JP2007047813A - Digital signal processing method, its program, and recording medium storing the program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To maintain continuity and efficiency with substantially no degradation by concluding filter processing requiring processing extending over preceding and succeeding frames, like, for example, an interpolation filter, autoregressive prediction coding, and decoding in only the current frame. <P>SOLUTION: An encoder performs encoding to a sample sequence at the tail end of the frame previous to the current frame or the sample sequence at the top of the current frame separately from the encoding to the current frame to generate an auxiliary code (410). The auxiliary code is employed as a part of the code of the current frame (19). A decoder, in case the previous frame of the current frame is lost, decodes the auxiliary code within the code of the current frame to obtain the sample sequence at the tail end of the frame before the current frame or the sample sequence at the top of the current frame. The processing extending over the preceding and succeeding frames is performed by using the obtained sample sequences. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to encoding and decoding of a digital signal in units of frames, a processing method related thereto, a program thereof, and a recording medium storing the program.

In processing of digital signals such as audio and images in units of frames, processing across frames such as prediction and filtering is frequently performed. By using samples from the front and back frames, continuity and efficiency can be improved. However, in packet transmission, the sample of the previous frame or the sample that follows may not be obtained, and processing from only the designated frame may be required. In these cases, continuity and compression efficiency decrease.
First, an encoding method and a decoding method that can be considered as an example partially using digital signal processing to which the digital signal processing method of the present invention can be applied will be described with reference to FIG. (This example is not publicly known.)
The digital signal of the first sampling frequency from the input terminal 11 is divided by the frame dividing unit 12 in units of frames, for example, every 1024 samples, and the digital signal for each frame is converted from the digital signal of the first sampling frequency by the downsampling unit 13. It is converted into a digital signal having a second sampling frequency lower than this. In this case, the high-frequency component is removed by the low-pass filter processing so that the aliasing signal is not generated by sampling at the second sampling frequency.

The digital signal of the second sampling frequency is subjected to irreversible or lossless compression encoding in the encoding unit 14 and is output as the main code Im. The main code Im is decoded by the local decoding unit 15, and the decoded local signal is converted by the upsampling unit 16 from the local signal of the second sampling frequency to the local signal of the first sampling frequency. At that time, as a matter of course, an interpolation process is performed. An error signal in the time domain between the local signal of the first sampling frequency and the digital signal of the first sampling frequency branched from the frame dividing unit 12 is calculated by the error calculating unit 17.
The error signal is supplied to the prediction error generation unit 51 to generate a prediction error signal of the error signal.

The prediction error signal is rearranged in the compression encoder 18 and is directly or further losslessly compressed and output as an error code Pe. The main code Im and the error code Pe from the encoding unit 14 are combined by the combining unit 19, packetized, and output from the output terminal 21.
For the rearrangement of the bit string and the lossless compression encoding, refer to, for example, pages 6 to 8 and FIG.

In the decoder 30, the code from the input terminal 31 is separated into the main code Im and the error code Pe by the separation unit 32, and the main code Im is decoded by the decoding unit 33 corresponding to the coding unit 14 of the encoder 10. The decoded signal of the second sampling frequency is obtained by irreversible or lossless decoding by the processing. The decoded signal of the second sampling frequency is upsampled by the upsampling unit 34 and converted into a decoded signal of the first sampling frequency. At this time, as a matter of course, interpolation processing is performed in order to increase the sampling frequency.
The separated error code Pe is subjected to processing for reproducing a prediction error signal in the decoding unit 35. The specific configuration and processing of the decryption unit 35 are disclosed in the above publication, for example. The sampling frequency of the reproduced prediction error signal is the first sampling frequency.

This prediction error signal is predicted and synthesized by the prediction synthesis unit 63 to reproduce the error signal. The prediction synthesis unit 63 corresponds to the configuration of the prediction error generation unit 51 of the encoder 10.
The sampling frequency of the reproduced error signal is the first sampling frequency, and this error signal and the decoded signal of the first sampling frequency from the upsampling unit 34 are added by the adding unit 36 to reproduce the digital signal. , And supplied to the frame synthesis unit 37. The frame synthesizing unit 37 sequentially connects the digital signals reproduced for each frame and outputs them to the output terminal 38.
In the upsampling units 16 and 34 in FIG. 1, one or more zero-value samples are inserted for each predetermined number of samples so that the sample sequence of the decoded signal becomes the sample sequence of the first sampling frequency. A sample string in which a zero-value sample is inserted is passed through an interpolation filter (generally a low-pass filter) composed of, for example, an FIR filter shown in FIG. To do. That is, a delay unit D having the delay of the first sampling frequency as a delay amount is connected in series, and a zero-padded sample string x (n) is input to one end of the series connection. Multiplier 22 ₁ for each output of D
To 22 filter coefficients in _{m h 1, h 2, ...} , h m is multiplied by these multiplication results are being added to the filter output y (n) by an adder 23.

As a result, for example, with respect to the solid-line decoded signal sample sequence shown in FIG. 2B, the inserted zero-value sample becomes a sample having a value obtained by linear interpolation as shown by the broken line.
In such FIR filter processing, as shown in FIG. 2C, each sample x (n), (n = 0,. The output y (n) is obtained by realizing the process of convolving the coefficient h _n with 2T + 1 = m samples in total of T point samples, that is, the calculation of the following equation.

Therefore, the first output sample y (0) of the current frame depends on T samples from x (-T) to x (-1) of the previous frame. Similarly, the last output sample y (L-1) of the current frame depends on T values from x (L) to x (L + T-1) of the next frame. The multiplication unit in FIG. 2A is referred to as a filter tap, and the number m of the multiplication units 22 _{1 to} 22 _m is referred to as the tap number.
In the encoding / decoding system as shown in FIG. 1, most of the samples of the previous and subsequent frames are also known. However, packet loss and random access (reproduction from the middle of audio and image signals) in the transmission path For this reason, information may be required to be completed within a frame. In this case, the unknown values of the preceding and following samples can all be assumed to be 0, but the continuity and efficiency are reduced.

In addition, the prediction error generation unit 51 of the encoder 10 in FIG. 1 uses an input sample string x (n) (in this example, an error signal from the error calculation unit 17 in the autoregressive linear prediction as shown in FIG. 3A, for example). ) Is input to one end of the serial connection of the delay unit D having the sample interval as a delay amount, and is input to the prediction coefficient determination unit 53. The prediction coefficient determination unit 53 includes a plurality of past input samples and an output prediction error y. since the (n) as the prediction error energy is minimized, a set of linear prediction coefficients _{{α 1, ..., α p} } is determined for each sample, these prediction coefficients α _1, ..., α _p is the delay Each corresponding output of the part D is multiplied by the multipliers 24 _{1 to} 24 _p , and the multiplication results are added by the adder 25 to generate a predicted value. In this example, the integer unit 56 Sample with this integer value predicted signal input Is subtracted in al subtraction unit 57, prediction error signal y (n) is obtained.

In such an autoregressive prediction process, as shown in FIG. 3B, the sample at the point p before each sample x (n), (n = 0,..., L-1) in the frame consisting of L samples is used. On the other hand, a prediction value is obtained by convolving the prediction coefficient α _i , and the prediction value is subtracted from the sample x (n), that is, the calculation of the following equation is executed to obtain the prediction error signal y (n).

However, [*] represents the integerization of the value *, for example, rounded down. Therefore, the prediction error signal y (0) at the head of the current frame depends on p input samples from x (-p) to x (-1) of the previous frame. It should be noted that integerization is not necessary for encoding that allows distortion. Further, integerization may be performed during the calculation.
In the auto-regressive predictive synthesis, the predictive synthesis unit 63 of the decoder 30 in FIG. 1 receives the input sample sequence y (n) (in this example, reproduced by the uncompressed encoding unit 35 as shown in FIG. 4A, for example). (Prediction error signal) is input to the adder 65, and the predictive synthesized signal x (n) is output from the adder 65 as will be understood later, and the predicted synthesized signal x (n) is the sample period of the sample sequence. Is input to one end of a serial connection of a delay unit D having a delay amount as well as input to the prediction coefficient determination unit 66. The prediction coefficient determination unit 66 determines the prediction coefficients α ₁ ,..., Α _p so that the error energy between the prediction signal x ′ (n) and the prediction composite signal x (n) is minimized, and the output of each delay unit D alpha _1, corresponding to ..., alpha _p is multiplied by the multiplication unit 26 ₁ ~ 26 _p, the prediction signal these multiplication results are added by an adder 27 is generated. The prediction signal is converted into an integer value by the integer converting unit 67, and the prediction signal x (n) ′ having an integer value is added to the prediction error signal y (n) input by the adding unit 65, thereby generating a predicted combined signal x (n ) Is output.

In such an autoregressive predictive synthesis process, as shown in FIG. 4B, for each input sample y (n), (n = 0,..., L-1) in a frame composed of L samples, the previous p The predicted composite signal x (n) is obtained by adding the predicted value obtained by convolving the prediction coefficient α _i to the predicted composite sample of the points, that is, by executing the following equation.

Therefore, the prediction synthesis sample x (0) at the head of the current frame depends on p prediction synthesis samples from x (-p) to x (-1) in the previous frame.
As described above, the autoregressive prediction processing and prediction synthesis processing require the input sample of the previous frame and the prediction synthesis sample of the previous frame. For example, in the coding / decoding system as shown in FIG. When information is required to be completed within a frame due to random access, it is possible to assume all unknown values of the previous samples to be 0, but continuity and prediction efficiency are reduced.

Conventionally, in a voice packet transmission system in which a voice signal is packet-transmitted only in a voiced section, packet transmission is not performed in a silent section, and pseudo background noise is inserted in the silent section on the receiving side, the levels of the voiced and silent sections are Patent Document 2 proposes a technique for correcting discontinuity so as not to cause a sense of incongruity at the beginning or end of a conversation. This method inserts an interpolated frame between the speech frame decoded in the sound period and the pseudo background noise frame on the receiving side, and in the case of the hybrid coding method, the filter coefficient and the noise codebook index are present. The sound section is used, and the gain coefficient takes an intermediate value of the background noise gain.
In the above-mentioned Patent Document 2, only a sound section is transmitted, and the beginning and end of the sound section are each processed in a state where there is no previous frame and no subsequent frame.
JP 2001-144847 JP 2000-307654 JP T.Moriya and 4 other authors “Sampling Rate Scalable Lossless Audio coding” 2002 IEEE Speech Coding Workshop proceedings 2002, October

  In processing for each frame, when using a processing method that improves continuity, quality and efficiency by processing the current frame using the sample before the current frame or the sample after the current frame, the receiving side (decoding side) ) To prevent degradation of continuity, quality, and efficiency even when the previous frame and the subsequent frame cannot be obtained, or even if only one frame is processed independently from the other frames It is desirable to obtain continuity, quality, and efficiency that are close to the same level as when they exist. Such signal processing is a part of encoding processing when encoding or transmitting or storing a digital signal for each frame, or decoding processing of a code received and received or a code read from a storage device. The present invention is not limited to the case where it is used for the processing of a part, but generally, the present invention is applied to a process for improving the quality and efficiency by using the sample of the previous frame and the subsequent frame in the process of a digital signal frame unit. It can be done.

  In other words, the object of the present invention is to perform processing for performing a digital signal in units of frames, using only the samples of that frame, and performing the same performance (continuity, quality, It is an object of the present invention to provide a digital signal processing method and its program that can obtain efficiency and the like.

The digital signal processing method according to the invention of claim 1
Used to encode the original digital signal in units of frames, the first sample series of the frame, the last sample series of the frame before the frame, the last sample series of the frame, the beginning of the frame after the frame A digital signal processing method for performing processing depending on at least one of the sample sequences,
A step of using an auxiliary code obtained by encoding the at least one sample series separately from the encoding for the frame as a part of the code of the frame.
According to a fifth aspect of the present invention, there is provided a digital signal processing method.
(a) Decoding the auxiliary code of the frame, the first sample sequence of the frame, the last sample sequence of the previous frame, the last sample sequence of the frame, the first sample sequence of the frame after the frame Determining at least one sample sequence of
(b) processing the at least one sample sequence for the frame as a decoded sample sequence at the end of a frame preceding the frame or a decoded sample sequence at the beginning of a frame after the frame;
Including.

A program for causing a computer to execute each step of the digital signal processing method according to the present invention is also included in the present invention.
The present invention also includes a readable recording medium in which a program capable of executing each step of the digital signal processing method according to the present invention by a computer is recorded.

As an effect common to the present invention, the processing can be completed within the frame while maintaining almost the continuity and efficiency in the case where the frame exists in the previous and / or subsequent frame. Therefore, it is possible to improve the performance when random access is required in units of frames or when a packet is lost.
According to the first aspect of the present invention, the sample sequence obtained as the auxiliary method is immediately used as a substitute sample sequence when there is a frame loss on the receiving side by separately preparing the head sample sequence or the tail sample sequence as auxiliary information. Can be used.
According to the invention of claim 5, random access to the frame is facilitated by immediately using the first sample sequence received as auxiliary information or the last sample sequence of the previous frame as a substitute sample sequence.

First Reference Embodiment Before describing the present invention, a reference example of a digital signal processing method related to the present invention will be described.
The first reference embodiment Figure 5A, as shown in FIG. 5B, for example, a digital signal of 1 frame stored like the buffer 100 (sample sequence) sample sequence ΔS that part of the continuous in S _FC is, that the buffer 100 The sample sequence ΔS is read by the substitute sample sequence generation unit 110 without being erased, and the sample sequence ΔS is processed as it is or as necessary, and is generated as a substitute sample sequence AS. Are connected by the sample string connection unit 120 before the first sample of the current frame FC and after the last sample of the current frame FC in the buffer 100, respectively, and this connected sample string PS (= AS + S _FC + AS, hereinafter processing sample string) Is called a linear combination processing unit 1 such as an FIR filter from the top of the substitute sample sequence AS. 30 and is subjected to linear combination processing. Of course, it is not necessary to connect the substitute sample sequence AS directly to the current frame in the buffer 100 in advance to form a series of processing sample sequences. The data may be stored and read out sequentially in the order of the sample sequences AS, S _FC , AS and supplied to the FIR filter.

The alternative sample sequence AS to connect after the last sample of the frame as indicated by the broken line in FIG. 5B, alternative sample using a sample sequence [Delta] S 'of consecutive that differ from the parts sample [Delta] S in the current frame digital signal S _FC You may connect as row | line AS '. Depending on the processing contents of the linear combination processing unit 130, the substitute sample string AS may be connected only before the first sample or only after the last sample.
The linear combination processing unit 130 needs a sample of the previous frame or a sample of the subsequent frame, but copies a part of the sample sequence in the current frame in place of the required sample sequence of the previous and subsequent frames. by using this as alternative sample sequence, it is possible to obtain a sample sequence S _FC only one frame of the processed digital signal (sample sequence) S _OU of the current frame without using the samples before and after the frame. In this case, compared with the case of processing for generating the alternative sample sequence from a portion sample sequence in the sample sequence S _FC of the current frame is simply the previous frame, the portion of the alternative sample sequence after 0, continuity, quality, Efficiency is improved.

Reference Example 1 in which the first reference form is applied to the FIR filter processing shown in FIG. 2A will be described.
The buffer 100 in FIG. 6A digital signal (sample sequence) S _FC of the current one frame shown in FIG. 6B are stored. Each sample of the digital signal S _FC x (n), and (n = 0, ..., L -1). The reading unit 141 in the substitute sample sequence generation / connection unit 140 uses the T samples from the second sample x (1) to x (T) from the beginning of the current frame FC as a part of continuous sample sequence ΔS. The T sample sequences ΔS read from the buffer 100 are replaced by sample sequences x (T),..., X (2), x (1) whose arrangement order is reversed by the reverse order arrangement unit 142. It is generated as a sample string AS. The alternative sample sequence AS is stored by the write unit 143 into the buffer 100 as lead before the first sample x of a frame FC of the digital signal S _FC in the buffer 100 (0).

Further, the reading unit 141 causes the T samples from the sample x (LT-1) T-1 before the last sample x (L-1) to the sample x (L-2) one before x (L-1). Are read from the buffer 100 as a part of the continuous sample sequence ΔS ′, and the sequence order of the sample sequence ΔS is reversed by the reverse order arrangement unit 142, and x (L−2), x (L−3),. x (LT-1) is generated as a substitute sample sequence AS ', and the substitute sample sequence AS' is stored by the writing unit 143 so as to be connected after the last sample x (L-1) of the current frame in the buffer 100.
Thereafter, a processing sample string x (-T),..., X (-1), x (0), x (1), from n = -T to n = L + T-1 is read from the buffer 100 by the reading unit 141. ..., x (L-2), x (L-1), x (L), ..., x (L + T-1) are read out and supplied to the FIR filter 150. As a result of the filtering process, y (0),..., Y (L-1) are output. In this example, the substitute sample sequence AS is arranged symmetrically with respect to the head sample x (0), and the samples in the frame FC are symmetrically arranged. Similarly, the substitute sample sequence AS ′ is within the frame FC with respect to the end sample x (L−1). These samples are symmetrically arranged, and these portions are symmetrical with respect to the first sample x (0) and the last sample x (L-1), respectively. As compared with the case where AS ′ is set to 0, filter processing outputs y (0),..., Y (L−1) are obtained with less disturbance of the frequency characteristics and less errors with respect to the case where there are frames before and after that.

6A is obtained by multiplying the substitute sample AS by a window function ω (n) whose weight decreases as it is ahead of the first sample x (0), for example, by the windowing unit 144 indicated by a broken line in FIG. 6A. In the same manner, the substitute function AS ′ multiplied by the window function ω (n) ′ whose weight becomes smaller as it comes after the last sample x (L−1) may be used. .
For the substitute sample AS ′, if it is performed on the sample sequence ΔS ′ before the window functions are arranged in reverse order, ω (n) can be used as the window function.
In the configuration of FIG. 6A, a processing sample sequence PS to which substitute sample sequences AS and AS ′ are added to the current frame in the buffer 100 is generated in the buffer 100, and the generated processing sample sequence PS is sequentially read from the head. In this case, the supply to the FIR filter 150 is shown. However, as is apparent from the above description, the substitute sample strings AS, AS ′ and the current frame sample string S _FC generated from the partial sample strings in the current frame are successively sequentially arranged in the order of AS, S _FC , AS ′. Therefore, even if the processing sample sequence PS to which the substitute sample sequences AS and AS ′ are added is not generated in the buffer 100, the partial sample sequence ΔS, the current frame sample sequence S _FC , and the partial sample sequence can be obtained. Samples may be taken from the current frame FC one by one in the order of ΔS ′ and supplied to the FIR filter 150.

  That is, for example, as shown in FIG. 7, n = −T is initially set (S1), x (−n) is read from the buffer 100, and x (n) as it is or multiplied by the window function ω (n) as necessary. ) Is supplied to the FIR filter 150 (S2), and it is checked whether n = -1 (S3). If not, n is incremented by 1 and the process returns to step S2 (S4). If n = -1, n is incremented by 1 (S5), x (n) is read from the buffer 100, supplied to the FIR filter 150 (S6), and it is checked whether n = L-1 or not. If not, the process returns to step S5 (S7). If n = L-1, n is incremented by 1 (S8), x (2L-n-2) is read from the buffer 100, and the window is read as it is or as necessary. The function ω (n) ′ is multiplied and supplied to the FIR filter 150 as x (n) (S9), and it is checked whether n = L + T−1. If not, the process returns to step S8, and n = L + If it is T-1, the process ends (S10).

Reference Example 2 in which the first reference embodiment is applied to FIG. 2A will be described. This is connected to the front sample x (0) and the last sample x (L-1) of the frame FC by using a part of the continuous sample sequence ΔS in the current frame FC.
That is, as shown in FIG. 8A, a part of continuous sample strings x (τ),..., X (τ + T−1) in the frame FC are read from the buffer 100 of FIG. 6A, and this sample string ΔS is read as a substitute sample string. The AS is stored in the buffer 100 so as to be connected before the first sample x (0), and the sample sequence ΔS is stored in the buffer 100 as the substitute sample sequence AS ′ so as to be connected after the last sample x (L−1). That is, in the substitute sample string generation / connection unit 140 of FIG. 6A, the output of the reading unit 141 is immediately supplied to the writing unit 143 as indicated by a broken line. In this method, it can be said that the copy of the partial sample sequence ΔS is shifted forward by τ + T + 1 to obtain the substitute sample sequence AS, and the copy of ΔS is shifted backward by L−τ to obtain the substitute sample AS ′. In this case, the window function ω (n) may be multiplied by the windowing unit 144 and the window function ω (n) ′ may be multiplied by the substitute sample string AS ′. Alternative sample sequences AS, sample sequence S _FC frame FC to AS 'was linked is being read supplied to the FIR filter 150 from the beginning of the alternative sample sequence AS, filtering result y (0), ..., y (L-1 )

As shown in FIG. 8B, the substitute sample string AS is connected before the first sample x (0) in the same manner as shown in FIG. 8A, and then x (τ ₁ ),..., X (τ ₁ + A sample sequence x (τ ₂ ),..., X (τ ₂ + T-1), which is part of a portion different from T-1), is extracted as a sample sequence ΔS ′, and this is used as a substitute sample sequence AS ′. It may be connected after the sample x (L-1). In this case, the substitute sample string AS ′ multiplied by the window function ω (n) ′ may be used.
In the case of this reference example 2, it is also possible to take out one sample at a time from the buffer 100 and supply it to the FIR filter 150. For example, as shown in parentheses in step S2 of FIG. 7, x (n + τ) is used as x (n) in the case of FIG. 8A and x (n + τ ₁ ) in the case of FIG. 8B, and in step S9. As shown in parentheses as x (n), x (n + τ ₁ ) may be used in the case of FIG. 8A and x (n + τ ₂ ) in the case of FIG. 8B.

Thus using only the sample sequence S _FC of Example one in 1 frame, the previous, it is possible to perform digital processing that requires some samples of subsequent frame, continuity, quality, Efficiency is improved.

The reference example 3 of the first reference form is the most preferable substitute by changing the extraction position of the partial sample sequence ΔS (or ΔS, ΔS ′) in the case of the generation method of various substitute sample sequences determined in advance or in the case of the reference example 2. Auxiliary information indicating one of the methods for generating the sample or / and auxiliary information indicating the extraction position of the sample sequence ΔS are output. This reference example is applied to the encoding / decoding system shown in FIG. 1, for example. A method for selecting the position will be described later.
For example, the following method can be considered as a method for generating a substitute sample string.
1. 1. Change τ in FIG. 8A of Reference Example 2, no window function 2. Change τ in FIG. 8A of Reference Example 2, no window function, reverse order arrangement. FIG. 8A of Reference Example 2 changes τ and has a window function. 4. Change τ in FIG. 8A of Reference Example 2, with window function, reverse order arrangement 5. Change τ ₁ and τ ₂ in FIG. 8B of Reference Example 2, no window function In Figure 8B tau ₁ of Reference Example 2, changing the tau _2, no window function, reverse arrangement 7. Change τ ₁ and τ ₂ in FIG. 8B of Reference Example 2, with window function. ₁ tau in Figure 8B of Reference Example 2, changing the tau _2, window function used, reverse arrangement 9. No window function in Reference Example 1 10. 10. There is a window function in Reference Example 11. 8. In FIG. 8A of Reference Example 2, τ is fixed and there is no window function. 8. τ fixed in FIG. 8A of Reference Example 2, no window function, reverse order array FIG. 8A of Reference Example 2 has τ fixed and window function. 8. τ fixed in FIG. 8A of Reference Example 2, with window function, reverse order array In FIG. 8B of Reference Example 2, τ ₁ and τ _{2 are} fixed, and no window function is provided. In FIG. 8B of Reference Example 2, τ ₁ and τ ₂ fixed, no window function, reverse order arrangement In FIG. 8B of Reference Example 2, τ ₁ and τ _{2 are} fixed, and there is a window function. In FIG. 8B of Reference Example 2, τ ₁ , τ ₂ fixed, with window function, reverse order arrangement Since methods 9 and 10 are included in methods 6 and 8, respectively, methods 9, 10 and methods 6 and 8 are selected at the same time. Never do. In general, since the alternative pulse trains of the methods 1 to 4 can be obtained better than the methods 11 to 14, they are not simultaneously selected. Similarly, the methods 5 to 8 and the methods 15 to 18 are not selected simultaneously. Accordingly, for example, one or more of the methods 1 to 8 are selected, or one or more of the methods 1 to 4 and any one of 9 and 10 are selected. As previously determined. In some cases, only one of the methods 1 to 8 is selected.

These predetermined generation methods are stored in the generation method storage unit 160 in FIG. 9A, and one of the substitute sample string generation methods is read from the generation method storage unit 160 under the control of the selection control unit 170 and is used as a substitute sample. Set in the column generation unit 110, the substitute sample sequence generation unit 110 starts operation, and extracts some continuous sample sequences ΔS in the current frame FC from the buffer 100 according to the set generation method, and substitute samples A sequence (candidate) is generated, and the candidate substitute sample sequence is supplied to the selection control unit 170.
The selection control unit 170 calculates the similarity between the candidate substitute sample sequence in the current frame FC and the corresponding sample sequence in the previous frame FB or the sample sequence in the next frame FF in the similarity calculation unit 171. In the similarity calculation unit 171, for example, as shown in FIG. 9B, in the previous frame FB used for FIR filter processing (for example, FIR processing executed in the upsampling unit 16 in FIG. 1) across the sample of the current frame FC. , X (-1) are stored in advance in the register 172 from the buffer 100 and used for FIR filter processing across the sample of the current frame FC in the next frame FF. First sample strings x (L),..., X (L + T−1) are stored in advance in the register 173 from the buffer 100.

If the input candidate substitute sample sequence is an AS for the sample sequence of the previous frame, it is stored in the register 174, and this sample sequence AS and the sample sequence x (-T), ..., x (-1) in the register 172 are stored. Is calculated by the distortion calculation unit 175. If the input candidate substitute sample is AS ′ for the sample sequence of the next frame, it is stored in the register 176, and this sample sequence AS ′ and the sample sequence x (L),..., X (L + T− in the register 173 are stored. The distortion error unit 175 calculates the square error with 1).
It can be said that the smaller the calculated square error (or weighted square error), the smaller the distortion of the candidate substitute sample sequence, that is, the higher the similarity with the last sample sequence of the previous frame or the first sample sequence of the next frame. The similarity may be determined by calculating the inner product (or cosine) of the sample sequence vectors corresponding to the current frame of each candidate substitute sample sequence, and the greater the value, the higher the similarity may be. In any of the methods 1 to 8, the position τ ₁ , τ ₂ is changed to τ = 0,..., L−1, for example, and the sample string at the position where the similarity is the maximum is the maximum similarity by the method. Candidate substitute sample string. When a plurality of methods 1 to 8 are selected as methods to be used, the candidate substitute sample sequence having the maximum similarity is selected from the candidate substitute sample sequences having the maximum similarity according to the selected method. To do.

Thus the highest alternative sample sequence similarity in the alternative sample sequence obtained in various ways AS, and supplies the AS 'to the FIR filter 150 by connecting after previous sample sequence S _FC of the current frame FC,. Further, information AI _AS indicating the method used to generate the substitute sample sequences AS and AS ′ adopted, and in the case of methods 1 to 8, the position τ (or τ ₁ and τ of the sample sequence ΔS (or this and ΔS ′) taken out) ₂ ) When only one of the auxiliary information AI including information AI _P and methods 1 to 8 is used, only the information AI _P is generated by the auxiliary information generation unit 180, and auxiliary information AI is supported as necessary. The information encoding unit 190 encodes the auxiliary code _CAI . For example auxiliary information AI or e Kuwawa the auxiliary code C _AI part of the code of the frame FC generated in the encoder 10 shown in FIG. 1, for transmission or recording.

If τ (or τ ₁ , τ ₂ ) is fixed in Reference Example 1 or Reference Example 2, it is not necessary to output auxiliary information if the decoding side is informed beforehand.
A processing procedure of the processing method shown in FIG. 9A will be described with reference to FIG.
First, a parameter m specifying a generation method is initialized to 1 (S1), the method m is read from the storage unit 160 and set in the substitute sample sequence generation unit 110 (S2), and the substitute sample sequence (candidate) AS, AS 'Is generated (S3). The similarity E _m between the substitute sample sequence AS, AS ′ and the previous frame sample sequence and the next frame sample sequence is obtained (S4), and it is checked whether the similarity E _m is higher than the maximum similarity E _M so far. (S5) If it is higher, E _M is updated to that E _m (S6), and the substitute sample sequence AS (or this and AS ′) stored in the memory 177 (FIG. 9A) is used as the substitute sample sequence (candidate). The update is saved at (S7). The memory 177 is also a maximum of similarity E _M until it has been saved.

If E _m is not larger than E _M in step S5, and after step S7, it is checked whether m = M (S8). If not, m is incremented by 1 in step S9 and the process returns to step S3. Move on to generation of substitute sample sequence by. If m = a M in step S8, then the Save and are alternative sample sequence AS (or AS and AS ') in front of the sample sequence S _FC of the current frame FC, connect after (S10), which FIR filtered (S11), and generates a supplementary information AI indicating information AI _AS / or position information AI _P shows the generation method of the adopted alternative sample sequences (S12).

In the methods 1 to 8 for changing the position τ or τ ₁ , τ ₂ , the generation of the substitute sample string having the highest similarity can be obtained in the same manner as steps S1 to S9 shown in FIG. For example, in the case of methods 1 to 4, for each m, as shown in parentheses in FIG. 10, τ = 0 is initially set in step S1, m is set in step S2, and a substitute sample string is generated in step S3. In step S4, the similarity Eτ is calculated. In step S5, it is checked whether it is greater than Eτ _M. If it is larger, Eτ _M is updated with Eτ in step S6, and the substitute sample string is updated and stored in step S7. If it is not, it is checked, and if not, τ is incremented by 1 in step S9 and the process returns to step S3. If τ = L−T + 1 in step S8, it is saved if M = 1 in step S10 When a certain substitute sample sequence AS is adopted and there are a plurality of _M , Eτ _M stored at that time is set as the similarity E _{m of the} method m.

Such a sample string S _FC of the current frame FC to the, to produce the most preferred alternative sample sequence, for outputting the auxiliary information AI as part of the code of the frame FC, decodes the code of this frame In the case where the processing of the digital signal necessary for the decoding requires samples of the previous (past) and subsequent (future) frames (for example, the upsampling unit 34 of the decoder 30 in FIG. 1), it is obtained during the decoding. Samples obtained by extracting some continuous sample sequences from the sample sequence S _FC (decoded) of the received frame FC by the method instructed by the auxiliary information AI and generating substitute sample sequences AS, AS ′ previous column S _FC, by connecting later by performing the digital signal processing, decoding the digital signal of one frame in only the sign of 1 frame (reproduction) child In addition, continuity, quality, and efficiency are improved.

This reference example is used for, for example, a part of encoding of a digital signal. A sample string similar to the head part (head sample string) in a frame is taken out from the frame, and a gain (gain 1 is included) in the similar sample string. ) Is subtracted from the first sample sequence, and a prediction error signal is generated from the sample sequence of the frame by autoregressive prediction, thereby preventing a decrease in prediction efficiency due to discontinuity. The smaller the prediction error, the better the prediction efficiency.
The reference example 4 is applied to, for example, the prediction error generation unit 51 in the encoder 10 of FIG. FIG. 11 shows an example of the functional configuration, FIG. 12 shows an example of a sample sequence in each process, and FIG. 13 shows an example of the flow of processing.

A digital signal (sample sequence) S _FC = {x (0),..., X (L-1)} of one frame FC to be processed is stored in, for example, the buffer 100 in FIG. 210, the sample sequence x (n + τ),..., X (n + τ + p-1) similar to the first sample sequence x (0),..., X (p-1) in the frame FC is buffered. read from sample sequence S _FC of the frame FC in 100 (S1). This similar sample sequence x (n + τ),..., X (n + τ + p-1) is converted into a similar sample sequence u (0),..., U (p-1) as shown in FIG. Shifting to the head position in the frame FC, the gain applying unit 220 multiplies the similar sample sequence u (n) by the gain β (0 <β ≦ 1) to obtain a sample sequence u (n) ′ = βu (n). (S2) The sample sequence u (n) ′ is subtracted from the sample sequence x (0),..., X (L−1) of the frame FC by the subtractor 230, and the result is sampled as shown in FIG. A sequence v (0),..., V (L-1) is set (S3). That is
n = 0,…, p-1 and v (n) = x (n) −u (n) ′
n = p,…, L-1 and v (n) = x (n)
And After multiplying x (n + τ),..., x (n + τ + p−1) by the gain β, the sample sequence may be shifted to the head position in the frame to be a sample sequence u (n) ′.

The p (predicted order) substitute sample sequences v (-p), ..., v (-1) are connected by the substitute sample sequence adding unit 240 as shown in FIG. 12 before the first sample v (0) ( S4). The substitute sample string v (-p), ..., v (-1) is 0, ..., 0, a fixed value d, ..., d, or the technique similar to the substitute sample string AS obtained in the first reference form. The obtained p sample strings may be used.
A sample sequence v (-p),..., V (L-1) connected with the substitute samples is input to the prediction error generator 51, and the prediction error signal y (0),. -1) is generated (S5).
The determination of the similar sample sequence x (n + τ), ..., x (n + τ + p-1), the determination of the gain β, for example, the power of the prediction error signal y (0), ..., y (L-1) Τ and β are determined so that is minimized. The calculation of the power of this error is such that the prediction error power is x (n + τ),..., X (n + τ) after the p samples after v (p) are used for the calculation of the predicted value. Since it does not matter from which part + p-1) is selected, the error power up to the prediction error signal y (2p) may be used to determine τ and β. In addition, the determination method is similar to the determination method of the substitute sample sequence AS described with reference to FIG. 10, and in this case, the error power calculation unit 250 (FIG. 11) calculates the error power each time τ is changed. When the error power is smaller than the minimum value p _EM so far, the error power is stored and updated in the memory 265 as the minimum value p _EM , and the similar sample sequence at that time is updated and stored in the memory 265. Further, the error power is obtained by changing τ ← τ + 1 and next τ, and if the error power is not small, the similar sample sequence at that time is updated and stored in the memory 265, and τ is changed from 1 to L-1-p. A similar sample sequence stored when the change is completed is adopted. Next, β is changed for this similar sample sequence, and the error power is calculated each time, and β when the error power is minimum is adopted. Such τ and β are determined under the control of the selection determination control unit 260 (FIG. 11).

A prediction error signal is generated for the sample sequence v (-p),..., V (L-1) generated using τ and β determined in this way, and auxiliary information representing τ and β used at that time the AI generated by the auxiliary information generation unit 270 (S6), the auxiliary information AI is encoded into code C _AI auxiliary information encoding unit 280 if necessary. The auxiliary information AI or the code _CAI is added to a part of the encoded code for the input digital signal of the frame FC by the encoder.
In the above, the value of τ is preferably larger than the predicted order p, and the sum ΔU + τ of the lengths ΔU and τ of the similar sample sequence u (n) is L−1 or less, that is, x (τ + ΔU) is the frame FC. Τ may be determined within a range that does not deviate from. The length ΔU of the similar sample sequence u (n) may be equal to or less than τ, and may be equal to or less than p, which is not related to the predicted order p, but is preferably equal to or greater than p / 2. Furthermore, the head position of the similar sample sequence u (n) may not necessarily coincide with the head position in the frame FC, that is, u (n) may be set to n = 3,..., 3 + ΔU, for example. The gain β applied to the similar sample sequence u (n) may be weighted depending on the sample, that is, u (n) may be multiplied by a predetermined window function ω (n). It only needs to represent τ.

A reference example of the prediction synthesis processing method corresponding to the reference example 4 will be described as a reference example 5. This predictive synthesis processing method is used in a part of the process of decoding the encoded code of the digital signal for each frame, for example, the predictive synthesis unit 63 in the decoder 30 in FIG. Even when decoding is performed from this frame, a decoded signal with good continuity and quality can be obtained. FIG. 14 shows an example of the functional configuration of the reference example 5, FIG. 15 shows an example of the sample sequence during the process, and FIG. 16 shows an example of the processing procedure.
A sample sequence y (0),..., Y (L-1) of the current frame FC of a digital signal (prediction error signal) to be subjected to prediction synthesis processing by autoregressive prediction is stored in, for example, the buffer 100 and read. The writing unit 310 reads the sample string y (0),..., Y (L−1).

  On the other hand, a substitute sample sequence AS = {v (−p),..., V (−1)} having the same length p as the predicted order p is generated from the substitute sample sequence generation unit 320 (S1). As the substitute sample string, predetermined ones such as 0,..., 0, fixed values d,..., D, and other predetermined sample strings are used. This substitute sample string v (-p), ..., v (-1) is sequentially applied from the head sample v (-p) to the prediction synthesis unit 63 at the last p samples of the prediction error signal of the frame immediately before the current frame FC. (S2), and subsequently, the sample sequence y (0),..., Y (L-1) to be subjected to the prediction synthesis process is sequentially supplied from the head to the prediction synthesis unit 63 to perform the prediction synthesis process. Then, a predicted synthesized signal v (n) (n = 0,..., L-1) is generated (S3). This predicted synthesized signal v (n) ′ is temporarily stored in the buffer 100.

The auxiliary information decoding unit 330 decodes the auxiliary code _CAI as a part of the code of the current frame FC to obtain auxiliary information, thereby obtaining τ and β (S4). The auxiliary information decoding unit 330 may receive the auxiliary information itself. The sample sequence acquisition unit 340 uses τ to determine a predetermined number from the composite signal (sample) sequence v (n), in this example, a sample sequence v (τ),. τ + p) is replicated, that is, v (τ),..., v (τ + p) is obtained with the predicted synthesized signal sequence v (n) as it is (S5). A sample sequence u (n) is shifted to the beginning position, and a gain β from the auxiliary information is multiplied by a gain applying unit 350 to generate a corrected sample sequence u (n) ′ = βu (n) (S6).

This corrected sample sequence u (n) ′ is added to the predicted synthesized sample (signal) sequence v (n) and output as a normal predicted synthesized signal x (n) (n = 0,..., L−1). (S7). The predicted composite sample sequence x (n) is
n = 0,…, p-1 and x (n) ＝ v (n) ＋ u (n) ′
n = p,…, L-1 and x (n) = v (n)
It is. As described above, the control unit 370 of the processing unit 300 performs control for causing each unit to execute processing.
In this way, a predictive synthesized signal with excellent continuity and quality can be obtained only from the frame FC. Since the reference example 5 corresponds to the reference example 4, the length ΔU of the correction sample sequence u (n) ′ is not limited to p, that is, it is irrelevant to the predicted order and is determined in advance. In addition, the position of the first sample of the corrected sample sequence u (n) ′ does not necessarily coincide with the first sample v (0) of the composite signal v (n), and this is also determined in advance. Furthermore, the gain β is not included in the auxiliary information, and may be weighted for each sample u (n) by a predetermined window function ω (n).
Second Reference Mode In the second reference mode of the present invention, samples x (1), x (2),... Before (past) the first sample x (0) of the frame or the last sample x (L- 1) The frame using the number of filter taps or the predicted order that depends only on the available sample (within the frame) without using the samples x (L), x (L + 1), ... after (future) The digital signal is processed.

Reference Example 6 will be described in which the second reference form is applied when performing autoregressive prediction. First, the case where the reference example 6 is applied to the process for obtaining the prediction error shown in FIG. 3A will be described with reference to FIG.
The prediction coefficient estimator 53 uses the current frame samples x (0),..., X (L-1) in the buffer in advance to obtain a primary prediction coefficient {α ⁽¹⁾ ₁ } and a secondary prediction coefficient {α ⁽²⁾ ₁ , α ⁽²⁾ ₂ }, ..., p-th order prediction coefficients {α ^(p) ₁ , ..., α ^(p) _p } are calculated in advance.
The first sample x (0) of the current frame FC is output as it is as the prediction error signal y (0).

For the next sample x (1), the primary prediction coefficient α ⁽¹⁾ ₁ from the prediction coefficient estimation unit 53 is used, and the product of this and x (0) is obtained by the calculation unit M ₁ to obtain the predicted value The prediction error signal y (1) is obtained by subtracting this prediction value from x (1).
When the next sample x (2) is input, the second-order prediction coefficients α ⁽²⁾ ₁ and α ⁽²⁾ ₂ from the prediction coefficient estimation unit 53 are used, and x (0) and x (1 ) And a convolution operation α ⁽²⁾ ₁ x (1) + α ⁽²⁾ ₂ x (0) is performed by the arithmetic unit M ₂ to obtain a prediction value, and the prediction value is subtracted from x (2) to obtain a prediction error. Find the signal y (2).
Hereafter, every time a sample is input, the prediction coefficient is incremented by one, and a prediction value is obtained by performing a convolution operation with this prediction coefficient and the past sample, and the prediction value is input at that time. Subtract from the sample to obtain the prediction error signal.

That is, on the encoding side (transmission side), although the previous frame FB of the frame FC exists, the sample of the previous frame is not used and the first sample (n = 0) of the current frame FC x (0 ) Is output as y (0) = x (0) without performing linear prediction. The second sample x (1) to the pth sample x (p-1) are nth order from the sample x (0),…, x (n) (n = 1, ..., p-1) coefficient prediction α ⁽ⁿ⁾ _1, ..., determine the alpha ⁽ⁿ⁾ _n convolution operation to the predicted value x (n) '. P samples x (np), ..., x (n-1) after the (p + 1) th sample x (p) of the current frame (n = p + 1, p + 2, ..., L-1) prediction coefficient alpha ^(p) ₁ of order p relative, ..., using the alpha ^(p) _p, obtains the convolution operation to the predicted value x (n) '. That is, the predicted value is obtained by a method similar to the conventional method. Note that the calculation of the p-th order prediction coefficient α ^(p) ₁ ,..., Α ^(p) _p in step S7 is performed in step S0 indicated by a broken line block. The next prediction coefficient may be calculated. Alternatively, n-th order (n = 1,..., P−1) prediction coefficients may be calculated in the process of calculating the p-th order prediction coefficient in step S0. Also, the calculated p-th order prediction coefficient is encoded and transmitted to the receiving side as auxiliary information.

ｎがｐになったかを調べ（Ｓ６）、なっていなければステップＳ３に戻りｐになっていれば、全サンプルx(0), …, x(L-1)から次数ｐの予測係数α^(p) ₁, …, α^(p) _pを求め（Ｓ７）、この予測係数を直前のｐ個の過去のサンプルx(n-p), …, x(n-1)に畳み込み演算して予測値を求め、これを現サンプルx(n)から減算して予測誤差信号y(n)を求める（Ｓ８）。つまり式（２）を演算する。処理すべきサンプルが終了したかを調べ（Ｓ９）、終了していなければｎを＋１してステップＳ８に戻り（Ｓ１０）、終了していれば処理を終りにする。 An example of this processing procedure is shown in FIG. First, n is initialized to 0 (S1), the sample x (0) is set as the prediction error signal y (0) (S2), n is incremented by 1 (S3), and all samples x (0),. The prediction coefficient α ⁽ⁿ⁾ ₁ ,..., Α ⁽ⁿ⁾ _n of degree n is obtained from (L-1) (S4), and the prediction coefficient is convolved with the samples x (0), ..., x (n-1). The calculation is performed, and the result is subtracted from the current sample x (n) fetched to obtain the prediction error signal y (n) (S5). That is, the following calculation is performed.

It is checked whether n has become p (S6). If not, the process returns to step S3, and if it is p, the prediction coefficient α ^{(of the} order p from all the samples x (0),..., x (L-1). ^p) ₁ , ..., α ^(p) _p is obtained (S7), and the prediction value is calculated by convolving the prediction coefficient with the previous p previous samples x (np), ..., x (n-1). Then, this is subtracted from the current sample x (n) to obtain a prediction error signal y (n) (S8). That is, the equation (2) is calculated. It is checked whether the sample to be processed is completed (S9). If not completed, n is incremented by 1 and the process returns to step S8 (S10). If completed, the process ends.

FIG. 19 shows prediction coefficients α ⁽ⁿ⁾ ₁ ,... Generated for each sample number n = 0,..., L−1 of the current frame to be used when Reference Example 6 is applied in FIG. α ⁽ⁿ⁾ _n is shown in a table. No prediction is performed for the sample x (0) of the first sample number n = 0 of the current frame. For each sample x (n) from the next sample number n = 1 to n = p−1, n-th order prediction coefficients α ⁽ⁿ⁾ ₁ ,..., Α ⁽ⁿ⁾ _n are set, and the rest ( pn) coefficients are set to α ⁽ⁿ⁾ _{n + 1} = α ⁽ⁿ⁾ _{n + 3} = ... = α ⁽ⁿ⁾ _p = 0. For each sample x (n) of n = p,..., L−1, p-th order prediction coefficients α ^(p) ₁ ,..., α ^(p) _p are calculated and set.
In order to perform p-th order linear prediction, the past p samples are required, so the samples x (0), ..., x (p-1) at the beginning of the frame are used for the prediction process. Although the rear end sample of the previous frame is required, as in Reference Example 6, from the sample number n = 0 to n = p−1, the prediction order is sequentially increased from 0 to p−1, and the sample number n = After p, by performing p-th order prediction (thus, even if the prediction process is performed without using the sample of the previous frame), the discontinuity of the prediction signals of the previous frame and the current frame can be reduced.

FIG. 20 shows a reference example 7 of the prediction synthesis process (reference example 6 is applied to FIG. 4A) corresponding to FIG. The prediction coefficient decoding unit 66 decodes the p-th order prediction coefficient from the received auxiliary information, and further calculates the n-th order prediction coefficient (n = 1,..., P−1) from the p-order prediction coefficient. When the first prediction error signal y (0) is input from the prediction error signal y (0),..., Y (L-1) of the current frame FC, this is directly used as the prediction composite signal x (0). When the next prediction error signal y (1) is input, α ⁽¹⁾ ₁ x (0) is calculated from the primary prediction coefficient α ⁽¹⁾ ₁ and x (0) obtained from the prediction coefficient decoding unit 66D. It calculates and in parts M ₁ obtains the predicted values, by adding this and y (1) and the synthesized signal x (1).

を行って予測値を求め、この予測値をy(n)と加算して予測合成信号x(n)を生成する。n=p以後は従来と同様に、つまり直前のｐ個の予測合成信号x(n-p), …, x(n-1)に対しｐ次の予測係数を式(3)により畳み込み演算し、y(n)と加算して予測合成信号x(n)を求める。この予測合成においても、予測係数は現フレームのサンプルy(n), n=0, ..., L-1, の入力に対し図１９の表で示した予測係数を設定することにより、前後フレームに跨らず、現フレーム内での予測合成を行うことができる。 When the next prediction error signal y (2) is input, the secondary prediction coefficients α ⁽²⁾ ₁ and α ⁽²⁾ ₂ from the prediction coefficient decoding unit 66D are calculated as x (0) and x (1). A convolution operation is performed in the part M ₂ to obtain a predicted value, and this predicted value and y (2) are added to obtain a composite signal x (2). Similarly, when y (n) is input until n = p, the nth-order prediction coefficients α ⁽ⁿ⁾ ₁ ,…, α ⁽ⁿ⁾ _n are changed to x (0),…, x (n− 1) Convolution operation

To obtain a predicted value and add this predicted value to y (n) to generate a predicted synthesized signal x (n). After n = p, the p-th order prediction coefficient is convolved with the previous p prediction composite signals x (np),... The prediction composite signal x (n) is obtained by adding (n). Also in this predictive synthesis, the prediction coefficients are changed by setting the prediction coefficients shown in the table of FIG. 19 for the inputs of the samples y (n), n = 0,..., L−1 of the current frame. Predictive synthesis within the current frame can be performed without straddling the frame.

As for the linear prediction coefficient, the i-th coefficient α ^(q) _i of the order q varies depending on the value of the order q. Thus, in Example 6 above, in FIG. 3A, for example, as described above, when the sample x (1) is input, the prediction coefficient of the first order as a prediction coefficient alpha ₁ alpha ⁽¹⁾ using the _1, sample x When (2) is input, second-order prediction coefficients α ⁽²⁾ ₁ and α ⁽²⁾ ₂ are used as prediction coefficients α ₁ and α ₂ (other α is 0), and x (3) is input The third-order prediction coefficients α ⁽³⁾ ₁ , α ⁽³⁾ ₂ , α ⁽³⁾ ₃ are used as the prediction coefficients α ₁ , α ₂ , α ₃ (other α is 0), and so on. It is necessary to change the prediction coefficient value for multiplying the past samples in each of the multiplication units 24 ₁ ,..., 24 _p for each input of the sample x (n).

On the other hand, the PARCO coefficient is the same as the i-th coefficient even if the value of the order q is different. That is, the Percoll coefficients k ₁ , k ₂ ,..., K _p are coefficients that do not depend on the order. It is well known that Percoll coefficients and linear prediction coefficients can be reversibly converted into each other. Therefore PARCOR coefficient k ₁ from the input sample, k _2, ..., determine the k _p, the prediction from the coefficient k ₁ 1 order coefficient α ^{₍₁₎ 1} determined, predicted from the coefficient k _1, k ₂ of the secondary The coefficients α ⁽²⁾ ₁ and α ⁽²⁾ ₂ are obtained, and similarly, the prediction coefficients α ^(p-1) ₁ ,…, α ⁽ p-1) order from the coefficients k ₁ ,…, k _p-1 ^p-1) _p-1 can be obtained. This calculation can be expressed as:
For i = 1, α ⁽¹⁾ ₁ = k ₁
For i = 2, ..., p, α ⁽ⁱ⁾ _i = -k ₁
α ⁽ⁱ⁾ _j = α ^(i-1) _j −k _i α ^(i-1) _ij , j = 1, ..., i-1
This calculation is the sample number n = 1 was described in Reference Example 6 and 7 above, ..., sequentially to ^{p-1 {α (1)} 1}, {α (2) 1, α (2) 2} , {Α ⁽³⁾ ₁ , α ⁽³⁾ ₂ , α ⁽³⁾ ₃ }, ..., {α ^(p-1) ₁ , α ^(p-1) ₂ , ..., α ^{(p- 1)} It is possible to efficiently perform _p-1 } in a shorter time than that obtained by linear prediction.
Therefore, in Reference Example 8, the linear prediction coefficients α ₁ ,..., Α _p are calculated from the Percoll coefficients and used by the prediction coefficient determination unit 53 in FIG.

The prediction coefficient determination unit 53 calculates p-order percoll coefficients k ₁ , k ₂ ,..., K _p from all samples S _FC = {x (0),. calculated, which is transmitted as auxiliary information C _A separately encoded.
The prediction coefficient determination unit 53 outputs the input sample x (0) as y (0) as it is.
a prediction coefficient determining section 53 x (1) is input is set to the multiplier to calculate the alpha ⁽¹⁾ ₁ from k _1. As a result, first-order prediction error y (1) = x (1) − [α ⁽¹⁾ ₁ x (0)] is output.
When x (2) is input, the prediction coefficient determination unit 53 calculates secondary prediction coefficients α ⁽²⁾ ₁ , α ⁽²⁾ ₂ from k ₁ and k ₂ and sets them in the multiplier. As a result, the second-order prediction error y (2) = x (2) − [α ⁽²⁾ ₂ x (0) + α ⁽²⁾ ₁ x (1)] is output.

When x (3) is input, the prediction coefficient determination unit 53 calculates third-order prediction coefficients α ⁽³⁾ ₁ , α ⁽³⁾ ₂ , α ⁽³⁾ ₃ from k ₁ , k ₂ and k _3. To set the multiplier. As a result, the third-order prediction error y (3) = x (3)-[α ⁽³⁾ ₃ x (0) + α ⁽³⁾ ₂ x (1) + α ⁽³⁾ ₁ x (2)] is output. Is done.
Similarly, the prediction order is sequentially increased up to the sample x (p), and thereafter, the p-order prediction coefficients α ^(p) ₁ ,..., Α ^(p) _p are used.

In the reference example 8 described above, the present invention is applied to the case where the autoregressive linear predictor shown in FIG. 3A is used as the prediction error generation unit 51 of FIG. FIG. 21A shows a configuration using a percoll filter as the prediction error generation unit 51 of FIG. As shown in FIG. 21A, a p-order percoll filter to which the present invention is applied has a configuration in which basic lattice structures are connected in p-stage cascade as is well known. The basic lattice structure of the j-th stage includes a delay unit D, a multiplier 24Bj that multiplies the delayed output by a Percoll coefficient k _j to generate a forward prediction signal, and subtracts the forward prediction signal from the input signal from the previous stage. A subtractor 25Aj that outputs a forward prediction error signal, a multiplier 24Aj that multiplies the input signal and the Percoll coefficient k _j to generate a backward prediction signal, and a backward prediction error signal by subtracting the backward prediction signal from the delayed output. And a subtractor 25Bj for outputting. The forward and backward prediction error signals are respectively given to the next stage. A prediction error signal y (n) by a p-order percoll filter is output from the subtracter 25Ap at the final stage (p-th stage). The coefficient determining unit 201 calculates the Percoll coefficients k ₁ ,..., K _p from the input sample string x (n) and sets them in the multipliers 24A1,..., 24Ap and 24B1,. These PARCOR coefficients are coded in the auxiliary information encoding unit 202, is output as auxiliary code C _A.

FIG. 22 is a table showing the coefficient k set in the p-order percoll filter of FIG. 21A so as to realize the prediction process based only on the sample of the current frame. As is apparent from this table, n coefficients k ₁ ,..., K _n are set for each input sample number n from sample number n = 0 to n = p, as shown in FIG. In addition, the remaining coefficients are set to k _{n + 1} = k _{n + 2} = ... = k _p = 0. Note that for each sample x (n) in this range, the only new coefficient that must be calculated is k _n , and the coefficients k ₀ , k ₁ , ..., k _n-1 have already been calculated. That is, the calculated coefficient can be used as it is.

In the case of the p-th order percall filter processing using the percoll coefficient k as described above, the prediction order is sequentially increased from 0 to p-1 from the sample number n = 0 to n = p−1, and after the sample number n = p. Can reduce the discontinuity of the prediction error signal of the previous frame and the current frame by performing the p-th order prediction.
FIG. 21B shows a configuration in which a prediction synthesis process corresponding to the prediction error generation process of FIG. Similar to the filter of FIG. 21A, the basic lattice structure is a p-stage cascade connection. The basic lattice of the j-th stage is a delay unit D, a multiplier 26Bj that generates a prediction signal by multiplying the output from the delay unit D by a coefficient k _j , and the prediction synthesis from the previous stage (j + 1). An adder 27Aj that adds the signals and outputs an updated predicted synthesized signal; a multiplier 26Aj that multiplies the updated predicted synthesized signal by a coefficient k _j to obtain a predicted value; and And a subtractor 27Bj that subtracts the output from the output of the output and gives the prediction error to the delay unit D of the preceding stage (j + 1). The auxiliary information decoding unit 203 decodes the input auxiliary code C _A to obtain the Percoll coefficients k ₁ ,..., K _p , and the corresponding multipliers 26A1, ..., 26Ap and 26B1, ... , Give to 26Bp.

The first stage (j = p) of the adder 27Ap sequentially enter the prediction error signal samples y (n), PARCOR coefficient k ₁ is set, ..., by performing the processing using the k _p, the last stage ( A predicted synthesized signal sample x (n) is obtained at the output of the adder 27A1 of j = 1). Also in this reference example in which prediction synthesis using the PARCOR filter, PARCOR coefficients k _1, ..., it may be set coefficients shown in FIG. 22 as k _p.

The procedure for executing the filter process of FIG. 21A by calculation will be described below.
The first sample x (0) is used as it is as the prediction error signal sample y (0).
y (0) ← x (0)
When the second sample x (1) is input, the error signal y (1) is obtained only by the first prediction.
y (1) ← x (1) −k ₁ x (0)
x (0) ← x (0) −k ₁ x (1)
When the third sample x (2) is input, a prediction error signal y (2) is obtained by the following calculation. However, x (1) is used to obtain y (3) in the next step.
t ₁ ← x (2) −k ₁ x (1)
y (2) ← t ₁ −k ₂ x (0)
x (0) ← x (0) −k ₂ t ₁
x (1) ← x (1) −k ₁ x (2)
When the fourth sample x (3) is input, y (3) is obtained by the following calculation. However, x (1) and x (2) are used to obtain y (4) in the next step.

t ₁ ← x (3) −k ₁ x (2)
t ₂ ← t ₁ −k ₂ x (1)
y (3) ← t ₂ −k ₃ x (0)
x (0) ← x (0) −k ₃ t ₂
x (1) ← x (1) −k ₂ t ₁
x (2) ← x (2) −k ₁ x (3)
Continue in the same manner. Thus, prediction processing can be performed only from the sample of the current frame. In addition, the k parameter can be used as it is until p + 1 samples x (n) are input, and one parameter can be obtained and the order increased by one. Once the coefficient is determined, the coefficient is updated one by one each time a sample is input.

Similarly, the prediction synthesis process by the Percoll filter shown in FIG. 21B can be executed by calculation as shown below. This process is the reverse of the prediction error generation process on the encoding side.
The first synthesized sample x (0) uses the input prediction error sample y (0) as it is.
x (0) ← y (0)
The second prediction synthesis sample x (1) is synthesized only by the first prediction.
x (1) ← y (1) + k ₁ x (0)
x (0) ← x (0) −k ₁ x (1)
The third prediction synthesis sample x (2) is obtained by the following calculation. However, x (0) and x (1) are used to obtain x (3) in the next step and are not output.

t ₁ ← y (2) + k ₂ x (0)
x (2) ← t ₁ + k ₁ x (1)
x (0) ← x (0) −k ₂ t ₁
x (1) ← x (1) −k ₁ x (2)
x (3) is obtained by the following calculation. However, x (0), x (1), and x (2) are used to obtain x (4) in the next step and are not output.
t ₂ ← x (3) + k ₃ x (0)
t ₁ ← t ₂ + k ₂ x (1)
x (3) ← t ₁ −k ₁ x (2)
x (0) ← x (0) −k ₃ t ₂
x (1) ← x (1) −k ₂ t ₁
x (2) ← x (2) −k ₁ x (3)
Continue in the same manner.

21A and 21B show a configuration example of a Percoll filter that performs linear prediction processing on the encoding side and a Percoll filter that performs predictive synthesis processing on the decoding side that is the inverse process thereof, but different configurations that perform processing equivalent to these are shown. There are many possible percoll filters, examples of which are given below. However, as described above, the linear prediction process and the prediction synthesis process are inverse processes, and the configuration of the Percoll filter is also symmetrical with each other. Therefore, an example of the Percoll filter on the decoding side will be described below.
In the Percoll filter of FIG. 23, no coefficient multiplier is provided between the forward path and the backward path of the signal, and a coefficient multiplier is inserted in the forward path.

In the percoll filter of FIG. 24, coefficient multipliers are inserted in the forward path and the backward path of each stage, respectively, and coefficient multipliers are also inserted between the forward path and the backward path.
The percoll filter of FIG. 25 has the same structure as that of FIG. 24, but the setting of the coefficients is different.
FIG. 26 shows an example of a Percoll filter configured without using the delay D, and a signal error between paths is obtained by subtracters inserted in parallel forward paths.
FIG. 27 shows the configuration of a Percoll filter that performs the inverse process corresponding to FIG.

In the reference example 9 described above, in the autoregressive linear prediction filter processing, the case where the order of linear prediction is sequentially increased from the start sample of the frame to the predetermined number of samples without using the sample of the past frame is shown. In the reference example 10, in the FIR filter process, the number of taps is sequentially increased without using the samples of the past frame.
FIG. 28A shows a reference example when the present invention is applied to the FIR filter processing in the sampling unit in FIG. 1, for example. The buffer 100 stores samples x (0),..., X (L−1) of the current frame FC. As described with reference to FIGS. 2A, 2B, and 2C, when the FIR filter processing is originally performed, the sample x (n) at each time point n is 2T + 1 pieces, that is, the sample and T pieces before and after the sample. The sample and the coefficients h ₁ ,..., H _{2T + 1} are convolved. However, when the present invention is applied, the sample of the previous frame is not used, as shown in the table of FIG. 28B. From the x (0) to the sample x (T), the number of taps of the FIR filter is increased for each sample, and after the sample x (T), a filter process with a predetermined number of taps is performed.

28A and 28B show examples of filter processing when T = 2 for simplicity. The prediction coefficient determination unit 101 is given samples x (0), x (1),..., And based on this, for each sample number n, as shown in the table of FIG. 28B, the prediction coefficients h ₀ , h ₁ ,. .. is calculated. Coefficients h ₀ to the sample x (0) of the current frame read from the buffer 100 is multiplied by the multiplier 22 _0, the output sample y (0) is obtained. Next, the multipliers 22 ₀ , 22 ₁ , 22 ₂ and the adder 23 ₁ perform the convolution operation of the samples x (0), x (1), x (2) and the coefficients h ₀ , h ₁ , h ₂ and output them. y (1) is obtained. Then the multiplier 22 _0, ..., 22 ₄ and adder 23 ₂ by the sample x (0), ..., x (4) and coefficients h _0, ..., performs convolution operation of h _4, Output y (2) is obtained. Thereafter, up to n = L-3, the sample x (n) and a total of five samples before and after the sample x are convolved with the coefficients h ₀ ,..., H ₄ to obtain the output y (n). Further, since the number of remaining samples in the current frame after that becomes smaller than T, the number of taps for filtering is sequentially reduced.

In this way, in the example of FIG. 28B, coefficients h ₀ , h ₁ , h ₂ are used for sample number L-2 at the end of the frame symmetrically with the start side of the frame, and only coefficient h ₀ is used for sample number L-1. use. That is, processing is performed so that the number of taps decreases symmetrically toward the front and rear ends of the frame. However, it is not necessarily symmetrical. In this example, each sample x (n) and the same number of samples are used symmetrically as the samples to be filtered, so the samples x (0) to x (T) are filtered. The number of taps is increased to 1, 3, 5, ..., 2T + 1. However, the sample to be filtered need not be selected symmetrically with respect to the sample x (n).

で実行し、y(n)を出力する。
ステップＳ３：ｔとｎをそれぞれ１歩進する。
ステップＳ４：n=Tとなったか判定し、なっていなければステップＳ２に戻り、再びステップＳ２，Ｓ３，Ｓ４を実行する。これによりｎの増加とともに増加されたタップ数で畳み込み処理が行われる。
ステップＳ５：n=Tとなっていれば次式

により畳み込み演算を行い、y(n)を出力する。
ステップＳ６：ｎを１歩進する。
ステップＳ７：n=L-Tとなったか判定し、なっていなければステップＳ５に戻って再びス
テップＳ５，Ｓ６，Ｓ７を実行する。これによりn=L-Tまでタップ数2T+1のフィルタ処理
が繰り返し実行される。
ステップＳ８：n=L-Tとなっていれば次式

により畳み込み演算を行い、y(n)を出力する。
ステップＳ９：n=L-1となったか判定し、なっていれば処理を終了する。
ステップＳ１０：n=L-1となっていなければｎを１歩進しTを１減少させ、ステップＳ８に戻り、再びステップＳ８，Ｓ９を実行する。これによりフレームの後端に向かってｎの増加とともにタップ数が漸次減少したフィルタ処理が行われる。 FIG. 29 shows the FIR filter processing procedure of Reference Example 10 described above.
Step S1: Initially set the sample number n and the variable t to 0.
Step S2: Convolution operation on the input sample is

And output y (n).
Step S3: Steps t and n one step each.
Step S4: It is determined whether n = T. If not, the process returns to Step S2, and Steps S2, S3, and S4 are executed again. As a result, the convolution process is performed with the number of taps increased as n increases.
Step S5: If n = T, then

Performs a convolution operation and outputs y (n).
Step S6: Advance n by one step.
Step S7: It is determined whether n = LT, and if not, the process returns to Step S5, and Steps S5, S6, and S7 are executed again. As a result, the filtering process with 2T + 1 taps is repeatedly executed until n = LT.
Step S8: If n = LT, then

Performs a convolution operation and outputs y (n).
Step S9: It is determined whether n = L-1, and if so, the process ends.
Step S10: If n = L-1 is not satisfied, n is incremented by 1 and T is decreased by 1. The process returns to Step S8, and Steps S8 and S9 are executed again. As a result, filter processing is performed in which the number of taps gradually decreases as n increases toward the rear end of the frame.

The reference example 11 is obtained by applying the technique of sequentially increasing the predicted order according to the reference example 10 without using the substitute sample string in the reference example 4, and will be described below with reference to FIGS. .
As illustrated in FIG. 30, the processing unit 200 has a configuration in which the substitute sample string adding unit 240 is removed from the configuration illustrated in FIG. In addition, the prediction error generation unit 51 executes the prediction error signal generation processing described with reference to FIGS.
11, 12, and 13, the digital signal (sample sequence) S _FC (= [x (0),..., X (L−1)]) of one frame FC to be processed is, for example, the buffer 100. Stored in
By the similar sample sequence selection unit 210, the sample sequence x (n + τ), ..., x (n + τ + p- similar to the first sample sequence x (0), ..., x (p-1) in the frame FC 1) is read from the sample sequence S _FC of the frame FC in the buffer 100 (S1). This similar sample sequence x (n + τ),..., X (n + τ + p-1) is made to become a similar sample sequence u (0),..., U (p-1) as shown in FIG. Shifting to the head position in the frame FC, the gain applying unit 220 multiplies the similar sample sequence u (n) by the gain β (0 <β ≦ 1) to obtain a sample sequence u (n) ′ = βu (n). (S2) The sample sequence u (n) ′ is subtracted from the sample sequence x (0),..., X (L−1) of the frame FC by the subtractor 230, and the result is sampled as shown in FIG. A sequence v (0),..., V (L-1) is set (S3). That is
n = 0,…, p-1 and v (n) = x (n) −u (n) ′
n = p,…, L-1 and v (n) = x (n)
And After multiplying x (n + τ),..., x (n + τ + p−1) by the gain β, the sample sequence may be shifted to the head position in the frame to be a sample sequence u (n) ′.

A sample sequence v (0),..., V (L-1) is input to the prediction error generation unit 51, and the prediction error signal y (0),... Is obtained by the autoregressive prediction described with reference to FIG. , y (L-1) is generated (S5).
Determination of the position τ and the gain β of the similar sample sequence x (n + τ),..., X (n + τ + p−1) is performed under the control of the selection determination control unit 260 as described in the reference example 4. To do.
A prediction error signal is generated for the sample sequence v (p),..., V (L-1) generated using τ and β determined in this way (S4), and τ and β used at that time are represented. the auxiliary information AI generated by the auxiliary information generation unit 270 (S5), further encoded into code C _AI auxiliary information AI auxiliary information encoding unit 280 as necessary. The auxiliary information AI or the code _CAI is added to a part of the encoded code for the input digital signal of the frame FC by the encoder.

  In the above description, the value of τ is preferably larger than the predicted order p, and the sum ΔU + τ of the lengths ΔU and τ of similar sample sequences u (n) is L−1 or less, that is, x (τ + ΔU) is the frame FC. Τ may be determined within a range that does not deviate from. The length ΔU of the similar sample sequence u (n) may be equal to or less than τ and may be equal to or less than p, which is not related to the predicted order p, but is preferably equal to or greater than p / 2. Furthermore, the head position of the similar sample sequence u (n) may not necessarily coincide with the head position in the frame FC, that is, u (n) may be, for example, n = 3,..., 3 + ΔU. The gain β applied to the similar sample sequence u (n) may be weighted depending on the sample, that is, u (n) may be multiplied by a predetermined window function ω (n). It only needs to represent τ.

A reference example of the prediction synthesis processing method corresponding to the reference example 11 will be described with reference to FIGS. As in the case of the reference example 4 described with reference to FIGS. 14, 15, and 16, this prediction synthesis processing method is used, for example, in the prediction synthesis unit 63 in the decoder 30 in FIG. 1. Even when decoding from a frame, a decoded signal with good continuity and quality can be obtained.
The functional configuration example illustrated in FIG. 33 is the same as the configuration in FIG. 14 in which the substitute sample string generation unit 320 in the processing unit 300 is removed. However, the prediction synthesis unit 63 performs the same prediction synthesis process as described in FIG. 20 or 21B of Reference Example 4.

A sample sequence y (0),..., Y (L-1) of the current frame FC of a digital signal (prediction error signal) to be subjected to prediction synthesis processing by autoregressive prediction is stored in, for example, the buffer 100 and read. The writing unit 310 reads the sample string y (0),..., Y (L−1).
Sample sequences y (0),..., Y (L-1) are sequentially supplied from the head to the prediction synthesis unit 63 (S1), and the prediction synthesis process is performed to produce a prediction synthesis signal v (n) ′ (n = 0. , ..., L-1) are generated (S2). This predicted synthesized signal v (n) ′ is temporarily stored in the buffer 100. The method described in FIG. 20 or 21B is used for this predictive synthesis.

The auxiliary information decoding unit 330 decodes the auxiliary code _CAI as a part of the code of the current frame FC to obtain auxiliary information, thereby obtaining τ and β (S3). The auxiliary information decoding unit 330 may receive the auxiliary information itself. The sample sequence acquisition unit 340 uses τ to determine a predetermined number from the composite signal (sample) sequence v (n), in this example, a sample sequence v (τ),. τ + p) is replicated, that is, v (τ),..., v (τ + p) is obtained with the predicted synthesized signal sequence v (n) as it is (S4). A sample sequence u (n) is shifted to the beginning position, and a gain β from the auxiliary information is multiplied by a gain applying unit 350 to generate a corrected sample sequence u (n) ′ = βu (n) (S5).

This corrected sample sequence u (n) ′ is added to the predicted synthesized sample (signal) sequence v (n) and output as a normal predicted synthesized signal x (n) (n = 0,..., L−1). (S6). The predicted composite sample sequence x (n) is
n = 0,…, p-1 and x (n) ＝ v (n) ＋ u (n) ′
n = p,…, L-1 and x (n) = v (n)
It is.
Since the reference example 12 corresponds to the reference example 11, the length ΔU of the correction sample sequence u (n) ′ is not limited to p, that is, it is irrelevant to the predicted order and is determined in advance. In addition, the position of the first sample of the corrected sample sequence u (n) ′ does not necessarily coincide with the first sample v (0) of the composite signal v (n), and this is also determined in advance. Furthermore, the gain β is not included in the auxiliary information, and may be weighted for each sample u (n) by a predetermined window function ω (n).
First Embodiment Hereinafter, a digital signal processing method according to the present invention will be described.
In the first embodiment of the present invention, for example, when an original digital signal is encoded in units of frames, as a part of the process, a process for generating an autoregressive prediction error signal is performed, or an interpolation filter process is performed. At this time, the last sample sequence of the frame immediately before (past) the current frame or the first sample sequence of the current frame is encoded separately, and the code (auxiliary code) is used as one of the encoding codes of the current frame of the original digital signal. Add to the part. When predicting and synthesizing the prediction error signal on the decoding side or when performing interpolation filter processing or the like, if there is no code of the previous (previous) frame of the frame, the auxiliary code is decoded and the decoded sample The column is used as the tail synthesis signal of the previous frame for the prediction synthesis of the frame.

Example 1 of the first embodiment will be described with reference to FIGS. Example 1 is a case where the first embodiment is applied to an encoder, for example, a prediction error generation unit 51 in the encoder 10 in FIG. The original digital signal S _M is encoded for each frame by the encoder 10 and a code is output for each frame. The prediction error generation unit 51 in a part of the encoding process predicts the input sample sequence x (n) in an autoregressive manner and performs the prediction error signal as described with reference to FIGS. 3A and 3B, for example. y (n) is generated and output every frame.
This input sample sequence x (n) is branched and the auxiliary sample sequence acquisition unit 410 generates a prediction error for the last sample x (-p),..., X (-1) of the frame immediately before (previous) the current frame FC. The prediction order p in the unit 51 is acquired and set as an auxiliary sample string. The auxiliary sample sequence x (-p), ..., x a (-1) is encoded with the auxiliary information encoding unit 420 generates an auxiliary code C _A, original digital signal of the auxiliary code C _A the current frame FC As a part of the encoding code. Main code Im in this example, by combining the error code Pe and auxiliary code C _A synthetic unit 19 outputs a set of codes of the current frame FC, transmits or records.

The auxiliary information encoding unit 420 outputs x (-p),..., X (-1) (generally a PCM code) without being encoded, with a code indicating an auxiliary sample sequence added. Also good. Preferably, compression encoding is performed using a differential PCM code, a prediction code (prediction error + prediction coefficient), a vector quantization code, or the like.
As shown by the broken line in FIG. 37, the last sample of the previous frame is not used, and the auxiliary sample is set to x (0),..., X (p-1) as auxiliary sample sequences for the predicted order of the first sample in the current frame FC You may acquire with the column acquisition part 410. FIG. The auxiliary code in this case is shown as C _A 'in FIG.

Example 2 of prediction synthesis processing corresponding to the prediction error generation of Example 1 will be described with reference to FIGS. 38 and 39. FIG. A set of codes obtained by encoding the original digital signal S _M for each frame is input to a decoder 30 such as the decoder 30 shown in FIG. 1 so that each frame can be distinguished. A set of codes for each frame is separated into each code in the decoder 30, and a decoding process is performed using these codes. As a part of the decoding process, a digital process for predicting and synthesizing the prediction error signal y (n) by the auto-regression type in the prediction synthesizing unit 63 is performed. This prediction synthesis process is performed as described with reference to FIGS. 4A and 4B, for example. That is, for the predictive synthesis of the leading portion y (0),..., Y (p-1) of the prediction error signal y (n) of the current frame FC, the last sample x ( -p),…, x (-1) is required.

However, if a packet in the middle of transmission is lost and the code set (Im, Pe, C _A ) of the previous frame cannot be obtained, or the code set of the frame in the middle of the code set of consecutive frames due to random access When a code set of the previous (past) frame does not exist, such as when performing decoding processing from the above, the missing detection unit 450 detects this and the auxiliary code C _A (or C _A ′) separated by the separation unit 32 (Auxiliary code C _A or C _A ′ described in the first embodiment) is decoded by the auxiliary information decoding unit 460 and auxiliary sample sequences x (−p),..., X (−1) (or x (0), ..., x (p-1)), and this auxiliary sample sequence is input to the prediction synthesis unit 63 as the prediction synthesis end sample sequence x (-p), ..., x (-1) of the previous frame, and then The prediction error signals y (0),..., Y (L-1) of the current frame are sequentially input to the prediction synthesis unit 63 to perform prediction synthesis processing, and the synthesized signals x (0),. ) Generated. The auxiliary code C _A (C _A ′) is duplicated and redundant, but a predictive synthesized signal with good continuity and quality can be obtained without depending on the previous frame. As a decoding processing method in the auxiliary information decoding unit 460, a method corresponding to the encoding processing method of the auxiliary information encoding unit 420 in FIG. 36 is used.

  36 to 39, for example, the digital signal processing related to the prediction error generation unit 51 in the encoder 10 and the prediction synthesizer 63 in the decoder 30 in FIG. 1 has been described. The present invention can also be applied to digital signal processing related to the FIR filter shown in FIG. In that case, instead of the prediction error generation unit 51 of FIG. 36 and the prediction synthesis unit 63 of FIG. 38, the FIR filters of FIG. The signal processing procedure is exactly the same as the processing described with reference to FIGS.

36 to 39, the greatest feature of the embodiment shown in FIG. 36 is that, in the encoding / decoding system shown in FIG. other code Im last sample sequence (or the top sample sequence of the current frame) as auxiliary code C _a of the current frame, since the delivery with Pe, if the missing frame is detected on the receiving side, the prediction in the next frame The synthesizer 63 has an advantage that the predictive synthesis process can be started immediately after adding the sample string obtained from the auxiliary code obtained in the current frame to the head of the error signal of the current frame.

While various codes as described above can be used as auxiliary code, the auxiliary sample sequence is a small number of samples of the order of the prediction order e.g., as auxiliary code C _A, for example when using the PCM code of the sequence of samples after the frame loss detecting the decoding side can initiate an available decode auxiliary code C _a of the current frame as it is as a raw auxiliary sample sequence data immediately. The same effect can be obtained when this method is applied to the FIR filter of the up-conversion unit.
Application Reference Example 1
For example, if the video, audio, etc. are distributed on the Internet, users are not able random access from any frame, generally at the beginning P _H of the start frame FH of frame sequence constituting the superframe SF shown in FIG. 40 Only random access is possible. Another prediction error code Pe of the prediction error signal subjected to digital signal processing described above for each frame, main code Im, auxiliary code C _A is inserted, a superframe FS consisting frames, for example, stored in the packet Is transmitted.

When the receiving side randomly accesses the start frame, it does not have information on a frame earlier than that, so the processing is completed with only the samples in the start frame. Even in such a case, by performing digital signal processing according to the present invention described in each of the above-described reference examples on the frame, the accuracy of linear prediction can be rapidly increased from the point of random access, and high quality can be achieved in a short time. You can start receiving.
Only for the start frame of random access, the digital processing is completed with only the samples in the start frame without using the samples of the past frame. For this reason, both the process of linearly predicting from the front in time and the process of predicting from the back in time are possible. On the other hand, in each frame boundary P _F, it is possible to start the linear prediction process using the samples of the previous frame.

FIG. 41A shows an application reference example applicable to the reference examples described in FIGS. In the reference example, the processing unit 500 of the encoder 10 includes a prediction error generation unit 51, a backward prediction unit 511, a determination unit 512, a selection unit 513, and an auxiliary information encoding unit 514. Although not shown, the encoder 10 includes an encoder that generates a main code, an encoder that encodes the prediction error signal y (n), and outputs a prediction error code Pe. Code Im, Pe, C _A is stored in the packet synthesis unit 19, is output.
In this applied reference example, the backward prediction unit 511 performs linear prediction processing in the past direction from the first symbol of the start frame. The prediction error generation unit 51 performs a forward linear prediction process on samples of all frames. The determination unit 512 encodes the prediction error obtained by performing the forward linear prediction process on the sample of the start frame by the prediction error generation unit 51, and is obtained by the backward linear prediction process of the sample of the start frame by the backward prediction unit 511. It encodes with a prediction error, compares these code amounts, and gives selection information SL for selecting the smaller one to the selection unit 513. The selection unit 513 selects and outputs the prediction error signal y (n) having the smaller code amount for the start frame, and selects and outputs the output of the prediction error generation unit 51 for the subsequent frames. Selection information SL is output as coded in the auxiliary information encoding unit 514 auxiliary code C _A.

FIG. 41B shows a decoder 30 corresponding to the encoder 10 of FIG. 41A, and is applicable to the reference examples of FIGS. The main code Im and the prediction error code Pe separated from the packet by the separation unit 32 are decoded by a decoder (not shown). The processing unit 600 includes a prediction synthesis unit 63, a backward prediction synthesis unit 631, an auxiliary information decoding unit 632, and a selection unit 633. The prediction error signal y (n) decoded from the prediction error code Pe is subjected to prediction synthesis processing by the prediction synthesis unit 63 for all the frame samples. On the other hand, the backward prediction synthesis unit 631 performs backward prediction synthesis only for the start frame. Selection information SL is decoded auxiliary information C _A by the auxiliary information decoder 632 is obtained, thereby either the output of the output or backward prediction synthesis unit 631, the prediction composer unit 63 for starting frame and controls the selection unit 633 Select. For all subsequent frames, the output of the predictive synthesis unit 63 is selected.
Application Reference Example 2
As described above, when the prediction error generation process is performed on the sample sequence on the encoding side according to the reference examples of FIGS. 17 and 21A, the first sample x (0) of the frame is output as it is as the prediction error sample y (0). For the samples x (1), x (2),..., X (p−1), first-order prediction processing, second-order prediction processing,. That is, the first sample of the random access start frame shown in FIG. 40 has the same amplitude as the original sample x (0), and the prediction accuracy increases as the second prediction value, the third prediction value, and the prediction order increase. Increasing, the amplitude of the prediction error decreases. By utilizing this fact, it is possible to reduce the amount of codes by adjusting entropy coding parameters. FIG. 42A shows the configuration of the encoder 10 and its processing unit 500 capable of adjusting such entropy encoding parameters, and FIG. 42B shows the configuration of the decoder 30 and its processing unit 600 corresponding to FIG. 42A. .

As illustrated in FIG. 42A, the processing unit 500 includes a prediction error generation unit 51, an encoding unit 520, an encoding table 530, and an auxiliary information encoding unit 540. The prediction error generation unit 51 performs the above-described prediction error generation processing of FIG. 17 or 21A on the sample x (n), and outputs a prediction error signal sample y (n). For example, the encoding unit 520 performs Huffman encoding with reference to the encoding table 530. In this example, the first sample x (0) having a large frame amplitude and the second sample x (1) are encoded using the dedicated table T1, and the third and subsequent samples x (2) are encoded. , x (3), ... find the maximum amplitude value for each of a plurality of predetermined samples, select a plurality of tables, here one of the two tables T2, T3, and each of the plurality of samples Encode and output error code Pe. In addition, selection information ST indicating which coding table is selected for each of the plurality of samples is output. Selection information ST is output as encoded by the auxiliary information encoding unit 54 auxiliary information C _A. Plurality of frames of code Pe, C _A is stored in the packet synthesis unit 19 along with the main code Im, it is sent.

As shown in FIG. 42B, the processing unit 600 of the decoder 30 includes an auxiliary code decoding unit 632, a decoding unit 640, a decoding table 641, and a prediction synthesis unit 63. The auxiliary information decoder 632 gives the decoding unit 640 the selection information ST decodes the auxiliary code C _A of the separation unit 32. The same decoding table 641 as the encoding table 530 in the encoder 10 of FIG. 42A is used. The decoding unit 640 decodes the beginning of the start frame and the next two prediction error codes Pe using the decoding table T1, and outputs prediction error signal samples y (0) and y (1). For the subsequent prediction error code Pe, one of the tables T2 or T3 designated by the selection information ST is selected and decoded for each of the plurality of codes, and a prediction error signal sample y (n) is output. The prediction synthesis unit 63 applies the prediction synthesis process of FIG. 20 or 21B described above, and performs a prediction synthesis process on the prediction error signal y (n) and outputs a prediction synthesis signal x (n).
Other Modifications The second reference embodiment and the first embodiment are not limited to the case where an autoregressive filter is used, but can be generally applied to processing such as an FIR filter as in the first reference embodiment. Further, in each of the reference examples described above, as the substitute sample strings AS and AS ′, only the upper digits (bits) of the samples may be used, or samples taken from the current frame that is the basis of AS and AS ′. AS and AS ′ may be obtained using only the upper digits (bits) of the samples in the columns ΔS and ΔS ′.

In the above description, the sample sequence in the current frame is used for the processing of the current frame as a substitute for the sample sequence of the previous or / and subsequent frame. However, the sample in the current frame is not used without using such a substitute sample sequence. You may make it complete only by.
For example, in a short filter with a small number of taps, simple extrapolation is possible, for example, when sample values are smoothed or interpolated after up-sampling. That is, for example, in FIG. 43 and FIG. 44, the sample sequence S _FC (= x (1), x (3), x (5),...) Of the current frame is stored in the buffer, and this sampling frequency is upsampled twice. In this case, as shown in FIG. 43A under the control of the control unit, the first sample x (0) of the current frame FC is extrapolated from the samples x (1), x (3), etc. close to that of the current frame FC. Extrapolated with the sample x (2), the average value of the samples x (1) and x (3) on both sides (interpolated) is obtained by interpolation, and after sample x (4) is interpolated by filtering presume. For example, the sample x (4) is estimated from x (1), x (3), x (5), x (7) by a 7-tap FIR filter. In this case, tap coefficients (filter coefficients) of every other three taps are set to zero. These estimated samples x (0), x (2) and input samples x (1) x (3) are synthesized by the synthesis unit with respect to the filter output so as to form a sample sequence shown in FIG. 43A.

  As a method of extrapolating the sample x (0), the nearest sample x (1) is used as it is as shown in FIG. 43B. Alternatively, as shown in FIG. 43C, a straight line 91 connecting two nearby samples x (1) and x (3) is extended to set the value at the time of sample x (0) as the value of sample x (0) (2 Point straight line extrapolation). Alternatively, as shown in FIG. 43D, the value at the time point x (0) is sampled by extending a straight line (minimum square line) 92 close to three nearby samples x (1), x (3), and x (5). Let x (0) (3-point linear extrapolation). Alternatively, as shown in FIG. 43E, a quadratic curve close to three nearby samples x (1), x (3), and x (5) is extended to obtain the value at the time of sample x (0) as sample x (0). (Three-point quadratic function extrapolation).

The digital signal to be processed in the above is generally processing in units of frames, but any signal that requires filter processing that performs processing before or after the frame is applicable. On the contrary, the present invention is intended for a process that requires such a filtering process, and is not limited to a part of the encoding process or the decoding process. When applied to the encoding process, it is used for any of the processes of lossless encoding, lossless decoding, lossy encoding, and lossy decoding.
The above-described digital processor of the present invention (some of which are shown as a processing unit in the figure) can be made to function by executing a program by a computer. That is, a program for causing a computer to execute each step of the various digital signal processing methods of the present invention described above is installed in a computer from a recording medium such as a CD-ROM or a magnetic disk or via a communication line, and the program is installed. Can be executed.

According to the reference example of the present invention described above, it can be said that the digital signal processing method according to the present invention used for encoding, for example, has the following configuration.
(A) Used in an encoding method for encoding a digital signal for each frame, and includes the current sample, at least the immediately preceding p (p is an integer of 1 or more) samples, and the immediately following Q (Q is an integer of 1 or more). This is a processing method using a filter that linearly combines any of the samples, and the sample may be an input signal or an intermediate signal such as a prediction error.
As p samples immediately before the first sample of the current frame, p substitute samples using some consecutive p samples in the current frame are arranged,
A linear combination of the first sample and at least a part of the substitute sample arranged immediately before by the filter, or a part of consecutive Qs in the current frame as Q samples immediately after the last sample of the current frame Distribute Q substitute samples using the samples,
The last sample and at least a part of the substitute sample arranged immediately thereafter are linearly combined by the filter.

Also, for example, the digital signal processing method according to the present invention used for decoding can be said to have the following configuration.
(B) Used in a decoding method for reproducing a digital signal for each frame, the current sample, at least the immediately preceding p (p is an integer of 1 or more) samples, and the immediately following Q (Q is an integer of 1 or more) samples , Where the sample is an intermediate signal such as a prediction error,
If the previous frame does not exist, use some consecutive p samples in the current frame as p substitute samples immediately before the first sample of the current frame, and at least a part of the first sample and substitute samples by the filter And
If there is no immediately following frame,
A part of Q samples in the current frame are used as Q substitute samples immediately after the end sample of the current frame, and the end sample and at least a part of the substitute samples are linearly combined by the filter. And

  By using the present invention, even if a transmission signal is accessed with a frame at an arbitrary time point, it can be immediately reproduced from that frame (random access is possible), so that it is distributed via the Internet, for example. Can be used to send and receive audio and video content.

The functional block diagram which shows the example of the encoder and decoder containing the part which can apply the reference example of a digital processor. A is a diagram illustrating an example of a functional configuration of a filter that requires processing over the preceding and following frames, B is a diagram illustrating an example of processing of an interpolation filter, and C is a diagram for explaining processing that spans the preceding and following frames. A is a figure which shows the function structural example of an autoregressive type | mold prediction error production | generation part, B is a figure for demonstrating the process. A is a figure which shows the function structural example of an autoregressive type | mold prediction synthetic | combination part, B is a figure for demonstrating the process. A is a figure which shows the function structural example of 1st reference form, B is a figure for demonstrating the process. A is a diagram showing a functional configuration example of the digital processor of Reference Example 1, and B is a diagram for explaining the processing. The figure which shows the example of the procedure of the digital processing method of the reference example 1. FIG. A is a figure which shows each example of the signal in the process of the reference example 2, B is a figure which shows the modification of A. A is a diagram showing a functional configuration example of the digital processor of Reference Example 3, and B is a diagram showing a functional configuration example of the similarity calculation unit. 10 is a flowchart showing an example of the procedure of the digital processing method of Reference Example 3. The figure which shows the function structural example of the digital processor of the reference example 4. FIG. The figure which shows each signal example in the process of the reference example 4. FIG. 10 is a flowchart showing an example of a procedure of a digital processing method of Reference Example 4; The figure which shows the function structural example of the reference example 5. FIG. The figure which shows the example of each signal in the process of the reference example 5. FIG. 10 is a flowchart showing an example of a procedure of a digital processing method of Reference Example 5. The figure for description of the reference example 6. FIG. 10 is a flowchart showing an example of a procedure of a digital processing method of Reference Example 6. The table | surface which shows the setting of the prediction coefficient in the reference example 6. FIG. The figure for description of the reference example 7. FIG. A is a figure which shows the filter structure which performs the prediction error signal production | generation process of the reference example 9, B is a figure which shows the filter structure which performs the prediction synthetic | combination process corresponding to FIG. 21A. The table | surface which shows the setting of the coefficient in the reference example 9. FIG. The figure which shows the other structural example of a filter. The figure which shows other structure of a filter. The figure which shows other structure of a filter. The figure which shows the structure of the filter which does not use a delay part. The figure which shows the structure of the filter which performs the reverse process of the filter of FIG. A is a diagram for explaining the reference example 10, and B is a table showing filter coefficient settings in the reference example 10. FIG. 10 is a flowchart showing a processing procedure of Reference Example 10. The figure for description of the reference example 11. FIG. The figure for demonstrating the process of the reference example 11. FIG. 10 is a flowchart showing a processing procedure of Reference Example 11. The figure for description of the reference example 12. FIG. The figure for demonstrating the process of the reference example 12. FIG. 14 is a flowchart showing a processing procedure of Reference Example 12. The figure which shows the function structural example of Example 1 of this invention. The figure for description of Example 1 of this invention. The figure which shows the function structural example of Example 2 of this invention. The figure for description of Example 2 of this invention. The figure which shows the example of a transmission signal frame structure. A is a figure for demonstrating the encoding side process part of the application reference example 1, B is a figure for demonstrating the decoding side process part corresponding to FIG. 41A. A is a figure for demonstrating the encoding side process part of the application reference example 2, B is a figure for demonstrating the decoding side process part corresponding to FIG. 42A. The figure for demonstrating another reference example. The functional block diagram of the reference example shown in FIG.

Claims (10)

Used to encode the original digital signal in units of frames, the first sample series of the frame, the last sample series of the frame before the frame, the last sample series of the frame, the beginning of the frame after the frame A digital signal processing method for performing processing depending on at least one of the sample sequences,
A digital signal processing method comprising a step of using, as a part of the code of the frame, an auxiliary code obtained by encoding the at least one sample series separately from the encoding for the frame. The digital signal processing method according to claim 1, wherein
The top sample series is a sample series of a predetermined number of consecutive samples including the top sample,
The sample sequence at the end is a sample sequence of a predetermined number of consecutive samples including the end sample,
The processing is to perform processing on the sample to be processed using at least a sample to be processed and a sample separated from the sample to be processed by the predetermined number of samples,
A digital signal processing method.   3. The digital signal processing method according to claim 2, wherein the process is a process of generating a prediction error signal by performing a linear prediction process on an input signal, and the number of samples of the sample series is the same as the order of the linear prediction process. A digital signal processing method.   3. The digital signal processing method according to claim 2, wherein the processing is FIR filter processing of an input signal, and the number of samples of the sample series is the same as the order of the FIR filter processing. (a) Decoding the auxiliary code of the frame, the first sample series of the frame, the last sample series of the previous frame, the last sample series of the frame, the first sample series of the frame after the frame Determining at least one sample sequence of
(b) processing the at least one sample sequence for the frame as a decoded sample sequence at the end of a frame preceding the frame or a decoded sample sequence at the beginning of a frame after the frame;
And a digital signal processing method. The digital signal processing method according to claim 5, wherein
The top sample series is a sample series of a predetermined number of consecutive samples including the top sample,
The sample sequence at the end is a sample sequence of a predetermined number of consecutive samples including the end sample,
The processing is to perform processing on the sample to be processed using at least a sample to be processed and a sample separated from the sample to be processed by the predetermined number of samples,
A digital signal processing method.   7. The digital signal processing method according to claim 6, wherein the process is a process of generating a prediction synthesized signal by linearly predicting and synthesizing an input error signal, and the number of samples of the sample series is the same as the order of the linear prediction process. A digital signal processing method.   7. The digital signal processing method according to claim 6, wherein the processing is FIR filter processing, and the number of samples of the sample series is the same as the order of the FIR filter processing.   The program for making a computer perform each step of the digital signal processing method in any one of Claims 1 thru | or 8.   9. A readable recording medium on which a program capable of executing each step of the digital signal processing method according to claim 1 is recorded by a computer.

Images