JPS6037658B2 - Time series waveform encoding device - Google Patents

Description

[Detailed description of the invention] The present invention relates to a time-series waveform encoding device.

As a time-series waveform encoding method, an output code sequence is obtained by quantizing the difference value between the sample value of the time-series waveform and the predicted value for the sample value according to the quantization step size controlled by the immediately previous output code. Adaptive differential quantization method (
ADPCM method) is known from the following documents.

“Ship A ptive QuantiZatjon Mth
a OneWordMemory,”BSTJ,V
ol. 52, No. 7, Sept. , 1973, ppl
ll9-1144, By N. S. Jayaut. The ADPCM encoding method selects a quantization code with a signal-to-noise ratio of at most 4 at each time when a time-series waveform is sampled. It is not necessarily the code sequence that minimizes the overall S/N among the possible ADPCM code sequences.

FIG. 1 shows an example of a case where the code sequence outputted by the encoding method is not the optimum code sequence that minimizes the overall S/N. In Figure 1, sample value S(t,) of time series waveform 1 given as input at time et al.
The amplitude value f(t
, ) are assumed to match. The reproduced waveform 2 generated by the encoding method has a large S/N at a time ratio of 2, but the dynamic range becomes too small at time t3 when the quantization step size is controlled by the output code at a time ratio of 2. This is an example in which the S/N becomes small due to a load. However, in the above case, curve 3
As shown in , the S/N is relatively low at 4.0% when the time is around, but the S/N is relatively high at the time when no overload occurs and the S/N increases as a whole. There will be an output code sequence in which N is larger than the encoding method. The ADPCM code sequence that minimizes the overall S/N has an S/N of all possible combinations of ADPCM codes.
It is clear that an optimal code sequence can be found by calculating , but it has the drawback that the amount of calculation becomes enormous when the number of samples is large. The purpose of the present invention is to prevent S/N deterioration caused by an inappropriate output code sequence in an ADPCM encoding method for time-series waveforms, prevent S/N deterioration of reproduced waveforms, and improve S/N of reproduced waveforms. An object of the present invention is to provide a time-series waveform encoding device that efficiently calculates an output code sequence.

The feature of the present invention is to efficiently calculate the ADPCM code sequence of time series waveform 1 that improves the S/N in a certain time interval using a dynamic programming method, thereby improving the S/N characteristics of the reproduced waveform. It has the effect of improving.

The time-series waveform encoding device of the present invention includes a waveform temporary memory circuit that stores a predetermined number of sample values of time-series waveforms, and a reproduced value for the sample value output from the waveform temporary memory circuit at the sample time. Means for calculating all possible combinations of code sequences at the immediately preceding sample time, and means for calculating the error between each of the reproduced values and the sample value, and calculating the difference between the sample value and the reproduced value at the past sample time that have already been calculated. means for calculating a new integrated value by adding it to the integrated value of errors, and comparing the integrated values of errors for code sequences having the same code value at the sampling time among the new integrated values, and calculating the minimum value for the code. The evaluation value storage section that stores the evaluation value of the sequence and the evaluation value for each code sequence outputted from the evaluation value storage section for each predetermined number of sample values are compared to detect the minimum evaluation value, and the minimum evaluation value is detected. and means for calculating a code sequence at each sample time associated with an evaluation value as an output code sequence. The present invention divides a sampled time series waveform into blocks for each fixed number (one) of sample values, and uses a dynamic programming method to efficiently generate an ADPCM code string that improves the S/N characteristic within each block. Calculated as follows.

The optimal ADPCM code sequence in the block can be found by solving the following saponification equations '1}, {2', {3', ■, where S is the time series waveform and J is the quantized Bell number. i=1,2......,l i=1,2......,J Here, G(i,j) is the i-th quantization at the i-th time point. This is an evaluation value for evaluating the signal and is given as an integrated value of quantization error power.

Here, Ei, j, and k are error powers and are expressed by the following equation (2).

Ei,j,k=(Si-Si)2...
...■Si is the regeneration value of Si and is expressed by the following formula [3'.

Si NibSi-Mizujugi-, kqi...
...[3} Here, b represents a prediction coefficient given in advance, and bSi-aki represents a predicted value for Si. Further, qi represents the i-th quantized Bell value given in advance. Furthermore, gi-,,k represent the adaptive multiplication of the Toyoko conversion step and are given by the following equation. g
i-,, k=akg-2, evening...
...(41 Here, ak is the k-th adaptation coefficient given in advance.

Further, the symbol represents the number of quantized bells selected for G(i-1,k) according to formula '11. G(1, i) (i=1,
2, . . . , J) and select the first value G(1, i') expressed by the formula {5'.

The recurrence formula {11
The optimal ADPCM code sequence is given by the (i, j) sequence that satisfies . - By performing the above operations for each block, an optimal ADPCM code sequence can be obtained in sequence.

According to the present invention, whereas the method of comparing all possible combinations of ADPCM codes in a round-robin manner requires comparison operations on the order of JI, S
An ADPCM code sequence with good /N characteristics can be obtained. Next, the present invention will be explained in detail with reference to the drawings.

FIG. 2 is a block diagram showing one embodiment of the present invention. In the following explanation, the prediction coefficient ([3}
b) in the formula is 1.

First, sample values of a time-series waveform are sequentially stored in the waveform data time storage circuit 3 via the waveform input terminal 1.

The sample values of the time-series waveforms stored in the waveform data temporary storage circuit 3 are stored every predetermined number of samples (number of samples in a block) according to control data given from the control circuit 2 via the waveform input/output control data transmission line 6. The waveform data is stored in the storage circuit 4. The waveform data is input one sample at a time from the storage circuit 4 to the error power calculation circuit 11 in the evaluation value calculation section 19 . Each circuit in the evaluation value calculation unit 19 is controlled by being given a control pulse from the control circuit 2 via the evaluation value calculation unit control data transmission line 5. Quantized bell value storage circuit 1
3, the quantized value output according to the control pulse and the adaptive multiplier output from the adaptive multiplication temporary storage circuit 14 according to the control pulse are balanced by the multiplication circuit 9, and then the predicted value is calculated. The predicted value outputted from the temporary storage circuit 15 in accordance with the control pulse is added from the addition circuit 1 and inputted to the error power calculation circuit 11 as a reproduced value at the sample time, and simultaneously stored in the predicted value temporary storage circuit 16. . In the error power calculation circuit 11, the square of the difference between the sample value of the audio waveform output from the waveform data storage circuit 4 and the reproduced value output from the addition circuit 10 is calculated as an error power value, and is output to the addition circuit 17. Ru. The evaluation value output from the evaluation value temporary storage circuit 12 in accordance with the control pulse is input to the addition circuit 17, and the error power calculation circuit 1
1 and is added to the error power value output from 1 and stored in the evaluation value temporary storage circuit 18 as a new evaluation value. The above operation is repeated several times for each quantized bell according to the control pulse output from the control circuit 2, and then the evaluation value for each quantized bell is output from the evaluation value temporary storage circuit 18 to the comparison circuit 22. .

The comparison circuit 22 compares the evaluation values according to the control data given from the control circuit 2 via the comparison circuit control data transmission line 7, stores the highest evaluation value of 4 in the evaluation value temporary storage circuit 12, and stores the minimum evaluation value. The optimally quantized bell data is optimally quantized and the bell data temporary storage circuit 24;
It is output to the evaluation value temporary storage circuit 18 , the predicted value temporary storage circuit 16 , and the adaptive multiplier temporary storage circuit 21 . The evaluation value temporary storage circuit 18 outputs the evaluation value according to the optimum quantized Verdei output given by the comparison circuit 22 and stores it in the evaluation value temporary storage circuit 12 . Predicted value temporary prediction circuit 1
2 outputs the predicted value according to the optimum quantized bell data given by the comparison circuit 22 and stores it in the predicted value temporary storage circuit 15. The adaptive multiplier temporary storage circuit 21 outputs the adaptive multiplier to the multiplier 20' in accordance with the optimum quantized signal data provided by the comparison circuit 22. On the other hand, the adaptation coefficient is output from the adaptation coefficient storage circuit 21 to the multiplier 2 in accordance with control data given from the control circuit 2 via the adaptation coefficient storage circuit control data transmission line 30. The adaptation multiplier input to the multiplier 2 and the adaptation coefficient are mixed and then stored in the adaptation multiplier temporary storage circuit 1.
4 is stored. The above operation is quantized for each sample value of the time-series waveform according to the control data outputted by the control circuit 2, and is repeated for the number of bells (J in equation 1).

Furthermore, the above operations are performed using the number of samples in the block (1 in formula m).
) times. After the above operations, comparison circuit 2
3 is an evaluation value temporary storage circuit 1 according to control data given from the control circuit 2 via the comparison circuit control data transmission line 8.
The evaluation values outputted from 2 are compared, and the minimum evaluation value is quantized and the bell data is output to the optimum code extraction circuit 25.

The optimal code extraction circuit 26 optimally quantizes the Bell data according to the quantized Bell data, sequentially extracts it as an output code sequence from the Bell data temporary storage circuit 24, and temporarily stores it in the optimal code extraction circuit 25.
It is output via the sign output terminal 26. Sign output terminal 26
The code sequence outputted via the code input terminal 27 is input to the decoding circuit 28 via the code input terminal 27.

A time-series waveform is synthesized from the code sequence in the decoding circuit 28 and outputted via the reproduced waveform output terminal 29. Note that the decoding circuit 28 used in the above-mentioned embodiment is as described in the first part of the above paper.
Since it can be realized using the configuration shown in the figure, a detailed description of the configuration is omitted.

In the above explanation, the quantization step size when quantizing the difference value between the sample value and the predicted value for the sample value is controlled by the output code at the past sample time, but the quantization step size This is also effective for the so-called DPCM encoding method in which the DPCM is fixed.

In the case of the DPCM encoding method, {4} in the above explanation
It is clear that ak in the formula becomes ak31 for all k, and therefore gi-,, k in the '3} formula is a special case where it becomes a constant. Further, in the above description, the integrated value of error power was used as the evaluation value, but it is clear that it can also be realized by using the integrated value of absolute error values or an appropriate error function value given in advance as the evaluation value.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining the present invention, where t on the axis represents time, the vertical axis represents amplitude, and 1 represents the time series waveform S(tj),
2 is the mid-range value f(ti) reproduced by the DPCM method,
3 is the midpoint value g(tj) reproduced by the optimal code sequence
shows. FIG. 2 is a block diagram showing an embodiment of the present invention, in which 1 is a waveform input terminal, 2 is a control circuit, 3 and 4 are waveform data temporary storage circuits, 5 is an evaluation value calculation control data transmission line, and 6 is a control circuit. Waveform input/output control data transmission line, 7 and 8 are comparison circuit control data transmission lines, 9 and 20 are multiplication circuits, 10 and 17 are addition circuits, 11 is an error power calculation circuit, 2 is an evaluation value temporary storage circuit, 13 is a Quantized bell value storage circuit; 14 is an adaptive multiplier temporary storage circuit; 15 and 16 are predicted value temporary storage circuits; 18 is an evaluation value temporary storage circuit; 19 is an evaluation value calculation unit; 21 is an adaptation coefficient storage circuit; 22 and 23 is a comparison circuit; 24 is an optimal quantization and bell data temporary storage circuit;
25 is an optimal code extraction circuit, 26 is a code output terminal, 2
7 is a code input terminal, 28 is a decoding circuit, 29 is a reproduced waveform output terminal, and 30 is an adaptation coefficient storage circuit control data transmission line. Buru′ Figure Buruz Figure

Claims (1)

[Claims] 1. In a time series waveform encoding device of the type that obtains an output code sequence by quantizing the difference value between the sample value of the time series waveform and the predicted value for the sample value, a predetermined number of sample values of the time series waveform are used. a waveform temporary memory circuit that stores
means for calculating reproduction values for the sample values output from the waveform temporary storage circuit for all possible combinations of code sequences at the sample time and the immediately preceding sample time; and means for calculating the error between each reproduction value and the sample value. means for calculating a new integrated value by calculating the respective magnitudes and adding them to the already calculated integrated value of the magnitude of the error between the sample value and the reproduced value at the past sample time; an evaluation value storage unit that compares accumulated values of errors for code sequences having the same sign value at the sample time and stores the minimum value as an evaluation value of the code sequence; It is characterized by comprising means for comparing the evaluation values for each code sequence output from the storage unit, detecting the minimum evaluation value, and calculating the code sequence at each sample time associated with the minimum evaluation value as an output code sequence. A time-series waveform encoding device.