KR100790534B1 - Signal processing method and apparatus for fast convolution using an overlap save scheme based on qdft - Google Patents

Abstract

A signal processing apparatus and method employing a convolution overlap-save scheme are provided to perform a QDFT(Quarter Discrete Fourier Transform) coefficient operation only on a new sample by using a filtering result for a previous sample existing in a single data block. A filtering result of the first sample inputted to a data block is stored in a buffer(S300). When the second sample is inputted to the data block, a half of the second sample is filled in the data block(S302). A QDFT coefficient of the data block filled with a half of the second sample and that of a filter are calculated(S304,S306). The calculated QDFT coefficients are multiplied by a multiplier and then converted into time domains by using IQDFT(Inverse QDFT)(S308). A real number part of the converted result and the filtering result value of the first sample stored in the buffer are added by an adder to obtain an output(S310).

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a signal processing apparatus and a signal processing method using a convolution superposition-

1 is a block diagram illustrating a signal processing technique to which a conventional convolution overlap-save technique is applied;

2 is a block diagram of a signal processing apparatus to which a convolution overlap-hold technique according to a preferred embodiment of the present invention is applied;

3 is a flowchart of a signal processing process to which a convolution overlap-hold technique according to a preferred embodiment of the present invention is applied.

Description of the Related Art

200: data block 202: filter

204: multiplier 206: buffer

208: adder 210: output block

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal processing technique and, more particularly, to an overlap-save technique suitable for reducing a calculation amount in performing convolution using Discrete Fourier Transform (DFT) To a signal processing apparatus and method to which the present invention is applied.

The process of changing the frequency characteristics of a signal in the field of signal processing is called filtering, and this filtering can be expressed mathematically as convolution such as the following [Equation 1].

[Equation 1]

In Equation (1), y denotes a converted signal, f denotes a filter, N denotes a length of a filter, and x denotes an input signal. These convolutions can be reduced by using DFT. This is because the DFT can be implemented by multiplying the convolution in the time domain by the frequency domain.

The convolution is implemented in the frequency domain by the overlap-add technique and the overlap-hold technique. Here, the overlap pending technique will be described as follows.

In the overlap-hold technique, when samples are input, the samples are filled in a block of a specific length, and the output is obtained by convolving the block and the filter in the frequency domain. When new samples come in, half of the existing sample is left in the block and the other half is filled with the new samples, repeating the same process. That is, consecutive blocks are always overlapped by half of the samples.

FIG. 1 is a block diagram illustrating a signal processing technique to which a conventional convolutional superposition-hold technique is applied, including a data block 100, a filter 102, a multiplier 104, and an output block 106.

The convolution process in the frequency domain to which the conventional superposition-hold technique is applied will be described with reference to FIG.

As shown in FIG. 1, a first sample of an input signal is filled into a data block 100, and a DFT coefficient of the data block 100 and a DFT coefficient of the filter 102 are calculated.

Thereafter, the output 106 is obtained by multiplying the respective DFT coefficients calculated through the multiplier 104 and transforming the filtering result into a time domain using an inverse discrete Fourier transform (hereinafter referred to as IDFT) . This can be expressed by the following equation (2).

&Quot; (2) "

Here, D represents the DFT matrix, F represents the DFT coefficient of the filter 102, and X represents the DFT coefficient of the input data block 100, respectively.

At this time, when the second sample is input to the data block 100, half of the first sample is left in the data block 100, and half of the second sample is filled in the data block 100.

Then, the DFT coefficient of the data block 100 filled with 1/2 of the first sample and 1/2 of the second sample and the filter 102 DFT coefficient are calculated, and the calculated DFT coefficients are multiplied by the multiplier 104, And then converts it into a time domain using IDFT to obtain an output 106. [

As described above, in the signal processing technique using the conventional convolutional superposition-suspend technique, it is necessary that a duplicate (1/2 of the previous sample) DFT coefficient operation is required for the previous sample every time a new sample is input.

This results in not only an excessive amount of signal processing in the signal processing system but also an increase in the overall system load.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems of the conventional art, and it is an object of the present invention to provide a method and apparatus for performing quadrature discrete Fourier transform (hereinafter abbreviated as QDFT) on new samples using filtering results for previous samples existing in one data block. ) Counting operation, thereby reducing the amount of signal processing computation and the overall system load, and to provide a signal processing apparatus and method to which the convolution overlap-hold technique is applied.

According to an aspect of the present invention, there is provided a signal processing method using a convolution superposition-suspend technique, the method comprising: storing a filtering result of a first sample input into a data block in a buffer; Filling the data block with a half of the second sample if the second sample is input; dividing the quadrature discrete Fourier transform coefficient of the half-filled data block of the second sample by a second discrete Fourier transform Transforming the real part of the transformed result into a time domain by performing a reverse quadrature discrete Fourier transform after multiplying each of the calculated quadrature Fourier transform coefficients by a transform coefficient, And adding the resultant value to obtain an output, wherein the convolutional superposition-suspending technique is applied to the signal processing method .

According to another aspect of the present invention, there is provided a signal processing apparatus to which a convolution overlap-hold technique is applied, comprising: a data block of a sub-block concept which is filled only with 1/2 of an input sample; A multiplier for multiplying a Fourier transform coefficient by a quadrature discrete Fourier transform coefficient of the filter, a buffer for storing a result of multiplication in the multiplier by using a quadrature discrete Fourier transform and storing the result of the multiplication in a time domain, And an output block for processing the addition result of the adder as an output. The present invention also provides a signal processing apparatus to which a convolution superposition-retention technique is applied, comprising an adder for adding a real part of a result to a conversion result stored in the buffer, and an output block for processing an addition result of the adder as an output.

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

Prior to the description of the embodiments, a key point of the present invention is that, in the signal processing technique using the overlap-hold technique, filtering of existing samples existing in one block is performed in the previous step, It is only used in the previous block. That is, according to the present invention, the filtering process for the existing samples existing in one block and the newly inputted samples are respectively separated, and then the filtering is not performed for the existing samples at the current stage, (Only DFT coefficient calculation is performed on newly input samples), thereby reducing the total amount of calculations. From this technical idea, the object of the present invention can be easily achieved.

FIG. 2 is a block diagram of a signal processing apparatus to which a convolution superposition-hold technique according to a preferred embodiment of the present invention is applied, including a data block 200, a filter 202, a multiplier 204, a buffer 206, (208), and an output block (210).

As shown in FIG. 2, the data block 200 means a half of a block in the overlap-hold technique, that is, a sub-block composed of newly input samples. In the overlap-hold technique, the first half of the sample is distorted, so only the back half is taken as the output.

The multiplier 204 multiplies the QDFT coefficients of the data block 200 and the QDFT coefficients of the filter 202. The result of the multiplication in the multiplier 204 is converted into a time domain using IQDFT and the result of the filtering is stored in the buffer 206. [

The buffer 206 is a means by which the buffering result of the initial sample is stored. Thus, the filtering result stored in the buffer 206 can be added to and outputted from the real part of the conversion result of the new sample. In addition, an imaginary part of the conversion result of the new sample is stored in the buffer 206 and can be utilized for filtering the sample of the next step.

The adder 208 adds the real part of the IQDFT-transformed result through the multiplier 204 and the filtering result stored in the buffer 206, and provides the result to the output block 210.

The resulting value of this output block 210 may be expressed as: < EMI ID = 3.0 >

&Quot; (3) "

은 M by M IQDFT 매트릭스를,

는 필터의 QDFT 계수를,

는 기존 샘플에 대한 QDFT 계수를, 그리고

는 새로 입력되는 샘플에 대한 QDFT 계수를 나타낸다. 또한, Re{}는 실수(Real Number) 부분을, Im{}은 허수(Imaginary Number) 부분을 각각 나타낸다.Where y is the output sample,

RTI ID = 0.0 > M by M IQDFT < / RTI &

The QDFT coefficient of the filter,

Is the QDFT coefficient for the existing sample, and

Represents a QDFT coefficient for a newly input sample. Re {} denotes a real number part, and Im {} denotes an imaginary number part.

The QDFT can be defined as the following equation (4).

&Quot; (4) "

과 새로 입력된 샘플에 대한 식

으로 표현됨을 알 수 있다. 여기서 중첩-보류 기법에서는 뒤의 절반만 올바른 샘플이므로

와

만을 올바른 출력으로 간주한다.Referring to Equation (3), if the output is expressed by Equation

And the equation for the newly input sample

As shown in Fig. Here, in the nested-hold technique, only the back half is the correct sample

Wow

Only the correct output.

At this time, since the results of the existing samples have already been obtained in the previous step, in this embodiment, desired output results can be obtained by adding only the equations for the newly inputted samples.

Hereinafter, with reference to FIG. 2, a signal processing process using a convolution superposition-hold technique according to a preferred embodiment of the present invention will be described in detail with reference to the flowchart of FIG.

First, in step S300, the filtering result of the first sample input to the data block 200 is stored in the buffer 206. [ The filtering result is obtained by filling the data block 200 with the first sample of the input signal and then computing the QDFT coefficient of the data block 200 and the QDFT coefficient of the filter 202, By multiplying the coefficients and then transforming them into the time domain using IQDFT.

At this time, if a second sample is input to the data block 200, as in step S302, half of the second sample is filled in the data block 200. [

Thereafter, in steps S304 and S306, the QDFT coefficient of the data block 200 filled with 1/2 of the second sample and the QDFT coefficient of the filter 202 are calculated.

In step S308, each of the calculated QDFT coefficients is multiplied by a multiplier 204, and is converted into a time domain using IQDFT.

가 버퍼(206)에 이미 저장되어 있기 때문에, 새로 입력된 샘플에 대한 필터링 연산

만 수행하고, 단지 버퍼(206)의 필터링 결과

를 가산함으로써 출력 결과를 얻을 수 있다.The real part of the converted result and the filtered result of the first sample stored in the buffer 206 are added through an adder 208 to obtain an output 210 (S310). That is, in this embodiment, the filtering result for the previous sample

Is already stored in the buffer 206, the filtering operation

Only the filtering results of the buffer 206

The output result can be obtained.

On the other hand, the imaginary part of the transformed result of the second sample is stored in the buffer 206, in order to obtain the output result in the next step, for example, when the third sample is input.

In the signal processing technique using the convolution superposition-hold technique according to the present invention, the amount of computation can be drastically reduced. The comparison with the conventional signal processing technology is as shown in Table 1 below.

[Table 1]

Multiplications Additions Conventional signal processing method

The signal processing method according to this embodiment

As can be seen from [Table 1], when the number of multiplications and the number of additions required for the two algorithms are obtained and the amounts of calculation when the length of the block is N are compared, Is significantly reduced.

According to the present invention, the QDFT coefficient operation is performed only on a new sample using the filtering result on the previous sample existing in one data block, thereby reducing the signal processing calculation amount and the overall system load.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims (9)

A signal processing method to which a convolution overlap-hold technique is applied, Storing a filtering result of a first sample input into a data block in a buffer; Filling a data block with a half of the second sample if a second sample is input to the data block; Calculating a quadrant Fourier transform coefficient of the filter and a quadrature Fourier transform coefficient of a data block filled with 1/2 of the second sample; Multiplying the calculated quadrature Fourier transform coefficients by the calculated quadrature discrete Fourier transform coefficients, and then performing a reverse quadrature discrete Fourier transform into the time domain; Adding the real part of the transformed result to the filtering result stored in the buffer to obtain an output Wherein the convolution superposition-suspend technique is applied. The method according to claim 1, Wherein the step of storing the filtering result of the first sample in the buffer comprises: Filling the data block with a first sample of the input signal, computing a quadrature Fourier transform coefficient of the data block and a quadrature Fourier transform coefficient of the filter, And multiplying the calculated quadrature Fourier transform coefficients by the calculated quadrature discrete Fourier transform coefficients, and then transforming the quadrature discrete Fourier transform coefficients to a time domain using the inverse quadrature discrete Fourier transform and storing the filtered results in the buffer. . The method according to claim 1, The method comprises: Wherein an imaginary part of the transformed result of the second sample is stored in the buffer. The method according to claim 1, The result of adding the real part of the transformed result to the filtering result stored in the buffer,

Lt; / RTI > remind

As a result of the filtering of the first sample, remind

Is a result of quadrature discrete Fourier transform coefficient calculation of the second sample, remind

M by M inverse quadrature discrete Fourier transform matrix, remind

Quot; is the quadrature discrete Fourier transform coefficient of the filter

A filter,

Is a quadratic discrete Fourier transform coefficient), remind

Is a quadrature discrete Fourier transform coefficient for the first sample

Is the previous sample), remind

Is a quadrature discrete Fourier transform coefficient for the second sample

Is a newly inputted sample), The Re {} is a real part, Im {} is an imaginary part And a convolutional superposition-suspend technique is applied. delete A signal processing apparatus to which a convolution overlap-hold technique is applied, A data block of a sub-block concept which is filled only with 1/2 of the input sample, A multiplier for multiplying a quadrant Fourier transform coefficient of the data block by a quadrature Fourier transform coefficient of the filter, A buffer in which the multiplication result in the multiplier is converted into a time domain using the inverse quadrature discrete Fourier transform and stored, An adder for adding the real part of the result of the inverse quadrature discrete Fourier transform through the multiplier and the transform result stored in the buffer, An output block for processing the addition result of the adder as an output; Wherein the convolution superposition-suspend technique is applied. The method according to claim 6, Wherein the filtering result of the first sample is stored in the buffer and the imaginary part of the conversion result of the second sample is stored and used for filtering the sample of the next step. . The method according to claim 6, Wherein the adder adds the filtered result of the first sample stored in the buffer and the real part of the transformed result of the second sample. 9. The method according to claim 7 or 8, Wherein the first sample is a previous sample input to the data block, and the second sample is a sample newly input to the data block.

Images