KR100322703B1 - Audio encoder/decoder - Google Patents

Abstract

PURPOSE: An audio encoder/decoder is provided to reduce the number of registers used to store coefficients by realizing a filter using properties of analysis window and mixing window coefficients and a compensation coefficient. CONSTITUTION: An audio encoder comprises an analysis sub-band filter for dividing an input audio signal into N sub-bands according to bands of a predetermined process unit, a sound mentality modeling part for considering a recognition ability of a human according to a frequency property of the audio signal, a quantization part for assigning and quantizing bits into each sub-band by outputs of the analysis sub-band and the sound mentality modeling part, and a packing part for forming a bit stream using quantized values and header information. An audio decoder comprises a header detecting part for unpacking the bit stream from the audio encoder, and a composition sub-band filter for restoring an original audio signal from the unpacked bit stream.

Description

Audio Encoder / Decoder

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoder / decoder. In particular, when the audio encoder / decoder is implemented in one chip, an audio encoder which reduces the registers used to store the analysis window coefficients and the synthesis window coefficients to 256 by using a polyphase filter. It is about a decoder.

The rapid development of semiconductor and computer technologies has enabled us to process much of the information we handle digitally and to enable high-speed processing. As a result, products that have been considered impossible to implement in the past are now appearing in this market. Among them, data compression / restore technology is the core technology of digital audio / video processing device. Recently, international standardization work related to compression, storage and testing of audio and video signals has been promoted. This work is being carried out by a group called Moving Picture Experts Group (MPEG).

The purpose of MPEG-1 proposed in MPEG is to reduce the number of bits used to represent data, and to read and write data at a speed of 1.5 Mbit / s, and to increase the amount of information that can be stored and represented on CDs, DATs, and magnetic disks. It is currently used in many karaoke systems, CD vision, and CD-FMV (Full Motion Video).

MPEG-2 is currently applied to CATV, Electronic News Gathering (ENG), IPC (Interpersonal Communication), ISM (Interactive Storage Media), and NDB (Network Database) in addition to satellite or terrestrial television broadcasting, HDTV, and digital audio broadcasting (DAB). It has various application fields such as Service, Digital Storge Media (DSM), Electronic Cinema (EC), Home Television Theater (HTT), and ISDN, and is particularly suitable for the exchange and utilization of information between broadcast and heterogeneous types.

Among the MPEG-1 and MPEG-2 compression schemes, new coding schemes for audio signals use recognition properties of human sounds by spatial and temporal masking effects. Efficient use of the quantizer using the masking effect ensures that the error signal generated in the processing unit band is within the limits that humans do not feel, and satisfies the condition where the listener cannot recognize the difference in sound quality, thereby obtaining the data compression effect. .

Meanwhile, the basic structures of the audio encoder / decoder using human psychoacoustics are shown in FIG. 1. Here, the structure of the audio encoder (a) is composed of a time / frequency mapping unit 11, a human psychoacoustic modeling unit 12, a quantization unit 13, and a bitpacking unit 14, and an audio decoder (b). Is composed of a frame unpacking unit 15, a restoring unit 16, and a reverse mapping unit 17.

In the audio encoder, the time / frequency mapping unit 11 divides the signals according to the processing unit bands that the human feels in order to consider the human psychology, and the human psychoacoustic modeling unit 12 QMF (Quadrature Mirror Filter) The band filtering method by the method or the method by the DFT or MDCT conversion is used for processing, and the masked thresholds which are threshold values which are not felt even if it considers a human psychoacoustic to an input signal are calculated using well-known laws. By using the output values of the time / frequency mapping unit 11 and the human psychoacoustic modeling unit 12, the bits are allocated according to the importance of each processing unit band, and the quantization unit 13 displays the values according to the determined number of bits. These quantized values are then packed in the bit packing section 14 for bit strip configuration.

In the time / frequency mapping section of MPEG-1 and MPEG-2 audio, signals are divided according to the processing unit bands that the human feels in order to consider human acoustic psychology. A polyphase structure was adopted and the coefficients were adjusted to reduce the effect of aliasing in the implementation. At this time, since the filter coefficients of the analysis window and the synthesis window used for encoding and decoding are different, each coefficient must be stored in the implementation. In this case, the filter coefficients are designed to have symmetry, and each of the 512 filter coefficients can be represented by 256 coefficients. An example of the filter coefficient is shown in FIG. 14, where 1 to 511 represent the index of the filter coefficient. If the index is 0, the coefficient is 0 and is not used because it does not affect the result of the operation. The coefficient when the index is 255 is the same as the case of 257, and the case of 2 is the same as the case of 510.

Meanwhile, the encoding / decoding process of MPEC audio is shown in FIG. 2. First, in the encoding process (a), each input signal is filtered, and then bits are allocated using masking thresholds calculated by FFT signals and human psychoacoustic laws, and transmitted according to a fixed bit rate. After the data is determined, the quantization step size is determined, and the samples of each band are quantized. Decoding process (b) shows that after receiving the encoded bitstream, reconstructing the samples of each band with the bit allocations belonging to the head information and the scale patterns of each band, and then passing the synthesized subband filter to Get signal.

Here, the analysis subband filter at the time of encoding and the synthesis subband filter at the time of decoding are as shown in FIG. Ci and Di, which are used to obtain Zi and Wi formed by 512 samples of each step in FIG. 3, are analysis window coefficients and composite window coefficients, respectively.

The psychoacoustic model of human being calculates the minimum masking threshold used to determine the noise value that can be felt in each processing band divided by band. The difference between the maximum signal magnitude and the minimum masking threshold is assigned bits or noise to determine the quantization step size of the actual quantizer of each processing band of each block. Two types of models are covered in the MPEG standard, each model available for all layers, but model 1 for layers 1 and 2 and model 2 for layer 3. In each psychoacoustic model, the final output is the signal to masked threshold ratio (SMR) of each processing band in the case of layers 1 and 2, and the SMR of the group of bands in layer 3.

These methods have been developed such as joint stereo coding (Joint Stereo Coding) method to improve the coding efficiency at a lower bit rate, and a lot of research is being conducted.

On the other hand, if only the encoder is implemented, an analysis window is required. If only the decoder is implemented, a synthesis window is required. Therefore, each coefficient may be stored in 256 registers. If implemented in ASIC or software, there is a problem that all 512 registers are used to store each coefficient value. For example, when video and audio signals are stored in real time, such as a camcorder, and in some cases connected to a TV and outputted, both an encoder and a decoder are required. In this case, 512 registers are used to store the analysis window coefficients and the composite window coefficients considering only symmetrical points.

Accordingly, an object of the present invention is to store the coefficients by implementing a filter using the properties of the analysis window coefficients and the synthesis window coefficients and the compensation coefficients when the audio encoder and the decoder are implemented in one chip to solve the above problems. An audio encoder / decoder is provided to reduce the registers used to 256.

Another object of the present invention is a modified synthesis window coefficient that can be used in MPEG audio with improved compensation coefficient, transfer function characteristics, modified analysis window coefficient for modified calculation coefficient between synthesis window and analysis window, modified synthesis An audio encoder / decoder for calculating the operation result of a modified analysis window with a window is provided.

In order to achieve the above objects, in the audio encoder / decoder according to the present invention,

Audio encoder

An analysis subband filter dividing the N audio subbands into N subbands according to a predetermined processing unit band with respect to an input audio signal:

A psychoacoustic modeling unit for considering a human cognitive ability according to the frequency characteristics of the audio signal:

A quantizer for allocating and quantizing bits to each subland by the output of the analysis subband filter and the psychoacoustic modeling unit; And

It consists of a packing unit for forming the quantized values in the bitstream with the header information,

Audio decoder

A header detector for solving a bitstream output from the audio encoder; and

And a synthesized subband filter for recovering the original audio signal from the unpacked bitstream.

Next, the present invention will be described in detail with reference to the accompanying drawings.

The filter used in the audio encoder / decoder according to the present invention is processed into 32 bands to consider human hearing psychology, which can be implemented by QMF filters as shown in FIG. In the case where the filter tap is L, it is required to calculate L * (1 + 2 + 4 + 8 + 16) = 31 * L = (number of bands-1) * L when processing by dividing into 32 bands. In order to reduce the calculation amount due to these filter coefficients, decimation is performed considering the Nyquist sampling, which is a sampling limit of the band-limited signals. As shown in FIG. 5, the number of operations when implemented by polyphase can be reduced to L. FIG.

If the sampling limit is considered in processing, such as QMF or polyphase, there is a problem that aliasing occurs due to decimation. This problem can be solved by modifying the filter coefficients. The analysis window and synthesis window used in MPEG use the filter coefficients by polyphase considering the aliasing.

6 and 7 show the transfer function and its filter coefficients in the case of the prototype of the polyphase filter considering the aliasing used in MPEG. In the present invention, the analysis window coefficients and the synthesis window coefficients as shown in Figs. 8 and 9, which have been de-aliased using the prototype filter coefficients of Fig. 7, are obtained and used for MPEG audio.

The sum of the prototype filter coefficients used in MPEG audio is not one due to rounding or the like. In the present invention, the prototype filter coefficients used in MPEG are modified to be 1 to obtain better transfer function characteristics. The transfer function and filter coefficient of the resulting prototype filter are as shown in FIG. 10 and FIG. Comparing FIG. 6 and FIG. 10, it can be seen that there is no significant difference in the transfer function except in the cases indicated by circles. The analysis window and the synthesis window which reduced the aliasing using the obtained prototype filter coefficients are as shown in FIG. 12 and FIG.

In the present invention, in order to make it easier to calculate the result of the calculation, the analysis window coefficient and the composite window coefficient have a coefficient difference of two squares so that it is possible to obtain the result of the operation by taking only the upper k bits of the operation result as a result. .

That is, when y = x ₁ + x ₂ + x _3, when x ₁ + kc ₁ , x ₂ = kc ₂ , x ₃ = kc ₃ ,

y = k (c ₁ + c ₂ + c ₃ ) or y ₁ = c ₁ + c ₂ = c ₃ . Same as obtaining y = ky ₁ . In this case, since k is an exponential multiple of 2, for example, if k is 32, a power of 2 is increased by 5 bits so that a 32 times effect can be obtained by shifting the y ₁ operation result by 5 bits.

Hereinafter, an embodiment of the polyphase filter applied to the present invention will be described.

The filter Where the filter tab is N. If you write this

Becomes

Polyphase filter applied to the present invention

and,

In this case, when g _O (i) and h _O (i) is g _O = kh _O (i), with g ₀ (i) or h ₀ (i) and k, y ₁ (n) And y ₂ (n) was found. That is, if only h ₀ (i) and k are known. If the y ₁ (i) first determined, and then, y ₂ (i) = can be obtained for y ₂ (i) by a _{ky 1 (i), g 0} (i) and know only k, y a ₂ (i) First, then Y ₁ (i) can be obtained. That is, since g ₀ (i) and h ₀ (i) need not be held at the same time, the number of registers used can be reduced. Here, we show that when the analysis window coefficient and the composite window coefficient have constant constants, it is not necessary to store both values.

The polyphase filter applied to the present invention can be applied to an audio device, an HDTV, a camcorder or the like employing the MPEG audio system.

As described above, the audio encoder / decoder according to the present invention is a codec that implements an encoder and a decoder that are compatible with MPEG audio, and has a number of registers used for storing analysis window coefficients and synthesized window coefficients from 512 to 256. This reduces the complexity of the hardware. In addition, by making the coefficient ratio of the analysis window and the synthesis window more than two exponents for easier hardware implementation, by shifting bits using the shifter or taking specific bits of the result when calculating using the analysis window or synthesis window. The result of the operation can be obtained.

1 is a diagram illustrating an audio codec considering human hearing psychology.

2 shows an MPEG audio encoder / decoder.

3 shows a synthesis / analysis filter of MPEG audio.

4 is a diagram for explaining a method for implementing a 32-band filter by QMF.

5 is a diagram for explaining a method of implementing a 32-band filter by polyphase.

6 is a transfer function of a prototype filter used in MPEG.

7 shows the coefficients of the prototype filter used in MPEG.

8 shows analysis window coefficients used in MPEG.

9 is the synthesis window coefficients used in MPEG.

10 is a transfer function of a prototype filter used in the present invention.

11 is the coefficient of the prototype filter used in the present invention.

12 is an analysis window coefficient used in the present invention.

13 is a synthesis window coefficient used in the present invention.

14 shows registers using symmetry of filter coefficients.

Claims (10)

Audio encoder An analysis subband filter for dividing the input audio signal into N subbands for each predetermined processing unit band; Psychoacoustic modeling unit for considering human cognitive ability according to frequency characteristics of the audio signal; A quantizer for allocating and quantizing bits to each subband by the output of the analysis subband filter and the psychoacoustic modeling unit; And It consists of a packing unit for forming the quantized values in the bitstream with the header information, Audio decoder A header detector for solving a bitstream output from the audio encoder; and An audio encoder / decoder comprising a composite subband filter for reconstructing the original audio signal from the unpacked bitstream. The audio encoder / decoder of claim 1, wherein values of analysis window coefficients of the analysis subband filter and synthesis window coefficients of the synthesis subband filter have a constant k times. 3. The audio encoder / decoder of claim 2, wherein the constant k is represented by N powers of two. 4. The audio encoder / decoder according to claim 3, wherein analysis window coefficients are obtained by dividing the synthesis window coefficients by k. 4. The audio encoder / decoder of claim 3, wherein a composite window coefficient is obtained by multiplying the analysis window coefficient by k. The audio encoder / decoder according to claim 1, wherein the analysis window of the analysis subband filter is symmetrically designed to represent all analysis window coefficients using only N of 2N coefficients. The audio coder / decoder according to claim 1, wherein the synthesis window of the synthesis subband filter is symmetrically designed to represent all synthesis window coefficients using only N out of 2N coefficients. The compound window according to claim 1, wherein when the analysis window coefficient and the synthesis window coefficient are designed with a constant k times, the calculation result by the analysis window coefficient is shifted to the right by N bits when the calculation is performed using the analysis window coefficient and the constant k. An audio encoder / decoder, which obtains processing results. The analysis window according to claim 1, wherein when the analysis window coefficient and the synthesis window coefficient are designed with a constant k times, when the calculation is performed using the synthesis window coefficient and the constant k, the calculation result by the synthesis window coefficient is shifted to the left by N bits. An audio encoder / decoder, characterized by obtaining processing results. 2. The audio encoder / decoding according to claim 1, wherein when the analysis window coefficient and the synthesis window coefficient are designed with a constant k times, the analysis window processing result is a value excluding the lower N bits from the operation result of the synthesis window coefficient. group.