CN105631098B - A Generalized Memory Effect Hatchback Nonlinear Model of a Broadband RF Power Amplifier - Google Patents

Abstract

The invention discloses a kind of General Memory effect wing-rooms on either side of a one-story house nonlinear models of wide band RF power amplifier, including the memoryless multinomial processing unit of static non linear, General Memory nonlinear processing unit and adder, the input terminal of the memoryless multinomial processing unit of static non linear and the input terminal of General Memory nonlinear processing unit are respectively connected to signal x (n+M), by the input terminal of wide band RF power amplifier current sample time sampled signal be x (n), x (n+M) be wide band RF power amplifier input terminal current sample time the following m-th sampling instant sampled signal, M=1, 2, 3, ...；The output signal y of the memoryless multinomial processing unit of static non linear₁(n) and the output signal y of General Memory nonlinear processing unit₂It (n) is two addends of the adder, output signal y (n)=y of adder₁(n)+y₂(n)；Advantage is that the easy training of model coefficient obtains, and modeling accuracy is higher.

Description

A Generalized Memory Effect Hatchback Nonlinear Model of a Broadband RF Power Amplifier

technical field

The invention relates to a nonlinear model of a broadband radio frequency power amplifier, in particular to a generalized memory effect hatchback nonlinear model of the broadband radio frequency power amplifier.

Background technique

There have been many related studies on the nonlinear modeling of strong static nonlinearity and memory effect of broadband RF power amplifiers, but these are not only memory polynomial nonlinear models, traditional generalized memory polynomial nonlinear models, or improved generalized fractional-order memory polynomial models. Linear models all have problems such as the higher the order, the more severe the coefficient divergence, and the worse the numerical stability. At the same time, with the increase of the polynomial order, the number of coefficients increases greatly, and the training difficulty increases.

The three main nonlinear models currently used are the traditional generalized memory effect polynomial model shown in Eq. (1), the improved Hammerstein nonlinear model shown in Fig. 1, and the serial dual nonlinear hatchback nonlinear model shown in Fig. 2 Model.

The traditional generalized memory effect polynomial model has high modeling accuracy, but there are problems such as complex model, large number of coefficients, and complicated training. The improved Hammerstein nonlinear model has only 0 and 1 orders in the memory effect part, and the order is small, and does not consider the advance memory effect, so the improved Hammerstein modeling accuracy is not high; the serial dual nonlinear hatchback nonlinear model is relatively Since the improved Hammerstein nonlinear model has many orders, it also does not consider the advance memory effect. Although its modeling accuracy is slightly improved compared with the improved Hammerstein nonlinear model, it is still not high, and the serial dual nonlinear hatchback The model cannot be solved at one time by the linear training method during model training, which makes the model coefficients difficult to obtain.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a generalized memory effect nonlinear model of a broadband radio frequency power amplifier with simple model, fewer coefficients, easy training of model coefficients and high modeling accuracy.

The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a generalized memory effect hatchback nonlinear model of a broadband radio frequency power amplifier, comprising a static nonlinear memoryless polynomial processing unit, a generalized memory nonlinear processing unit and an adder. The input end of the static nonlinear memoryless polynomial processing unit and the input end of the generalized memory nonlinear processing unit are respectively connected to the signal x(n+M), and the sampling signal of the input end of the broadband radio frequency power amplifier at the current sampling time is x (n), x(n+M) is the sampling signal of the input end of the broadband radio frequency power amplifier at the Mth sampling time in the future at the current sampling time, M=1, 2, 3, ......; the described The output signal y 1 (n) of the static nonlinear memoryless polynomial processing unit and the output signal y ₂ (n) of the generalized memory nonlinear processing unit are the _two addends of the adder, and the addition The output signal y(n)=y ₁ (n)+y ₂ (n) of the device.

The static nonlinear memoryless polynomial processing unit is expressed in the form of a static nonlinear memoryless polynomial as:

Among them, f is the highest order of the static nonlinear memoryless polynomial, e is the lowest order of the static nonlinear memoryless polynomial, f﹥e, e and f are all positive integers, is the coefficient of the static nonlinear memoryless polynomial, and the symbol "||" is the modulo symbol.

The generalized memory nonlinear processing unit is expressed by an improved generalized memory polynomial as:

in, is the complex envelope leading term of the improved generalized memory polynomial, is the complex envelope alignment term of the improved generalized memory polynomial; m is a positive integer, m=1, 2, ..., M; g is the order of the complex envelope leading term, k is the memory depth of the complex envelope leading term, Both g and k are positive integers, k≥1, g<e, is the coefficient of the complex envelope leading term; the symbol “| |” is the modulo symbol; r is the order of the complex envelope alignment term, s is the memory depth of the complex envelope alignment term, r and s are both positive integers, s ≥1, r<e, is the coefficient of the complex envelope alignment term; when q=0, x(nq) is the sampling signal of the input end of the broadband RF power amplifier at the current sampling moment, and x(n-(qm)) is the input end of the broadband RF power amplifier at The sampling signal of the mth sampling time in the future at the current sampling time; when q≠0 and q≥m, x(nq) is the sampling signal of the input end of the broadband RF power amplifier at the qth sampling time in the past of the current sampling time, x(n-(qm)) is the sampling signal of the input end of the broadband RF power amplifier at the qmth sampling time in the past of the current sampling time; when q≠0 and q<m, x(nq) is the input of the broadband RF power amplifier The sampling signal of the terminal at the qth sampling moment in the past at the current sampling moment, x(n-(qm)) is the sampling signal of the input terminal of the wideband radio frequency power amplifier at the mqth sampling moment in the future at the current sampling moment;

At this time, the generalized memory effect hatchback nonlinear model of the broadband RF power amplifier can be expressed as:

In the generalized memory effect hatchback nonlinear model of the broadband RF power amplifier, and It is obtained by training using the existing nonlinear model training method. In this method, the model coefficients and It can be solved at one time by the linear training method, and the training method is simple.

Compared with the prior art, the present invention has the advantage that the high-order nonlinear characteristics of the broadband radio frequency power amplifier are calculated by the static nonlinear memoryless polynomial processing unit, and the low-order nonlinear characteristics of the broadband radio frequency power amplifier are calculated by the generalized memory nonlinear processing unit. , and then add the high-order nonlinear characteristics of the broadband radio frequency power amplifier and the low-order nonlinear characteristics of the broadband radio frequency power amplifier using an adder to obtain the nonlinear characteristics of the broadband radio frequency power amplifier, thus avoiding the occurrence of high-order nonlinearity in the generalized memory nonlinear processing unit. term, considering the advanced memory effect comprehensively, the model coefficients can be solved at one time by the linear training method, and the modeling accuracy is high.

Description of drawings

Fig. 1 is the structure diagram of the existing improved Hammerstein nonlinear model;

FIG. 2 is a structural diagram of an existing serial dual nonlinear hatchback nonlinear model;

FIG. 3 is a structural diagram of the generalized memory effect hatchback nonlinear model of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the embodiments of the accompanying drawings.

Embodiment 1: As shown in Figure 3, a generalized memory effect hatchback nonlinear model of a broadband radio frequency power amplifier is characterized in that it includes a static nonlinear memoryless polynomial processing unit, a generalized memory nonlinear processing unit and an adder. The input end of the memoryless polynomial processing unit and the input end of the generalized memory nonlinear processing unit are respectively connected to the signal x(n+M), and the sampling signal of the input end of the broadband RF power amplifier at the current sampling time is x(n), x( n+M) is the sampling signal of the input end of the broadband RF power amplifier at the Mth sampling time in the future at the current sampling time, where M=1 is taken; the output signal y ₁ (n) of the static nonlinear memoryless polynomial processing unit and the generalized The output signal y ₂ (n) of the memory nonlinear processing unit is the two addends of the adder, and the output signal y(n)=y ₁ (n)+y ₂ (n) of the adder.

Embodiment 2: As shown in Figure 3, a generalized memory effect hatchback nonlinear model of a broadband radio frequency power amplifier is characterized in that it includes a static nonlinear memoryless polynomial processing unit, a generalized memory nonlinear processing unit and an adder. The input end of the memoryless polynomial processing unit and the input end of the generalized memory nonlinear processing unit are respectively connected to the signal x(n+M), and the sampling signal of the input end of the broadband RF power amplifier at the current sampling time is x(n), x( n+M) is the sampling signal of the input end of the broadband RF power amplifier at the Mth sampling time in the future at the current sampling time, where M=1 is taken; the output signal y ₁ (n) of the static nonlinear memoryless polynomial processing unit and the generalized The output signal y ₂ (n) of the memory nonlinear processing unit is the two addends of the adder, and the output signal y(n)=y ₁ (n)+y ₂ (n) of the adder.

In this embodiment, the static nonlinear memoryless polynomial processing unit is expressed in the form of a static nonlinear memoryless polynomial as:

Among them, f is the highest order of the static nonlinear memoryless polynomial, e is the lowest order of the static nonlinear memoryless polynomial, f﹥e, e and f are all positive integers, is the coefficient of the static nonlinear memoryless polynomial, and the symbol "| |" is the modulo symbol.

Embodiment 3: As shown in FIG. 3, a generalized memory effect hatchback nonlinear model of a broadband radio frequency power amplifier is characterized in that it includes a static nonlinear memoryless polynomial processing unit, a generalized memory nonlinear processing unit and an adder. The input end of the memoryless polynomial processing unit and the input end of the generalized memory nonlinear processing unit are respectively connected to the signal x(n+M), and the sampling signal of the input end of the broadband RF power amplifier at the current sampling time is x(n), x( n+M) is the sampling signal of the input end of the broadband RF power amplifier at the Mth sampling time in the future at the current sampling time, where M=1 is taken; the output signal y ₁ (n) of the static nonlinear memoryless polynomial processing unit and the generalized The output signal y ₂ (n) of the memory nonlinear processing unit is the two addends of the adder, and the output signal y(n)=y ₁ (n)+y ₂ (n) of the adder.

In this embodiment, the static nonlinear memoryless polynomial processing unit is expressed in the form of a static nonlinear memoryless polynomial as:

Among them, f is the highest order of the static nonlinear memoryless polynomial, e is the lowest order of the static nonlinear memoryless polynomial, f﹥e, e and f are all positive integers, is the coefficient of the static nonlinear memoryless polynomial, and the symbol "| |" is the modulo symbol.

The generalized memory nonlinear processing unit is expressed as an improved generalized memory polynomial:

in, is the complex envelope leading term of the improved generalized memory polynomial, is the complex envelope alignment term of the improved generalized memory polynomial; m is a positive integer, m=1, 2, ..., M; g is the order of the complex envelope leading term, k is the memory depth of the complex envelope leading term, Both g and k are positive integers, k≥1, g<e, is the coefficient of the complex envelope leading term; the symbol “| |” is the modulo symbol; r is the order of the complex envelope alignment term, s is the memory depth of the complex envelope alignment term, r and s are both positive integers, s ≥1, r<e, is the coefficient of the complex envelope alignment term; when q=0, x(nq) is the sampling signal of the input end of the broadband RF power amplifier at the current sampling moment, and x(n-(qm)) is the input end of the broadband RF power amplifier at The sampling signal of the mth sampling time in the future at the current sampling time; when q≠0 and q≥m, x(nq) is the sampling signal of the input end of the broadband RF power amplifier at the qth sampling time in the past of the current sampling time, x(n-(qm)) is the sampling signal of the input end of the broadband RF power amplifier at the qmth sampling time in the past of the current sampling time; when q≠0 and q<m, x(nq) is the input of the broadband RF power amplifier The sampling signal of the terminal at the qth sampling moment in the past at the current sampling moment, x(n-(qm)) is the sampling signal of the input terminal of the wideband radio frequency power amplifier at the mqth sampling moment in the future at the current sampling moment;

At this time, the generalized memory effect hatchback nonlinear model of the broadband RF power amplifier can be expressed as:

Generalized Memory Effect Hatchback Nonlinear Model of Broadband RF Power Amplifier and It is obtained by training using the existing nonlinear model training method.

In order to verify the superiority of the generalized memory effect hatchback nonlinear model of the broadband radio frequency power amplifier of the present invention, 6000 baseband data at the input end of the Doherty power amplifier are collected as the model input training data, and 6000 baseband data at the corresponding power amplifier output end are used as the model output training data. The existing model training method performs model parameter training on the generalized memory effect hatchback nonlinear model of the broadband radio frequency power amplifier of the present embodiment to obtain model parameters and After the training of the generalized memory effect hatchback nonlinear model of the broadband radio frequency power amplifier of this embodiment is completed, 12000 baseband data at the input end of the power amplifier are additionally selected as the input data of the model to verify its modeling accuracy.

The generalized memory effect hatchback nonlinear model of the broadband radio frequency power amplifier in this embodiment has e=5, f=7, g=4, k=4, r=4, and s=4.

Table 1 shows the comparison of the NMSE and the number of model coefficients of the generalized memory effect hatchback nonlinear model, the traditional generalized memory polynomial model and the serial dual nonlinear hatchback nonlinear model of the broadband RF power amplifier of this embodiment.

Table 1. Performance comparison of three different models

It can be seen from the analysis of Table 1 that the number of coefficients of the generalized memory effect hatchback nonlinear model of the present invention is reduced by 40% compared with the traditional generalized memory effect polynomial model, and the modeling accuracy is only reduced by 0.5dB. Compared with the serial dual nonlinear hatchback nonlinear model, the modeling accuracy of the generalized memory effect hatchback nonlinear model is improved by 3.6dB.

As mentioned above, the generalized memory effect hatchback nonlinear model of the present invention has the best compromise between modeling ability, training complexity and number of coefficients, not only has the advantages of high modeling accuracy and simple training, but also has the advantages of less number of coefficients, etc. advantage.

Claims (1)

1. The generalized memory effect two-compartment nonlinear system of the broadband radio frequency power amplifier is characterized by comprising a static nonlinear memoryless polynomial processing unit, a generalized memory nonlinear processing unit and an adder, wherein the input end of the static nonlinear memoryless polynomial processing unit and the input end of the generalized memory nonlinear processing unit are respectively connected with a signal x (n + M), the sampling signal of the input end of the broadband radio frequency power amplifier at the current sampling moment is x (n), x (n + M) is the sampling signal of the input end of the broadband radio frequency power amplifier at the Mth sampling moment in the future of the current sampling moment, and M is 1,2,3 and …; what is needed isThe output signal y of the static nonlinear memoryless polynomial processing unit₁(n) and the output signal y of said generalized memory nonlinear processing unit₂(n) is two addends of the adder, and the output signal y (n) of the adder is y₁(n)+y₂(n)；

The static nonlinear memoryless polynomial processing unit adopts a static nonlinear memoryless polynomial form to express as follows:

wherein f is the highest order of the static non-linear memoryless polynomial, e is the lowest order of the static non-linear memoryless polynomial, f > e, e and f are positive integers,the symbol, | | | is a modulus symbol, and the symbol Σ is a consecutive plus symbol;

the generalized memory nonlinear processing unit is expressed by adopting an improved generalized memory polynomial as follows:

wherein,in order to improve the complex envelope antecedent term of the generalized memory polynomial,a complex envelope alignment term that is an improved generalized memory polynomial; m is a positive integer, M is 1,2, …, M; g is the order of the complex envelope antecedent term, k is the memory depth of the complex envelope antecedent term, g and k are positive integers, k is more than or equal to 1, g<e，Coefficients that are complex envelope antecedent terms; the symbol "|" is a modulo symbol; r is the order of the complex envelope alignment term, s is the memory depth of the complex envelope alignment term, r and s are positive integers, s is more than or equal to 1, r<e，Coefficients that are complex envelope alignment terms; when q is 0, x (n-q) is a sampling signal of the input end of the broadband radio frequency power amplifier at the current sampling moment, and x (n- (q-m)) is a sampling signal of the input end of the broadband radio frequency power amplifier at the mth sampling moment in the future of the current sampling moment; when q is not equal to 0 and q is not less than m, x (n-q) is a sampling signal of the input end of the broadband radio frequency power amplifier at the past q-th sampling moment of the current sampling moment, and x (n- (q-m)) is a sampling signal of the input end of the broadband radio frequency power amplifier at the past q-m-th sampling moment of the current sampling moment; when q is not equal to 0 and q is not equal to<When m is reached, x (n-q) is a sampling signal of the input end of the broadband radio frequency power amplifier at the past q sampling time of the current sampling time, and x (n- (q-m)) is a sampling signal of the input end of the broadband radio frequency power amplifier at the future m-q sampling time of the current sampling time;

at this time, the generalized memory effect two-compartment nonlinear system of the broadband radio frequency power amplifier is represented as follows:

the generalized memory effect two-compartment nonlinear system of the broadband radio frequency power amplifierAndthe training method is obtained by adopting the existing training method of the nonlinear system.