KR0138300B1 - Apparatus and method for filtering digital audio - Google Patents

Abstract

Disclosed is a digital audio filtering method and apparatus for performing the method. The digital audio filtering method inputs 32 subband samples in which encoded digital audio data is decoded and dequantized, thereby reducing the factor i to 32 and the factor k to 16 in 2048 inverse discrete cosine transform coefficients (N _ik ). _an inverse discrete cosine vector step in which _ik is 128 different values and the 32 inverse discrete cosine vectors (V _i ) having subband samples (S) are obtained according to the following equations,

After the inverse discrete cosine vector step, 512 current blocks (n) of the inverse discrete cosine vector (V _i ), the window wing vector (D _i ), and the window vector (W _i ^n-1 ) of the previous (n-1) blocks A window vector step of obtaining a window vector (W _i ⁿ ) according to the following equation,

(i% p × 2 refers to the remainder after dividing i by a multiple of p, where% is the remainder operator.) Digital audio filtering device, characterized in that the window vector is stored and initialized and output as a PCM sample. Is a second storage means for storing the inverse discrete cosine transform coefficient ( _Nik ) and the windowing vector by inputting the decoded and inversely quantized subband sample encoded digital audio data, and the inverse discrete cosine transform coefficient and the subband sample. An inverse discrete cosine transform vector, a windowing vector, an inverse discrete cosine transform vector, and a window vector of a previous block, and an operation means for performing an operation to obtain a current window vector. And controlling means for generating a band sample, controlling the inverse discrete cosine transform coefficient and the windowing vector to be input to the calculation means, and controlling the operation. The circuit is simpler, and the circuit operation is simpler because it operates at a lower operating frequency, which is about 1/4 less than the conventional operating frequency, and the manufacturing cost of the circuit is low, and the number of inverse discrete cosine conversion coefficients is reduced, thereby reducing the ROM table. You can see a lot of effects in size.

Description

Digital audio filtering method and device

1 is a block diagram of a typical MPEG1-audio decoder.

2 is a flowchart for explaining a decoding method of the MPEG-1 audio decoder shown in FIG.

FIG. 3 is a flowchart illustrating an overlap add method among the filtering methods of the audio filter unit illustrated in FIG. 1.

FIG. 4 is a flowchart for explaining a method adopted in MPEG1 DIS among the filtering methods of the audio filter unit shown in FIG.

5 is a flowchart illustrating a filtering method of the digital audio filtering apparatus according to the present invention.

6 is a flowchart according to the present invention for explaining in detail the flowchart shown in FIG.

7 is a block diagram of a digital audio filtering apparatus according to the present invention for performing the flowchart shown in FIG.

8 is a detailed block diagram according to the present invention of the digital audio filtering device shown in FIG.

The present invention relates to a MPEG (MOVING PICTURE EXPERT GROUP) 1-audio decoder, and more particularly, to a filtering method for reducing the number of operations of a multiplication when performing an inverse discrete cosine transform to output a PCM sample, and an apparatus for performing the method. will be.

Moving Pictures Expert Group (MPEG) is the main task of proposing international standards for the encoding of moving picture and audio information, mainly for Digital Storage Media (DSM). Group. There are several subgroups under this MPEG, one of which is the audio subgroup, which defines the requirements for the encoding of speech and produces the standard.

When the sampling frequency of the decoder is 44.1 KHz during the IDCT (Inverse Discret Cosine Transform or lower IDCT) process of the overlap-add method proposed in the MPEG1 DIS (Draft International Standard), Before one sample is printed and the next sample is printed, i.e. for 1 / 44.1 KHz time, all operations of the decoder-sample splitting, dequantization, IDCT, filtering operations-must be completed. In the case of audio filtering using the proposed method in MPEG DIS, only 2048 multiplications are performed on 32 subband samples in the IDCT process. The minimum operating frequency is 44.1KHz × (2048 ÷ 32) = 2.8MHz. Since the frequency required only in IDCT is 2.8MHz, the minimum operating frequency should be about 5MHz or more considering inverse quantization and many other calculations. Since the most multiplication is done during the multiplication process, there is a limit to the slower operating frequency, which requires the use of expensive chips with precise circuits, and a large current consumption, and the size of the ROM table in which the values of various parameters are stored. There is a problem in that the size of the chip is small.

An object of the present invention is to provide a digital audio filtering method and apparatus which drastically reduced the number of multiplications in the IDCT process in order to solve the above conventional problems.

In order to achieve the above object, a digital audio filtering method according to the present invention includes a decoding step of decoding additional information by inputting encoded digital audio data, an inverse quantization step of dequantizing the decoded samples and outputting a subband sample; In the digital audio decoder, the audio filtering step comprises inputting the subband sample to inverse discrete cosine transform to produce a window vector, and then outputting PCM samples. A band sample is input and the value of the factor i is determined from an inverse discrete cosine transform coefficient ( _Nik : where i and k are factors for designating an address where the inverse discrete cosine transform coefficient is stored). Reduce to a third constant, reduce the value of the factor k to a fourth constant, and make the inverse discrete cosine transform coefficient have different values of the fifth constant. And that the sub-band sample (S) has along the first inverse discrete cosine vector (V _i) of the predetermined number of the following formula to obtain the vector inverse discrete cosine phase,

After the inverse discrete cosine vector step, the sixth constant current from the inverse discrete cosine vector (V _i ), the windowing vector (D _i ), and the window vector (W _i ^n-1 ) of the previous (n-1) block. (n) a window vector step of obtaining a window vector (W _i ⁿ ) of the blocks according to the following equation,

(i% p × 2 means the remainder after dividing i by a multiple of p, where% is the remainder operator.) When vector columns of the window vectors consecutive with the first integer are the fifth integer. A storage step of sequentially storing the first vector column to the last vector column in the PCM output buffer; and all the vector columns not yet stored as empty positions of the window vector after the vector string is stored in the PCM output buffer. And an initializing step of initializing the position of the window vector remaining after moving them, and an outputting step of outputting the vector strings stored in the PCM output buffer to the PCM samples after the initializing step.

In order to achieve the above object, the digital audio filtering apparatus according to the present invention includes decoding means for decoding additional information by inputting encoded digital audio data, and first memory for dequantizing the decoded samples and storing a subband sample. Dequantization means including means and an address generating means for generating an address storing said dequantized subband sample, and inputting said subband sample to inverse discrete cosine transform to form a window vector, and then output a PCM sample. In the digital audio decoder comprising an audio filtering means, the audio filtering means is an inverse discrete cosine transform coefficient ( _Nik : where i and k are factors for designating an address stored in the inverse discrete cosine transform coefficient) and a window. Input second storage means for storing a wing vector, the inverse discrete cosine transform coefficient and the subband sample; A calculation means for obtaining the inverse discrete cosine transform vector, receiving the windowing vector, the inverse discrete cosine transform vector, and the window vector of the previous block, and calculating a current window vector; and generating the address. And controlling means for generating an additional subband sample, controlling the inverse discrete cosine transform coefficient and the windowing vector to be input to the calculating means, and controlling the calculation.

Hereinafter, a digital audio filtering method and an apparatus for performing the method will be described with reference to the accompanying drawings.

Before describing the present invention, a configuration and operation of a digital audio decoder including a digital audio filter for performing multiplication and a conventional audio filtering method will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram of a general MPEG1-audio decoder, which is composed of a demultiplexer and an error detector 200, an additional information decoder 202, an inverse quantizer 204, and an audio filter 206.

2 is a flowchart for explaining a decoding method of the MPEG-1 audio decoder shown in FIG.

The audio data consisting of a series of bits encoded by the encoder is input to IN shown in FIG. 1 (step 400), and an error is detected by the demultiplexer and the error detector 200, and the additional information decoder 202 is used. Receives the outputs of the demultiplexer and the error detection unit 200 and divides the header information, bit allocation for storing information of the data to be input in the future (step 402), scale factor selection information to be decoded, scale factor, and sample, respectively. The inverse quantization unit 204 is supplied (step 404), and the inverse quantization unit 204 receives the input and dequantizes the sample (step 406), and the audio filter unit 206 inputs the dequantized sample. The original audio PCM data is restored (step 408). After step 408, the audio filter unit 206 outputs a PCM sample (step 410).

3 is a flowchart for explaining an overlap add method among the filtering methods of the audio filter unit 206 shown in FIG.

FIG. 4 is a flowchart for explaining a method adopted by MPEG1 DIS among the filtering methods of the audio filter unit 206 shown in FIG.

First, the overlap add method will be described in detail with reference to FIG. 3.

The audio filter unit 206 receives 32 subband samples (S _k ) (step 600) to generate an inverse discrete cosine vector V _i (i is a vector factor) by performing IDCT as shown in Equation 1 below. Step 602).

In equation (1), i and k are factors.

64 V _i are created in Equation (1), and then periodically expanded to form a median vector U _i of 512 arguments (step 604). After step 604, the window vector W _i is calculated by multiplying Ui by the windowing vector F _i as shown in Equation (2) (step 606).

W _i = U _i F _i i = 0-511

The thus obtained W _i (i = 0-511), the window vectors are stored in the PCM output buffer (step 608). After the window vector W _{0 to} W ₃₁ is output to the PCM output buffer, after step 608, W _32- W ₅₁₁ is moved to the left of W _0- W ₄₇₉ , and W _480- W ₅₁₁ is initialized to '0'. This initialization is continued until all the window vectors are stored in the PCM output buffer (step 610).

After operation 610, the window vectors stored in the PCM output buffer are output to the PCM sample (operation 612).

The filtering method of the audio filter unit 206 adopted in MPEG1-DIS will be described in detail as follows with reference to FIG.

As in the overlap add method, the audio filter unit 206 inputs 32 dequantized 32 new subband samples (step 800). After operation 800, ₆₄ inverse discrete cosine vectors V ₀ to V ₉₅₉ are moved from V ₆₄ to V ₁₀₂₃ (step 802). The 32 subband samples S _k newly introduced into the inverse discrete cosine vectors V ₀ to V ₆₃ emptied after step 802 are matrixed by Equation (1) (step 804). After step 804, 64 elements are taken in the middle of the 1024 inverse discrete cosine vectors V _i to form an intermediate vector U _i having 512 elements (step 806). After operation 806, the window vector W _i is formed by multiplying the intermediate vector U _i by the windowing vector D _i (operation 808). After step 808, the vector is extracted by skipping at 31 window vectors W _i and then 32 output samples are obtained as shown in Equation (3) (step 810). After operation 810, 32 output samples T _j are output as PCM samples (operation 812).

T _31n = W (31) + W (63) +... + W (551).

Here, n denotes a block time, and block time means that digital audio data encoded in the MPEG1 standard is input as a frame composed of 12 blocks which are sequentially input. In this case, each block is input time.

5 is a flowchart illustrating a filtering method of the digital audio filtering apparatus according to the present invention.

6 is a flowchart according to the present invention for explaining in detail the flowchart shown in FIG.

FIG. 7 is a block diagram of the digital audio filtering apparatus according to the present invention for performing the flowchart shown in FIG. 6, which includes a dequantizer 1400, a storage 1402, a controller 1404, and a calculator 1406. Reference numeral 1408 corresponds to the audio filter unit 206 shown in FIG. 1, and reference numeral 1400 corresponds to the inverse quantization unit 204 shown in FIG.

FIG. 8 is a detailed block diagram of the digital audio filtering apparatus according to the present invention shown in FIG. 7 and includes an address generator 1600, a first storage unit 1602, and an audio filter constituting the inverse quantization unit 1400. The second storage unit 1604, the third storage unit 1606, the operation unit 1608, the first control unit 1610, and the second control unit 1612 constituting the unit 1408 are configured.

A digital audio filtering method and an apparatus for performing the same according to the present invention in a digital audio decoder will be described in detail with reference to FIGS. 5, 6, 7 and 8 as described above.

Inverse quantization unit 1400 shown in FIG. 7 dequantizes a sample of digital audio data, and stores it in the first storage unit 1602 shown in FIG. 8 through terminal IN1, and the address generator 1600 32 subband samples stored in the first storage unit 1602 are controlled by the control signals i and k (where i and k are factors of inverse discrete cosine transform coefficients) generated by the first control unit 1610. The multiplication, addition and accumulation are supplied to the operation unit 1608 that performs multiplication, addition, and accumulation through IN2, and the operation unit 1603 supplies the inverse discrete cosine transform coefficient N _ik stored in the second storage unit 1604 of the first control unit 1610. an input to an inverse discrete cosine transform vector (V _i) calculated in accordance with the control as follows (the step 1000).

Here, the second control unit 1612 shown in FIG. 8 plays a role of determining what is to be calculated and which one is to be calculated first. The first control unit 1610 determines what is to be input to the operation unit 1608. It also determines whether an operation is addition or subtraction.

The method for calculating one V _i is shown in the following equations (4), (5), (6), (7) and (8).

First, N _ik , the inverse discrete cosine coefficient used in IDCT, is given by Equation (4) below.

Rewrite equation (4) above to form equation (5).

In Equation (5), since the cos function has a period of 2π, only 128 cos coefficients are needed according to i and k. Of course, 64 and 32 cos coefficients may be necessary, but considering the complexity of the hardware, 128 coefficients are the best. In the present invention, this property was used to reduce the different _Nik values from 2048 to 128.

Listing Equation (1) gives the following equation (6).

Vi = N _i0 × S ₀ + N _i1 × S ₁ + N _i2 × S ₂ +... N _i30 × S ₃₀ + N _i31 × S ₃₁ . i = 0-63 (6)

Equation (5) is cos function symmetric about the Y axis in the second plane system.

N _i0 = | N _i31 |

N _i1 = | N _i30 |

N _i2 = | N _i32 |

:, i = 0-63

N _i14 = | N _i17 |

N _i15 = | N _i16 | and equation (6) becomes the following equation (7).

Vi = N _i0 × (S ₀ ± S ₃₁ ) + N _i1 × (S ₁ ± S ₃₀ ) + N _i2 × (S ₂ ± S ₂₉ ) +... + N _i14 × (S ₁₄ ± S17) + N _i31 × (S ₁₅ ± S ₁₆ ). Formula (7)

Where i = 0-63, and the value of the subband sample (S _k ) from above is (+) if i is even and (-) if i is odd.

Therefore, summarizing equation (7) can be written as equation (8).

As described above, it can be seen that the number of factors k is reduced from 32 to 16 (step 1200).

Next, look at the properties of the argument i as follows:

The inverse discrete cosine vector V _i of equation (8) has the following properties.

V ₀ , V ₁ ,. , V ₁₅ exists, but when i = 16, it becomes odd multiple of N _ik = cos (π / 2) in Eq. (5), and is '0' regardless of k value.

And if i = 17

The reason for the same terms ② and ③ in Equation (9) is that the cos function is symmetric about the Y axis of the secondary plane system.

Therefore, the following relationship holds for i = 17 to i = 32, as in equation (9).

V ₁₇ = -V ₁₅

V ₁₈ = -V ₁₄

:

V ₃₁ = -V ₁

V ₃₂ = -V ₀

That is, V _i = -V _(32-i) , (i = 17-32) -------- Equation (10)

Becomes

Likewise

V ₄₉ = -V ₄₂

V ₅₀ = -V ₄₆

:

V ₆₂ = -V ₃₄

V ₆₃ = -V ₃₃

That is, V _i = V _(96-i) , (i = 49-63) ------- Equation (11)

Becomes

As described above, it can be seen that the number of factors i has been reduced from 64 to 32 (step 1202), and an inverse discrete cosine vector is obtained using Equation (8) (step 1204).

As a result, the multiplication operation performed in the conventional digital audio filter unit 206 for obtaining the inverse discrete cosine vector V _i was 2048 (= 64 × 32) times because i = 0-63 and k = 31. As can be seen from equations (11) and (8), the multiplication operation of the digital audio filtering device according to the present invention required for IDCT is i = 0-15, i = 33-48 and k = 0. Since it is -15, the total number is 512 (= 32 x 16), which is reduced to 1/4 compared to the total number of multiplications by the filtering method proposed in the conventional MPEG DIS.

In the operation of the audio filtering device for reducing the multiplication operation in IDCT, the first control unit 1610 generates only the values of the factors i and k described above, and supplies them to the address generator 1600 shown in FIG. The control unit 1602 generates a subband sample, and the second storage unit 1604 controls the supply of necessary inverse discrete cosine transform coefficients to the operation unit 1608.

The order to store the PCM output buffer, obtain a window de vector (W _i) after step 1000 (the 1002 phase) under the control of the operation unit 1608 of the first controller 1610 illustrated in Fig. 8 a third storage unit Input the windowing vector D _i from 1606, multiply it by V _{(i% 64)} , and then add the window vector W _i ^n-1 of the previous (n) block stored in the accumulator of the calculator 1608 to add one. 512 window vectors (W _i ^n-1 ) that can be obtained from V _i are obtained through Equation (12) (step 1206).

In Equation (12), the operator% is the operator to take the remainder, and V _{(i% 64)} is the inverse discrete cosine vector of the factor corresponding to the remainder after dividing the argument i by 64.

The calculating unit 1608 sequentially stores 32 vectors W ₀ to W ₃₁ from the front of the 512 W _i obtained after the step 1206 in the PCM output buffer (step 1208).

In order to initialize the window vector after step 1208 (step 1004), the operation unit 1608 moves W ₃₂ to W ₅₁₁ from W ₀ to W ₄₇₉ , and then moves W ₄₈₀ to W ₅₁₁ from '0'. Initialize to step 1210.

After step 1210, the window vector stored in the PCM output buffer is output as a PCM sample (steps 1006 and 1212).

In addition, although a buffer for storing V _i is conventionally required, the audio filtering apparatus according to the present invention eliminates V _i using Equations (10) and (11). That is, it can be seen that after calculating the V ₀ is calculated for all of W _i, which is available using only V ₀ method in the equation (13).

If W _i is obtained as above, V _i buffer is not needed.

In conclusion, when comparing the audio filtering method and apparatus proposed by MPEG DIS with the audio filtering device according to the present invention, the minimum frequency required for IMDCT is 44.1KHz × (512 ÷ 32) at 44.1KHz. 0.7MHz, that is, the total required operating frequency is about 1.4MHz, and it operates at a lower operating frequency, which is about 1/4 less than the conventional operating frequency. Since the number of N _ik ) is reduced, the ROM table is reduced, and thus, many effects can be seen in terms of chip size.

Claims (6)

A decoding step of inputting coded digital audio data to decode additional information, an inverse quantization step of inversely quantizing the decoded samples and outputting a subband sample, and inputting the subband sample to inverse discrete cosine transform to window In the digital audio decoder including an audio filtering step of generating a vector and then outputting PCM samples, the audio filtering step is performed. First receiving the sub-band samples of a predetermined number (p) of claim 2 N _ik (inverse discrete cosine transform coefficients of a predetermined number (x): where i and k is the inverse discrete cosine transform coefficients for specifying the stored address, The value of the factor i is reduced to a third constant, the value of the factor k is reduced to a fourth constant, and the inverse discrete cosine transform coefficient has a different value of the fifth constant. An inverse discrete cosine vector step of obtaining the inverse discrete cosine vector V _i of the first predetermined integer with the subband sample S according to the following equation; After the inverse discrete cosine vector step, the sixth constant current from the inverse discrete cosine vector (V _i ), the windowing vector (D _i ), and the window vector (W _i ^n-1 ) of the previous (n-1) block. (n) a window vector step of obtaining the window vector (W _i ⁿ ) of the block according to the following equation: (i% p × 2 is the remainder after dividing i by a multiple of p, where% is the remainder operator.) A storage step of sequentially storing from the first vector column to the last vector column in the PCM output buffer when there is a fifth constant of the vector columns consisting of the window vectors successive to the first constant; An initializing step of initializing the position of the window vector remaining after moving all the vector sequences not yet stored to the position of the window vector emptied after the vector sequence is sequentially stored in the PCM output buffer; And an output step of outputting the vector strings stored in the PCM output buffer to the PCM samples after the initialization step. The method of claim 1, The inverse discrete cosine vector step and the window vector step The first predetermined number is 32, the second predetermined number is 2048, the third predetermined number is 32, the fourth predetermined number is 16, the fifth predetermined number is 128, and the sixth predetermined number is Is 512, and the number of multiplications required to obtain the inverse discrete cosine vector is a sixth constant. The method of claim 1, And wherein the window vector step represents the inverse discrete cosine vectors whose factor i is an even multiple of the first predetermined number as an inverse discrete cosine vector having the factor i of zero. Decoding means for inputting coded digital audio data to decode additional information, first storage means for inversely quantizing the decoded samples, and storing an subband sample and an address in which the dequantized subband sample is stored; An audio quantizer comprising: an inverse quantization means including an address generating means, and an audio filtering means for inputting the subband sample to inverse discrete cosine transform to form a window vector and then outputting a PCM sample Sudan An inverse discrete cosine transform coefficients _(ik N: where i and k is a factor Im for designating the address of the inverse discrete cosine transform coefficient is stored) second memory means for storing the windowed vector; The inverse discrete cosine transform coefficient and the subband sample are input to obtain the inverse discrete cosine transform vector, and the window wing vector, the inverse discrete cosine transform vector, and the window vector of the previous block are input to obtain a current window vector. Computing means for performing an operation to obtain; And a control means for controlling the address generator to generate a subband sample, controlling the inverse discrete cosine transform coefficient and the windowing vector to be input to the calculation means, and controlling the calculation. Filtering device. The method of claim 4, wherein The second storage means comprises: third storage means for storing the inverse discrete cosine transform coefficient; And a fourth storage means for storing the window wing vector. The method of claim 4, wherein The control means calculates the values of the factors i and k necessary to obtain the inverse discrete cosine transform vector, supplies them to the address generator, specifies addresses of the second and third storage means, and controls the calculation means. First control means; And a second control means for controlling said calculating means.