KR19980084170A - Combined filter for MPEG and Dolby AC3 audio decoders - Google Patents

Abstract

Disclosed is a combined filter for MPEG and Dolby AC3 audio decoders. The combined filter of the present invention includes a first mux, a second mux, a first ram, a second ram, a ROM table, a butterfly module, an address generator, a state machine, and a third mux. The first mux and the second mux multiplex the MPEG bit stream and the AC-3 bit stream, and perform a function of determining whether to write to RAM or accept another input depending on whether the FFT module is completed. The first and second RAMs are used to temporarily store transform coefficients received as inputs and to store the intermediate results of each calculation. The ROM table provides a table of coefficients to the butterfly module. The butterfly module is a basic butterfly module that performs the FFT. The address generator generates an address of the first and second RAM where input and output samples are stored and an appropriate address of the butterfly module for performing an FFT. The state machine is used to control each module and is connected to the entire module. The third mux outputs a time domain sample.

Therefore, according to the present invention, despite the use of a decoder in which the hardware units are integrated by using the combined filter, both the signals encoded by the two audio compression algorithms can be reproduced, thereby increasing the efficiency and the added value.

Description

Combined filter for MPEG and Dolby AC3 audio decoders

The present invention relates to an audio decoder of a semiconductor device, and more particularly to a combined filter for an MPEG and a Dolby AC3 audio decoder.

Digital audio is used in video and multimedia applications, and occupies a significant portion of the overall signal bandwidth, making its demand for efficient compression algorithms indispensable. Typical compression algorithms include MPEG and Dolby AC3. The MPEG compression algorithm was first standardized as an international standard for audio compression, and effective compression was realized by using human acoustic recognition characteristics that react differently according to frequency bands. AC3 has been adopted as the audio standard for North American HDTV systems and is now being applied to DVD (Digital Versatile Disk), DBS (Direct Broadcasting System), Set Top Box (STB), and Digital Cable. Like MPEG, human acoustic recognition was used for compression. Both MPEG and AC3 can be used for compression of voice and high quality audio signals by not restricting the type of input signals.

On the other hand, the audio decoder is divided roughly into a bit allocation portion and a reproduction filter portion which restores a signal in the time domain using the allocated bits. Actually, the bit allocation algorithms of MPEG and AC3 differ in many ways. Because of this, blocks of the same function cannot be found in the bit allocation part at the decoder. On the other hand, in the reproduction filter part, an inverse modified discrete cosine transform having a large amount of calculation may be integrated into one structure by the method of the present invention.

It is therefore an object of the present invention to provide a combined filter for MPEG and Dolby AC3 audio decoders that can reproduce both signals encoded with MPEG and Dolby AC3 audio compression algorithms.

1 is an operation flowchart of a combined filter according to the present invention.

2 is a block diagram of a combined filter for MPEG and Dolby AC3 audio decoders according to the present invention.

The combined filter of the present invention for achieving the above object includes a first mux, a second mux, a first ram, a second ram, a ROM table, a butterfly module, an address generator, a state machine, and a third With mux.

The first mux and the second mux multiplex the MPEG bit stream and the AC-3 bit stream, and perform a function of determining whether to write to RAM or accept another input depending on whether the FFT module is completed.

The first and second RAMs are used to temporarily store transform coefficients received as inputs and to store the intermediate results of each calculation.

The ROM table provides a table of coefficients to the butterfly module.

The butterfly module is a basic butterfly module that performs the FFT.

The address generator generates an address of the first and second RAM where input and output samples are stored and an appropriate address of the butterfly module for performing an FFT.

The state machine is used to control each module and is connected to the entire module.

The third mux outputs a time domain sample.

Therefore, according to the present invention, despite the use of a decoder in which the hardware units are integrated by using the combined filter, both the signals encoded by the two audio compression algorithms can be reproduced, thereby increasing the efficiency and the added value.

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

The present invention aims to design a single conversion block by appropriately modifying the different conversion equations adopted by MPEG and AC3. In other words, MPEG and AC3 adopt a subband structure, which is effective for dividing human acoustic recognition into frequency bands. The frequency characteristics of each subband are expressed by a simple equation, which is called an inverse modified discrete cosine transform. First, AC3 supports three conversion equations, which determine at the encoding stage which conversion equation to use, depending on the characteristics of the input signal. And in order to reduce the amount of computation, this transform is transformed to use the Fast Fourier Transform (FFT) structure. In contrast, MPEG uses a single conversion equation, with some differences compared to that of AC3. Therefore, although the FFT structure designed for AC3 can be dealt with separately from the conversion formula of MPEG in form, it is clarified in the present invention that the FFT structure designed for AC3 can be used through some modifications.

In order to explain the present invention, in Chapter I, the inverse modified cosine transform equation of AC3 is examined and a series of mathematical interpretations are given to bring it to the FFT structure. And in Section II, we can use the existing FFT structure for AC3 through the mathematical analysis of inverse modified cosine transform of MPEG. Chapter III describes the basic structure for the actual implementation.

I. Inverse Modified Cosine Transform of Dolby AC3

AC3 uses three transformations. If the input signal is relatively unitary in the time domain (indicates the times occupied by the 512 input signals, which depends on the sample frequency), use a so-called long transform if there is no sudden amplitude and frequency shift. In contrast, in the unit section in which the transition of the input signal occurs, inaccuracies in the transition occurring when using the long transform are compensated by using two short transforms. Equation 1a represents three conversion equations used by AC3.

Equation 1a

In this case, K is 512 for the long transform and 256 for the short transform. And 0 ≦ k ≦ K / 2-1, and α is as follows.

Next, the derivation process to the FFT structure that can significantly reduce the amount of calculation using Equation 1a will be described.

Ⅰ-1. Inverse Modified Cosine Transform of Long Transform

It can be seen from Equation 1a that the Long Transform is represented by the following Equation 2b. Here the constant -2 / K is ignored for convenience.

[Equation 1b]

The value obtained by multiplying the analysis window (Analysis Window Coefficients) by the input signal is called g _m (r).

Therefore, the following relation holds.

Equation 1c

At this time

Is given by

In the same way

Equation 1d

Use it to define the following new array:

[Equation 1e]

In the above formula, 'means replace r with r + K / 4. In simpler terms, this can be given by

Equation 1f

At this time, z _m (r) = {(g ₁ -g ' ₂ ) + j (-g ₂ -g' ₁ )} is established. Equation 1f is defined as a new array Is given in the form of a Discrete Fourier Transform (DFT) of z _m (r). Thus, restoring z _m (r) means that it is possible through a new array of inverse DFTs and a series of scaling processes. This suggests that FFT can be used to reduce the operation. The whole process is as follows.

Procedure 1. _{Form a} new array X _{K / 2-2K-1} (m) + jX _2K (m) from the given conversion values X _K (m) (k = 0,1, ..., 255).

Procedure 2. Multiply the shear element exp {2π (8k + 1) / 8K} by the new array.

Step 3. Take the result of Step 2 from the K / 4-point inverse FFT.

Step 4. Multiply the result of step 3 by the trailing element exp {2π (8r + 1) / 8K}.

It can be seen that the mathematical model included in the standard provided by Dolby is based on this. Next, we will look at how two short transforms can be expressed as Equation 1e.

I-2. Inverse Modified Cosine Transform of Short Transforms

(1) Short Transform of the first block

As given in Equation 1a, the short transform of the first block is represented as follows.

[1 g of equation]

Again, the process is similar to that given in Section I-1, which is summarized below.

[Equation 1h]

And

Equation 1i

Is displayed as: So the new array

Equation 1j

And is defined as f _m (r) = (g ' ₁ -g' ₂ ) + j (g ₁ -g ₂ ). Equation 1j also indicates that the new array can be represented in the form of a DFT.

Procedure 1. Define a new array X _{K / 2-2K-1} (m) + jX _2K (m) from the given conversion values X _K (m) (k = 0,2,4, ..., 126). . Where K is 256.

Procedure 2. Multiply the new array by the shear element exp {2π (8k + 1) / 8K}.

Step 3. Take the result of Step 2 from the K / 4-point inverse FFT.

Step 4. Multiply the result of step 3 by the trailing element exp {2π (8r + 1) / 8K}.

(2) Short Transform of the second block

It can be seen from Equation 1a that the short transform of the second block is given as follows.

] 1k Equation

therefore

Equation 1l

If you define a new array and clean it up,

[1m Equation]

Therefore, the decoding order is as follows.

Procedure 1. _{Define a} new array X _{K / 2-2K-1} (m) + jX _2K (m) from the given conversion values X _K (m) (k = 1,3,5, ..., 127). Where K is 256.

Procedure 2. Multiply the new array by the shear element exp {2π (8k + 1) / 8K}.

Step 3. Take the K / 4-point inverse FFT from the result of step 2.

Step 4. Multiply the result of step 3 by the trailing element exp {2π (8r + 1) / 8K}.

As shown above, the inverse modified cosine transform of AC3 can be implemented with 128-point FFT.

In the next chapter, we show that we can use the 128-point FFT structure implemented above by modifying the inverse modified cosine transform of MPEG.

Ⅱ transform of MPEG inverse modified cosine transform

This chapter discusses how to implement an inverse modified cosine transform using FFT.

II-1. Transformation of the formula

The inverse modified cosine transform used in MPEG is as follows.

Equation 2a

Let's define a new array as follows:

[Equation 2b]

Therefore, if the original conversion equation 2a is rearranged using the newly defined array equation 2b, it may be expressed as follows.

[Equation 2c]

Therefore, the following relation is established.

[Equation 2d]

The following relation can be derived from Equation 2c.

Theorem 1.

[Equation 2e]

Theorem 2.

[Equation 2f]

If you clean up the following array

[Equation 2g]

Theorem 3.

[Equation 2h]

Theorem 3 shows that the array v _m (r) has the form of a discrete cosine transform (DCT). DCT is given by

Equation 2i

In this case, the constant α is defined as follows.

[Equation 2j]

The expression given in Theorem 3 can be expressed as a vector.

[Equation 2k]

V = CX _K

In this case, the elements of matrix C are as follows.

[Equation 2l]

In the same manner, Equation 2i is expressed as D _DCT = ACD _N in a vector form, and matrix A is a square matrix having diagonal elements α _k . Comparing the square matrix A to both sides of Equation 2k and comparing it with the vector form of Equation 2i shows that the array V is equal to the DCT of X _K multiplied by the inverse matrix of A.

[Equation 2m]

V = A ^-1 dct (X _k )

Therefore, considering the result of Equation 2m, it can be seen that the original inverse modified cosine transform Equation 2a is related to DCT. The sequence of steps given below describes how to calculate Equation 2a using DCT.

Procedure 1. Take a given input 32-point DCT.

Step 2. Multiply the result of step 1 by the inverse matrix of A as given in Equation 2m.

Step 3. Find the array of v 'using the relation given in Theorem 2.

Step 4. Use theorem 1 and 2 to extend the 32-point result from step 3 to 64-point.

Step 5. Obtain the original array v using the relation given in Equation 2b.

In the next section, we briefly examine the correlation between DCT and FFT.

II-2. Relationship between DCT and DFT

This section shows that N-point DCT can be represented as a 2N-point DFT.

Assume that the input signal is x (n), and define y (n) as follows.

[Equation 2n]

The 2N-point DFT of y (n) is as follows.

[Equation 2o]

therefore,

[Equation 2p]

The N-point DCT of x (n) is defined as follows.

[Equation 2q]

Therefore, the relationship between DCT and DFT can be implied by the following equation.

[Equation 2r]

This indicates that the N-point DCT can be implemented by multiplying the 2N-point DFT or FFT by an appropriate constant.

The formal conclusions of Sections II-1 and II-2 suggest that the original inverse modified cosine equation (2a) can be implemented as a 64-point forward FFT. This means that the inverse modified cosine transform of MPEG can be included in this structure in the light of AC3 being implemented as a 128-point inverse FFT.

III. Structure of the combined filter

This chapter discusses how to implement a practical filter.

1 is an operation flowchart of a combined filter according to the present invention. Referring to FIG. 1, it is determined by AC3 or MPEG encoding from an input bit stream. First, in the case of AC3 encoded data, pre-twisting and then performing a 128 point FFT. The output is then post-twisted. Next, in case of MPEG encoded data, a new array is formed according to Equation 2n, and then a 128 point FFT is performed. The computed real output is then tweened, and the output thus calculated is rearranged.

2 is a block diagram of a combined filter for MPEG and Dolby AC3 audio decoders according to the present invention. Referring to FIG. 2, the combined filter of the present invention includes a first mux 210, a second mux 212, a first 128 * 24 ram 214, a second 128 * 24 ram 216, The ROM table 218, the butterfly module 220, the address generator 222, the state machine 224, and the 3rd mux 226 are provided.

The first mux 210 and the second mux 212 multiplex the MPEG bit stream and the AC-3 bit stream to determine whether to write to RAM or accept another input depending on whether the FFT module is completed. Do this. At this time, 256 conversion coefficients are input.

The first 128 * 24 RAM 214 and the second 128 * 24 216 RAM are used to temporarily store 256 conversion coefficients received as inputs and to store intermediate results of each calculation.

The ROM table 218 provides the table coefficients to the butterfly module.

The butterfly module 220 is a basic butterfly module that performs an FFT.

The address generator 222 generates an address of the RAM 214 and 216 where input and output samples are stored and an appropriate address of the butterfly module for performing an FFT.

The state machine 224 is used for control of each module and is connected to the entire module.

The third mux 226 outputs 512 time domain samples.

With the structure as described above, coding is completed in Verilog-HDL (Hardware Description Language) and has been verified.

The highest sample frequency for AC3 and MPEG is 48KHz, and the time required to perform the operation on one audio block is 5.33ms and 0.66ms, respectively. This means that if the system is operating at 27 MHz and each module is in a pipelined structure, 24,000 clocks and 9,000 clocks are allocated to each channel. Using the Verilog simulator, the above combined filter resulted in computations of approximately 19,500 and 6,800 clocks for 24-bit audio signals. Multiplication of the actual combined filter is implemented using Bit-Serial Multiplier.

As a result, the present invention is designed as one filter based on the 128-point real FFT structure, each of the reconstruction filters required for decoding the bit stream encoded with Dolby AC3 and MPEG, which are audio compression standards.

The present invention is not limited to the above embodiments, and it is apparent that many modifications are possible by those skilled in the art within the technical spirit of the present invention.

As described above, according to the present invention, despite the use of a decoder in which hardware units are integrated using a combined filter, both signals and signals encoded by two audio compression algorithms can be reproduced, thereby increasing efficiency and added value.

Claims (1)

A first mux for multiplexing the MPEG input bit stream and the AC3 input bit stream and determining whether to write a transform coefficient to RAM or accept another input according to a state of whether a fast Fourier transform is completed; A second mux for multiplexing the MPEG input bit stream and the AC3 input bit stream and determining whether to write a transform coefficient to RAM or accept another input according to a state of whether a fast Fourier transform is completed; A first RAM connected to the first mux to temporarily store transform coefficients and to store intermediate results of each calculation; A second RAM connected to the second mux to temporarily store transform coefficients and to store intermediate results of each calculation; A ROM table for providing the table coefficients to the butterfly module; A butterfly module for performing fast Fourier transform as the transform coefficients and the ROM table coefficients of the first and second RAMs; An address generator for generating an address of the first and second RAMs to store input / output samples and an appropriate address of the butterfly module for performing a fast Fourier transform; A state machine used for control of each module and connected to the entire module; And And a third mux connected to the first RAM and the second RAM and outputting time-domain samples in response to the signal of the address generator.