CN101577525A - Device and method for estimating model of amplifier - Google Patents

Abstract

The invention relates to a device and a method for estimating a model of an amplifier. The device estimates the model of the amplifier according to the input signal and the output feedback signal of an amplifier and comprises a first data selector, a timer, a corresponding sampling point acquiring unit, a time delayer, a second data selector and a model estimating unit; the first data selector is used for selecting and storing the sampling point in the input signal in accordance with preset threshold and acquiring the time point of the sampling point; the timer is used for determining the time delay of the output feedback signal relative to the input signal; the corresponding sampling point acquiring unit is used for acquiring the sampling point of the input signal aligned to the output feedback signal in time in accordance with the time delay determined by the timer; the time delayer is used for delaying the time point of the sampling point of the input signal in accordance with the time delay determined by the timer; the second data selector is used for acquiring the corresponding feedback sampling point of the output feedback signal in a corresponding time point in accordance with the time point of the delayed sampling point; and the model estimation unit is used for estimating the model of the amplifier in accordance with the sampling point and the corresponding feedback sampling point.

Description

Amplifier model estimation unit and method

Technical field

The present invention relates to a kind of amplifier model estimation unit and method.

Background technology

In wireless communication system, sending signal will amplify analog radio-frequency signal through high power amplifier, goes out by antenna transmission then.For improving spectrum efficiency, increasing non-permanent envelope modulation mode (such as 64QAM/16QAM) and CDMA/OFDM combination are widely used in the modern wireless communication systems.The denominator of these signals before entering amplifier all be peak-to-average force ratio (PAPR) than higher, such as the PAPR theoretical value of the ofdm signal that adopts 1024 FFT up to 10dB.The signal of these high PAPR can cause the severe nonlinear distortion by power amplifier, causes signal constellation point damage and out-of-band radiation power to increase.For reducing nonlinear distortion, can adopt the way of rollback transmitted power, but can cause the efficient of amplifier to reduce like this.For this reason, for many years, people have done deep research in this field, and wherein digital pre-distortion technology has obtained people's extensive concern.

Fig. 1 shows the structure chart of conventional power amplification device.Power amplification device shown in Figure 1 has adopted disclosed traditional feedback system digital pre-distortion technology in the document 1.

As shown in Figure 1, the digital signal that send predistorter (PD, PreDistion) 1 carry out predistortion after, become analog signal through digital to analog converter 2.By modulator 4 up-conversions, enter amplifier (PA) 5 then and amplify behind these analog signal process low pass filter 3 filtering out of band signals, the signal after the amplification becomes two parts by power splitter 6.A part is gone out by antenna transmission, and another part is as the feedback signal of predistortion.Feedback signal becomes digital signal through demodulator 7 down-conversions by A-D converter 8.Amplifier model estimation unit 10 utilizes PD output signal and feedback signal to carry out the simulated estimation of amplifier.Owing to require PD output signal and feedback signal aligns in time, thereby, give respective signal point acquiring unit 11 time delay information then by the time delay of timer 12 detection feedback signals.The signal of 11 pairs of inputs of respective signal point acquiring unit is stored, and when receiving the indication of self-timer 12, obtains the signal of the input of aliging in time with feedback signal.Amplifier model estimates to adopt adaptive algorithm to try to achieve the coefficient that model is estimated, the criterion of disclosed coefficient estimation is to make the signal behind the PD and the mean square error minimum of the signal that feeds back in the document 1.Specific algorithm is gradient method at random.After having obtained model and estimating, and then try to achieve inverse model by anti-PA model solution unit 9, the gain after making signal through predistortion and amplifier is 1 or constant, the i.e. linearisation of amplifier.

Device shown in Figure 1 has following shortcoming:

(1) whole signaling points of use transmission and received signal before convergence, amount of calculation is big;

(2) when conditional number is big, convergence rate may be very slow;

(3) time delay of feedback loop is bigger usually, so respective signal acquiring unit 11 circuit among the figure need bigger memory space.

Summary of the invention

The above-mentioned shortcoming that the present invention is directed to prior art is made, and a kind of amplifier model estimation unit and method are provided, and is used to overcome one or more shortcoming of prior art.

To achieve these goals, the application provides following invention.

Invention 1, a kind of amplifier model estimation unit, carry out the estimation of amplifier model according to the signal of input amplifier and the output feedback signal of described amplifier, it is characterized in that described amplifier model estimation unit comprises first data selector, second data selector, delayer, corresponding sampled point acquiring unit, timer and model estimation unit:

Described first data selector is used for selecting and storing according to predetermined threshold the sampled point of described input signal, and obtains the time point of described sampled point;

Described timer is determined the time-delay of described output feedback signal with respect to described input signal;

Corresponding sampled point acquiring unit according to the time-delay that described timer is determined, is obtained the sampled point of the described input signal that aligns in time with described output feedback signal;

The time-delay that described delayer is determined according to described timer is delayed time to the time point of the described sampled point of described input signal;

Described second data selector obtains the corresponding feedback sample point of described output feedback signal at corresponding time point according to the time point of the described sampled point through delaying time of described delayer output;

Described model estimation unit is determined the model of described amplifier according to described sampled point and described corresponding feedback sample point.

The invention 2, according to the invention 1 described amplifier model estimation unit, it is characterized in that the number of described sampled point is between 1 to 128.

The invention 3, according to the invention 2 described amplifier model estimation units, it is characterized in that the number of described sampled point is between 4 to 64.

The invention 4, according to the invention 1 described amplifier model estimation unit, it is characterized in that, the model of described amplifier is the memory multinomial model, described model estimation unit utilization is with the formula of the higher harmonics exponent number time gained of the real part of described memory multinomial model pair amplifier input signal or the described memory multinomial model of imaginary part differentiate, use the value of sampled point of described input signal and the value of described output feedback sampling point, determine the parameter of described memory multinomial model, thereby estimate described memory multinomial model.

Invention 5, according to invention 4 described amplifier model estimation units, it is characterized in that described model estimation unit is at first determined the coefficient of high-order harmonic wave when determining described polynomial module shape parameter.

Invention 6, according to invention 4 described amplifier model estimation units, it is characterized in that the higher harmonics exponent number of described multinomial model is between 3-9.

The invention 7, according to the invention 4 described amplifier model estimation units, it is characterized in that, described amplifier model estimation unit also comprises noise suppression unit, and described noise suppression unit is used for suppressing the noise of the selected output feedback sample point of described second data selector.

Invention 8, according to invention 1 described amplifier model estimation unit, it is characterized in that described predetermined threshold is definite with sampling number purpose ratio according to the difference of the maximum amplitude of the real part of described input signal or imaginary part and minimum amplitude.

Invention 9, according to invention 4 described amplifier model estimation units, it is characterized in that, described first data selector select imaginary part be 0 and the input signal of difference in preset range of real part and described predetermined threshold as described input sample point.

Invention 10, according to invention 4 described amplifier model estimation units, it is characterized in that, described first data selector select real part be 0 and the input signal of difference in preset range of imaginary part and described predetermined threshold as the sampled point of described input signal.

Invention 11, a kind of amplifier model method of estimation, this method is carried out the estimation of amplifier model according to the output feedback signal of signal in the input amplifier and described amplifier, it is characterized in that, and described amplifier model method of estimation may further comprise the steps:

Select and store sampled point in the described input signal according to predetermined threshold, and obtain the time point of described sampled point;

Determine of the time-delay of described output feedback signal with respect to described input signal;

According to described time-delay, obtain the sampled point of the described input signal that aligns in time with described output feedback signal, and the time point of obtained sampled point is delayed time;

The time point through the sampled point of time-delay according to the output of described delayer obtains the corresponding feedback sample point of described output feedback signal at corresponding time point; And

Model according to described sampled point and the definite described amplifier of described corresponding feedback sample point.

Invention 12, a kind of computer program, this computer program is by computer or logical block execution the time or when making an explanation or compile the back execution, can make described computer or logical block carry out the estimation of amplifier model according to the output feedback signal of signal in the input amplifier and described amplifier, be characterised in that, make computer or logical block carry out following steps:

Select and store sampled point in the described input signal according to predetermined threshold, and obtain the time point of described sampled point;

Determine of the time-delay of described output feedback signal with respect to described input signal;

According to described time-delay, obtain the sampled point of the described input signal that aligns in time with described output feedback signal, and the time point of obtained sampled point is delayed time;

According to the time point of sampling point, obtain the correspondence feedback sampling point of described output feedback signal at corresponding time point through delaying time; And

According to described sampling point and the described corresponding model that sampling point is determined described amplifier that feeds back.

Invention 13, a kind of computer-readable recording medium, storage aforementioned calculation machine readable program.

The present invention is directed to present circuit complexity problem of higher, only utilize some special sampling points, in model is estimated, adopt the method for asking local derviation directly to try to achieve polynomial coefficient in the model estimation.Finishing model by local derviation calculating estimates.

Therefore following advantage is arranged:

1, directly tries to achieve the multinomial coefficient of model in estimating, fast convergence rate.

2, only utilize some special sampling points, then amount of calculation reduces greatly; And the respective signal acquisition cuicuit only needs storage seldom.

Emulation shows that the method according to this invention and device can accurately be tried to achieve amplifier model.

Description of drawings

When the explanation below reading in conjunction with the accompanying drawings, above and other purpose of the present invention, feature and advantage will become clearer.In the accompanying drawings,

Fig. 1 shows the structure chart of conventional linear power amplification device;

Fig. 2 shows the structured flowchart according to the amplifier model estimation unit of one embodiment of the present invention;

Fig. 3 shows the structured flowchart according to the amplifier model estimation unit of another execution mode of the present invention;

Fig. 4 is the sequential chart of selecting according to the data of one embodiment of the present invention;

Fig. 5 obtains circuit with the respective signal of prior art and shows the minimizing situation that corresponding sampled point of the present invention is obtained the unit required storage with contrasting;

Fig. 6 is the input range of amplifier and the characteristic curve of output amplitude;

Curve when Fig. 7 has represented that amplifier output signal asked the single order partial derivative to the input signal real part;

Fig. 8 has represented the triple-frequency harmonics coefficient and the NMSE between the ideal value (normalization mean square error) performance that estimate; And

Fig. 9 shows the flow chart according to the amplifier model method of estimation of one embodiment of the present invention.

Embodiment

Below in conjunction with accompanying drawing the specific embodiment of the present invention is described.

Fig. 2 shows the structured flowchart according to the linear power amplification device of one embodiment of the present invention.Among Fig. 2 with Fig. 1 in identical parts illustrate with identical label, and omitted detailed description to them.Should be noted that, although parts 1-9 etc. has been shown among Fig. 2, should be clear, linear power amplification device of the present invention is therefore not limited, and it can be applied to needing in other device that the pair amplifier model estimates of other.

As shown in Figure 2, complex signal x (n)=xr (the n)+jxq (n) of PD1 output (n represents the time sequence number of sampling point, and n i.e. moment n*T, and wherein T is the sampling point gap length) only selects some special sampling point x by first data selector 14 _s(n) use for model estimation unit 15, same feedback signal y (n) only selects some special sampling point y by second data selector 13 _s(n) select to use to data.

Introduce the operation of data selector 14 below.

Data selector 14 is compared real part xr (n) or the imaginary part xq (n) of received signal x (n)=xr (n)+jxq (n) with predetermined threshold value, when when the absolute value of the difference of predetermined threshold value is in predetermined scope, then this sampling point is considered as special sampling point, takes a sample.

Should note, the present invention will select some special sampling points in input signal x (n)=xr (n)+jxq (n), thereby, although input signal x (n)=xr (n)+jxq (n) itself is exactly a sampling point, in this article, unless have clear and definite indication to obtain explanation in the context, generally it be called signal or signaling point, and selected special sampling point is called sampling point or sampled point.

One time model estimates that the number of selected special sampling point is scheduled to, and for example can determine according to emulation.The number that should be scheduled to for example can be the arbitrary positive integer among the 1-128.And preferably between 4-64.And more preferably between 8-32.And generally can be taken as 2 index doubly, for example get 8,16 and 32.For the convenience that illustrates, be that 32 situation describes with this predetermined number below.And situation about comparing with the amplitude of getting real part describes.

In concrete execution mode, can obtain the threshold value of each point in advance by following mode.

At first, get the maximum in certain timing statistics scope (being max (| xr (n) |)) of the value of real part xr (n) and real part xr (n) the minimum value of the value of timing statistics scope necessarily (being min (| xr (n) |)).This maximum and minimum value can obtain in advance, for example add up maximum and the minimum value of signal xr (n) in the certain hour scope.These two values are common by the digital modulation formula decision that sends signal, and irrelevant with other factors.

Then, the difference of maximum and minimum value divided by predetermined sampling number num, is obtained the threshold value reference point $\mathrm{ref}=\frac{\max \left(\right|x_r\left(n\right)\left|\right)-\min \left(\right|x_r\left(n\right)\left|\right)}{\mathrm{num}}.$ Notice that as mentioned above, the num here can get positive integers such as 32.

Ask for the threshold value of each sampled point at last.For example can obtain sampled point k (the threshold value Threshold of 0＜k＜num+1) locate according to following formula _k

Threshold _k＝min(|xr(n)|)+k×ref。

Can obtain uniform sampled point like this.As required, can increase coefficient factor to each sampled point _kFactor _kBe illustrated in the coefficient of k sampled point.

Be Threshold _k=min (| xr (n) |)+k * ref * factor _k

Like this, by adjusting coefficient factor _k, can obtain close or thin sampled point.

It should be noted, on regard to threshold value calculating can in first data selector 14, carry out, but more preferably, can calculate in advance and be stored in the data selector 14 after finishing, or in data selector 14 storage device that can read.

Like this, the real part xr (n) of the signal x (n) of input is compared with corresponding threshold value, if their difference is then classified this point as sampled point in predetermined scope.Particularly, real part and first threshold values can be compared, if satisfy condition (condition for example is | xr (n)-Threshold ₀|＜ε, ε are predetermined number) then it is taken as sampled point, preserve this sampled point x (n) (containing real part and imaginary part); That determines this sampled point is numbered 0; Produce this sampled point fixed timing mark (the fixed timing mark level is for high constantly such as this, and other moment level are low) in time simultaneously.If do not satisfy condition, then continue with other threshold values relatively, when satisfy condition with k threshold values (condition for example is | xr (n)-Threshold _k|＜ε, ε are predetermined number), preserve this sampled point (containing real part and imaginary part) similarly; Determine that this sampled point is numbered k-1, produce this sampled point fixed timing mark in time simultaneously.If more do not satisfy condition with all threshold values, then abandon this signal x (n), promptly fixed timing mark is a low level, the operation of not preserving sampled point.

Next input signal x (n+1) is done above-mentioned similar operations, but should be noted that the pairing threshold values of sampled point that need not this moment and compiled number compares.

When the pairing sampled point of all threshold values all being had corresponding the preservation, then a secondary data selection course is finished.This moment, each threshold values was all to there being a sampled point to preserve.

Timer 12 is determined from the output feedback signal of analog to digital converter 8 outputs and the time delay between the input signal, and is given corresponding sampled point acquiring unit 19 and delayer 18 with the time delay information of determining.The signal of 19 pairs of inputs of corresponding sampled point acquiring unit is stored, and when (comprising time delay information), obtains the sampling point of the input signal that aligns in time with output feedback signal in the indication that receives self-timer 12.Delayer 18 is according to the indication that comes self-timer 12 that receives, and the time point of the sampled point of described corresponding sampled point acquiring unit 19 input signal that aligns in time with output feedback signal that obtained is delayed time.Second data selector, 13 bases are sampled from the time point of the sampled point through delaying time of delayer, obtain the output feedback signal sampled point corresponding with input sample point.Input sample point that amplifier model estimation unit 15 is obtained according to corresponding sampled point acquiring unit 19 and described second data selector obtain to export sampled point to carry out amplifier model and estimates.

Introduce amplifier model estimation unit 15 below.

At first, because the existence of first data selector 14 and second data selector 13 only utilizes some special sampling points to carry out model and estimates, amount of calculation is greatly reduced.Therefore, according to one embodiment of the present invention, amplifier model estimation unit 15 can adopt any model estimation unit known in the field.

Secondly, preferably, can adopt employing memory multinomial model estimation unit well known in the art.Can be about the memory multinomial model referring to document 3.

Introduce the preferred amplifier model estimation unit 15 of the present invention below.

According to preferred implementation of the present invention, the model of amplifier can adopt the memory polynomial repressentation as follows:

$y\left(n\right)=\underset{p=1}{\overset{P}{\text{\Σ}}}\underset{l=0}{\overset{L-1}{\mathrm{\Σ}}}{a}_{p,l}x{(n-l)\left|x(n-l)\right|}^{p-1}$ (formula 1)

Wherein x (n) is the PA input signal, and y (n) is the output of analog to digital converter 8.P is higher harmonics exponent number (generally only considering odd-order), and L is the maximal memory degree of depth.Because formula 1 is the power function form about integer p, if therefore considering to adopt asks the method for local derviation to carry out depression of order, that must simply try to achieve a after y (n) is tried to achieve P rank local derviation _{P, l}, and then backward is tried to achieve a _{P-2, l}... until a _{1, l}In order to illustrate, be described by a simple model earlier, promptly suppose amplifier model P=3, L=0.In fact generally the P with amplifier model is arranged between 3 to 8 just can satisfy the demand (seeing document 3).

Y (n)=k ₁X (n)+k ₃| x (n) | ²X (n) (formula 2)

For complex signal x (n)=xr (the n)+jxq (n) of input, wherein xr (n) and xq (n) are respectively the real part and the imaginary part of input signal.For simplifying, saved the n in the bracket, i.e. x=x (n).Local derviation is asked to xr continuously in formula 2 both sides, then can obtain:

$\frac{\&PartialD;y}{{\&PartialD;x}_r}={\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_1+k_3({3\mathrm{<figure-callout\; id="2"\; label="x\; r"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">x</figure-callout>}}_r^{}+x_{\mathrm{<figure-callout\; id="2"\; label="q"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">q</figure-callout>}}^2+2{\mathrm{jx}}_r{x}_q)$

$\frac{{\&PartialD;}^{2}y}{\&PartialD;{x^2}_r}=k_3({6x}_r+2j{x}_q)$ (formula 3)

$\frac{{\&PartialD;}^{3}y}{{{\&PartialD;x}^3}_r}={6\mathrm{<figure-callout\; id="3"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_3$

Under the situation of known x (n) and y (n), can be by finding the solution by equation

And equation

The linear equation in two unknowns group that constitutes is tried to achieve coefficient k ₁And k ₃:

${\mathrm{<figure-callout\; id="3"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_3=\left(\frac{{\&PartialD;}^{3}y}{{{\&PartialD;x}^3}_r}\right)/6$

(formula 4)

${\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_1=\frac{\&PartialD;y}{{\&PartialD;x}_r}-k_3({3\mathrm{<figure-callout\; id="2"\; label="x\; r"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">x</figure-callout>}}_r^{}+x_{\mathrm{<figure-callout\; id="2"\; label="q"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">q</figure-callout>}}^2+2{\mathrm{jx}}_r{x}_q)$

If the k value is enough little, i.e. the interval of the neighbouring sample point that differentiate is used is enough little, the derivative in (formula 3) $\frac{\&PartialD;y}{{\&PartialD;x}_r}\&ap;\frac{y(n+k)-y\left(n\right)}{x_r(n+k)-x_r\left(n\right)};$

Also can try to achieve by similar method.

It is further noted that if according to the algorithm of backward, promptly earlier from high-order harmonic wave coefficient k ₃Find the solution beginning, then can avoid common Gaussian elimination process, calculate simple relatively for N unit linear function group.

If same asks local derviation to imaginary part:

$\frac{\&PartialD;y}{{\&PartialD;x}_q}=k_{1}j+k_3({3\mathrm{<figure-callout\; id="2"\; label="jx\; q"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">jx</figure-callout>}}_q^{}+{\mathrm{<figure-callout\; id="2"\; label="jx\; r"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">jx</figure-callout>}}_r^{}+{2x}_r{x}_q)$

$\frac{{\&PartialD;}^{2}y}{{{\&PartialD;x}^2}_q}=k_3({6\mathrm{jx}}_q+{2x}_r)$ (formula 5)

$\frac{{\&PartialD;}^{3}y}{{{\&PartialD;x}^3}_q}=6\mathrm{<figure-callout\; id="3"\; label="j\; k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">j</figure-callout>}{k}_{}{}_3$

${\mathrm{<figure-callout\; id="3"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_3=\left(\frac{{\&PartialD;}^{3}y}{{{\&PartialD;x}^3}_q}\right)/\left(6j\right)$

The similar result that obtains: ${\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_1=-j\&CenterDot;[\frac{\&PartialD;y}{{\&PartialD;x}_q}-k_3({3\mathrm{<figure-callout\; id="2"\; label="jx\; q"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">jx</figure-callout>}}_q^{}+{\mathrm{<figure-callout\; id="2"\; label="jx\; r"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">jx</figure-callout>}}_r^{}+2x_r{x}_q)]$ (formula 6)

For simplicity, the method that imaginary part is found the solution repeats no more, below all can simply be generalized at the method for real part being asked local derviation imaginary part is asked on the method for local derviation.

Should be noted that because derivative calculations is actually and calculate the rate of change of described feedback signal to described input signal real part or imaginary part, therefore required sampling number must be greater than differentiating number of times.

In addition, if certain condition is set, amount of calculation can further reduce.For example: only imaginary part is about 0 data in the selection transmission data, and the difference of real part and pre-set threshold value is in predetermined scope.

Formula (formula 3) above that is petty can further be reduced to:

${\mathrm{<figure-callout\; id="3"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_3=\left(\frac{{\&PartialD;}^{3}y}{{{\&PartialD;x}^3}_r}\right)/6$

(formula 7)

${\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="A20081010022100171.png,A20081010022100181.png"\; state="\{\{state\}\}">k</figure-callout>}}_1=\frac{\&PartialD;y}{{\&PartialD;x}_r}-k_3\left({3x}_r^2\right)$

Contrast equation 5 and 7 can be seen by such data and select amount of calculation further to reduce.It is to be noted that imaginary part is 0 is not the criterion of a strictness, all requiring the direct current biasing (DC component) of signal for most communication systems is 0, and to be about 0 probability of happening still very high for imaginary part like this.Such as for the WiMAX signal, xr and xq are that average is 0 Gaussian random variable, and it is that probability of happening is the highest that xq equals that 0 situation compares with other values.

It is 0 that the criterion here is not limited to the imaginary part that we mention, and also comprises the system of selection that other can formula of reduction 3.

For more general model, provided the general expression formula of finding the solution below:

Because signal x (the n)=x in the formula 1 _r(n)+jx _q(n), thus formula 1 can be written as:

$y\left(n\right)=\underset{p=1}{\overset{P}{\mathrm{\&Sigma;}}}\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{a}_{p,l}{[x_r(n-l)+{\mathrm{jx}}_q(n-l)][{x_r}^2(n-l)+{{\mathrm{jx}}^2}_q(n-l)]}^{\frac{p-1}{2}}$

(formula 8)

$=\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{p=1}{\overset{P}{\mathrm{\&Sigma;}}}\underset{k^{\&prime;}=0}{\overset{\frac{p-1}{2}}{\mathrm{\&Sigma;}}}{a}_{p,l}{c}_{\frac{p-1}{2}}^{k^{\&prime;}}{j}^{\frac{p-1}{2}-k^{\&prime;}}\left[{x_r}^{{2k}^{\&prime;}}(n-l)\right]\left[{x_q}^{p-1-{2k}^{\&prime;}}(n-l)\right][x_r(n-l)+{\mathrm{jx}}_q(n-l)]$

Can find that by following formula y (n) forms for the integer rank power series of a series of real parts and imaginary part, its high reps is P.If real part is asked local derviation, then imaginary part can be regarded as and the irrelevant constant of real part.Order $\mathrm{\&beta;}(k^{\&prime;},p,n,l)=c_{\frac{p-l}{2}}^{k^{\&prime;}}{j}^{\frac{p-1}{2}-k^{\&prime;}}\left[{x_q}^{p-1-2k^{\&prime;}}(n-l)\right],$ We know ought $k^{\&prime;}=\frac{p-1}{2}$ The time, β (k ', p, n, l)=1.

Formula can be written as:

$y\left(n\right)=\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{p=1}{\overset{P}{\mathrm{\&Sigma;}}}\underset{k^{\&prime;}=0}{\overset{\frac{p-1}{2}}{\mathrm{\&Sigma;}}}{a}_{p,l}\mathrm{\&beta;}(k^{\&prime;},p,n,l)\left[{x_r}^{{2k}^{\&prime;}}(n-l)\right][x_r(n-l)+{\mathrm{jx}}_q(n-l)]---\left(9\right)$

For example to l=l ₀, its coefficient sets

P=1,3 ... finding the solution of P can the branch following steps be carried out:

$\frac{\&PartialD;y\left(n\right)}{{\&PartialD;x}_r(n-l_0)}=a_{1,{\mathrm{<figure-callout\; id="0"\; label="l"\; filenames="A20081010022100231.png"\; state="\{\{state\}\}">l</figure-callout>}}_0}+\underset{p=3}{\overset{P}{\mathrm{\&Sigma;}}}\underset{k^{\&prime;}=1}{\overset{\frac{p-1}{2}}{\mathrm{\&Sigma;}}}{a}_{p,l}\mathrm{\&beta;}(k^{\&prime;},p,n,l_0)\left\{({2k}^{\&prime;}-1)x_r^{2k^{\&prime;}-1}(n-l_0)\right[x_r(n-l_0)+{\mathrm{jx}}_q(n-l_0)]+x_r^{2k^{\&prime;}}(n-l_0)\}---\left(10\right)$

$\frac{{\&PartialD;}^{3}y\left(n\right)}{{{\&PartialD;x}_r}^3(n-l_0)}=a_{3,{\mathrm{<figure-callout\; id="0"\; label="l"\; filenames="A20081010022100231.png"\; state="\{\{state\}\}">l</figure-callout>}}_0}+\underset{p=5}{\overset{P}{\mathrm{\&Sigma;}}}\underset{k^{\&prime;}=2}{\overset{\frac{p-1}{2}}{\mathrm{\&Sigma;}}}{a}_{p,l}\mathrm{\&beta;}(k^{\&prime;},p,n,l_0)\left\{({2k}^{\&prime;}-1)({2k}^{\&prime;}-2)({2k}^{\&prime;}-3)x_r^{{2k}^{\&prime;}-3}(n-l_0)\right[x_r(n-l_0)$

$+{\mathrm{jx}}_q(n-l_0)]+\left({4k}^{\&prime;2}\right)x_r^{{2k}^{\&prime;}-2}(n-l_0)\}$ (formula 11)

...

$\frac{{\&PartialD;}^{P}y\left(n\right)}{{{\&PartialD;x}_r}^P(n-l_0)}=P!a_{P,l}$ (formula 12)

If we are known input signal series x (n), x (n-1), x (n-l ₀) and output, formula (10)～formula (12) has formed (P-1)/2 yuan linear function group so, is enough to find the solution (P-1)/2 unknowm coefficient

P=1 wherein, 3 ... P.For other a _{P, l}, l=0,1...L-1 equally can in the hope of, so also just finished the estimation of PA model.If its equation solution method adopts backward to find the solution, promptly find the solution a earlier _{P, l}, a again _{P-2, l}... .a _{P-4, l}Up to a _{1, l}, the forward steps that can avoid equation group to find the solution then.

Disturb in order to tackle the thermal noise that adds in the feedback signal, can carry out further noise suppressed the data that feed back to.Fig. 3 shows the structured flowchart according to the amplifier model estimation unit of another execution mode of the present invention.As shown in Figure 3, the output signal of selector 13 is sent in the noise suppression unit 16.After treating that noise is further suppressed, send into again in the model estimation unit 15 and go.Remainder is identical with device shown in Figure 2, thereby will not give unnecessary details.

A simple case of noise suppression circuit is exactly that difference reception data y (n)=x0+n (t) that identical transmission data x (n)=x0 is obtained carries out arithmetic mean, can suppress The noise.General communication system is WiMAX for example, all can send the training symbol of repetition among the IEEE802.11a every a fixing cycle, such as preamble or the like, can be used.

The noise suppression circuit that the present invention carried is not limited to above-mentioned arithmetic average method, also can comprise the circuit of other the inhibition noise of knowing, such as well-designed filter etc.

Fig. 4 is the sequential chart of selecting according to the data of one embodiment of the present invention, can see the relation according to input signal and threshold values of the present invention, and first data selector 14 has been stored xs (0), xs (1) ... xs (k) ..., produced fixed timing mark simultaneously.Second data selector 13 goes to select in the middle of amplifier output signal according to the fixed timing mark of having delayed time, and its criterion is then selected institute's amplifier output signal in the corresponding moment for as long as the fixed timing mark after the current delay is a high level.The present invention only need save required storage more greatly according to seldom special sampling point and conventional method that the fixed timing mark after the time-delay is left required calculating as can be seen.

Fig. 5 obtains circuit with the respective signal of prior art and shows the minimizing situation that corresponding sampled point of the present invention is obtained the unit required storage with contrasting.In common broadband system, because signal rate is higher, so usually corresponding thousands of the signaling points of feedback control loop time delay.In order to delay time, common amplifier model algorithm for estimating needs the internal memory that can store thousands of signaling points.And the present invention only chooses tens sampled points by the data selection algorithm, and required memory is reduced to tens sampled points like this.

At WiMAX, carried out simple algorithm emulation under the 20MHz bandwidth forward link, wherein amplifier model is that a memory depth is 0, higher harmonics number of times is 3 multinomial model: amplifier model and simulation result are shown in Fig. 6～8:

Fig. 6 is the input range of amplifier and the characteristic curve of output amplitude.Transverse axis is an input signal amplitude, and the longitudinal axis is an amplitude output signal, can see that the amplifier output signal amplitude is a non-linear relation for input signal amplitude.

Curve when Fig. 7 has represented that the pair amplifier output signal is asked with the single order partial derivative to the input signal real part.Transverse axis is the input signal real part, and the longitudinal axis is an output signal, and wherein curve is ideal value or theoretical value, and circle is the single order local derviation for trying to achieve according to the present invention then.Can see that both coincide finely.

Fig. 8 has represented the triple-frequency harmonics coefficient and the NMSE between the ideal value (normalization mean square error) that estimate.Transverse axis is for finding the solution number of times, and the longitudinal axis is NMSE.Since had repeatedly find the solution average, can see find the solution for 40 times average after, NMSE＜5%, this shows that performance is fine.

In sum: the present invention only utilizes some special sampling points to adopt the method for asking local derviation to try to achieve polynomial coefficient in the model estimation in model is estimated, fast convergence rate, amount of calculation and required memory circuit reduce greatly, are a kind of methods of Linear Power Amplifier of real low complex degree.

Fig. 9 shows the flow chart according to the amplifier model method of estimation of one embodiment of the present invention.As shown in the figure, at first in step 901, select and store sampled point in the described input signal, and obtain the time point of described sampled point according to predetermined threshold.Predetermined threshold can determine as previously mentioned, to the selection of sampled point in the input signal also as previously mentioned.Determine of the time-delay of described output feedback signal in step 902 then with respect to described input signal.This can be determined by timer 12.Then, in step 903, according to described time-delay, obtain the sampled point of the described input signal that aligns in time with described output feedback signal, and delay time by the time point of 18 pairs of obtained sampled points of delayer by corresponding sampled point acquiring unit 19.In step 904,, obtain the corresponding feedback sample point of described output feedback signal then at corresponding time point by the time point of second data selector 13 according to the sampled point through delaying time of described delayer output.At last, in step 905, determine the model of described amplifier according to described sampled point and described corresponding feedback sample point by amplifier model estimation unit 15.Obviously, also can increase the step of noise remove.

The present invention can be realized by simple hardware, also can be realized by software, also can be realized by the mixing of hardware and software.When being realized by software, this software can make above-mentioned method or the device of computer reality when being carried out by computer.The present invention also comprises the storage medium of storing this software.Storage medium for example can be CD, DVD, floppy disk, MO, flash memory, tape etc.

More than explanation only is exemplary, is not limitation of the present invention.Scope of the present invention is only determined by claim and equivalent thereof.

List of references:

[1]Hyun?Woo?Kang，Yong?Soo?Cho，Dae?Hee?Youn.“On?compensatingnonlinear?distortions?of?an?OFDM?system?using?an?efficient?adaptivepredistorter”，Communications，IEEE?Trans.1999，47(4)：522-526.

[2]CH.H.Cheng，E.J.Powers，′Optimal?Volterra?Kernel?EstimationAlgorithms?for?a?nonlinear?Communication?System?for?PSK?and?QAM?inputs′，IEEE?Transactions?on?Signal?processing，vol.49，No.1，January?2001，pp.147-163.

[3]Behzad?Razavi，”RF?Microelectronics”，Prentice?Hall.

Claims (10)

1, a kind of amplifier model estimation unit, carry out the estimation of amplifier model according to the signal of input amplifier and the output feedback signal of described amplifier, it is characterized in that described amplifier model estimation unit comprises first data selector (14), second data selector (13), delayer (18), corresponding sampled point acquiring unit (19), timer (12) and model estimation unit (15):

Described first data selector (14) is used for selecting and storing according to predetermined threshold the sampled point of described input signal, and obtains the time point of described sampled point;

Described timer (12) is determined the time-delay of described output feedback signal with respect to described input signal;

The time-delay that described corresponding sampled point acquiring unit (19) is determined according to described timer obtains the sampled point of the described input signal that aligns in time with described output feedback signal;

The time-delay that described delayer (18) is determined according to described timer is delayed time to the time point of the described sampled point of described input signal;

Described second data selector (13) obtains the corresponding feedback sample point of described output feedback signal at corresponding time point according to the time point of the described sampled point through delaying time of described delayer output;

Described model estimation unit (15) is determined the model of described amplifier according to described sampled point and described corresponding feedback sample point.

2, amplifier model estimation unit according to claim 1 is characterized in that, the number of described sampled point is between 4 to 64.

3, amplifier model estimation unit according to claim 1, it is characterized in that, the model of described amplifier is the memory multinomial model, described model estimation unit (15) is utilized the formula with the higher harmonics exponent number time gained of the real part of described memory multinomial model pair amplifier input signal or the described memory multinomial model of imaginary part differentiate, use the value of described input sample point and described output feedback sample point, determine the parameter of described memory multinomial model, thereby estimate described memory multinomial model.

4, amplifier model estimation unit according to claim 3 is characterized in that, described model estimation unit is at first determined the coefficient of high-order harmonic wave when determining described polynomial module shape parameter.

5, amplifier model estimation unit according to claim 3 is characterized in that, the higher harmonics exponent number of described multinomial model is between 3-9.

6, amplifier model estimation unit according to claim 3, it is characterized in that, described amplifier model estimation unit also comprises noise suppression unit (16), and described noise suppression unit (16) is used for suppressing the noise of the selected output feedback sample point of described second data selector.

7, amplifier model estimation unit according to claim 1 is characterized in that, described predetermined threshold is determined according to the maximum amplitude of the real part of described input signal or imaginary part and the difference and the sampling number purpose ratio of minimum amplitude.

8, amplifier model estimation unit according to claim 3 is characterized in that, described first data selector select imaginary part be 0 and the input signal of difference in preset range of real part and described predetermined threshold as the sampled point of described input signal.

9, amplifier model estimation unit according to claim 3 is characterized in that, described first data selector select real part be 0 and the input signal of difference in preset range of imaginary part and described predetermined threshold as the sampled point of described input signal.

10, a kind of amplifier model method of estimation, this method is carried out the estimation of amplifier model according to the output feedback signal of signal in the input amplifier and described amplifier, it is characterized in that, and described amplifier model method of estimation may further comprise the steps:

Select and store sampled point in the described input signal according to predetermined threshold, and obtain the time point of described sampled point;

Determine of the time-delay of described output feedback signal with respect to described input signal;

According to described time-delay, obtain the sampled point of the described input signal that aligns in time with described output feedback signal, and the time point of obtained sampled point is delayed time;

Time point according to the sampled point through delaying time obtains the corresponding feedback sample point of described output feedback signal at corresponding time point; And

Model according to described sampled point and the definite described amplifier of described corresponding feedback sample point.

Images