CN101025919A - Synthetic sub-band filtering method for audio decoding and synthetic sub-band filter - Google Patents

Abstract

The invention relates to a synthesis subband filtering method and a synthesis subband filter during the audio decoding process, which includes: (a) It inputs a number of frequency-domain samples to form a new point group. (b) During the first stage of IDCT, it converts the new samples group into two intermediate variable arrays and puts them into memory for temporary preservation. (c) During the second stage of IDCT, it gets the intermediate variable arrays from the memory and eliminates redundant IDCT of symmetry by using of the intermediate variables arrays to obtain the IDCT data of time domain samples and puts them into the memory for temporary preservation. (d) It addresses the memory and takes out a number of time domain IDCT data of different frequency of a point sample. (e) It adds window for the IDCT sample data, then it outputs the PCM samples. (f) It repeats step (d) and (e) till it completes all the window-adding about a point sample group.

Description

Synthetic subband filtering method and synthon band filter in the audio decoder

Technical field

The present invention relates to audio decoder, relate in particular to synthetic subband filtering method and synthon band filter in the audio decoder.

Background technology

The mpeg audio standard is a kind of audio code decode algorithm of widespread use, and this algorithm has significantly reduced the transmission bandwidth and the memory capacity of voice data, has guaranteed the transmission quality of voice data simultaneously.Adopted method of coding subband in the mpeg audio coding.Method of coding subband uses one group of bandpass filter that the frequency band of input audio signal is divided into several frequency range, and each frequency range is a subband, adopts independent encoding scheme to remove coding to the sound signal in each subband.Like this, can carry out adaptive control respectively to each subband signal, the size after the quantification can be regulated according to the energy level of each subband, simultaneously, can represent each sample value to the different figure place of each allocation of subbands according to each subband signal in sensorial importance.Usually when mpeg audio is encoded, adopt to analyze sub-filter, and adopt the synthon band filter to carry out corresponding inverse operation during decoding.

The fundamental operation flow process of synthon band filter as shown in Figure 1 in the mpeg standard.This flow process comprises roughly 7 steps, and step S1 is 32 new subband sampling points for example for ready for; Step S2 is shifted, and farthest 64 values are cleaned out 1024 coefficient arrays, for the result of IDCT (inverse discrete cosine transform) vacates the position; Step S3 carries out the idct transform matrix operation of 32*64; Step S4 extracts 512 windowing coefficients, prepares to carry out windowing and rebuilds; Step S5 carries out the windowing conversion with the windowing module to the sampling point data; Step S6 mixes 32 PCM sampling points of stack; 32 PCM sampling points after step S7 output is rebuild.Wherein carry out IDCT calculating and windowing among step S3 and the S5 respectively and calculate, need respectively to use multiplicaton addition unit to finish, whole computation process needs 5 multiplicaton addition units and a totalizer, and the intermediate result in the computing simultaneously also needs a large amount of storage unit of temporary.When hardware was realized, the storage unit of these multiplicaton addition units and intermediate result need take a large amount of areas, particularly in system single chip, the reduction of entire chip cost had been proposed a difficult problem.

Summary of the invention

For overcoming the defective that exists in the prior art, the present invention proposes a kind of synthetic subband filtering method and synthon band filter that is used for audio decoder, computational resource and storage resources have been saved, reduced the area overhead of demoder to a great extent, improve counting yield simultaneously, and had good expandability.On the other hand, the present invention just can finish decoding on time under the relatively low situation of the frequency of operation of demoder, satisfies the requirement of at the uniform velocity playing.

According to an aspect of the present invention, provide a kind of synthetic subband filtering method in the audio decoder, may further comprise the steps:

(a) a new sampling point group of a plurality of frequency domain sampling points formations of input;

(b), the new sampling point group of importing is processed into two intermediate variable arrays and sends into preservation temporarily in the storer as the IDCT phase one;

(c) as the IDCT subordinate phase, from storer, take out the intermediate variable array that obtains in the IDCT phase one, and utilize the symmetry redundancy among this intermediate variable array elimination IDCT, obtain the IDCT data of sampling point time domain, and send in the storer and temporarily preserve;

(d) storer is carried out addressing, take out a plurality of time domain IDCT data of a sampling point under the different frequency value;

(e) data of the IDCT sampling point that takes out are carried out the windowing operation, export the PCM sampling point then;

(f) repeating step (d) and (e), all sampling points in a sampling point group are all finished the windowing operation.

According to described method, further comprise the steps:

(g) carry out and change output by the mathematics interpolation and count finishing described windowing operation back, the re-sampling operations of counting with change sampling rate and fine setting output.

According to described method, in step (a), import the new sampling point group that 32 frequency domain sampling points constitute.

According to described method, in step (b), the new sampling point group that will import by following formula operation is processed into two intermediate variable arrays:

S_add[k]＝s _t[k]+s _t[31-k]，k＝0...15。

S_sub[k]＝s _t[k]-s _t[31-k]

According to described method, in step (c), utilize described intermediate variable array, obtain the IDCT data of sampling point time domain by following formula operation:

$\left\{\begin{array}{c}V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\Σ}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{sub}\left[k\right],u=2i\\ V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\Σ}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{add}\left[k\right],u=2i+1.\end{array}\right.$

According to described method, in step (d), take out 32 the time domain IDCT data of a sampling point under the different frequency value, according to the order of the addressing shown in the following addressing formula storer is carried out addressing, once take out the time domain IDCT data of a sampling point under 32 different frequency values

i∈(0...15)；

j∈(0...31)

I is an odd number $\left\{\begin{array}{cc}j<16& \mathrm{index}=-(i*32+15-j),\\ j=16& \mathrm{index}=0,\\ j>16& \mathrm{index}=i*32+j-17;\end{array}\right.$

I is an even number $\left\{\begin{array}{cc}j<=16& \mathrm{index}=-(i*31+15+j),\\ j>16& \mathrm{index}=i*32+47-j\end{array}\right.$

Wherein, index represents the reference address of the storer of buffer memory in the middle of carrying out is represented directly export 0 value during index=0, preceding "-" of index value number represent to the value of index appropriate address taking-up by turn negate add 1.

According to described method, in step (e), the data of the IDCT sampling point of taking-up are carried out the windowing operation by following formula operation:

$V_i=\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{U}_j+32i*D_j+32,i=\mathrm{0,1},...31.$

According to described method, in step (c), described sampling point is handled according to the order of the addressing shown in the described addressing formula.

According to a further aspect in the invention, provide a kind of synthon band filter of realizing above-mentioned synthetic subband filtering method, comprising:

The intermediate computations unit, the new sampling point group that a plurality of frequency domain sampling points that will import in the IDCT phase one constitute is processed into the intermediate variable array;

Multiplicaton addition unit, reusable ground are carried out the operation comprise the multiply-add operation that carries out in multiply-add operation that the IDCT subordinate phase is carried out described intermediate variable array and the windowing operation at least;

Address-generation unit is carried out required addressing operation in the computing;

Data storage cell comprises the data that produced in the computing of described intermediate variable array in order to storage;

The coefficient storage unit is in order to deposit the calculating coefficient table that comprises IDCT coefficient table and windowing coefficient table;

Control module is in order to control the computing flow process in the whole synthetic subband filtering.

According to described synthon band filter, also comprise separation layer, the unified impact damper of accepting the memory read write request of each unit as storer.

According to described synthon band filter, comprise that also changing output by the mathematics interpolation counts, and exports the resampling unit of counting with change sampling rate and fine setting.

In one embodiment of the invention, described data storage cell is static RAM (SRAM).

In one embodiment of the invention, control module is unified state machine.

In other embodiments of the invention, computational resource that can be multiplexing also comprises shift unit etc.

In one embodiment of the invention, control module is controlled whole computing flow process, controls at least one multiplexing computational resource, finishes the process that comprises decoding and aftertreatment.

The algorithm of MPEG synthon band filter provided by the invention allows single-point to calculate, and needn't once calculate all for example sampling points of 32 subbands, thereby makes the computation process of streamline become possibility.Just can start calculating when it allows input to reach 17 sampling points simultaneously, carry out the IDCT operation again and needn't receive full all for example 32 sampling points by the time, thereby improved work efficiency.

MPEG I﹠amp; II Layer I﹠amp; In the II audio decoder process, synthetic sub-band filter and follow-up post-processing module, for example resampling, volume are regulated or the like, all need to use MAC (multiplicaton addition unit) resource, reality of the present invention has only been used a MAC unit, and be responsible for calculating scheduling by a unified state machine, reach the multiplexing of MAC unit in a plurality of different operatings.By the improvement of algorithm, make whole process can adopt streamline (Pipeline) mode to carry out, thereby saved computational resource, greatly improve expandability simultaneously and revised convenience.This structure has reduced the area overhead of demoder to a great extent, and required a plurality of multiply-add operations and additive operation can be reduced to by a MAC and finish in the calculation procedure, because the unification of structure and the independence of front and back state also are beneficial to debugging.Because the opening of state machine structure increases or reduces treatment step and also is easy to, it is also very convenient to revise computational accuracy simultaneously simultaneously.

Description of drawings

The following drawings is the aid illustration to exemplary embodiment of the present, to the elaboration of the embodiment of the invention, be to disclose feature of the present invention place, but do not limit the present invention in conjunction with the following drawings for further, identical reference number is represented respective element or step among the embodiment in the accompanying drawing, wherein:

Fig. 1 is the fundamental operation flow process of existing subband synthesis filter in the mpeg standard.

Fig. 2 A is the histogram of data in 32 subbands of 1 sampling point.

Fig. 2 B is the PCM sampled value figure that the data of 1 sampling point are passed through the time domain that obtains after IDCT calculates.

Fig. 3 is the process flow diagram according to synthetic subband filtering method in a kind of mpeg audio decoding of the present invention.

Fig. 4 is the structured flowchart of a synthon band filter of realization synthetic subband filtering method shown in Figure 3.

Fig. 5 is the state machine diagram of the intermediate computations unit of synthon band filter shown in Figure 4.

Fig. 6 is the kernel state machine synoptic diagram of synthon band filter shown in Figure 4.

Fig. 7 is the SRAM address assignment synoptic diagram of synthon band filter shown in Figure 4.

Embodiment

During audio decoder, synthon band filter (SubBand Synthesis Filter) is commonly used to a plurality of (for example 32) the subband frequency domain sampling point behind the demultiplexing is mapped to time domain, the reverse mapping of audio decoder (inverse mapping) part is finished in final stack output for example 32 PCM sampling points, is operand maximum in the whole M PEG decoding algorithm, the most complicated unit.

The matrix of coefficients size of IDCT for example is 32*64 in the mpeg decode algorithm, and the generation formula of IDCT coefficient is as follows:

$\mathrm{IM}\left[i\right]\left[k\right]=\cos \left[\frac{(i+16)*(2*k+1)*\mathrm{\&pi;}}{64}\right],$ I=0...63, k=0...31 (formula one)

Idct transform is the initial step of whole synthon band filter, the resource consumption aspect, a maximum block storage is promptly from 1024 buffer memory requirement behind the idct transform in the synthetic sub-band filter fundamental operation flow process of mpeg standard, according to 24 accuracy computation, need the SRAM of about 3K byte to deposit, this has also determined the granularity size of whole calculating.

Because cosine function has symmetry, and is as follows:

cosθ＝cos(2π-θ)

And the IDCT coefficient is a cosine function, therefore, has a large amount of symmetry in the algorithm of IDCT, is optimized on this basis and can cut down a large amount of redundances.

By IDCT coefficient expression formula as can be seen, each column element of IDCT matrix has constituted a cosine waveform that frequency is different, and its time shift on transverse axis is-16 points.If to put 16 is the center, IM _0-0... IM _15-0With IM _17-0... IM _32-0For the odd symmetry relation, be the center to put 48, IM _33-0... IM _47-0With IM _49-0... IM _63-0Be the even symmetry relation, and this symmetric relation is set up all for the element in any row.On this basis, can be formula as follows with the IDCT variation in coefficient matrix:

$\mathrm{IM}\left[i\right]\left[k\right]=\cos \left[\frac{(i+33)*(2*k+1)*\mathrm{\&pi;}}{64}\right],$ I=0...31, k=0...31 (formula two)

As optimized Algorithm, the symmetry redundancy that is caused by i in the IDCT coefficient is eliminated, and the IDCT matrix of coefficients is reduced to 32*32.

K in the IDCT coefficient is each row element of IDCT matrix of coefficients, and the redundancy that is caused by k presents and the identical symmetric relation of IDCT rectangular array element.Following formula relation is just arranged:

IM[i] [k]=(1) ⁱIM[i] [m], k=(31-m), m=15...0 (formula three)

When IDCT coefficient place behavior even number line, be the even symmetry relation between preceding 16 coefficients and back 16 coefficients, preceding 16 coefficients and back 16 coefficients are the odd symmetry relation when IDCT coefficient place behavior odd-numbered line.Therefore we are in the first step of carrying out idct transform, utilize two intermediate variable arrays shown in for example 32 frequency domain sampling point constructive formulas four of input:

S_add[k]=s _t[k]+s _t[31-k], k=0...15 (formula four)

S_sub[k]＝s _t[k]-s _t[31-k]

Second of the idct transform step can utilize two intermediate variable arrays of first step acquisition to finish like this, and its computing formula is a formula five as follows.

$\left\{\begin{array}{cc}V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{sub}{\left[k\right]}^{\&prime;}& u=2i\\ V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{add}{\left[k\right]}^{\&prime;}& u=2i+1\end{array}\right.$ (formula five)

Can eliminate the symmetry redundancy among the IDCT in this way fully, the IDCT matrix of coefficients further is reduced to 16*32.The cache size that needs behind the idct transform is 256 sampling points like this, greatly reduces the middle buffer memory requirement of synthetic sub-band filter, has reached the purpose that reduces chip area from an aspect.

In the synthon band filter,, following computing formula is arranged for the operation of the windowing after the idct transform:

W _i=U _i* D _i, i=O, 1 ... 511 (formula six)

The computing formula of synthon band filter stack output sample is as follows:

$\mathrm{Vi}=\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{W}_{j+32i,}i=\mathrm{0,1},...31$ (formula seven)

In conjunction with formula six and formula seven, can obtain following formula:

$V_i=\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{U}_j+32i*D_j+32,i=\mathrm{0,1},...31$ (formula eight)

V in the formula eight _iFor rebuilding the PCM sampling point of back output.

With reference to figure 2A and 2B, Fig. 2 A is a kind of histogram of data in 32 subbands of 1 sampling point.Fig. 2 B is that the data of 1 sampling point are passed through the PCM sampled value figure of the time domain that calculates.When carrying out windowing, we need calculate the data of each sampling point under 32 different frequency values together, obtain the PCM sampled value on the time domain of a sampling point.Transverse axis is a frequency values among Fig. 2 A, and transverse axis is a time value among Fig. 2 B.For formula eight, after Fig. 2 A need wait until that whole 32 subband sampling points are finished DCT calculating, could by formula eight carry out windowing operation respectively to each sampling point: and Fig. 2 B once lists the data under all 32 frequency values under the sampling point time domain, can by formula eight handle a sampling point.

If before the windowing operation, DCT sampling point data are selected for use suitable access order, can be after the time domain IDCT data that obtain under 32 different frequency values of each sampling point, directly carry out the windowing operation and also immediately export the PCM sampled value of this sampling point, and need not wait until that receiving all sampling points just begins to carry out the windowing operation after the time domain DCT data under 32 different frequency values, after finishing the windowing operation of all sampling points then, export the PCM sampled value of each sampling point again.

In one embodiment of the invention, adopting size is the storer of 512byte, obtains carrying out the time domain DCT data of the sampling point that windowing operates by the addressing mode of following formula nine:

i∈(0...15)；

j∈(0...31)

I is an odd number $\left\{\begin{array}{cc}j<16& \mathrm{index}=-(i*32+15-j),\\ j=16& \mathrm{index}=0,\\ j>16& \mathrm{index}=i*32+j-17;\end{array}\right.$

I is an even number $\left\{\begin{array}{cc}j<=16& \mathrm{index}=-(i*32+15+j),\\ j>16& \mathrm{index}=i*32+47-j\end{array}\right.$ (formula nine)

In the formula nine, index is the reference address to storer.In addition, represent directly to export 0 value during index=0, the value that number expression of "-" before the index value is taken out index appropriate address negate by turn adds 1.

In one embodiment of the invention, when calculating according to formula five, can correspondingly change common computation sequence, carry out the idct transform operation in proper order according to the addressing that formula nine requires to sampling point, the required sampling point of windowing operation can be exported more in time, and guarantee that last output result can pass through compliance test (Compliance Test) requirement.

Obtain the time domain IDCT related data of sampling point behind the idct transform, adopt the suitable access mode shown in formula nine then, can finish windowing conversion and stack output sample with a multiplicaton addition unit to storer.

According to formula four, five and eight, and in conjunction with the addressing method that formula nine provides, the algorithm of the synthon band filter that can be optimized makes synthetic sub-band filter to calculate by single-point.Begin to calculate the output of first PCM sampling point from idct transform, need 32*16+16 to take advantage of and add the cycle at least, under the requirement of at the uniform velocity playing, must take advantage of the longest 32*16+16 to add the cycle and treat as general data, the frequency of operation of demoder requires than higher.And adopt after the single-point account form, the interval of PCM sampling point output is very even accordingly.Under the relatively low situation of the frequency of operation of demoder, just can finish decoding on time, satisfy the requirement of at the uniform velocity playing.

Fig. 3 represents the operating process according to a kind of synthetic subband filtering method in the mpeg audio decoding of the present invention.Step S31 imports the new sampling point group that a plurality of (for example being 32) frequency domain sampling point constitutes; Step S32 as the IDCT phase one, is processed into two intermediate variable arrays with the new sampling point group of importing and sends into preservation temporarily in the storer; Step S33, as the IDCT subordinate phase, from storer, take out the intermediate variable array that obtains in the IDCT phase one, and utilize the symmetry redundancy among this intermediate variable array elimination IDCT, handle the IDCT data that the back obtains the sampling point time domain, and send in the storer and temporarily preserve; Step S34 carries out addressing to storer, takes out a sampling point a plurality of (for example being 32) time domain IDCT data under the different frequency value; Step S35 carries out the windowing operation to the data of the IDCT sampling point that takes out, exports the PCM sampling point then; Step S36 judges whether all (for example the being 32) sampling points in the sampling point group are all finished, and as not, operates and returns step S34, and all sampling points in a sampling point group are all finished the windowing operation.

Wherein, step S32 is the IDCT phase one, and the new sampling point group of utilizing step S31 to import is finished the computing of formula four, and the intermediate operations result is sent into preservation temporarily in the storer.According to formula four, only need the corresponding in twos calculating of the sampling point in 0-15 and the 16-31 scope, therefore, input just can begin the calculating of back when reaching 17 sampling points, just do not carry out the IDCT operation and for example need not wait until that 32 sampling points are all imported, thereby improved work efficiency.

Step S33 is the IDCT subordinate phase, takes out the intermediate operations result that the IDCT phase one obtains from storer, carries out the computing of formula five, obtains for example related data of 32 time domain IDCT sampling points, and sends in the storer and temporarily preserve.

Step S34 by formula nine addressing methods that provide takes out the related data of time domain IDCT sampling points under 32 different frequency values for example.

Step S35 carries out windowing operation, to the data of a time domain IDCT sampling point taking out, carries out the computing of formula eight, exports the PCM sampling point then.The each data of a time domain IDCT sampling point under 32 different frequency values for example of taking out, immediately this column data is entered the windowing calculation stages, and directly export a corresponding PCM sampling point, and needn't wait in 1 sampling point group to be removed the data of all sampling points under the different frequency value just to carry out windowing.Like this, the PCM sampling point can be exported quickly, and the work efficiency of whole module also can increase.

Step S36 judges that whether for example 32 sampling points in the sampling point group are all finished, if all do not finish as yet, then proceed abovementioned steps; If all finish, then finish the synthetic sub-band filter of this sampling point group.Can get back to step S31 according to the requirement of processor, notice is sent into next new sampling point group input.

In one embodiment of the invention, the synthon band filter can also further carry out post-processing operation such as first order resampling (Resample) and second level resampling (Resample) operation after the windowing operation is finished.First order re-sampling operations changes output by the mathematics interpolation counts the change sampling rate.Second level re-sampling operations changes output by the mathematics interpolation counts, and fine setting output is counted, and makes things convenient for audio/video synchronous.

When realizing the mpeg audio algorithm, whole synthon band filter can be divided into 5 parts: the vector U of displacement (Shifting), IDCT, 512 components sets up, windowing is rebuild, PCM sampling point stack output.When realizing formula four, five and eight computing flow process, to start with from the coupling of computing, the computing work of whole synthon band filter can be defined as following 3 parts by fractionation again: (1) idct transform: the matrix multiplication operation of being responsible for IDCT; (2) address produces (DAG): be responsible for that displacement (Shifting), IDCT, vector U set up, all addressing operation during the stack of PCM sampling point; (3) windowing operation (Window): the overlap-add operation when matrix multiplication when being responsible for windowing and PCM rebuild.

Be decoded as example with two channel audios in the following examples and illustrate that the present invention realizes the structure of a kind of synthon band filter of synthetic subband filtering method in the mpeg audio decoding.Fig. 4 is the structured flowchart of a kind of synthon band filter of realization method shown in Figure 3.Synthon band filter shown in Figure 4 comprises: intermediate computations unit 1, and the new sampling point group that a plurality of (for example being 32) the frequency domain sampling point that will import in the IDCT phase one constitutes is processed into the intermediate variable array; Multiplicaton addition unit 2 is carried out the multiplexing operation comprise the multiply-add operation that carries out in multiply-add operation that the IDCT subordinate phase is carried out described intermediate variable array and the windowing operation; Address-generation unit 3 is carried out required addressing operation in the computing; Data storage cell 4 comprises the data that produced in the computing of described intermediate variable array in order to storage; Coefficient storage unit 5 is in order to deposit the calculating coefficient table that comprises IDCT coefficient table and windowing coefficient table; Control module 6 is in order to control the computing flow process in the whole synthetic subband filtering.Wherein, the intermediate computations unit is responsible for the unified control module of a part of evaluation work of synthetic subband filtering algorithm and is controlled whole synthetic subband filtering operation flow process, and controls the multiplexing a plurality of different steps in the computing flow process of at least one computational resource.Described data storage cell for example is a static RAM (SRAM).

According to the method flow that Fig. 3 describes, the work of idct transform phase one is responsible in the intermediate computations unit, and with after for example 32 frequency domain sampling points are processed into required intermediate variable according to formula four of input, the SRAM that restores uses for the idct transform subordinate phase.The work of intermediate computations unit is controlled by state machine.When resetting, state machine is from IDLE state 50 beginning (see figure 5)s at every turn, enters each state successively according to the calculation requirement of formula four.Fig. 5 is the state machine diagram of intermediate computations unit.The COMM state is 51 times among Fig. 5, calculates (IQ) by inverse quantization and draws preceding 16 sampling points, directly writes to the 0-15 address of SRAM.CAL state 52 times whenever the IQ sampling point that comes out, is read corresponding sampling point according to formula four from SRAM.SUBWR state 53 times will write SRAM by the intermediate result one that subtraction draws.ADDWR state 54 times will write SRAM by the intermediate result two that addition draws.Can be got by algorithm characteristics, sampling point was exported in the middle of the COMM state can be kept 16 IDCT.Return CAL by ADDWR and can keep 16 data write cycles.

The MAC unit carries out corresponding multiply-add operation.Under the control of unified control module, multiplicaton addition unit can carry out multiplexing in the operation of each different step of computing flow process.According to the suitable access mode shown in formula nine to storer, the for example data of 32 sampling points in 1 IDCT sampling point group of windowing action need no longer include relevance, can take out corresponding data respectively to each sampling point and carry out the windowing operational computations, and control idct transform and windowing operation with unified state machine, carry out single-point and calculate.Because unified control module is controlled down, the multiplexing MAC unit of multiply-add operation in the multiply-add operation of IDCT subordinate phase and the windowing operation, thus saved very large tracts of land expense.The MAC unit can adopt the general adder and multiplier of industry, does not repeat them here.

In one embodiment of the invention, unified control module is controlled whole computing flow process, controls one or more multiplexing computational resources, finishes the process that comprises decoding and aftertreatment.The audio decoder that comprises surround sound during for example aftertreatment is calculated, utilize unified control module, taking advantage of in a plurality of different calculation step adds calculating can multiplexing same MAC unit, and the displacement in simultaneously a plurality of different calculation step is calculated also can multiplexing same shift unit.

The DAG unit carries out the addressing operation in the synthon band filter under the control of control module, for example displacement (shifting), idct transform, vector U sets up, involved various addressing operations during the stack of PCM sampling point.

Impact damper 7 is as the separation layer of storer, and the memory read write request of each module of system is accepted in unification.

Depositing in the coefficient storage unit and needing the coefficient table used in the algorithm.According to previously described algorithm, comprise IDCT coefficient table and windowing coefficient table.The coefficient storage unit can adopt circuit to solidify and generate, and perhaps utilizes other read-only mode storage.

In one embodiment of the invention, the synthon band filter further comprises also that in the aftertreatment part resampling unit 8, resampling unit can be provided with first order resampling (Resample) operational module and second level resampling (Resample) operational module.First order re-sampling operations module changes output by the mathematics interpolation counts the change sampling rate.Second level re-sampling operations module changes output by the mathematics interpolation counts, and fine setting output is counted, and makes things convenient for audio/video synchronous.In first order re-sampling operations module and the second level re-sampling operations module required finish take advantage of add calculating equally can be with the multiplexing MAC of idct transform and windowing operation unit.

Control module can be unified kernel state machine, and its principle of work can be referring to Fig. 6.Kernel state machine control idct transform phase one, idct transform subordinate phase, windowing operation, the first order resample, the second level resamples totally 5 modules enable (Enable) signal.In the integrality control, according to IDCT Step1 Enable (the idct transform phase one enables)-＞IDCT Step2 Reset (the IDCT subordinate phase resets)-＞IDCT Step2 Enable (the IDCT subordinate phase enables)-＞Windows Enable (the windowing operation enables)-＞Resample1 Reset (first order resamples and resets)-＞Resample1 Enable (first order resamples and enables)-＞computation sequence of Resample2 Enable (second level resamples and enables) forms the drive pattern of fixed priority.Wherein, IDCT Step1 state carries out the work of idct transform phase one; IDCT Step2 state carries out the work of idct transform subordinate phase; The Windows state carries out the windowing operation; The Resample1 state carries out first order re-sampling operations; The Resample2 state carries out second level re-sampling operations.

After the computing of finishing the IDCT subordinate phase, reset the MAC unit MAC unit opportunity (Cycle) of (Reset) that resets when the kernel state machine also is provided with simultaneously module and switches, and promptly is initialized as 0.IDCTStep2 Reset state is finished resetting of MAC unit in the state machine.In addition, Resample1 Reset state is finished resetting of first order re-sampling operations module.

When carrying out the algorithm realization, fetching data from storer is the operation that consumes the clock period most, and the storage to middle computational data simultaneously also will consume a large amount of areas.Therefore, can offer a static buffer, utilize the DAG unit to finish the corresponding addressing operation of each sampling point.Like this, finish the calculating of the related data of a sampling point whenever the IDCT subordinate phase, after obtaining the related data of IDCT sampling point of a time domain, the DAG unit just can be directly from SRAM corresponding memory location read the related data of this sampling point, send into the MAC unit and carry out windowing calculating, and directly export a PCM sampling point, and needn't wait for that the data of all sampling points under the different frequency value are just carried out windowing in 1 sampling point group of input, thus improved the efficient of whole audio decoder process.In the present embodiment, the SRAM size for example is 1088*24, and address assignment for example as shown in Figure 7.The input sample data that passage 0 obtains after IDCT phase one conversion are temporary in the address 1023-1055 of SRAM, and the input sample data that passage 1 obtains after IDCT phase one conversion are temporary in the address 1055-1087 of SRAM; The IDCT sampling point that passage 0 obtains after the conversion of IDCT subordinate phase is temporary in the address 0-511 of SRAM, and the IDCT sampling point that passage 1 obtains after the conversion of IDCT subordinate phase is temporary in the address 511-1023 of SRAM.

Present embodiment is just in order further more clearly to describe the present invention, but not limitation of the present invention.Be to be understood that the present invention is not limited to the elaboration that embodiment does, anyly all should be encompassed within the spirit and scope of claim of the present invention based on modification of the present invention and equivalent of the present invention.

Claims (10)

1. a kind of synthetic subband filtering method in the audio decoder may further comprise the steps:

(a) a new sampling point group of a plurality of frequency domain sampling points formations of input;

(b), the new sampling point group of importing is processed into two intermediate variable arrays and sends into preservation temporarily in the storer as the IDCT phase one;

(c) as the IDCT subordinate phase, from storer, take out the intermediate variable array that obtains in the IDCT phase one, and utilize this intermediate variable array to eliminate symmetry redundancy among the IDCT, handle the IDCT data that the back obtains the sampling point time domain, and send in the storer and temporarily preserve;

(d) storer is carried out addressing, take out a plurality of time domain IDCT data of a sampling point under the different frequency value;

(e) data of the IDCT sampling point that takes out are carried out the windowing operation, export the PCM sampling point then;

(f) repeating step (d) and (e), all sampling points in a sampling point group are all finished the windowing operation.

2. the method for claim 1 is characterized in that further comprising the steps:

(g) carry out and change output by the mathematics difference and count finishing described windowing operation back, the re-sampling operations of counting with change sampling rate and fine setting output.

3. method as claimed in claim 1 or 2 is characterized in that, the new sampling point group that 32 frequency domain sampling points of input constitute in step (a); In step (b), the new sampling point group that will import by following formula operation is processed into two intermediate variable arrays:

S_add[k]＝s _t[k]+s _t[31-k]，k＝0...15。

S_sub[k]＝s _t[k]-s _t[31-k]

4. method as claimed in claim 1 or 2, it is characterized in that the new sampling point group that 32 frequency domain sampling points of input constitute in step (a) is in step (c), utilize described intermediate variable array, obtain the time domain IDCT data of sampling point by following formula operation:

$\left\{\begin{array}{c}V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{sub}\left[k\right],u=2i\\ V\left[u\right]=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}\mathrm{IM}\left[u\right]\left[k\right]*S\_\mathrm{add}\left[k\right],u=2i+1.\end{array}\right.$

5. method as claimed in claim 1 or 2 is characterized in that, the new sampling point group that 32 frequency domain sampling points of input constitute in step (a); In step (d), according to the order of the addressing shown in the following addressing formula storer is carried out addressing, once take out the time domain IDCT data of a sampling point under 32 different frequency values,

i∈(0...15)；

j∈(0...31)

I is an odd number $\left\{\begin{array}{c}j<16\mathrm{index}=-(i*32+15-j),\\ j=16\mathrm{index}=0,\\ j>16\mathrm{index}=i*32+j-17;\end{array}\right.$

I is an even number $\left\{\begin{array}{c}j<=16\mathrm{index}=-(i*32+15+j),\\ j>16\mathrm{index}=i*32+47-j\end{array}\right.$

Wherein, index represents the reference address of the storer of buffer memory in the middle of carrying out is represented directly export 0 value during index=0, preceding "-" of index value number represent to the value of index appropriate address taking-up by turn negate add 1.

6. method as claimed in claim 1 or 2 is characterized in that, the new sampling point group that 32 frequency domain sampling points of input constitute in step (a); In step (d), take out 32 the time domain IDCT data of a sampling point under the different frequency value; In step (e), the data of the IDCT sampling point of taking-up are carried out the windowing operation by following formula operation:

$V_i=\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{U}_j+32i*D_j+32,i=\mathrm{0,1},...31.$

7. method as claimed in claim 5 is characterized in that, in step (c), according to the order of the addressing shown in the described addressing formula described sampling point is handled.

8. synthon band filter of realizing the described synthetic subband filtering method of claim 1 comprises:

The intermediate computations unit, the new sampling point group that a plurality of frequency domain sampling points that will import in the IDCT phase one constitute is processed into the intermediate variable array;

Multiplicaton addition unit is carried out the multiplexing operation comprise the multiply-add operation that carries out in multiply-add operation that the IDCT subordinate phase is carried out described intermediate variable array and the windowing operation;

Address-generation unit is carried out required addressing operation in the computing;

Data storage cell comprises the data that produced in the computing of described intermediate variable array in order to storage;

The coefficient storage unit is in order to deposit the calculating coefficient table that comprises IDCT coefficient table and windowing coefficient table;

Control module is in order to control the computing flow process in the whole synthetic subband filtering.

9. synthon band filter as claimed in claim 8 is characterized in that also comprising the separation layer as storer, the unified impact damper of accepting the memory read write request of each unit.

10. synthon band filter as claimed in claim 8 or 9 is characterized in that also comprising that changing output by the mathematics interpolation counts, the resampling unit of counting with change sampling rate and fine setting output.

Images