KR100903109B1 - Lossless Coding/Decoding apparatus and method - Google Patents

Abstract

An apparatus and method for lossless encoding / decoding of an audio signal are disclosed. After generating a first bit string for each level from the quantization indexes for the frequency coefficients of the current frame, and generating a symbol consisting of the run length of the second bit string in a line of the first bit string for each level, The symbol is encoded into a third bit string. As a result, the encoding performance of the audio signal can be improved.

Description

Lossless coding / decoding apparatus for audio signal and method thereof {Lossless Coding / Decoding apparatus and method}

1 is a diagram showing the configuration of an embodiment of a conventional audio encoder;

2 is a diagram illustrating an example of conventional Huffman coding;

3 is a diagram illustrating an example of a quantization index of a frequency coefficient;

4 is a diagram illustrating a configuration of an embodiment of a lossless encoder according to the present invention;

5 is a diagram showing the configuration of an embodiment of a lossless decoder according to the present invention;

6 is a diagram illustrating a flow of an embodiment of a lossless encoding method according to the present invention;

7 is a flowchart illustrating an embodiment of a lossless decoding method according to the present invention.

The present invention relates to encoding and decoding audio signals, and more particularly, to an apparatus and a method for lossless encoding / decoding.

1 is a diagram showing the configuration of an embodiment of a conventional audio encoder.

Referring to FIG. 1, an audio encoder includes a frequency converter 100, a psychoacoustic analyzer 110, a quantizer 120, and a lossless encoder 130.

The frequency converter 100 generates spectral coefficients by converting the audio signal into a frequency domain. The psychoacoustic analyzer 110 generates masking information on the audio signal by analyzing the audio characteristics of the audio signal.

The quantization unit 120 quantizes the frequency coefficients, outputs a quantization index of the frequency coefficients, determines a quantization standard for each frequency band according to masking information, and quantizes the frequency coefficients according to the determined quantization standard. In this case, the quantization unit 120 may use various quantization techniques such as prediction quantization in order to improve quantization performance.

The lossless encoder 130 generates a bitstream by encoding the quantization index of the frequency coefficient. Bit strings are conveyed through a storage medium or communication channel.

The audio decoder extracts a quantization index from the transmitted bit string, obtains a frequency coefficient, and then converts the frequency coefficient into a time domain to restore the final audio signal.

The input data to be encoded by the lossless encoder 130 is generally called a symbol. Inverse conversion between an input symbol and an output code is possible (i.e., restoration of an input symbol from an output code), and no additional distortion occurs is called 'lossless' coding.

The goal of lossless coding is to encode given input symbols with the minimum amount of bits and to have the minimum amount of bits on average, taking into account the probability distribution of the symbols. For example, a symbol having a high probability of occurrence is encoded by a short bit of code, and a symbol having a low probability of occurrence is encoded by a long bit of code. As a result, in the lossless encoding, the number of bits is reduced on average compared with the conventional method of encoding all symbols with the same length.

The conventional lossless coding unit directly encodes a quantization index of a frequency coefficient by an entropy coding method such as Huffman coding or arithmetic coding.

Huffman coding is given by a symbol corresponding to each symbol, and in general, each symbol has a different length. In addition, even if the bits of the corresponding symbol code are listed in a line without separately indicating the symbol boundary, the code of each symbol is allocated to completely restore each symbol. Given the probability of occurrence of symbols, a method is known in the art that produces a sign that allows recovery with a minimum average number of bits.

2 is a diagram illustrating an example of conventional Huffman coding.

Referring to FIG. 2, when the probability of each symbol is given for four symbols, a code and a number of bits for Huffman coding are given, and a symbol can be recovered from a bit string without ambiguity. When encoding with respect to FIG. 2, an average of 1.6 bits per symbol is required, and in general, bits are saved as compared with the case of allocating 2 bits per symbol.

As an example of Huffman coding according to the prior art, the operation specifications of the lossless coding unit used in MPEG-2 / 4 AAC are as follows.

1024 quantization indexes (symbols) are input to each frame, and are combined into two-dimensional or four-dimensional vectors to define symbols and Huffman coding. Eleven Huffman tables (7 for 2D and 4 for 4D) are given, and a table capable of encoding the symbol with the least bits is selected and used.

In addition, the order of the vector is selected from two-dimensional and four-dimensional cases where the encoding performance is excellent. Since different Huffman tables are used for different vector orders, the vector order is automatically defined when the table used is specified.

In the conventional lossless coding method, the quantization index of the frequency coefficient is used as a symbol. That is, the concept of directly encoding without additional processing by inputting the frequency coefficient index.

3 is a diagram illustrating an example of a quantization index of a frequency coefficient.

Referring to FIG. 3, when a quantization index of a frequency coefficient is given, two- or four-dimensional vectors are sequentially generated in a direction in which a frequency increases (that is, a direction in which n increases), and Huffman coding of each vector is performed.

The ultimate goal of lossless encoding is to encode the symbol using fewer bits. However, the conventional method uses only Huffman coding for the frequency coefficient index and does not perform additional processing for improving the performance, and thus does not provide the optimal performance.

An object of the present invention is to provide a lossless encoding / decoding apparatus and method capable of improving the performance of an audio encoder by compressing an audio signal with a small number of bit streams having the same encoding distortion.

Another object of the present invention is to provide a computer-readable recording medium having recorded thereon a program for executing a lossless encoding / decoding method on a computer.

In order to achieve the above technical problem, an embodiment of the lossless encoder according to the present invention includes a bit converter for generating a first bit string for each level from quantization indexes for frequency coefficients of a current frame; A run length converter configured to generate a symbol including a run length of a second bit string in which the first bit string for each level is arranged in a line; And a run length encoder to encode the symbol into a third bit string.

In order to achieve the above technical problem, an embodiment of the lossless coding method according to the present invention comprises: generating a first bit string for each level from quantization indices of frequency coefficients of a current frame; Generating a symbol comprising a run length of a second bit string in which the first bit string for each level is arranged in a line; And encoding the symbol into a third bit string.

In order to achieve the above technical problem, an embodiment of a lossless decoding apparatus according to the present invention includes: a run length decoder for decoding a bit string to obtain a symbol having a run length; A run-bit converter configured to arrange bit values according to the run length to generate a first bit string; And a bit inverse transform unit reconstructing quantization indices of frequency coefficients from the first bit string.

In order to achieve the above technical problem, an embodiment of the lossless decoding method according to the present invention includes: obtaining a symbol having a run length by decoding a bit string; Generating a first bit string by arranging bit values according to the run length; And restoring quantization indices for frequency coefficients from the first bit string.

As a result, the encoding performance of the audio signal can be improved.

Hereinafter, a lossless encoding / decoding apparatus and method thereof according to the present invention will be described in detail with reference to the accompanying drawings.

First, a process of lossless encoding and decoding according to the present invention will be described with reference to FIG. 3.

라 하고, 부호를 제거한 절대값을

이라고 하자. 먼저,

을 0 레벨(k=0)부터 시작하여 각 레벨 k에 대한 비트값 Q_k[n]을 다음 수학식 1과 같이 구한다. 여기서, k및

은 모두 0 이상의 정수이고, Q_k[n]은 비트값이다.Integer quantization index of N frequency coefficients corresponding to the current frame

The absolute value without the sign

Let's say. first,

Starting from the 0 level (k = 0), the bit value Q _k [n] for each level k is obtained as in Equation 1 below. Where k and

Are all integers greater than or equal to zero, and Q _k [n] is a bit value.

For example, referring to the frequency coefficient of FIG. 3, N = 10 and Q _k [n] are given by Equation 2 below.

Q ₃ [n] = {,, 0,0,,,,,,}

Q ₂ [n] = {, 0,1,1,,,,,,}

Q ₁ [n] = {0,1,1,1,0,,,, 0,0}

Q ₀ [n] = {1,1,1,1,1,0,0,0,1,1}

Here, Null values are shown except.

The n position with a null value in Q _k [n] is excluded from coding because it can be known without additional information by the "Stack" property in the restoration process. In addition, if the bit is all 0 at a certain level, the bit generation is terminated automatically because there is no need for more levels.

을 정의하여 전송하면 필요한 모든 정보가 전달된다. 또한 비트열이 종료되는 시점이 자동적으로 정해지므로 이에 대한 추가 정보는 필요없다. 수학식 2에 대한

값은 다음 수학식 3과 같 다.The restoration process starts at k = 0 and increments k and if Q _p [n ₀ ] = 0 at a specific n = n ₀ , then Q _r [n ₀ ] = Null for all levels r with r> p. to be. Therefore, the n information of each bit in Q _k [n] does not need to be transmitted separately, and a new bit string is obtained by arranging bits in one dimension while increasing k starting from k = 0.

After defining and sending, all necessary information is delivered. In addition, when the bit string ends is automatically determined, no additional information is required. For equation (2)

The value is shown in Equation 3 below.

로부터

을 복원하는 과정은 다음과 같다. 먼저 다음 수학식 4에 따라

로부터 Q_k[n]을 복원한다.

from

The process of restoring is as follows. First, according to the following equation (4)

Restore Q _k [n] from.

을 복원한다.From Q _k [n] according to the following equation (5)

Restore

를 전송하면 주파수 계수의 인덱스 절대값

의 복원이 가능하다. 따라서 종래 기술에 따라

을 무손실 부호화하는 대신에, 본 발명에서는

으로부터 새로운 정보

을 정의하여 무손실 부호화하는 방법을 사용함으로써, 부호화에 필요한 비트수를 감소한다. Bit strings defined in this way

Transmits the absolute value of the index of the frequency coefficient

Restoration is possible. Thus according to the prior art

Instead of lossless coding, in the present invention

New information from

The number of bits required for encoding is reduced by using the method for lossless encoding by defining the s i n.

에는 일반적으로 동일한 비트값이 연속적으로 발생하게 되며, 연속적으 로 발생하는 동일한 비트열을 런(Run)이라 한다. 비트열에는 '1'의 런과 '0'의 런이 반복적으로 발생하므로, 각 런의 길이만 전달하면 정보의 손실 없이 복원이 가능하다. 예를 들어, 도 3의 주파수 계수에 대하여 다음 수학식 6과 같은 런 길이 열만 전송하면 된다.

In general, the same bit value is continuously generated, and the same bit string that is continuously generated is called a run. Since the run of '1' and the run of '0' are repeatedly generated in the bit string, the length of each run can be restored without loss of information. For example, only the run length column as shown in Equation 6 may be transmitted with respect to the frequency coefficient of FIG. 3.

Run Length column = {5,3,2,1,3,4,2,2}

In the present invention, a bit string is generated by converting a frequency coefficient quantization index into bits for each level, the bit string is converted into a run length string, and a new symbol to be finally encoded is newly defined, and then the symbol is losslessly encoded.

를 생성할 때 각 레벨 Q_k[n]의 비트 나열 방향을 변경할 수 있다. 일 예로, 특정 레벨에서 주파수가 증가하는 방향으로 비트를 나열하고 다음 레벨에서는 주파수가 감소하는 방향으로 비트를 나열하여 인접한 레벨에서 비트 반향이 반대가 되게 할 수 있다. 이 방법을 적용하면 도 3에 대한

와 그에 따른 런 길이는 다음 수학식 7과 같다. 여기서 k=0에서 주파수 증가 방향으로 비트를 나열하였다.From Q _k [n]

You can change the bit-ordering direction of each level Q _k [n] when generating. For example, bits may be arranged in a direction of increasing frequency at a specific level, and bits may be arranged in a direction of decreasing frequency at a next level so that bit reflections may be reversed at adjacent levels. Applying this method to FIG. 3

And the resulting run lengths are shown in Equation 7 below. Here the bits are listed in the direction of increasing frequency at k = 0.

비트열과 런 길이 열이 다르고, 그에 따라 부호화에 필요한 비트수가 다르게 된다. 따라서 각 프레임에 대하여 모든 레벨에서 비트 나열 방향을 선택하여 심볼을 구하는 것도 가능하다. In other words, when generating a bit stream, two directions can be used for each level.

The bit string and the run length string are different, and accordingly, the number of bits required for encoding is different. Therefore, it is also possible to obtain a symbol by selecting bit ordering directions at all levels for each frame.

4 is a diagram illustrating a configuration of an embodiment of a lossless encoder according to the present invention.

Referring to FIG. 4, a lossless encoder includes an absolute value module 400, a code processor 410, a symbol converter 420, a run length encoder 430, and a bit multiplexer 440. 420 includes a bit converter 422 and a run length converter 424.

When the absolute value module 400 receives the quantization index of the frequency coefficient, the absolute value module 400 separates the quantization index into a sign and an absolute value and outputs the quantization index to the code processor 410 and the symbol converter 420, respectively.

The code processor 410 outputs a first code bit string obtained by encoding a sign output from the absolute value module 400.

The symbol converter 420 receives the first symbol that is the absolute value of the quantization index, and then converts the first symbol into a second symbol to which encoding is to be applied. Specifically, the symbol converter 420 includes a bit converter 422 and a run length converter 424.

The bit converter 422 converts the first symbol into bit values for each level by performing the operation of Equation 1, arranges them in one dimension, and generates and outputs a first bit string. The bit converter 422 is not limited to the operation of Equation 1, and various other methods for obtaining the same output by performing an operation of the same concept may be applied. The run length converter 424 obtains a run of the same bit from the first bit string and obtains and outputs a second symbol corresponding to the length of each run.

The run length encoder 430 encodes a second symbol having a run length and outputs a first absolute value bit string. The run length encoder 430 may use conventional Huffman coding or Arithmetic Coding.

The bit multiplexer 440 combines the first code bit string output from the code processor 410 and the first absolute value bit string output from the run length encoder 430 to output the second bit string of the final lossless encoder. .

5 is a diagram illustrating a configuration of an embodiment of a lossless decoder according to the present invention.

Referring to FIG. 5, the lossless decoder includes a bit demultiplexer 500, a code decoder 510, a run length decoder 520, a symbol inverse transform unit 530, and a code combiner 540. The symbol inverse converter 530 includes a run-bit converter 532 and a bit inverse converter 534. The lossless decoder according to the present invention decodes a bit string output from the lossless encoder illustrated in FIG. 4.

When the bit demultiplexer 500 receives the first bit string, the bit demultiplexer 500 demultiplexes the first bit string into a first code bit string and a first absolute value bit string.

The code decoder 510 restores the first code bit string to a coefficient sign.

The run length decoder 520 outputs a first symbol corresponding to the run length through a reverse process of the run length encoder 430 illustrated in FIG. 4.

The symbol inverse transform unit 530 inversely transforms the first symbol and outputs a second symbol. In detail, the symbol inverse transform unit 530 includes a run-bit transform unit 532 and a bit inverse transform unit 534. The run-bit converter 532 generates a second bit stream from the run length of the first symbol, and the bit inverse converter 534 generates the absolute coefficient value from the second bit stream according to the operations of Equations 4 and 5 below. Outputs a second symbol corresponding to The bit inverse converter 534 is not limited to the operations of Equations 4 and 5, and various other methods for obtaining the same output by performing operations of the same concept may be applied.

The sign combiner 540 combines the coefficient sign and the absolute coefficient value and outputs the final coefficient index.

6 is a flowchart illustrating an embodiment of a lossless encoding method according to the present invention.

Referring to FIG. 6, the lossless encoder separates the quantization index of the frequency coefficient of the current frame into a sign and an absolute value (S600). The lossless encoder generates a code bit string by encoding a sign (S610). Lossless coding converts the absolute values into bit strings for each level according to Equation 1, and generates the second bit string by listing the converted bit strings for each level in one dimension (S620). The lossless encoder converts the second bit string into the run length string (S630) and encodes the run length string (S640). The lossless encoder generates a final output bit string by combining the encoded values of the code bit string and the run length string (S650).

7 is a flowchart illustrating an embodiment of a lossless decoding method according to the present invention.

Referring to FIG. 7, the lossless decoder separates a code bit string and an absolute value bit string from an input bit string (S700). The lossless decoder restores the code bit string to obtain the coefficient code (S710), and restores the absolute value bit string to obtain the run length (S720). The lossless decoder generates a bit string from the run length (S730) and obtains the absolute value of coefficients according to Equations 4 and 5 (S740). The lossless decoder combines the coefficient encoding absolute values to obtain a quantization index of the final frequency coefficient (S750).

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

According to the present invention, since the performance of lossless coding of frequency coefficients is improved in the audio encoder, the audio signal can be compressed with a smaller number of bit streams with the same encoding distortion. In addition, if the target number of bits is the same, having the same compression rate can reduce distortion and improve the performance of the audio encoder.

Claims (17)

A bit converter configured to generate a first bit string for each level from quantization indexes of frequency coefficients of the current frame; A run length converter configured to generate a symbol including a run length of a second bit string in which the first bit string for each level is arranged in a line; And And a run length encoder to encode the symbol into a third bit string. The method of claim 1, An absolute value module that separates a quantization index of a predetermined number of frequency coefficients into a sign and an absolute value; A code processor which encodes the sign into a fourth bit string; And And a multiplexer configured to multiplex the third bit string and the fourth bit string. And a quantization index of the bit transform unit is an absolute value separated from the absolute value module. The method of claim 1, wherein the bit converter, And the first bit string is generated by allocating 1 when the value of the quantization indexes is greater than each level value at each level, 0 when the same value, and null otherwise. The method of claim 1, wherein the run length conversion unit, And a second bit string obtained by arranging the first bit string for each level in one of a frequency increasing direction and a frequency decreasing direction. The method of claim 1, wherein the run length conversion unit, And a second bit string obtained by alternately arranging the first bit string for each level in a frequency increasing direction and a frequency decreasing direction. Generating a first bit string for each level from quantization indices for the frequency coefficient of the current frame; Generating a symbol comprising a run length of a second bit string in which the first bit string for each level is arranged in a line; And And encoding the symbol into a third bit string. The method of claim 6, Separating quantization indices of a predetermined number of frequency coefficients into sign and absolute values; Encoding the sign into a fourth bit string; And Multiplexing the third bit stream and the fourth bit stream; And a quantization index of the first bit string generation step is the absolute value. The method of claim 6, wherein the symbol generation step, Generating the first bit string by allocating 1 when the value of the quantization indices at each level is greater than each level value, 0 in the same case, and null in other cases. Coding method. The method of claim 6, wherein the symbol generation step, And generating a symbol comprising a run length of a second bit string in which the first bit string for each level is arranged in one of a frequency increasing direction and a frequency decreasing direction. The method of claim 6, wherein the symbol generation step, And generating a symbol comprising a run length of a second bit string alternately arranging the first bit string for each level in a frequency increasing direction and a frequency decreasing direction. A run length decoder configured to decode a bit string encoded by using the bit string for each level generated from the quantization indexes of the frequency coefficients to obtain a symbol having a run length; A run-bit converter configured to arrange bit values according to the run length to generate a first bit string; And And a bit inverse transform unit for recovering quantization indices of frequency coefficients from the first bit string. The method of claim 11, A demultiplexer for separating a received bit string into a sign bit string and the absolute value bit string; A code decoding unit which decodes the code bit string to determine a sign; And And a sign combiner configured to combine the sign and the quantization index. And the bit string of the run length decoder is an absolute value bit string separated by the demultiplexer. The method of claim 11, wherein the bit inverse converter, And a second bit string for each level is obtained from the first bit string, and the quantization indices are obtained from the second bit string. Decoding a bit string to obtain a symbol having a run length; Arranging bit values according to the run length to generate a first bit string; And Restoring quantization indices for frequency coefficients from the first bit stream. The method of claim 14, Dividing a received bit string into a sign bit string and the absolute value bit string; Determining a sign by decoding the code bit string; And Combining the sign and the quantization index; And the bit string of the step of obtaining the symbol is the absolute value bit string. The method of claim 14, wherein the quantization index recovery step, Obtaining a second bit string for each level from the first bit string, and obtaining the quantization indices from the second bit string. A computer-readable recording medium having recorded thereon a program for executing the method according to any one of claims 6 to 10 and 14 to 16 on a computer.

Images