JPH04286421A - Run length 1/n compression floating coding method and its decoding method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention]

[Industrial Application Field] The present invention relates to a code compression method used for high-quality recording and transmission of image data, and in particular a run-length 1/n compression floating method with good instantaneous S/N ratio characteristics over a wide input range. It relates to a coding method and its decoding method.

[Prior Art] Conventional satellite PCM broadcasting and digital audio tape recorders (hereinafter abbreviated as DAT) have a relatively light compression rate, taking into consideration the trade-off with the amount of hardware while suppressing quality deterioration due to compression. is adopted. Examples of these are line codes, DAT
In this case, a 12-bit 13-line code is used. Detailed technical standards are “EI” published by Japan Electronics Industries Association.
AJ CP-2305DAT Cassette
System Part1: Dimensio
and Characteristics”. FIG. 6(b) is a diagram illustrating a method of converting a 13-line code (12 bits) used in DAT, and shows a positive range. In the frame in the figure, numbers in parentheses are decimal numbers, and other numbers are binary numbers. The most significant bit (hereinafter referred to as MSB) of the binary number is the polarity bit P, which is "0" in the positive range and "1" in the negative range. "ABCDEFGH" in the figure is a value from "00000000" to "11111111", and "*" is an arbitrary binary number. Further, L0 is the run length of continuous data Q0 in which "0" continues in the upper part excluding the polarity bit of the original data. The procedure for the encoding method described above will be explained below. FIG. 6(b) is a diagram specifically showing conversion between encoding and decoding. First, a method of encoding a 13-line code (12 bits) will be explained. 16
The bit area is divided into seven areas within a positive range, and the shift amount is manipulated according to each area. As is clear from the figure, region division can be performed by starting from the position of "1" when viewed from the MSB side. In other words, it can be determined by the run length L0 of continuous data of "0" excluding the polarity bit. There are eight run lengths L0 from "0" to "7 or more", and the complement of L0 is expressed as an index as range data. For example, if the original data is "1216 (decimal)" or "000001001"
1000000 (binary)", the run length L0 of the continuous data Q0 is "4". The range data Q2 is “
It is "011" which is the inversion of "100". Also, “ABC
DEFGH” is “00110000”. Therefore,
The output data becomes "001100110000". FIG. 6(a) is a diagram schematically showing the above concept. The output data is a polarity bit P and range data Q representing the area.
2 and the shifted mantissa data D2 are combined into 12 bits. Range data Q2 has a fixed word length of 3 bits. If the original data is 2S (tooth) complementary, the negative range can be encoded in exactly the same way as the positive range by inverting all bits below P. A code whose bits are inverted in the negative range in this way is called a folded binary code. In the following explanation, unless otherwise specified, folded binary codes will be used.
Conversion with S complementary is mutually possible. Next, a method of decoding a 13-line code (12 bits) will be explained. Decoding uses polygonal code range data Q2
Since the position of the polarity bit P is known, the exponent part, that is, the range, is reproduced by looking at the three bits after the polarity bit P. Since the position of the mantissa data D2 is also known, the run length L0 of the linear code (16 bits) is restored using the range data Q2, and the inverted bit T0 and the mantissa data D2 are added after it. When the code length of the original data is less than W0, a fixed value is used. Decryption processing is performed according to such a procedure. When the encoding and decoding described above are performed, the precision of representation is changed due to rounding of information. Rounding of information occurs due to shift operations. Run length L0 is “6” and “7 or more”
Since they are the same, there are seven types of shifts. Therefore,
There are 14 types for positive and negative, but since the same shift operation is performed for positive and negative in the central area, the input/output characteristics of the entire area can be represented by 13 types of polygonal lines. In each region of FIG. 6(b), the representation accuracy, that is, the resolution changes depending on the run length L0. When the run length L0 is "0" to "6", the resolution is 10 bits to 16 bits in word length conversion. The contents explained above are summarized in (Table 1).

[Table 1] FIG. 7 is a characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a polygonal line code. In the figure, the horizontal axis shows the input level, and the vertical axis shows the instantaneous S/N ratio. However, for the sake of simplicity, the instantaneous S/N ratio is omitted for the sake of simplicity, which is about 2 dB, which is the S/N ratio improvement of the rectangular wave. In the figure, Z1 indicates the characteristics of a 13-line code (12 bits). Z1 will be explained. The input level of 0 dB is based on the maximum amplitude of a sine wave that can be represented by a code. Linear code “P1111111111
11111" corresponds to this. In the input level range of 0dB to -6dB, the linear code is "P1****..."
According to (Table 1), since the run length L0 is "0", the resolution is 10 bits. The quantization noise of 10-bit data is -60 dB, assuming -6 dB per bit. Therefore, the instantaneous S/N ratio for an input level of 0 dB to -6 dB is 60 dB to 54 dB. Next, in the input level range of -6 dB to -12 dB, the linear code becomes "P01***...", and as shown in Table 1, since the run length L0 is "1", the resolution is 11 bits. The quantization noise for 11-bit data is -66 dB. Therefore, the instantaneous S/N ratio is between 60 dB and 54 dB. In the same manner, the instantaneous S/N ratio was obtained in the region up to -96 dB, and the characteristic curve was obtained as Z1. Here, at an input level of -36 dB or less, the characteristic curve is the same as that of a linear code (16 bits).

[Problems to be Solved by the Invention] However, with the above conventional method, the instantaneous S/N ratio is only 60 dB at best, and at even smaller input levels, the instantaneous S/N ratio is as low as with linear codes (16 bits). descend. Professional digital recording and reproducing equipment, which requires a margin (headroom) against excessive input, records with a low average input level, which effectively narrows the dynamic range that can be used, and Since the instantaneous S/N ratio becomes insufficient over a considerable range, there is a problem in that editing in which amplification and attenuation are repeated increases deterioration. As mentioned above, this problem is the same even with linear codes (16 bits). Preferably, the hearing detection limit is about 48 dB.
Input level 0 to -96d
B is required to be satisfied over a wide range. As an improvement method for this purpose, a method of expanding the conventional 13-line code (12 bits) to 16 bits can be easily considered. (Table 2) ranges from linear code (20 bits) to polygonal code (1
6 bits). In Table 2, the word length of the mantissa data D2 becomes 12, so the overall resolution increases by 4 bits.

[Table 2] The characteristic diagram of the restoration code that decodes and expands the compressed code using the polyline code (16 bits) shown in (Table 2) is Z in FIG.
It will be like 2. This improved method improves the instantaneous S/N ratio by 24 dB over the entire range, but at an input level of -72 dB.
In the range from dB to -96dB, the instantaneous S/N ratio is 48d.
It is below B and is insufficient. Therefore, if this signal is amplified by an amplifier, there is a problem in that the quantization noise is audibly detected. The present invention solves the above-mentioned conventional problems, and has an instantaneous S/N ratio of approximately 48 dB, which is the auditory detection limit in a wide range of input levels from 0 to -96 dB.
The aim is to secure a grade of B or higher.

[Means for Solving the Problems] In order to achieve this object, the run length 1/n compression floating code method of the present invention uses a straight line code as the original data, that is, continuous data Q0 in which bits of a predetermined logic are consecutive in the upper part. A linear code consisting of an inverted bit T0 that breaks the continuity of the continuous data Q0, and lower data D0 after the inverted bit T0 is encoded to obtain compressed data, that is, a run length of the continuous data Q0. Compressed continuous data Q1 obtained by compression, an inversion bit T1 that breaks the continuity of the compressed continuous data Q1, compressed residual data C1 representing the remainder F1 generated when compressing the run length, and the lower order data D0. Mantissa data D obtained by rounding
The data is converted into compressed data consisting of 1 and 1. However, when the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1, and n is an integer of 2 or more, L1=int(L0/n) F1=L0 mod n.

[Operation] The present invention compresses the run length L0 of the original data to approximately 1/n by the method described above. The exponent part, that is, the range, is specified using compressed continuous data having a compressed run length, inverted bits, and compressed residual data. The exponent part, ie, the range, has the function of self-expanding according to the run length of the original data. That is, when the run length is small, the exponent part or range is specified using a small number of bits, and when the run length becomes large, the exponent part or range is subdivided, and the exponent part or range is specified using a large number of bits. Further, when the word length of the entire code is finite and set to a predetermined value, the number of bits allocated to the mantissa part is changed according to the run length. Through these effects, the space expressed by the compression code is expanded, and the expression precision can be maximized.

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1(a) is a conceptual diagram showing the structure of a run-length 1/2 compression floating code in the first embodiment of the present invention, and FIG. 1(b) is a diagram showing a specific example for explaining the conversion process. It is. In the figure, parts similar to those of the conventional example are given the same symbols. Hereinafter, the compression (encoding) procedure will be described first with reference to the figures. The first embodiment uses a straight line code (30 bits) with a run length of 1/2.
This is a coding method that compresses and expands into a compressed floating code (16 bits). Compressed continuous data Q1 is continuous data Q0
This is continuous data having a run length L1, which is obtained by dividing the run length L0 by 2 and converting it into an integer. That is, L1=int(L0/2). Further, if the remainder term of integer division is the compression remainder F1, then F1=L0 mod 2 . The inversion bit T1 is an inversion bit that breaks the run of the compressed continuous data Q1. The compression remainder data C1 is the complement representation of the compression remainder F1. Further, the mantissa data D1 is upper partial data of the data D0. The run length 1/2 compressed floating code is arranged in the following order: polarity bit P, compressed continuous data Q1, inversion bit T1, compressed remainder data C1, and mantissa data D1. Below, Figure 1
Based on (b), the run length L0 is “0” to “2”.
8" will be explained. When L0=0, L1 and F1 are as follows: L1=int(0/2)=0 F1=0 mod 2 =0. Since the run length L1 is "0", there is no compressed continuous data Q1. The compressed residual data C1 is "1". Data D0 has 30 bits, of which the upper 13 bits "ABCDEFGHIJKLM" are mantissa data D1. A run-length 1/2 compressed floating code is created by arranging the polarity bit P, inversion bit T1, compressed remainder data C1, and mantissa data D1 in this order.
EFGHIJKLM”. Similarly, find the following. When L0 = 1, L1 = int (1/2) = 0 F1 = 1 mod 2 = 1, so there is no compressed continuous data Q1, and compressed residual data C1
is "0". The upper 13 bits of data D0 “ABC
DEFGHIJKLM" is the mantissa data D1. The run length 1/2 compressed floating code is “P10AB
CDEFGHIJKLM”. When L0=2, L1=int(2/2)=1 F1=2 mod 2 =0 Therefore, the continuous data Q1 is "0" and the compressed residual data C1 is "1". The upper 12 bits of data D0 “A
"BCDEFGHIJKL" is the mantissa data D1. The run length 1/2 compressed floating code is “P011
ABCDEFGHIJKL”. When L0=3, since L1=int(3/2)=1 F1=3 mod 2 =1, the continuous data Q1 is "0" and the compressed residual data C1 is "0". The upper 12 bits of data D0 “A
"BCDEFGHIJKL" is the mantissa data D1. The run length 1/2 compressed floating code is “P010
ABCDEFGHIJKL”. Also, L0=20
When L1=int(20/2)=10 F1=20 mod 2 =0, the continuous data Q1 is "0000000000" and the compressed residual data C1 is "1". The upper three bits "ABC" of data D0 are mantissa data D1. Therefore, the run-length 1/2 compressed floating code is “P
000000000011ABC”. Furthermore, L
When 0=28, L1=int(28/2)=14 F1=28 mod 2 =0, so continuous data Q1 is “0000000000000
0", and the compressed residual data C1 is "1". There is no data D0 and no mantissa data D1. Therefore, the run length 1/2 compressed floating code is “P00000
0000000001”. In this way, a straight line code (30 bits) is compressed into a run length 1/2 compressed floating code (16 bits). Next, the decompression (decoding) procedure will be explained. In the decoding procedure, the polarity bit P is removed from the compressed data, and the run length L1 is obtained from the compressed continuous data Q1 and the inverted bit T1. Furthermore, compressed remainder data C1 (1 bit) and mantissa data D1 immediately after the inversion bit T1 are obtained. A compression remainder F1 is obtained by inverting the compression remainder data C1. From these, the run length L0 of the continuous data Q0 of the original data is obtained from L0=2*L1+F1. Continuous data Q0 is restored by concatenating "0"s having a length of run length L0. An inverted bit T0 is added after continuous data Q0, and mantissa data D1 is added after that. A polarity bit P is added to the beginning to form the original data, but if this code length is less than W0, a fixed value is assigned to the lower part of the mantissa data D1 to make the code length W0. Follow these steps:
Performs decryption processing. Below, some cases of the run length L1 will be specifically explained with reference to FIG. 1(b). When L1=0 and F1=0, since L0=2*0+0=0, there is no continuous data Q0. The mantissa data D1 has 13 bits and is "ABCDEFGHIJKLM". At this time, the original data has a polarity bit P, an inversion bit T0
and mantissa data D1, "P1ABCDEF
GHIJKLM＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊". In addition, "***..." is a fixed value "011..."
shall be allocated. Similarly, L1=10 and F1
When =0, L0=2*10+0=20, so the continuous data Q0 is "0000000000"
0000000000”, and the mantissa data D1 is “AB
C”. At this time, the original data is ``P000000000000000000'' by arranging the polarity bit P, continuous data Q0, inverted bit T0, and mantissa data D1.
001ABC＊＊＊＊＊”. In this way, a linear code (30 bits) is decoded and expanded from a run-length 1/2 compressed floating code (16 bits). In the same manner as above, the results summarized for all cases are shown in (Table 3).

Table 3 In Table 3, the linear code (30 bits) is a folded binary code, and the floating code is a folded run length 1/2 compressed floating code. Run length L0, run length L1 and resolution columns are 1
It is a 0-base number. The expression accuracy, that is, the resolution of the restored code obtained by decoding and expanding the compressed code is determined by the rounding accuracy of the linear code, and changes depending on the run length L0. As shown in Table 3, according to the first embodiment, a maximum accuracy of 30 bits to 15 bits can be obtained. Here, a description will be given of the instantaneous S/N ratio when a linear code is input, compressed and encoded using a run-length 1/2 compression floating code, and this code is decoded to output a decompressed restored code. However, for simplicity, the instantaneous S/N ratio omits the S/N ratio improvement (approximately 2 dB) of the rectangular wave. The input level is based on the maximum amplitude of the sine wave that can be represented by a sign (0 dB)
Make it. The straight line code "P1111..." corresponds to this. The range of the input level from 0 dB to -6 dB is represented by "P1ABC..." in linear code, and the resolution is 15 bits from (Table 3). The quantization noise of 15-bit data is -90 dB, assuming -6 dB per bit. Therefore, in the input level range of 0dB to -6dB, instantaneous S
/N ratio is 90 dB to 84 dB. Input level -
In the range of 6 dB to -18 dB, the linear code “P01
ABC..." or "P001ABC...", and from (Table 3), the resolution is 16 bits. Since the quantization noise of 16-bit data is -96 dB, the instantaneous S
/N ratio is 90dB to 78dB. Input level -
In the range of 18 dB to -30 dB, the linear sign “P0
001ABC..." or "P00001ABC...
.'' According to (Table 3), the resolution is 17 bits. Since the quantization noise of 17-bit data is -102 dB, the instantaneous S/N ratio is 84 dB to 72 dB. Similarly, instantaneous S/N ratios are obtained in the input level range up to -180 dB. FIG. 3 is an instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a run-length 1/2 compressed floating code. In the figure, the horizontal axis shows the input level, and the vertical axis shows the instantaneous S/N ratio. Characteristic curve X in the same figure
1 represents the instantaneous S/N ratio of the run length 1/2 compressed floating code shown in (Table 3). From the figure, it can be said that the characteristic curve X1 almost exceeds the auditory detection limit of 48 dB for the instantaneous S/N ratio in the input level range of 0 dB to -96 dB. Compared to the conventional linear code (16 bits),
Instantaneous S of about 42 dB at input level -96 dB
/N ratio improvement is achieved. The instantaneous S/N ratio improvement works over a wide range of input levels from -18 dB to -180 dB. Instantaneous S/ in the range of input level 0dB to -6dB
The N ratio deteriorates, but the deterioration is at most 6 dB, and it is still 90 dB.
It has an instantaneous S/N ratio of dB to 84 dB. This area is reserved as a margin and is usually rarely used, so it poses no problem at all. Also, in comparison with the polygonal code (16 bits) shown in the characteristic curve Z2 of FIG. 7, there is an improvement of 6 dB in the input level range of 0 dB to -12 dB, and an improvement of 18 dB at -96 dB. As described above, according to the first embodiment, even if a minute level signal is amplified, the quantization noise can be made smaller than the auditory detection limit over a wide range up to the input level of -96 dB. Therefore, the recording level can be set with sufficient headroom, and the recording level can be amplified arbitrarily in later editing. Further, in order to improve the problem of quantization noise of the characteristic spectrum that occurs in a minute level signal, it is possible to eliminate processing such as adding white noise (dither) to disperse the noise spectrum. Input level 0 with current technology
Amplify a wide range of signals from dB to -180dB,
Although it is difficult to digitize with an AD converter, the effects of the present invention will be maximized with the future development of functional elements and circuit technology that utilize high-temperature superconductivity. In addition, in (Table 3), the increase characteristics of the run length 1/2 compression floating code will be considered. In the first embodiment, by using the compressed remainder data C1 as the complement of the compressed remainder F1, the numerical value in the upper range is always larger than the lower range at the boundary of the ranges. Therefore, the run length 1/2 compressed floating code of the first embodiment has a monotonically increasing characteristic. Even if compressed data is mistakenly reproduced as a linear code (16 bits), the monotonically increasing characteristic is advantageous because it will not cause level reversals or jumps. This effect is due to the compression remainder data C1 being the complement of the compression remainder F1, but it is not necessarily necessary to use the complement and other expressions may be used. As mentioned above, with the current technology, it is difficult to amplify a signal with a wide range of input levels from 0 dB to -180 dB or to digitize it with an AD converter. Therefore, if a 30-bit linear code cannot be prepared, the linear code may be shortened. (Table 4
) represents a code decoding method for compressing and expanding a linear code (28 bits) into a run length 1/2 compressed floating code (16 bits).

[Table 4] In this case, the run length L0 is truncated at a maximum of 26. When run length L0=26, inversion bit T0 is omitted,
Let the remaining data be D0. Encoding is performed in the same manner, but the compressed remainder data C1 is omitted and mantissa data D1, that is, D0 is placed in its place. By truncating the run length L0, the instantaneous S/N ratio characteristic becomes a characteristic curve X2 in FIG. 3. The input level range from -162 dB to 0 dB is the same as X1 in the figure. Also, in exactly the same way,
The linear code may be shortened to 24 bits. (Table 5)
represents a code decoding method for compressing and expanding a straight line code (24 bits) into a run length 1/2 compressed floating code (16 bits).

[Table 5] In this case, the run length L0 is cut off at the maximum of "18". When the run length L0=18, the inversion bit T0 is omitted and the remaining data is set as D0. The encoding is done in the same way, but since it is known that the maximum value of the run length L1 is "9", the inversion bit T1 can be anything. Further, the compressed remainder data C1 is omitted and mantissa data D1, that is, D0 is arranged in its place. Run length L
By truncating at 0, the instantaneous S/N ratio characteristic becomes a characteristic curve X3 in FIG. Input level -114dB to 0
In the dB range, it is the same as X1 in the same figure. Any of the characteristic curves X1, X2, and X3 described above can almost clear the auditory detection limit of the instantaneous S/N ratio within the input level range of 0 dB to -96 dB. Next, a second embodiment of the present invention will be described with reference to the drawings. Figure 2 (a
) is the run length 1/4 in the second embodiment of the present invention
It is a conceptual diagram showing the structure of a compressed floating code, and FIG. 2(b) is a diagram showing a specific example for explaining the conversion process. In the figure, parts similar to those of the conventional example are given the same symbols. Hereinafter, the compression (encoding) procedure will be described first with reference to the figures. The second embodiment converts a linear code (32 bits) into a run length 1/4 compressed floating code (16 bits).
This is a coding method that compresses and expands the data into bits. The compressed continuous data Q1 is continuous data having a length of a run length L1 obtained by dividing the run length L0 of the continuous data Q0 by "4" and converting it into an integer. L1=int(L0/4) Also, if the remainder term of integer division is the compression remainder F1, then F1
= L0 mod 4. The inversion bit T1 is an inversion bit that breaks the run length of the compressed continuous data Q1. The compression remainder data C1 is a complement representation of the compression remainder F1. Further, the mantissa data D1 is partial data on the upper side of the data D0. The run length 1/4 compressed floating code is
Polarity bit P, compressed continuous data Q1, inversion bit T1,
The compressed remainder data C1 and the mantissa data D1 are arranged in this order. Hereinafter, the run length L0 is "0" based on FIG. 2(b).
The case of "28" to "28" will be explained. When L0=0, L1=int(0/4)=0 F1=0 mod 4 =0 Since the run length L1 is "0", there is no compressed continuous data Q1. The compressed residual data C1 is "11". Data D0 has 30 bits, of which the upper 12 bits "ABCDEFGHIJKL" are mantissa data D1. A run-length 1/4 compressed floating code has a polarity bit P, an inversion bit T1, a compressed remainder data C1, and a mantissa data D1 arranged in this order.
DEFGHIJKL”. Similarly, when L0 = 1, L1 = int (1/4) = 0 F1 = 1 mod 4 = 1, so there is no compressed continuous data Q1, and compressed residual data C1
is "10". The upper 12 bits of data D0 “AB
"CDEFGHIJKL" is the mantissa data D1. The run length 1/4 compressed floating code is “P110A
BCDEFGHIJKL”. When L0=3, since L1=int(3/4)=0 F1=3 mod 4 =3, the continuous data Q1 is "0" and the compressed residual data C1 is "00". The upper 12 bits of data D0
ABCDEFGHIJKL” is the mantissa data D1. The run length 1/4 compressed floating code is “P10
0ABCDEFGHIJKL”. Furthermore, when L0=20, L1=int(20/4)=5 F1=20 mod 4 =0, so the continuous data Q1 is "00000" and the compressed residual data C1 is "11". The upper 7 bits "ABCDEFG" of data D0 are mantissa data D1. Therefore, the run-length 1/4 compressed floating code is “P
00000111ABCDEFG”. Furthermore, when L0=23, since L1=int(23/4)=5 F1=23 mod 4 =3, the continuous data Q1 is "00000" and the compressed residual data C1 is "00". The upper 7 bits "ABCDEFG" of data D0 are mantissa data D1. Therefore, the run-length 1/4 compressed floating code is “P
00000100ABCDEFG”. In this way, a straight line code (32 bits) is compressed into a run length 1/4 compressed floating code (16 bits). Next, the decompression (decoding) procedure will be explained. In the decoding procedure, the polarity bit P is removed from the compressed data, and the run length L1 is obtained from the compressed continuous data Q1 and the inverted bit T1. Also, the compressed residual data C immediately after the inversion bit T1
1 (2 bits) and mantissa data D1 are obtained. A compression remainder F1 is obtained by inverting the compression remainder data C1. From these, the run length L0 of continuous data Q0 of the original data is L0=
Calculate from 4*L1+F1. Continuous data Q0 is restored by concatenating "0"s having a length of run length L0. An inverted bit T0 is added after continuous data Q0, and mantissa data D1 is added after that. A polarity bit P is added to the beginning to form the original data, but if this code length is less than W0, a fixed value is assigned to the lower part of the mantissa data D1 to make the code length W0. Follow these steps:
Performs decryption processing. Below, some cases of the run length L1 will be specifically explained with reference to FIG. 2(b). When L1=0 and F1=0, since L0=4*0+0=0, there is no continuous data Q0. The mantissa data D1 is 12 bits and is "ABCDEFGHIJKL". At this time, the original data is "P1ABCDEFG" by arranging the polarity bit P, inversion bit T0, and mantissa data D1.
HIJKL＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊". It is assumed that "***..." is assigned a fixed value "1000...". Similarly, when L1=5 and F1=0, L0=4*5
+0=20, so the continuous data Q0 is “0000000000
0000000000”, and the mantissa data D1 is “AB
CDEFG”. At this time, the original data is ``P00000000000000'' by arranging the polarity bit P, continuous data Q0, inverted bit T0, and mantissa data D1.
0000001ABCDEFG****”. In this way, a linear code (32 bits) is decoded and expanded from a run-length 1/4 compressed floating code (16 bits). In the same manner as above, the results summarized for all cases are shown in (Table 6).

[Table 6] According to Table 6, according to this embodiment, a maximum precision of 32 bits to 14 bits can be obtained. Here, a description will be given of the instantaneous S/N ratio when a linear code is input, compressed and encoded using a run-length 1/4 compression floating code, and this code is decoded to output a decompressed restored code. The input level range from 0 dB to -6 dB is represented by the linear code "P1ABC...", and from (Table 6) the resolution is 1.
It is 4 bits. The quantization noise of 14-bit data is 1
Assuming -6 dB per bit, it becomes -84 dB. Therefore,
Instantaneous S/N in input level range 0dB to -6dB
The ratio will be 84dB to 78dB. Input level -6d
In the range from B to -12 dB, the linear code “P01AB
C...'', and from (Table 6), the resolution is 15 bits. Since the quantization noise of 15-bit data is -90 dB, the instantaneous S/N ratio is 84 dB to 78 dB. In the input level range of -12dB to -18dB, the linear code is "P001ABC..." (Table 6)
The resolution is 16 bits. Since the quantization noise of 16-bit data is -96 dB, the instantaneous S/N ratio is 8.
It becomes 4dB to 78dB. In the input level range of -18 dB to -30 dB, the linear code “P0001AB
C..." or "P00001ABC...", and from (Table 6), the resolution is 17 bits. Since the quantization noise of 17-bit data is -102 dB, the instantaneous S/N ratio is 84 dB to 72 dB. Similarly, instantaneous S/N ratios are obtained in the input level range up to -180 dB. FIG. 4 is an instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a run-length 1/4 compression floating code. In the figure, the horizontal axis shows the input level, and the vertical axis shows the instantaneous S/N ratio. The characteristic curve Y1 in the same figure is (Table 6
) represents the instantaneous S/N ratio of the run-length 1/4 compressed floating code. It is clear from the figure that the characteristic curve Y1 exceeds the auditory detection limit of 48 dB for the instantaneous S/N ratio in the input level range of 0 dB to -96 dB. Input level -96 compared to conventional linear code (16 bits)
An instantaneous S/N ratio improvement of about 60 dB in dB is achieved. The instantaneous S/N ratio improvement works over a wide range of input levels from -18 dB to -196 dB. The instantaneous S/N ratio deteriorates in the input level range of 0 dB to -12 dB, but the deterioration is 12 dB at most, and still 84 dB to 7 dB.
It has an instantaneous S/N ratio of 8 dB. Since this area is not normally used as a margin, it poses no problem at all. Also, in comparison with the polygonal code (16 bits) shown in the characteristic curve Z2 in FIG.
The improvement ranged from 4 dB to -196 dB, with -96 d
In B, an improvement of 36 dB is obtained. As described above, according to the second embodiment, even if a signal with an input level of −144 dB, which is even smaller than that in the first embodiment, is amplified, the quantization noise is not audibly detected at all. Therefore, the recording level can be set with sufficient headroom, and the recording level can be amplified arbitrarily in later editing. (Table 7) is a straight line code (29 bits) with a run length of 1/
This is a code decoding method for compressing and expanding a 4-compression floating code (16 bits).

[Table 7] In this case, the run length L0 is cut off at the maximum of "20". When the run length L0=20, the inversion bit T0 is omitted and the remaining data is set as D0. Encoding is done in the same way, but since it is known that the maximum value of run length L1 is "5", the inversion bit T1 is omitted. Furthermore, the compressed residual data C1 may be of any type. Run length L0
By terminating , the instantaneous S/N ratio characteristic becomes a characteristic curve Y2 in FIG. 4 . Input level -126dB to 0d
The range B is the same as Y1 in the same figure. Furthermore, in exactly the same way, the linear code may be shortened to 26 bits. (Table 8) is a linear code (26 bits) with a run length of 1/
This is a code decoding method for compressing and expanding a 4-compression floating code (16 bits).

[Table 8] In this case, the run length L0 is cut off at the maximum of "16". When the run length L0=16, the inversion bit T0 is omitted and the remaining data is set as D0. Encoding is done in the same way, but since it is known that the maximum value of run length L1 is 4, the inversion bit T1 is omitted. Furthermore, the compressed residual data C1 may be of any type. Run length L0
By terminating , the instantaneous S/N ratio characteristic becomes a characteristic curve Y3 in FIG. 4 . Input level -102dB to 0d
The range B is the same as Y1 in the same figure. No matter which of the characteristic curves Y1, Y2, and Y3 mentioned above, the input level is 0.
It can completely clear the auditory detection limit of instantaneous S/N ratio in the range from dB to -96 dB. Next, a third embodiment of the present invention will be described. The third example is a linear code (24
This is a coding method that compresses and expands a run-length 1/4 compression floating code (8 bits). That is, the number of bits of the code of the second embodiment is changed. Therefore, the compression (encoding) and decompression (decoding) procedures are the same as in the second embodiment, so their explanation will be omitted. Table 9 shows the results summarized for all exponent parts, that is, range cases.

[Table 9] According to Table 9, according to this embodiment, a maximum precision of 24 bits to 6 bits can be obtained. FIG. 5 is an instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a run-length 1/4 compressed floating code according to the third embodiment. In the figure, the horizontal axis shows the input level, and the vertical axis shows the instantaneous S/N ratio. The characteristic curve V1 in the figure represents the instantaneous S/N ratio of the run length 1/4 compression floating code shown in Table 9. It is clear from the figure that the characteristic curve Y1 has a linear sign (
8-bit), it exhibits a substantially flat instantaneous S/N ratio characteristic over a much wider input level range. Although this is still below the instantaneous S/N ratio auditory detection limit of 48 dB, by combining it with differential codes, etc.
You can take advantage of the wide input level range and high resolution. It can be used for redundant compression that utilizes the correlation between samples of image data. In the first to third embodiments described above, the linear code is a folded binary code, but other linear codes such as 2'S complementary code or offset binary code can also be converted into each other. Alternatively, it can be applied in exactly the same way by simply changing a predetermined logical value. Further, although only the cases where n is "2" and "4" have been described, n may be any integer that is "2 or more". In this case, since the number of compression remainder cases changes depending on the value of n, it goes without saying that the word length of the compression remainder data may be changed.

Effects of the Invention As described above, the present invention uses compressed continuous data obtained by compressing the upper run length of the original data to about 1/n, inversion bits, and residual data derived when compressing the run length. It constitutes a section or a range. The exponent part, that is, the word length of the range changes depending on the run length of the original data. Due to this effect, when the run length is small, the exponent part, that is, the range, is made up of a small word length, and the word length of the mantissa part is made large, so that expression accuracy can be maximized. In addition, when the run length is large, the exponent part, or range, is constructed with a large word length, so the type of the exponent part, or range, can be expressed subdivided and the input level range can be widened, expanding the space expressed by the compression code. be done. This maintains both accuracy and input level range above specified characteristics, so when applied to digital recording and playback equipment for business editing in particular, quantization noise remains audible even after repeated mixing or dubbing that involves amplification and attenuation. No longer detected, input level 0 to -96dB
It is possible to secure an instantaneous S/N ratio of approximately 48 dB or more, which is the auditory detection limit, over a wide range of .

[Brief explanation of the drawing]

FIG. 1 (a) is a conceptual diagram showing the structure of a run-length 1/2 compression floating code in the first embodiment of the present invention; (b) is an explanatory diagram also illustrating the conversion process;

[Figure 2
] (a) is a conceptual diagram showing the structure of a run-length 1/4 compression floating code in the second embodiment of the present invention; (b) is an explanatory diagram showing a specific example for explaining the conversion process.

FIG. 3: Run length 1 in the first embodiment of the present invention
Instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a /2 compression floating code

FIG. 4 is an instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a run-length 1/4 compressed floating code in the second embodiment of the present invention.

FIG. 5 is an instantaneous S/N ratio characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a run-length 1/4 compressed floating code in the third embodiment of the present invention.

[Figure 6] (a) Conventional 13-line code (
Conceptual diagram (b) showing the configuration of 12 bits) is also an explanatory diagram showing a specific example to explain the conversion method.

[Figure 7] Characteristic diagram of a restored code obtained by decoding and expanding a compressed code using a conventional polygonal line code

[Explanation of symbols]

P Polarity bit Q0 Continuous data Q1 Compressed continuous data Q2 Range data L0 Run length L1 of continuous data Q0 Run length L2 of compressed continuous data Q1 Range data Q2
Run length T0, T1 Inverted bit D0 Remaining data D1, D2 Mantissa data W0 Code length of original data W1, W2 Code length of compressed data

Claims (12)

[Claims] 1. A linear code serving as the original data, that is, continuous data Q0 in which bits of a predetermined logic are continuous in the upper part, and an inverted bit T that breaks the continuity of the continuous data Q0.
0 and lower-order data D0 after the inverted bit T0 is encoded to produce compressed data, that is,
Compressed continuous data Q1 obtained by compressing the run length of the continuous data Q0, an inversion bit T1 that breaks the continuity of the compressed continuous data Q1, and compressed residual data representing a remainder F1 generated when compressing the run length. C
1 and mantissa data D1 obtained by rounding the lower order data D0. However, when the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1, and n is an integer of 2 or more, L1=int(L0/n) F1=L0 mod n. 2. The run length 1 according to claim 1, wherein the compressed data is arranged in the order of compressed continuous data Q1, inverted bit T1, compressed remainder data C1, and mantissa data D1 from the highest order.
/n compression floating code method. 3. A linear code serving as the original data, that is, a polarity bit P, continuous data Q0 in which upper bits of a predetermined logic are consecutive except for this polarity bit P, and an inversion that breaks the continuity of the continuous data Q0. A compressed sequence obtained by encoding a linear code consisting of bit T0 and lower data D0 after the inverted bit T0 and compressing compressed data, that is, polarity bit P and the run length of the continuous data Q0. data Q1 and the compressed continuous data Q
Compressed data consisting of an inversion bit T1 that breaks the continuity of 1, compressed remainder data C1 representing a remainder F1 generated when compressing the run length, and mantissa data D1 obtained by rounding the lower order data D0. A run-length 1/n compressed floating code method that converts to . However, when the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1, and n is an integer of 2 or more, L1=int(L0/n) F1=L0 mod n. 4. The run length 1/n according to claim 1 or 3, wherein the compression remainder data C1 is a complement of the compression remainder F1.
Compressed floating code method. 5. The run length 1/n compression floating coding method according to claim 1, wherein lower bits of the data D1 are deleted so that the compressed data has a predetermined code length W1. [Claim 6] Predetermined limit length LS and run length L0
When the smaller of is Lmin, L1 and F1 are
4. The run length 1/n compression floating code method according to claim 1 or 3, wherein L1=int(Lmin/n) F1=Lmin mod n. [Claim 7] The compressed data includes the polarity bit P from the upper
4. The run length 1/n compression floating code method according to claim 3, wherein the compressed continuous data Q1, the inverted bit T1, the compressed remainder data C1, and the mantissa data D1 are arranged in this order. [Claim 8] Predetermined limit length LS and run length L0
When the smaller of is Lmin, L1 and F1 are
7. The method according to claim 6, wherein L1=int(Lmin/n) F1=Lmin mod n, and when the run length L0 is larger than the limit length LS, the inversion bit T0 and the inversion bit T1 are omitted to perform front-justification. Run length 1/n compression floating code method. [Claim 9] Predetermined limit length LS and run length L0
When the smaller of is Lmin, L1 and F1 are
L1=int(Lmin/n) F1=Lmin mod n When the run length L0 is larger than the limit length LS, the inversion bit T0 and the inversion bit T1 are omitted and left-justified to represent the compression remainder F1. Compressed residual data C1
The run length 1/ of claim 6 is set to a predetermined fixed value.
n compression floating code method. 10. Compressed data, that is, compressed continuous data Q1 in which bits of a predetermined logic are continuous in the upper part, and an inverted bit T1 that breaks the continuity of the compressed continuous data Q1.
The compressed data consisting of the compressed remainder data C1 representing the remainder F1 generated when compressing the run length and the mantissa data D1 is decoded to form a linear code that becomes the original data, that is, bits of a predetermined logic are consecutive in the upper part. Continuous data Q0
, an inversion bit T0 that breaks the continuity of the continuous data Q0, and lower data D0 after the inversion bit T0.
Run length 1/n to be converted to a linear code consisting of
A method for decoding compressed floating codes. However, when the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1, the remainder obtained from the compressed residual data C1 is F1, and n is an integer of 2 or more, L0=
Let L1*n+F1 D0=D1. Claim 11: Compressed data, that is, polarity bit P
, compressed continuous data Q1 in which upper bits of a predetermined logic are consecutive except for this polarity bit P, and the compressed continuous data Q
Compressed data is composed of an inversion bit T1 that breaks the continuity of 1, compressed remainder data C1 representing a remainder F1 generated when compressing the run length, and mantissa data D1.
A linear code that is decoded and becomes the original data, that is, a polarity bit P, continuous data Q0 in which upper bits of a predetermined logic are continuous except for this polarity bit P, and the continuous data Q
A method for decoding a run-length 1/n compressed floating code that converts it into a linear code composed of an inverted bit T0 that breaks the continuity of 0s and lower-order data D0 after the inverted bit T0. However, when the run length of the continuous data Q0 is L0, the run length of the compressed continuous data Q1 is L1, the remainder obtained from the compressed residual data C1 is F1, and n is an integer of 2 or more, L0=L1*n+F1 D0= Let it be D1. [Claim 12] The decoded linear code has a predetermined code length W.
A fixed value ``100...'' is added to the lower part of data D0 so that
12. The run length 1/n compression floating code according to claim 10 or 11, wherein the run length 1/n compressed floating code adds ``. How to decrypt.