 0. Since a silent interval occurs every two minutes, the input signal level drops at that point and it is possible to judge whether the encoder is stable or unstable, so even if the input signal level is high, the encoder may become unstable. However, it can be restored to a stable state after a short period of time.

In addition, in this case, one threshold value is provided for the difference signal level between the input signal Xj and the local decoded signal father j, but this threshold value may be changed depending on the level of the input signal You can also do it. That is, by increasing this threshold value as the input signal level increases, it is possible to detect an unstable state of the encoder and quickly restore it to a stable state, regardless of the input level of the encoder. Further, in this embodiment, the encoder shown in FIG. 1 is used in the portion 400 surrounded by the broken line in FIG. 5, but the encoder shown in FIG. You can also do it. That is, an input signal
51 and output from the determination circuit 490 via the signal line 491, it is sufficient to provide the adaptive filter with a reset signal. As explained above, the present invention has a function that uses a simple method to detect prediction filter oscillations and unstable operation of the encoder itself that occur in conventional adaptive predictive differential encoding devices, and automatically restores the stable state. The present invention provides an adaptive predictive differential encoding method with
It is extremely useful.

Note that the various processes described above may be performed by an electronic computer.

[Brief explanation of the drawing]

FIGS. 1, 2, and 3 are block diagrams showing conventional adaptive predictive differential encoding devices, and FIGS. 4 and 5 are block diagrams showing embodiments of the present invention, respectively.

Claims (1)

[Claims] 1. Quantizing the difference between a digital input signal to be encoded and a predicted value for the input signal, outputting the quantized signal, and dequantizing the quantized difference signal; In the adaptive predictive differential encoding method for obtaining a predicted value for the input signal by adaptive prediction, the locally decoded signal obtained by adding the predicted value and the dequantized reproduced difference signal and the input signal are interchanged. Obtaining an error signal by reducing the error signal, and determining whether the adaptive prediction operation is normal by monitoring the level of the error signal,
1. An adaptive predictive differential encoding method, characterized in that the predictive coefficients used in the adaptive prediction are reset only when it is determined that the adaptive prediction is abnormal, so that the adaptive prediction operation becomes normal. 2. Quantizes the difference between a digital input signal to be encoded and a predicted value for the input signal, outputs the quantized signal, inversely quantizes the quantized difference signal, and adaptively predicts the input signal. In the adaptive predictive differential encoding method for obtaining a predicted value for the input signal, an error signal obtained by subtracting a locally decoded signal obtained by adding the predicted value and the dequantized reproduced difference signal from the input signal. By monitoring the input signal level and the level of the input signal, it is determined whether the adaptive prediction operation is normal or not, and only when it is determined that the operation is abnormal, the prediction coefficients used for the adaptive prediction are reset. An adaptive predictive differential encoding method characterized by making adaptive predictive operation stable.