JPS6128223A - Forecast coding device - Google Patents

Abstract

PURPOSE:To prevent error in code inversion produced by overflow or underflow by using an overflow detection signal to replace a forecast error respectively into a prescribed value within a level range possible for positive or negative transmission if the forecast error causes a positive or a negative overflow error. CONSTITUTION:A full adder 8 obtains a difference between an input A and a forecast error B and a full adder 9 inputs the input A and a code of the forecast value B. If the result of operation overflows in case of 4-bit (>=8), when it is regarded as 8-bit operation, the output is G00001*** (* can be either ''1'' or ''0''), and if the result underflows (<=-9), the result is 11110***. When neither underflow nor overflow takes place (7--8), the result is 00000*** when positive and 11111*** when negative. Thus, the overflow and underflow are detected by observing the MSB (most significant bit) of the output of the full adder 8 and the LSB (least significant bit) of the output of the full adder 9.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to a predictive coding device used for compressing a signal band in communications, etc.

The configuration of the conventional example and its problems The predictive coding method takes advantage of the correlation of the input signal.
This method compresses the signal transmission band by predicting the input signal and transmitting only the difference component between the input signal and the predicted value, and is used in the field of communication records.

FIG. 1 shows the configuration of a conventional predictive encoding device.

In Figure 1, 1 is a subtraction circuit that subtracts predicted value B from input A, 2 is a quantization circuit that quantizes the prediction error between input A and predicted value B, and output C of quantization circuit 2 is a prediction code decoding circuit. Served in a container. 3 is an inverse quantization circuit that has a quantization characteristic opposite to that of the quantization circuit 2 and dequantizes the output C of the quantization circuit 2; 4 is an inverse quantization circuit that combines the output of the inverse quantization circuit 3 and the subtraction circuit 1;
An adder circuit calculates a predicted value corresponding to the next input by adding the input predicted value B, and 5 is a holding circuit that holds the output of the adder circuit 4 until the next input signal comes in.

The problem with the configuration of the conventional example is that for inputs containing many high-frequency components with large amplitudes and rapid changes, the prediction error is larger than the maximum value that can be expressed with the original number of bits (overflow) or smaller than the minimum value ( Underflow), the sign may be reversed and an error may occur.

Now, let's focus on subtraction circuit 1. Input A and predicted value B are both 16 bits, and the output is also 16 bits.

Table 1 shows an example of computer simulation output in the case of low frequency and large amplitude, and Table 2 shows a computer simulation output in the case of high frequency and large amplitude.

Margin below Table 1 Table 2 68.

This is a sine wave with an amplitude of half the full scale (32767 level). In this embodiment, the period of the low frequency is set to % of the sampling frequency, and the period of the high frequency is set to %.

Due to the circuit configuration, the predicted value always follows the input one clock ago. Therefore, the difference between the input value and the predicted value is always approximately the same as the difference between the current input value and the input value one clock ago. Therefore, in the case of the low frequency shown in Table 1, the difference between the input value and the input value one clock ago is at most 280 in the case of this embodiment.
00 level, the minimum prediction error is 128368
become the level. This value can be expressed in 16 bits including the sign, so there is no problem.

However, in the case of high frequency and large amplitude in Table 2, the maximum difference between adjacent sample values is about 57,000, and the minimum value of the prediction error in the tree table is -56,735. This value is 17 bits including the sign.

In other words, the prediction error can be expressed in 16 bits (32767
~-32768), it has underflowed. If this 17-bit value of -56375 is transmitted only by the lower 16 bits, that is, by the output bit count of 16 bits, the value becomes -8801, and when the sign is inverted and reproduced by a new decoder, it becomes the original value. The value is completely different. Therefore, if the input is audio data, this sign inversion causes a sudden change in the playback output during playback, producing noise that is not present in the original musical tone, making it extremely difficult to listen to. Generally speaking, g-sound signals have few signals with high-frequency components with large amplitudes, but they are not completely absent.

Conversely, if the input level is lowered to prevent such a phenomenon, the S/N ratio of the system will deteriorate.

OBJECTS OF THE INVENTION The present invention solves the above-mentioned drawbacks, and an object of the present invention is to provide a predictive coding device that can prevent sudden changes in output due to sign inversion caused by overflow or underflow of prediction errors. That is.

Structure of the Invention The present invention includes a first means for determining a prediction error that is the difference between an input signal and a predicted signal, and a first means for detecting whether the prediction error overflows the transmittable level range in a positive or negative manner. 2 and the overflow detection signal obtained by the second means, when the prediction error has caused a positive or negative overflow, a range of levels at which the prediction error can be transmitted as positive or negative, respectively. It is configured to have a third means for replacing it with a constant value within,
This makes it possible to prevent sign inversion errors caused by overflow or underflow that occur when a large-amplitude, high-frequency signal is input.

DESCRIPTION OF THE EMBODIMENTS FIG. 2 shows a block diagram of a predictive coding apparatus according to an embodiment of the present invention. In FIG. 2, the same members as in FIG. 1 are given the same numbers. Reference numeral 6 denotes a detection circuit for detecting overflow or underflow when the prediction error overflows or underflows.Here, the operation of the detection circuit 6 and switching circuit 7 for detecting overflow or underflow will be mainly explained. FIG. 3 is a circuit diagram showing a specific circuit configuration of the detection circuit 6. In this embodiment, for convenience of explanation, a case is shown in which the input and predicted values are 4 bits expressed in two's complement. The 4-bit full adder 1o for both 8 and 9 is a detection section,
By adopting the configuration shown in FIG. 3, input A and predicted value B are subtracted.

Of these, full adder 8 calculates the difference between input A and predicted value B, and full adder 9 receives the signs of input A and predicted value B as its inputs. In other words, this corresponds to the fact that the subtraction between the input and predicted values is treated as an 8-bit operation instead of a 4-bit operation.

In the case of 4 bits, if the operation result overflows (8 or more), if it is regarded as an 8-bit operation, the output will be
1＊＊＊ (* can be 0 or 1. The same applies below)
, at one time underflow (-9 or less) 11110**)
c. If there is no underflow or overflow (7 to -8), 00000*** if positive, 11111***108 if negative.

becomes.

Therefore, by looking at the MSB (most significant bit) of the output of the full adder 8 and the LSB (least significant bit) of the output of the full adder 9, the overflow/underflow signal H is 0, and when there is an underflow, the overflow signal F is detected. is 0. The underflow signal H is 1, and in other cases both signals are O. Further, the error signal E becomes 1 when there is an overflow or an underflow.

Next, a specific circuit configuration diagram of the switching circuit 7 is shown in FIG. Reference numeral 11 denotes a selector, which changes the prediction error to a constant positive or negative value according to the error signal E. Reference numeral 12 denotes a circuit that provides a constant positive or negative value as the output I in response to an underflow signal H when there is an overflow or an underflow. When there is an overflow, the maximum positive value (7 for 4 bits and 7 bits) is output instead of the prediction error, and when there is an underflow, the minimum negative value (-8 for 4 bits) is output.

In this way, when an overflow or underflow occurs, the predicted value will not follow for several clocks because it is replaced with a constant value, but no code error will occur, so no unpleasant noise will occur in the case of audio.

In this example, 4 bits and 2's complement are used for explanation purposes, but
It goes without saying that the same method can be used even if the number of bits is different or the code is different.

Effects of the Invention As described above, according to the present invention, it is possible to prevent code inversion errors caused by overflow or underflow that occur when a large-amplitude, high-frequency signal is input to a predictive encoding device.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a conventional predictive coding device, Fig. 2 is a block diagram of a predictive coding device according to an embodiment of the present invention, Fig. 3 is a block diagram of the same detection circuit, and Fig. 4 is a quantization diagram. Circuit, 4...
...Addition circuit, 5...Holding circuit, 6...
Detection circuit, 7...Switching circuit.

Claims (1)

[Claims] a first means for determining a prediction error that is a difference between an input signal and a predicted signal; a second means for detecting whether the prediction error overflows a transmittable level range in a positive or negative manner; If the overflow detection signal obtained by means 2 indicates that the prediction error has caused a positive or negative overflow, the prediction error is replaced with a constant value within a level range that allows positive or negative transmission, respectively. 3. A predictive encoding device characterized by having the means of 3.