KR19990041758A - Digital audio encoding device - Google Patents

Abstract

A digital audio encoding apparatus comprising: a plurality of A / D conversion units for converting an analog signal input from each channel into a digital signal; A plurality of slave blocks for performing subband analysis and psychoacoustic modeling on signals input from the plurality of A / D converters; A master block receiving a processing result output from a plurality of slave blocks and generating and transmitting a final bit string; A bit slave block for assisting the master block;

AES / EBU digital audio format. It is designed to fit in real time. In order to make the optimized size considering the actual size of the encoder, the redundant blocks are deleted and designed in the minimum unit required. Each channel is processed as a slave block, and as a result of the processing, a bit stream is generated by collecting in the master block, thereby performing digital encoding. The data processed in the slave block is stored and processed in the dual port RAM and is stored and processed in the dual port RAM after the processing. The data to be processed is compliant with the AES / EBU standard so as to comply with AES / EBU standards and is compatible with broadcasting equipment because it uses it as a PCM signal and uses an input signal.

Description

Digital audio encoding device

The present invention relates to a high-definition television, and more particularly to a digital audio encoding apparatus capable of high compression rate and fast data processing time of a digital signal.

As digital audio is used in various fields such as communication, computers, and home appliances, there has been a problem of storing and transmitting a large amount of data. In order to solve this problem, the Motion Picture Experts Group (MPEG) standard of the International Organization for Standardization (MPEG-1) was established in 1991 and expanded to multiple channels. In November 1994, The standard has been completed.

MPEG adopts MUSICAM (MUSICAM) method which can obtain sound quality of CD (Compact Disc) level at about 128kbit / s as a standard of audio coding method. The MUSICAM scheme is a sub-band (sub-band) coding scheme using auditory characteristics. The MUSICAM scheme codes the sound so as to minimize perceptual noise in each sub-band. The MUSICAM scheme performs a subjective restoration from 96 to 128 kbit / It was selected as a layer 2 method of MPEG in such a way that the sound can be obtained.

FIG. 5 is a block diagram showing a configuration of a general MPEG audio encoding apparatus. The A / D converter 51 receives the analog audio signal, samples it at 48 KHz, and converts it into a digital signal. The filter bank unit 52 divides the 48 KHz digital signal input from the A / D conversion unit 51 into 32 subbands and outputs the 32 subbands. The maximum value detector 53 detects a maximum value for each subband in a digital signal divided into 32 subbands output from the filter bank unit 52 and outputs the maximum value. The scale factor detector 54 detects a scale factor of each subband in a digital signal divided into 32 subbands output from the filter bank unit 52 and outputs the index. The layer and pattern detector 55 receives the index of the scale factor from the scale factor detector 54 and divides the index into hierarchies and detects the pattern. The A / D converter 51 performs fast Fourier transform The input 48 KHz digital signal is Fourier-transformed and the spectrum is output. The absolute audio masking unit 57 compares the spectrums output from the fast Fourier transform unit 56 with the decibel (dB) of the digital signal divided into 32 subbands output from the filter bank 52, And outputs it. The bit allocation unit 58 allocates a bit to the maximum spectrum output from the absolute audible masking unit 57. The encoding unit 59 converts the signals output from the layer and pattern detection unit 55 and the bit allocation unit 58 into frames and outputs the frames.

The MPEG audio standard compression method is generally divided into MPEG-1 and MPEG-2 according to the application purpose of the encoding, and is divided into the layer 1, layer 2, and layer 3 according to the bit rate. Although the basic algorithms of the MPEG-1 audio coding method and the MPEG-2 audio coding method are the same, in the case of the MPEG-1 audio coding method, only the stereo signals of up to two channels (left and right) are coded, so that the sound image localization is unstable So that there is a disadvantage that the realism can not be faithfully reproduced.

Therefore, in MPEG-2, a central channel (hereinafter referred to as C), a surround channel (hereinafter referred to as LS and RS) are assigned to the left (hereinafter referred to as L) (L, R, C, LS, RS, LFE), which is a 3/2 + 1 adopted as a recommendation of SMPTE and ITU by adding a low frequency effect channel ) Encoding algorithm.

The basic algorithms of MPEG-1 and MPEG-2 pass the input signal through 32 weighted overlap-add equal spaced filter banks in order to eliminate statistical redundancy. Band sample. At the same time, the psychoacoustic model using FFT (Fast Fourier Trandform) removes the perceptual redundancy, obtains a mask threshold value, and gives bit allocation information used for quantization. That is, bit allocation is performed so that the signal-to-noise can be masked by the signal with the output value of the filter bank and the masking threshold value. When the noise can not be completely masked, bits are allocated to each sub-band to minimize subjective noise and bit strings are created with quantized sub-band samples and side information.

In layer 1 and 2 of MPEG, we use 32 SSB filter banks of the same weighted overlap-add method. The filter used for the sub-band analysis is based on a 512-tap low-pass filter and is frequency shifted by matrix operation to become 32 sub-bands of equal size.

Using a psychoacoustic model with sub-band analysis, it is possible to determine the maximum noise level that can be masked in each subband and use it to allocate bits that determine the actual quantizer of each subband. MPEG provides two psychoacoustic models, which can be applied to appropriate applications.

The first method is to divide the FFT spectrum by the pure tone and the noise component, to obtain the masking threshold value by each component, and then to obtain the masking threshold value by considering the absolute audible limit. In layer 2, psychoacoustic model 1 is generally used to obtain the masking threshold value, which is used when a low compression ratio is required. The second method is to calculate the masking threshold value by convolving the FFT spectrum with the spreading function of the auditory neuron model, which requires a large amount of computation. However, since the masking characteristic can be more accurately modeled, it is used in applications requiring a high compression ratio . As described above, since the algorithms of MPEG-1 and MPEG-2 are very complicated and require a large amount of computation, it takes a lot of computation time and complicated structure hardware to implement it.

SUMMARY OF THE INVENTION The present invention has been proposed in order to solve the above problems,

An object of the present invention is to provide a digital audio encoding apparatus capable of realizing high compression rate and real-time processing of a digital audio signal by implementing algorithms of MPEG-1 and MPEG-2 in hardware.

It is another object of the present invention to provide a digital audio encoding apparatus capable of storing and transmitting a large amount of data using digital audio.

According to an aspect of the present invention, there is provided an apparatus comprising: a plurality of slave blocks for performing subband analysis and psychoacoustic modeling for each channel; A master block receiving the processing results output from the plurality of slave blocks and generating and transmitting a final bit string; A bit slave block for assisting the master block; AES / EBU digital audio format.

1 is a block diagram showing a configuration of a digital audio encoding apparatus according to the present invention.

2 is a flowchart illustrating a subband analysis process performed by the slave block according to the present invention.

3 is a flowchart illustrating a method of psychoacoustic model 1 performed by a slave block according to the present invention.

4 is a flowchart illustrating a bit allocation process performed by the master block according to the present invention.

5 is a block diagram showing a configuration of a general MPEG audio encoding apparatus.

Description of the Related Art

10 .... Master control unit 11 .... First local memory

12 .... Programmable logic part 13 .... Program memory

14 .... Reset switch part 15 .... Display part

16 .... clock section 21 .... slave control section

22 .... second local memory 23,24 .... dual port memory

25 .... Address decoder 100 .... Master block

201-206 .... Slave block 207 .... Bit slave block

Hereinafter, the operation and effect of the embodiment according to the technical idea of the present invention will be described in detail.

1 is a block diagram showing a configuration of an MPEG audio coding apparatus according to the present invention.

The MPEG audio encoding apparatus according to the present invention includes a plurality of A / D conversion units 31-37 for converting an analog signal input from each channel into a digital signal; A plurality of slave blocks 201-206 for performing subband analysis and psychoacoustic modeling on signals input from the plurality of A / D converters 31-37; A master block 100 receiving a processing result output from the plurality of slave blocks 201-206 and generating and transmitting a final bit string; A bit slave block 207 for assisting the master block; And a digital input unit 300 that satisfies the AES / EBU digital audio standard.

The master block 100 includes a main control unit 10 for implementing digital signal processing and each part to implement an MPEG algorithm; A first local memory unit 11 used as a subband analysis buffer and an FFT buffer; A first local memory unit 11 for storing program codes and various data necessary for the master block 100 and the slave blocks 201-206; The control signal of the dual port memory unit 24 required for information exchange between the master block 100 and the slave block 201 and the control signal of the programmable logic unit for generating the control signals of the first local memory unit 11 and the program memory unit 13 (12); A program memory unit 13 for storing a software program implementing an MPEG algorithm; A reset switch unit 14 for resetting the master block 100, the slave blocks 201-206 and the bit slave blocks 207 by generating a reset signal after power is applied; A power indicator 15 for indicating that power is applied; And a clock section 16 for supplying a main clock to the master block 100, the slave blocks 201-206 and the bit slave block 207 and synchronizing them.

The slave block 201 includes a slave controller 21 for performing subband analysis and psychoacoustic modeling on the digital signal input from the A / D converter 31-35; A second local memory unit 22 for storing necessary program codes and various data between the master block 100 and the slave block 201; A pair of dual port memory units (23, 24) for sequentially transmitting a processing result between a master block (100) and a slave block (201) storing results processed by the slave controller (21); And an address decoder unit 25 for decoding the control signal. The slave block is composed of six slave blocks configured as described above, and includes five sub-blocks 201-205 for processing signals input from five channels (L, R, C, LS, RS) And a bit slave block 206 for processing a signal input from the bit slave block 206. [

The operation and effect of the present invention will be described below.

The slave control unit 21 of the slave block 201 reads the software from the program memory 13 of the master block 100 and performs a subband analysis process.

FIG. 2 is a flowchart illustrating a subband analysis process performed by the slave block according to the present invention. The slave controller 21 reads the next subband analysis process from the program memory 13 of the master block 100 and performs the subband analysis process.

In step S21, the slave control unit 21 receives 32 new audio samples from the A / D conversion unit 31 and stores them in the internal buffer. In step S22, a vector having 512 input samples is generated using the audio samples stored in the internal buffer. The input sample is placed from 31 to 0, the most recent sample is placed at 0, and the oldest 32 samples are discarded. In step S23, the input vector generated in step S22 is multiplied by an analysis window (Window). In step S24, 512 samples are divided into 8 64 sample blocks, and each block is added to construct a new vector. In step S25, 32 sub-band samples are generated by the analysis matrix of the following equation (1) and stored in the second local memory 22.

......(One)

Using a psychoacoustic model with sub-band analysis, it is possible to determine the maximum noise level that can be masked in each subband and use it to allocate bits that determine the actual quantizer of each subband.

MPEG provides two psychoacoustic models, which can be applied to appropriate applications. The first method is to divide the FFT spectrum into pure tone and noise components and obtain the masking value by each component and then obtain the masking threshold value by considering the absolute audible limit. In layer 2, generally, the psychoacoustic model 1 is used to determine the masking threshold Value and is used when a low compression ratio is required.

The second method is to calculate the masking threshold value by convolving the FFT spectrum with the spreading function of the auditory neuron model, which requires a large amount of computation. However, since the masking characteristic can be more accurately modeled, it is used in applications requiring a high compression ratio .

3 is a flowchart illustrating a method of psychoacoustic model 1 performed by a slave block according to the present invention.

In step S31, the Hanning window given in Equation 2 shown below is applied to 1024 sample input signals except for the first 64 samples and the last 64 samples among the 1152 samples inputted in the step S31, and the FFT is calculated, Obtain the spectrum. With this spectrum, a new bit is allocated for 1152 samples.

.....(2)

..... (3)

In step S32, the sound pressure level in each subband is calculated by Equation (4).

, n: subband index .... (4)

In the hypothesis, X (k) is the spectral coefficient in each subband region and scfmax (n) means the maximum of the three scale factors per sub-band. The absolute audible limit is taken into consideration by using the data obtained through psychoacoustic experiments in step S33. An offset of -12dB is given for more than 96kbit / s, and it is used at 96kbit / s or more.

In step S34, a tonal component and a non-tonal component are found in the spectrum information. First, a pure tone component is found, and then one noise component is found in one critical band from the remaining spectrum excluding the pure tone component. The method for obtaining the pure tone is regarded as the local maxima if it is larger than the front and back signals, and it is regarded as the pure tone if it is 7dB or more higher than the surrounding signal in the frequency region given below. It searches for a narrow region in the low frequency region and searches for a wide region in the high frequency component. Each frequency domain is as follows.

.... (5)

The sound pressure level of the pure tone thus obtained is obtained by calculating X _tm = X (k-1) + X (k) + X (k + 1). The noise component is obtained as the sum of the components excluding the pure tone component in the critical band. That is, the energy sum in the critical band becomes a masking component due to noise.

When the sum of energy in the critical band is the same in psychoacoustic, it is assumed that the noise component is concentrated at the position which is the center of the geometric center in the critical band by using the same sound pressure level at any sound level, and a masking curve is obtained. The critical band is divided into 23 to 26 according to the sampling frequency. The critical band size is about 0.1 KHz in the low frequency range and up to 4 KHz in the high frequency range.

When calculating the total masking threshold value at step S35, the number of masking components is reduced to simplify the calculation. It is not considered when the sound pressure level of the pure tone masking component and the noise masking component is smaller than the absolute audible limit.

In step S36, the masking threshold value is obtained after performing sub-sampling without obtaining a threshold value for each FFT coefficient k in order to reduce the amount of calculation. In the layer 2, the sub- Without sampling, skip one of the four subbands in the next three subbands, and select one out of the four bands in the next three subbands, and in the remaining bands (sampling frequencies up to 20KHz for 1KHz or 48KHz and up to 15KHz for 32KHz) Only one threshold value is obtained. All pure tone masking components and noise components are transformed into the sub-sampled coefficients closest to the FFT coefficients k. The masking threshold by pure tone and noise is calculated by the following equation.

(6)

In the above equation, av _tm , av _nm , and vf are masking indices of pure tone and noise, respectively, and masking function, and are calculated as follows.

... (7)

-3? Dz? -1 Bark, dB

-1? Dz? 0 Bark, dB

0? Dz? 1 Bark, dB

1? Dz? 8 Bark, dB (8)

The masking function determines the slope of the masking curve according to the distance of the bark unit (dz = z (i) -z (j)). Where X [z (j)] is the sound pressure level (dB) of the jth masking element. If it is more than 8Bark than -3Bark, masking is no longer considered.

In step S37, the total masking threshold is determined as the sum of the individual masking threshold and the audible limit. When the number of pure tone masking components is m and the number of noise masking components is n, the total masking threshold value is obtained by the following equation. The minimum masking threshold value for each subband is determined by the following equation with the total masking threshold value obtained by the above method.

..... (9)

Then, the Signal to Masking Ratio (SMR) in each subband is calculated by the following Expression (10).

... (10)

As you can see from the transplant, the signal mask ratio can be obtained by subtracting the minimum masking threshold from the sound pressure level of each band. As a result, if the signal mask ratio is small, the sound pressure level of the signal is small or the masking is large. Therefore, the smaller the SMR, the more efficient quantization can be performed using fewer bits.

In order to find the scale factor for normalizing the sample value of each subband, the maximum value of 12 normalized absolute values must first be found. This value is compared with the 63 scale factors proposed by MPEG, and the scale factor immediately next to the normalized maximum value is set as the scale factor of the frame. The scale factor found in this way is not encoded, but the index of the scale factor corresponding to the value is encoded. Of course, since the decoding stage already has the codebook of the index, if only six bits required to encode 64 indexes are transmitted, the scale factor sent from the encoder can be found.

In layer 2, 36 samples (3 frames) are analyzed for each subband at a time, and three scale factors must be transmitted since the bits are uniformly allocated. When encoding this scale factor, a scale factor must be transmitted. When encoding an escalation factor, the number of bits used is reduced by using scale factor information (SCFSI).

The scale factor information is divided into five classes according to how similar the scale factors of three frames are. To classify classes, we define dscf1 and dscf2 variables as shown in Equation 11, and each dscf is given a class value according to its value. Here, scf1, scf2, and scf3 denote the scale factors of the first, second, and third frames, respectively.

... (11)

Class values assigned to dscf1 and dscf2, that is, Class 1 and Class 2, are indexed and used to find scale factor selection information (SCFSI). The number of intensities in the transmission pattern refers to the scale factor actually used in the first, second, and third frames, respectively. The numbers "1", "2" and "3" represent the scale factors of the first, second and third frames, respectively, and "4" represents the largest scale factor of the intensity.

In step S38, the scale factor is encoded by sending only one or two representative values, instead of independently transmitting the scale factors of similar size using the class and scale factor selection information. Encoding the scale factor in this way is one of the differences between layer 1 and layer 2.

The number of bits (adb) usable for encoding the sample and the scale factor is determined by the bits (bhdr) necessary for the header information (header) in the total number of bits cb and the error correction bits (bcrc) (bbal) for the auxiliary data and the bits (banc) for the auxiliary data are subtracted.

... (12)

The sample and scale factor can be encoded using the number of bits obtained by the above equation. The basic rule used in bit allocation is to minimize the noise to mask ratio (NMR) of the entire frame without exceeding the number of bits available in a frame. The number of usable bits per sample is 0 to 16 bits, and 1 bit is not used because there is no increase in signal-to-noise ratio.

Unlike at layer 1, at layer 2, the number of bits used for bit allocation information in the high-frequency band is reduced. The number of bits allocated to the high frequency band using the statistical characteristic is found to be smaller than the number of bits allocated to the low frequency band. Therefore, in the layer 2, the number of bits used for the 3-bit and 2-bit allocation information is reduced from 4 bits in the low-frequency band to the high-frequency band using a given bit allocation table. In the 30th and 31st subbands, Do not. For each subband, the MNR is obtained by subtracting the SMR from the SNR. Initially, 0 bits are assigned to all bands. After the bit (bspl) for the sample and the bit (bscf) for the scale factor are placed, the following procedure is repeated.

FIG. 4 is a flowchart illustrating a bit allocation process performed by the master block according to the present invention. The master block 100 reads the subband analysis result and the psychoacoustic model result stored in the local memory 22 of the slave blocks 201-206 through the dual port memories 23 and 24 and outputs them to the first local memory 11, And performs the next bit allocation process.

In step S41, a subband having a minimum noise-to-mask ratio (MNR) among the subbands of all channels is determined. In step S42, the step is increased in accordance with the bit allocation assigned to the subband having the minimum noise-to-mask ratio (MNR) to improve the signal-to-noise ratio (SNR). In step S43, a new noise-to-mask ratio (MNR) of the subband with the minimum noise-to-mask ratio (MNR) is calculated. In step S44, the number of usable bits is calculated by the following equation.

... (13)

It is determined in step S45 whether the number of usable bits adb is smaller than the sum of the number of sample bits bspl and the number of scale factor bits bscf. And ends the assignment process.

In order to quantize the subband samples, a mid-tread is used which is symmetric about zero, that is to say an error with respect to a small value near zero. The subband samples are divided by a scale factor and quantized through the following normalization process.

■ AX + B calculation

■ Only N MSBs (Most Significant Bits) are taken.

■ Invert the MSB.

Where N is the number of bits allocated.

First, when AX + B is calculated, the value between -1 and +1 is converted into the range of ¹ to (1-2 ^{N + 1} ), so that it becomes easy to quantize to the level of 2 ^N-1 . The reason for reversing the MSB is that the synchronization signal uses twelve " 1 "

Since three consecutive samples are coded according to the allocated bits, the use bits can be reduced. The gain that occurs when three consecutive samples are coded together is as follows. If three steps are coded, six bits are required for each encoding, but only five bits (32 steps) are sufficient because the encoding requires only 27 (= 3 ³ ) steps. If you are building a group, you only need 3, 5, or 9 steps. When three consecutive samples are coded together, the continuous samples x, y, and z are made into one encoded value v _m by the following relation.

, v ₃ is 0 .... 26

, v ₅ is 0 .... 124

, v ₉ is 0 .... 728

As described above, according to the present invention, by implementing the algorithms of MPEG-1 and MPEG-2 in hardware, a high compression rate of a digital audio signal can be achieved, real time processing can be performed, and a large amount of data can be stored and transmitted using digital audio It is possible to perform encoding of a complicated calculation amount suitable for real-time processing.

Claims (4)

A digital audio encoding apparatus comprising: A plurality of A / D conversion units for converting an analog signal input from each channel into a digital signal; A plurality of slave blocks for performing subband analysis and psychoacoustic modeling on signals input from the plurality of A / D converters; A master block receiving the processing results output from the plurality of slave blocks and generating and transmitting a final bit string; A bit slave block for assisting the master block; And a digital input unit that satisfies the AES / EBU digital audio standard. 2. The method of claim 1, A main control unit for implementing digital signal processing and each part to implement an MPEG algorithm; A first local memory unit for storing program codes and various data necessary for the master block and the slave block; A programmable logic unit for generating a control signal for each memory unit necessary for information exchange between the master block and the slave block; A program memory unit for storing a software program implementing an MPEG algorithm; A reset switch unit for generating a reset signal after power is applied to reset the master block, the slave block and the bit slave block; And a clock unit for supplying a main clock to the master block, the slave block, and the bit slave block and synchronizing them. 2. The apparatus of claim 1, wherein the slave block A slave controller for performing subband analysis and psychoacoustic modeling on the digital signal input from the A / D converter; A second local memory unit for storing necessary program codes and various data between the master block and the slave block; A pair of dual port memory units for sequentially transmitting a processing result between the master block and the slave block storing a result processed by the slave controller; And an address decoder for decoding the control signal. The digital audio encoding apparatus of claim 1, wherein the slave block comprises five slave blocks for processing a signal input in five channels and a signal input in a channel of 0.1.