CN102984099A - Algorithm for digital predistortion fraction time delay estimation and signal alignment and achievement thereof - Google Patents

Abstract

The invention discloses an estimation method for achieving digital predistortion (DPD) system signal fraction ratio sampling interval time delay on the basis of a three-point two-times algorithm. The invention discloses a method for designing a front insertion filter by combining a window function and a Nyquist interpolation filter so as to achieve signal alignment. First, N points of DPD input signals with the same rate and N points of PA feedback output signals are simultaneously cut to perform correlated operation; then an integer time delay interval of the signals is estimated through the maximum module value positions of cross-correlation values, and two second largest values are taken simultaneously; then the three-point two-times algorithm is adopted to estimate the fraction ratio sampling interval time delay of two signals; and finally, and front insertion alignment of the PA feedback output signals is finished by combining a window function method according to total time delay. Achievement results of the algorithm prove that alignment signals excellent in performance can be provided for a DPD system. Compared with existing algorithms of the system, the method has the advantages that the calculated amount is small and digitized achievement is simple and the like. Simultaneously, the invention discloses an achievement scheme of the algorithm.

Description

The digital pre-distortion fractional delay is estimated and signal alignment algorithm and realization

Technical field

The present invention relates to a kind of signal time delay method of estimation of digitlization pre-distortion system and a kind of signal interpolation alignment schemes of digitlization pre-distortion system, belong to signal of communication digital processing technology field.

Background technology

Digital pre-distortion (DPD) is the main stream approach that power amplifier (PA) linearisation is proofreaied and correct, and its performance locks into temporal alignment always.Because the DPD system need consider the several above distortions of PA bandwidth, so sampling rate is very high, need to there be the well behaved Time Delay Estimation Algorithms algorithm that aligns with signal interpolation to ensure the realization of this system.

At present, the Time Delay Estimation Algorithms about signal in the DPD system mainly contains five kinds.One, the iterative method Time Delay Estimation Algorithms.Utilize to have some linear between signal, realize that by alternative manner DPD output and PA feed back output and align.Facts have proved: the method stability is affected by iteration step length, and the time-delay estimated accuracy is poor.Two, delay-locked loop (DLL) algorithm.Utilize the output clock of the voltage controlled oscillator VCO in the error signal control loop that the loop phase discriminator provides, realize that the arteries and veins position of the sampling pulse of PA feedback loop changes the purpose that reaches signal alignment.Facts have proved: the realization of this algorithm is complicated, convergence rate is slow.Three, the correlation function method.Divide again two kinds of amplitude difference correlation function method and Amplitude correlation function methods.The former gets amplitude difference to the corresponding sampling point of input signal and feedback signal, then to the poor cross-correlation of asking.Again by integer (times T between the maximum estimated signal _s) time delay; Compare with Amplitude correlation function signal delay time estimation method to be discussed below, increased and asked Interpolation Process.The latter is most popular signal synchronizing method, also is the estimation that utilizes the maximum value position realization signal time delay of correlation function, equally also realizes integer (times T _s) the time delay estimation.In the practice, in order to adopt the method realization to the estimation of fractional delay, need to carry out the high power interpolation to the input signal of DPD or to the feedback loop output signal of PA, extract by inserting space before again, after the formation multichannel, relevant respectively with the signal that does not carry out interpolation operation again, then get the peak value of each road correlated results, the comparison of again these peak values being put together is got and is wherein got maximum peak value, and the extraction signal of its correspondence is namely thought and alignd with another coherent signal.Though the method has lot of advantages, but shortcoming is also quite obvious: (1) can not fundamentally eliminate the fuzzy problem of time delay; (2) operand is excessive, and the DPD system can not follow the pace of change of PA parameter.In order to guarantee the performance of DPD, generally need the interpolation of 32 or 64 times even higher multiple to extract again.Four, based on the wavelet analysis delay time estimation method.After the Amplitude correlation function of signal is done wavelet analysis, realize the estimation of signal time delay.This searching optimal solution process operand based on the overall situation is also not little.Five, realize the estimation of time delay based on Fibonacci method.Suppose that at first the time delay scope is [a, b], adopt Fibonacci method optimizing in [a, b].Detailed process is: utilize golden section approach, 1 τ in the order sharp [a, b] ₁, and estimate with its coefficient to the raised cosine compensator, retell the filter that estimates coefficient and the feedback output convolution of PA and obtain exporting v (n), with v (n) and PA feedback loop output signal related operation, repeat above-mentioned this process again, until | τ _k-τ _K-1| less than a little constant ε who sets, namely exported synchronously v this moment _k(n).As seen there is priori to determine the computational complexity of system to the time delay scope of DPD system.Its essence is a kind of exhaustive method, and in this a series of calculating process, the numerical computations problem of algorithm occurs unavoidably.

The alignment schemes that realizes signal mainly contains three kinds of methods: one, and based on heavily adopting filter method.Adopt because of speed issue or hardware consumption problem and be not suitable for the DPD system.Two, based on polynoimal approximation.Utilized the multinomial of a low order at certain area the interpolation filter coefficient to be estimated, with the variation of adaptation signal.Its essence is a time varing filter, and the error of the new coefficient of the filter that the method realizes depends on the quantified precision of interpolation algorithm and old coefficient; Three.Based on the FRROW filter method.The method be with multinomial coefficient as after filter coefficient and the input convolution, realize signal alignment also needing to the filter weights correction in the calculating process with the time deviation convolution again, the numerical computations problem of existence also can have influence on the performance of DPD system.

In sum, propose based on the DPD signal time delay precise Estimation Method of " 3 secondaries " interpolation method with insert aliging of signal of filter method realization based on the signal " front " of window function method and have certain practical significance.By analysis, the approximate Rice channel model that meets Gauss's white noise of the channel of DPD and PA formation.Its cross-correlation has the quadratic function feature during two-forty sampling near its maximum.When the number of winning the confidence when being of convenient length, cross-correlation function is unimodal, for optimal method estimated signal time delay provides condition.Simultaneously, in conjunction with the multi-rate digital signal processing technology, the realistic existing interpolation filter of available window method can be realized aliging of signal with input after the convolution.

Summary of the invention

The present invention is intended to estimate (accurately part) and signal alignment problem for the signal time delay that solves the DPD system.

Solution thinking about the time delay estimation problem: similar with its digital system, also there is the sampling fuzzy problem in the DPD system.At labor think behind the PA channel model based on DPD: for the high-speed sampling of DPD, can utilize cross-correlation function near peak point, to have the quadratic function feature, get three points of its maximum and second largest value, can adopt " 3 secondaries " interpolation algorithm to estimate the signal time delay less than the sampling interval, thoroughly solve the offset issue that causes because of sampling between discrete correlation function and serial correlation peak of function.

Solution thinking about the interpolation alignment problem of signal: after having obtained the precise delay information of signal, according to the multi-rate digital signal processing technology, adopt the window function method that the PA feedback is exported and carry out " front inserting " recursion, realize the alignment of the signal in the DPD system.

The method of dealing with problems and step:

(1) PA feedback signal y (n) and DPD input signal d (n) are carried out the sampling of phase same rate, and get respectively simultaneously N point (L=2 ^M〉=N, M ∈ Z ⁺), it is temporary in dual port RAM;

(2) sampling 2L point FFT calculates the cross-correlation of y (n) and d (n); And extract among the r (n) and get maximum and two second largest values;

(3) according to the fractional spaced time delay τ between optimum theory estimation y (n) and the d (n);

(4) according to h _w(n)=(τ-n) calculates the interpolation filter coefficient to win (n) sa;

(5) according to Realize inserting before the signal.

Finish.

Algorithm of the present invention has following distinguishing feature:

Operand was low when (1) signal time delay of DPD was estimated.Concretely: the algorithm for estimating of fraction time sampling blanking time only needs just can finish by three displacements, three plus and minus calculations and twice multiplying.

(2) algorithm has used the cross-correlation function range value, and the Gaussian noise in the channel is the estimated accuracy of influence time not.According to the Optimum Theory method, the time delay valuation of fraction time sampling of the present invention interval is exactly the true value of the signal time delay fractional part in the DPD system.

(3) based on sa () " forward " the interpolated signal alignment schemes of window function method, have equally the low property of complexity, and inherited the various advantages of FIR filter.

(4) algorithm discussed of the present invention can be functional for the aligned signal that the DPD system provides.

Description of drawings

Fig. 1 is the DPD circuit system structural principle schematic diagram based on direct mode of learning.Wherein, empty frame has partly related to algorithm of the present invention among Fig. 1.

Fig. 2 is the whole implementation schematic block diagram of algorithm of the present invention.

Fig. 3 is the internal structure schematic block diagram of the signal alignment device of algorithm realization of the present invention.

Fig. 4 is the use flowage structure schematic block diagram of the present invention in the DPD system.

Fig. 5 is the Signal cross correlation computing schematic flow sheet in the algorithm of the present invention.

Fig. 6 is the fractional delay algorithm for estimating realization flow schematic diagram in the algorithm of the present invention.

Fig. 7 is that the signal " front " in the algorithm of the present invention is inserted alignment algorithm realization flow schematic diagram.

1. digital predistorters (DPD) among the figure, 2. digital to analog converter (DAC) becomes frequency mixer on 3., 4. power amplifier (PA), 5. antenna, 6. attenuator, 7. local oscillator, the time-delay estimator (T), 9. front plug in filter, 10. analog to digital converter, 11. lower change frequency mixer, 12. controller, 13. memories, 14. signal correction devices.Plug in filter (9) wherein and be made of three parts, they are: 91. filter coefficient calculators, 92. window added devices, 93. front plug in filters; Wherein window added device (92) is made of two parts, and they are: 921. window function coefficient memories, 922. multipliers.

Embodiment

1. signal fraction time sampling interval Time Delay Estimation Algorithms that is used for digital pre-distortion (DPD) system and a kind of in conjunction with window function and the Nyquist interpolation filter (method of the signal of sin (x)/x) design " front inserting " filter realization signal alignment.Algorithm of the present invention is applicable to proofread and correct based on the power amplifier of DPD technology (PA) linearisation.Wherein the input signal of DPD with the sampling rate of PA feedback loop output signal identical.The basic of algorithm is characterised in that: at first, carry out relevant to DPD with the output N point signal of PA.Then, maximum and two second largest values by this correlation function value adopt " 3 secondaries " method estimated signal fraction time sampling interval time delay.At last, adopt window function to realize that PA feedback loop output signal " front " plugs in filter and finish aliging of DPD input signal and PA feedback loop output signal.

2. implementation that is used for right 1 described algorithm.Its basic being characterised in that: lower synchronously at controller, intercept simultaneously the DPD input and export each N point with the feedback of PA, be put into two independently in the memory; Correlator to two independently the data in the memory do computing cross-correlation, two second largest values in the Output rusults and maximum are to the time-delay estimator; The time-delay estimator adopts the fraction time sampling interval time delay of " 3 secondaries " interpolation algorithm estimated signal according to input signal; The interpolation filter coefficient calculator is calculated the interpolator coefficient to window added device according to the output of time-delay estimator; To the windowing process of interpolation filter coefficient, and the result delivered to " front " interpolation filter and the PA feedback loop output signal carries out convolution algorithm by window added device, export the result of a DPD input signal alignment.Aforesaid operations is all worked in the presence of controller; Controller is subjected to total line traffic control.

3 realizations according to right 1 described algorithm is characterized in that: employed cross correlation algorithm comprises the steps:

The 3a intercepting has the input baseband signal d (n) and each N point of PA feedback loop output signal y (n) of the DPD of identical sampling rate, and wherein N is positive integer.

3b calculates the time domain cross-correlation r (n) of (2N-1) point of d (n) and y (n) according to Cross-Correlation Theory.

4 realizations according to right 1 described algorithm is characterized in that: the cross-correlation function r (n) that provides according to step 3 is provided employed fraction time sampling interval Time Delay Estimation Algorithms, extracts its maximum and second largest value; Again according to " 3 secondaries " interpolation algorithm estimated score sampling interval time delay (τ), and output.

5. according to the realization of right 1 described algorithm, it is characterized in that: employed " front " interpolation alignment algorithm comprises the steps:

5a adopts sampling function to calculate interpolation filter coefficient h (n) according to the fraction time sampling interval time delay (τ) of step 4, and lengthening is the window function of M to h (n), tries to achieve the interpolation filter coefficient h _w(n).

" front " interpolation filter coefficient h that 5b provides according to step 5a _w(n) with PA feedback loop output signal y (n) convolution and, the signal that aligns with the DPD input signal of output.

6 realizations according to right 1 described algorithm is characterized in that: adopt " 3 secondaries " algorithm to realize that estimation and signal " front " the interpolation alignment schemes of signal fraction time sampling interval time delay (τ) comprise the steps:

6a. the method according to the described intercept signal of step 3a is characterized in that: described step 3a carries out in the following manner: read simultaneously DPD input signal d (n) and each N point of PA feedback loop output signal y (n), and it is put into respectively the degree of depth is L=2 ^m, wherein (d (n) and y (n) be respectively according to formula (1) for m ∈ Z, the individual unit of L＞2N),

$\left\{\begin{array}{c}d\left(n\right)=d_i\left(n\right)+j\·d_Q\left(n\right)\\ y\left(n\right)=y_i\left(n\right)+j\·y_Q\left(n\right)\end{array}\right.---\left(1\right)$

Operation.

6b. according to the method for the cross-correlation of two limited long datas of the described calculating of step 3b, it is characterized in that: described step 3b carries out in the following manner: according to circumference correlation computations linear correlation algorithm, be at the same time or separately the 2L=2 of d (n) and y (n) ^M+1Point FFT is designated as respectively D _2L(k) and Y _2L(k), (k=0,1,2 ... 2L-1).And according to formula (2),

R _2L(k)＝D _2L(k)Y ^* _2L(k) (2)

Carry out, again according to formula (3),

r _2L(n)＝1/2N·iFFT _2L(R(k)) (3)

Operate.Wherein, n=0,1 ... 2L-1.

6c. according to the method for the described estimated signal fraction time sampling of step 4 interval time delay, it is characterized in that: described step 4 is undertaken by following operation: the maximum of getting { | r (n) | } is designated as f ₁, second largest value is designated as respectively f ₂, f ₃, again by formula (4),

$\mathrm{\&tau;}=\frac{f_3-{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000810890500011.png,HSA00000810890500031.png"\; state="\{\{state\}\}">f</figure-callout>}}_1}{{4f}_2-2f_1-2{\mathrm{<figure-callout\; id="3"\; label="f"\; filenames="HSA00000810890500011.png"\; state="\{\{state\}\}">f</figure-callout>}}_3}\&CenterDot;T_s---\left(4\right)$

Estimated score sampling interval time delay.If the integer time delay of signal is Γ, according to formula (5),

γ＝Γ+τ (5)

The overall delay value γ of estimated signal.

6d. the method according to the described calculating interpolation filter of step 5a coefficient is characterized in that: described step 5a is undertaken by following operation: adopt the window function method to calculate " front " and insert alignment filter coefficient h _wIf the length of window is the M point, by formula (6),

h _w(n)＝Window(k)×sin(γ-KT _s)/(γ-kT _s)，k＝0，1，…M (6)

Carry out.H wherein _wCan be symmetrical, also can be asymmetrical.

6e. insert the alignment algorithm according to step 5b described " front ", it is characterized in that: described step 5b is undertaken by following operation: with the input signal of PA feedback loop output signal y (n) as " front inserting " digital filter of pressing the described realization of step 5d, this filter output is by formula (7)

$\hat{d}\left(n\right)=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}y(i+n)\&CenterDot;h_w\left(i\right),n=\mathrm{0,1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N-1---\left(7\right)$

Carry out." front inserting " exportable signal that aligns with DPD input signal d (n) of digital filter

7. the implementation method of described algorithm according to claim 2, it is characterized in that: the realization of described algorithm is carried out as follows: d (n) and y (n) connect memory 13, memory 13 connects signal correction device 14, signal correction device 14 connects time-delay estimator 8, time-delay estimator 8 connects front slotting filter coefficient calculator 91, the front filter coefficient calculator 91 of inserting connects window added device 92, window added device 92 connects front plug in filter 93, the front filter 93 of inserting connects signal y (n) and window added device 92, front plug in filter 93 output signals

Interpolation alignment algorithm adopts functional unit as described below: front interpolation filter coefficient calculations calculator (91), window added device (92), front interpolation filter (93) are connected with controller 12.Controller 12 provides external reference interface function andlogic control.

According to algorithm of the present invention, realized the power amplifier prototype of a cover based on the base band digital predistortion device.Repeatedly test result shows: adopt algorithm of the present invention that the time unifying signal of a function admirable can be provided for the DPD system.Algorithm operation quantity of the present invention is little, and implementation structure is simple, and algorithm finds no any convergence problem in running and stable problem occurs.

Claims (7)

1. signal fraction time sampling interval Time Delay Estimation Algorithms that is used for digital pre-distortion (DPD) system and a kind of in conjunction with window function and the Nyquist interpolation filter (method of the signal of sin (x)/x) design " front inserting " filter realization signal alignment.Algorithm of the present invention is applicable to proofread and correct based on the power amplifier of DPD technology (PA) linearisation.Wherein the input signal of DPD with the sampling rate of PA feedback loop output signal identical.The basic of algorithm is characterised in that: at first, the output N point signal of DPD and PA is carried out related operation.Then, maximum and two second largest values by this correlation function value adopt " 3 secondaries " method estimated signal fraction time sampling interval time delay.At last, adopt window function to realize that PA feedback loop output signal " front " plugs in filter and finish aliging of DPD input signal and PA feedback loop output signal.

2. implementation that is used for right 1 described algorithm.Its basic being characterised in that: lower synchronously at controller, intercept simultaneously the DPD input and export each N point with the feedback of PA, be put into two independently in the memory; Correlator to two independently the data in the memory do computing cross-correlation, two second largest values in the output operation result and maximum are to the estimator of delaying time; The time-delay estimator adopts the mark fraction time sampling interval time delay of " 3 secondaries " interpolation algorithm estimated signal according to input signal; The interpolation filter coefficient calculator is calculated the interpolator coefficient to window added device according to the output of time-delay estimator; To the windowing process of interpolation filter coefficient, and the result delivered to " front " interpolation filter as the branch road weight coefficient of this filter by window added device; " front " interpolation filter and PA feedback loop output signal carry out convolution algorithm, export the result of a DPD input signal alignment.Aforesaid operations is all worked in the presence of controller; Controller is subjected to total line traffic control.

3. according to the realization of right 1 described algorithm, it is characterized in that: employed cross correlation algorithm comprises the steps:

The 3a intercepting has the input baseband signal d (n) and each N point of PA feedback loop output signal y (n) of the DPD of identical sampling rate, and wherein N is positive integer.

3b calculates the time domain cross-correlation r (n) of (2N-1) point of d (n) and y (n) according to Cross-Correlation Theory.

4. according to the realization of right 1 described algorithm, it is characterized in that: the cross-correlation function r (n) that provides according to step 3 is provided employed fraction time sampling interval Time Delay Estimation Algorithms, extracts its maximum and second largest value; Again according to " 3 secondaries " interpolation algorithm estimated score sampling interval time delay (τ), and output.

5. according to the realization of right 1 described algorithm, it is characterized in that: employed " front " interpolation alignment algorithm comprises the steps:

5a adopts Nyquist interpolation filter coefficient h (n) according to the fraction time sampling interval time delay (τ) of step 4, and lengthening is the window function of M to h (n), tries to achieve the interpolation filter coefficient h _w(n).

" front " interpolation filter coefficient h that 5b provides according to step 5a _w(n) with PA feedback loop output signal y (n) convolution and, the signal that aligns with the DPD input signal of output.

6. according to the realization of right 1 described algorithm, it is characterized in that: adopt " 3 secondaries " algorithm to realize that estimation and signal " front " the interpolation alignment schemes of signal fraction time sampling interval time delay (τ) comprise the steps:

6a. the method according to the described intercept signal of step 3a is characterized in that: described step 3a carries out in the following manner: read simultaneously DPD input signal d (n) and each N point of PA feedback loop output signal y (n), and it is put into respectively the degree of depth is L=2 ^m, wherein (d (n) and y (n) be respectively according to formula (1) for m ∈ Z, the individual unit of L＞2N),

$\left\{\begin{array}{c}d\left(n\right)=d_i\left(n\right)+j\&CenterDot;d_Q\left(n\right)\\ y\left(n\right)=y_i\left(n\right)+j\&CenterDot;y_Q\left(n\right)\end{array}\right.---\left(1\right)$ Operation.

6b. according to the method for the cross-correlation of two limited long datas of the described calculating of step 3b, it is characterized in that: described step 3b carries out in the following manner: according to circumference correlation computations linear correlation algorithm, be at the same time or separately the 2L=2 of d (n) and y (n) ^M+1Point FFT is designated as respectively D _2L(k) and Y _2L(k), (k=0,1,2 ... 2L-1).And according to formula (2),

$R_{2L}\left(k\right)=D_{2L}\left(k\right)Y_{2L}^{*}\left(k\right)---\left(2\right)$ Carry out, again according to formula (3),

r _2L(n)＝1/2N·iFFT _2L(R(k)) (3)

Operate.Wherein, n=0,1 ... 2L-1.

6c. according to the method for the described estimated signal fraction time sampling of step 4 interval time delay, it is characterized in that: described step 4 is undertaken by following operation: the maximum of getting { | r (n) | } is designated as f ₁, second largest value is designated as respectively f ₂, f ₃, again by formula (4),

$\mathrm{\&tau;}=\frac{f_3-f_1}{{4f}_2-2f_1-2f_3}\&CenterDot;T_s---\left(4\right)$ Carrying out the fraction time sampling interval estimates.If the integer sampling interval time delay of signal is Γ, according to formula (5),

The overall delay value γ of γ=Γ+τ (5) estimated signal.

6d. the method according to the described calculating interpolation filter of step 5a coefficient is characterized in that: described step 5a is undertaken by following operation: adopt the window function method to calculate " front " and insert alignment filter coefficient h _wIf the length of window is the M point, by formula (6),

h _w(n)＝Window(k)×sin(γ-kT _s)/(γ-kT _s)，k＝0，1，…M (6)

Carry out.H wherein _wCan be symmetrical, also can be asymmetrical.

6e. insert the alignment algorithm according to step 5b described " front ", it is characterized in that: described step 5b is undertaken by following operation: with the input signal of PA feedback loop output signal y (n) as " front inserting " digital filter of pressing the described realization of step 5d, this filter output is by formula (7)

$\hat{d}\left(n\right)=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}y(i+n)\&CenterDot;h_w\left(i\right),n=\mathrm{0,1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N-1---\left(7\right)$

Carry out." front inserting " exportable signal that aligns with DPD input signal d (n) of digital filter

7. the implementation method of described algorithm according to claim 2, it is characterized in that: the realization of described algorithm is carried out as follows: d (n) and y (n) connect memory 13, memory 13 connects signal correction device 14, signal correction device 14 connects time-delay estimator 8, time-delay estimator 8 connects front slotting filter coefficient calculator 91, the front filter coefficient calculator 91 of inserting connects window added device 92, window added device 92 connects front plug in filter 93, the front filter 93 of inserting connects signal y (n) and window added device 92, front plug in filter 93 output signals

Interpolation alignment algorithm adopts functional unit as described below: front interpolation filter coefficient calculations calculator (91), window added device (92), front interpolation filter (93) are connected with controller 12.Controller 12 provides external reference interface function andlogic control.

Images