CN115037580B - Self-learning-based radio frequency predistortion system and method - Google Patents

Abstract

The present invention proposes a radio frequency predistortion system and method based on self-learning. The system includes a digital predistortion DPD processor, a power amplifier, a 1/G module and an operation module, as well as a DPD learner and a self-learning module; the implementation steps of the method are: initialization parameters; the DPD processor performs pre-distortion processing on the RF signal; the power amplifier performs power amplification on the pre-distorted RF signal; the 1/G module performs 1/G amplitude reduction on the amplified signal; the DPD learner performs non-linearity on the reduced signal Characteristic inverse processing; the computing module compares the predistortion RF signal and the predistortion estimation signal; the self-learning module estimates the predistortion parameters; determines whether the iteration conditions are met; and obtains the RF predistortion result. The radio frequency predistortion system of the present invention has a simple structure and low cost; the predistortion method improves the accuracy of the radio frequency signal predistortion processing results, thereby reducing the bit error rate of the communication system.

Description

Radio frequency predistortion system and method based on self-learning

Technical field

The invention belongs to the field of wireless communication technology and relates to a radio frequency predistortion system and method. Specifically, it relates to a system and method for realizing radio frequency predistortion based on self-learning, which can be used in various wireless communication systems.

technical background

Wireless communication system includes random bit generator, Gray encoder, mapper, signal converter, low-pass filter, modulator, power amplifier, channel, demodulator, demapper, Gray decoder, bit error rate detection controller, delayer and other components. The signal processing process is: at the transmitting end, the random bit generator generates a baseband signal and transmits the baseband signal to the Gray encoder; the Gray encoder performs Gray encoding on the baseband signal and transmits the encoded baseband signal to the mapper; mapper The encoded baseband signal is mapped to the constellation diagram, and then the baseband signal is converted into two signals by a signal converter and transmitted to the low-pass filter; the low-pass filter filters the two signals separately to reduce the interference between the signals. interference, and transmits the filtered signal to the modulator; the modulator up-converts the filtered signal into a radio frequency signal, and transmits the radio frequency signal to the power amplifier; the power amplifier amplifies the radio frequency signal, and passes the additive Gaussian white signal The noise channel is transmitted to the receiving end; at the receiving end, the demodulator down-converts the RF amplified signal into a baseband signal, and transmits the baseband signal to the low-pass filter; the low-pass filter filters out the baseband signal out-of-band noise, and then The two signals are converted into a complex signal through the signal converter, and the complex signal is transmitted to the demapper; the demapper maps the complex signal to the nearest point on the constellation diagram and outputs the point; the Gray encoder outputs the signal The output signals of the demapper are respectively transmitted to the bit error rate detector to detect the bit error rate of the output signal of the demapper. The bit error rate can reflect the quality of wireless communication, and the inherent nonlinear characteristics of the power amplifier will seriously affect the quality of wireless communication, causing the bit error rate to increase.

In order to ensure the linear performance of the power amplifier and reduce the bit error rate of signalless communication, the power amplifier needs to be linearized. At present, the main methods for linearizing power amplifiers include power back-off, negative feedback, feed forward, digital predistortion (DPD), etc. The principle of digital predistortion technology is to add a digital predistortion DPD processor at the front end of the power amplifier. The DPD processor will generate a distortion signal component for the input signal, which is opposite in phase to the distortion signal generated after the original input signal passes through the power amplifier. The amplitude is the same, thereby linearly compensating the distorted signal and offsetting the nonlinear effect of the power amplifier. Compared with other technologies, the predistortion method has high efficiency, good adaptability, and moderate complexity, so it has attracted the attention of many scholars.

The application publication number is CN114050795A, and the patent application titled "A digital pre-distortion processing system and method" discloses a digital pre-distortion processing system and method. The system structure is shown in Figure 1, including a predistorter, a power amplifier, a 1/G module, an operation module, a first learner and a second learner. This method first performs coefficient fitting estimation through the first learner, quickly determines the coefficient estimate interval, roughly adjusts the coefficient interval value, and then fine-tunes the coefficient interval value through the second learner. This invention requires at least two signal processes to achieve the digital predistortion effect. However, its shortcomings are that using a double-stage learner has a complex system structure, high cost, and low work efficiency; and the fine-tuning result of the second signal processing heavily depends on the first coarse-tuning result, and the difference between the two signal processing processes is The coupling is too high and it is not easy to maintain and adjust. Furthermore, the coefficient fitting estimation accuracy of the first learner is low, and the nonlinear processing effect of the power amplifier is poor, resulting in a high wireless communication bit error rate and poor communication quality.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the above-mentioned existing technologies, and propose a radio frequency pre-distortion system and method based on self-learning, which is used to solve the existing radio frequency pre-distortion systems' complex structure, high module coupling degree, and The radio frequency predistortion method has the technical problem of low calculation accuracy.

In order to achieve the above objects, the technical solutions adopted by the present invention are:

A radio frequency pre-distortion system based on self-learning includes a digital pre-distortion DPD processor, a power amplifier, a 1/G module and an arithmetic module connected in sequence. The output end of the digital pre-distortion DPD processor is also connected to the arithmetic module; so A digital predistortion DPD learner is loaded between the output end of the 1/G module and the input end of the operation module, and a self-learning module is loaded between the input end of the digital predistortion DPD learner and the output end of the operation module, in:

Digital predistortion DPD processor, used to predistort the input radio frequency signal;

Power amplifier, used to power amplify the pre-distorted radio frequency signal output by the DPD processor;

1/G module, used to reduce the 1/G amplitude of the amplified signal output by the power amplifier;

Digital predistortion DPD learner is used to inversely process the nonlinear characteristics of the reduced amplitude signal output by the 1/G module;

The operation module is used to perform difference operation on the pre-distortion radio frequency signal output by the DPD processor and the pre-distortion estimation signal output by the digital pre-distortion DPD learner;

The self-learning module includes a decision module and a parallel first predistortion parameter calculation module and a second predistortion parameter calculation module; the decision module is used to judge the error threshold function of the estimated error as the initialized nth radio frequency signal changes and the initialized The size of the error threshold; the first predistortion parameter operation module or the second predistortion parameter operation module estimates the predistortion parameters based on the judgment results of the decision module, and updates the parameters of the digital predistortion DPD learner through the predistortion parameters.

A radio frequency pre-distortion method based on self-learning, including the following steps:

(1)Initialization parameters:

Initialize the N channels of RF signals input to the RF predistortion system as x={x(1),x(2),...,x(n),...,x(N)}, and the error threshold is η ₀ , the nonlinear order of the power amplifier is l, the maximum nonlinear order is L, and let n=1, where N≥50, x(n) represents the nth radio frequency signal, L≥1, 1≤l ≤L;

(2) The digital predistortion DPD processor performs predistortion processing on RF signals:

The digital predistortion DPD processor performs predistortion processing on the nth RF signal x(n) input to the RF predistortion system and outputs the predistorted RF signal z(n);

(3) The power amplifier amplifies the power of the predistorted RF signal:

The power amplifier amplifies the predistorted radio frequency signal z(n) output by the digital predistortion DPD processor and outputs the amplified predistorted radio frequency signal y(n);

(4) The 1/G module reduces the amplified signal by 1/G amplitude:

The 1/G module reduces the amplitude of the amplified signal y(n) output by the power amplifier and outputs the reduced signal y(n)/G;

(5) The digital predistortion DPD learner performs inverse processing of nonlinear characteristics of the reduced signal:

The digital predistortion DPD learner performs inverse processing of nonlinear characteristics on the reduced signal y(n)/G output by the 1/G module, and outputs the predistortion estimation signal z′(n);

(6) The operation module compares the pre-distorted RF signal and the pre-distorted estimated signal:

The operation module compares the predistortion RF signal z(n) output by the digital predistortion DPD processor with the predistortion estimation signal z′(n) output by the digital predistortion DPD learner, and obtains the estimation error e(n)=|z (n)-z′(n)|;

(7) The self-learning module estimates the pre-distortion parameters:

The decision module determines the error threshold function of the estimated error e(n) as it changes with the initialized nth radio frequency signal x(n). Does it satisfy the initialized error threshold value eta ₀ /> If so, the first predistortion parameter operation module uses the recursive least squares RLS algorithm and estimates the predistortion parameters through e(n). Otherwise, the second predistortion parameter operation module uses the minimum mean square error LMS algorithm and estimates the predistortion parameters through e(n). ) Estimate the predistortion parameters, and then update the parameters of the digital predistortion DPD learner through the predistortion parameters;

(8) Determine whether the iteration conditions are met:

Determine whether n=N is true. If so, obtain the trained digital predistortion DPD learner and perform step (9). Otherwise, set n=n+1 and perform step (2);

(9) Obtain RF pre-distortion results:

The digital predistortion DPD processor extracts the parameters of the trained digital predistortion DPD learner and performs predistortion processing on the radio frequency signal x(n) through the parameters to achieve nonlinear processing of the power amplifier.

Compared with the prior art, the present invention has the following advantages:

1. The radio frequency predistortion system of the present invention uses a digital predistortion DPD learner to inversely process the nonlinear characteristics of the reduced amplitude signal output by the 1/G module, avoiding the problems caused by the dual-stage learner used in the existing radio frequency predistortion system. Compared with the existing technology, the disadvantages of complex structure and high coupling between learners can reduce the cost and computational complexity.

2. The present invention uses two predistortion parameter operation modules in the self-learning module loaded between the input end of the digital predistortion DPD learner and the output end of the operation module, and uses different methods to estimate the predistortion parameter according to the judgment results of the decision module. distortion parameters, and then update the parameters of the digital predistortion DPD learner through the predistortion parameters. After multiple rounds of training, the final parameters of the digital predistortion DPD learner are obtained, which avoids the existing technology's use of coefficient fitting methods to obtain the parameters of the learner. The influence of parameters on the accuracy of RF signal predistortion processing results effectively improves the nonlinear processing effect of the power amplifier, thereby reducing the bit error rate of the communication system.

Description of the drawings

Figure 1 is a schematic structural diagram of a radio frequency predistortion system in the prior art.

Figure 2 is a schematic structural diagram of the radio frequency predistortion system of the present invention.

Figure 3 is a flow chart of the implementation of the radio frequency predistortion method of the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

Referring to Figure 2, the radio frequency predistortion system of the present invention includes a digital predistortion DPD processor, a power amplifier, a 1/G module and an operation module that are connected in sequence. The output end of the digital predistortion DPD processor is also connected to the operation module; A digital predistortion DPD learner is loaded between the output end of the 1/G module and the input end of the operation module, and a self-learning module is loaded between the input end of the digital predistortion DPD learner and the output end of the operation module, where :

Digital predistortion DPD processor, used to predistort the input RF signal to compensate for the nonlinear distortion of the power amplifier;

Power amplifier, used to power amplify the pre-distorted radio frequency signal output by the DPD processor;

1/G module, used to reduce the 1/G amplitude of the amplified signal output by the power amplifier;

The digital predistortion DPD learner is used to inversely process the nonlinear characteristics of the reduced amplitude signal output by the 1/G module; the existing technology uses a second-order learner, which has a complex structure and high cost, and the second learner is serious Relying on the first learner, the modification of the parameters of the second learner needs to take the first learner into consideration; the present invention only uses a first-order learner, has a simple structure, low cost, and no coupling between modules;

The operation module is used to perform difference operation on the pre-distortion radio frequency signal output by the DPD processor and the pre-distortion estimation signal output by the digital pre-distortion DPD learner;

The self-learning module includes a decision module and a parallel first predistortion parameter calculation module and a second predistortion parameter calculation module; the decision module is used to judge the error threshold function of the estimated error as the initialized nth radio frequency signal changes and the initialized The size of the error threshold; the first predistortion parameter operation module or the second predistortion parameter operation module estimates the predistortion parameters based on the judgment results of the decision module, and updates the parameters of the digital predistortion DPD learner through the predistortion parameters.

Referring to Figure 3, the radio frequency predistortion method of the present invention includes the following steps:

Step 1) Initialization parameters:

Initialize the N channels of RF signals input to the RF predistortion system as x={x(1),x(2),...,x(n),...,x(N)}, and the error threshold is η ₀ , the nonlinear order of the power amplifier is l, and the maximum nonlinear order is L. The nonlinear order is the order of the input-output relationship expression of the power amplifier, which is used to characterize the degree of nonlinearity of the power amplifier, and let n= 1. Among them, N≥50, x(n) represents the nth radio frequency signal, L≥1, 1≤l≤L, and N=65 in this embodiment.

Step 2) The digital predistortion DPD processor performs predistortion processing on the radio frequency signal:

The digital predistortion DPD processor performs predistortion processing on the nth RF signal x(n) input to the RF predistortion system to compensate for the nonlinear distortion of the power amplifier and outputs the predistorted RF signal z(n); z(n) The calculation formula is:

Among them, ∑ represents the summation operation, M represents the maximum memory depth, the memory depth represents the number of signals between the current signal and the historical signal, h _lm represents the predistortion parameter when the nonlinear order is l and the memory depth is m, when l=1 , h _lm = 1 when m = 0, h _lm = 0 at other times, x (nm) represents the mth historical radio frequency signal before the current radio frequency signal x (n), 0 ≤ m ≤ M.

Step 3) The power amplifier amplifies the power of the pre-distorted RF signal:

The power amplifier amplifies the predistorted RF signal z(n) output by the digital predistortion DPD processor and outputs the amplified predistorted RF signal y(n); the calculation formula of y(n) is:

Among them, z(nm) represents the mth historical radio frequency signal before the current predistorted radio frequency signal z(n), a _lm represents the filter coefficient when the nonlinear order of the power amplifier is l and the memory depth is m, m represents the memory depth. , 0≤m≤M, M is the maximum memory depth.

Step 4) The 1/G module reduces the amplified signal by 1/G amplitude:

The 1/G module reduces the amplitude of the amplified signal y(n) output by the power amplifier to 1/G times the output signal of the power amplifier. The value of G is consistent with the amplification factor of the power amplifier and is used to offset the effect of the power amplifier on Amplification of signal amplitude, output reduction signal y(n)/G;

Step 5) The digital predistortion DPD learner performs inverse processing of nonlinear characteristics of the reduced signal:

The digital predistortion DPD learner performs inverse processing of nonlinear characteristics on the reduced signal y(n)/G output by the 1/G module, and outputs the predistortion estimated signal z′(n); the calculation formula of z′(n) is:

Among them, y(nm)/G represents the m-th historical signal before the current signal y(nm)/G, h _lm represents the pre-distortion parameter when the nonlinear order is l and the memory depth is m, m represents the memory depth, When l=1, m=0 other times 0≤m≤M, M is the maximum memory depth.

Step 6) The operation module compares the pre-distorted radio frequency signal and the pre-distorted estimated signal:

The operation module compares the predistortion RF signal z(n) output by the digital predistortion DPD processor with the predistortion estimation signal z′(n) output by the digital predistortion DPD learner, and obtains the estimation error e(n)=|z (n)-z′(n)|;

Step 7) The self-learning module estimates the pre-distortion parameters:

The decision module determines the error threshold function of the estimated error e(n) as it changes with the initialized nth radio frequency signal x(n). Does it satisfy the initialized error threshold value eta ₀ /> If so, the first predistortion parameter operation module uses the recursive least squares RLS algorithm and estimates the predistortion parameters through e(n). Otherwise, the second predistortion parameter operation module uses the minimum mean square error LMS algorithm and estimates the predistortion parameters through e(n). ) estimates the predistortion parameters, and then updates the parameters of the digital predistortion DPD learner through the predistortion parameters.

The recursive least squares RLS algorithm has the advantages of powerful target parameter tracking capability and fast function convergence. It can be used in time-varying systems for real-time parameter estimation and is suitable for adaptive control; the first predistortion parameter calculation module uses recursive least squares The RLS algorithm estimates the predistortion parameters, and the estimation formula is:

h _lm =min||e(n)|| ²

The minimum mean square error LMS algorithm has a simple calculation process, linear computational complexity, efficient calculation, is not susceptible to external disturbances, and has good Lupine performance. The second predistortion parameter calculation module uses the minimum mean square error LMS algorithm to estimate the predistortion parameters. The estimation formula is:

Among them, min{·} represents the minimum operation, ||·|| represents the norm operation, and E{·} represents the mean operation. Represents the predistortion parameters when the nonlinear order is l and the memory depth is m.

Step 8) Determine whether the iteration conditions are met:

Determine whether n=N is true. If so, obtain the trained digital predistortion DPD learner and perform step (9). Otherwise, set n=n+1 and perform step (2);

Step 9) Obtain RF pre-distortion results:

The digital predistortion DPD processor extracts the parameters of the trained digital predistortion DPD learner and performs predistortion processing on the radio frequency signal x(n) through the parameters to achieve nonlinear processing of the power amplifier.

In the prior art, the first learner performs coefficient fitting estimation. Coefficient fitting refers to knowing several discrete function values of a certain function. By adjusting several undetermined coefficients in the function, the difference between the estimated function and the expected function is reduced, thereby obtaining The estimated value is close to the expected value, but it is generally difficult to select a suitable known discrete function value, resulting in a large gap between the estimated value and the expected value, a large deviation between the goodness of fit and 1, and a low estimation accuracy; and the present invention Through step-by-step iteration, the gap between the estimated value and the expected value is continuously narrowed, thereby obtaining higher estimation accuracy.

Claims (6)

1. The radio frequency predistortion system based on self-learning comprises a digital predistortion DPD processor, a power amplifier, a 1/G module and an operation module which are sequentially connected, wherein the output end of the digital predistortion DPD processor is also connected with the operation module, and the radio frequency predistortion system based on self-learning is characterized in that a digital predistortion DPD learner is loaded between the output end of the 1/G module and the input end of the operation module, and a self-learning module is loaded between the input end of the digital predistortion DPD learner and the output end of the operation module, wherein:

a digital predistortion DPD processor for predistortion processing of an input radio frequency signal;

the power amplifier is used for amplifying the power of the predistortion radio frequency signal output by the DPD processor;

the 1/G module is used for carrying out 1/G amplitude reduction on the amplified signal output by the power amplifier;

the digital predistortion DPD learner is used for carrying out nonlinear characteristic inverse processing on the signal with reduced amplitude output by the 1/G module;

the operation module is used for carrying out difference operation on the predistortion radio frequency signal output by the DPD processor and the predistortion estimated signal output by the digital predistortion DPD learner;

the self-learning module comprises a judging module, a first predistortion parameter operation module and a second predistortion parameter operation module which are parallel; the judging module is used for judging the size of an error threshold function of the estimation error along with the change of the initialized nth path of radio frequency signals and the initialized error threshold value; the first predistortion parameter operation module or the second predistortion parameter operation module estimates predistortion parameters according to the judgment result of the judgment module, and updates parameters of the digital predistortion DPD learner through the predistortion parameters.

2. The self-learning-based radio frequency predistortion method is characterized by comprising the following steps of:

(1) Initializing parameters:

initializing N paths of radio frequency signals input into a radio frequency predistortion system into x= { x (1), x (2), x (N), N is more than or equal to 50, x (N) represents the N paths of radio frequency signals, and the error threshold value is eta ₀ The nonlinear order of the power amplifier is L, the maximum nonlinear order is L, L is more than or equal to 1 and less than or equal to L, and n=1;

(2) The digital predistortion DPD processor performs predistortion processing on the radio frequency signal:

the digital predistortion DPD processor performs predistortion processing on an n-th path radio frequency signal x (n) input into the radio frequency predistortion system and outputs a predistortion radio frequency signal z (n);

(3) The power amplifier performs power amplification on the predistortion radio frequency signal:

the power amplifier is used for amplifying the power of the predistortion radio frequency signal z (n) output by the digital predistortion DPD processor and outputting an amplified predistortion radio frequency signal y (n);

(4) The 1/G module performs 1/G amplitude reduction on the amplified signal:

the 1G module performs amplitude reduction on an amplified signal y (n) output by the power amplifier and outputs a reduced signal y (n)/G;

(5) The digital predistortion DPD learner performs nonlinear characteristic inverse processing on the reduced signal:

the digital predistortion DPD learner performs nonlinear characteristic inverse processing on the reduced signal y (n)/G output by the 1/G module and outputs a predistortion estimated signal z' (n);

(6) The operation module compares the predistortion radio frequency signal with the predistortion estimation signal:

the operation module compares a predistortion radio frequency signal z (n) output by the digital predistortion DPD processor with a predistortion estimated signal z '(n) output by the digital predistortion DPD learner to obtain an estimated error e (n) = |z (n) -z' (n) |;

(7) The self-learning module estimates predistortion parameters:

the judgment module judges an error threshold function of the estimated error e (n) along with the change of the initialized nth path radio frequency signal x (n)With an initialized error threshold value eta ₀ Whether or not to meet->If yes, the first predistortion parameter operation module adopts a Recursive Least Square (RLS) algorithm and estimates predistortion parameters through e (n), otherwise, the second predistortion parameter operation module adopts a Least Mean Square (LMS) algorithm and estimates predistortion parameters through e (n), and then the parameters of the digital predistortion DPD learner are updated through the predistortion parameters;

(8) Judging whether iteration conditions are met:

judging whether n=n is true, if so, obtaining a trained digital predistortion DPD learner, executing the step (9), otherwise, enabling n=n+1, and executing the step (2);

(9) Acquiring a radio frequency predistortion result:

the digital predistortion DPD processor extracts parameters of the trained digital predistortion DPD learner, and carries out predistortion processing on the radio frequency signal x (n) according to the parameters, thereby realizing nonlinear processing on the power amplifier.

3. The self-learning based rf predistortion method according to claim 2, wherein the predistortion rf signal z (n) in step (2) has a calculation formula:

wherein Σ represents the summation operation, M represents the maximum memory depth, h _lm Representing a predistortion parameter for a non-linear order of l memory depth of m, h when l=1, m=0 _lm =1, other times h _lm =0, x (n-M) represents the mth historical radio frequency signal preceding the current radio frequency signal x (n), 0.ltoreq.m.ltoreq.m.

4. The self-learning based rf predistortion method according to claim 2, wherein said amplified predistortion rf signal y (n) of step (3) is calculated by the formula:

wherein z (n-m) represents the mth historical radio frequency signal, a, preceding the current predistortion radio frequency signal z (n) _lm The filter coefficient when the nonlinear order of the power amplifier is l and the memory depth is M is represented by M, M represents the memory depth, M is more than or equal to 0 and less than or equal to M, and M is the maximum memory depth.

5. The self-learning based radio frequency predistortion method according to claim 2, wherein the predistortion estimation signal z' (n) in step (5) has a calculation formula:

wherein y (n-m)/G represents an mth history signal, h, preceding the current signal y (n-m)/G _lm Representing a predistortion parameter for a non-linear order of l memory depth of m, m representing memory depth, h when l=1, m=0 _lm =1, other times h _lm =0, 0.ltoreq.m.ltoreq.m, M being the maximum memory depth.

6. The self-learning-based radio frequency predistortion method according to claim 2, wherein the first predistortion parameter operation module in step (7) estimates predistortion parameters by using a recursive least squares RLS algorithm, and the second predistortion parameter operation module estimates predistortion parameters by using a least mean square error LMS algorithm, and the estimation formulas are respectively:

h _lm ＝min||e(n)|| ²

h _lm ＝min E{||e(n)|| ² }

wherein, min { · } represents minimum-value-seeking operation, |·|| represents norm-seeking operation, E { · } represents mean-seeking operation, h _lm Representing the predistortion parameters when the nonlinear order is l and the memory depth is m.