FI84682C - FOERFARANDE OCH ANORDNING FOER PREDIKTERANDE KODNING. - Google Patents

Description

1 84682

Method and apparatus for predictive coding

The present invention relates to a predictive coding method comprising generating a pleated difference signal proportional to the difference between the digital input signal and the digital prediction signal, quantizing the difference signal, summing the prediction signal and the quantized difference signal from the input signal and reconstructing the input signal .

Predictive coding, or DPCM (Differential Pulse code modulation), utilizes the correlation between the signal values of the input signal by encoding and shifting the difference between the signal values.

Figure 1 shows a schematic block diagram of a typical DPCM encoder. Block 1 forms the input signal x (n) and the so-called the difference signal e (n) between the prediction signal p (n), which is quantized in the quantization block 3. The quantized difference signal e '(n) is summed in block 4 to the prediction signal p (n), whereby a so-called a reconstructed signal x '(n) representing the sum of the input signal x (n) and the quantization error. Based on the reconstructed signal x '(n), a prediction signal p (n) is generated in the prediction block 5, which is input to the block 1.

The signals x (n) and p (n) are generally digital signals in parallel, comprising, for example, eight bits, with a reading range of 0 to 255. In standard DPCM coding, the difference signal e (n) can have both positive and negative values, and one bit more is used to represent it unambiguously than to represent the signal x (n) and p (n). For example, in a 9-bit 2-complement representation, the entire value range of the difference signal e (n) is -256 ... + 255. The value P of the prediction signal p (n) determines the possible reading range XMIN-P of the difference signal e (n) at each moment. . .x ^ -p, i.e. in the example case -P ... 255-P, when the maximum and minimum values of the input signal are X ^ and XMIH. In this case, only a part of the values of the difference signal e (n) and the corresponding quantization levels are used at any given time, as illustrated in Fig. 2A.

Folding quantization takes advantage of the above-mentioned property of the difference signal by describing (folding) the negative values of the difference signal e (n) to a positive reading range before quantization, as illustrated in Fig. 2B. Folding quantization is described, for example, in "A Simple High Quality DPCM codec for Video Telephone Using 8 Mbit per second", Nachtrictentechenische Zeitung, 1974, Häft 3, pages 115-117. In Fig. 2B, the negative region of the difference signal, -1 ...- P, is mapped to 15 positive regions above the value 255-P so that the values X and 256-X correspond to the same quantized value. Thus, below the value 255-P there is the largest possible positive and above the maximum possible negative signal value. In practice, folding can be performed by omitting the sign bit of the difference signal e (n) when the difference signal is in the 2-complement form. Since one bit less is now used to represent the difference signal, the reading range of the quantization block is half that of normal predictive coding. Thus, the quantization levels are about two to 25 times denser compared to normal predictive coding if the same number of levels are available to represent the quantized difference signal.

The quantization levels Q are typically less than 64. Then, in each quantization interval Qn 30 .., Qn + i of the quantization block, several values of the difference signal e (n) are located. With small and large values of the input signal, in the quantization interval to which the difference signal value 255-P belongs, the value of the difference signal can be described as incorrect, which is hereinafter referred to as erroneous quantization. In Fig. 2B, in the quantization interval Qn ... Qntl above the value 255-P, the negative difference signal value folded into the positive region is depicted in quantization as a relatively large positive value of the quantized difference signal e '(n), as illustrated in Fig. 2C. For example, this incorrect quantum-5 tone may appear as a change of white pixel to black or vice versa in video image transfer.

An attempt has been made to prevent such an error in folding quantization by limiting the signal from above and below to the maximum possible quantization error. In this case, the difference signal cannot obtain values from the quantization range Qn ... Qntl, which includes both positive and folded negative values. However, such signal limitation significantly reduces the dynamics of the input signal, especially in the case of coarse (low-quantization) quantizers.

It is an object of the present invention to provide a method and apparatus for pleated quantization that prevents erroneous quantization without limiting the dynamics of the input signal.

According to the invention, this is achieved by identifying erroneous quantizations by means of overflows occurring at different times in the difference signal generation event and the reconstructed signal generation event, and correcting them depending on the value of the prediction signal.

In the method according to the invention, an erroneous quantization event can be easily detected by observing the events of generating the difference signal and the reconstructed signal. In normal operation, when a negative value is pleated in the positive range when the difference signal is generated, i.e., an overflow has occurred, an overflow is also automatically generated when the reconstructed signal is generated, which cancels out the folding effect distorting the quantization result and results in a correct reconstructed value. If only one of these 4 84682 formation events overflows, it indicates an inaccuracy in the quantization.

When an erroneous quantization event is detected, the value of the prediction signal can be used to determine whether the erroneous quantization description occurred from a positive to a negative value or vice versa. In this way, the prediction signal can be used to correct the quantization error by selecting the nearest correct value of the quantized difference signal or the reconstructed signal. Most preferably, this is accomplished by storing in memory for each combination of the value of the difference signal and the value of the prediction signal a correction value that, in the event of an error, replaces the value distorted by erroneous quantization.

In a preferred embodiment of the invention, the value of the reconstructed signal, which has been distorted by erroneous quantization, is replaced by a correction value stored in the memory. In this case, a smaller and faster memory can be used than by correcting the quantized difference signal itself. Fast memory is essential for speed-intensive applications, such as 34 Mbit / s video codecs. In any case, since the reconstructed signal must be generated before an erroneous quantization event can be identified, correcting the reconstructed signal is also advantageous in terms of speed requirements. In this case, the signal processing 25 takes place as a continuously advancing process, in which there is no need to go back to correct the signals already generated.

The invention also relates to a device according to claim 6.

The invention will now be described in more detail by means of directional examples with reference to the accompanying drawings, in which Fig. 1 shows a block diagram of a prior art predictive coding apparatus, Fig. 2A shows a difference signal reading range a difference signal, and Fig. 2C illustrates the generation of a quantization error when quantizing the difference signal of Fig. 2B, and Fig. 3 shows a block diagram of an apparatus according to the invention for pleated quantization.

In Fig. 3, the subtractor circuit 1 subtracts the prediction signal p (n) from the input signal x (n) to form a folded difference signal e (n). The signals x (n) and p (n) are preferably 8-bit digital parallel signals. The subtractor circuit 1 may be a conventional circuit which results in a signal in the 2-complement format comprising eight bits and a sign bit. Said eight bits form a pleated difference signal e (n) and the sign bit forms an overflow signal 0F1. The difference signal e (n) is passed to the quantization circuit 3, where it is quantized. The quantized difference signal e '(n) generated by the quantization circuit 3 is summed in the adder circuit 4 to the prediction signal p (n). The adder circuit 4 may be a conventional circuit which generates a signal in 2-complement format comprising eight bits and a sign bit. Said eight bits form the first reconstructed signal x'N (n). The sign bit forms the overflow signal 0F2. The first reconstructed signal x'K (n) is applied to one input of the selector circuit 6. A quantization error correction signal x'or (n) is applied to the second input of the selector circuit 6. The selector circuit 6 connects it to one of the two inputs to its output 30 according to the control signal OF applied to its control input. The output signal x '(n) of the selector circuit 6 forms a final reconstructed signal, on the basis of which a prediction signal p (n) is generated in the prediction block, which is input to the subtractor circuit 1.

The implementation of the prediction block 5 may vary very much depending on the application, but typically it comprises at least a calculation algorithm which calculates the value P of the new prediction signal p (n) from the reconstructed signal x '(n) either without delay or with a predetermined delay. A typical application is video image transmission, whereby the value of the prediction signal p (n) can be calculated, for example, by means of previous pixels on the same line and a delay of the line period by adjacent pixels on the previous line or by a delay of 10 frames in the previous image field. using the corresponding pixel.

The overflow signal 0F1 generated by the subtractor circuit 1 and the overflow signal 0F2 generated by the adder circuit 4 are input to an exclusive OR circuit 8, the output signal 15 of which forms a control signal OF. If there is no overflow in either of the circuits 1 and 4, i.e. OF1 = 0 and 0F2 = 0, then also OF = 0, in which case the selector circuit 6 connects the first reconstructed signal x'N (n) to its output. The signal x'N (n) is connected to the output of the selector circuit 6 also when an overflow occurs in each of the circuits 1 and 4, i.e. 0F1-1, 0F2 = 1 and OF = 0. Finally, if only one of the circuits 1 and 4 overflows, i.e. 0F1 = 1 and 0F2 = 0 or OF1 = 0 and 0F2 * 1, then the control signal OF acquires a state of 1, whereby the selector circuit connects the overflow correction signal x'ol, (n) to its output. 25 to output the value of the reconstructed signal x '(n) in the event of an error.

The apparatus of Fig. 3 further comprises an erroneous quantization correction register 7, which generates a quantization correction signal x'0F (n). An en-30 ink signal p (n) is applied to register 7. The quantization correction register 7 is preferably a memory circuit in which a corrected value of the reconstructed signal is defined for each value of the prediction signal p (n). This value at any given time generates a quantization correction signal x’or (n) which is continuously selected at the second input of the selector circuit 6, regardless of whether or not incorrect quantization has taken place. When erroneous quantization occurs and the state of the signal OF is 1, the selector circuit 6 can quickly correct the value of the reconstructed signal x '(n) by connecting a signal 5 x'0F (n) to its output.

In a preferred embodiment of the invention, the quantizer circuit 3 includes an intermediate coding circuit 3a for subsequent VLC (variable length coding) coding, which generates an intermediate code IC (Intermediate code) based on the quantization result 10, which is typically the sequence number of the quantization level selected as the quantization result. There are typically less than 64 quantization levels, as well as the values of the difference signal e '(n), so that the intermediate code IC can be represented by six Bitil-15s. This allows for smaller and faster VLC encoding memory.

In the embodiment of Fig. 3, both overflow signals 0F1 and 0F2 are applied to the quantization circuit 3. The intermediate coding circuit 3a includes two memories: a normal intermediate coding memory and an erroneous quantization correction memory. The normal intermediate encoding memory contains the sequence number of each quantization level. When the overflow signals 0F1 and 0F2 are the same, the intermediate code IC is generated by this memory. The quantization correction memory, in turn, comprises a corrected intermediate code value for each combination of the 25 quantization results and the different overflow signals 0F1 and 0F2, which in the event of an error corresponds to the nearest correct quantization level which is not erroneous. The intermediate code IC is generated by this memory when incorrect quantization occurs and the states of sig-30 signals 0F1 and 0F2 are different.

By means of the preferred embodiment of the above invention, the identification and correction of an erroneous quantization event takes place as quickly as possible, which makes it suitable for high-speed applications such as the transmission of a good quality video signal.

8 84682

However, especially in slower applications, the quantized difference signal e ’(n) can be corrected as an alternative solution. In this case, instead of register 7, a correction memory is needed, in which the nearest correct value of the quantized difference signal stored for each prediction signal p (n) and 5 difference signal e (n) or e '(n) is stored, which does not cause an error and which replaces the quantized difference signal value in an error situation. Such a memory may be included in the quantization circuit 3 or it may be a separate circuit of the register 7 type.

In practical applications, there may be several parallel quantizers 3 according to Fig. 3, each of which has its own correction table in the register 7 and in the intermediate coding circuit 3a. In this case, information 15 is also brought to the register 7 as to which quantizer the signal is coded at at any given time.

The figures and the related description are intended to illustrate the present invention only. The details of the method and apparatus of the present invention may vary within the scope of the appended claims.

Claims (8)

9,84682 A method for predictive coding, comprising generating a folded difference signal proportional to the difference between the digital input signal and the digital prediction signal, quantizing the difference signal, summing the prediction signal and the quantized difference signal, generating a known signal, generating a signal representative of the input signal erroneous quantization events are identified by the overflows occurring at different times in the difference signal generation event and the reconstructed signal generation event and are corrected depending on the value of the prediction signal. A method according to claim 1, characterized in that the erroneous quantization is corrected by replacing the erroneous value of the quantized difference signal with the nearest error-free value selected by the prediction signal. A method according to claim 1, characterized in that the erroneous quantization is corrected by replacing the error-distorted value of the reconstructed signal with the nearest correct value selected by means of the prediction signal. Method according to claim 2 or 3, characterized in that an intermediate code is generated for subsequent coding on the basis of the quantization result, which is preferably the sequence number of the selected quantization level, and that the intermediate code generated in the erroneous quantization event is changed to another intermediate code by means of a prediction signal or said overflow signals. corresponding to the closest correct value of the quantized difference signal. Method according to one of the preceding claims, characterized in that the folded difference signal is formed by removing the sign bit of the difference signal in the 2-complement form. An apparatus for predictive coding, the apparatus comprising a subtracting means (1) for generating a pleated difference signal (e (n)) proportional to the difference between the digital input signal (x (n)) and the digital prediction signal (p (n)), quantizing means (3) ) for quantizing the difference signal, summing the prediction signal (p (n)) of the adder means (4) and the quantized difference signal (e '(n)) to form a reconstructed signal (x' (n)) representing the input signal, the prediction signal of the prediction means (5) (p (n)) depending on the reconstructed signal (x '(n)), characterized in that the subtractor means (1) and the adder means (4) comprise overflow signal outputs (OF1, OF2) indicating overflow, and that the device further comprises quantization correction means (6, 7, 8) responsive to the different occurrence of said overflow signals (0F1, OF2) for detecting and correcting erroneous quantization; si depending on the value of the prediction signal (p (n)). Apparatus according to claim 6, characterized in that the means for correcting the erroneous quantization comprise a memory means (7) for indicating the prediction signal, in which a correction value of the reconstructed signal is stored for each value of the prediction signal, and a selector means responsive to said overflow signals (0F1, 0F2). (6, 8) to replace the value of the reconstructed signal (x '(n)) distorted by erroneous quantization il 84682 with said correction value when said overflow signals (0F1, 0F2) occur at different times. Device according to claim 6 or 7, characterized in that the quantization means (3) comprises coding means (3a) which, depending on the quantization result, generates an intermediate code (IC), which is preferably a sequence number of the selected quantization level, and said coding means (3a) a quantization correction memory 10 in which a corrected intermediate code value is stored for each of the values of the difference signal and the overflow signals, which in the event of an erroneous quantization event replaces the original intermediate code. 12 84682