JPS6129220A - Method of digital signal transmission - Google Patents

Abstract

PURPOSE:To heighten efficiency of bit compression or bit reduction rate by making forecast filter processing on word string in front part from a reference word and word string in rear part independently and transmitting with the reference word. CONSTITUTION:Inputted digital signals are made to blocks for specified number of words along time axis and stored in a block memory 3. The content of the memory 3 is processed and then given to a mode selection.adoptive information calculating circuit 6. The calculating circuit 6 selects most suitable forecast filter characteristic for word string in the front part of reference word and word string in the rear part of reference word in the block. Forecast filter processing of front part word string and rear part word string is made independently basing on the selection, and transmitted together with the reference word. Thus, efficiency of bit compression or bit reduction rate can be heightened.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a digital signal transmission method for transmitting digital signals such as PCM signals, and particularly relates to a digital signal transmission method with improved bit compression efficiency or bit reduction rate. .

[Conventional technology] In recent years, with the advancement of digital technology, analog signals such as audio signals and video signals are sampled, quantized and encoded, and transmitted as so-called PCM (pulse code modulation) signals ( This may include recording and playback).

In this way, when converting an analog signal into a PCM digital signal and transmitting it, generally speaking, the higher the sampling frequency, the wider the analog signal band that can be transmitted, and the higher the number of quantization bits, the wider the dynamic range. known to be widespread. Therefore, the original analog signal can be reproduced with high fidelity, i.e. wideband and highly dynamic.
Digital transmission over a range requires a high sampling frequency and a large number of quantization bits, resulting in a high number of bits to be transmitted per unit time, the so-called bit rate.

However, transmission media (which may also include recording media)
The above-mentioned bit rate is limited by the characteristics of By considering economic efficiency, cost performance, etc. when supplying products such as devices, it is important to perform high-quality signal transmission or recording/playback at as low a bit rate as possible.

By the way, as a technique for transmitting a signal in a large Guinamink range at a relatively low bit rate, a digital signal processing technique using linear prediction such as a differential PCM method is known.

For example, the patent application No. 58-976 proposed earlier by the inventor of the present invention
No. 88, etc., the digital signal data to be transmitted is divided into blocks every certain number of words and transmitted through one of a plurality of prediction filters. (Straight PCM
) data, one that outputs differential PCM data, and one that outputs summation PGM data. I try to select each block. This optimal PCM mode selection is based on the maximum absolute value of straight PCM data within one block, the maximum absolute value of differential PCM data, and the sum PCM mode.
Detect the maximum absolute value of each CM data, and calculate these three
This is done by comparing the maximum absolute values within two blocks and selecting the PCM mode that yields the smallest value.

[Problem that the invention seeks to solve]

By the way, when differential PCM data or summation PCM data is selected as data that has been subjected to such predictive filter processing, one word per block is used as the standard in order to restore the original sampling peak value data. Sampling peak value data (straight PCM data) is required.

Since straight PCM data is always selected as the reference data word regardless of the optimal prediction filter according to the input signal, this is an obstacle to improving the bit compression efficiency or bit reduction rate.

In view of these circumstances, it is an object of the present invention to provide a digital signal transmission method that can further improve bit compression efficiency or bit reduction rate.

[Means for solving the problem] That is, the digital signal transmission method according to the present invention divides an input digital signal into blocks along the time axis for each fixed number of words, and blocks each block at a predetermined position in the time axis direction. 1
The present invention is characterized in that a word is used as a reference word, and prediction filter processing is performed independently on a word string in the front part and a word string in the rear part of the base word in the block, and the results are transmitted together with the reference word.

[For production]

Therefore, even though predictive filter processing is performed independently on the front and rear parts of blocks, only one reference word is required per block, which substantially improves bit compression efficiency or bit reduction rate. [Embodiment] Hereinafter, an embodiment of the digital signal transmission method according to the present invention applied to an audio bit rate reduction system will be described with reference to the drawings.

FIG. 1 shows an example of the configuration of the encoder side of the audio bitrate reduction system. The input terminal 1 of the encoder shown in FIG. 1 receives, for example, an l-word 14-hit digital PCM signal ( A sampled peak value data signal) is supplied. This input digital signal is sent to a pre-emphasis circuit 2. This pre-emphasis circuit 2 is used to improve the signal-to-noise ratio by emphasizing the high frequency range of the signal, and has a time constant of, for example, about 50 μs. A digital signal (for example, 1 word of 14 bits) from the pre-emphasis circuit 2 is sent to a block memory 3. Here, the input digital signal has a fixed number of words (
This is assumed to be 2n+1 words. ) are divided into blocks,
The data of 20+1 words corresponding to this l block is stored in the block memory 3. The data read from the block memory 3 is sent to a difference/sum calculation circuit 4, and, for example, differential PCM data, straight PCM data, and sum FCM data each having the number of words for one block are calculated. Now, data of one block (2n11 words) of the input digital signal is sequentially written as X-H, X n+1+ ''' r
Xl g xQ , x11x2. ...+xn 11
When Xn is the data x at the center position.

The above differential PCM data is as follows: dl=xo7Xl, d-XO-XIdrl=Xn
-1"n r '"2 x-, 41-x.

2n words of are calculated for each block, and the summed PCM data is as follows: al = x6 +xt, a-1 = X6 +
x-1az = x1+X2, a-2:X-1+
2n words are calculated for each block, such as X-2. In addition, the above straight PCM data excludes the data of the above reference word, X-0 to X-1, xl to xn.
2n words are obtained per block.

For example, data of each PCM mode (differential PCM, summation PCM, and straight PCM) of 2n words per l block is sent to the maximum absolute value detection comparison circuit 5 in % block,
The maximum absolute value for each mode is detected and compared for the data in the first half of the block (each data with subscripts from -n to -1), and the maximum absolute value of the data in the second half of one block (each data with subscripts 1 to n) is detected and compared. The maximum absolute value for each mode of data) is detected and compared. In other words, for the first half 93A block along the time axis, the above (
Difference PCM data (maximum absolute value d of L, ~d-0
=-1 Maximum absolute value aM- of the summed PCM data a-1 to a-0 and the straight 20-bar day X-
1-, X-n, and compare these three maximum absolute values dM-+ ``M- and XM- with each other to find the data corresponding to the smallest value. The CM mode is selected by the mode selection/adaptive information calculation circuit 6. Also, for the latter % block, the data dlxdn, at of each PCM mode is
Each maximum absolute value dM+ of ~an and Xl−Xn
, aM+ and XM+, and compare them to select the PCM mode that yields the smallest value.
- Selected by the selection/adaptive information calculation circuit 6.

Such mode selection is performed to maximize the compression efficiency when digitally transmitting the input signal, and generally corresponds to the selection of predictive filter characteristics, and the applicant is Previously proposed patent application 1976-976
No. 87-9, Patent Application No. 163054/1982, Patent Application No. 1982
Although it is disclosed in detail in Japanese Patent No. 166267, etc., the operating principle will be briefly explained below.

Figure 3 shows each PCM mode F(
Curve A shows the frequency characteristics of the dynamic range of differential PCM mode, curve B shows the frequency characteristic of sum PCM mode, and curve C shows the frequency characteristics of straight PCM mode. It shows. Note that the sampling frequency fs during digitization is 32 kHz. As is clear from Fig. 3, in the differential PCM mode, the lower the input frequency fi, the better the S/N ratio against quantization noise and the larger the dynamic range, and in the summation PCM mode, the higher the ft, the greater the dynamic range.・The range becomes larger. Therefore, by selecting the mode that provides the largest dynamic range according to the frequency of the input signal,
It performs digital signal transmission with high bit compression efficiency. For example, the dynamic range of the differential PCM mode and the dynamic range of the straight PCM mode become equal when the input signal frequency fi is % of the sampling frequency, and when this frequency is fs/6, the dynamic range of the straight PCM mode becomes equal. The maximum absolute value of the data becomes equal to the maximum absolute value of the differential PCM data. Also,
When the input signal frequency f1 is % of the sampling frequency fs (i = fs / 3), the straight PC
, M mode and the sum PCM mode become equal, and at this time, generally, the maximum absolute values of the straight and sum PCM data become equal to each other. Therefore, the maximum absolute value of the data in each of the differential, summation, and straight PCM modes is detected, and these maximum absolute values are compared, and the PCM mode that yields the minimum value has the largest dynamic range. This is the best mode available. In the case of the present invention, such selection of the optimum PCM mode is performed for each % block.

Referring again to FIG. 1, the mode selection/adaptive information calculation circuit 6 outputs mode selection information according to the optimal PCM mode selected for each % block, and this mode selection information is sent to the mode processing circuit 7 and the multiplexer. It has been sent to 8.

The mode processing circuit 7 is supplied with data from the difference/sum calculation circuit 4 and the block memory 3, and the data for the above % blocks in the PCM mode according to the above mode selection information is supplied to the mode processing circuit 7. The data is sent again to the block memory 3, and from this block memory 3, the above data is sent to the range processing circuit 11. This block
The data sent from the memory 3 to the range processing circuit 11 is
The data is in the optimal mode of difference, sum, or straight selected for each % block above, and is 14 bits or 15 hits (in the case of difference and sum PCM data) of 1 word. Since the dynamic range is large, normal level signals are expressed with an extremely small number of bits. That is, most of the 14-bit or 15-bit data for one word is expressed by a small number of bits, about several bits.

The range processing circuit 11 is a shifter that performs a so-called bit shift, and operates to normalize all data for a % block by a peak value. For example, this bit shift amount is used as an exponent value, and the shifted This is for so-called floating point display where data is used as a mantissa value. In this case, the bit corresponding to the above exponent value
The shift amount information is output as readaptive information or range information from the mode selection/adaptive information calculation circuit 6, and is sent to the multiplexer 8 and the range processing circuit 11, respectively.

The bit-shifted data from the range processing circuit 11 is sent via an adder 12 to a requantizer 13, where it is requantized into data of 7 bits per word, for example. Associated with this requantizer 13 is an error feedback circuit 14 for so-called noise shaping processing.

This error feedback circuit 14 is a requantizer 13
An adder 15 that performs a subtraction operation takes out the error amount (error amount) that has occurred between the input and output of The signal is attenuated by a predetermined amount, and sent to an adder 12 that performs a subtraction operation, which is provided on the input side of the redoji converter 13, and is fed back. Such error feedback allows the spectral pattern of the requantization noise to change.

Here, when the signal sent from the requantizer 13 to the adder 15 is El, and the error signal from the adder 15 is E2, as shown in Japanese Patent Application No. 58-97689, the following equation is obtained. K in this formula is the feedback amount adjustment circuit 1
7 and fs is the sampling frequency. The frequency response when K in this equation is changed is as shown in FIG. 4, and as K approaches 1, a noise shaping effect is obtained in which the noise spectrum concentrates on the high frequency side. Such noise shaping processing is performed for each % block according to the mode selection information from the mode selection/adaptive information calculation circuit 6. For example, when the straight or summation PCM mode is selected, the input signal frequency is Since it is relatively high, the above coefficient K
By increasing (bringing it close to 1), the noise is biased toward the high frequency side, and the noise is masked by the input signal, thereby improving the audible S/N ratio.

As described above, the data of n words and 17-7 hits per % block, which has been requantized by the requantizer 13 and simultaneously subjected to the noise shaping process, is sent to the multiplexer 8.

This multiplexer 8 receives one word of reference data per block from the block memory 3, and one word of mode selection information M and adaptive information (range) per % block from the mode selection/adaptive information calculation circuit 6. Information) R is also sent, and when these are put together for one block, it becomes as shown in FIG.

In this FIG. 5, the center reference word WQ corresponds to the reference data of the lower 14 bits of the above l2, and the n words of Wl-Wn correspond to the data x1-wxn and the reference data xo for the first half % block of the input data. 1 obtained based on
This corresponds to the transmission signal data of 7 bits per word, and furthermore, range information f1. lJ: and mode selection information M+ of 1 word and 2 bits are provided. Also, W-n-W- in FIG.

R- and M- are the transmission signal data, range information, and mode selection information for the latter % block, respectively.

Each piece of data as shown in FIG. 5 is converted into, for example, a serial signal according to a predetermined transmission format in the multiplexer 8, outputted via the output terminal 9, and transmitted via a transmission medium (which may include a recording medium). transmitted.

Next, FIG. 6 shows an example of a decoder for decoding the digital signal transmitted from the above-mentioned encoder via the transmission system and converting it into the original digital PCM signal (sampling peak value data signal).

In FIG. 6, the transmitted digital signal is supplied to a multiplexer 22 via an input terminal 21. In FIG. This multiplexer 22 uses the block synchronization signal and word synchronization signal in the transmission digital signal, for example, to input each word Wo, Wl-Wn,
M+, R+, W-1 to W-, ;, M-, a- are separated from each other and adaptively processed (bit shifted) to transmit signal words Wl-Wn (or W
-1 to W-n) to the range processing circuit 23. This processing circuit 23 is based on the content of adaptive information (range information) R+ (or IL) from the multiplexer 22,
Perform range restoration operation. For example, the 7-bit word W1
The MSB (bit indicating the sign) of ~Wn (or W-1 ~ W-n) is sign-extended by the hit shift, and further L
Following the SB, necessary invalid hits are added to convert the data into 14 or 15 bits as a whole. This data is PCM mode data specified by mode selection word M+ (or M-), and is straight PCM mode data.
When the mode is selected, the data x0 to xn
(or X-1 to X-n), the data di-dn (or d-1 to d-n), sum P
In the CM mode, the data a1 to an (or a-1,
-a-re). Such a range processing circuit 2
The data from 3 is sent to the block memory 24, and this block memory 24 and the difference/sum calculation circuit 2
Data is being sent and received between 5 and 5. The difference/sum calculation circuit 25 is supplied with the mode information M+ (or M-) from the multiplexer 22 and the data of the reference word Wo, and based on these data, the original sampling peak value data xl ""'Xr+ (or X-1
˜X−n) is performed. That is, regarding the data for each % block, for example, when the difference PCM mode is selected on the encoder side, the difference/sum calculation circuit 25 on the decoder side performs summation processing, and the summation PCM mode is selected on the encoder side. When this happens, the difference/sum calculation circuit 25 performs differential processing to restore the original straight 20M data. It goes without saying that when the straight PCM mode is selected on the encoder side, the data from the range processing circuit 23 of the decoder can be used as is.

In this way, the difference/sum calculation circuit 25 performs difference/sum processing as necessary, and the block memory 24
from the original straight PCM data x1 to xn (or
−1+”+x−n) is output and sent to multiplexer 2.
Sent to 6. The data x, ) of the reference word WO from the multiplexer 22 is also sent to this multiplexer 26, and all the strays 1-P for one block are sent to this multiplexer 26.
CM data (sampling wave height data (X-□~xn
is output from the multiplexer 26 and taken out from the decode output terminal 29 via the de-emphasis circuit 28. Note that the de-emphasis circuit 28 has a characteristic opposite to that of the pre-emphasis circuit 2 on the encoder side.

In the digital signal transmission system consisting of the encoder and decoder configured as described above, the reference word Wo
(Set the word length of data X(1) to LO bits (for example, 14
bits), and transmission signal words W1 to Wn and W-1-W.
-, the word length is Ll bits (for example, 7 bits),
When the word length of range information R+ or R- is LR bits (e.g. 3 bits) and the word length of mode selection information M+ or M- is LM bits (e.g. 2 bits), the total amount transmitted per block is The number of bits K A 211+1 is KA2n+1=Lo +2(nLl +Ln+LM)
Good luck. Here - as an example, n=8. LO-14゜L1
=7, LR=3, and LM=2, the number of words of sampling data transmitted for each l block is 17 words, and the total number of bits KA,7 is 136 bits. On the other hand, the total number of bits KCl7 for 17 words of the original 1-word 14-bit digital PCM data (sampled peak value data) is 14X17'=238 (bits), and 0゛, which is the compression efficiency, is Reduction rate KA
I”/KCsr.

is approximately 0.57.

By the way, the patent application filed in 1983-1997, which was previously proposed by the applicant
In the prior art of the present invention, such as in No. 687, one type of mode selection and range selection is performed for each block. For example, when transmitting sampling data of n+1 words per block, the word length LO bit standard is selected. When transmitting one word, n words of transmission signal word of word length L1 bits, one word of word length, R bits of range information 19-1-', and one word of mode selection information of word length Ly bits. The total number of focuses in one block KBrI+1 is )KB n+1 = L
'o +nLl+L R1LM tonal. = Example c! :L,
Te, 1l=8', Lo=14, L1=7,
If LR=3, LM=2, 9 in 1 block
Word sampling data can be transmitted, and the total number of bits KBo=75 (bits). Also, 1
The total number of bits KC9 when transmitting 9 words of sampling peak value data of 14 bits is 14X9=12
Since it is 6 (bits), the bit reduction rate KB9/KCq
is approximately 0.60, and it can be seen that the bit reduction rate of 0.57 shown in the example of the embodiment of the present invention is superior by approximately 0.03.

The bit reduction rate KBn according to the prior art of the present invention is
+1/KCn+1 and the bit reduction rate KAzn+x/KC2n+1 according to the embodiment of the present invention are compared and their ratio is taken, KCn+1=(n+1)Lo, K
Czn+1=(2n+1)L.

Since , then . A specific example of this bit reduction rate ratio is shown in FIG. In this FIG. 7, curve A has LO=12, Ll-
4, LR=3. The ratio of the hit reduction rate when LM=1 and n is changed from 2 to 40, and curve B is Lo
=16. Ll:14. LR=:3, LM""1 insertion, n
It shows the ratio of the bit reduction rate when changing from 2 to 40.

As is clear from FIG. 7, according to the embodiment of the present invention, it is possible to achieve further bit compression or bit reduction compared to the above-mentioned new technology previously proposed by the applicant. be.

Incidentally, in the above embodiment, three types of prediction filters, differential, summation, and straight PCM modes, can be selected, but in general, one of a plurality of types of prediction filters with different characteristics is selected. You can also do this. That is, as the simplest prediction filter, output data y1 is composed of two input data X1 . X
When considering a prediction filter represented by a linear linear combination yi ``Xi kxl-1 of □-1, the three modes of the above embodiment correspond to setting the coefficient k to one of 1. However, these 1.0.
Coefficients other than -1 may also be selected, and one of four or more different predictive filter characteristics may be selected for each of the % blocks in accordance with the input signal.

In addition, using a high-order prediction filter that obtains output data by a linear combination of three or more input data that are continuous on the time axis, the characteristics of this prediction filter are
The configuration may be such that the optimum one is selected for each block.

It should be noted that the present invention is not limited to the above-mentioned embodiments. For example, the transmission order of the data to be transmitted as shown in FIG. 5 can be arbitrarily set. Send mode selection information M-, M+ and range information R-, R+, then transmit signal words W-1-W-n
and Wl-Wn may be sent sequentially. Also,
The word at the center position along the time axis of one block of sampling data to be transmitted in Fig. 2 is the reference word WO, but the word at a position either before or after this center position A reference word may be used, and the number of words before and after the reference word may be set to be different from each other. Furthermore, in addition to basic modes such as differential, summation, and straight PCM modes, the PCM modes to be selected include a plurality of prediction filters with mutually different characteristics, and the characteristics of these prediction filters are applied to the input signal. The selection may be made depending on the

〔Effect of the invention〕

Brown riceAccording to the digital signal transmission according to the present invention≠cloud, ■ A block is divided into two parts, a standard word of one word is set in common to each divided part, and each divided part is independently assigned the optimum P.
Since adaptive processing such as CM mode selection and range selection is performed, only one reference word is required for two adaptively processed parts4, making it possible to significantly improve bit compression efficiency or bit reduction rate. .

[Brief explanation of drawings]

Fig. 1 is a block circuit diagram showing the encoder side configuration of an embodiment of the present invention, Fig. 2 is a diagram showing sample link data for 2 minutes to be transmitted, and Fig. 3 is a diagram showing the dynamic . A graph showing the frequency characteristics of the range, Fig. 4 a graph showing the noise shaping characteristics due to error feedback, Fig. 5 a diagram showing the word structure of one block transmitted, and Fig. 6 a graph showing the noise shaping characteristics due to error feedback. FIG. 7 is a block circuit diagram showing the decoder side configuration of the embodiment, and a graph showing the ratio of bit reduction rates between the embodiment of the present invention and the prior art.

Claims (1)

[Claims] The input digital signal is divided into blocks along the time axis for each fixed number of words, one word at a predetermined position in the time axis direction within one block is used as a reference word, and the word string in the front part and the rear part of the reference word in the block is A digital signal transmission method characterized in that predictive filter processing is performed independently on each of the word strings and the results are transmitted together with the reference word.