CN101771639A - Predistortion parameter processing method and device - Google Patents

Abstract

The invention provides a predistortion parameter processing method and a device; wherein the method comprises the following steps: an initial signal matrix is built by utilizing data which carries out predistortion treatment and power amplification feedback data after periodic filtering processing begins; the data which carries out predistortion treatment is a reference signal, and the data for power feedback is an input signal; QR decomposition processing is carried out to the built initial signal matrix, and the predistortion parameters are determined by carrying out backward iteration operation to the matrix which carries out the QR decomposition; by adopting the method and the device, the arithmetic workload and the arithmetic complexity can be effectively reduced; in addition, the determined predistortion parameters begin to be stored from Nth sampling point symbol, so as to well save the expenses of the system operation.

Description

A kind of processing method of pre-distortion parameters and device

Technical field

The present invention relates to communication technical field, relate in particular to a kind of processing method and device of pre-distortion parameters.

Background technology

At present, Large scale construction along with global 3G network, operator more and more pays attention to reducining the construction costs and maintenance cost, and power amplifier is as one of device the most expensive in the middle of the communication system, its requirement to efficient is also more and more higher, thereby makes digital pre-distortion DPD (Digital Pre-Distortion) technology obtain very fast development.Wherein, power amplifier has closely with filter again gets in touch: when filter was output as the linear function of input, this filter was a linear filter, otherwise was nonlinear filter, if parameter conversion in time is referred to as time varing filter again; And according to the characteristic of radio-frequency power amplifier (PA), when the statistical property conversion of input process, the process of the own parameter of sef-adapting filter adjustment is referred to as " tracking " process, and the purpose of predistortion is exactly to follow the tracks of the variation of PA.At present, when the statistical property of input process changes, the own parameter of sef-adapting filter adjustment is when satisfying the requiring of certain criterion, and promptly when carrying out pre-distortion, the criterion of employing is minimum mean square error criterion (MMSE) or criterion of least squares (LS) normally.

But, above-mentioned two kinds of different criterions have different characteristics respectively, when carrying out pre-distortion, at first want the characteristic of clear and definite selected algorithm: least mean-square error (MMSE) linear filter when being steady to input process is called Weiner filter, and Weiner filter satisfies normal equation, it directly is pre-distortion parameters to the signal that matrix inversion obtains, promptly

But adopt this method operand bigger, canned data is also more simultaneously; Thereby, prior art has proposed a kind of LMS algorithm of effectively separating normal equation on this basis, this method is by the one group of filter coefficient that square obtains to the class data long-time statistical error that obtains, but for power amplifier, can only obtain one group of data in a period of time, thereby can only estimate and be similar to, so this method and be not suitable for pre-distortion to PA to the long-time statistical characteristic.

And least square is the optimum filtering to one group of given data, and this method generally is applicable to the pre-distortion of PA at present: utilize M rank linear filter to estimate known signal d (i), wherein being estimated as follows d (i) according to known N data x (n):

0≤i≤N wherein, for least-squares algorithm, w _m(i) optimum value makes weighted cumulative square error performance function for minimum, that is:

Though this least square method also can obtain pre-distortion parameters by the inverse matrix of separating normal equation, operand is still bigger, and prior art proposes to try to achieve pre-distortion parameters, that is: w by RLS (Recursive Least-Squares, recursive least-squares) for this reason _m(n)=w _m(n-1)+g _m(n) e (n), wherein error signal is: The recurrence formula of this RLS and LMS is similar, and difference only is a gain coefficient, promptly

And C _m(n-1) also obtain, i.e. autocorrelation matrix according to recurrence formula:

Therefore, as long as adopt this kind method to utilize the initial pre-distortion parameters w of predistortion _m(0), signals after pre-distortion x _m(0) and intermediate variable C _m(0) just can be in the hope of described pre-distortion parameters; Though RLS can carry out optimum filtering to known one group of data, it mainly still carries out recursion at the correlation matrix of input data, calculates relative complex, the big (O[N of operand ³]), and under the environment of measuring noise or non-stationary, also can have the problem of stability;

Therefore, propose a kind of QRD-RLS of utilization (QR decomposes recursive least-squares) algorithm in the prior art and carried out the method for pre-distortion, this QRD-RLS is a kind of improvement to the RLS algorithm, this kind algorithm is not that the correlation matrix of importing data is carried out recursion, but directly input data matrix is carried out recursion, operand significantly reduces, and good data stability is arranged simultaneously, so be fit to for the DPD systematic comparison; But this algorithm still has the defective of himself, and promptly each output sample all needs to carry out O[N ²] inferior multiplying, and for a real-time system, the multiplication amount is still still very big, for example only carries out 1000 samplings, N=60 then needs to carry out multiplication 3,600,000 times, implements the comparison difficulty for existing chip.

Summary of the invention

In view of this, the problem that the present invention solves provides a kind of processing method and device of pre-distortion parameters, effectively low operand of valency and computational complexity.

For addressing the above problem, technical scheme provided by the invention is as follows:

A kind of processing method of pre-distortion parameters comprises: after periodic filter is handled beginning, utilize data and power amplifier feedback data after the pre-distortion to make up the initialize signal matrix; Data after the described pre-distortion are reference signal, and described power amplifier feedback data is an input signal; Initialize signal matrix to described structure carries out the QR resolution process; Carry out the back by the matrix after QR is decomposed and determine pre-distortion parameters to interative computation.

Preferably, described initialize signal matrix to described structure carries out the QR resolution process and is specially: adopt quadrature diagonalization mode to carry out the iteration diagonalization and handle being increased to each new data vector in the described initialize signal matrix.

Preferably, the vector of each row is the data of same sampling instant input in the initialize signal matrix.

Preferably, this method also comprises: after periodic filter begins each time, preserve the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine.

Preferably, this method also comprises: whether the error signal of periodically judging described definite pre-distortion parameters exceeds default concussion scope in the default time period, if, then with the alternative fixed pre-distortion parameters of new default true parameter.

Preferably, adopt systolic structures that described initialize signal matrix is carried out the QR resolution process.

A kind of processing unit of pre-distortion parameters comprises: construction unit, resolving cell and arithmetic element; Wherein, described construction unit is used for after periodic filter is handled beginning, is reference signal, is that input signal makes up the initialize signal matrix with the power amplifier feedback data with the data after the pre-distortion; Described resolving cell is used for the initialize signal matrix that construction unit makes up is carried out the QR resolution process; Described arithmetic element is used for carrying out the back by the matrix after QR is decomposed and determines pre-distortion parameters to interative computation.

Preferably, described resolving cell comprises: first processing module and second processing module; Wherein, described first processing module is used for adopting quadrature diagonalization mode that each the new data vector that is increased to described initialize signal matrix is carried out iteration diagonalization processing; Described second processing module is used to adopt systolic structures that described initialize signal matrix is carried out the QR resolution process.

Preferably, this device also comprises: memory cell; Wherein, described memory cell is used for after periodic filter begins each time, preserves the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine.

Preferably, this device also comprises: judging unit; Described judging unit is used for periodically judging whether the error signal of the pre-distortion parameters that described arithmetic element is determined exceeds default concussion scope in the default time period, if then substitute fixed pre-distortion parameters with new default true parameter.

As can be seen, adopt method and apparatus of the present invention, after periodic filter is handled beginning, utilize the data after the pre-distortion to make up the initialize signal matrix as input signal as reference signal, power amplifier feedback data, again described initialize signal matrix is carried out the QR resolution process, and by the back determine pre-distortion parameters to interative computation, thereby effectively reduced operand and computational complexity; And just preserve the pre-distortion parameters that computing is determined since N sampling point symbol, thereby well saved system's operation expense.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, to do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below, apparently, accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the system configuration schematic diagram that utilizes data after the predistortion and power amplifier feedback data to determine pre-distortion parameters in the embodiment of the invention;

Fig. 2 is the method flow schematic diagram of the embodiment of the invention 1;

Fig. 3 is the structural representation of pulsing and handling in the method for the embodiment of the invention 1;

Fig. 4 is the apparatus structure schematic block diagram of the embodiment of the invention 2.

Embodiment

After basic thought of the present invention is that periodic filter is handled beginning, utilize the data after the pre-distortion to make up the initialize signal matrix as input signal as reference signal, power amplifier feedback data, again described initialize signal matrix is carried out the QR resolution process, and by then determining pre-distortion parameters to interative computation, thereby effectively reduced operand and computational complexity, and well saved system's operation expense.

For the ease of understanding, at first QRD-RLS is simply introduced: the RLS algorithm that decomposes based on QR can be so that the information matrix diagonalization, to avoid matrix inversion; Promptly be actually a matrix Y and carry out the QR decomposition, because Y generally is a nonsingular matrix, so can change into quadrature (tenth of the twelve Earthly Branches) matrix Q (Q ^HQ=I) with the product of nonsingular upper triangular matrix R.This is not that the correlation matrix of input data is inverted, but directly input data matrix is carried out the mode of recursion, good data stability is arranged, and can realize fast.As in embodiments of the present invention,, finally can obtain pre-distortion coefficients W according to feedback signal Y and input intermediate-freuqncy signal X; Certainly, the algorithm that decomposes based on QR has multiple, and wherein GIVENS rotation algorithm performance is best, and it provides optimal way of realization aspect numerical characteristic and hardware enforcement, does not repeat them here.

Mainly be to utilize data and power amplifier feedback data after the predistortion to determine pre-distortion parameters in the present embodiment; It should be noted that wherein scheme of the prior art is that the data y that PA returns is carried out finding the solution of matrix as a reference, the filter coefficient that obtains so also needs to invert and just can get to the end pre-distortion parameters; And present embodiment proposes the data z after the predistortion as the reference signal, and the pre-distortion coefficients of then trying to achieve is exactly the inverse function of power amplifier: w=pa ^H, wherein H represents inversion operation, and the data y that returns with PA is as input signal then, and the data y that described pre-distortion parameters w and PA are returned multiplies each other, and the pre-distortion parameters w that relatively obtains of the data z that comes out with predistorter is both by being asked again; Concrete, as shown in Figure 1, the supposition of the pre-distortion parameters of the high speed predistorter among the figure (vector adjuster) is pa ^H, z=xpa is then arranged ^H, the data of PA are thought y=zpa=(xpa after the adding predistorter ^H) pa, then can obtain following derivation:

From then on formula can find out that the pre-distortion parameters w that the data utilized after the predistortion and power amplifier feedback data are determined is exactly the inverse function of PA, and this w can directly be stored in and uses among the LUT, therefore save the needed operand of inverting on the one hand, avoided the error of inverting in addition on the one hand.

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described; Obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that is obtained under the creative work prerequisite.

Referring to Fig. 2, be the processing method of the embodiment of the invention 1 pre-distortion parameters, this method comprises:

In step 201, after periodic filter is handled beginning, utilize data and power amplifier feedback data after the pre-distortion to make up the initialize signal matrix; Data after the described pre-distortion are reference signal, and described power amplifier feedback data is an input signal; Concrete, present embodiment is that data and the power amplifier feedback data after QRD-RLS algorithm technical utilizes predistortion determined pre-distortion parameters; Wherein, the QRD-RLS algorithm mainly is that one group of equation is found the solution minimum error signal, and following group of equation promptly is to ask (w ₀, w ₁... w _N) so that e (1), e (2) ... e (M) minimum:

y ₁(1)w ₀+y ₂(1)w ₁+…y _N(1)w _N＝z(1)+e(1)

y ₁(2)w ₀+y ₂(2)w ₁+…y _N(2)w _N＝z(2)+e(2)

.

.

.

y ₁(M)w ₀+y ₂(M)w ₁+…y _N(M)w _N＝z(M)+e(M)

Wherein the target of least square (LS) is exactly the data of finding in a period of time, makes

Minimum, λ is a memory fact; And As time goes on, the shared ratio of the error in past is more and more littler, thereby top formula can be write as matrix form: Yw=z+e, wherein Y is the matrix of input signal in a period of time, z is a reference signal vector, and e is an error signal, and W is a pre-distortion parameters; Wherein, can decompose matrix Y by matrix Q and matrix R has Y=QR, and Y decomposes by QR and obtains upper triangular matrix R, and Z decomposes by QR and obtains another one vector z ':

$Y\·w=Z\⇒Q^{H}Y\·w=Q^{H}Z\⇒Q^H\mathrm{QR}\·w=Q^{H}Z,$ $Q^{H}Q=I\⇒R\·w=z^{\′},z^{\′}=Q^{H}Z\⇒w=R^H\·z^{\′}$

Wherein, input signal is multiplied by matrix Q ^HThe QR that both can finish matrix Y decomposes; If the data Y initial set that PA returns is dressed up upper triangular matrix, then the Y matrix after the assembling equals matrix R, that is: Y=R: thereby after periodic filter was handled beginning, the initial assembling of matrix R was as shown in the formula signal:

$\mathrm{BIG}\_R\left(0\right)=\left[\begin{array}{cc}{u}^T\left(0\right)& z_0\left(0\right)\\ R\left(0\right)& \mathrm{\Γ}\left(0\right)\end{array}\right]=\begin{array}{}\left[\begin{array}{ccccc}{u}_{\mathrm{0,1}}\left(0\right)& u_{\mathrm{0,2}}\left(0\right)& ...& u_{0,N}\left(0\right)& z_0\left(0\right)\\ u_{\mathrm{1,1}}\left(0\right)& u_{\mathrm{1,2}}\left(0\right)& ...& u_{1,N}\left(0\right)& z_1\left(0\right)\\ 0& u_{\mathrm{2,2}}\left(0\right)& ...& u_{2,N}\left(0\right)& z_2\left(0\right)\\ & & .& .& .\\ 0& 0& .& .& .\\ & & .& .& .\\ 0& ...& 0& u_{N-1,N}\left(0\right)& z_{N-1}\left(0\right)\\ & 0& 0& u_{N,N}\left(0\right)& z_N\left(0\right)\end{array}\right]\end{array}$

Wherein, the size of upper triangular matrix is N*N (N=Q* (M+1)),

u ^T(0)=u _0,1(0), u _0,2(0) ... u _{0, N}(0)=and y (N+1), y (N+1) ³... y (N+1-M) ^Q, Γ (0)=z ₁(0), z ₂(0) ... z _N-1(0), z _N(0); Initial set is dressed up after the merit, and every input data line moment point adds 1, above each element of R (0) matrix as follows, the rest may be inferred:

u _1，1(0)，u _1，2(0)，…u _1，N(0)＝y(N)，y(N) ³，…y(N-M) ^Q

...

u _N-1，N-1(0)，u _N-1，N(0)＝y(1)，y(1) ³

u _N，N(0)＝y(0)

Initialization with respect to prior art makes up Γ (0)=0 ^T(δ 〉=0), the mode that present embodiment makes up the initialize signal matrix can make pre-distortion parameters reach more stable state fast, and promptly the initialization time of predistortion is shorter; While is owing to the matrix y (N) of input signal, y (N) ³... y (N-M) ^QBoth considered signal non-linear (1,2 ... Q), also considered signal Memorability (0,1 ... M), thus make that having comprised the nonlinear distortion that signal is had a Memorability when generating pre-distortion parameters handles.

In step 202, the initialize signal matrix of described structure is carried out the QR resolution process; In step 203, carry out the back by the matrix after QR is decomposed and determine pre-distortion parameters to interative computation;

Wherein, present embodiment proposes to carry out QR by GIVENS rotation and systolic structures and decomposes, but is not limited thereto;

A, the initial assembling that periodically updates each time make up finish after, by the GIVENS rotation every vector y that imports delegation being eliminated is zero, thereby obtain upper triangular matrix R, promptly use the diagonalizable method of quadrature (GIVENS rotation) that each the new data vector that is increased among the BIG_R (0) is carried out iteration diagonalization processing; And input signal has only simple multiplying when GIVENS rotates, and does not have matrix operation, and operand significantly reduces, and is that input signal is found the solution all the time, but not the coherent signal of input, the error of calculation reduces; The vector that it should be noted that each row in the signal matrix that makes up after the initialization is the data of same sampling instant input.In the GIVENS rotation, with a series of GIVENS spin matrixs of initialize signal matrix B IG_R (0) premultiplication that make up, with each element of the row of first in the cancelling; For example, finish after the GIVENS rotation for the first time, obtain following formula and illustrate:

$\mathrm{BIG}\_\stackrel{\‾}{R}\left(0\right)=\left[\begin{array}{cc}0& \tilde{e}\left(1\right)\\ R\left(1\right)& \mathrm{\Γ}\left(1\right)\end{array}\right]$

Concrete, at n sampled point constantly, the signal matrix that initial assembling makes up is as follows:

$\mathrm{BIG}\_R\left(n\right)=\left[\begin{array}{cc}{u}^T\left(n\right)& z_0\left(n\right)\\ R\left(n\right)& \mathrm{\Γ}\left(n\right)\end{array}\right]$

U wherein ^T(n) be the input signal that returns at n moment PA, z ₀(n) be n reference signal constantly, then have:

u ^T(n)＝u _0，0(n)，u _0，1(n)，…u _0，N(n)＝y(n+1)，y(n+1) ³，…y(n+1-M) ^Q

Γ(n)＝z ₁(n)，z ₂(n)，…z _N-1(n)，z _N(n)

Use the GIVENS rotation then, make the signal u of new input ^T(n) all be rotated and be null value, simultaneously z ₀(n) also be rotated; But because of z ₀(n) often can not be rotated and be null value, so the later z of rotation ₀(n) signal is error signal; In this GIVENS rotation, with a series of GIVENS spin matrixs of matrix B IG_R (n) premultiplication, with the element in the row of first in the cancelling, wherein each element in this first row all has corresponding rotating vector: Q _m(n), m=1,2 ... N, then the twiddle factor of this N that imports constantly signal correspondence is exactly

Expression formula as follows:

$\tilde{Q}\left(n\right)=Q_1\left(n\right)\·Q_2\left(n\right)\·\·\·Q_N\left(n\right)$

$=\left[\begin{array}{cccccc}\cos {\mathrm{\θ}}_1\left(n\right)& -\sin {\mathrm{\θ}}_1\left(n\right)& 0& ...& & 0\\ {\sin }^{*}{\mathrm{\θ}}_1\left(n\right)& \cos {\mathrm{\θ}}_1\left(n\right)& 0& ...& & \\ & & & & .& \\ 0& 0& 1& 0& .& \\ & & & & .& \\ .& .& & & & \\ .& .& 0& 1& 0& \\ .& .& & & & \\ & & & & .& \\ & & & & .& 0\\ & & & & .& \\ 0& ...& & & 0& 1\end{array}\right]\·\left[\begin{array}{cccccc}\cos {\mathrm{\θ}}_2\left(n\right)& 0& -\sin {\mathrm{\θ}}_2\left(n\right)& ...& & 0\\ 0& 1& 0& ...& & \\ & & & & .& \\ {\sin }^{*}{\mathrm{\θ}}_2\left(n\right)& 0& \cos {\mathrm{\θ}}_2\left(n\right)& 0& .& \\ & & & & .& \\ .& .& & & & \\ .& .& 0& 1& 0& \\ .& .& & & & \\ & & & & .& \\ & & & & .& 0\\ & & & & .& \\ 0& ...& & & 0& 1\end{array}\right]\·\·\·$

$\left[\begin{array}{cccccc}\cos {\mathrm{\θ}}_N\left(n\right)& 0& & ...& & -\sin {\mathrm{\θ}}_N\left(n\right)\\ 0& 1& 0& ...& & \\ & & .& & .& \\ & 0& .& 0& .& \\ & & .& & .& \\ .& .& & & & \\ .& .& 0& 1& 0& \\ .& .& & & & \\ & & & & 1& 0\\ {\sin }^{*}{\mathrm{\θ}}_N\left(n\right)& ...& & & 0& \cos {\mathrm{\θ}}_N\left(n\right)\end{array}\right]$

With this twiddle factor The signal matrix that makes up with described new assembling multiplies each other and promptly obtains

At first consider matrix Q ₁(n) and the product of BIG_R (n), i.e. rotation for the first time:

$Q_1\left(n\right)\&CenterDot;\mathrm{BIG}\_R\left(n\right)=\left[\begin{array}{cccccc}\cos {\mathrm{\&theta;}}_1\left(n\right)& -\sin {\mathrm{\&theta;}}_1\left(n\right)& 0& ...& & 0\\ {\sin }^{*}{\mathrm{\&theta;}}_1\left(n\right)& \cos {\mathrm{\&theta;}}_1\left(n\right)& 0& ...& & \\ & & & & .& \\ 0& 0& 1& 0& .& \\ & & & & .& \\ .& .& & & & \\ .& .& 0& 1& 0& \\ .& .& & & & \\ & & & & .& \\ & & & & .& 0\\ & & & & .& \\ 0& ...& & & 0& 1\end{array}\right]\&CenterDot;\left[\begin{array}{ccccc}{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}\left(n\right)& {\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,2}}\left(n\right)& ...& {\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{0,N}\left(n\right)& z_0\left(n\right)\\ {\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}\left(n\right)& {\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,2}}\left(n\right)& ...& {\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{1,N}\left(n\right)& z_1\left(n\right)\\ 0& {\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{2,2}}\left(n\right)& ...& {\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{2,N}\left(n\right)& z_2\left(n\right)\\ & & .& .& .\\ 0& 0& .& \text{.}& .\\ & & .& .& .\\ 0& ...& u_{N-1,N-1}\left(n\right)& u_{N-1,N}\left(t\right)& z_{N-1}\left(n\right)\\ & 0& 0& u_{N,N}\left(t\right)& z_N\left(n\right)\end{array}\right]$

$={\left[\begin{array}{ccccc}0& {\stackrel{\&OverBar;}{u}}_{\mathrm{0,1}}\left(n\right)& ...& {\stackrel{\&OverBar;}{u}}_{0,N}\left(n\right)& \stackrel{\&OverBar;}{z}\left(n\right)\\ {\stackrel{\&OverBar;}{u}}_{\mathrm{1,1}}\left(n\right)& {\stackrel{\&OverBar;}{u}}_{\mathrm{1,2}}\left(n\right)& ...& {\stackrel{\&OverBar;}{u}}_{1,N}\left(n\right)& \stackrel{\&OverBar;}{z}(n-1)\\ 0& {\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{2,2}}\left(n\right)& ...& {\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{2,N}\left(n\right)& z(n-2)\\ & & .& .& .\\ 0& 0& .& .& .\\ & & .& .& .\\ 0& ...& u_{N-1,N-1}\left(n\right)& u_{N-1,N}\left(n\right)& z(n-N-1)\\ & 0& 0& u_{N,N}\left(n\right)& z(n-N)\end{array}\right]}_{(N+1)\&times;(N+1)}$

After rotation for the first time, be positioned at the element u of (1,1) position _0,1(n) promptly be null value by being eliminated, first row of matrix and all data of second row are also all rotated, but other elements of going remain unchanged, without any computing; And wherein, first rotation element cos θ of new input ₁(n) and sin θ ₁(n) satisfy following equation:

Can try to achieve rotation element cos θ thus ₁(n) and sin θ ₁(n), as follows:

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1=\sqrt{{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}{\left(n\right)}^2+{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}{\left(n\right)}^2}\\ \cos {\mathrm{\&theta;}}_1\left(n\right)=\frac{{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}\left(n\right)}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}\\ \sin {\mathrm{\&theta;}}_1\left(n\right)=\frac{{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}\left(n\right)}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}\end{array}\right.$

Because u _0,1(n) and u _1,1(n) all being plural number, still is the selection problem of real arithmetic so exist complex operation this moment; Prior art obtain r1 be by

Perhaps

Obtain; The r that the former obtains ₁Also will be plural number, and the complete reservation of the latter u _1,1(n) if phase information is signal u _0,1(n), u _1,1(n) only be real number, like this to r ₁Find the solution all correctly, but then step is accurate for complex operation; Therefore present embodiment proposes r ₁Calculating should be

Promptly rotate the phase theta of element ₁Not only by u _1,1(n) decision also is subjected to r simultaneously ₁The influence of phase place; Corresponding rotation element cos θ ₁(n) and sin θ ₁(n) should be as follows:

$\left\{\begin{array}{cc}\cos {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}\left(t\right)-\sin {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}\left(n\right)=0& \\ {\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">cos</figure-callout>}}^2{\mathrm{\&theta;}}_1\left(n\right)+{\left|\sin {\mathrm{\&theta;}}_1\left(n\right)\right|}^2=1,& \sin {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\left[\sin {\mathrm{\&theta;}}_1\left(n\right)\right]}^{*}={\left|\sin {\mathrm{\&theta;}}_1\left(n\right)\right|}^2\end{array}\right.$

And the rotation element cos θ that tries to achieve thus ₁(n) and sin θ ₁(n), as follows:

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1=\sqrt{{\left|{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}\left(n\right)\right|}^2+{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}{\left(n\right)}^2}\\ \cos {\mathrm{\&theta;}}_1\left(n\right)=\frac{{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{1,1}}\left(n\right)}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}\\ \sin {\mathrm{\&theta;}}_1\left(n\right)=\frac{{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{\mathrm{0,1}}\left(n\right)}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}\end{array}\right.$

By rotation element cos θ ₁(n) and sin θ ₁(n) can eliminate the element that first row first is listed as, next rotate other elements of preceding two row, handle as follows:

${\stackrel{\&OverBar;}{u}}_{0,m}\left(n\right)=\cos {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{0,m}\left(n\right)-\sin {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{1,m}\left(n\right)$ (m＝2，…，N)

${\stackrel{\&OverBar;}{u}}_{1,m}\left(n\right)={\sin }^{*}{\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{0,m}\left(n\right)+\cos {\mathrm{\&theta;}}_1\left(n\right)\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="H2008102470504E0000011.png,H2008102470504E0000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_{1,m}\left(n\right)$ (m＝1，…，N)

Wherein

The numeric ratio that obtains is more special,

So the numerical value of this position does not need to calculate again later on:

Finish at last after N the rotation, at the element u of the n new input of first row constantly _0,1:N(n) all be rotated and be null value, finished this GIVENS rotation of n constantly,

Mathematic(al) representation as follows:

$\mathrm{BIG}\_\stackrel{\&OverBar;}{R}\left(n\right)=\left[\begin{array}{ccccc}0& 0& ...& 0& e\\ {\stackrel{\&OverBar;}{u}}_{\mathrm{1,1}}\left(n\right)& {\stackrel{\&OverBar;}{u}}_{1,2}\left(n\right)& ...& {\stackrel{}{\stackrel{\&OverBar;}{u}}}_{1,N}\left(n\right)& {\stackrel{\&OverBar;}{z}}_1\left(n\right)\\ 0& {\stackrel{\&OverBar;}{u}}_{2,2}\left(n\right)& ...& {\stackrel{\&OverBar;}{u}}_{2,N}\left(n\right)& {\stackrel{\&OverBar;}{z}}_2\left(n\right)\\ & & .& .& .\\ 0& 0& .& .& .\\ & & .& .& .\\ 0& ...& {\stackrel{\&OverBar;}{u}}_{N-1,N-1}\left(n\right)& {\stackrel{\&OverBar;}{u}}_{N-1,N}\left(n\right)& {\stackrel{\&OverBar;}{z}}_{N-1}\left(n\right)\\ & 0& 0& {\stackrel{\&OverBar;}{u}}_{N,N}\left(n\right)& {\stackrel{\&OverBar;}{z}}_N\left(n\right)\end{array}\right]$

After finishing n renewal constantly, enter n+1 constantly, these stylish input data are as follows:

$R(n+1)={\left[\begin{array}{cccc}{\stackrel{\&OverBar;}{u}}_{\mathrm{1,1}}(n+1)& {\stackrel{\&OverBar;}{u}}_{\mathrm{1,2}}(n+1)& ...& {\stackrel{\&OverBar;}{u}}_{1,N}(n+1)\\ 0& {\stackrel{\&OverBar;}{u}}_{\mathrm{2,1}}(n+1)& ...& {\stackrel{\&OverBar;}{u}}_{2,N}(n+1)\\ 0& 0& ...& \\ & & .& .\\ 0& 0& .& .\\ & & .& .\\ 0& ...& {\stackrel{\&OverBar;}{u}}_{N-1,N-1}(n+1)& {\stackrel{\&OverBar;}{u}}_{N-1,N}(n+1)\\ 0& ...& 0& {\stackrel{\&OverBar;}{u}}_{N,N}(n+1)\end{array}\right]}_{N\&times;N}$

U wherein ^T(n+1) and the mathematic(al) representation of Γ (n+1) as follows:

u ^T(n+1)＝u _0，0(n+1)，u _0，1(n+1)，…u _0，N(n+1)＝y(n+2)，y(n+2) ³，…y(n+2-M) ^QΓ(n+1)＝z ₁(n+1)，z ₂(n+1)，…z _N-1(n+1)，z _N(n+1)

Begin (n+1) GIVENS rotation constantly again, method is constantly identical with (n), repeats no more herein.

From the above, be specially by GIVENS rotation carrying out QR decomposition:

$\tilde{Q}\left(n\right)\left[\begin{array}{cc}{u}^T\left(n\right)& z_0\left(n\right)\\ \sqrt{\mathrm{\&lambda;}}\&CenterDot;R\left(n\right)& \sqrt{\mathrm{\&lambda;}}\&CenterDot;\mathrm{\&Gamma;}\left(n\right)\end{array}\right]=\left[\begin{array}{cc}{0}^T& \tilde{e}\left(n\right)\\ R(n+1)& \mathrm{\&Gamma;}(n+1)\end{array}\right]$

$\widetilde{\mathrm{\&gamma;}}\left(n\right)=\underset{i=1}{\overset{N}{\mathrm{\&Pi;}}}\cos {\mathrm{\&theta;}}_i$

$e\left(n\right)=20\&CenterDot;{\log }_{10}(\tilde{e}\left(n\right)\&CenterDot;\widetilde{\mathrm{\&gamma;}}\left(n\right))$

And then, use the back just can finish asking for of pre-distortion parameters to interative computation by R (n) and Γ (n); Promptly

$w_N(n+1)=\frac{z_N(n+1)}{R_{\mathrm{NN}}(n+1)},$ N＝(M+1)·(Q+1)/2

$w_i(n+1)=\frac{1}{R_{i,i}(n+1)}(z_i(n+1)-\underset{j=i+1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_{i,j}(n+1)w_j(n+1)),$ for i＝N-1，…1

Wherein: 0＜＜λ＜1, preferred λ gets 0.95＜λ＜0.99999, in order to simplify operand, is provided with when specific implementation Thereby the adding of memory fact only needs simple shift operation, reduced multiplying.It should be noted that, because the pre-distortion parameters that obtains after the initial assembling is in initial convergence phase, pre-distortion coefficients also plays pendulum, abandon obtaining pre-distortion parameters, but the GIVENS rotation is being proceeded always, after carrying out L iteration renewal, think that the later matrix R (L) of GIVENS rotation enters stable state, pre-distortion parameters begins to preserve.

B, in the specific implementation, in order to make QR decompose more effectively with orderly, also can adopt systolic structures to realize: described mainly is the streamline sequence that algorithm is mapped as the basic calculating unit by the systolic structures realization, execute the task with parallel mode in these unit, make that all unit all are in active state in each clock cycle, therefore can use the pulsation form to finish equation R (n) to the finding the solution of R (n+1), the processing structure of pulsation as shown in Figure 3;

Wherein, processing procedure shown in Figure 3 is the situation of Systolic battle array by beat work: being located at n=1 is first beat constantly, then u during first count ₁₁Boundary element is received u ₁(1), finishes u ₁₁New value calculate, preserve, and with the c1 that calculates, s1 sends; During second beat, u ₁₂The unit is according to c1, s1 and the u that receives ₂(1) calculates new value

With With c1, s1 continues to send to the right, handles

Deliver to u downwards ₂₂When the 3rd beat, u ₁₃The unit receives u ₁(3), c1, s1 finish similar u ₁₂Operation, u simultaneously ₂₂The unit basis Calculate c2, s2 also transmits to the right, goes on by beat like this, and data flow will be by left oblique upper incident Systolic battle array, and press beat and advance to each cell node, after the outflow of product unit; Array element u as can be seen samples for the first time ^r(n) input will just all be absorbed by the Systolic battle array after N claps, and the output of error signal e (n) need just can obtain through the 2N+1 bat.This Systolic battle array realizes that the QR decomposition is highly susceptible to FPGA and realizes, the algorithm that the sampled signal that PA of every input returns is just finished 1～Q beat flows, after M+1 input signal flows in the pulsation battle array, just finished the renewal of pulsation battle array, next only needing to finish interative computation gets final product, wherein fall to such an extent that computing can be adopted the similar method of interative computation with above-mentioned GIVENS rotation, do not repeat them here.

In addition, the method for present embodiment also can comprise step 204: after periodic filter begins each time, preserve the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine;

Wherein, whether prior art comes decision system to restrain according to the size of error signal e (n), but because adaptive algorithm is after finding the optimum filter coefficient, adaptive process should stop, and because the randomness of system data, the restriction of filter coefficient length and precision, the fluctuation that filter coefficient does not stop about its optimal values in a kind of mode at random, sef-adapting filter just can reach the steady operation mode behind certain hour as a result, its performance also will stop to continue to improve, simultaneously adaptive-filtering reach error signal e (n) after the stable state also along with the randomness of input data in the fluctuation that does not stop, thereby this mode is easy to cause bigger error.And present embodiment proposes after periodic filter begins each time, carrying out adaptive-filtering handles, but do not preserve filter parameter, by the time finish after the adaptive-filtering processing of N sampling symbol, just begin to preserve filter parameter, and filter coefficient after this will preserve also, if had filter coefficient in identical entry address then upgrade replacement, no longer carry out the calculating and the comparison of error signal e this moment, thereby can the saving system move expense; It is total, sampling symbol numbers N can be according to the actual requirements or system emulation determine that this paper repeats no more.

In addition, this method also can comprise step 205: whether the error signal of periodically judging described definite pre-distortion parameters exceeds default concussion scope in the default time period, if then substitute fixed pre-distortion parameters with new default true parameter; May receive unusual input signal or misoperation at the base station, thereby cause dispersing or the problem of collapse of algorithm, present embodiment proposes the periodic error signal of differentiating, for example after 1000 adaptive-filterings are handled, whether the error signal e of differentiating for the 1001st this cycle vibrates within the specific limits, if error signal e in a period of time continues to exceed certain limit, then empty the pre-distortion parameters among the LUT, reinitialize the pre-distortion parameters among the LUT; And error signal e continues in a period of time within the specific limits by the time, then continues buffer memory and upgrades the sef-adapting filter parameter, thereby can reduce the operand of error signal.

As can be seen, adopt the method for the embodiment of the invention, after periodic filter is handled beginning, utilize the data after the pre-distortion to make up the initialize signal matrix as input signal as reference signal, power amplifier feedback data, again described initialize signal matrix is carried out the QR resolution process, and by the back determine pre-distortion parameters to interative computation, thereby effectively reduced operand and computational complexity; And just preserve the pre-distortion parameters that computing is determined since N sampling point symbol, thereby well saved system's operation expense.

Based on above-mentioned thought, the embodiment of the invention 2 has proposed a kind of processing unit of pre-distortion parameters again, and as shown in Figure 4, this device 400 comprises: construction unit 401, resolving cell 402 and arithmetic element 403; Wherein,

Described construction unit 401 is used for after periodic filter is handled beginning, is reference signal, is that input signal makes up the initialize signal matrix with the power amplifier feedback data with the data after the pre-distortion; Described resolving cell 402 is used for the initialize signal matrix that construction unit 401 makes up is carried out the QR resolution process; Described arithmetic element 403 is used for carrying out the back by the matrix after QR is decomposed and determines pre-distortion parameters to interative computation.

Wherein, described resolving cell 402 comprises: first processing module and second processing module; Wherein, described first processing module is used for adopting quadrature diagonalization mode that each the new data vector that is increased to described initialize signal matrix is carried out iteration diagonalization processing; Described second processing module is used to adopt systolic structures that described initialize signal matrix is carried out the QR resolution process.

Preferably, this device also comprises: memory cell; Wherein, described memory cell is used for after periodic filter begins each time, preserves the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine.

In addition, this device also comprises: judging unit; Described judging unit is used for periodically judging whether the error signal of the pre-distortion parameters that described arithmetic element is determined exceeds default concussion scope in the default time period, if then substitute fixed pre-distortion parameters with new default true parameter.

It will be understood by those skilled in the art that and to use many different technologies and in the technology any one to come expression information, message and signal.For example, the message of mentioning in the above-mentioned explanation, information can be expressed as voltage, electric current, electromagnetic wave, magnetic field or magnetic particle, light field or above combination in any.

The professional can also further should be able to recognize, the unit and the algorithm steps of each example of describing in conjunction with embodiment disclosed herein, can realize with electronic hardware, computer software or the combination of the two, for the interchangeability of hardware and software clearly is described, the composition and the step of each example described prevailingly according to function in the above description.These functions still are that software mode is carried out with hardware actually, depend on the application-specific and the design constraint of technical scheme.The professional and technical personnel can use distinct methods to realize described function to each specific should being used for, but this realization should not thought and exceeds scope of the present invention.

The method of describing in conjunction with embodiment disclosed herein or the step of algorithm can directly use the software module of hardware, processor execution, and perhaps the combination of the two is implemented.Software module can place the storage medium of any other form known in random asccess memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or the technical field.

To the above-mentioned explanation of the disclosed embodiments, make this area professional and technical personnel can realize or use the present invention.Multiple modification to these embodiment will be conspicuous concerning those skilled in the art, and defined herein General Principle can realize under the situation that does not break away from the spirit or scope of the present invention in other embodiments.Therefore, the present invention will can not be restricted to these embodiment shown in this article, but will meet and principle disclosed herein and features of novelty the wideest corresponding to scope.

The above only is preferred embodiment of the present invention, and is in order to restriction the present invention, within the spirit and principles in the present invention not all, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1. the processing method of a pre-distortion parameters is characterized in that, comprising:

After periodic filter is handled beginning, utilize data and power amplifier feedback data after the pre-distortion to make up the initialize signal matrix; Data after the described pre-distortion are reference signal, and described power amplifier feedback data is an input signal;

Initialize signal matrix to described structure carries out the QR resolution process;

Carry out the back by the matrix after QR is decomposed and determine pre-distortion parameters to interative computation.

2. method according to claim 1 is characterized in that, described initialize signal matrix to described structure carries out the QR resolution process and is specially:

Adopt quadrature diagonalization mode that each the new data vector that is increased in the described initialize signal matrix is carried out iteration diagonalization processing.

3. method according to claim 2 is characterized in that:

The vector of each row is the data of same sampling instant input in the described initialize signal matrix.

4. method according to claim 2 is characterized in that, this method also comprises:

After periodic filter begins each time, preserve the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine.

5. method according to claim 4 is characterized in that, this method also comprises:

Whether the error signal of periodically judging described definite pre-distortion parameters exceeds default concussion scope in the default time period, if, then with the alternative fixed pre-distortion parameters of new default true parameter.

6. method according to claim 1 is characterized in that:

Adopt systolic structures that described initialize signal matrix is carried out the QR resolution process.

7. the processing unit of a pre-distortion parameters is characterized in that, comprising: construction unit, resolving cell and arithmetic element; Wherein,

Described construction unit is used for after periodic filter is handled beginning, is reference signal, is that input signal makes up the initialize signal matrix with the power amplifier feedback data with the data after the pre-distortion;

Described resolving cell is used for the initialize signal matrix that construction unit makes up is carried out the QR resolution process;

Described arithmetic element is used for carrying out the back by the matrix after QR is decomposed and determines pre-distortion parameters to interative computation.

8. device according to claim 7 is characterized in that, described resolving cell comprises: first processing module and second processing module; Wherein,

Described first processing module is used for adopting quadrature diagonalization mode that each the new data vector that is increased to described initialize signal matrix is carried out iteration diagonalization processing;

Described second processing module is used to adopt systolic structures that described initialize signal matrix is carried out the QR resolution process.

9. device according to claim 8 is characterized in that this device also comprises: memory cell; Wherein, described memory cell is used for after periodic filter begins each time, preserves the pre-distortion parameters that computing is determined since N sampling point symbol, described N can be according to the actual requirements or system emulation determine.

10. device according to claim 9 is characterized in that this device also comprises: judging unit; Described judging unit is used for periodically judging whether the error signal of the pre-distortion parameters that described arithmetic element is determined exceeds default concussion scope in the default time period, if then substitute fixed pre-distortion parameters with new default true parameter.

Images