Broadband power amplifier linearization method and system based on predistortion

Abstract

The invention belongs to the technical field of data processing of the computer information disciplinary, and relates to a broadband power amplifier linearization method and system based on predistortion, and the method comprises the steps: collecting and preprocessing a radio frequency input baseband signal, an actual output baseband signal and tube core heat flow data of a broadband power amplifier, and obtaining an initial data set; calculating a transient heat dissipation gradient at each sampling moment based on the signal energy difference and the tube core temperature change rate; in combination with broadband distribution characteristics of radio frequency input baseband signals, broadband thermal teratogenesis factors at all sampling moments are obtained; performing dynamic tracking compensation on a cross term coefficient of a generalized memory polynomial based on the distortion factor to obtain an update coefficient; and the pre-distortion signal is reconstructed by utilizing the updating coefficient, so that high-linearity output of the broadband power amplifier is realized. The dynamic response speed of coefficient updating can be improved, thermally induced nonlinear distortion of a power amplifier is effectively offset, and the linearity and stability of a high-power broadband radio frequency test are improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology in computer information science, and relates to a method and system for linearizing a broadband power amplifier based on predistortion, used to correct or eliminate the distorted signal output by the power amplifier. Background Technology

[0002] In the field of precision test and measurement, wideband power amplifiers are core components in RF test instruments such as signal generators and network analyzers, enabling high-power output. To eliminate the nonlinear distortion generated by wideband power amplifiers during high-power operation, the industry widely employs digital predistortion technology to perform inverse compensation on the input signal, thereby improving the linearity of the test output. Generalized memory polynomial algorithms, as a mainstream core technology in the industry, can significantly improve the fitting and compensation accuracy for the strong memory effect of broadband signals by introducing lead and lag cross terms in the signal envelope.

[0003] To address the impact of power amplifier thermal effects on linearity, several improvement schemes have been proposed in existing technologies. For example, patent application 1 (CN117040450B) proposes a piecewise linear digital predistortion system and method for power amplifier temperature compensation. This system tests the power amplifier characteristics at different temperature points, fits the relationship function between the piecewise linear model parameters and temperature, and updates the model parameters based on real-time temperature during actual operation, thereby compensating for the impact of temperature changes on the power amplifier characteristics. However, this scheme is essentially still a quasi-static temperature compensation method. Its parameter updates rely on a pre-fitted temperature-parameter function relationship, which cannot track the dynamic thermal memory effect caused by rapid transient fluctuations in chip temperature under high-power continuous or pulsed output conditions in real time. When the temperature changes drastically, deviations easily occur between the pre-fitted relationship and the actual nonlinear characteristics of the power amplifier. Patent application document 2 (US7577211B2) discloses a digital predistortion system and method for high-efficiency transmitters, which adopts a three-parallel architecture of linear path, memoryless predistortion path and memory-based predistortion path, and uses an IIR filter library to compensate for higher-order reactive memory effect and thermal memory effect. However, the model structure of this scheme is fixed and the parameter adaptive update cycle is long. When faced with the rapid transient drift of nonlinear characteristics caused by instantaneous heating of wideband high-power power amplifiers, it is difficult to achieve real-time tracking and rapid parameter updates.

[0004] Currently, existing generalized memory polynomial algorithms have significant limitations when facing high-power precision measurement scenarios. When high-power RF power amplifiers output continuous or burst pulses, the chip temperature rises rapidly and fluctuates violently, triggering a strong dynamic thermal memory effect. This causes the nonlinear characteristic curve of the power amplifier to drift rapidly with temperature. Because the generalized memory polynomial algorithm relies on a large mathematical model with high-order, deep memory dimensions when fitting broadband cross features, its autocorrelation matrix has extremely high dimensions. This results in massive high-dimensional matrix inversion operations being required when updating predistortion parameters. This extremely high computational load makes the cycle of updating predistortion coefficients very long, and the algorithm's response speed lags far behind the speed of temperature drift. This quasi-static, massive computational mode cannot track the nonlinear mutations caused by instantaneous heating in real time. As a result, the algorithm cannot adjust the model weights in time during the later stages of long-term full-load testing, ultimately leading to linearization lock-out, which seriously undermines the extremely high stability and measurement accuracy required by the test instrument. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology, seek to solve the technical problem of slow tracking of dynamic thermal memory effect caused by high computational load, and provide a broadband power amplifier linearization method based on predistortion.

[0006] To achieve the aforementioned objectives, this invention provides a predistortion-based method for linearizing a broadband power amplifier. The specific steps include: acquiring the RF input baseband signal, actual output baseband signal, and die heat flux data of the broadband power amplifier, and preprocessing them to obtain an initial data set; obtaining the transient heat dissipation gradient corresponding to each sampling moment based on the energy difference between the RF input baseband signal and the actual output baseband signal in the initial data set, and the rate of change of the die heat flux data per unit time; calculating the broadband thermal distortion factor corresponding to each sampling moment based on the transient heat dissipation gradient and the broadband distribution characteristics of the RF input baseband signal at each sampling moment; dynamically tracking and compensating the cross-term coefficients of the generalized memory polynomial based on the broadband thermal distortion factor at each sampling moment to obtain the updated coefficients of the generalized memory polynomial after dynamic tracking and compensation; and reconstructing the predistortion signal based on the updated coefficients after dynamic tracking and compensation at each sampling moment to achieve high linearity output of the broadband power amplifier.

[0007] This invention extracts effective thermal and signal features through multi-source synchronous data acquisition and preprocessing. First, it quantifies the transient thermal dissipation gradient corresponding to the dynamic thermal memory effect of a broadband power amplifier. Then, it fuses the broadband distribution characteristics of the RF input baseband signal to obtain a broadband thermal distortion factor, which serves as the basis for dynamic correction. During the intervals of traditional long-period full-matrix coefficient updates of the generalized memory polynomial, a high-frequency, short-period fast tracking mechanism is introduced. Scalar operations are used to accurately and dynamically compensate the cross-term coefficients of the generalized memory polynomial, replacing the traditional high-dimensional matrix inversion coefficient update method. This achieves low-latency, high-dynamic updates of the predistortion model coefficients. Finally, based on the updated coefficients, the predistortion cancellation signal is reconstructed, accurately canceling the amplitude and phase frequency nonlinear distortions caused by die temperature drift and thermal memory effects in the broadband power amplifier. This solves the technical problems of high computational load, slow thermal memory effect tracking, and easy linearization lock-up in traditional algorithms. It significantly improves the linearity and output stability of the broadband power amplifier under high-power conditions, while reducing the computational complexity of predistortion coefficient updates, adapting to the application requirements of RF test instruments for high-precision, high-dynamic linearization compensation.

[0008] The method for obtaining an initial data set as described in this invention includes: mapping the RF input baseband signal, the actual output baseband signal, and the die thermal flow data to the same time domain index to achieve high-precision timestamp alignment; then sequentially performing outlier removal, noise filtering, and missing value interpolation; introducing a data truncation mechanism based on a sliding window, setting a fixed preset length as the length of the sliding window, and setting a sliding step size to control the movement of the sliding window, wherein the sliding step size is half of the preset length; and truncating the preprocessed data stream into multiple valid data segments through the sliding window, wherein the valid data segments constitute the initial data set.

[0009] The present invention describes obtaining the transient heat dissipation gradient corresponding to each sampling time, including: In the formula, It is the first Transient heat dissipation gradient at each sampling time; It is an index of the sampling time; , These are the first two in the sliding window. The RF input baseband signal and the RF input baseband signal envelope at each sampling point; , These are the first two in the sliding window. The actual output baseband signal and the actual output baseband signal envelope at each sampling point; It is the index of the sampling point within the sliding window; It is the theoretical linear gain of a broadband power amplifier; It is the first Core heat flow data at each sampling time; It is the first Core heat flow data at each sampling time; It is the sampling time interval; It is the preset length of the sliding window; It is a constant to prevent the denominator from being zero.

[0010] This invention couples the energy loss difference between the RF input baseband signal and the actual output baseband signal with the unit-time change rate of the die heat flow data, and combines this with a sliding window to accumulate continuously sampled data, constructing a quantitative calculation model for the transient heat dissipation gradient. This enables accurate characterization of the transient heating impact intensity caused by power loss and temperature abrupt changes in broadband power amplifiers. It overcomes the limitation that relying solely on the absolute die temperature cannot capture the abrupt change nodes of the thermal memory effect, and by using the time-domain constraints of the sliding window to fit the time-domain characteristics of the memory effect in broadband power amplifiers, the calculated transient heat dissipation gradient can accurately match the real-time changes of the dynamic thermal memory effect. This provides accurate and effective thermal characteristic parameters for the subsequent quantification and compensation of thermal distortion. At the same time, the parameters in the model are highly compatible with the pre-processed sampled data and the sliding window configuration, ensuring the consistency of the calculation process and the reliability of the results.

[0011] The method for obtaining the theoretical linear gain of a broadband power amplifier according to the present invention includes: when the broadband power amplifier is in the low-power linear operating range, inputting a set of standard test baseband signals with known amplitudes to the broadband power amplifier through the baseband processing unit of the test instrument, and simultaneously acquiring the standard test output baseband signal output by the broadband power amplifier; calculating the ratio of the envelope amplitude of the standard test output baseband signal to the envelope amplitude of the standard test baseband signal, and taking the average value of multiple test results, which is the theoretical linear gain.

[0012] This invention avoids interference from nonlinear distortion under high-power conditions on gain calculation by conducting standard signal tests in the low-power linear operating range of a broadband power amplifier, ensuring the accuracy of the gain reference. Simultaneously, it employs a standard test baseband signal with known amplitude as input and synchronously acquires the output signal. The gain is calculated by the ratio of their envelope amplitudes, and the average of multiple test results is taken to effectively offset random errors during the testing process, obtaining a precise and stable theoretical linear gain for the broadband power amplifier. This theoretical linear gain serves as the core reference parameter for subsequent signal energy difference calculations, providing a reliable reference for the accurate quantification of transient thermal dissipation gradients and ensuring the accuracy of thermal memory effect feature extraction. Furthermore, the entire acquisition method is simple to operate, compatible with conventional testing procedures of testing instruments, and possesses good engineering feasibility.

[0013] The broadband distribution characteristics of the RF input baseband signal corresponding to each sampling time described in this invention include: the statistical variance of the RF input baseband signal envelope within the sliding window to which each sampling time belongs, the physical cross-band bandwidth limit set by the test system, and the preset ambient reference temperature.

[0014] The present invention describes obtaining the broadband thermal distortion factor corresponding to each sampling time, including: In the formula, It is the first Broadband thermal distortion factor corresponding to each sampling time; It has dimensions The normalized total coefficient; It is the first Transient heat dissipation gradient at each sampling time; It is the first The statistical variance of the envelope of the RF input baseband signal within the sliding window of each sampling time; It is the physical cross-frequency bandwidth limit; It is the ambient standard room temperature; Based on the natural constant The logarithmic function with base 0; It is an index of the sampling time; It is the absolute value symbol.

[0015] This invention couples the transient thermal dissipation gradient, which characterizes the thermal-induced heating impact of a power amplifier, with the statistical variance of the RF input baseband signal envelope, which reflects the intensity of signal amplitude modulation. It also introduces physical crossband bandwidth limits and ambient temperature as normalization constraints, and constructs a broadband thermal distortion factor calculation model using logarithmic and square root nonlinear mappings. This enables precise quantification of the degree of thermally induced nonlinear distortion in broadband power amplifiers. The model is specifically tailored to the significant distortion at the bandwidth edges of broadband power amplifiers, deeply integrating thermal characteristics with broadband signal distribution characteristics. It can sensitively reflect the strength of thermal memory effects based on changes in the thermal dissipation gradient, and also demonstrate the influence of broadband signal characteristics on distortion through signal envelope variance and physical bandwidth. The introduction of ambient temperature eliminates the interference of environmental temperature differences on distortion quantification, allowing the calculated broadband thermal distortion factor to accurately characterize the distortion depth of the power amplifier under different sampling times and broadband operating conditions due to dynamic thermal memory effects. This provides a basis for dynamic compensation of the predistortion coefficient that is consistent with actual broadband operating conditions.

[0016] The present invention describes a dynamic tracking and compensation method for the cross-term coefficients of a generalized memory polynomial to obtain updated coefficients of the generalized memory polynomial after dynamic tracking and compensation, comprising: In the formula, These are the update coefficients of the generalized memory polynomial after dynamic tracking and compensation; These are the original cross-term coefficients of the generalized memory polynomial obtained after updating the full matrix coefficients of the previous long period; It is an index of the signal envelope delay memory depth; It is an index of the signal envelope's advance memory depth; It is the nonlinear order of the generalized memory polynomial; It is an index of the sampling time; It is the memory time span corresponding to the current cross item; It is the first Broadband thermal distortion factor corresponding to each sampling time; It is the maximum distortion tolerance constant; It is the hyperbolic tangent function.

[0017] This invention couples a wideband thermal distortion factor with the cross-term memory time span and inputs the result into a hyperbolic tangent function to construct a dynamic correction multiplier. This multiplier is then multiplied by the original cross-term coefficients of the generalized memory polynomial, forming a dynamic tracking compensation model for the cross-term coefficients. This model enables rapid updates to the predistortion coefficients. The model uses the wideband thermal distortion factor to characterize the distortion degree of the thermal memory effect and uses the memory time span to weight the degree of thermal sensitivity. This allows long-span cross-terms, which are more sensitive to the thermal memory effect, to receive stronger compensation corrections. The hyperbolic tangent function constrains the correction amplitude between 0 and 1, ensuring the smooth stability of the coefficient compensation. The maximum distortion tolerance constant further limits the overall correction amplitude, preventing model instability. The entire compensation process is completed only through scalar operations, eliminating the need for high-dimensional matrix inversion in traditional algorithms. This allows for high-frequency execution during long-period full matrix coefficient updates, achieving low-latency dynamic updates of the predistortion coefficients. This enables the coefficients to follow the dynamic thermal memory effect of the power amplifier in real time, accurately matching the current thermally induced nonlinear distortion state and significantly improving the dynamic response speed and adaptation accuracy of predistortion compensation.

[0018] The index of signal envelope delay memory depth described in this invention Index of signal envelope advance memory depth Nonlinear order of generalized memory polynomials and the span of memory The setup method includes: indexing the signal envelope delay memory depth. The value is determined based on the memory effect depth of the broadband power amplifier, ranging from 0 to 32; and The matching constitutes the memory span range of the memory polynomial, with values ​​ranging from 0 to 16; The value is preset to 3 to 5 based on the nonlinear characteristics and compensation accuracy requirements of the broadband power amplifier; Depend on and The result is obtained by taking the absolute value after calculating the difference.

[0019] The reconstruction of the predistortion signal described in this invention to achieve high linearity output of a broadband power amplifier includes: substituting the updated coefficient set of the dynamically tracked and compensated generalized memory polynomial into the generalized memory polynomial operation engine of the test instrument one by one; calling the updated coefficient set and performing an inner product operation with the real-time acquired broadband RF input baseband signal to generate a predistortion cancellation signal, wherein the signal format and bandwidth range of the predistortion cancellation signal are consistent with the RF input baseband signal; and sending the predistortion cancellation signal into the RF analog link of the test instrument for preprocessing, and then inputting it into the broadband power amplifier for high-power amplification to achieve high linearity output of the broadband power amplifier.

[0020] This invention also provides a predistortion-based broadband power amplifier linearization system, the main structure of which includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned predistortion-based broadband power amplifier linearization method. By adopting the above technical solution, the aforementioned predistortion-based broadband power amplifier linearization method is generated into a computer program and stored in the memory for loading and execution by the processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0021] Compared with existing technologies, the advantages of this invention are as follows: It effectively solves the technical problem of slow tracking of dynamic thermal memory effect in broadband power amplifiers due to the high computational load of traditional generalized memory polynomial algorithms. By extracting effective thermal and signal features through multi-source synchronous data acquisition and preprocessing, it first quantizes the transient thermal dissipation gradient characterizing the intensity of transient heating impact, and then fuses the broadband distribution characteristics of the RF input baseband signal to generate a broadband thermal distortion factor that accurately characterizes the degree of thermal distortion. Based on this, a high-frequency short-period fast tracking mechanism is introduced into the gap of traditional long-period full matrix coefficient update. Through scalar operation, the cross-term coefficients of the generalized memory polynomial are dynamically compensated with weights, replacing the traditional coefficient update method of high-dimensional matrix inversion. This not only significantly reduces computational complexity and improves the dynamic response speed of predistortion coefficient update, but also allows... The coefficients follow the transient changes in die temperature drift and thermal memory effect in real time, achieving precise adaptive compensation for thermally induced nonlinear distortion. At the same time, the dual constraints of the hyperbolic tangent function and the maximum distortion tolerance constant ensure the smooth stability of coefficient compensation. Based on the pre-distortion cancellation signal reconstructed by the updated coefficients, it can effectively cancel the amplitude and phase frequency nonlinear distortions caused by the dynamic thermal memory effect in broadband power amplifiers, significantly improving their output linearity under high power conditions and stability under long-term full-load testing. Moreover, the parameters of this invention can be flexibly configured according to the power amplifier characteristics and test accuracy, achieving high compatibility with the conventional architecture of RF test instruments without the need for additional complex hardware. It balances compensation accuracy, computational efficiency, and engineering feasibility, and can well meet the application requirements of precision RF test and measurement for wideband, high-precision, and low-distortion signal output. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the linearization method for a broadband power amplifier based on predistortion that relates to the present invention.

[0023] Figure 2 This is a schematic diagram of the broadband thermal distortion factor variation in the broadband power amplifier linearization method based on predistortion involved in this invention.

[0024] Figure 3 This is a schematic diagram comparing the compensation effects of the predistortion-based broadband power amplifier linearization method involved in this invention. Detailed Implementation

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings and examples.

[0026] Example 1: This embodiment discloses a predistortion-based broadband power amplifier linearization method, referring to... Figure 1 This includes steps S1 to S5: S1. Collect the RF input baseband signal, the actual output baseband signal, and the die heat flow data, and preprocess them to obtain the initial data set.

[0027] It should be noted that this step eliminates time-domain deviations, acquisition noise, and abnormal data interference between the RF input baseband signal, the actual output baseband signal, and the die thermal flow data through multi-source synchronous acquisition and standardized preprocessing. This ensures a high degree of consistency in timing, amplitude, and validity of various data, thereby providing real and reliable observation samples for subsequent dynamic thermal memory effect modeling. At the same time, it reduces the redundant load of high-dimensional model calculations and provides a stable and reliable data foundation for the rapid real-time update of predistortion parameters.

[0028] Specifically, the baseband processing unit of the testing instrument synchronously acquires the RF input baseband signal and the actual output baseband signal of the broadband power amplifier. Simultaneously, a thermistor pre-installed on the broadband power amplifier chip substrate acquires the temperature data of the broadband power amplifier die in real time, i.e., die heat flow data. The units of the RF input baseband signal and the actual output baseband signal are both volts, and the unit of the die heat flow data is Kelvin. The system sampling time interval is... Set to a fixed value in the nanosecond range, for example. Nanosecond; The RF input baseband signal, the actual output baseband signal, and the die heat flow data are mapped to the same time domain index to complete high-precision timestamp alignment. Then, the RF input baseband signal, the actual output baseband signal, and the die heat flow data are sequentially processed by outlier removal, noise filtering, and missing value interpolation. The timestamp alignment, outlier removal, noise filtering, and missing value interpolation are all well-known technologies and will not be described in detail here.

[0029] Furthermore, a data truncation mechanism based on a sliding window is introduced: a fixed preset length is set. The length of the sliding window is set, and the sliding step is set to control the movement of the sliding window. A preset length is used. The settings need to take into account both the depth of the memory effect of the broadband power amplifier and the real-time requirements of the model calculation.

[0030] Specifically, in a preferred embodiment, for a typical broadband signal with a bandwidth of 100MHz, considering that the memory effect of a broadband power amplifier typically covers several to tens of symbol periods, a preset length is used... The sampling point is set to 1024. This length is sufficient to capture the historical dependencies of the RF input baseband signal envelope, while avoiding data redundancy and excessive computational burden for subsequent model training caused by an excessively large sliding window length. The sliding window has a preset length. The window size is determined by sliding backwards from the beginning of the data stream after timestamp alignment and preprocessing, with a sliding step size typically set to a preset length. Half of the data, or 512 sampling points, are used to ensure data overlap between adjacent sliding windows and improve the continuity of the data stream. Each time the sliding window slides, the continuous RF input baseband signal, the corresponding actual output baseband signal, and the time-synchronized die thermal flow data contained in the current sliding window are extracted as a valid data segment.

[0031] Specifically, by performing the above sliding window interception operation on all data streams, a data set consisting of multiple valid data segments is finally formed. Each valid data segment ensures that the RF input baseband signal, the actual output baseband signal, and the die heat flow data correspond one-to-one. All the above valid data segments together constitute the initial data set required for this step, providing high-quality data support for the accurate identification of the dynamic thermal memory effect in the future.

[0032] S2. Based on the energy difference between the RF input baseband signal and the actual output baseband signal in the initial data set, and the rate of change of the die heat flow data per unit time, the transient heat dissipation gradient corresponding to each sampling moment is obtained.

[0033] It should be noted that this step first obtains the theoretical linear gain of the broadband power amplifier, and then, based on the initial data set, combines the energy difference between the RF input baseband signal and the actual output baseband signal with the instantaneous temperature change rate of the die heat flow data to construct the transient heat dissipation gradient, providing key thermal characteristic parameters for accurate modeling of the dynamic thermal memory effect.

[0034] Specifically, the theoretical linear gain of the broadband power amplifier is first obtained through the following steps: When the broadband power amplifier is in its low-power linear operating range, a set of standard test baseband signals with known amplitudes is input to the broadband power amplifier through the baseband processing unit of the test instrument. Simultaneously, the standard test output baseband signal output by the broadband power amplifier is acquired. Within the low-power linear operating range, the broadband power amplifier exhibits no significant nonlinear distortion and conforms to linear operating characteristics. The ratio of the envelope amplitude of the standard test output baseband signal to the envelope amplitude of the standard test baseband signal is calculated, and the average value of multiple test results is taken as the theoretical linear gain of the broadband power amplifier. It is used for the accurate calculation of the energy difference between the input and output signals, and remains constant during the normal operation of the wideband power amplifier, so there is no need for repeated calculation.

[0035] It should be noted that the signal envelope refers to the instantaneous amplitude characteristic sequence obtained by performing modulus calculation on the quadrature components of the RF input baseband signal or the actual output baseband signal. It is used to characterize the instantaneous energy intensity of the signal at each sampling moment and belongs to the prior art, so it will not be elaborated here.

[0036] Furthermore, for all sampled data within each valid data segment in the initial dataset, within a set preset length... Within the corresponding sliding window range, a transient heat dissipation gradient calculation model is constructed, taking the first... Taking the sampling time as an example, we will calculate the first sampling time. The transient heat dissipation gradient at each sampling time is expressed as follows: ; In the formula, It is the first The transient heat dissipation gradient at each sampling time is used to characterize the intensity of the transient heating impact caused by the power loss and temperature change of the broadband power amplifier at the current time. It is an index of the sampling time; It is the first one in the sliding window The RF input baseband signal at each sampling point It is the first one in the sliding window The envelope of the RF input baseband signal at each sampling point; It is the first one in the sliding window The actual output baseband signal at each sampling point It is the first one in the sliding window The actual output baseband signal envelope of each sampling point; It is the index of the sampling point within the sliding window; It is the theoretical linear gain of a broadband power amplifier; It is the first Core heat flow data at each sampling time; It is the first Core heat flow data at each sampling time; It is the sampling time interval; It is the preset length of the sliding window; It is a constant to prevent the denominator from being zero, and its value ranges from 0.01 to 0.1. For example, .

[0037] Specifically, Reflecting the radio frequency input baseband signal at the first Input energy at each sampling point It is the ideal input energy after the actual output baseband signal is normalized according to the theoretical linear gain. The difference between the two characterizes the internal power loss of the broadband power amplifier at this sampling point due to nonlinear loss and thermal loss. This is the rate of change of the die heat flux data per unit time, directly reflecting the drastic instantaneous temperature change of the broadband power amplifier die; multiply the power loss term by the temperature change rate term, and then... (The sentence is incomplete and requires more context to translate accurately.) The transient heat dissipation gradient obtained by accumulating within the corresponding sliding window can dynamically and sensitively reflect the transient heating impact intensity caused by sudden load changes and temperature drift under high power output of broadband power amplifiers. This enables the accurate extraction of key features of dynamic thermal memory effect, providing core thermal feature basis for subsequent thermal sensing modeling and real-time parameter updates of predistortion model.

[0038] Furthermore, by performing the above calculations at each sampling time, the transient heat dissipation gradient corresponding to each sampling time can be finally obtained.

[0039] S3. Based on the transient thermal dissipation gradient corresponding to each sampling time, and combined with the broadband distribution characteristics of the RF input baseband signal corresponding to each sampling time, the broadband thermal distortion factor corresponding to each sampling time is calculated.

[0040] It should be noted that this step combines the acquired transient thermal dissipation gradient with the envelope statistical characteristics and signal operating bandwidth characteristics of the RF input baseband signal to construct a broadband thermal distortion factor. This factor accurately characterizes the degree of distortion at different frequency points within a broadband range affected by the dynamic thermal memory effect, providing frequency-domain related thermal distortion characteristics for the subsequent construction of the thermal sensing predistortion model.

[0041] Specifically, based on the set sliding window, determine the first... The sliding window containing the sampling time is defined, and the RF input baseband signal envelope of all sampling points within that sliding window is extracted. Statistical variance calculation is then performed on the variance of all extracted RF input baseband signal envelopes, and the result is the RF input baseband signal envelope. The statistical variance of the RF input baseband signal envelope within the sliding window of each sampling time is denoted as . It is used to characterize the intensity of amplitude modulation and bandwidth fluctuation characteristics of the RF input baseband signal within the current sliding window.

[0042] Furthermore, before the broadband power amplifier test begins, based on the test requirements and system design parameters, the physical cross-band bandwidth limit of the RF input baseband signal is preset using the parameter configuration module of the test instrument. The physical cross-band bandwidth limit The test system has fixed configuration parameters, corresponding to the maximum frequency range that the RF input baseband signal can cover. These parameters remain constant throughout the test and do not require repeated settings; for example, the physical cross-band bandwidth limit. .

[0043] Specifically, before the broadband power amplifier test begins, the initial ambient temperature of the test environment is measured and used as the preset ambient reference temperature. The unit is Kelvin, used to subsequently eliminate the influence of ambient temperature reference differences on the calculation of broadband thermal distortion factor; if the test ambient temperature does not fluctuate significantly. The measurement should remain constant throughout the test; if the ambient temperature changes significantly, the measurement can be repeated and updated. For example, .

[0044] It should be noted that the broadband distribution characteristics of the RF input baseband signal at each sampling time specifically include: the statistical variance of the RF input baseband signal envelope within the sliding window of each sampling time, the physical cross-band bandwidth limit set by the test system, and the preset ambient temperature. These three types of parameters, from the three dimensions of signal amplitude fluctuation characteristics, system operating bandwidth range, and ambient temperature reference, together constitute the broadband distribution characteristics of the RF input baseband signal, providing complete frequency domain and environmental parameter support for the calculation of the broadband thermal distortion factor.

[0045] Furthermore, based on the transient heat dissipation gradient corresponding to each sampling time, combined with the statistical variance of the RF input baseband signal envelope within the sliding window to which each sampling time belongs, the physical cross-band bandwidth limit, and the ambient reference room temperature, the broadband thermal distortion factor corresponding to each sampling time is obtained; taking the first... Taking the sampling time as an example, we will calculate the first sampling time. The broadband thermal distortion factor corresponding to each sampling time is expressed by the following formula: ; In the formula, It is the first The broadband thermal distortion factor corresponding to each sampling time is used to characterize the degree of broadband nonlinear distortion of the broadband power amplifier caused by the dynamic thermal memory effect at the current time. It has dimensions The normalized total coefficient; It is the first Transient heat dissipation gradient at each sampling time; It is the first The statistical variance of the envelope of the RF input baseband signal within the sliding window of each sampling time; It is the physical cross-frequency bandwidth limit; It is the ambient standard room temperature; Based on the natural constant The logarithmic function with base 0; It is an index of the sampling time; It is the absolute value symbol.

[0046] When the transient heat dissipation gradient increases rapidly, if the statistical variance of the RF input baseband signal envelope is large and the signal physical cross-band bandwidth limit is wide, it indicates that the broadband signal as a whole is under the combined effect of severe heat shock and amplitude fluctuation. The band edge components are more susceptible to nonlinear compression and distortion stretching caused by thermal memory effect. The broadband thermal distortion factor, through nonlinear mapping between the logarithmic and square root domains, can sensitively and accurately characterize the distortion depth affected by thermal memory effect within the broadband, especially at the band edge, and realize the quantitative characterization of broadband thermal nonlinear distortion.

[0047] It should be noted that, The dimensions are (Voltage squared) Kelvin / second); The dimensions are (The square of the voltage); The dimensions are (Kelvin); The dimension of is Hertz, that is Therefore, the fraction The remaining dimensions after calculation are (Voltage to the fourth power), therefore, to avoid the problem of inconsistent dimensions in the relation, a normalized total coefficient is introduced. , Dimensional That is, the normalized total coefficient of the voltage to the negative fourth power, used to offset the dimensions generated by the calculation of the other variables in the formula. This achieves dimensional matching and numerical normalization of calculations within the logarithmic domain, thus balancing computational sensitivity and numerical stability.

[0048] Furthermore, by calculating for each sampling time, the broadband thermal distortion factor corresponding to each sampling time can be finally obtained. Please refer to [link to relevant documentation]. Figure 2 As shown, Figure 2This is a schematic diagram of the change in the broadband thermal distortion factor. As shown in the figure, the broadband thermal distortion factor exhibits dynamic changes over time that are highly correlated with the die heat flow data and transient heat dissipation gradient. Its peak time strictly corresponds to the peak time of temperature change and transient heat dissipation gradient, fully demonstrating the design concept of this invention that deeply integrates thermal characteristics with the broadband distribution characteristics of the RF input baseband signal. When the broadband power amplifier is in the stage of sudden pulse output or rapid die temperature rise, the broadband thermal distortion factor increases significantly, accurately reflecting the intensification of nonlinear distortion caused by the enhanced dynamic thermal memory effect. In the stage of stable temperature and small fluctuation of RF input baseband signal amplitude, the broadband thermal distortion factor remains at a low level, which matches the good linearity of the broadband power amplifier. Its fluctuation amplitude is positively correlated with the intensity of transient heat dissipation gradient and statistical variance of RF input baseband signal envelope, which is completely in line with the core logic of this invention of coupling and quantifying thermal distortion with heat dissipation gradient and signal broadband distribution characteristics. This provides accurate and real-time thermal distortion characteristic basis for the subsequent dynamic tracking and compensation of generalized memory polynomial cross term coefficients.

[0049] S4. Based on the broadband thermal distortion factor corresponding to each sampling time, the cross term coefficients of the generalized memory polynomial are dynamically tracked and compensated to obtain the updated coefficients of the generalized memory polynomial after dynamic tracking and compensation.

[0050] It should be noted that this step, based on the broadband thermal distortion factor corresponding to each sampling time, introduces a high-frequency, short-period fast tracking mechanism on the basis of traditional long-period full matrix coefficient update. The broadband thermal distortion factor is used as a dynamic correction factor to quickly compensate and correct the cross-term coefficients of the generalized memory polynomial, thereby realizing the dynamic update of the model coefficients. Finally, the updated coefficients after dynamic tracking compensation are obtained, providing low-latency and high-dynamic coefficient support for the nonlinear predistortion of broadband power amplifiers.

[0051] Specifically, setting It is an index of the signal envelope delay memory depth, used to characterize the delay order of the current sampling moment due to the influence of the historical signal envelope, based on the memory effect depth of the broadband power amplifier and the preset length of the sliding window. Confirmed; for example, combined with the preset length of the sliding window. Each sampling point is set. The value range is from 0 to 32, for example, This corresponds to a medium-delayed memory depth, which can both cover the historical impact of hot memory effects and avoid model redundancy; [Setting...] It is an index of the signal envelope lead memory depth, used to characterize the lead order of the influence of the current sampling moment on the future signal envelope, and is related to the index of the signal envelope delay memory depth. Together they constitute the memory span of the memory polynomial; setting The value range is from 0 to 16, for example, Index of signal envelope delay memory depth Matching is used to form a reasonable memory span, balancing model sensitivity and computational cost; settings are made. It is the nonlinear order of the generalized memory polynomial, used to characterize the fitting order of the model to the nonlinear distortion of the broadband power amplifier, and is preset according to the amplifier's nonlinear characteristics and compensation accuracy requirements; for example, Values ​​range from 3 to 5, with preference given to the latter. This reduces model complexity while maintaining fitting accuracy.

[0052] Further, settings It is the memory time span corresponding to the current cross term, and is an index of the memory depth delayed by the signal envelope. Indexing with signal envelope advance memory depth The difference is calculated, and then the absolute value is taken to obtain the value, which is used to characterize the time span of the memory effect corresponding to the current cross term. The larger the time span, the more sensitive the cross term is to long-term memory effect and hot memory effect; combined with the above... , Examples of possible values, It corresponds to a medium memory time span, has moderate sensitivity to thermal memory effects, and is suitable for the dynamic thermal characteristics of broadband power amplifiers.

[0053] Specifically, before the broadband power amplifier test begins, the maximum allowable distortion tolerance constant of the system is preset according to the performance indicators and compensation accuracy requirements of the test system. The maximum distortion tolerance constant Fixed configuration parameters are set for the system to limit the magnitude of coefficient correction and ensure the stability of model operation; for example, the maximum allowable distortion tolerance constant is... , is a dimensionless parameter.

[0054] Furthermore, based on the broadband thermal distortion factors acquired at each sampling time, a generalized memory polynomial coefficient dynamic tracking compensation model is constructed, with the first... Taking a single sampling time as an example, the cross-term coefficients of the generalized memory polynomial are updated through rapid compensation. The updated coefficients of the generalized memory polynomial after dynamic tracking compensation are calculated, and the relationship is as follows: ; In the formula, These are the update coefficients of the generalized memory polynomial after dynamic tracking compensation, used to characterize the model cross term weights after thermal distortion adaptive correction; These are the original cross-term coefficients of the generalized memory polynomial obtained after updating the full matrix coefficients of the previous long period; It is an index of the signal envelope delay memory depth; It is an index of the signal envelope's advance memory depth; It is the nonlinear order of the generalized memory polynomial; It is an index of the sampling time; It is the memory time span corresponding to the current cross item; It is the first Broadband thermal distortion factor corresponding to each sampling time; It is the maximum distortion tolerance constant; It is the hyperbolic tangent function.

[0055] This compensation mechanism eliminates the need for high-dimensional matrix inversion operations, achieving rapid model coefficient updates solely through scalar operations. This effectively reduces computational complexity and overcomes the computational bottleneck of traditional generalized memory polynomials. (The last sentence appears to be incomplete and possibly refers to a different mechanism or method.) The hyperbolic tangent function increases with the increase of broadband power amplifier die temperature and the enhancement of thermal memory effect. The output is a dynamic correction value between 0 and 1, achieving smooth and stable correction of the model coefficients; at the same time, it remembers the time span. The larger the value, the more sensitive the corresponding cross-term coefficient is to the thermal memory effect. As an internal weighting term, it enables higher-order, long-memory cross-term coefficients that are more sensitive to thermal changes to obtain stronger compensation and correction, thereby accurately and adaptively tracking thermally induced nonlinear distortion changes and ensuring that the updated model coefficients can match the current thermal memory effect state.

[0056] It should be noted that by performing the above-mentioned rapid coefficient compensation update at high frequency during the intervals of traditional long-cycle full coefficient updates, the model coefficients can still follow the temperature drift and thermal memory distortion trajectory in real time between two full matrix updates, which greatly improves the dynamic response speed and adaptation accuracy of the model coefficients.

[0057] Furthermore, by performing the above-mentioned dynamic tracking compensation on the coefficients at each sampling time, the coefficients of the generalized memory polynomial model can be updated in real time, and the updated coefficients after dynamic tracking compensation at each sampling time can be obtained, forming a set of updated coefficients.

[0058] S5. Based on the updated coefficients after dynamic tracking compensation at each sampling time, the predistortion signal is reconstructed to achieve high linearity output of the wideband power amplifier.

[0059] It should be noted that, based on the updated set of generalized memory polynomials obtained after dynamic tracking compensation, and combined with the real-time input wideband RF baseband signal, the predistortion signal is reconstructed through generalized memory polynomial operations, thereby offsetting the thermal nonlinear distortion of the wideband power amplifier, and finally achieving high linearity high-power signal output, thus completing the entire predistortion compensation process for the dynamic thermal memory effect of the wideband power amplifier.

[0060] Specifically, the updated coefficient set of the generalized memory polynomial after dynamic tracking compensation corresponding to each sampling time is substituted into the generalized memory polynomial operation engine of the test instrument one by one; wherein, the updated coefficient set has been compensated by dynamic tracking of thermal distortion, which can accurately match the current dynamic thermal memory effect state of the broadband power amplifier, and provide reliable coefficient support for the accurate reconstruction of the predistortion signal.

[0061] Furthermore, the generalized memory polynomial operation engine calls the aforementioned updated coefficient set and performs an inner product operation with the real-time acquired wideband RF input baseband signal. By performing inverse fitting and compensation on the nonlinear distortion of the RF input baseband signal, a pre-distortion cancellation signal is generated that can offset the thermally induced nonlinear distortion of the wideband power amplifier. The signal format and bandwidth range of the pre-distortion cancellation signal are consistent with the RF input baseband signal to ensure compatibility with the RF analog link.

[0062] Specifically, the generated pre-distortion cancellation signal is sent to the RF analog link of the test instrument. After pre-processing by the link, it is input to the wideband power amplifier for high-power amplification. Since the pre-distortion cancellation signal has pre-compensated for the nonlinear distortion caused by the dynamic thermal memory effect of the wideband power amplifier, it can effectively cancel the amplitude and phase frequency distortion caused by die temperature drift and thermal memory effect during the operation of the amplifier. Finally, the test instrument can achieve wideband, high-precision, and low-distortion signal output, which meets the high linearity requirements of high-power wideband RF testing.

[0063] Please see Figure 3 As shown, Figure 3The comparison chart shows the compensation effects. As can be seen, the uncompensated original output signal exhibits significant amplitude fluctuations and nonlinear distortion, with a chaotic waveform that deviates severely from the ideal linear output reference. This distortion is further aggravated during the thermal shock phase of sudden temperature changes in the die. While the traditional GMP-compensated output signal reduces distortion to some extent, its reliance on long-period full-matrix coefficient updates prevents real-time tracking of transient changes in the dynamic thermal memory effect, resulting in significant residual distortion and insufficient waveform stability during rapid temperature drift. In contrast, the amplitude fluctuations of the compensated output signal from this method are significantly suppressed, resulting in a smooth, regular waveform that closely adheres to the ideal linear output reference. Even at the moment of temperature change corresponding to the peak of the thermal dissipation gradient, it maintains excellent linearity and stability. The comparison clearly demonstrates that this invention, by dynamically adjusting the cross-term coefficients of the generalized memory polynomial through a wideband thermal distortion factor, achieves precise and real-time cancellation of nonlinear distortion caused by the dynamic thermal memory effect. Its compensation effect is significantly superior to both the traditional uncompensated method and the traditional GMP compensation method, effectively improving the output linearity and long-term operational stability of the wideband power amplifier under high-power conditions.

[0064] This invention also discloses a predistortion-based broadband power amplifier linearization system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a predistortion-based broadband power amplifier linearization method according to the present invention.

[0065] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for linearizing a broadband power amplifier based on predistortion, characterized in that, include: The RF input baseband signal, actual output baseband signal, and die heat flow data of the broadband power amplifier are collected and preprocessed to obtain an initial dataset. Based on the energy difference between the RF input baseband signal and the actual output baseband signal in the initial data set, and the rate of change of the die heat flow data per unit time, the transient heat dissipation gradient corresponding to each sampling moment is obtained. Based on the transient thermal dissipation gradient corresponding to each sampling time, and combined with the broadband distribution characteristics of the RF input baseband signal corresponding to each sampling time, the broadband thermal distortion factor corresponding to each sampling time is calculated. Based on the broadband thermal distortion factor corresponding to each sampling time, the cross term coefficients of the generalized memory polynomial are dynamically tracked and compensated to obtain the updated coefficients of the generalized memory polynomial after dynamic tracking compensation. Based on the updated coefficients after dynamic tracking compensation corresponding to each sampling time, the predistortion signal is reconstructed to achieve high linearity output of the broadband power amplifier.

2. The method for linearizing a broadband power amplifier based on predistortion according to claim 1, characterized in that, The process of obtaining the initial data set includes: mapping the RF input baseband signal, the actual output baseband signal, and the die thermal flow data to the same time domain index to achieve high-precision timestamp alignment; then sequentially performing outlier removal, noise filtering, and missing value interpolation; introducing a sliding window-based data truncation mechanism, setting a fixed preset length as the length of the sliding window, and setting a sliding step size to control the movement of the sliding window, wherein the sliding step size is half of the preset length; and truncating the preprocessed data stream into multiple valid data segments through the sliding window, wherein the valid data segments form the initial data set.

3. The method for linearizing a broadband power amplifier based on predistortion according to claim 2, characterized in that, The process of obtaining the transient heat dissipation gradient corresponding to each sampling time includes: ； In the formula, It is the first Transient heat dissipation gradient at each sampling time; It is an index of the sampling time; , These are the first two in the sliding window. The RF input baseband signal and the RF input baseband signal envelope at each sampling point; , These are the first two in the sliding window. The actual output baseband signal and the actual output baseband signal envelope at each sampling point; It is the index of the sampling point within the sliding window; It is the theoretical linear gain of a broadband power amplifier; It is the first Core heat flow data at each sampling time; It is the first Core heat flow data at each sampling time; It is the sampling time interval; It is the preset length of the sliding window; It is a constant to prevent the denominator from being zero.

4. The method for linearizing a broadband power amplifier based on predistortion according to claim 3, characterized in that, The method for obtaining the theoretical linear gain of the broadband power amplifier includes: When the wideband power amplifier is in the low-power linear operating range, a set of standard test baseband signals with known amplitudes are input to the wideband power amplifier through the baseband processing unit of the test instrument, and the standard test output baseband signal output by the wideband power amplifier is acquired simultaneously; the ratio of the envelope amplitude of the standard test output baseband signal to the envelope amplitude of the standard test baseband signal is calculated, and the average value of multiple test results is taken as the theoretical linear gain.

5. The method for linearizing a broadband power amplifier based on predistortion according to claim 2, characterized in that, The broadband distribution characteristics of the RF input baseband signal corresponding to each sampling time include: the statistical variance of the RF input baseband signal envelope within the sliding window to which each sampling time belongs, the physical cross-band bandwidth limit set by the test system, and the preset ambient temperature.

6. The method for linearizing a broadband power amplifier based on predistortion according to claim 2, characterized in that, The process of obtaining the broadband thermal distortion factor corresponding to each sampling time includes: ； In the formula, It is the first Broadband thermal distortion factor corresponding to each sampling time; It has dimensions The normalized total coefficient; It is the first Transient heat dissipation gradient at each sampling time; It is the first The statistical variance of the envelope of the RF input baseband signal within the sliding window of each sampling time; It is the physical cross-frequency bandwidth limit; It is the ambient standard room temperature; Based on the natural constant The logarithmic function with base 0; It is an index of the sampling time; It is the absolute value symbol.

7. The method for linearizing a broadband power amplifier based on predistortion according to claim 1, characterized in that, The dynamic tracking and compensation of the cross-term coefficients of the generalized memory polynomial to obtain the updated coefficients of the generalized memory polynomial after dynamic tracking and compensation includes: ； In the formula, These are the update coefficients of the generalized memory polynomial after dynamic tracking and compensation; These are the original cross-term coefficients of the generalized memory polynomial obtained after updating the full matrix coefficients of the previous long period; It is an index of the signal envelope delay memory depth; It is an index of the signal envelope's advance memory depth; It is the nonlinear order of the generalized memory polynomial; It is an index of the sampling time; It is the memory time span corresponding to the current cross item; It is the first Broadband thermal distortion factor corresponding to each sampling time; It is the maximum distortion tolerance constant; It is the hyperbolic tangent function.

8. The method for linearizing a broadband power amplifier based on predistortion according to claim 7, characterized in that, The index of the signal envelope delay memory depth Index of signal envelope advance memory depth Nonlinear order of generalized memory polynomials and the span of memory The setup methods include: Index of signal envelope delay memory depth The value is determined based on the memory effect depth of the broadband power amplifier, ranging from 0 to 32; and The matching constitutes the memory span range of the memory polynomial, with values ​​ranging from 0 to 16; The value is preset to 3 to 5 based on the nonlinear characteristics and compensation accuracy requirements of the broadband power amplifier; Depend on and The result is obtained by taking the absolute value after calculating the difference.

9. The method for linearizing a broadband power amplifier based on predistortion according to claim 1, characterized in that, The reconstructed predistortion signal enables the high linearity output of the broadband power amplifier, including: The updated set of coefficients of the generalized memory polynomial after dynamic tracking compensation is substituted one by one into the generalized memory polynomial operation engine of the test instrument; the updated set of coefficients is called and the inner product operation is performed with the real-time acquired wideband RF input baseband signal to generate a predistortion cancellation signal. The signal format and bandwidth range of the predistortion cancellation signal are consistent with the RF input baseband signal; the predistortion cancellation signal is sent to the RF analog link of the test instrument for preprocessing, and then input to the wideband power amplifier for high-power amplification to achieve high linearity output of the wideband power amplifier.

10. A broadband power amplifier linearization system based on predistortion, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a predistortion-based broadband power amplifier linearization method according to any one of claims 1-9.

Images