CN101540749A - Implementation method and device of pretreatment unit capable of being configured with length-variable DFT - Google Patents

Abstract

The invention provides an implementation method of a pretreatment unit capable of being configured with a length-variable DFT; firstly the relation of length M data output after the variable length N and sampling rate are converted is limited correspondingly, so that the designed digital low pass filter is applicable to all variable lengths; then the sampling multiples L and the sub-filter length P of the integral multiples of a multi-phase filter used in the pretreatment unit are determined, and the design of the multi-phase filter is altered to the design of a digital low pass filter with the designed length being L.P; for different variable lengths N, the values of L and P are the same, and the same frequency domain parameters are adopted to design the digital low pass filter applicable to all variable lengths. For the implementation method, the multi-phase filter is combined with a linear interpolator to convert the interpolation of the output result into a calculation operation of a corresponding sub-filter coefficient. Different DFT variable lengths N are fixed by adopting the structure of the pretreatment unit designed by the method of the invention, and the position of the output interpolation data is determined by the collocated parameters N and M.

Description

The implementation method of the pretreatment unit of configurable length-variable DFT and device

Technical field

The present invention relates to Digital Signal Processing and digital information transmission technical field, particularly a kind of implementation method that is used for the pretreatment unit of DFT (discrete Fourier transform (DFT)) system that transform length can dispose.

Background technology

OFDM/OFDMA and relevant SC-FDMA technology thereof are used widely in modern wireless communication systems.Why the OFDM technology receives much concern, wherein one very important reasons be that it can utilize contrary discrete Fourier transform (DFT)/discrete Fourier transform (DFT) (IDFT/DFT) to carry out the modulation and demodulation of multicarrier, simplified system configuration and design greatly.Therefore, as the key modules of these technology of realization, DFT/IDFT has important role in the wireless communication system based on the OFDM technology.

The DFT conversion has several expression-forms commonly used, has only the difference of an invariant between these several expression-forms, and the content of its expression is the same.In communication, use following formula to represent N point DFT conversion usually:

$X\left(k\right)=\sqrt{\frac{1}{N}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}x\left(n\right)e^{-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;k}\frac{n}{N}}---\left(1\right)$

Accordingly, N point IDFT is expressed as:

$x\left(n\right)=\sqrt{\frac{1}{N}}\underset{k=0}{\overset{N-1}{\mathrm{\&Sigma;}}}X\left(k\right){\mathrm{<figure-callout\; id="2"\; label="e\; j"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">e</figure-callout>}}^j---\left(2\right)$

As can be known complex conjugate is got in the input of DFT computing by (1) and (2) formula, the output result gets complex conjugate again, just can obtain the IDFT transformation results, thus all relevant argumentation all can be generalized to IDFT naturally to DFT.

Employed DFT in the wireless communication system was from only using transform length N=2 originally ⁿDFT so that carry out the FFT computing and develop into and brought into use transform length non-2 now ⁿDFT, use 3780 DFT as the China Digital TV ground transmission standard; Develop into the DFT that uses multiple configurable transform length from the DFT that uses single transform length, as LTE and mobile WiMAX.

In the present communication standard, for transform length N ≠ 2 ⁿDFT, the product that all needs N can resolve into several little prime factors is N=N ₁N ₂N _I(N ₁, N ₂... N _IBe little prime number), by prime factor decomposition method PFA, Winograd fft algorithm etc. calculate the DFT of small point earlier, and the processing by a series of complex steps obtains final N point DFT then.Yet this method is not suitable for the DFT of any transform length, and particularly those transform length can not resolve into the DFT of little prime factor product.In addition, if calculate the DFT of configurable transform length with this kind method, corresponding various transform length must have common prime factor, will be very complicated otherwise calculate.

In fact, can calculate the DFT of small point, if the transform length of the DFT that counts greatly is chosen as M=2 with the DFT that counts greatly ⁿThen can realize with FFT.The following describes the DFT of how ordering by M accurately obtain N point DFT result (M＞N), in the middle of analysis subsequently, the invariant of DFT/IDFT transformation for mula

To not consider.

According to digital signal processing theory, discrete Fourier transform (DFT) (DFT) is exactly the result of the one-period of the fourier series (DFS) of getting the discrete time periodic signal in fact.Owing to relate to problems such as sample rate conversion, use DFS very not convenient, the present invention uses the Fourier transform (DTFT) of discrete time periodic signal to carry out associated description.The spectral line of DTFT is the corresponding δ function of intensity and DFS progression.In description subsequently, the invariant that does not influence operation result character in each formula is not considered.

Fig. 1 is a continuous periodic signal, and its cycle is T _Sym, among the figure this periodic signal to be done sampling and obtained periodically discrete signal, there be N equally distributed sampled point in its each cycle, and the sampling period is T _S1, then following relation is set up.

N·T _s1＝T _sym (3)

x(n)＝x(nT _s1)，n＝0，1，…，N-1 (4)

N spectral response value shown in the DFT window is consistent in the result that this N point is DFT and the DTFT frequency spectrum shown in Figure 2, and DFT result can be by the convenient calculating (notes: the invariant of all not considering the front in the DFT formula afterwards) of following formula.

$X\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}x\left(n\right)e^{-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;}\frac{\mathrm{kn}}{N}},$ k＝0，1，…，N-1 (5)

The frequency interval of side frequency point

$\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="A20091004978300264.png,A20091004978300282.png"\; state="\{\{state\}\}">f</figure-callout>}}_1=\frac{1}{{\mathrm{<figure-callout\; id="1"\; label="NT\; s"\; filenames="A20091004978300264.png,A20091004978300282.png"\; state="\{\{state\}\}">NT</figure-callout>}}_s}=\frac{1}{T_{\mathrm{sym}}}---\left(6\right)$

According to sampling theorem, DTFT shown in Figure 2 can recover periodic signal continuous time shown in Figure 1 after by low pass filter shown in Figure 3.And the frequency spectrum of Fig. 2 by low pass filter shown in Figure 3 after remaining the intensity of spectral line can obtain according to the DFT transformation results of (5) formula.Therefore, according to the character of Fourier transform, can obtain undistorted original periodic signal continuous time according to following formula reconstruct.

Fig. 4 is periodic signal continuous time after the reconstruct shown in (7) formula, and this signal is the same with signal shown in Figure 1.Now the signal shown in (7) formula is carried out up-sampling with respect to Fig. 1, (M＞N), sampled result is illustrated among Fig. 4 each cycle uniform sampling M sampling point equally.

M·T _s2＝T _sym (8)

m＝0，1，…，M-1 (9)

Spectral line in the DFT window of M point x ' DFT (m) and DTFT frequency spectrum shown in Figure 5 is corresponding, and its corresponding frequency interval is as follows.

$\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">f</figure-callout>}}_2=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="MT\; s"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">MT</figure-callout>}}_s}=\frac{1}{T_{\mathrm{sym}}}={\mathrm{\&Delta;f}}_1---\left(10\right)$

Therefore, the position of Frequency point shown in Figure 5 and Frequency point shown in Figure 2 is accurately alignment.X ' (m) can also try to achieve with following formula with IDFT.

$x^{\&prime;}\left(m\right)=\underset{k=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{X}^{\&prime;}\left(k\right){\mathrm{<figure-callout\; id="2"\; label="e\; j"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">e</figure-callout>}}^j,m=\mathrm{0,1},...,M-1---\left(11\right)$

X ' is an x ' DFT transformation results (transform length is M) (m) (k), what (11) formula and (9) formula were represented is same group of data, make its every periodic feature that equates and utilize complex-exponential function, we can obtain following relation (note: all derivations have all been omitted the invariant of DFT/IDFT formula front, and this does not influence result's correctness) herein.

In sum, the DFT that can come any N littler of accurate Calculation to order in theory fully with the DFT that M is ordered than M.If M=2 ⁿ, then the M DFT of ordering can realize with FFT (fast fourier transform), just is to say that the DFT of any transform length can finish by FFT through design rightly.

According to above theory analysis, calculate N point DFT with M point DFT and need do unlimited cycle expansion and use ideal low-pass filter filtering N point input data.In the implementation procedure of reality, ideal low-pass filter is impossible realize, also can not do unlimited cycle expansion to the input data.But we can pass through some Digital Signal Processing means, make the N point DFT result who calculates with M point DFT infinitely approach the DFT transformation results that it calculates according to (1) or (5) formula.

Configurable DFT with any transform length form, its transform length N is variable, so its corresponding M also may have nothing in common with each other with the low pass filter of accordingly input data realization up-sampling rate being changed.At this situation, it is 200910045186.1 patent of invention that the inventor has proposed application number to China national Department of Intellectual Property.The system that this patent is described can calculate the DFT/IDFT of any transform length with a covering device, such as the DFT of the employed 35 kinds of transform length of LTE uplink SC-FDMA, thereby cleared away a big obstacle for the further application of DFT in the wireless communication technology in future.

See also Fig. 6, it is 200910045186.1 for number of patent application, is used for the top level structure block diagram of DFT transformation system of the configurable transform length of 3GPP LTE/4G radio communication.This system comprises pretreatment unit 101, configurable fft processing unit 102, post-processing unit 103 and corresponding control logic unit 104.The main effect of post-processing unit is to carry out frequency domain compensation, and data pick-up and pre-mapping are handled, and the present invention will not discuss

The operation principle of this DFT transformation system is as follows: the input of pretreatment unit 101 receives the input data and it is carried out exporting configurable fft processing unit 102 to after the conversion of up-sampling rate, these input data are complex datas, with the piece is unit, every blocks of data comprises the N point data, obtains a M=2 through behind the up-sampling of pretreatment unit 101 ^LDNThe output of point data, M=2 ^LDNIt is 2 integer power.Configurable fft processing unit 102 receives the M=2 of pretreatment unit output ^LDNPoint data, and it is carried out the FFT conversion that length is M, export the data after the conversion to post-processing unit 103 then.After 103 pairs of data of post-processing unit carry out obtaining final DFT transform data after frequency domain compensation, data pick-up and pre-mapping are handled.

Please consult Fig. 7 again, it is the structured flowchart of the pretreatment unit 101 of above-mentioned configurable length-variable DFT system, mainly comprises loop restructuring circuit 111 on data path, configurable multiphase filter circuit 112 and interpolation circuit 113.The main effect of pretreatment unit 101 is to be that the input signal of N carries out up-sampling to length, and the length that obtains 2 integer power is M=2 ^LDNDateout, also finish the function of corresponding low pass filter simultaneously.

The present invention mainly solves the method for designing problem of the pretreatment unit of configurable length-variable DFT, can be applied to 3GPP LTE uplink SC-FDMA and modulate used DFT conversion.DFT is a key algorithm of realizing the LTE uplink SC-FDMA, and its transform length N depends on that the uplink data for designated user sends the number of sub-channels of distributing, and is defined by following formula:

N＝12×2 ^a3 ^b5 ^c≤1320 (13)

Wherein N is the quantity of subcarrier, a, b and c under N≤1320 conditions all more than or equal to 0 (during the 20MHz bandwidth).For the user of appointment, the N scope can (a, b c=0) to 1296, always have 35 different selections from 12.

By the derivation of (13) formula and front as can be known, realize for convenient, the pretreatment unit that need design has unified and regular structure, makes the DFT of various transform length can use same set of device to finish low pass filter and up-sampling function, thereby reduces hardware implementation cost.

Summary of the invention

The implementation method of the pretreatment unit of configurable length-variable DFT, its main purpose are that design is a kind of to all general pretreatment unit structure of any transform length, finishing corresponding sample rate conversion function, and hardware implementation cost are reduced greatly.

The implementation method of the pretreatment unit of configurable length-variable DFT is used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, comprising:

(1) with N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

(2) the configuration multiphase filter comes the dateout behind the loop restructuring is carried out the integral multiple up-sampling, and for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed;

(3) utilize interpolation device that the integral multiple up-sampling data of multiphase filter output are carried out interpolation, obtain the dateout on the final outgoing position, to different transform length, the structure of the interpolation device that is maintained fixed.

Step (2) further comprises:

For the up-sampling multiple is that L and subfilter length are the design of the multiphase filter of P, is translated into a design with wave digital lowpass filter of even symmetry and linear phase shift characteristic, all uses same digital filter h for all transform length N _d(n), its frequency domain parameter:

Free transmission range [0,1/L-Δ f ₁].

Transition band scope [1/L-Δ f ₁, 1/L+ Δ f ₂]

Stopband range [1/L+ Δ f ₂, 1]

Filter length N _p=LP

Δf ₁≈Δf ₂， $0<\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">f</figure-callout>}}_2<\frac{2}{L}\&CenterDot;\frac{{\mathrm{\&delta;}}_{\min }}{1-{\mathrm{\&delta;}}_{\min }},$ $0<{\mathrm{\&delta;}}_{\min }<\frac{1}{2}$

Δ f ₁With Δ f ₂Be adjustable parameter.

Input data length N and corresponding dateout length M relation are made respective limits, and making has a minimum relative distance between M and the N:

N≤M(1-δ _min)， $0<{\mathrm{\&delta;}}_{\min }<\frac{1}{2}$

Making M is that the form that satisfies the minimum of following formula is 2 ⁿInteger:

M＝2 ^LDN。

Optimally, to wave digital lowpass filter h _d(n) window function w (n) is set reducing passband and stopband ripple,

h(n)＝h _d(n)·w(n)，n＝0，1，…，N _p-1

H (n) is the unit impulse response of final filter, h _d(n) response that is to use the digital filter design method to obtain, w (n) is the response of window function.

The configuration multiphase filter further comprises: the function that L subfilter realizes lowpass digital filter h (n) and L times of up-sampling is set, and the coefficient of i subfilter is

c _i(n)＝h(nL+i)，i＝0，1，…，L-1，n＝0，1，…，P-1

The structure of each subfilter is identical, all has P tap.

The configuration multiphase filter further comprises and adopts the variable filter construction of coefficient to realize, total P tap, only need use multiphase filter to calculate and interpolation dateout adjacent two L times of up-sampling data on the position for linear interpolation, the position of interpolation dateout is determined by following formula:

pos(m)＝pos(m-1)+N/M，m＝1，2，…，pos(0)＝α ₀

α ₀Be initial position value, N and M are configurable, and corresponding location parameter is as follows:

Position for input data x (n)

Δ=pos (m)-n is interpolation dateout y (m) and x (n) alternate position spike;

Calculate first L times of used subfilter sequence number of up-sampling data:

The relative position on L sampling rate of interpolation dateout and first L times of up-sampling data is poor:

ε＝Δ·L-i

Calculating second used subfilter sequence number of L times of up-sampling data can obtain according to the i reckoning.

The invention also discloses a kind of implementation method of pretreatment unit of configurable length-variable DFT, be used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, comprising:

(1) with N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

(2) under the situation of linear interpolation, the operation of configurable multiphase filter and linear interpolation device is combined, constitute the sample rate conversion step of an integral body: use a variable filter construction of coefficient to realize the function of multiphase filter h (n) and linear interpolation, for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed, utilize interpolation device that multiphase filter subfilter coefficient is carried out interpolation, multiply each other with corresponding input data then and sue for peace and obtain dateout on the final outgoing position, to different transform length, the structure of the interpolation device that is maintained fixed.

Under the situation of linear interpolation, configurable multiphase filter and linear interpolation device are combined, its final dateout obtains according to the following steps;

Step (1) obtains the tap coefficient of equivalence to corresponding subfilter coefficient interpolation:

$\stackrel{\&OverBar;}{c}\left(k\right)=\left\{\begin{array}{cc}{c}_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_{i+1}\left(k\right)-c_i\left(k\right)],& i<L-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0(k+1)-c_i\left(k\right)],& i=L-1,k<P-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0\left(0\right)-c_i\left(k\right)],& i=L-1,k=P-1\end{array}\right.$

Step (2) is calculated modifying factor:

${\mathrm{\&delta;}}_m\left(i\right)=\left\{\begin{array}{cc}0,& i<L-1\\ [x(n+1)-x(n-P+1)\&CenterDot;\mathrm{\&epsiv;}\&CenterDot;c_0\left(0\right),& i=L-1\end{array}\right.$

Step (3) calculates final dateout:

$y\left(m\right)=\underset{k=0}{\overset{P-1}{\mathrm{\&Sigma;}}}\stackrel{\&OverBar;}{c}\left(k\right)\&CenterDot;x(n-k)+{\mathrm{\&delta;}}_m\left(i\right)$

Above computational methods will be converted into the interpolation calculation to two corresponding subfilter coefficients of adjacent position to the interpolation of the dateout of configurable multiphase filter, and the output result is added the final process result that a modifying factor obtains pretreatment unit.

A kind of pretreatment unit that is used for configurable length-variable DFT is used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, it is characterized in that, comprising:

The loop restructuring circuit: be used for N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

The sample rate conversion circuit: it realizes the function of multiphase filter and interpolation device, multiphase filter is used for the dateout behind the loop restructuring is carried out the integral multiple up-sampling, for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed, interpolation device is used for the integral multiple up-sampling data of multiphase filter output are carried out interpolation, obtain the dateout on the final outgoing position, for different transform length, the interpolation device structure that is maintained fixed, the sample rate conversion circuit combines multiphase filtering and interpolation operation on implementation structure and finishes, and will be converted into the interpolation to corresponding subfilter coefficient to the interpolation equivalence of integral multiple up-sampling dateout.

The loop restructuring circuit further comprises the loop restructuring control logic circuit, reconstruct buffer memory, MUX, output FIFO;

Output FIFO is used for and the sample rate conversion circuit of back carries out rate-matched.

The sample rate conversion circuit further comprises tap renewal logical circuit, Coefficient Look-up Table, tap data delay line, and the operator circuit of finally finishing filtering and interpolation function;

Coefficient Look-up Table is deposited the tap coefficient of multiphase filter according to the subfilter order, and the tap coefficient of same subfilter leaves the adjacent position in;

The loop restructuring circuit is exported to tap data delay line with the result, the data of loop restructuring output in tap data delay line through after the phase delay factor, form one group of data [x (n+1), x (n), x (n-1) ..., x (n-P+1)];

The operator circuit comprises some multipliers and an add tree circuit, and contains a correction circuit, the dateout of tapped delay line [x (n), x (n-1) ..., x (n-P+1)] and corresponding tap coefficient [c ₀, c ₁..., c _P-1] multiply each other, the result who obtains and the result of correction circuit send into the addition of add tree circuit together, obtain the final output y (m) of pretreatment unit;

Tap is upgraded logical circuit and is received the control input, reads the coefficient of multiphase filter correspondence position subfilter from Coefficient Look-up Table, produces the tap coefficient [c of operator circuit through interpolation ₀, c ₁..., c _P-1].

Tap is upgraded logical circuit and is comprised a configurable digital controlled oscillator circuit and a coefficient interpolation circuit;

Configurable digital controlled oscillator receives the control input, produces location of interpolation information and outputs it to the coefficient interpolation circuit; Configurable digital controlled oscillator comprises that a N/M produces circuit, an adder, a MUX, and the delayer that delay is D; N/M produces circuit, all control of controlled input signal of MUX and delayer D; The function of N/M can be finished by simple shifting function; Parameter N and M be by control input configuration, this determined the structure of multiphase filter be fix but its function is configurable;

The coefficient interpolation circuit receives the location of interpolation of configurable digital controlled oscillator generation and the corresponding subfilter tap coefficient of multiphase filter that Coefficient Look-up Table provides, and produces and offers the used tap coefficient [c of operator circuit ₀, c ₁..., c _P-1].

It is a kind of to all general pretreatment unit structure of any transform length that the present invention can provide, and finishing corresponding sample rate conversion function, and hardware implementation cost reduced greatly.

Description of drawings

Fig. 1 is a continuous periodic signal, and its cycle is T _Sym, this periodic signal has been done sampling, there be N equally distributed sampled point in its each cycle, and the sampling period is T _S1

Fig. 2 is the frequency spectrum of the discrete time periodic signal Fourier transform (DTFT) after the sampling shown in Figure 1, and the N bar spectral line in the DFT window that provides among the figure is consistent with N point DFT transformation results;

Fig. 3 is the frequency domain response of an ideal low-pass filter;

Fig. 4 is periodic signal continuous time after the reconstruct, and this signal is carried out up-sampling with respect to Fig. 1, and each cycle uniform sampling M sampling point (M＞N);

Fig. 5 has M bar spectral line consistent with M point DFT transformation results for the frequency spectrum of continuous time signal being made the Fourier transform (DTFT) of the discrete time periodic signal that obtains behind the up-sampling shown in Figure 4 in the DFT window among the figure;

Fig. 6 is the top level structure block diagram of DFT transformation system that is used for the configurable transform length of 3GPP LTE/4G radio communication;

Fig. 7 is the structured flowchart commonly used of the pretreatment unit of configurable length-variable DFT system;

Fig. 8 is one and is applicable to the present invention, the amplitude-frequency response template figure of the low pass filter that available digital Design of Filter technology realizes;

Fig. 9 is corresponding with Fig. 8, but frequency adopts the represented amplitude-frequency response template figure of relative value;

Figure 10 is the schematic diagram from N point input data to the sample rate conversion process of M point dateout;

Figure 11 is the structured flowchart of a device for carrying out said of pretreatment unit described in the invention;

Figure 12 is the structured flowchart of loop restructuring circuit;

Figure 13 is the block diagram of the used Coefficient Look-up Table of pretreatment unit;

Figure 14 upgrades the structured flowchart of logical circuit for tap;

Figure 15 is the structured flowchart of configurable digital controlled oscillator.

Embodiment

The main effect of pretreatment unit is that N point sampling input data rate is converted to M point dateout.Usually, this can realize with interpolation decimation filter commonly used, promptly does M interpolation doubly earlier and then does N extraction doubly again.But for configurable length-variable DFT, the N value is the configurable multiple value that has, and we can not design such interpolation decimation filter to every kind of transform length N.

At this situation, the present invention adopts configurable multiphase filter and linear interpolation device to finish the function of N point input data to the sample rate conversion of M point dateout.Configurable multiphase filter is the nucleus module of pretreatment unit, and it finishes the function of integral multiple up-sampling and low pass filter.For the various DFT transform length that needs are supported, design the fixing multiphase filter of a structure, this multiphase filter has fixing length and fixing unit impulse response, and the up-sampling multiple of realization is also fixed.

1. Design of Filter

The design of configurable multiphase filter coefficient (being unit impulse response) is the same with the coefficient design process of general digital filter, and just its specific implementation structure is different.Fig. 8 has provided the amplitude-frequency response template schematic diagram of the low pass filter of an available digital Design of Filter technology realization.For good digital filter, its passband amplitude-frequency response is constant substantially, and the transition band amplitude-frequency response is level and smooth, and stopband is fully decayed.F among the figure _S1Be the sample frequency of input signal, f _S2Be that the N point data is upsampled to M=2 ^LDNThe sample frequency of dateout during point data, f _OsBe the integral multiple up-sampling frequency that multiphase filter uses, following relational expression is set up.

f _os＝L·f _s1 (14)

$f_{s2}=f_{s1}\&CenterDot;\frac{M}{N}---\left(15\right)$

$L>>\frac{M}{N},$ L is an integer up-sampling multiple.

The following describes and how to design the fixing multiphase filter of an implementation structure.According to the correlation theory of Digital Signal Processing, Design of Digital Filter, at be relative frequency rather than absolute frequency, below we all discuss the design of this multiphase filter on the basis of relative frequency.The data that frequency axis shown in Figure 8 is provided are divided by f _Os/ 2 just can obtain corresponding relative frequency, and Fig. 9 is the amplitude-frequency response template schematic diagram of representing with the corresponding employing relative frequency of Fig. 8.

Amplitude-versus-frequency curve shown in Figure 9 has provided two transition band parameter Δ f ₁With Δ f ₂Signal, their the expression also be the relative frequency value, in order to the restriction transition band scope.The design of transition band is extremely important, if make the FFT that follow-up processing is ordered by M accurately obtain the DFT result that transform length is N, must make neither to produce spectral aliasing in [0,1/L] frequency range and divide decay only, therefore Δ f in theory ₂Should satisfy following relation.

$0<\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">f</figure-callout>}}_2<\frac{2}{L}(\frac{M}{N}-1)---\left(16\right)$

As shown in Figure 9, the important frequencies point 1/L among the figure is the inverse of integral multiple up-sampling multiple L, does not have direct relation with the transform length N of DFT, and another important frequencies point is also only relevant with the relative value M/N of M and N.Therefore, to all transform length N of configurable length-variable DFT, can design a unified multiphase filter and finish low-pass filtering and the operation of integral multiple up-sampling.

For Design of Digital Filter, transition band is precipitous more, and then the length of filter needs also is difficult to more greatly and more realize.In order to widen transition band, we are pushed into 1/L-Δ f with the starting point of transition band forward by 1/L ₁To situation shown in Figure 9, the width of transition band is relevant with M/N, thus we need transition band enough wide make filter be easy to the design and length less; Simultaneously wish that again the ratio of M/N is enough little, otherwise the too high processing that is unfavorable for back level module of the sample rate of pretreatment unit dateout.To this, we can do following restriction to the relation of M and N, and making has a minimum relative distance between M and the N:

N≤M(1-δ _min)， $0<{\mathrm{\&delta;}}_{\min }<\frac{1}{2}---\left(17\right)$

The M that selects should be 2 integer power, so that use FFT, can be met the M of the minimum of (17) formula thus.

M＝2 ^LDN (19)

(18) in the formula

The smallest positive integral more than or equal to x is got in expression.

According to (17) formula, Δ f ₂Span further limited, become following formula:

$0<\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">f</figure-callout>}}_2<\frac{2}{L}\&CenterDot;\frac{{\mathrm{\&delta;}}_{\min }}{{1-\mathrm{\&delta;}}_{\min }}---\left(20\right)$

δ _MinValue according to performance with realize that complexity selects.Δ f ₁+ Δ f ₂Be the transition band width, Δ f ₁With Δ f ₂Size can select Δ f as required flexibly ₁With Δ f ₂Can select to such an extent that to be more or less the same be Δ f ₁≈ Δ f ₂

For the realization of multiphase filter, the length of the unit impulse response of the digital filter of design need satisfy following relation

N _p＝L·P (21)

Also corresponding to the number of the subfilter of multiphase filter, P is the length of subfilter to L corresponding to the multiple of up-sampling in the formula.The structure of each subfilter is identical, only coefficient difference several delays.

Like this, according to top analysis, selected L, Δ f ₁And Δ f ₂It is as follows that we just can obtain the relevant parameter of required filter:

The ■ free transmission range [0,1/L-Δ f ₁].

■ transition band scope [1/L-Δ f ₁, 1/L+ Δ f ₂]

■ stopband range [1/L+ Δ f ₂, 1]

■ filter length N _p=LP

After obtaining these parameters, utilize digital computer can design well behaved Finite Impulse Response filter easily with even symmetry and linear phase shift characteristic.According to the digital filter that usual method designs, the amplitude-frequency characteristic of its passband and stopband generally has bigger ripple.In order to reduce the ripple of passband and stopband, and the decay that strengthens stopband, can be to the above FIR filter windowed function that designs.Concrete operations are as follows.

h(n)＝h _d(n)·w(n)，n＝0，1，…，N _p-1 (22)

H in the formula (n) is final filter unit impulse response, h _d(n) response that is to use filter design method to obtain, w (n) is the response of window function.After the windowing, the transition band of the filter that original design obtains is broadened, therefore design h _d(n) parameter Δ f the time ₂The scope of selecting wants appropriateness to dwindle.Select Hamming window or Kaiser window commonly used just can obtain reasonable result, wherein the following institute of Hamming window formula.

$w\left(n\right)=0.54-0.46\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="A20091004978300281.png,A20091004978300283.png"\; state="\{\{state\}\}">cos</figure-callout>}\frac{2\mathrm{\&pi;n}}{M-1},$ n＝0，1，…，N _p-1 (23)

2. sample rate conversion algorithm

According to the analysis of front, need expand low-pass filtering then as the cycle to the input data by the DFT that the FFT calculating N that M is ordered is ordered.In the practice, only need with N import data [x (0), x (1) ..., x (N-1)] loop restructuring be [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, and needn't do unlimited cycle expansion.For different transform length N, the P value is maintained fixed, and P also is the subfilter length of the described multiphase filter in back.

Loop restructuring is a simple relatively process, below the main sample rate conversion process of discussing.For the input data of mentioning in the sample rate conversion process, all be meant through the data after the loop restructuring.

After obtaining filter coefficient h (n), the data after just can adopting the multiphase filter method to loop restructuring are carried out integral multiple up-sampling and low-pass filtering.The method for designing of multiphase filter resolves into L subfilter with original filter h (n) in fact exactly, and each subfilter can be represented with following formula.

c _i(n)＝h(nL+i)，i＝0，1，…，L-1 (24)

The structure of each subfilter all is the same, has P tap, tap sequence number n=0, and 1 ..., P-1.Adopt the multiphase filter structure to carry out the integral multiple up-sampling, producing each L times of up-sampling data only needs P multiplication rather than LP original multiplication, and this dateout is produced through corresponding subfilter by the input data.For L times of up-sampling, be equivalent to corresponding each input data and all will produce L dateout, its relative position can be used the i=0 of following formula, and 1 ..., L-1 describes.Dateout at position i is passed through subfilter c by the input data _i(n) produce.

Figure 10 has provided the process that is obtained M point dateout by N point input data.Final purpose is will carry out sample rate conversion by the N point data to obtain M point dateout, and this dateout is to carry out linear interpolation by the data that obtain behind L times of up-sampling of input data to obtain.In realization, there is no need the data that obtain behind all L times up-sampling are all calculated, only need to calculate before the final dateout of next-door neighbour and L times of up-sampling data afterwards, operand will significantly reduce like this.The input data x (n) that derive below, behind the L times of up-sampling data x ' (nL+i), the relation between the dateout y (m).Each signals sampling cycle is satisfied following relation.

$T_{s1}=L\&CenterDot;T_{\mathrm{os}}\&DoubleRightArrow;T_{\mathrm{os}}=T_{s1}/L---\left(25\right)$

$N\&CenterDot;T_{s1}=M\&CenterDot;T_{s2}\&DoubleRightArrow;T_{s2}=T_{s1}\&CenterDot;N/M---\left(26\right)$

At first, need to determine concerning of dateout y (m) with respect to input data x (n) and L times of up-sampling data x ' position (nL+i).If the location interval of the adjacent data of input data is normalized to 1, then following relation is satisfied in the position of adjacent final dateout:

pos(0)＝α ₀

pos(m)＝pos(m-1)+N/M，m＝1，2，… (27)

α ₀Be to look to realize convenient and selected initial value, order

Δ＝pos(m)-n (29)

ε＝Δ·L-i (31)

Two then adjacent with dateout y (m) position L times of up-sampling data can be obtained by following formula:

$x^{\&prime;}(\mathrm{nL}+i)=\underset{k=0}{\overset{P-1}{\mathrm{\&Sigma;}}}{c}_i\left(k\right)\&CenterDot;x(n-k)---\left(32\right)$

$x^{\&prime;}(\mathrm{nL}+i+1)=\left\{\begin{array}{cc}\underset{k=0}{\overset{P}{\mathrm{\&Sigma;}}}{c}_{i+1}\left(k\right)\&CenterDot;x(n-k),& i<L-1\\ \underset{k=0}{\overset{P-1}{\mathrm{\&Sigma;}}}{c}_0\left(k\right)\&CenterDot;x(n+1-k),& i=L-1\end{array}\right.---\left(33\right)$

Dateout is obtained by the linear interpolation of these two adjacent datas, i.e. following formula.

y(m)＝(1-ε)·x′(nL+i)+ε·x′(nL+i+1) (34)

According to above analysis, realization is exactly to obtain two L times of up-sampling data adjacent with required dateout position with L times of up-sampling of a configurable multiphase filter realization integer earlier by N point input data to the most direct method of the sample rate conversion of M point dateout, carries out linear interpolation with these two data then and obtains final dateout.According to (32), (33) formula realizes that the function of this configurable multiphase filter only needs P tap just much of that, but its tap coefficient need upgrade according to the position dynamic of dateout.Can be known that by (27) to (33) formula the mode that the multiphase filter tap coefficient upgrades finally depends on N/M, just the transform length N of the DFT that is disposed has determined the pattern that the multiphase filter tap coefficient upgrades.The described sample rate conversion of pretreatment unit among Fig. 7 just realizes with the method.

Yet, by anatomizing (32), (33) and (34) formula, we find that the method for above realization sample rate conversion is not optimum, can also improve.(32) c and in (33) formula _i(k) be real number, and x (n) is a plural number, the calculating of (32) and (33) formula needs 4P multiplication altogether, and (34) formula also needs 4 multiplication in addition, and then above sample rate conversion algorithm needs 4P+4 multiplication altogether.Change a kind of thinking and consider, if we are earlier to the subfilter coefficient c in (32) and (33) formula _i(k) carry out and the corresponding computing of linear interpolation, and then multiply each other with corresponding input data correspondence, the function of Shi Xianing is identical with aforementioned algorithm like this, but required multiplication number can reduce.Concrete form such as following various description.

$\stackrel{\&OverBar;}{c}\left(k\right)=\left\{\begin{array}{cc}{c}_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_{i+1}\left(k\right)-c_i\left(k\right)],& i<L-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0(k+1)-c_i\left(k\right)],& i=L-1,k<P-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0\left(0\right)-c_i\left(k\right)],& i=L-1,k=P-1\end{array}\right.---\left(35\right)$

$y\left(m\right)=\underset{k=0}{\overset{P-1}{\mathrm{\&Sigma;}}}\stackrel{\&OverBar;}{c}\left(k\right)\&CenterDot;x(n-k)+{\mathrm{\&delta;}}_m\left(i\right)---\left(36\right)$

(36) δ in the formula _m(i) be a modifying factor, the special circumstances when being used for handling i=L-1.

${\mathrm{\&delta;}}_m\left(i\right)=\left\{\begin{array}{cc}0,& i<L-1\\ [x(n+1)-x(n-P+1)\&CenterDot;\mathrm{\&epsiv;}\&CenterDot;c_0\left(0\right),& i=L-1\end{array}\right.---\left(37\right)$

For the low pass filter h (n) of the even symmetry linear phase shift that obtains according to previously described method design, following formula is set up.

c ₀(0)＝c _L-1(P-1)，c ₀(0)＜＜1 (38)

Therefore, can ignore modifying factor δ generally speaking _m(i) promptly the influence of (37) formula.(35) formula of employing to (38) described method realizes the sample rate conversion of N point input data to M point dateout, and required multiplication number is that 3P+3 or 3P are individual.

Said method is that the operation with configurable multiphase filter and linear interpolation device combines, constitute the sample rate conversion device of an integral body, to make interpolation to the interpolation of the integral multiple up-sampling data of configurable multiphase filter output into, thereby save calculation resources multiphase filter subfilter coefficient.That is to say that the present invention also provides the implementation method of the pretreatment unit of another kind of configurable length-variable DFT, be used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, comprising:

(1) with N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

(2) under the situation of linear interpolation, the operation of configurable multiphase filter and linear interpolation device is combined, constitute the sample rate conversion step of an integral body: use a variable filter construction of coefficient to realize the function of multiphase filter h (n) and linear interpolation, for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed, utilize interpolation device that multiphase filter subfilter coefficient is carried out interpolation, multiply each other with corresponding input data then and sue for peace and obtain dateout on the final outgoing position, to different transform length, the structure of the interpolation device that is maintained fixed.

3. specific implementation device

Figure 11 is that it is realized in digital circuit according to the implement device of the pretreatment unit of a configurable length-variable DFT of method design described above.The function that this device is realized is a common N point data of input to be carried out up-sampling obtain M point dateout.The implement device of pretreatment unit shown in Figure 11 comprises loop restructuring circuit 201, and logical circuit 203, Coefficient Look-up Table 206, tapped delay line 202, and the operator circuit 208 of finally finishing filtering and interpolation function are upgraded in tap.Loop restructuring circuit 201 receives N point input data, and it is carried out loop restructuring, then the result is exported to tap data delay line 202.Loop restructuring output data in tap data delay line 202 through after some delay factors, form one group of data [x (n+1), x (n), x (n-1) ..., x (n-P+1)].Operator circuit 208 comprises some multipliers 204 and add tree circuit 207, and contains a correction circuit 205 of finishing the described function of (37) formula.Operator circuit 208 with the dateout of tapped delay line 202 [x (n), x (n-1) ..., x (n-P+1)] and corresponding tap coefficient [c ₀, c ₁..., c _P-1] multiply each other, the result of result who obtains and correction circuit 205 sends into 207 additions of add tree circuit together, obtains the output y (m) of pretreatment unit.Tap is upgraded logical circuit 203 and is received the control input, reads the coefficient of multiphase filter correspondence position subfilter from Coefficient Look-up Table 206, produces the required tap coefficient [c of operator circuit 208 through interpolation ₀, c ₁..., c _P-1].Coefficient Look-up Table 206 is deposited the tap coefficient of multiphase filter according to the subfilter order.Entire circuit structurally is actually function with configurable multiphase filter and linear interpolation device and combines and realize.

Figure 12 is the structural representation of loop restructuring circuit 201, and its chief component comprises corresponding loop restructuring control logic circuit 301, reconstruct buffer memory 302, MUX 303, and output FIFO304.The effect of loop restructuring circuit 201 be with the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output.The effect of output FIFO304 is to carry out rate-matched with the processing procedure of back one-level.

Figure 13 is the structural representation of Coefficient Look-up Table 206, and it is generally realized with ROM.What Coefficient Look-up Table 206 was deposited is the tap coefficient of multiphase filter, and the coefficient of identical subfilter is deposited in that same address is adjacent among the figure, and that deposit as address 0 is the tap coefficient [c of subfilter 0 ₀(0), c ₀(1) ..., c ₀And deposit address 1 is the tap coefficient [c of subfilter 1 (P-1)], ₁(0), c ₁(1) ..., c ₁(P-1)], or the like.The data that deposit at last L place, address do not have corresponding subfilter, and being mainly used to is the special circumstances during i=L-1 in reply (35) formula.

Figure 14 is the structure chart that logical circuit 203 is upgraded in tap, and it comprises a configurable digital controlled oscillator circuit 401 and a coefficient interpolation circuit 402.Configurable digital controlled oscillator 401 receives the control input, produces the location of interpolation shown in (27) formula, outputs it to the coefficient interpolation circuit.Coefficient interpolation circuit 402 receives location of interpolation and the tap coefficient of the corresponding subfilter of multiphase filter that obtains from Coefficient Look-up Table 206, produces according to (28) to the algorithm of (31) formula and (35) formula and offers the used tap coefficient [c of operator circuit among Figure 11 208 ₀, c ₁..., c _P-1].

Figure 15 is the structure chart of configurable digital controlled oscillator 401, and it produces the positional information by the represented interpolated data of (27) formula.Configurable digital controlled oscillator 401 comprises that a N/M produces 502, one MUX 503 of 501, one adders of circuit, and the delayer 504 that delay is D.N/M produces circuit 501, MUX 503 and all control of controlled input signal of delayer D circuit 504.Because M=2 ^LDN, N/M produces circuit 501 does not need division, and simple shifting function just can be finished the function of N/M.According to the narration of front, for any N ≠ 2 ⁿTransform length, the basic structure of the pretreatment unit of configurable length-variable DFT described in the invention all is the same; For different transform length N, the main difference part that its pretreatment unit is finished corresponding function is the difference of N/M, and the relation of N and M can be determined by (17) to (19) formula in the present invention, and they are by control input configuration.Therefore, the function of multiphase filter is configurable, and the mode of the position of the interpolated data of generation changes with control input N and M.

Below be further specifying to the function of correction circuit among Figure 11:

With reference to (37) formula, correction circuit designs for the special circumstances during i=L-1 in the processing formula.The final output y (m) of pretreatment unit obtains by linear interpolation, and be used for data x ' that interpolation obtains y (m) (nL+i) and x ' (nL+i+1) input data x (n) are carried out obtaining behind the L times of up-sampling, they tighten adjacent y (m) in the position, one just in the front of y (m), and another one is then just in the back of y (m).See also (32) formula and (33) formula, (being under the situation of i＜L-1) under normal conditions, x ' (nL+i) and x ' (nL+i+1) can by same group import data [x (n), x (n-1) ..., x (n-P+1)] and multiply by different subfilter coefficient [c _i(0), c _i(1) ..., c _iAnd [c (P-1)] _I+1(0), c _I+1(1) ..., c _I+1(P-1)] then summation obtains.Because the input data of using are identical, the tap coefficient that we only need to change the subfilter of using just can obtain these two L times of up-sampling data with same circuit.But under the situation of i=L-1, above process is with regard to some difference.This moment x ' (nL+i) and original the same, by import data [x (n), x (n-1) ..., x (n-P+1)] and subfilter coefficient [c _i(0), c _i(1) ..., c _i(P-1)] multiplying each other then, summation obtains.And x ' (nL+i+1) be by the input data [x (n+1), x (n) ..., x (n-P+2)] and subfilter coefficient [c ₀(0), c ₀(1) ..., c ₀(P-1)] multiplying each other then, summation obtains.

Common method for designing, in order to comprise above two kinds of situations, need with one group of MUX come from two groups of input data [x (n), x (n-1) ..., x (n-P+1)] and [x (n+1), x (n) ..., x (n-P+2)] in choose one group and do the multiplication add operation.The resource of the device usefulness of Shi Xianing can be more like this, and scheduling is also difficult relatively.

The present invention has changed mentality of designing, and interpolation arithmetic and multiphase filter are combined, and will make the interpolation of antithetical phrase filter coefficient to the interpolation of L times of up-sampling data into.Realize for convenience, the main situation of considering i＜L-1 design in, just all finally export all by same group import data [x (n), x (n-1) ..., x (n-P+1)] produce, saved corresponding MUX, shown in (36) formula.At the special circumstances of i=L-1, handle by the modifying factor shown in (37) formula.Correction circuit finish corresponding to function shown in (37) formula.If required precision is not very high, correction circuit can not wanted, and this is in (38) formula explanation.

Multiplier that Figure 11 is required and adder be mainly in order to realize (35)-(38) formula, is in order to realize formula (35)-(37) more strictly speaking, and needing the multiplier number is the 3P+3 number; If do not want correction circuit, the multiplier number that needs is 3P.

Preferred embodiment of the present invention, the multiphase filter that pretreatment unit uses adopts L=32 times of up-sampling, subfilter length is elected P=6 as, carry out required sample rate conversion of pretreatment unit and low-pass filtering operation respectively for the data of importing at random that 35 kinds of DFT transform length according to LTE produce, its performance can satisfy the requirement of associated communication system.In this case, if comprise correction circuit, pretreatment unit shown in Figure 11 needs 21 multipliers; If omit correction circuit, the multiplier number that then needs is reduced to 18.If improve the clock frequency of operation, then can be in multiplexing corresponding calculation resources of different clocks cycle, required multiplier number will reduce at double.

In conjunction with the accompanying drawings a preferred embodiment of the present invention is had been described in detail above, but the present invention is not restricted to the foregoing description, under the spirit and scope situation of the claim that does not break away from the application, those skilled in the art can make various modifications and remodeling.

Claims (11)

1, the implementation method of the pretreatment unit of configurable length-variable DFT is used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, it is characterized in that, comprising:

(1) with N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

(2) the configuration multiphase filter comes the dateout behind the loop restructuring is carried out the integral multiple up-sampling, and for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed;

(3) utilize interpolation device that the integral multiple up-sampling data of multiphase filter output are carried out interpolation, obtain the dateout on the final outgoing position, to different transform length, the structure of the interpolation device that is maintained fixed.

2, the implementation method of pretreatment unit as claimed in claim 1 is characterized in that, step (2) further comprises:

For the up-sampling multiple is that L and subfilter length are the design of the multiphase filter of P, is translated into a design with wave digital lowpass filter of even symmetry and linear phase shift characteristic, all uses same digital filter h for all transform length N _d(n), its frequency domain parameter:

Free transmission range [0,1/L-Δ f ₁].

Transition band scope [1/L-Δ f ₁, 1/L+ Δ f ₂]

Stopband range [1/L+ Δ f ₂, 1]

Filter length N _p=LP

Δf ₁≈Δf ₂， $0<\mathrm{\&Delta;}{f}_2<\frac{2}{L}\&CenterDot;\frac{{\mathrm{\&delta;}}_{\min }}{1-{\mathrm{\&delta;}}_{\min }},$ $0<{\mathrm{\&delta;}}_{\min }<\frac{1}{2}$

Δ f ₁With Δ f ₂Be adjustable parameter.

3, according to the implementation method of the pretreatment unit of claim 1 or 2 described configurable length-variable DFTs, it is characterized in that, also comprise: input data length N and corresponding dateout length M relation are made respective limits, and making has a minimum relative distance between M and the N:

N≤M(1-δ _min)， $0<{\mathrm{\&delta;}}_{\min }<\frac{1}{2}$

Making M is that the form that satisfies the minimum of following formula is 2 ⁿInteger:

M＝2 ^LDN。

4, the implementation method of the pretreatment unit of configurable length-variable DFT as claimed in claim 2 is characterized in that, also comprises:

To wave digital lowpass filter h _d(n) window function w (n) is set reducing passband and stopband ripple,

h(n)＝h _d(n)·w(n)， n＝0，1，…，N _p-1

H (n) is the unit impulse response of final filter, h _d(n) response that is to use the digital filter design method to obtain, w (n) is the response of window function.

5, as the implementation method of the pretreatment unit of claim 1 or 4 described configurable length-variable DFTs, it is characterized in that, the configuration multiphase filter further comprises: the function that L subfilter realizes lowpass digital filter h (n) and L times of up-sampling is set, and the coefficient of i subfilter is

c _i(n)＝h(nL+i)，i＝0，1，…，L-1，n＝0，1，…，P-1

The structure of each subfilter is identical, all has P tap.

6, the implementation method of the pretreatment unit of configurable length-variable DFT as claimed in claim 1, it is characterized in that, the configuration multiphase filter further comprises and adopts the variable filter construction of coefficient to realize, total P tap, only need use multiphase filter to calculate and interpolation dateout adjacent two L times of up-sampling data on the position for linear interpolation, the position of interpolation dateout is determined by following formula:

pos(m)＝pos(m-1)+N/M，m＝1，2，…，pos(0)＝α ₀

α ₀Be initial position value, N and M are configurable, and corresponding location parameter is as follows:

Position for input data x (n)

Δ=pos (m)-n is interpolation dateout y (m) and x (n) alternate position spike;

Calculate first L times of used subfilter sequence number of up-sampling data:

The relative position on L sampling rate of interpolation dateout and first L times of up-sampling data is poor:

ε＝Δ·L-i

Calculating second used subfilter sequence number of L times of up-sampling data can obtain according to the i reckoning.

7, a kind of implementation method of pretreatment unit of configurable length-variable DFT is used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, it is characterized in that, comprising:

(1) with N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

(2) under the situation of linear interpolation, the operation of configurable multiphase filter and linear interpolation device is combined, constitute the sample rate conversion step of an integral body: use a variable filter construction of coefficient to realize the function of multiphase filter h (n) and linear interpolation, for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed, utilize interpolation device that multiphase filter subfilter coefficient is carried out interpolation, multiply each other with corresponding input data then and sue for peace and obtain dateout on the final outgoing position, to different transform length, the structure of the interpolation device that is maintained fixed.

8, the implementation method of the pretreatment unit of configurable length-variable DFT as claimed in claim 7 is characterized in that, under the situation of linear interpolation, configurable multiphase filter and linear interpolation device is combined, and its final dateout obtains according to the following steps;

Step (1) obtains the tap coefficient of equivalence to corresponding subfilter coefficient interpolation:

$\stackrel{\&OverBar;}{c}\left(k\right)=\left\{\begin{array}{cc}{c}_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_{i+1}\left(k\right)-c_i\left(k\right)],& i<L-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0(k+1)-c_i\left(k\right)],& i=L-1,k<P-1\\ c_i\left(k\right)+\mathrm{\&epsiv;}\&CenterDot;[c_0\left(0\right)-c_i\left(k\right)],& i=L-1,k=P-1\end{array}\right.$

Step (2) is calculated modifying factor:

${\mathrm{\&delta;}}_m\left(i\right)=\left\{\begin{array}{cc}0,& i<L-1\\ [x(n+1)-x(n-P+1)]\&CenterDot;\mathrm{\&epsiv;}{\&CenterDot;c}_0\left(0\right),& i=L-1\end{array}\right.$

Step (3) calculates final dateout:

$y\left(m\right)=\underset{k=0}{\overset{P-1}{\mathrm{\&Sigma;}}}\stackrel{\&OverBar;}{c}\left(k\right)\&CenterDot;x(n-k)+{\mathrm{\&delta;}}_m\left(i\right)$

Above computational methods will be converted into the interpolation calculation to two corresponding subfilter coefficients of adjacent position to the interpolation of the dateout of configurable multiphase filter, and the output result is added the final process result that a modifying factor obtains pretreatment unit.

9, a kind of pretreatment unit that is used for configurable length-variable DFT is used to finish N point input data to M=2 ^LDNThe sample rate conversion of some dateout is finished the low-pass filtering function to the input data simultaneously, it is characterized in that, comprising:

The loop restructuring circuit: be used for N the input data [x (0), x (1) ..., x (N-1)] be reconstructed into [x (0), x (1) ..., x (N-1), x (0), x (1) ..., x (P)] and output, for different transform length N, the value of P is maintained fixed;

The sample rate conversion circuit: it realizes the function of multiphase filter and interpolation device, multiphase filter is used for the dateout behind the loop restructuring is carried out the integral multiple up-sampling, for different transform length, multiphase filter adopts identical frequency domain parameter, and its up-sampling multiple L and subfilter length P are maintained fixed, interpolation device is used for the integral multiple up-sampling data of multiphase filter output are carried out interpolation, obtain the dateout on the final outgoing position, for different transform length, the interpolation device structure that is maintained fixed, the sample rate conversion circuit combines multiphase filtering and interpolation operation on implementation structure and finishes, and will be converted into the interpolation to corresponding subfilter coefficient to the interpolation equivalence of integral multiple up-sampling dateout.

10, pretreatment unit as claimed in claim 9 is characterized in that, the loop restructuring circuit further comprises the loop restructuring control logic circuit, reconstruct buffer memory, MUX, output FIFO;

Output FIFO is used for and the sample rate conversion circuit of back carries out rate-matched.

11, pretreatment unit as claimed in claim 9 is characterized in that, the sample rate conversion circuit further comprises tap renewal logical circuit, Coefficient Look-up Table, tap data delay line, and the operator circuit of finally finishing filtering and interpolation function;

Coefficient Look-up Table is deposited the tap coefficient of multiphase filter according to the subfilter order, and the tap coefficient of same subfilter leaves the adjacent position in;

The loop restructuring circuit is exported to tap data delay line with the result, the data of loop restructuring output in tap data delay line through after the phase delay factor, form one group of data [x (n+1), x (n), x (n-1) ..., x (n-P+1)];

The operator circuit comprises some multipliers and an add tree circuit, and contains a correction circuit, the dateout of tapped delay line [x (n), x (n-1) ..., x (n-P+1)] and corresponding tap coefficient [c ₀, c ₁..., c _P-1] multiply each other, the result who obtains and the result of correction circuit send into the addition of add tree circuit together, obtain the final output y (m) of pretreatment unit;

Tap is upgraded logical circuit and is received the control input, reads the coefficient of multiphase filter correspondence position subfilter from Coefficient Look-up Table, produces the tap coefficient [c of operator circuit through interpolation ₀, c ₁..., c _P-1];

Tap is upgraded logical circuit and is further comprised a configurable digital controlled oscillator circuit and a coefficient interpolation circuit;

Configurable digital controlled oscillator receives the control input, produces location of interpolation information and outputs it to the coefficient interpolation circuit; Configurable digital controlled oscillator comprises that a N/M produces circuit, an adder, a MUX, and the delayer that delay is D; N/M produces circuit, all control of controlled input signal of MUX and delayer D; The function of N/M can be finished by simple shifting function; Parameter N and M be by control input configuration, this determined the structure of multiphase filter be fix but its function is configurable;

The coefficient interpolation circuit receives the tap coefficient of the location of interpolation and the corresponding subfilter of multiphase filter that Coefficient Look-up Table provides of configurable digital controlled oscillator generation, produces and offers the used tap coefficient [c of operator circuit ₀, c ₁..., c _P-1].

Images