Spectrum Control Method of Pulse Modulation Waveform for All-Solid-State Transmitter

Abstract

This application relates to the field of intelligent spectrum control and specifically discloses a spectrum control method for pulse-modulated waveforms in all-solid-state transmitters. This method generates an ideal waveform and performs pre-distortion processing. After that, the actual output spectrum of the power amplifier is compared with the target spectrum template at a deep, fine-grained level. Using spectrum sensing and difference reasoning, it accurately quantifies spectrum deviation and forms an informative spectrum leakage index. Based on this index, the system dynamically optimizes the DPD coefficient and adjusts the output spectrum in a closed-loop manner to achieve precise suppression of spectrum leakage, ensuring that the signal strictly complies with the spectrum template requirements, thereby significantly improving transmitter performance and compatibility.

Description

Technical Field

[0001] The present application relates to the field of intelligent spectrum control, and more specifically, to a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter. Background Art

[0002] All-solid-state transmitters, owing to their high efficiency, excellent reliability, and compact design, have found widespread application in modern communications, radar, and other fields. However, to achieve higher efficiency, the power amplifiers in these transmitters, particularly all-solid-state power amplifiers, often operate in a nonlinear region. This nonlinearity significantly impacts the pulsed waveforms amplified by them. Ideally, pulsed waveforms, due to their discontinuous nature and rapid rise and fall edges, possess wide spectral bandwidths. However, the nonlinearity of the power amplifier introduces additional distortion, further broadening the spectrum and causing so-called spectral leakage. This spectral leakage refers to the spread of energy into adjacent frequency bands, which not only reduces the spectral efficiency of the transmitted signal but also interferes with other systems operating in adjacent frequency bands, potentially degrading performance or even causing system failure. This violates strict spectrum management regulations and system compatibility requirements. Therefore, effectively controlling the spectral leakage of pulsed waveforms in all-solid-state transmitters to meet stringent spectral mask requirements remains a significant technical challenge.

[0003] To mitigate the spectral broadening caused by PA nonlinearity, digital predistortion (DPD) technology is currently commonly used in the industry. DPD introduces predistortion at the PA input that counteracts the PA's nonlinear characteristics, compensating for the distortion and improving the output signal's linearity and spectral purity. For continuous wave (CW) signals, DPD technology is relatively mature, with performance evaluation and coefficient adaptive adjustment often based on metrics such as the output signal's error vector magnitude (EVM) or adjacent channel power ratio (ACPR). However, for pulse-modulated waveforms, the inherent transient nature of traditional CW DPD methods can limit performance evaluation and fast and accurate coefficient adjustment, especially when meeting complex and stringent spectral mask requirements. Many existing DPD methods may struggle to directly and effectively optimize spectral mask compliance, or their feedback during performance evaluation and coefficient adjustment may not fully and accurately reflect the detailed differences between the actual output spectrum and the target spectral mask. This can lead to slow convergence of the DPD adaptive process or suboptimal spectral control, especially when spectral leakage characteristics are complex and distributed across multiple non-adjacent frequency bands.

[0004] Therefore, an optimized spectrum control scheme for pulse modulation waveforms of all-solid-state transmitters is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter, which generates an ideal waveform and performs pre-distortion processing, and then performs a deep and fine-grained semantic feature comparison between the actual output spectrum of the power amplifier and the target spectrum template. The spectrum perception and difference reasoning method is used to accurately quantify the spectrum deviation and form a spectrum leakage index with rich information. Based on this index, the system dynamically optimizes the DPD coefficient and adjusts the output spectrum in a closed loop to achieve precise suppression of spectrum leakage, ensuring that the signal strictly meets the spectrum template requirements, thereby significantly improving the transmitter performance and compatibility.

[0006] According to one aspect of the present application, a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter is provided, comprising:

[0007] Obtain pulse parameter requirements and spectrum mask requirements;

[0008] generating an ideal shaped baseband waveform based on the pulse parameter requirements and the spectrum template requirements;

[0009] Inputting the ideal shaped baseband waveform into a digital predistortion function defined by the current effective DPD model coefficients to obtain a predistorted baseband waveform;

[0010] After performing digital-to-analog conversion on the predistorted baseband waveform, the predistorted baseband waveform is passed through an all-solid-state power amplifier to obtain a power amplifier output baseband waveform;

[0011] Performing Fourier transform on the power amplifier output baseband waveform to obtain an actual output spectrum;

[0012] Comparing the actual output spectrum with a spectrum template to obtain a spectrum leakage index;

[0013] Based on the comparison between the spectrum leakage indicator and a preset threshold, evaluating the current DPD performance status to obtain a DPD performance status evaluation result;

[0014] Based on the DPD performance status evaluation result, the current effective DPD model coefficients are adaptively adjusted to obtain updated DPD model coefficients.

[0015] Compared to existing technologies, this application provides a spectrum control method for pulse-modulated waveforms in all-solid-state transmitters. This method generates an ideal waveform and performs pre-distortion processing. It then performs a deep, fine-grained semantic feature comparison between the actual output spectrum of the power amplifier and the target spectrum template. Using spectrum sensing and difference inference, it accurately quantifies spectrum deviations, generating an informative spectrum leakage index. Based on this index, the system dynamically optimizes the DPD coefficient and adjusts the output spectrum in a closed-loop manner, achieving precise suppression of spectrum leakage and ensuring that the signal strictly complies with the spectrum template requirements, thereby significantly improving transmitter performance and compatibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 Flowchart of a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter according to an embodiment of the present application;

[0018] Figure 2 Schematic diagram of data flow of a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter according to an embodiment of the present application;

[0019] Figure 3 A flowchart of a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter according to an embodiment of the present application for comparing the actual output spectrum with a spectrum template to obtain a spectrum leakage index;

[0020] Figure 4 A flowchart of the spectrum control method of the all-solid-state transmitter pulse modulation waveform according to an embodiment of the present application for performing fine-grained transfer inference aggregation on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference inference coding vector. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0022] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0023] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0024] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0025] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0026] All-solid-state transmitters are widely used due to their advantages. However, to improve efficiency, the power amplifiers often operate in a nonlinear region. This, combined with the characteristics of the pulse modulation waveform, leads to significant spectral broadening, also known as spectral leakage. Spectral leakage not only reduces efficiency but also interferes with adjacent systems and violates spectrum regulations. Therefore, effectively controlling it to meet strict spectral templates is a major challenge. Digital pre-distortion (DPD) technology is currently commonly used to compensate for nonlinearities. However, when processing pulse waveforms and meeting complex templates, traditional DPD methods based on simple metrics (such as ACPR) lack sufficient detailed and accurate feedback to reflect the detailed differences between the actual spectrum and the template. This results in poor DPD adaptive adjustment and ineffective suppression of complex spectral leakage.

[0027] To address the aforementioned technical issues, the technical solution of this application proposes a spectrum control method for pulse-modulated waveforms in all-solid-state transmitters. This solution addresses the spectrum leakage challenge faced by all-solid-state transmitters when processing pulse-modulated waveforms by constructing a high-precision adaptive digital predistortion (DPD) control system. The core concept of this system is to implement a refined spectrum compliance feedback mechanism to drive the optimization of DPD model coefficients. Specifically, the system first generates an ideal waveform based on given pulse and spectrum requirements and applies predistortion processing defined by the current DPD model. This signal is then converted digitally to analog and fed into an all-solid-state power amplifier (PA). Crucially, the system obtains the PA's actual output spectrum and, rather than simply calculating a simple metric, employs a "spectrum sensing" and "difference reasoning" approach. By analyzing and understanding the intrinsic semantic features of the actual output spectrum and the spectral template, the system performs in-depth, fine-grained comparative reasoning, accurately identifying and quantifying the degree and specific form of the actual spectrum's deviation from the template, thereby generating a rich spectrum leakage metric. This metric serves as the core feedback signal, accurately evaluating the current DPD operating status and its effectiveness in spectrum control. Finally, based on this precise evaluation result, the DPD model coefficients are adaptively adjusted in a targeted manner, forming a closed-loop optimization process. This allows the output spectrum of the power amplifier to be dynamically and precisely shaped, minimizing spectrum leakage and ensuring that the final signal strictly meets the stringent spectrum mask requirements, effectively improving the performance and compatibility of the transmitter.

[0028] In the technical solution of the present application, a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter is proposed. Figure 1 Flowchart of a spectrum control method for a pulse modulation waveform of an all-solid-state transmitter according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to the embodiment of the present application. Figure 1 and Figure 2As shown, according to an embodiment of the present application, the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter includes the following steps: S100, obtaining pulse parameter requirements and spectrum template requirements; S200, generating an ideal shaped baseband waveform based on the pulse parameter requirements and the spectrum template requirements; S300, inputting the ideal shaped baseband waveform into a digital predistortion function defined by the current effective DPD model coefficient to obtain a predistorted baseband waveform; S400, performing digital-to-analog conversion on the predistorted baseband waveform and passing it through an all-solid-state power amplifier to obtain a power amplifier output baseband waveform; S500, performing Fourier transform on the power amplifier output baseband waveform to obtain an actual output spectrum; S600, comparing the actual output spectrum with the spectrum template to obtain a spectrum leakage index; S700, evaluating the current DPD performance status based on a comparison between the spectrum leakage index and a preset threshold to obtain a DPD performance status evaluation result; S800, based on the DPD performance status evaluation result, adaptively adjusting the current effective DPD model coefficient to obtain an updated DPD model coefficient.

[0029] Specifically, in steps S100 and S200, pulse parameter requirements and spectral mask requirements are obtained, and based on these requirements, an ideal shaped baseband waveform is generated. It should be understood that since the nonlinearity of all-solid-state power amplifiers is the key cause of spectral leakage, DPD technology corrects distortion by applying predistortion at the PA input that is opposite to the PA's nonlinear characteristics. To achieve effective predistortion, it is first necessary to know the ideal output waveform (i.e., when using a completely linear PA) and its corresponding spectrum. This "ideal shaped baseband waveform" represents the baseband signal that the system expects to obtain without nonlinear distortion, and its spectral characteristics must strictly conform to the preset spectral mask requirements. Therefore, generating this ideal waveform provides a "linear reference point" for DPD operation and a "gold standard" or target for subsequent spectral comparisons and error calculations.

[0030] Specifically, in one specific example of this application, the system first obtains and analyzes input pulse parameter requirements (such as pulse width, rise / fall time, pulse repetition interval, etc.) and spectrum mask requirements (such as mainlobe width, out-of-band suppression level, etc.). These requirements together define the characteristics that an ideal pulse signal should possess in the time and frequency domains. Next, based on the pulse parameters and spectrum mask requirements, an appropriate pulse shaping filter or window function is designed or selected. Because the original rectangular pulse produces strong sidelobes in the frequency domain and does not meet the strict spectrum mask requirements, pulse shaping is required, such as using raised cosine roll-off, Gaussian shaping, or other customized window functions. The selected or designed shaping method should ensure that, under ideal linear conditions, the spectrum of the shaped pulse signal meets or exceeds the specifications specified by the spectrum mask, particularly in terms of out-of-band suppression. Then, based on the determined pulse parameters and shaping function, an ideal continuous-time baseband pulse signal is generated. This involves constructing a time-domain representation of the signal based on the pulse start time and duration, as well as the selected shaping waveform. Subsequently, the ideal continuous-time baseband pulse signal is digitized. According to the principles of digital signal processing, a continuous-time signal must be sampled at a rate higher than the Nyquist rate to obtain its discrete-time representation, namely, the ideally shaped baseband waveform. This sampling process converts the continuous signal into a series of digital samples, which constitute the digital input to the DPD system. Finally, if a pulse train is required (such as a pulse train in radar applications), the individual ideally shaped baseband pulses are temporally concatenated or superimposed (if interpulse modulation is present) according to the pulse repetition interval to form a complete sequence of ideally shaped baseband waveforms. This sequence serves as the ideal input signal for the subsequent DPD processing.

[0031] Specifically, in step S300, the ideal shaped baseband waveform is input into a digital predistortion function defined by the currently effective DPD model coefficients to obtain a predistorted baseband waveform. It should be understood that, due to the pursuit of efficiency, all-solid-state power amplifiers often operate in a nonlinear region, resulting in signal distortion and spectral broadening. Digital predistortion (DPD) technology processes the signal before it enters the power amplifier, restoring it to its ideal state as closely as possible after passing through the amplifier's nonlinear path, thereby suppressing distortion and spectral leakage. The purpose of this step is to calculate, based on the current nonlinear characteristics of the power amplifier (characterized by the DPD model coefficients), how the ideal baseband waveform needs to be "distorted" to offset the nonlinear effects of the amplifier, thereby obtaining the predistorted baseband waveform that is fed into the amplifier.

[0032] Specifically, in a specific example of the present application, the system first loads or obtains the currently valid DPD model coefficients. These coefficients are parameters of the predistortion function, and they characterize the degree to which the current DPD model fits the nonlinear characteristics of the power amplifier. These coefficients may be initial values ​​obtained through calibration when the system starts, or the latest values ​​updated based on feedback information during the adaptive process. Next, the ideal shaped baseband waveform generated in the previous step is used as input and sent to the digital predistortion function. The digital predistortion function is typically a digital signal processing module based on a memory polynomial model (MP), a generalized memory polynomial model (GMP), or other advanced models. These models can capture the nonlinear and memory effects of the power amplifier. The predistortion function calculates the corresponding predistortion output value based on the input ideal shaped baseband waveform sample value and the currently loaded DPD model coefficients.

[0033] Then, for each input ideal shaped baseband waveform sample, the predistortion function performs complex nonlinear operations based on its internal algorithm (e.g., polynomial calculation, table lookup, filtering, etc.) and the DPD model coefficients. For example, a DPD function based on the MP model might calculate various power terms of the input sample and their products with delayed samples, multiply these terms by the corresponding DPD coefficients, and sum them to generate a predistorted output sample. This calculation process is then repeated for all samples in the ideal shaped baseband waveform sequence, resulting in a baseband waveform sequence with the same length as the input sequence and containing predistortion information. This sequence is known as the predistorted baseband waveform. Ideally, when this predistorted baseband waveform is fed into the power amplifier (PA), the resulting nonlinear distortion precisely offsets the PA's inherent nonlinearity, making the PA's output signal as close to ideal as possible. Finally, the resulting predistorted baseband waveform sequence is fed into a subsequent digital-to-analog converter (DAC) for processing before being fed into the all-solid-state power amplifier (PSA).

[0034] Specifically, in step S400, the predistorted baseband waveform undergoes digital-to-analog conversion and is then passed through an all-solid-state power amplifier to obtain a power amplifier output baseband waveform. It should be understood that since the predistorted baseband waveform is a digital signal, while the all-solid-state power amplifier operates in the analog domain and its primary function is to amplify low-power input signals to the required transmit power level, the digital predistorted baseband waveform must first be converted to an analog signal before it can be fed to the power amplifier for high-power amplification. The purpose of this step is precisely to complete this digital-to-analog conversion and signal power amplification process, thereby generating the actual power amplifier output baseband waveform, which serves as the direct feedback source for subsequent spectrum analysis and DPD adaptive adjustment.

[0035] Specifically, in one specific example of this application, a digital predistorted baseband waveform is first fed into a digital-to-analog converter (DAC). The DAC's function is to convert discrete-time digital samples into continuous-time analog voltage or current signals. The DAC requires a precise clock to synchronize its conversion rate. Typically, the sampling rate is much higher than the bandwidth of the baseband signal to ensure conversion accuracy and facilitate subsequent filtering. The DAC then outputs a stepped analog signal containing high-frequency image components. To remove these undesirable image components and smooth the signal waveform, the DAC output typically passes through an analog low-pass filter. The filter's cutoff frequency is designed to preserve the baseband signal's valid spectral components while suppressing images above the Nyquist frequency. The filtered signal becomes a continuous, smooth analog baseband signal.

[0036] This analog baseband signal (typically a low-frequency or intermediate-frequency signal), after digital-to-analog conversion and filtering, then needs to be upconverted to the desired radio frequency (RF) carrier frequency. This is typically achieved using a mixer, which multiplies the baseband signal with a carrier signal generated by a local oscillator (LO) to generate an RF signal containing the baseband signal information. During the upconversion process, a bandpass filter may also be required to select the desired RF signal sidebands and suppress other undesirable mixing products. The upconverted RF signal is then fed into a solid-state power amplifier (SSPA). The SSPA is the core component of the transmitter, responsible for amplifying the low-power RF signal to a power level sufficient for long-distance transmission. A solid-state PA typically consists of a multi-stage transistor amplifier, whose design balances performance metrics such as efficiency, linearity, and power output. During this step, the PA receives the analog RF signal and amplifies it to produce a high-power RF output signal.

[0037] Finally, the amplified RF signal is obtained from the power amplifier's output. In some DPD systems, for feedback and analysis, a portion of the RF output signal needs to be coupled down and down-converted back to baseband. This is typically achieved through a coupler, mixer, and low-pass filter. The RF signal is mixed with the local oscillator signal, and then filtered to produce an analog baseband signal in the same frequency band as the original baseband signal. This down-converted analog baseband signal undergoes analog-to-digital conversion (ADC) to produce the digitized power amplifier output baseband waveform for subsequent spectrum analysis and DPD adaptive processing.

[0038] Specifically, in step S500, the baseband waveform output by the power amplifier is Fourier transformed to obtain the actual output spectrum. It should be understood that the spectrum is the core dimension for evaluating the quality of the transmitter output signal and determining whether it meets the spectrum template requirements. In addition, the nonlinearity of the all-solid-state power amplifier will lead to spectrum broadening and leakage, which is exactly the problem that needs to be suppressed by DPD technology. The baseband waveform output by the power amplifier is a time domain signal representation. It is difficult to intuitively judge the spectrum characteristics and the degree of compliance with the spectrum template by directly observing the time domain waveform. The Fourier transform can convert the time domain signal into a frequency domain representation, that is, the spectrum, thereby revealing the energy distribution of the signal at different frequency components. Therefore, the purpose of Fourier transforming the baseband waveform output by the power amplifier is to obtain the frequency characteristics of the output signal under the actual working state of the power amplifier, that is, the actual output spectrum, which serves as the basis for evaluating the DPD effect, discovering spectrum leakage problems and guiding DPD adaptive adjustment.

[0039] Specifically, the process of performing a Fourier transform on the PA output baseband waveform to obtain the actual output spectrum typically follows a standard digital signal processing workflow. The specific implementation is as follows: First, the system acquires a digitized PA output baseband waveform sequence, obtained through analog-to-digital conversion (ADC). This sequence is a discrete-time representation of a continuous-time analog signal at a specific sampling rate. Next, to perform an effective Fourier transform, the acquired baseband waveform sequence typically requires preprocessing. This may include removing DC offsets, applying a window function (such as a Hamming window or a Henning window) to reduce spectral leakage (the window function's spectral sidelobes can affect the results), and selecting a data segment of appropriate length for analysis (typically a sequence containing several complete pulse cycles). The window function is applied to smooth the edges of the data segment to avoid truncation-induced spectrum broadening.

[0040] Next, the preprocessed baseband waveform sequence is subjected to a discrete Fourier transform (DFT). More commonly, the efficient fast Fourier transform (FFT) algorithm is used. The FFT is an efficient algorithm for computing the DFT, rapidly converting discrete time-domain sequences into discrete frequency-domain sequences. The length of the input time-domain sequence (e.g., N samples) determines the frequency resolution and total frequency range of the output frequency-domain sequence. The FFT output is a set of complex numbers, each representing the amplitude and phase of a specific discrete frequency component. The amplitude and phase information is then extracted from the complex FFT output. The energy distribution of the spectrum is typically of interest, so the power spectrum or power spectral density is calculated for each frequency component. The power spectrum is typically calculated by calculating the square of the complex amplitude and may be normalized. To present the spectrum more intuitively, it is often converted to a power spectrum in decibels (dB). Finally, the calculated power spectrum data is plotted on a graph with frequency on the horizontal axis and power (usually in dBm or relative power) on the vertical axis, visualizing the actual output spectrum. This spectrum clearly shows the energy distribution of the amplifier output signal at different frequencies, including the main lobe, side lobes, and harmonics and intermodulation products caused by nonlinearity, all of which constitute the specific manifestations of spectral leakage. This actual output spectrum serves as input for subsequent comparison with the spectrum template, spectral feature extraction, and difference inference.

[0041] Figure 3 The flowchart of the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to the embodiment of the present application is as follows: Figure 3 As shown, according to the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to the embodiment of the present application, step S600 includes: S610, extracting spectrum features from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector; S620, performing fine-grained transfer inference aggregation on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference inference coding vector; S630, performing feature decoding on the spectrum semantic difference inference coding vector to obtain the spectrum leakage indicator.

[0042] Specifically, in step S610, spectrum features are extracted from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector. It should be understood that the nonlinearity of the power amplifier caused by the pursuit of efficiency in the all-solid-state transmitter is the main reason for the spectrum leakage of the pulse modulation waveform. This leakage not only reduces the spectrum efficiency, but also seriously interferes with the adjacent channel system. The existing DPD method based on simple indicators such as ACPR is difficult to accurately reflect the detailed differences between the actual output spectrum and the template when dealing with the transient characteristics of the pulse waveform and the complex and strict spectrum template, resulting in poor DPD adaptive effect and difficulty in achieving optimal spectrum control. Therefore, in order to solve this problem, in the technical solution of the present application, spectrum features are extracted from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector. That is, the original spectrum data may contain power values ​​at a large number of frequency points. Direct comparison of these values ​​is difficult to capture complex and important information such as the overall shape, texture, key peaks, notches, and sideband roll-off characteristics of the spectrum. In particular, in a specific example of this application, a spectral feature extractor based on a Transformer is used to process the raw spectral data. The purpose of this process is to analyze and extract the key patterns, structures, and intrinsic connections between the two spectral curves (the actual output spectrum and the spectral template), and encode them into a representative semantic feature encoding vector. The semantics here refers to the physical meaning and distortion characteristics behind the spectral data. For example, an abnormal increase in energy in a certain frequency band may correspond to a specific nonlinear product.

[0043] It is worth mentioning that by converting the actual output spectrum and spectrum template into this more abstract and information-dense semantic feature encoding vector, the subsequent comparison is no longer a simple point-to-point numerical comparison, but is based on these high-level feature vectors. These feature vectors can effectively compress the high-dimensional information of the original spectrum while retaining the structural features that are critical for evaluating spectrum compliance. This semantic feature-based encoding provides a high-quality input foundation for subsequent fine-grained transfer reasoning and aggregation analysis, enabling the system to more deeply understand the specific differences between the actual spectrum and the template in different aspects, thereby generating a more accurate and rich spectrum leakage indicator, ultimately serving a more effective and precise DPD coefficient adaptive adjustment process, and better achieving strict control of complex pulse waveform spectra.

[0044] Specifically, in a specific example of the present application, spectral features are extracted from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector, including: passing the actual output spectrum and the spectrum template through a Transformer-based spectral feature extractor respectively to obtain the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector.

[0045] Specifically, in step S620, fine-grained transfer inference aggregation is performed on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference inference coding vector. It should be understood that when all-solid-state transmitters process pulse-modulated waveforms, the nonlinearity of the power amplifier is a key factor leading to spectrum leakage. Traditional DPD feedback mechanisms, based on simple indicators, are unable to fully reflect the subtle differences between the actual output spectrum and the complex spectrum template, and are unable to effectively guide the precise adjustment of DPD coefficients to meet strict compliance requirements. The semantic feature coding vectors obtained after extracting the actual output spectrum and the spectrum template are essentially high-dimensional representations that condense spectral structural information. Although they are more abstract than the original spectrum data, accurately measuring the differences between the two and extracting effective information to guide DPD optimization remains a challenge. Directly comparing these high-dimensional vectors globally may overlook key local details and the interactions between features. Therefore, in the technical solution of the present application, fine-grained transfer inference aggregation is further performed on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference inference coding vector. That is, spectral leakage is not always uniform and global; it can result from specific distortion patterns caused by amplifier nonlinearities in different frequency regions. These local effects and their interactions are crucial for accurately diagnosing the problem and adjusting DPD. Therefore, through fine-grained transfer inference aggregation analysis, we can deeply model the interaction between the two sets of spectral semantic feature vectors, capturing their subtle differences, response patterns, or information transfer mechanisms at different "granularities" (corresponding to local intervals sorted by eigenvalue magnitude). The ordered arrangement based on eigenvalue magnitude eliminates the arbitrariness of the original input order, focusing the comparison on the intrinsic strength and relative distribution of features. Equal-granularity feature segmentation decomposes the problem into interactive analysis of multiple corresponding local regions, achieving a "fine-grained" understanding. The resulting spectral semantic difference inference encoding vector is no longer a simple difference quantization value, but rather integrates a deep understanding of the local interaction relationship between the actual output spectral semantics and the spectral template semantics within different feature strength intervals, and a comprehensive encoding of the global perspective of cross-interval dependencies. This vector comprehensively and compactly expresses the deep-level deviation patterns and structural differences of the actual spectrum relative to the template, providing high-quality input for subsequent feature decoding. This enables the generated spectrum leakage index to more accurately reflect the actual spectrum problem, making the adaptive adjustment of DPD more targeted and efficient, ultimately achieving high-precision control of the pulse modulation waveform spectrum of the all-solid-state transmitter, ensuring that the spectrum template requirements are strictly met.

[0046] Figure 4The present invention provides a flow chart of the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to the embodiment of the present application, which performs fine-grained transfer reasoning aggregation on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain the spectrum semantic difference reasoning coding vector. Figure 4 As shown, according to the spectrum control method of the pulse modulation waveform of the all-solid-state transmitter of the embodiment of the present application, step S620 includes: S621, ordering and equally granularly dividing the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a sequence of actual output spectrum local semantic feature ordered coding vectors and a sequence of spectrum template local semantic feature ordered coding vectors; S622, performing transfer response inference and information transmission on the sequence of actual output spectrum local semantic feature ordered coding vectors and the sequence of spectrum template local semantic feature ordered coding vectors to obtain the spectrum semantic difference inference coding vector.

[0047] Specifically, in an embodiment of the present application, step S621, the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are arranged in an ordered manner and divided into equal granularity to obtain a sequence of actual output spectrum local semantic feature ordered coding vectors and a sequence of spectrum template local semantic feature ordered coding vectors, including: the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are arranged in an ordered manner based on the size of the eigenvalues ​​to obtain the actual output spectrum semantic feature ordered arrangement coding vector and the spectrum template semantic feature ordered arrangement coding vector; the actual output spectrum semantic feature ordered arrangement coding vector and the spectrum template semantic feature ordered arrangement coding vector are divided into equal granularity features to obtain the sequence of the actual output spectrum local semantic feature ordered coding vector and the sequence of the spectrum template local semantic feature ordered coding vector.

[0048] More specifically, the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are arranged in an ordered manner based on the eigenvalue size to obtain the actual output spectrum semantic feature ordered arrangement coding vector and the spectrum template semantic feature ordered arrangement coding vector, which can be expressed as follows:

[0049]

[0050]

[0051] in, and represent the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector respectively, Indicates sorting of vector elements. and They respectively represent the ordered arrangement coding vector of the actual output spectrum semantic features and the ordered arrangement coding vector of the spectrum template semantic features.

[0052] It should be understood that although the original actual output spectrum semantic feature coding vector captures the semantic information of the spectrum, the order of its internal elements may be just a convention or by-product of the feature extraction process. This arbitrary arrangement order may introduce irrelevant variables and interfere with the network's learning and capture of truly meaningful spectrum semantic difference patterns. Therefore, in order to ensure that the subsequent difference reasoning process can focus on the intrinsic strength and distribution of the features themselves, rather than their arbitrary arrangement order in the original coding vector, in the technical solution of the present application, the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are ordered based on the size of the eigenvalues. In specific implementation, this step is sorted according to the numerical size (i.e., the eigenvalue size) of each feature element in the feature vector. For the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector, the system will respectively arrange the elements in their respective vectors in ascending or descending order according to their numerical values, thereby generating the corresponding actual output spectrum semantic feature ordered arrangement coding vector and spectrum template semantic feature ordered arrangement coding vector. This process introduces a normalized representation based on intrinsic numerical attributes. This operation is performed to remove any arbitrariness that may be introduced by the original feature order, and to endow the subsequent feature response encoding network with a certain degree of invariance or equivariance to the input order. By sorting the feature vectors by their intrinsic strength, the network can focus on the distribution of feature values, their relative magnitudes, and the combinations of features within different strength intervals, rather than their positions in the original unsorted vector.

[0053] More specifically, the ordered arrangement coding vector of the actual output spectrum semantic features and the ordered arrangement coding vector of the spectrum template semantic features are subjected to equal-granularity feature segmentation to obtain a sequence of ordered coding vectors of the actual output spectrum local semantic features and a sequence of ordered coding vectors of the spectrum template local semantic features, which can be expressed as follows:

[0054]

[0055]

[0056] in, represents the feature segmentation function, They represent the first, second, and third in the sequence of ordered encoding vectors of local semantic features of the actual output spectrum. and The actual output spectrum local semantic feature ordered encoding vector, They represent the first, second, and third in the sequence of ordered encoding vectors of local semantic features of spectrum templates. and The ordered encoding vector of local semantic features of the spectrum template.

[0057] It should be understood that even after the ordered arrangement based on the size of the eigenvalues, the feature vector is still a holistic representation, and the subsequent transfer response inference unit is designed to capture the complex interaction relationship between the local feature sets between the actual output spectrum semantics and the spectrum template semantics. Directly performing interaction modeling on the entire ordered vector is computationally intensive and difficult to effectively capture localized difference patterns. Therefore, it is necessary to decompose these global actual output spectrum semantic feature ordered arrangement encoding vectors and spectrum template semantic feature ordered arrangement encoding vectors into a series of smaller local feature segments corresponding to specific eigenvalue intervals, so as to provide structured input for subsequent local interaction modeling, thereby achieving a detailed analysis of the differences in the spectrum within different feature intensity intervals, and supporting the architectural characteristics of "fine-grained" transfer reasoning aggregation.

[0058] In specific implementation, this step takes as input the actual output spectral semantic feature ordered permutation encoding vector and the spectral template semantic feature ordered permutation encoding vector obtained in the previous step. These ordered vectors are then synchronously segmented along their feature dimensions. This segmentation is performed by decomposing the actual output spectral semantic feature ordered permutation encoding vector and the spectral template semantic feature ordered permutation encoding vector into multiple continuous sub-vector segments with the same dimensions, i.e., "equal granularity." This "granularity," or the length of each sub-vector segment, is an important hyperparameter that determines the refinement of subsequent local interaction modeling and the size of the local region. For example, if the original vector has 100 features and a granularity of 10 is selected, each ordered vector will be segmented into 10 sub-vector segments of length 10. The purpose of this operation is to convert the strength-sorted actual output spectral semantic feature ordered permutation encoding vector and the spectral template semantic feature ordered permutation encoding vector into a sequence of actual output spectral local semantic feature ordered permutation encoding vectors and a sequence of spectral template local semantic feature ordered permutation encoding vectors that can be processed in parallel or sequentially by the transfer response inference unit. Through this equal-granularity segmentation, the system divides the entire feature space into a series of ordered local regions, each representing a set of features within a specific intensity range. This allows for independent or correlated interaction modeling of pairs of local feature sets corresponding to the same intensity range (one from the actual spectrum, one from the template spectrum).

[0059] Specifically, in an embodiment of the present application, step S622 performs transfer response inference and information transfer on the sequence of the actual output spectrum local semantic feature ordered coding vectors and the sequence of the spectrum template local semantic feature ordered coding vectors to obtain the spectrum semantic difference inference coding vector, including: inputting each group of corresponding actual output spectrum local semantic feature ordered coding vectors and spectrum template local semantic feature ordered coding vectors in the sequence of the actual output spectrum local semantic feature ordered coding vectors and the sequence of the spectrum template local semantic feature ordered coding vectors into a transfer response inference unit to obtain a sequence of spectrum local semantic transfer response coding matrices; performing transfer response inference sequence transfer on the sequence of the spectrum local semantic transfer response coding matrices to obtain the spectrum semantic difference inference coding vector.

[0060] More specifically, each corresponding set of the sequence of the actual output spectrum local semantic feature ordered coding vectors and the sequence of the spectrum template local semantic feature ordered coding vectors is input into the transfer response inference unit to obtain a sequence of spectrum local semantic transfer response coding matrices, which can be expressed as follows:

[0061]

[0062] in, for and The spectrum local semantic transfer response encoding matrix between is matrix multiplication, for activation function, is the trainable modulation weight matrix.

[0063] It should be understood that the aforementioned equal-granularity segmentation has decomposed the overall spectral feature difference problem into a series of local difference analysis tasks corresponding to different feature intensity ranges. The nonlinear effects of all-solid-state amplifiers are complex and diverse, which may cause the actual output spectrum and the template spectrum to exhibit different correlations, dependencies, or "transfer" patterns in features across different frequency ranges or energy levels (for example, energy transfer to shoulders due to clipping, or sidelobe asymmetry caused by memory effects). Simply comparing the numerical differences of local vectors is insufficient to capture these complex, nonlinear local interaction patterns. Therefore, a specially designed module (transfer response inference unit) is required to deeply model each pair of local vectors corresponding to the same intensity range. This module accurately quantifies and encodes how the actual spectral features "respond" to or deviate from the template spectral features in that local region, capturing their inherent complex relationships and "transfer" mechanisms. This operation is performed to generate a series of spectral local semantic transfer response encoding matrices. Each spectral local semantic transfer response encoding matrix is ​​the output of a pair of local vectors processed by the transfer response inference unit. It is designed to encapsulate the rich and structured interaction information between the local features of the actual output spectrum and the local features of the spectral template within a specific feature intensity range. These spectrally localized semantic transfer response encoding matrices express the spatial measure of the transfer response space based on their row vectors. Their structure (e.g., element-to-element correlation strength) preserves more interaction details than simple scalar scores. In the practical scenario of spectrum control for all-solid-state transmitters, these matrices specifically encode information such as subtle deformation patterns of the mainlobe shape in the highest intensity region, deviation patterns from the template in the shoulder region (medium intensity), and specific spurious or noise floor elevation patterns in the far sidelobe region (low intensity).

[0064] More specifically, the sequence of the spectrum local semantic transfer response coding matrix is ​​transferred by transfer response inference sequence to obtain the spectrum semantic difference inference coding vector, which is expressed as follows:

[0065]

[0066]

[0067]

[0068] in, Indicates flattening the matrix into vector processing, is the spectrum local semantic transfer response encoding vector, for of norm, is the value of the natural exponential function with the natural constant e as the base, To transfer the response inference weighted operation, for encoder, Inferring an encoding vector for the spectral semantic difference.

[0069] It should be understood that while the aforementioned steps generate a deep encoding of local differences and interactions within each feature intensity range (i.e., a sequence of spectral local semantic transfer response encoding matrices), these local matrices are discrete and fail to capture dependencies and overall trends across different intensity ranges. The nonlinear effects of solid-state amplifiers are typically global, and distortions in different intensity regions are not completely independent but rather correlated. For example, mainlobe clipping can affect shoulder and sidelobe elevation. To accurately understand the overall spectral leakage pattern and infer useful metrics, it is necessary to effectively integrate these discrete local interaction information, capturing their sequential dependencies and forming a unified representation that comprehensively reflects the overall spectral semantic differences. This approach integrates local interaction information distributed across different feature intensity ranges, captures the evolution of these local interaction patterns with feature intensity, and learns the dependencies and global context across different intensity ranges. Through this sequential transfer and aggregation, the network is able to extract a comprehensive and concise understanding of the overall spectral semantic differences from the details of a series of local interactions in the spectral local semantic transfer response. For example, the network might learn that when the interaction patterns in the highest-intensity regions show differences of a particular shape, this is often accompanied by a particular pattern of power rise in the shoulders of the intermediate-intensity regions, and a certain asymmetry in the distal sidelobes of the lower-intensity regions.

[0070] Preferably, in a specific example of the present application, it can be seen that the segmentation granularity of the feature segmentation function, while affecting the local partition range of the spectrum local semantic transfer response coding matrix, will also directly affect the expression of the coupling relationship between the corresponding actual output spectrum local semantic feature ordered coding vector and the spectrum template local semantic feature ordered coding vector.

[0071] Since the spectrum local semantic transfer response encoding matrix is ​​the transfer response between two local regions, it actually uses its row vector as a benchmark to express the structured measure of the transfer response space, and the row vector dimension is also the representation of the local region scale mentioned above, if the local region scale, that is, the row vector length If it is introduced as a local size strength constraint, then the low-dimensional structure measurement expression of the spectrum local semantic transfer response encoding matrix is ​​F norm It should follow a Poisson-like relationship:

[0072]

[0073] That is, the row vector length The number of low-dimensional structure measures acting on the local region transfer response space as a local size strength constraint is Second-rate, express The factorial of . From this, we can solve the parameter .

[0074] Thus, the transfer-response interaction is introduced by and the mean expectation is In the case of a Poisson-like process, and Describing the edge connection representation of the spatial measure, the transfer response association probability between two local regions is further determined as:

[0075]

[0076] in, represents the two-norm of the vector, is an absolute value.

[0077] Then the transfer response association probability To local regional scale Iteratively adjust to obtain :

[0078]

[0079] That is, under the strict guarantee that the expected degree is In this case, the regularization constraint of the mean expected connection probability fluctuation of each row is used to determine the regularization of the overall structured measure, so that the coupling correlation information within the local transfer response relationship can avoid local overfitting and enhance the global expression ability of the sequence of the spectral local semantic transfer response encoding matrix.

[0080] Specifically, in step S630, the spectral semantic difference reasoning coding vector is feature decoded to obtain the spectral leakage indicator. Accordingly, in a specific example of the present application, the spectral semantic difference reasoning coding vector is feature decoded to obtain the spectral leakage indicator, including: passing the spectral semantic difference reasoning coding vector through a decoder-based spectral semantic difference reasoner to obtain an inference decoding value of the spectral leakage indicator. It should be understood that although the spectral semantic difference reasoning coding vector generated by the previous step contains rich difference semantic information, it is essentially a high-dimensional, abstract coding structure and is difficult to be directly used for feedback control of the DPD system. Therefore, the spectral semantic difference reasoning coding vector is further feature decoded to obtain the spectral leakage indicator. The local interaction details, cross-interval dependencies and global difference patterns implicit in the spectral semantic difference reasoning coding vector are reversely mapped through the decoding network, and the original high-dimensional abstract features are converted into a spectral leakage indicator with clear engineering significance. This indicator not only needs to reflect the overall intensity of spectrum leakage (such as the overall energy overflow), but also needs to reveal its distribution characteristics (such as the excess amplitude in a specific frequency band, the correlation between multi-band leakage) and possible causes of distortion (such as the characteristic response corresponding to certain nonlinear orders).

[0081] Specifically, the process of decoding the spectral semantic difference inference encoding vector through a decoder-based reasoner typically relies on a pre-trained neural network or other machine learning model. The implementation is as follows: First, the system obtains the spectral semantic difference inference encoding vector generated in the previous step. This is a compact representation of the interaction and difference patterns between the actual spectrum and the template at different semantic feature levels. Next, the spectral semantic difference inference encoding vector is fed as input to a decoder-based spectral semantic difference reasoner. This decoder is typically a deep neural network whose structure and parameters are trained offline using a large amount of data pairs containing different spectral leakage patterns and their corresponding real-world spectral leakage indicators. The decoder contains fully connected layers, recurrent layers (such as LSTM or GRU), or other network layers capable of processing sequence or vector inputs. Furthermore, the decoder processes the input encoding vector layer by layer and extracts features based on the complex nonlinear mapping relationships it has learned internally. This process can be understood as the decoder attempting to identify the underlying factors that lead to specific spectral leakage patterns from the encoding vector and associate them with the desired output indicators. For example, the encoding vector contains information about the intensity of differences in a local region (corresponding to a specific frequency range). The decoder learns that this intensity pattern is highly correlated with the degree of power exceedance within that frequency range. The final decoder layer typically outputs one or a set of numerical values ​​representing the inferred decoded values ​​of the spectral leakage index. These metrics can be diverse, such as the out-of-band suppression margin (or exceedance) for multiple specific frequency bands, total out-of-band leakage power, distortion metrics associated with specific nonlinear orders, or more complex spectral shape conformance scores. The decoder's output layer is designed based on the desired type and number of metrics (for example, one neuron per quantified metric). Finally, the system takes these numerical values ​​output by the decoder and uses them as the current inferred decoded values ​​of the spectral leakage index. These values ​​directly reflect the system's quantitative assessment of the degree of difference between the current power amplifier output spectrum and the template. These quantitative metrics are used as feedback signals for the DPD adaptive algorithm, guiding the algorithm to adjust the DPD model coefficients to reduce these values, thereby suppressing spectral leakage and bringing the actual output spectrum closer to the target template.

[0082] Specifically, in step S700, the current DPD performance status is evaluated based on a comparison between the spectrum leakage index and a preset threshold to obtain a DPD performance status evaluation result. It is worth noting that the DPD performance status evaluation results include excellent, critical, and degraded status. It should be understood that while the spectrum leakage index quantifies the difference between the actual output spectrum and the target template, simply knowing the difference is insufficient to guide effective DPD adaptive adjustments. The system needs to determine the current spectrum performance level: whether it meets requirements (excellent), approaches the performance boundary and requires caution (critical), or deviates significantly and requires significant adjustment (degraded). This hierarchical status evaluation provides more granular feedback, enabling the DPD adaptive algorithm to adopt the most appropriate adjustment strategy based on the current performance "health." For example, in excellent conditions, adjustments can be slowed or paused; in critical conditions, small or preventative adjustments can be made; and in degraded conditions, more aggressive or larger adjustments can be made. This improves the efficiency, stability, and convergence speed of the adaptive process, ultimately enabling more precise spectrum control and maintaining long-term compliance.

[0083] Specifically, the process of evaluating the DPD performance status based on the comparison between the spectrum leakage indicator and the preset threshold is implemented as follows: First, the system obtains the inferred decoded value of the spectrum leakage indicator calculated in the above steps. These indicators are a quantitative representation of the degree of deviation and shape of the actual output spectrum relative to the spectrum template. Next, the system accesses multiple sets of pre-set thresholds. Unlike a simple pass / fail judgment, in order to distinguish between the three states of excellent, critical, and degraded, it is usually necessary to set at least two thresholds for each or each group of key spectrum leakage indicators. These thresholds define the boundaries between different performance states. For example, you can set an "excellent threshold" (stricter than the template requirements) and a "degraded threshold" (corresponding to the boundary of the template requirements or slightly looser), and the state between the two is a "critical state."

[0084] The acquired spectrum leakage index values ​​are then compared against these preset multi-level thresholds. Comparison logic determines the threshold range within which each key indicator falls. For example, if a metric (such as the out-of-band suppression margin in a key frequency band) exceeds the "excellent threshold," performance in that area is excellent; if it lies between the "degraded threshold" and the "excellent threshold," performance is in the critical range; and if it falls below the "degraded threshold" (i.e., exceeding the mask requirements), performance has degraded. Subsequently, based on the comparison results of all key spectrum leakage indicators, the system determines the current overall DPD performance status using preset decision rules or state machine logic. These decision rules may be based on a "worst indicator determines the status" (i.e., if any key indicator is degraded, the overall status is degraded) or more complex logic such as weighted summation or expert systems. For example, if all indicators exceed the "excellent threshold," the status is considered excellent; if all indicators exceed the "degraded threshold," but at least one indicator falls between the "degraded threshold" and the "excellent threshold," the status is considered critical; and if at least one key indicator falls below the "degraded threshold," the status is considered degraded. Finally, the DPD performance status assessment result is output, clearly indicating whether the current system is in "excellent state," "critical state," or "degraded state." This assessment result serves as input to the subsequent adaptive adjustment module, guiding the algorithm to adopt differentiated coefficient update strategies based on different states, such as adjusting the learning rate, adjusting the regularization strength, or triggering a more aggressive calibration procedure.

[0085] Specifically, in step S800, based on the DPD performance status assessment results, the currently effective DPD model coefficients are adaptively adjusted to obtain updated DPD model coefficients. It should be understood that the nonlinear characteristics of a power amplifier are not constant and can drift due to various factors, such as temperature, supply voltage, component aging, and variations in the specific waveform characteristics being processed. Static DPD coefficients cannot effectively and continuously compensate for these dynamically changing distortions. By introducing an adaptive adjustment mechanism based on the DPD performance status assessment results, the system can perceive in real time the impact of changes in the power amplifier operating state on the spectrum (as reflected by the spectrum leakage index). Based on the assessment results (excellent, marginal, or degraded), the DPD model coefficients are intelligently adjusted, dynamically tracking changes in the power amplifier characteristics and ensuring that the transmitter output spectrum consistently meets strict mask requirements under various operating conditions. This adaptability is the fundamental guarantee for system robustness and long-term performance stability.

[0086] Specifically, in one specific example of this application, the system first obtains the DPD performance status assessment results generated in the previous steps (indicating whether the current status is excellent, critical, or degraded) and detailed spectrum leakage indicators. At the same time, the system also holds the currently used DPD model coefficient set.

[0087] Next, the system triggers the corresponding coefficient adjustment strategy based on the received DPD performance status evaluation results. This is a state-based decision-making process.

[0088] If the evaluation result is "Excellent," DPD compensation is effective, spectrum leakage is well below the threshold, and system performance is stable and meets requirements. In this case, a conservative strategy is typically adopted: maintain the current DPD model coefficients unchanged or make only very slow, small tracking adjustments (if very slow drift is required). This strategy aims to avoid unnecessary adjustments that could introduce oscillation or degrade performance.

[0089] If the evaluation result is "critical," DPD compensation is effective, but certain spectral leakage indicators are approaching or slightly exceeding preset critical thresholds, posing a potential risk of performance degradation. At this point, the system initiates an online fine-tuning algorithm. This algorithm leverages detailed spectral leakage indicators (specifically, information indicating which frequency bands or distortion modes are approaching or exceeding critical thresholds) to make targeted, small adjustments to the DPD model coefficients that significantly impact those specific spectral regions or distortion modes. The direction and magnitude of these adjustments are calculated based on the changing trends of the spectral indicators by an optimization algorithm (e.g., gradient descent, least mean squares (LMS) variants, or recursive least squares (RLS) variants, but with a small step size). The goal is to accurately suppress excessive or critical spectral components back to a safe range to avoid triggering further degradation. At this stage, additional amplifier operating status information (such as temperature and voltage) and pre-trained predictive models can optionally be combined to provide auxiliary coefficient pre-tuning.

[0090] If the assessment result is "Degraded," the current DPD compensation is ineffective or severely insufficient, one or more key spectral leakage indicators have significantly exceeded degradation thresholds, and system performance cannot meet basic requirements. In this case, fine-tuning is insufficient to correct the problem, and a more comprehensive calibration is required. The system will issue a recalibration request signal, notifying the control system or operator to initiate a complete DPD coefficient identification or calibration process. This process may require suspending normal transmission, injecting special calibration signals, collecting large amounts of input and output data, and running complex identification algorithms (e.g., based on iterative least squares or deep learning) in the background or in a dedicated identification module to recalculate a new set of DPD model coefficients that effectively compensate for the current power amplifier nonlinearities. While waiting for recalibration to complete, the system may implement temporary strategies to temporarily mitigate losses or maintain basic spectral compliance, such as applying critical fine-tuning logic or falling back to a set of pre-stored, known, and relatively safe backup coefficients.

[0091] Finally, based on the adjustment strategies implemented in different states, the system generates updated DPD model coefficients (which may be the same as the current coefficients or new coefficients after fine-tuning or re-identification). These updated coefficients are loaded into the DPD module and applied to the ideal shaped baseband waveform in the next processing cycle, thus forming a new predistorted baseband waveform, continuing the entire closed-loop control process to continuously optimize and maintain the transmitter's spectral performance.

[0092] In summary, the spectrum control method for a pulse-modulated waveform in an all-solid-state transmitter according to the embodiments of this application is illustrated. After generating an ideal waveform and performing pre-distortion processing, the method performs a deep, fine-grained semantic feature comparison between the actual output spectrum of the power amplifier and the target spectrum template. Using spectrum sensing and difference reasoning, the method accurately quantifies spectrum deviations, forming an informative spectrum leakage index. Based on this index, the system dynamically optimizes the DPD coefficient and adjusts the output spectrum in a closed-loop manner to achieve precise suppression of spectrum leakage, ensuring that the signal strictly complies with the spectrum template requirements, thereby significantly improving transmitter performance and compatibility.

[0093] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A spectrum control method for a pulse modulation waveform of an all-solid-state transmitter, characterized in that: include: Obtain pulse parameter requirements and spectrum mask requirements; generating an ideal shaped baseband waveform based on the pulse parameter requirements and the spectrum template requirements; Inputting the ideal shaped baseband waveform into a digital predistortion function defined by the current effective DPD model coefficients to obtain a predistorted baseband waveform; After performing digital-to-analog conversion on the predistorted baseband waveform, the predistorted baseband waveform is passed through an all-solid-state power amplifier to obtain a power amplifier output baseband waveform; Performing Fourier transform on the power amplifier output baseband waveform to obtain an actual output spectrum; Comparing the actual output spectrum with the spectrum template to obtain a spectrum leakage index includes: Extracting spectrum features from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector; Performing fine-grained transfer reasoning aggregation on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference reasoning coding vector; Performing feature decoding on the spectrum semantic difference reasoning encoding vector to obtain the spectrum leakage indicator; Based on the comparison between the spectrum leakage indicator and a preset threshold, evaluating the current DPD performance status to obtain a DPD performance status evaluation result; Based on the DPD performance status evaluation result, the current effective DPD model coefficients are adaptively adjusted to obtain updated DPD model coefficients.

2. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 1, characterized in that: Extracting spectrum features from the actual output spectrum and the spectrum template to obtain an actual output spectrum semantic feature coding vector and a spectrum template semantic feature coding vector, including: passing the actual output spectrum and the spectrum template through a Transformer-based spectrum feature extractor respectively to obtain the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector.

3. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 2, characterized in that: Performing fine-grained transfer inference aggregation on the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector to obtain a spectrum semantic difference inference coding vector, including: The actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are arranged in order and divided into equal granularity to obtain a sequence of actual output spectrum local semantic feature ordered coding vectors and a sequence of spectrum template local semantic feature ordered coding vectors; Transfer response inference and information transfer are performed on the sequence of ordered coding vectors of local semantic features of the actual output spectrum and the sequence of ordered coding vectors of local semantic features of the spectrum template to obtain the spectrum semantic difference inference coding vector.

4. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 3, characterized in that: The actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector are arranged in an orderly manner and divided into equal granularity to obtain a sequence of actual output spectrum local semantic feature ordered coding vectors and a sequence of spectrum template local semantic feature ordered coding vectors, including: Performing an ordered arrangement of the actual output spectrum semantic feature coding vector and the spectrum template semantic feature coding vector based on the size of the eigenvalues ​​to obtain an actual output spectrum semantic feature ordered arrangement coding vector and a spectrum template semantic feature ordered arrangement coding vector; The ordered arrangement coding vector of the actual output spectrum semantic features and the ordered arrangement coding vector of the spectrum template semantic features are subjected to equal-granularity feature segmentation to obtain a sequence of the ordered coding vectors of the actual output spectrum local semantic features and a sequence of the ordered coding vectors of the spectrum template local semantic features.

5. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 4, characterized in that: Performing transfer response inference and information transfer on the sequence of the actual output spectrum local semantic feature ordered coding vectors and the sequence of the spectrum template local semantic feature ordered coding vectors to obtain the spectrum semantic difference inference coding vector, including: Inputting each corresponding set of the sequence of the actual output spectrum local semantic feature ordered coding vectors and the sequence of the spectrum template local semantic feature ordered coding vectors into a transfer response inference unit to obtain a sequence of spectrum local semantic transfer response coding matrices; The sequence of the spectral local semantic transfer response coding matrices is transferred by transfer response reasoning sequence to obtain the spectral semantic difference reasoning coding vector.

6. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 5, characterized in that: Feature decoding is performed on the spectral semantic difference inference coding vector to obtain the spectral leakage index, including: passing the spectral semantic difference inference coding vector through a spectral semantic difference reasoner based on a decoder to obtain an inference decoding value of the spectral leakage index.

7. The spectrum control method of the pulse modulation waveform of the all-solid-state transmitter according to claim 6, characterized in that: The DPD performance status evaluation results include excellent status, critical status and degraded status.

Images