JP2000268509A - Encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an encoding device capable of sharply enhancing the encoding efficiency even when a GHA(general harmonic analysis) is used in the device. SOLUTION: A local decoding circuit 15 locally decodes an output bit stream. A waveform extending circuit 16 extends a waveform by the local decoding of an already encoded frame to a next frame which is not yet encoded based on the locally decoded signal to make it the predictive signal of the next frame. The difference signal (predictive remainder) between the predictive signal and the original signal of the next frame from a framing circuit 11 is obtained from a subtracting circuit 12 and this signal is supplied to a frequency component extracing circuit 13 as an original signal and the extracing of frequency components is performed by the GHA. Thus, the number of extracted components per unit time can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an encoding device,
In particular, the present invention relates to an encoding device that encodes an audio signal.

[0002]

2. Description of the Related Art A conventional encoding method for audio signals is based on a time-frequency conversion method. this is,
This is a method in which an input signal in the time domain is converted into a frequency domain and then coding is performed. This is an efficient coding method using the bias of the signal in the frequency domain and psychoacoustics of human beings. But,
The fast Fourier transform (FFT) and the modified discrete cosine transform (MD) used in the encoding method by the time-frequency transform method
The frequency analysis method such as CT) is a theory that analyzes a periodic and harmonic signal, and assumes that a waveform in an observation section is periodically repeated outside the observation section where the signal is observed.

However, in the above-described frequency analysis method, since many different frequency components are actually extracted depending on the observation section, the frequency domain signal already contains an error, and the coding efficiency is reduced. However, since the frequency resolution is inversely proportional to the length of the observation section, even when analyzing a transient signal, the observation section length cannot be shortened too much, the frequency resolution becomes insufficient and the coding efficiency decreases, Furthermore, since it is difficult to predict the waveform outside the observation interval, it is not possible to expect an improvement in coding efficiency by coding the prediction residual,
Further, there is a drawback that it is difficult to analyze the frequency of an unsteady signal.

On the other hand, the theory of Fourier analysis extended to non-harmonic signals is also known. This theory is called Generalized Harmonic Analysis (GHA), which extracts the most dominant sine wave with the minimum residual energy from the original waveform in the observation section and repeats the same process for the residual component It is an analysis method that can accurately extract frequency components even for small frequency fluctuations that are not stationary, and the observation section length and frequency resolution can be set freely independently of each other. And the ability to predict signals beyond the threshold.

[0005]

Therefore, the above-mentioned GHA
Can perform frequency analysis with higher accuracy than harmonic analysis such as FFT and MDCT. Therefore, it is conceivable to use GHA instead of FFT or the like used as frequency analysis means in a conventional encoding device. In this case, a discontinuous waveform does not occur if all of the frequency-analyzed components are transmitted, but a discontinuous waveform is highly likely to occur when a large amount of components are discarded to reduce the bit rate. That is, when analyzing a non-stationary signal such as a music signal,
The frequency components constituting the adjacent frames A and B are greatly different. Therefore, for example, when only the 10 most dominant frequency components are left out of 100 frequency components,
If the same frequency components remain in frames A and B, no discontinuous waveform will be generated. However, in general, this is not the case, and the number of remaining 10 frequencies in frames A and B is reduced by the smaller number. Even when the difference between the components is small, there is a high possibility that a discontinuous waveform will occur.

As described above, even when GHA is employed as the frequency analysis means in the encoding method, a discontinuous waveform at the joint of the observation section occurs at the time of decoding, particularly when the bit rate of the encoded bit stream is low. Therefore, in the encoding unit, in order to prevent the discontinuous waveform from occurring at the time of decoding, it is necessary to overlap observation sections at the time of encoding, and there is a problem that a significant improvement in encoding efficiency cannot be expected. .

[0007] The present invention has been made in view of the above points,
It is an object of the present invention to provide an encoding device that can greatly improve encoding efficiency even when GHA is used.

[0008]

SUMMARY OF THE INVENTION To achieve the above object, the present invention provides a framing circuit for dividing an input digital audio signal into a frame composed of a plurality of samples.
A subtraction circuit for obtaining a difference signal between the signal in the frame extracted from the framing circuit and the prediction signal, and frequency analysis of the difference signal from the subtraction circuit by general harmonic analysis to output a plurality of frequency extraction components A frequency component extraction circuit, an information amount reduction circuit that performs encoding with a reduced amount of information of a plurality of frequency extraction components extracted by the frequency component extraction circuit, and transmits the information as a bit stream to a transmission path; A local decoding circuit for locally decoding the output bit stream, and a waveform of an output decoded signal of the local decoding circuit,
And a waveform extension circuit for generating a prediction signal by extending the waveform to the section of the next frame that is not encoded and outputting the prediction signal to the subtraction circuit.

According to the present invention, based on a locally decoded signal,
Since a predicted signal whose waveform is extended is generated in the section of the next frame that is not coded, the frequency analysis by the frequency component extraction circuit is performed on the prediction residual between the original signal of the section of the next frame and the predicted signal. Can be.

[0010]

Next, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of an encoding apparatus according to the present invention. In FIG. 1, the encoding apparatus includes a frequency analysis circuit 10 based on GHA, an information amount reduction circuit 14, a local decoding circuit 15 for outputting a decoded signal, and a waveform extension circuit 16 for outputting a prediction signal. The frequency analysis circuit 10 includes a framing circuit 11 for dividing the input digital audio signal into frames,
The subtraction circuit 12 subtracts the output audio signal of the framing circuit 11 from the prediction signal from the waveform extension circuit 16, and the frequency component extraction circuit 13 extracts a plurality of frequency components from the output signal of the subtraction circuit 12. I have.

Next, the operation of this embodiment will be described with reference to FIG.
3 and the signal waveform diagrams of FIGS. 4 to 6. The digital audio signal is supplied to the framing circuit 11 shown in FIG. 1, where it is divided into observation sections (frames) including a plurality of samples. At this time, the frames do not necessarily have to overlap. When the prediction signal is not output from the waveform extension circuit 16, the digital audio signal divided into frames by the framing circuit 11 and taken out is subtracted from the subtraction circuit 12.
Is supplied to the frequency component extraction circuit 12, where the frequency components of the signals in the frame are extracted by the general harmonic analysis (GHA) according to the procedure of the flowcharts shown in FIGS.

That is, when a digital audio signal of one frame is input (step 10 in FIG. 2).
1) Extract only one frequency component that minimizes the residual signal from the signal of one frame (step 102 in FIG. 2). Subsequently, it is determined whether or not the original signal from which the extracted frequency components have been removed, that is, the residual is sufficiently small (step 103 in FIG. 2). If the residual is larger than a predetermined threshold value, the processing in step 102 is performed. Is performed again on the residual, and the analysis is terminated when the residual becomes sufficiently small. Then, the obtained analysis result is output to the information amount reducing circuit 14 in FIG. 1 (Step 104 in FIG. 2).

FIG. 3 is a flowchart showing the detailed processing of the above-mentioned step 102, and also shows the above-mentioned step 103. The frequency component analysis process of step 102 will be described in more detail with reference to the flowchart of FIG.
Is selected (step 201 in FIG. 3), and the continuous signal x ₀ (t) observed in the short-time observation section [0, L].
Fourier coefficients S (f) and C (f) uniquely determined from
Is calculated by the following equation (Step 202 in FIG. 3).

[0014]

(Equation 1) Here, in the equations (1) and (2), f is an arbitrary frequency,
T is 1 / f in cycle, n is an integer, and nT ≦ L.

Subsequently, using the above Fourier coefficients S (f) and C (f), the residual signal ε (t, f) in the observation section and its energy E (f) are calculated by the following equation (FIG. 4). Step 203 in FIG. 3).

[0016]

(Equation 2) Subsequently, it is determined whether or not the frequency f is the last frequency component (step 204 in FIG. 3). If the frequency f is not the last frequency component, the processing of steps 201 to 203 is performed again. In this manner, the calculations of the above equations (1) to (4) are performed for all the frequency components of one frame, and the frequency f that minimizes the energy E (f) expressed by the equation (4) is obtained. ₁ and Fourier coefficients S ₁ (f ₁ ) and C ₁ (f ₁ ) are obtained. The frequency f _{1 at} this time is the most dominant spectrum,
This is selected (step 205 in FIG. 3).

Next, the following equation _{x 1 (t) = x 0} (t) -S 1 (f 1) sin (2πf 1 t) -C 1 (f 1) cos (2πf 1 t) (5), The residual x ₁ (t) is determined from the original signal x ₀ (t) by deleting the most dominant component of the frequency f ₁ determined in step 205 by the dominant spectrum (step 20 in FIG. 3)
6).

Subsequently, it is determined whether or not the residual x ₁ (t) is sufficiently small (step 103 in FIG. 3). Here, the determination in step 103 may be made based on a comparison result of magnitude comparison between a predetermined threshold value set in advance and the residual x ₁ (t), but here, the residual x ₁ (t) ) Is expected to be sufficiently small
Times, the residual x ₁ (t) is regarded as the original signal and step 201
The determination is made based on whether or not the processing of steps S206 to S206 has been repeated.

Thus, the original signal x ₀ (t) is represented by the following equation x (t) in the observation section [0, L].

[0020]

(Equation 3) As shown in equation (6), f ₁ , f ₂ ,. . . , Of f _N, a superposition of sinusoidal total N present (frequency extraction component), it can be estimated with high accuracy. These sine waves are not necessarily harmonic (frequency that is an integral multiple of the fundamental wave). Note that the power level P (f _k ) of the signal x (t) represented by the equation (6) is S ² _k + C ² _k
It is represented by

The frequency extraction component obtained in this way is supplied to the information amount reduction circuit 14 in FIG. 1, where the frequency extraction component having a power level lower than the set power level is not output, thereby reducing the information amount. At the same time, the number of frequency extraction components is recorded on data obtained by reducing the amount of information by lossless compression encoding such as Huffman encoding for frequency extraction components having a power level equal to or higher than the set power level. A bit stream obtained by adding the header information is transmitted to an arbitrary transmission path.

The bit stream is supplied to a local decoding circuit 15 and decoded. This decoded signal is supplied to the waveform extension circuit 16, and the waveform obtained by local decoding of the already coded frame is extended to the next frame which has not been coded, and is used as a prediction signal of the next frame.
For example, if the decoded signal extracted from the local decoding circuit 15 is a signal having a waveform as indicated by I in FIG.
6 is based on the local decoded signal in the section A, using x (t) obtained by setting the time t in the equation (6) as the range of the section A,
The signal of the section B is predictively combined with the time t as the range of the section B to generate a prediction signal indicated by a dotted line III in FIG.

This prediction signal III is supplied to the subtraction circuit 12, and is subtracted from the original signal of the next section B (indicated by II in FIG. 4) outputted from the framing circuit 11, and is indicated by IV in FIG. The difference signal is supplied to the frequency component extraction circuit 13 as the original signal of the section B, and the above-described frequency component extraction is performed, and a plurality of frequency components are extracted.

The frequency extraction component obtained for the prediction residual is supplied to the information amount reduction circuit 14 in FIG. 1, where the extracted component having a power level smaller than the set power level is output as described above. Is not output, the amount of information is reduced, and the frequency extraction component having a power level equal to or higher than the set power level is further reduced in the information amount by lossless compression processing such as Huffman coding. A bit stream obtained by adding header information recording the number is transmitted to an arbitrary transmission path.

[0025]

As described above, according to the present invention,
By generating a prediction signal whose waveform has been extended to the next uncoded frame section based on the local decoded signal, the prediction residual between the original signal of the next frame section and the prediction signal is generally reduced. Since the frequency components are extracted by the harmonic analysis, the frequency components per unit time can be extracted with as small a number as possible, which can greatly improve the coding efficiency.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a flowchart for explaining an operation of a main part in FIG. 1;

FIG. 3 is a flowchart illustrating in detail a frequency component extraction step of a main part in FIG. 2;

FIG. 4 is a waveform diagram illustrating an example of an original signal waveform output from the framing circuit of FIG. 1;

FIG. 5 is a diagram showing an example of a predicted signal waveform in FIG.

FIG. 6 is a diagram showing an example of a differential signal waveform in FIG.

[Explanation of symbols]

 Reference Signs List 10 frequency analysis circuit 11 framing circuit 12 subtraction circuit 13 frequency component extraction circuit 14 information amount reduction circuit 15 local decoding circuit 16 waveform extension circuit

Continued on the front page F term (reference) 5D044 AB05 DE03 GK08 HL11 5D045 DA02 DA20 5J064 AA02 BA09 BB01 BC02 BC08 BC22 BD02 9A001 BB02 BB04 EE04 GG12 HH15 HH16

Claims (2)

[Claims] 1. A framing circuit for dividing an input digital audio signal into a frame including a plurality of samples; a subtraction circuit for obtaining a difference signal between a signal in a frame extracted from the framing circuit and a prediction signal; A frequency component extraction circuit that performs frequency analysis on the difference signal from the subtraction circuit by general harmonic analysis and outputs a plurality of frequency extraction components, and information on the plurality of frequency extraction components extracted by the frequency component extraction circuit An information amount reduction circuit that performs encoding with a reduced amount and transmits the bit stream as a bit stream; a local decoding circuit that locally decodes an output bit stream of the information amount reduction circuit; and an output decoding signal of the local decoding circuit. The waveform is extended to a next frame section that is not encoded to generate the prediction signal and output the signal to the subtraction circuit. Encoding apparatus characterized by having a waveform extending circuit. 2. The information amount reduction circuit records the number of frequency extraction components in data obtained by performing compression encoding on a frequency extraction component having a power level higher than a set power level among input frequency extraction components. The encoding apparatus according to claim 1, wherein the bit stream to which the header information added is generated.