KR100300957B1 - Digital audio encoding method using lookup table and apparatus for the same - Google Patents

Abstract

PURPOSE: A digital audio encoding method using a lookup table and an apparatus for the same are provided to reduce a delay time in an audio encoding process by storing the amount of bit allocation according to a characteristics of an input audio signal in a lookup table. CONSTITUTION: A frequency mapping portion(31) is used for dividing time area input audio data into 32 frequency bands by passing the audio data through a band resolution filter. A characteristic value acquirement portion(33) obtains an address of a lookup table(35) by calculating occupation ratios for a scale factor and a mean square of an input signal and selecting a larger one of the occupation ratios as the address of the lookup table(35). The characteristic value acquirement portion(33) determines an address of the lookup table(35) according to the occupation ratios. The lookup table(35) has the number of allocated bits for protecting a frequency band. A bit allocation and quantization portion(37) allocates bits according to each band by using the obtained lookup table(35) and controls the amount of total bit. A frame packing portion(39) forms the quantized audio signal as a bit stream.

Description

Digital audio encoding method and apparatus using lookup table

1 is a block diagram showing a typical MPEG audio coder.

2A is a diagram for explaining an apparatus for generating a lookup table.

2b is a view for explaining a process of generating a lookup table.

3 is a block diagram showing a digital audio encoding apparatus using a lookup table according to the present invention.

4 is a graph comparing the performance of the MPEG apparatus according to the NMR of the left channel and the encoding apparatus according to the present invention.

5 is a graph comparing the performance of the MPEG apparatus according to the NMR of the right channel and the encoding apparatus according to the present invention.

The present invention relates to a digital audio encoding method and apparatus, and more particularly, to a digital audio encoding method and apparatus using a look-up table (hereinafter referred to as LUT).

Today's communication technology is changing everything from analog to digital. In response to this trend, digital transmission is indispensable for any audio device or audio transmission. The transmission of digital audio is more resistant to ambient noise than conventional analog transmission, and the sound quality can be reproduced very cleanly as in a compact disc (CD). However, as the amount of data to be transmitted increases, various problems such as the amount of memory to be stored or the capacity of a transmission line have caused.

The technology required to solve this problem is data compression technology. In the case of audio, the goal of audio compression technology is to compress the original sound, transmit it, and then uncompress and play it almost like the original sound when heard. In other words, it is possible to transmit less information per unit time while playing the same sound quality.

This technology is currently underway around the world, and the starting point is the mini disk (MD) made by SONY of Japan in 1992 and the digital compact cassette (DCC) made by Philips. In the case of MD, for example, the CD-level sound quality is relatively smaller than a conventional CD, and the compression ratio is about 5: 1, so that a much larger amount of data can be stored than a CD. Has characteristics.

Meanwhile, the International Standards Organization for Digital Compression Encoding Technology, or Moving Picture Experts Group (MPEG), was established worldwide. MPEG is largely composed of system, video and audio subdivisions. The audio part is divided into three layers again.

MPEG compares, analyzes, and tests many low-rate coding techniques proposed to establish international standards for moving images and the encoded surfaces of audio signals. Once these international standards have been produced, all future digital storage media will need to encode and store the data in accordance with this standard. Here, the digital storage medium includes a CD-ROM, a digital audio tape (DAT), a magneto-optical disk (MOD) and a computer disk.

A technique commonly used in compression encoding of an audio signal is a method using a human psychoacoustic model. Using masking phenomenon and critical band among auditory characteristics, it removes signals that humans cannot feel and encodes only the signals that must be present and allocates bits by using less bits than the original signals. The sound quality is almost the same as that of the original sound.

Here, the masking phenomenon refers to a phenomenon in which a human face does not feel at all by masking another signal due to interference among audio signals. In addition, the critical band is a kind of unit in which humans distinguish frequencies of sound, and is generally divided into 24 bands. As we move towards the higher frequencies, the magnitude of this force becomes larger on a log scale. Therefore, the human ear is not easy to distinguish frequencies for high frequency signals rather than low frequencies.

In order to allocate bits using these auditory characteristics, a signal-to-noise ratio (hereinafter referred to as SNR) and a signal-to-mask ratio (hereinafter referred to as SMR) are obtained as these values. The mask-to-noise ratio (hereinafter referred to as MNR) must be calculated again. Here, the mask level means the minimum signal level that humans do not feel. Therefore, it is not necessary to allocate bits for signals below this mask level.

Through the above process, the final MNR is obtained and bits are repeatedly allocated based on this value. However, it takes a lot of computation time during this series of processes, which means that the real-time delay in the encoder is increased. Therefore, there is a need to reduce the complexity of the computation.

Meanwhile, a general MPEG audio encoder will be briefly described with reference to FIG. 1 as follows.

The frequency mapping unit 11 converts the audio data in the time domain into 32 equal bands in the frequency domain by using a band decomposition filter. In this case, there are 12 samples in each band in layer I and 36 samples in layer II. On the other hand, the scale factor is a value that decreases by 1 / {(2 ^1/3 )} ^N every three steps from the maximum value 2.0, and there are a total of 63 steps as shown in Table 1. Therefore, the number of bits required to encode this is 6 bits.

The encoding method has some differences depending on the hierarchies. In Layer I, the largest value among the 12 samples in each band is obtained, and a value equal to or slightly larger than this value is used as a scale factor of the corresponding band. On the other hand, in layer II, since three scale factors exist in each band, the similarity of each scale factor is examined to determine how many of the three are encoded. In other words, the difference between the neighboring scale factors is selected differently according to the range of the value. Therefore, unlike in layer I, information that selects a scale factor is additionally needed. In this case, two bits are encoded.

The psychoacoustic model 13 is a part of the computational complexity that is the largest in the encoder, and the final output value of the psychoacoustic model is an SMR of each band and serves as a criterion for bit allocation. The SMR value is calculated by the following series of steps. In the first step, the audio signal in the time domain is converted into the frequency domain by a fast Fourier transform (FFT), in the second step, a sound pressure level of each band is calculated, and in the third step, an absolute masking threshold ( Absolute Threshold) is calculated, in the fourth step the voiced and unvoiced components of the audio signal are determined, in the fifth step the masker is determined, in the sixth step the respective masking thresholds are calculated, and in the seventh step, the overall masking is performed. The threshold value is calculated, and in the eighth step, the minimum masking threshold of each band is calculated, and in the ninth step, the SMR value of each band is calculated.

In the bit allocation and quantization unit 15, first, the bit allocation process repeatedly performs the following series of steps based on the SMR value in the psychoacoustic model 13 to obtain the bit allocation of each band. In the first step, the initial allocation bit is set to 0, and in the second step, an MNR value is obtained for each band, where the MNR value is obtained by subtracting the SNR value from the SNR value. In the third step, the band having the minimum MNR is found among the MNR values obtained for each band, and the number of allocated bits is increased by one. In the fourth step, if the required number of bits is not exceeded, the second to third steps are repeated for the remaining bands. do.

On the other hand, the quantization process is performed through a series of steps as follows. In the first step, samples are divided into scale factors within each band, and X is calculated. In the second step, A * X + B (where A and B are predetermined table values) is calculated. As many as the number of allocation bits obtained in the bit allocation process are taken, and in the fourth step, the most significant bit (MSB) is reversed.

As described above, in the conventional digital audio encoder, since the psychoacoustic model is used, a nine-step process is required to obtain an SMR value, which increases the complexity of the operation and greatly affects the overall execution time. In addition, since the MNR is calculated again using the SMR value obtained by the above method, and the bit allocation loop is repeatedly performed based on the MNR, time delay also occurs in this process.

As a result of the actual experiment, it can be seen that the computational complexity is high enough that the auditory psychology model and bit allocation process takes about 49.9% in the execution time of the entire encoder as shown in Table 2 below. Therefore, the need for an alternative to this process has emerged.

Accordingly, an object of the present invention is to create a lookup table in advance that the bit allocation according to the characteristics is made by identifying the characteristics of the input audio signal while being compatible with the existing MPEG to reduce the delay time required for audio compression encoding in order to solve the above problems. Next, there is provided a digital audio encoding method for allocating bits by using a created lookup table.

Another object of the present invention is to provide an apparatus most suitable for realizing the digital audio encoding method.

In order to achieve the above object, the digital audio encoding method according to the present invention includes the number of allocation bits for encoding the frequency band, using the occupancy ratio of the characteristics of the frequency band including the scale factor and the root mean square. Creating a look-up table stored in the calculated address in advance; Dividing the time domain input audio signal into a plurality of same frequency bands; Calculate the scale factor and square average of each frequency band, determine the share of scale factor and square average of each band, compare the share of each frequency band, calculate the address of each frequency band using the largest share, Determining the number of allocated bits for each frequency band, including extracting the number of allocated bits from a lookup table prepared in advance using an address calculated for each frequency band; Allocating bits to frequency bands corresponding to the extracted number of allocated bits and quantizing the allocated bits; And forming a quantized audio signal from the quantized bits into a bitstream.

In order to achieve the above object, the digital audio encoding apparatus according to the present invention comprises: a frequency mapping unit for dividing a time domain input audio signal into a plurality of same frequency bands; A lookup table including allocation bit numbers for encoding a frequency band, wherein the allocation bit numbers are stored at an address calculated using a share of characteristics of the frequency band including a scale factor and a squared average; A characteristic acquisition unit for determining an address of a lookup table according to a share of the characteristics of the input audio signal; A bit allocation and quantization unit for allocating bits corresponding to addresses in the lookup table to each frequency band and quantizing the allocated bits; And a frame packing unit configured to form a quantized audio signal from the quantized bits into a bitstream, wherein the characteristic obtaining unit comprises: a calculating unit calculating a characteristic of each frequency band of the input audio signal; And an address calculator for comparing the determining unit and the occupancy rate of each frequency band and calculating an address for each frequency band using the largest occupancy rate.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

In the present invention, the most important bit allocation method by LUT is to remove the psychoacoustic model part and bit allocation loop part of FIG. 1 and to perform bit allocation by using a previously obtained LUT. The digital audio encoding apparatus proposed by the present invention is shown in FIG. First, the LUT generation process required for bit allocation will be described.

FIG. 2A is a diagram for explaining an apparatus for creating a LUT. The frequency mapping unit 21, the psychoacoustic model 23, the bit allocation unit 25, the characteristic value acquisition unit 27, and the LUT preparation unit 29 are shown. ).

2B is a flowchart for explaining a method for creating a LUT. Referring to FIG. 2B, the operation of the apparatus for creating the LUT is described in detail.

First, exactly the input audio data is subjected to band resolution filtering through the frequency mapping unit 21 and then divided into a plurality of frequency bands (step 210). The scale factor and the root mean square are calculated as the characteristics in the time domain for each band through the characteristic value obtaining unit 27 (operation 220).

Specifically, the LUT is a table that determines bit allocation for each band, and is used to minimize the time required for bit allocation. To do this, the overall execution time can be reduced without analyzing the input signal in the frequency domain such as the FFT or power distribution used in the psychoacoustic model. Therefore, consider the characteristics in the time domain for the input signal. In other words, when an input signal comes in, it does not interpret it in the frequency domain, but finds the characteristic immediately in the time domain and sets the values as a standard for LUT creation. In the present invention, these characteristics according to the psychoacoustic model are defined by the scale factor and the root mean square. Of course, these characteristics are used to determine the LUT address immediately after passing through the bandpass filter of the frequency mapping unit 11 in FIG. First, we will briefly discuss these features and how to apply them to an encoding algorithm.

The final output value of the psychoacoustic model 23 is SMR, which means a difference between a signal level and a masking level for each band. In order to obtain this value, the signal level must first be calculated. In the psychoacoustic model 23, a larger value of power of each band and power spectrum sf _max of the scale factor is set as the signal level of the band. This is represented by the following equation. Here, Lsb (n) is the sound pressure level of the band n, which is the same as the S level of the SMR.

Lsb (n) = MAX [power (n), 20log (sf _max (n) * 32768) -10] dB ... (1)

In the first equation, the factors that determine the signal level of each band are the scale factor and power value of each band. In the psychoacoustic model, this power was obtained through frequency conversion, but inferred from the time domain, it can be calculated as the squared mean of sample values for each band after passing through the band-pass filter. Since the scale factor is the largest sample value in each band, it can be directly obtained in the time domain.

In general, if the power of the audio signal is large, a lot of bits should be allocated, so the power in the frequency domain may be determined by the magnitude of the sample values in the time domain. Considering this, it can be said that the scale factor, which is the largest value of each band, is related to power, thereby determining bit allocation.

Specifically, in writing a LUT using the characteristics of the audio signal, the most important consideration is how to use these characteristics to find the address of the LUT. That is, the method of writing LUT using these two characteristics, namely scale factor and root mean, is as follows.

Referring back to FIG. 2B, after step 220, the feature value acquisition unit 27 calculates the occupancy ratio of the scale factor and the root mean square for each band for the entire band (step 230). Next, the share of each band is compared and an address for each frequency band is calculated based on the largest share (step 240).

The LUT is used to store the address and bit allocation for each band and is used to minimize the time required for bit allocation. In other words, the address determined by the above two parameters should be found in the LUT and assigned to the corresponding band immediately. The bit allocation method in MPEG increases by one bit from the band having the largest value from the band-specific SMR values obtained from the psychoacoustic model. This is a method of processing relative SMR values within one frame in consideration of cross-band correlation. In consideration of this point, the present invention introduces the concept of occupancy SR (n) to obtain a relative value of each band of the scale factor and the root mean square when obtaining the address of the LUT.

The occupancy SR _sf (n) for the scale factor of the band n is a value obtained by dividing the scale factor _sf (n) of the band n by the sum of the scale factors sf of the entire band and can be expressed as in the second equation.

On the other hand, occupancy SR _pwr (n) with respect to the square root of the band n is a value obtained by dividing the square root of the band n (pwr (n)) by the sum of the square roots of all the bands, and can be expressed as in Equation 3 below.

Here, the whole band means both channels, for the purpose of association with the joint mode. This occupancy is an indicator of how much the value of one band occupies. Therefore, occupancy is a relative value that takes into account the total values, which is a concept similar to the iterative bit allocation loop of MPEG.

A large value is selected from the scale factors for each band and the occupancy rate of the root mean square, and used to obtain the address of the band. The reason for this is that both characteristics can be used to calculate the address of the LUT, but the disadvantage is that the larger the address, the larger the total LUT size. In the psychoacoustic model, as shown in Equation 1, the larger of the two is determined as the signal level of the band, and thus, the present invention has similar processing.

The occupancy is all less than 1. In the present invention, these values are converted to integer values using the scale factor 63 mapping table of Table 1 and then addressed to the LUT. In this case, 62 may be the largest address in each band.

Referring to FIG. 2B again, the number of split bits for each band is calculated through the bit allocator 25 based on the SMR obtained using the psychoacoustic model 23 (step 250). After operation 250, since the bit allocation by the psychoacoustic model and the occupancy ratio of the two characteristics are obtained for each band, the allocation of the number of bits for each band is obtained and stored in the corresponding address (operation 260).

After the final occupancy rate of each band is obtained, the bit allocation corresponding to these values is investigated using a lot of audio data and stored in the LUT. That is, the bit allocation through the selected audio data is investigated using the MPEG algorithm using the psychoacoustic model, and the occupancy rate of each band is determined to determine the bit corresponding to the occupancy rate as the bit corresponding to the occupancy rate.

For example, in one band, if the bit allocation by the psychoacoustic model is 7 in the band 0 and the occupancy ratio is 5, the frequency is increased by 1 in the 7-bit region of the address 5. This process is repeated for a plurality of audio data (step 270), and the most frequent value is determined as the bit allocation of the band (step 280). As shown in Table 3 below, 8 bits are used for sculpture 5, 7 bits for address 30, and 0 bits for address 61 and 62, respectively. If there is more than one band with the highest frequency, the higher bit is selected.

On the other hand, since the LUTs thus obtained have a total of 32 bands of 62 for each band, a total of 1984 pieces of information are stored. Since this is a large capacity of about 25 kbytes, the optimization process is added to reduce the total size and use the memory more efficiently (step 290). That is, if the last bit allocation is 0 as shown in addresses 61 and 62 of Table 3, the bit value is changed by processing to store only the bit allocation of the smaller address if the same bit allocation is made between consecutive addresses. Only the address of the boundary is used as information. This optimization can significantly reduce the amount of memory to store the LUT.

Next, audio encoding using the generated LUT will be described.

3 is a block diagram showing a digital audio coding apparatus using a look-up table according to the present invention, which includes a frequency mapping section 31, a characteristic value obtaining section 33, a LUT 35, a bit allocation and quantization section 37, The frame packing part 39 is comprised.

Referring to the operation based on the configuration of FIG. 3, the frequency mapping unit 31 first divides the audio data through a band-pass filter to be divided into 32 frequency bands, and in each band, 12 samples for layer I and 12 samples for layer II. There are 36 samples.

In order to obtain the address of the LUT 35, the characteristic value acquisition unit 33 calculates the occupancy ratio of the scale factor and the root mean square of the input signal by the above-described method for each band, and then selects a larger value among them to select the LUT 35. ) As an address.

The bit allocation and quantization unit 37 allocates bits for each band by the LUT 35 obtained by the method described above, and then checks whether there are remaining bits or exceeded in comparison with the required number of bits, and then uses the total bit usage. Adjust the Of course, when the bit is allocated by the LUT 35, it is preferable to allocate the bit in accordance with the required number of bits. The reason is that the processing speed can be reduced by that, because no additional adjustment is required.

First, when the bits are allocated by the LUT 35 and then the processing process when the required number of bits is exceeded, the first step is to examine the high bands for both channels and have the smallest occupancy and the band allocated to one or more bits. Decreases by one bit from the command until it does not exceed the required number of bits. At this time, the band that has been reduced once is assigned the priority as the last, so that all bands can be reduced evenly. And, from the high band, it is generally because important information is concentrated in the low frequency side.

On the contrary, after the bit allocation by the LUT 35, if there is a bit of whether the total number of bits used is smaller than the required number of bits and can be allocated, the low band is used to investigate the largest occupancy and the maximum bit for each band. It increments by one bit from the band not exceeding the quota, and repeats until the required number of bits is not exceeded.

This bit adjustment process greatly affects the reduction of the overall execution time required for audio encoding. That is, it is necessary to create an accurate and reliable LUT 35 because the number of repetitions of this bit adjustment process additionally depends on how exactly the bit allocation by the LUT 35 is matched to the required number of bits. have.

Joint-stereo mode, on the other hand, derives from the human psychoacoustic characteristics and generally takes advantage of the fact that the audio source's ability to accurately locate audio sources decreases toward higher frequencies. In general, the hardware complexity is almost the same as the stereo coding scheme, and the delay time of the encoder is almost unchanged. The main purpose of this mode is to increase the audio quality of the encoding, which can be reduced to about 10 to 30 kbps at a bit rate.

Looking at the joint-stereo coding scheme standardized in MPEG audio, a specific band is determined, and above that band, the samples are summed and only one is encoded without obtaining the samples. In the range determination of a specific band to be joint-stereo coded, the bit allocation for causing the noise of each band to be lowered below the mask level is determined to examine whether or not the number of bits is greater than the required number of bits. Select one of four bands. In the joint-stereo band thus determined, a large bit allocation among left and right channels is selected and quantized.

The problem when applying the concept to the LUT in the present invention and encoding is to determine the range of band to be encoded by joint-stereo. In the present invention, by examining the occupancy of each band, the range is determined by examining where the total of the occupied bands exceeds 99%. In the band to be encoded with joint stereo, a scale factor and a root mean square value of the left and right channels are respectively selected and used to obtain the occupancy ratio. When each occupancy rate is obtained, the sum of all bands is smaller than in stereo mode, so each occupancy rate is larger than in stereo. Therefore, when coding in joint-stereo mode, the other bits are calculated so that the bit allocation is different. Therefore, it is possible to encode in stereo and joint-stereo mode using one LUT.

In fact, to test the performance of the present invention, twelve audio data were used to specify the total execution time. The experimental environment uses SUN SPARC-10 under Unix system, and audio data is obtained from CD.

The following Table 4 shows the execution time in the encoder and compares the performance of the encoding method using the UUT proposed in the present invention and the existing MPEG algorithm.

In Table 4, twelve audio data are given an arbitrary name, and the numbers represent actual execution times. The performance improvement is obtained by the following equation.

First, looking at the execution speed of the algorithm proposed in the present invention, as shown in Table 4, it can be seen that a slight difference depending on the audio data, but the speed improvement of about 44% on average compared to the conventional method. In Table 4, data 1-6 show the performance rate for layer II and the rest for layer I. This means that the real-time processing speed of the encoder is improved, so that the processing delay in the encoder is reduced when the hardware is implemented. However, the bit allocation process proposed in the present invention is not performed only by the LUT, but because there is a process of processing the extra bits or the missing bits after this process, depending on how quickly the bit allocation adjustment process is performed. The overall run time can be much faster. In addition, the sound quality of the audio is almost the same as that of the conventional MPEG algorithm. In other words, up to 128 kbps in layer I and 96 kbps in layer II, almost the same CD sound quality can be heard.

On the other hand, the performance evaluation method using the psychoacoustic characteristics proposed by Karlheinz Brandenburg was used to see how the performance of the algorithm proposed in the present invention is different from the conventional MPEG. This method uses NMR as a reference scale. If this value is negative, it means that the noise is below the mask level. Therefore, it is estimated that the noise is reproduced with cleaner sound quality because it is less noise.

4 and 5 are graphs comparing the evaluation results of the two algorithms on a channel-by-channel basis, and the method using the LUT shows a relatively stronger noise performance than the LPEG method in the low frequency region. In other words, even in the same amount of required bits, more efficient use of bits is achieved in the encoding method using the LUT.

As described above, the digital audio encoding method and apparatus using the look-up table according to the present invention do not use the psychoacoustic model, thereby greatly reducing the computation time delay in the encoder, thereby increasing the possibility of real-time processing in the encoder when implemented in hardware. As an actual application example, encoding of music through a network is similar, which is similar to the VOD concept, and is a system that receives and transmits music required by a user through a network in real time.

In addition, the psychoacoustic model is difficult to implement in actual hardware because the algorithm is too complex, and has the greatest influence on determining the value of the encoder. In the present invention, since the psychoacoustic model is not used, only the memory is added. This makes it easy to implement in hardware and can reduce the cost of the encoder.

In addition, the present invention can be applied to a small device such as a portable camera because a much smaller chip can be obtained than the conventional MPEG encoder. In recent years, the use of the MPEG chip in the computer is a general trend, and when the present invention is used, it is possible to reduce the board size of the computer because it maintains the full compatibility with the MPEG, and thus it is possible to manufacture a product that pursues light and small size.

Claims (9)

A digital audio encoding method comprising: a lookup table including allocation bits for encoding a frequency band, and storing the allocation bits at an address calculated using a share of the characteristics of the frequency band including a scale factor and a root mean square. Creating in advance; Dividing the time domain input audio signal into a plurality of same frequency bands; Calculate the scale factor and square average of each frequency band, determine the share of scale factor and square average of each band, compare the share of each frequency band, calculate the address of each frequency band using the largest share, Determining the number of allocated bits for each frequency band, including extracting the number of allocated bits from a lookup table prepared in advance using an address calculated for each frequency band; Allocating bits to frequency bands corresponding to the extracted number of allocated bits and quantizing the allocated bits; And forming a quantized audio signal from the quantized bits into a bitstream. The digital audio encoding method of claim 1, wherein the generating of the lookup table comprises optimizing bit allocation by storing only a boundary portion where a bit value changes, except for an address having an allocated number of bits in the lookup table. . The method of claim 1, further comprising: comparing the requested number of bits with the actual number of allocated bits after allocating bits using a lookup table, and adjusting the allocation bits according to the comparison result. Digital audio coding method. 4. The method of claim 3, wherein when the actual bit allocation is allocated less than the required number of bits, the band having the largest occupancy rate is found from the low frequency band and increased by one bit, and the actual bit allocation is allocated more than the required number of bits. If it is, the digital audio encoding method characterized by finding the band having the smallest occupancy from the high frequency band and decreases by one bit. The digital audio encoding method of claim 1, wherein allocation numbers of bits are determined using the lookup table in a stereo mode and a joint-stereo mode without changing the lookup table. 6. The method of claim 5, wherein the method finds boundaries of frequency bands to be encoded in a joint-stereo mode by obtaining occupancy rates for each frequency band. A digital audio encoding apparatus for encoding an input audio signal, comprising: a frequency mapping unit for dividing a time domain input audio signal into a plurality of same frequency bands; A lookup table including allocation bits for encoding a frequency band, wherein the allocation bits store the characteristics of the frequency band including the scale factor and the root mean square at an address calculated using occupancy rates; A characteristic acquisition unit for determining an address of a lookup table according to a share of characteristics of the input audio signal; A bit allocation and quantization unit for allocating bits corresponding to addresses in the lookup table to each frequency band and quantizing the allocated bits; And a frame packing unit configured to form a quantized audio signal from the quantized bits into a bitstream, wherein the characteristic obtaining unit comprises: a calculating unit calculating a characteristic of each frequency band of the input audio signal; And an address calculator which compares a decision unit to determine and an occupancy rate of each frequency band, and calculates an address for each frequency band by using the largest occupancy rate. 10. The digital audio encoding apparatus of claim 7, wherein the lookup table excludes an address having an allocated number of bits from 0 and stores only a boundary portion at which a bit value changes to optimize bit allocation. The method of claim 1, wherein generating the lookup table comprises: (a) dividing the test audio input signal into a plurality of frequency bands; (b) calculating a scale factor and a root mean square for each frequency band; (c) calculating occupancy rates corresponding to the scale factor and the root mean square of each frequency band; (d) comparing the occupancy rates of each frequency band and calculating an address for each frequency band based on the largest occupancy rate; (e) determining the number of allocated bits for each frequency band by using the psychoacoustic model; (f) storing the number of allocated bits for each frequency band at an address corresponding to each frequency band; (g) repeating steps (a) to (f) and storing a plurality of allocated bit numbers in each address; And and (h) generating the lookup table by storing each address and the number of allocation bits having the highest frequency in each address.