A baseband radio frequency signal data feedback processing method based on vector modulation

Abstract

The application provides a baseband radio frequency signal data feedback processing method based on vector modulation, which first receives a baseband signal data set transmitted by a target radio frequency link in a preset frequency band, which contains multiple signal frame original modulation parameters and corresponding radio frequency feedback signal waveforms, then calculates feedback error indicators composed of phase offset, amplitude distortion and spectral leakage factor according to the radio frequency feedback signal waveforms, quantizes the modulation error, analyzes the feedback signal waveforms in the time-frequency domain, extracts error feature vectors, inputs the pre-trained model for cross-frame correlation learning, generates a dynamic adjustment parameter sequence containing time sequence dependence, and finally constructs a closed-loop feedback control link based on the dynamic adjustment parameter sequence and the original modulation parameters, so as to realize accurate processing and optimized control of the baseband radio frequency signal and effectively improve the signal transmission performance.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically, to a method for baseband radio frequency signal data feedback processing based on vector modulation. Background Technology

[0002] In the field of communication technology, baseband radio frequency signal processing is crucial for achieving efficient and accurate information transmission. With the increasing demands for signal quality and transmission efficiency in communication systems, how to effectively process baseband radio frequency signals to reduce transmission errors and improve signal performance has become a hot research topic.

[0003] For example, when analyzing signals, most existing technologies process signals based on single frames, failing to fully recognize the temporal dependencies and correlations between signal frames. Since signals in actual communication are continuous and dynamically changing, this approach of ignoring inter-frame relationships makes signal processing unable to adapt to the dynamic characteristics of signals, unable to respond to signal changes in a timely and effective manner, and reducing the real-time performance and effectiveness of signal processing.

[0004] Moreover, existing signal control methods are mostly open-loop control or fixed-parameter adjustment mechanisms. Open-loop control cannot be adjusted in real time according to the actual signal transmission conditions, while fixed-parameter adjustment mechanisms are difficult to cope with changes in different communication environments and signal characteristics, and cannot flexibly adapt to complex and ever-changing practical application scenarios. This results in unstable signal transmission quality and fails to meet the high precision and high reliability requirements of modern communication systems for signal processing. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a baseband radio frequency signal data feedback processing method based on vector modulation, the method comprising:

[0006] The receiver receives a baseband signal dataset transmitted by the target radio frequency link within a preset frequency band. The baseband signal dataset contains the original modulation parameters of multiple signal frames arranged in a time sequence, as well as the radio frequency feedback signal waveform corresponding to each signal frame.

[0007] The corresponding feedback error index is calculated based on the RF feedback signal waveform. The feedback error index consists of phase offset, amplitude distortion and spectral leakage factor, and is used to quantify the modulation error of the signal frame during transmission.

[0008] Based on the feedback error index, the time-frequency domain joint analysis is performed on the RF feedback signal waveform of each signal frame to extract the error feature vector of each signal frame. The error feature vector is then input into the pre-trained error prediction model for cross-frame correlation learning to generate a dynamic adjustment parameter sequence containing time-series dependencies.

[0009] Based on the dynamically adjusted parameter sequence and the original modulation parameters of the multiple signal frames, a closed-loop feedback control link is constructed.

[0010] In another aspect, embodiments of the present invention also provide a communication service system, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this application embodiment receives a baseband signal dataset containing the original modulation parameters and the RF feedback signal waveform. Then, it defines a feedback error index composed of phase offset, amplitude distortion, and spectral leakage factor. This index can highly accurately quantify the modulation error of the signal frame during transmission. Next, it uses time-frequency domain joint analysis combined with a pre-trained error prediction model for cross-frame correlation learning to generate a dynamic adjustment parameter sequence containing time-series dependencies. This fully explores the signal characteristics in different domains. Then, it uses the error prediction model for cross-frame learning, effectively capturing the dynamic change patterns between frames, thereby generating a dynamic adjustment parameter sequence that reflects the overall trend of the signal change. Finally, a closed-loop feedback control link is constructed based on the dynamic adjustment parameter sequence and the original modulation parameters, realizing intelligent and adaptive control of baseband RF signal transmission. This closed-loop feedback mechanism can automatically adjust relevant parameters in real time according to the signal transmission error and dynamic changes, effectively reducing modulation errors during signal transmission, improving the accuracy and stability of signal transmission, and significantly enhancing the overall performance and reliability of the baseband RF signal processing system. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the baseband radio frequency signal data feedback processing method based on vector modulation provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the communication service system provided in the embodiments of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a schematic flowchart of a vector modulation-based baseband radio frequency signal data feedback processing method provided in one embodiment of the present invention. The following is a detailed description of the vector modulation-based baseband radio frequency signal data feedback processing method.

[0015] Step S110: Receive the baseband signal dataset transmitted by the target RF link within a preset frequency band. The baseband signal dataset includes the original modulation parameters of multiple signal frames arranged in a time sequence, and the RF feedback signal waveform corresponding to each signal frame.

[0016] The target RF link refers to the RF link in a wireless communication system responsible for transmitting baseband signals within a specific frequency band. For example, the RF link in a wireless communication base station is responsible for converting baseband signals into RF signals and transmitting them. The preset frequency band refers to the specific frequency range in which the target RF link operates. For example, the 2.4GHz band is one of the commonly used frequency bands in wireless communication. The signal frame is the basic unit in the baseband signal data set. Each signal frame contains a certain length of data, which can be transmitted with specific modulation parameters. The original modulation parameters define how to convert digital information into a signal form suitable for RF transmission. For example, initial settings for phase and amplitude. The RF feedback signal waveform is the RF signal waveform obtained at a specific feedback point of the RF link after the RF feedback signal has been transmitted. It carries various information about the RF signal during actual transmission, such as interference and nonlinear effects of devices.

[0017] In detail, consider a scenario where a wireless communication base station transmits data to multiple mobile terminals. The target radio frequency (RF) link in the wireless communication base station operates in a specific preset frequency band, such as the 2.4 GHz band. Within this 2.4 GHz band, the base station continuously transmits a baseband signal dataset containing various information. This baseband signal dataset contains the raw modulation parameters of multiple signal frames arranged in a time sequence. These raw modulation parameters define how digital information is converted into a signal form suitable for RF transmission. For example, the raw modulation parameters may include initial settings for phase and amplitude. Simultaneously, the RF feedback signal waveform corresponding to each signal frame is also included in the baseband signal dataset. This RF feedback signal waveform is obtained from a specific feedback point on the link after the signal has passed through the RF link. For instance, a feedback acquisition point is set after the power amplifier at the base station transmitter. When the RF signal passes through this feedback acquisition point, its waveform is acquired as the RF feedback signal waveform. This RF feedback signal waveform carries various information about the RF feedback signal during actual transmission, such as interference and the nonlinear effects of devices.

[0018] Step S120: Calculate the corresponding feedback error index based on the RF feedback signal waveform. The feedback error index consists of phase offset, amplitude distortion and spectral leakage factor, and is used to quantify the modulation error of the signal frame during transmission.

[0019] The phase offset is the difference between the actual transmitted signal phase and the ideal transmitted signal phase, reflecting the phase change of the signal during transmission. The amplitude distortion is the degree of deviation between the actual transmitted signal amplitude and the ideal transmitted signal amplitude, reflecting the amplitude change of the signal during transmission. The spectral leakage factor is the degree to which the energy distribution of the signal in the spectrum does not conform to the ideal situation, such as spectral leakage caused by imperfections in the filter or modulation process.

[0020] For example, in the scenario of the aforementioned wireless communication base station, for each received RF feedback signal waveform, it is necessary to calculate the corresponding feedback error index. Taking the signal frame transmitted by the base station as an example, the RF feedback signal waveform is first subjected to quadrature demodulation. The quadrature demodulation process can be understood as decomposing the RF signal into a baseband in-phase component sequence and a baseband quadrature component sequence. Then, based on the obtained baseband in-phase component sequence and baseband quadrature component sequence, the instantaneous phase sequence and instantaneous amplitude sequence of each signal frame are calculated. For example, when calculating the instantaneous phase sequence, the phase value of each sampling point is obtained by using the relationship between the in-phase component and the quadrature component.

[0021] Next, the instantaneous phase sequence is compared point-by-point with a preset reference phase sequence to generate a phase difference sequence. This reference phase sequence is pre-defined based on an ideal signal transmission model and represents the phase change of the signal under no error conditions. Then, the phase difference sequence is averaged using a sliding window to obtain the phase offset for each signal frame. For example, if the actual phase deviates from the reference phase within a signal frame due to instability of certain components in the RF link (such as an oscillator), this deviation will be quantified as a phase offset.

[0022] To calculate amplitude distortion, the instantaneous amplitude sequence is normalized and the ratio of it to a preset reference amplitude sequence is calculated to generate an amplitude difference sequence. The reference amplitude sequence is also based on ideal transmission settings. In actual transmission, if the nonlinear characteristics of the power amplifier cause changes in signal amplitude, these changes will be reflected in the amplitude difference sequence. Then, statistical variance analysis is performed on the amplitude difference sequence to obtain the amplitude distortion for each signal frame.

[0023] Finally, a windowed Fourier transform is performed on the RF feedback signal waveform to extract the ratio of main lobe energy to side lobe energy, and the spectral leakage factor for each signal frame is calculated based on this ratio. In wireless communication, if the energy distribution of the signal in the spectrum does not conform to ideal conditions, for example, spectral leakage may be due to imperfections in the filter or modulation process, and the spectral leakage factor can quantify this phenomenon. By fusing the phase offset, amplitude distortion, and spectral leakage factor, a feedback error index that comprehensively reflects the modulation error of the signal frame during transmission is generated.

[0024] Step S130: Based on the feedback error index, perform time-frequency domain joint analysis on the RF feedback signal waveform of each signal frame, extract the error feature vector of each signal frame, and input the error feature vector into the pre-trained error prediction model for cross-frame correlation learning to generate a dynamic adjustment parameter sequence containing time-series dependencies.

[0025] Joint time-frequency domain analysis is a method that analyzes signals simultaneously from both the time and frequency domains. Time-domain analysis focuses on how the signal changes over time, while frequency-domain analysis focuses on the energy distribution of the signal in the frequency domain. The error feature vector, extracted through joint time-frequency domain analysis, is a vector that describes the error characteristics of the signal frame and includes error information described from different perspectives (such as phase, amplitude, and spectrum).

[0026] The pre-trained error prediction model is a model that has been trained on a large amount of data and is capable of predicting signal frame errors. Examples include Long Short-Term Memory (LSTM) networks. Cross-frame correlation learning is a method that utilizes the temporal correlation between signal frames to predict the error characteristics of future signal frames by analyzing the error features of historical signal frames. The dynamically adjusted parameter sequence is a parameter sequence containing temporal dependencies, generated based on the feedback error index and the results of cross-frame correlation learning. It is used to dynamically adjust the transmission parameters of the radio frequency link to optimize signal transmission quality.

[0027] For example, based on the feedback error index calculated earlier, a joint time-frequency domain analysis can be performed on the RF feedback signal waveform of each signal frame. First, a phase error variation curve is generated based on the difference between the phase offset and a preset reference phase trajectory, which represents the path of signal phase change over time under ideal transmission conditions. Simultaneously, the amplitude distortion is normalized using a sliding window to obtain an amplitude error distribution histogram. The amplitude error distribution histogram can visually display the distribution of amplitude error in different parts of the signal frame.

[0028] The phase error variation curve, amplitude error distribution histogram, and spectral leakage factor are concatenated using multi-channel features to generate an initial error feature vector. This initial error feature vector contains information describing signal error from different perspectives. Then, a convolutional neural network (CNN) is used to enhance the local features of the initial error feature vector. For example, in the CNN, multi-channel convolution processing is performed on the initial error feature vector. Convolutional kernels of different scales are used to perform parallel convolution operations on the phase error variation curve, amplitude error distribution histogram, and spectral leakage factor in the initial error feature vector, generating a first feature map containing different frequency response features. Convolutional kernels of different scales can capture error features of different frequency components.

[0029] Next, spatial pyramid pooling is applied to the first feature map to extract the maximum values ​​of local regions at different spatial levels, generating a second feature map with multi-resolution features. This step is similar to observing features in an image at different resolutions, enabling the acquisition of more comprehensive error feature information. Then, based on a frequency domain decomposition filter bank, the second feature map is used to separate the high-frequency and low-frequency error components, generating high-frequency and low-frequency feature sub-maps. This is analogous to separating a sound signal according to its high and low frequencies for more detailed analysis.

[0030] High-frequency and low-frequency feature sub-maps are input into a cross-fusion layer. Coupled feature maps of the high-frequency and low-frequency error components are generated through element-wise multiplication and channel attention weighting. This approach highlights the interrelationships between error components of different frequencies. The coupled feature maps are then subjected to depthwise separable convolution, with the convolution kernel sliding along both the time and frequency axes to integrate multi-scale contextual dependencies and generate a multi-scale fused feature map. Finally, the multi-scale fused feature map is residually concatenated with the initial error feature vector, and after normalization using an activation function, an error feature vector containing multi-scale error features is output.

[0031] Finally, the aforementioned error feature vectors are input into the pre-trained error prediction model for cross-frame correlation learning. Taking a Long Short-Term Memory (LSTM) network as an example, the error feature vectors are input into the LSM network in the order of signal frames to predict the phase shift trend and amplitude distortion trend of the next signal frame, generating preliminary adjustment parameters. Simultaneously, the set of error feature vectors for the target RF link within the historical transmission cycle is obtained, and the correlation weight between the current error feature vector and historical error feature vectors is calculated using an attention mechanism. For example, if the error features of the current signal frame are similar to the error features at a specific point in the past, then the error feature vector at that historical moment will be assigned a higher correlation weight. Based on the correlation weights, the historical adjustment parameters are weighted and fused to generate a historically dependent compensation correction amount. The preliminary adjustment parameters and the compensation correction amount are then superimposed to generate a dynamic adjustment parameter sequence containing temporal dependencies.

[0032] Step S140: Based on the dynamically adjusted parameter sequence and the original modulation parameters of the multiple signal frames, a closed-loop feedback control link is constructed.

[0033] The closed-loop feedback control link is constructed based on the dynamically adjusted parameter sequence and the original modulation parameters. It is a control link that can provide real-time feedback and adjust the signal transmission quality, realizing intelligent and adaptive control of baseband radio frequency signal transmission.

[0034] Based on the above steps, this embodiment of the application receives a baseband signal dataset containing the original modulation parameters and the RF feedback signal waveform. Then, it defines a feedback error index composed of phase offset, amplitude distortion, and spectral leakage factor, which can highly accurately quantify the modulation error of the signal frame during transmission. Next, it uses time-frequency domain joint analysis combined with a pre-trained error prediction model for cross-frame correlation learning to generate a dynamic adjustment parameter sequence containing time-series dependencies. This fully explores the signal characteristics in different domains. Then, it uses the error prediction model for cross-frame learning, effectively capturing the dynamic change patterns between frames, thereby generating a dynamic adjustment parameter sequence that reflects the overall trend of the signal change. Finally, based on the dynamic adjustment parameter sequence and the original modulation parameters, a closed-loop feedback control link is constructed, realizing intelligent and adaptive control of baseband RF signal transmission. This closed-loop feedback mechanism can automatically adjust relevant parameters in real time according to the signal transmission error and dynamic changes, effectively reducing modulation errors during signal transmission, improving the accuracy and stability of signal transmission, and significantly enhancing the overall performance and reliability of the baseband RF signal processing system.

[0035] In one possible implementation, step S120 includes:

[0036] Step S121: Perform quadrature demodulation on the RF feedback signal waveform to generate a baseband in-phase component sequence and a baseband quadrature component sequence.

[0037] In this embodiment, the radio frequency (RF) feedback signal in the base station's RF link contains complex information. Quadrature demodulation is a process based on mathematical principles and algorithms. For example, through specific circuit structures or digital signal processing algorithms, the RF feedback signal can be decomposed into a baseband in-phase component sequence and a baseband quadrature component sequence. Taking a specific signal frame as an example, assuming the RF feedback signal is a complex waveform with various changes after transmission, the baseband in-phase component sequence obtained after quadrature demodulation may be represented as a series of discrete values. These values ​​reflect the component characteristics of the RF feedback signal in the in-phase direction. Similarly, the baseband quadrature component sequence also exists in a similar discrete numerical form, corresponding to the in-phase component sequence in time, and together they describe the characteristics of the RF feedback signal in the baseband.

[0038] Step S122: Calculate the instantaneous phase sequence and instantaneous amplitude sequence of each signal frame based on the baseband in-phase component sequence and the baseband quadrature component sequence.

[0039] When calculating the instantaneous phase sequence, the relationship between the baseband in-phase components and the baseband quadrature components can be utilized. For example, for each sampling point, the instantaneous phase value of that sampling point is obtained through mathematical operations such as calculating the arctangent function. These instantaneous phase values, arranged in chronological order, constitute the instantaneous phase sequence. For calculating the instantaneous amplitude sequence, the instantaneous amplitude value of each sampling point can be obtained through mathematical operations such as taking the square root of the sum of the squares of the in-phase and quadrature components. These instantaneous amplitude values, arranged in chronological order, form the instantaneous amplitude sequence.

[0040] Step S123: Calculate the point-by-point difference between the instantaneous phase sequence and the preset reference phase sequence to generate a phase difference sequence, and perform sliding window averaging on the phase difference sequence to obtain the phase offset of each signal frame.

[0041] In the ideal transmission model of a base station, a preset reference phase sequence is established, representing the phase change pattern of the signal under conditions of no interference or error. When the instantaneous phase sequence during actual transmission is compared point-by-point with this reference phase sequence, a phase difference sequence is obtained. For example, if at a certain moment the value of the reference phase sequence is at a specific angle, and the value of the actual instantaneous phase sequence deviates from it, this deviation is recorded in the phase difference sequence. To obtain a more stable and representative phase offset, a sliding window averaging process is applied to the phase difference sequence. Assuming the sliding window size is a certain number of sampling points, the phase difference is averaged within this window, and the result is the phase offset of the signal frame. This phase offset reflects the average phase deviation of the signal throughout the entire signal frame transmission process, which may be due to unstable oscillator frequency in the RF link or other factors.

[0042] Step S124: Calculate the normalized ratio of the instantaneous amplitude sequence and the preset reference amplitude sequence to generate an amplitude difference sequence, and perform statistical variance analysis on the amplitude difference sequence to obtain the amplitude distortion of each signal frame.

[0043] In this embodiment, the preset reference amplitude sequence is also based on ideal transmission settings. In actual transmission, devices such as power amplifiers may exhibit nonlinear characteristics, causing changes in signal amplitude. An amplitude difference sequence is obtained by calculating the normalized ratio between the instantaneous amplitude sequence and the reference amplitude sequence. For example, if the instantaneous amplitude at a certain sampling point is 1.2 times or 0.8 times the reference amplitude, this ratio is recorded in the amplitude difference sequence. Statistical variance analysis is performed on the amplitude difference sequence; the variance reflects the dispersion of the amplitude difference throughout the entire signal frame. A larger variance indicates greater amplitude fluctuation within the signal frame, i.e., higher amplitude distortion; conversely, a smaller variance indicates lower amplitude distortion.

[0044] Step S125: Perform a windowed Fourier transform on the RF feedback signal waveform, extract the ratio of main lobe energy to side lobe energy, and calculate the spectral leakage factor of each signal frame based on the ratio.

[0045] In this embodiment, the windowed Fourier transform is a mathematical tool for analyzing the frequency domain characteristics of a signal. By performing a windowed Fourier transform on the RF feedback signal waveform, the energy distribution of the signal in the frequency domain can be obtained, including the main lobe energy and the side lobe energy. The main lobe energy represents the frequency band where the main energy of the signal is concentrated, while the side lobe energy is the energy distribution outside the main lobe. After calculating the ratio of the main lobe energy to the side lobe energy, the spectral leakage factor is calculated based on this ratio. For example, if the main lobe energy is very strong and the side lobe energy is relatively weak, the spectral leakage factor may be small, indicating that the energy distribution of the signal in the spectrum is relatively concentrated in the main lobe, and the spectral leakage is small; conversely, if the side lobe energy is not negligible relative to the main lobe energy, the spectral leakage factor is large, which may be due to imperfections in the filter or modulation process.

[0046] Step S126: The phase offset, amplitude distortion and spectral leakage factor are fused to generate the feedback error index.

[0047] For example, the fusion process can be based on a weighted algorithm or mathematical formula. For instance, different weights might be assigned according to the importance of each factor in the overall error, and then the phase offset, amplitude distortion, and spectral leakage factor are weighted and summed or otherwise calculated to obtain a feedback error index that comprehensively reflects the modulation error of the signal frame during transmission. This feedback error index will serve as an important basis for subsequent analysis and processing.

[0048] In one possible implementation, step S130 includes:

[0049] Step S131: Based on the difference between the phase offset and the preset reference phase trajectory, a phase error change curve is generated, and the amplitude distortion is normalized by a sliding window to obtain an amplitude error distribution histogram.

[0050] In this embodiment, the preset reference phase trajectory is the complete path of signal phase change over time under ideal transmission conditions. Therefore, by calculating the difference between the phase offset and this reference phase trajectory, the change of phase error over time within the entire signal frame can be obtained. Plotting these differences in chronological order generates the phase error change curve. For amplitude distortion, sliding window normalization is performed. For example, the amplitude distortion is normalized within a sliding window of a certain length, making the amplitude distortion within different windows comparable. Then, the frequency of each normalized amplitude distortion value is counted. Using these frequencies as the ordinate and the range of amplitude distortion values ​​as the abscissa, an amplitude error distribution histogram is plotted. This amplitude error distribution histogram can visually display the distribution of amplitude error across different amplitude ranges within the signal frame.

[0051] Step S132: The phase error change curve, amplitude error distribution histogram and spectral leakage factor are spliced ​​together using multi-channel features to generate an initial error feature vector.

[0052] In this process, these three pieces of information describing the error from different perspectives can be combined in a specific order and format. For example, the values ​​in the phase error change curve, the frequency values ​​in the amplitude error distribution histogram, and the values ​​of the spectral leakage factor may be arranged into a vector in a certain order. This vector becomes the initial error feature vector, which contains a preliminary comprehensive description of the signal frame error, but further processing is needed to extract more comprehensive error features.

[0053] Step S133: The initial error feature vector is enhanced locally using a convolutional neural network to extract the coupling relationship between high-frequency error components and low-frequency error components, thereby generating the error feature vector containing multi-scale error features.

[0054] In one possible implementation, step S133 includes:

[0055] Step S1331: Perform multi-channel convolution processing on the initial error feature vector. Use convolution kernels of different scales to perform parallel convolution operations on the phase error change curve, amplitude error distribution histogram and spectral leakage factor in the initial error feature vector to generate a first feature map containing different frequency response features.

[0056] In this embodiment, convolutional kernels of different scales can be understood as filters of different sizes, capable of capturing error features of different frequency components. For example, smaller-scale convolutional kernels may focus more on error features in the high-frequency part, being more sensitive to rapid changes in the phase error curve, local fluctuations in the amplitude error distribution histogram, and high-frequency changes in the spectral leakage factor; while larger-scale convolutional kernels focus more on error features in the low-frequency part, capable of capturing the changing trend of intra-frame errors over a longer time scale. Through parallel convolution operations, the resulting first feature map contains rich frequency response features, which describe the error situation from different frequency perspectives.

[0057] Step S1332: Perform spatial pyramid pooling on the first feature map to extract the maximum values ​​of local regions at different spatial levels and generate a second feature map with multi-resolution features.

[0058] In this process, local maxima of different regions are extracted from the first feature map. For example, at a finer spatial level, local extrema of the error features within a small range may be captured. These extrema may correspond to significant errors at certain specific moments or frequency points within a signal frame. At a coarser spatial level, the overall trend of the error features over a larger range can be obtained. The combination of these local maxima of different spatial levels forms a second feature map with multi-resolution features. This second feature map can more comprehensively reflect the distribution of error features at different spatial scales.

[0059] Step S1333: Based on the frequency domain decomposition filter bank, the second feature map is subjected to frequency band separation of high-frequency error components and low-frequency error components to generate high-frequency feature sub-maps and low-frequency feature sub-maps.

[0060] The frequency domain decomposition filter bank separates the high-frequency and low-frequency components of the second feature map according to pre-defined frequency limits. The high-frequency feature map mainly contains the high-frequency components of the error features, such as the high-frequency components that may reflect rapidly changing phase errors within a signal frame, short-term amplitude fluctuations, and spectral leakage. The low-frequency feature map mainly contains low-frequency components, such as the slow changing trend of phase errors within a signal frame and long-term amplitude fluctuations.

[0061] Step S1334: Input the high-frequency feature sub-map and the low-frequency feature sub-map into the cross-fusion layer, and generate a coupled feature map of high-frequency error components and low-frequency error components by element-wise multiplication and channel attention weighting.

[0062] Element-wise multiplication emphasizes the relationship between high-frequency and low-frequency error components, while channel attention weighting adjusts this relationship according to the importance of different channels (i.e., high-frequency and low-frequency channels). For example, if there is a specific correlation pattern between high-frequency and low-frequency error components in a particular signal frame, this method can highlight the correlation pattern, and the generated coupling feature map can more accurately describe this correlation.

[0063] Step S1335: Perform depthwise separable convolution processing on the coupled feature map, slide the convolution kernel along the time axis and frequency axis respectively, integrate multi-scale contextual dependencies, and generate a multi-scale fused feature map.

[0064] In this process, depthwise separable convolution can more effectively process the data in the coupled feature map. Sliding the convolution kernel along the time and frequency axes can be understood as scanning error features in different time and frequency directions. In this way, contextual dependencies at different scales can be integrated. For example, the relationship between high-frequency and low-frequency error components at different times and this relationship at different frequencies can be comprehensively considered. The generated multi-scale fused feature map can more comprehensively reflect the comprehensive characteristics of the error at multiple scales.

[0065] Step S1336: The multi-scale fused feature map is residually concatenated with the initial error feature vector, and the error feature vector containing multi-scale error features is output after normalization by the activation function.

[0066] In this embodiment, the residual connection can retain some original information in the initial error feature vector while fusing new features from the multi-scale fusion feature map. Activation function normalization adjusts the result, making the error feature vector numerically more compatible with the requirements of subsequent processing. This final error feature vector contains error features extracted from multiple scales and perspectives, providing more comprehensive and accurate input information for subsequent operations such as cross-frame correlation learning.

[0067] In one possible implementation, step S130 further includes:

[0068] Step S135: Input the error feature vector into the long short-term memory network in the order of signal frames to predict the phase shift trend and amplitude distortion trend of the next signal frame and generate preliminary adjustment parameters.

[0069] In detail, each signal frame has a corresponding error feature vector, which contains error information of that signal frame obtained from multi-scale and multi-faceted analysis. Long Short-Term Memory (LSTM) networks are neural network structures specifically designed for processing sequential data. By inputting the error feature vectors arranged sequentially by signal frame into the LSTM, the LSTM can process the sequential data using its internal memory units and gating mechanisms. For example, to predict phase shift trends, the LSTM analyzes phase-related information in the error feature vectors of previous signal frames, such as features in the phase error change curve, and learns the patterns of this information over time to predict the possible phase shift trend in the next signal frame. Similarly, to predict amplitude distortion trends, the LSTM integrates amplitude-related information such as the amplitude error distribution histogram in the error feature vectors of previous signal frames to derive a prediction of the amplitude distortion trend in the next signal frame. Based on these predictions, preliminary adjustment parameters are generated. These preliminary adjustment parameters are initial estimates of the phase and amplitude adjustments for the next signal frame, and may include values ​​such as the direction and approximate magnitude of the phase adjustment, and the proportion of the amplitude adjustment. These values ​​will serve as the basis for further adjustments.

[0070] Step S136: Obtain the set of error feature vectors of the target radio frequency link within the historical transmission period, and calculate the correlation weight between the current error feature vector and the historical error feature vector through an attention mechanism.

[0071] In this embodiment, a large set of error feature vectors from historical transmission cycles is accumulated during the long-term transmission process of the base station. These historical error feature vectors reflect the error situation of the target radio frequency link under different transmission conditions. The attention mechanism is a mechanism that can dynamically allocate weights based on the current input. In this scenario, when calculating the correlation weight between the current error feature vector and the historical error feature vector, the attention mechanism considers multiple factors. For example, if some features in the current error feature vector are very similar to features at a specific historical moment in the historical error feature vector, such as the shape of the phase error change curve or the pattern of the amplitude error distribution histogram, then the error feature vector at that historical moment will be assigned a higher correlation weight. Specifically, the correlation weight may be determined by mathematical methods such as calculating the distance metric (e.g., Euclidean distance) or correlation coefficient between the two. If the distance is close or the correlation is high, the correlation weight is large, indicating that the current error feature vector and the historical error feature vector have a strong correlation; conversely, if the distance is far or the correlation is low, the correlation weight is small.

[0072] Step S137: The historical adjustment parameters are weighted and fused according to the correlation weight to generate the historical dependency compensation correction amount, and the preliminary adjustment parameters are superimposed with the compensation correction amount to generate the dynamic adjustment parameter sequence.

[0073] Historical adjustment parameters correspond to historical error feature vectors and represent adjustment parameters applied to different signal frames during previous transmissions. These historical adjustment parameters are weighted and fused using correlation weights. For example, if the correlation weight between a historical error feature vector and the current error feature vector is high, then that historical adjustment parameter will have a larger weight in the weighted fusion process. Assuming that the historical adjustment parameters contain different adjustment values ​​for phase and amplitude, the resulting historically dependent compensation correction after weighted fusion can reflect the impact of historical transmission experience on the adjustment of the current signal frame. The preliminary adjustment parameters are superimposed with the compensation correction amount. The preliminary adjustment parameters are obtained by LSTM prediction based on the error feature vector of the current signal frame. The compensation correction amount is a weighted fusion result based on historical transmission experience. The superposition of the two generates a dynamic adjustment parameter sequence. This dynamic adjustment parameter sequence integrates the prediction result of the current signal frame and historical transmission experience. It contains complete parameter information for phase and amplitude adjustment. These parameters have time-dependent relationships and can provide accurate adjustment basis for subsequent construction of closed-loop feedback control link. This optimizes the radio frequency signal transmission of the base station, reduces phase offset, amplitude distortion and other problems, improves the quality and stability of signal transmission, and ensures that the mobile terminal can accurately receive the data sent by the base station.

[0074] In one possible implementation, step S140 includes:

[0075] Step S141: Based on the phase compensation weight and amplitude compensation ratio in the dynamically adjusted parameter sequence, perform multi-level vector modulation compensation on the phase component and amplitude component in the original modulation parameters to generate a target modulation parameter set containing frame-by-frame correction parameters. Each correction parameter contains a phase compensation curve and amplitude compensation coefficient matrix that are dynamically adjusted according to the error characteristics of adjacent signal frames.

[0076] In detail, consider a signal frame within which a small phase shift may occur due to the characteristics of certain components in the RF link or slight external interference. This shift may be approximately linear; for example, the phase changes at a relatively stable rate over time. In this case, the linear compensation coefficient for the phase shift within the signal frame comes into play, calculating the amount of compensation needed based on the trend of this shift, thus providing an initial correction to the phase component in the original modulation parameters. Simultaneously, within this signal frame, the amplitude may be distorted due to factors such as the nonlinearity of the power amplifier. The mean correction amount for this amplitude distortion can then be used to adjust the amplitude component in the original modulation parameters based on the average amplitude distortion. For example, if the average amplitude within the signal frame deviates from the ideal value, the mean correction amount can bring the average value back to a level close to the ideal value.

[0077] Step S142: The correction parameters in the target modulation parameter set are jointly encoded with the corresponding radio frequency feedback signal waveform to generate a parameter optimization sample set. The parameter optimization sample set is then iteratively optimized through a feedback control network to output real-time modulation parameters that meet the preset error convergence conditions. The feedback control network uses a gradient descent-based weight update mechanism to perform nonlinear correction on the phase compensation curve and amplitude compensation coefficient matrix.

[0078] Step S143: Load the real-time modulation parameters into the parameter configuration interface of the vector modulator, drive the vector modulator to perform multi-carrier quadrature modulation and waveform reconstruction on the baseband signal according to the real-time modulation parameters, generate an RF output signal that suppresses phase noise and amplitude distortion, and synchronously feed back the waveform characteristics of the RF output signal to the error compensation module of the target RF link to form a closed-loop feedback control link.

[0079] In one possible implementation, step S141 includes:

[0080] Step S1411: Decompose the phase compensation weight and amplitude compensation ratio in the dynamically adjusted parameter sequence into primary compensation parameters, intermediate compensation parameters, and advanced compensation parameters according to a preset compensation level. The primary compensation parameters include the linear compensation coefficient for phase shift within the signal frame and the mean correction amount for amplitude distortion. The intermediate compensation parameters include the gradient suppression coefficient for phase jump between adjacent signal frames and the boundary constraint value for amplitude fluctuation range. The advanced compensation parameters include the phase equalization factor based on the time-frequency distribution matrix and the amplitude distortion compensation weight.

[0081] Step S1412: Based on the primary compensation parameters, perform preliminary linear compensation on the phase component and amplitude component in the original modulation parameters of the current signal frame to obtain primary correction parameters, wherein the primary correction parameters include the average offset correction value of the phase component and the normalized scaling factor of the amplitude component.

[0082] In the base station signal processing flow, for the phase component, the average offset correction value is calculated based on the linear compensation coefficient of the phase offset within the signal frame. For example, if the linear compensation coefficient indicates that a certain angle adjustment of the phase is needed, this adjustment value is the average offset correction value, which directly corrects the phase component in the original modulation parameters. For the amplitude component, the normalized scaling factor is obtained based on the mean correction amount of amplitude distortion. Assuming that the original amplitude deviates from the normal proportional relationship due to distortion, the normalized scaling factor can adjust the amplitude to a relatively reasonable range, making the proportional relationship between the amplitude component and other related parameters more in line with the ideal signal transmission requirements.

[0083] Step S1413: Input the primary correction parameters into the nonlinear correction module corresponding to the intermediate compensation parameters, perform smoothing filtering on the instantaneous jump of the phase component according to the gradient suppression coefficient, and truncate and correct the extreme values ​​of the amplitude component in combination with the boundary constraint values ​​to generate intermediate correction parameters.

[0084] When a signal switches from one frame to the next, a momentary phase jump may occur. This jump may be caused by a signal type conversion or a sudden change in link state. In base station signal processing, the gradient suppression coefficient in the intermediate compensation parameters smooths out these momentary phase jumps. For example, if a large phase jump occurs suddenly, the gradient suppression coefficient will smooth the jump according to its amplitude and direction using a specific filtering algorithm, making the phase change more continuous and gradual, thus preventing such sudden jumps from adversely affecting subsequent signal processing and transmission. Simultaneously, for the amplitude component, various factors may cause large fluctuations in amplitude, even resulting in extreme values ​​exceeding the normal range. The boundary constraint value for the amplitude fluctuation range in the intermediate compensation parameters comes into play, truncating and correcting extreme values ​​of the amplitude component. For example, if the maximum amplitude value exceeds the set boundary constraint value, it is corrected to the boundary constraint value to prevent excessive amplitude fluctuations from severely affecting signal quality. This processing generates the intermediate correction parameters.

[0085] Step S1414: Input the intermediate correction parameters and the advanced compensation parameters into the time-frequency domain joint optimization module, perform energy balance adjustment on the frequency domain distribution of the phase component according to the phase equalization factor, and perform weighted suppression on the time domain fluctuation of the amplitude component through the amplitude distortion compensation weight to generate advanced correction parameters.

[0086] In radio frequency signal transmission at base stations, the phase of the signal may exhibit uneven energy distribution in the frequency domain. The phase equalization factor in the advanced compensation parameters addresses this issue within the time-frequency domain joint optimization module. For example, the phase energy at certain frequencies may be too high or too low. The phase equalization factor adjusts the energy distribution of the phase components in the frequency domain based on the phase energy at different locations, resulting in a more uniform energy distribution and improved frequency domain characteristics. Simultaneously, amplitude components may exhibit fluctuations in the time domain, which can affect signal stability. The amplitude distortion compensation weights suppress these time-domain fluctuations. For instance, if the amplitude fluctuates significantly within certain time periods, the amplitude distortion compensation weights suppress these fluctuations using a specific weighting algorithm based on the degree and characteristics of the fluctuations, resulting in a more stable amplitude in the time domain. This process generates the advanced correction parameters.

[0087] Step S1415: The primary correction parameters, intermediate correction parameters and advanced correction parameters are superimposed and integrated in the order of signal frames to generate a target modulation parameter set containing frame-by-frame correction parameters. The correction parameters of each signal frame include the three-level compensation superposition result of the phase component and the multi-level constraint fusion value of the amplitude component.

[0088] In the signal modulation process of the base station, for each signal frame, the three-level compensation superposition result of the phase component is obtained by sequentially superimposing the phase average offset correction value from the primary correction parameters, the phase value after smoothing and filtering from the intermediate correction parameters, and the phase value after frequency domain energy equalization adjustment from the advanced correction parameters. This superposition result integrates adjustments from multiple aspects, including preliminary linear compensation within the signal frame, smoothing processing between frames, and optimization in the frequency domain, enabling a more comprehensive correction of the phase component. For the amplitude component, the multi-level constraint fusion value is obtained by fusing the normalized scaling coefficient from the primary correction parameters, the amplitude value after extreme value truncation correction from the intermediate correction parameters, and the amplitude value after time-domain fluctuation weighted suppression from the advanced correction parameters. This fusion value considers amplitude adjustments at different levels, thus enabling more effective optimization of the amplitude component. These frame-by-frame correction parameters combine to form the target modulation parameter set, providing more accurate modulation parameters for subsequent signal processing.

[0089] In one possible implementation, step S142 includes:

[0090] Step S1421: Divide the parameter optimization sample set into a training set and a validation set, and perform feature mapping on the samples in the training set through the fully connected layer of the feedback control network to output intermediate optimization parameters.

[0091] For example, each sample in the training set contains various information obtained from previous processing, such as correction parameters and RF feedback signal waveforms. The fully connected layer performs complex mathematical operations on the information in these samples according to the pre-set connection weights and neuron structure, thereby mapping the input sample information into intermediate optimization parameters. These intermediate optimization parameters are a transformation and refinement of the original sample information, containing feature information related to error optimization.

[0092] Step S1422: Calculate the mean square error between the intermediate optimization parameters and the corresponding RF feedback signal waveforms in the verification set, and update the weight parameters of the feedback control network based on the mean square error using the backpropagation algorithm.

[0093] In the optimization process of base stations, mean square error (MSE) is a crucial indicator for measuring the difference between intermediate optimized parameters and the actual RF feedback signal waveform. For example, for each RF feedback signal waveform in the validation set, the MSE between it and the corresponding intermediate optimized parameters is calculated. This calculation involves mathematical operations such as summing the squared differences for each sampling point. Based on this MSE, the weight parameters of the feedback control network are updated using the backpropagation algorithm. The backpropagation algorithm is an optimization algorithm based on the gradient descent principle. It adjusts the weight parameters according to the gradient of the MSE relative to the weight parameters, with a certain step size, so that the weight parameters are updated in the direction that reduces the MSE. For example, if a certain weight parameter has a significant impact on the MSE and the current direction increases the MSE, then the backpropagation algorithm will decrease the value of that weight parameter; conversely, it will increase it, thereby continuously optimizing the weight parameters of the feedback control network so that it can better process samples and reduce errors.

[0094] Step S1423: When the rate of decrease of the mean square error in the iteration of a preset number of consecutive rounds is less than a preset threshold, the iteration is terminated and the real-time modulation parameters are output; otherwise, the number of hidden layer nodes is increased and the parameters are remapped.

[0095] In the iterative optimization process of a base station, the preset number of rounds and the preset threshold are crucial pre-defined parameters. If the rate of decrease of the mean square error (MSE) is less than the preset threshold during consecutive iterations of the preset number of rounds, it means that the error reduction has become very slow, and the optimization process is close to convergence. For example, assuming a preset number of rounds of 10 and a preset threshold of 0.01, if the rate of decrease of the MSE is less than 0.01 for 10 consecutive iterations, a relatively optimal state is considered to have been reached, at which point the iteration is terminated and the real-time modulation parameters are output. However, if the rate of decrease of the MSE has not yet reached the preset threshold, it indicates that the feedback control network may not have fully learned the information in the samples, and the network complexity needs to be increased. Increasing the number of hidden layer nodes can increase the network's expressive power, and then the parameters are remapped, and iterative optimization is performed again until the preset error convergence condition is met.

[0096] In one possible implementation, step S143 includes:

[0097] Step S1431: The real-time modulation parameters are parsed into multiple subcarrier modulation parameter groups according to the frequency band division rules of orthogonal subcarriers, and the bandwidth and center frequency of each subcarrier modulation parameter group are adapted and mapped according to the preset subcarrier frequency band allocation strategy.

[0098] For example, in a multi-carrier communication system, different subcarriers have different frequency band ranges and communication requirements. Real-time modulation parameters contain modulation information for the entire signal and need to be decomposed onto each subcarrier. Based on the frequency band allocation rules for orthogonal subcarriers, the relevant information in the real-time modulation parameters is parsed into multiple subcarrier modulation parameter groups. Each subcarrier modulation parameter group contains specific modulation information for that subcarrier. Then, according to a preset subcarrier frequency band allocation strategy—for example, some frequency bands are suitable for high-speed data transmission, while others are suitable for low-power transmission—bandwidth and center frequency are matched and mapped for each subcarrier modulation parameter group. This means that, based on the function and requirements of each subcarrier, a suitable bandwidth and center frequency are set for each subcarrier to ensure that the subcarrier can transmit signals within the optimal frequency band range.

[0099] Step S1432: The quadrature modulation unit of the vector modulator generates corresponding baseband IQ modulation signals based on the phase compensation curve and amplitude compensation coefficient matrix in each subcarrier modulation parameter group, and orthogonally superimposes the baseband IQ modulation signals of all subcarriers to generate a multi-carrier baseband quadrature modulation signal.

[0100] In a base station's vector modulator, the quadrature modulation unit (QMoU) is a crucial component for signal modulation. For each subcarrier modulation parameter set, the phase compensation curve and amplitude compensation coefficient matrix provide key information for adjusting the phase and amplitude of that subcarrier. Based on the phase compensation curve, the QMoU adjusts the phase of the baseband signal to generate the I-path modulated signal; based on the amplitude compensation coefficient matrix, it adjusts the amplitude of the baseband signal to generate the Q-path modulated signal. For example, if the phase compensation curve for a specific subcarrier indicates that the phase needs to be gradually increased over a certain time period, then the phase will be adjusted according to this curve when generating the I-path modulated signal. Similarly, if the amplitude compensation coefficient matrix specifies the amplitude scaling ratio, the amplitude will be adjusted according to this ratio when generating the Q-path modulated signal. Then, the baseband IQ modulation signals of all subcarriers are orthogonally superimposed, that is, the I and Q modulation signals of each subcarrier are superimposed in an orthogonal manner to generate a multi-carrier baseband orthogonal modulation signal. This multi-carrier baseband orthogonal modulation signal contains information from multiple subcarriers, and the orthogonal superposition method ensures the effectiveness and anti-interference of the signal.

[0101] Step S1433: Perform digital-to-analog conversion processing on the multi-carrier baseband quadrature modulation signal to generate an analog baseband waveform, and upconvert the analog baseband waveform to a preset radio frequency carrier frequency through a mixer to generate an initial radio frequency signal waveform.

[0102] During signal conversion at a base station, the multi-carrier baseband quadrature modulation signal is in digital form and needs to be converted into an analog signal for radio frequency (RF) transmission. Digital-to-analog (DAC) conversion transforms the digital multi-carrier baseband quadrature modulation signal into an analog baseband waveform. This process involves algorithms and circuit operations that convert the sampled values ​​of the digital signal to analog signals. The generated analog baseband waveform is a baseband analog signal with a relatively low frequency range. To transmit the signal to the RF band, a mixer is needed to upconvert the analog baseband waveform to a preset RF carrier frequency. For example, if the preset RF carrier frequency is 2.4 GHz, the mixer will mix the analog baseband waveform with a 2.4 GHz carrier signal, thereby boosting the frequency of the analog baseband waveform to 2.4 GHz and generating an initial RF signal waveform. This initial RF signal waveform then possesses frequency characteristics suitable for RF transmission.

[0103] Step S1434: Based on the nonlinear predistortion compensation coefficient in the real-time modulation parameters, perform predistortion correction processing on the initial radio frequency signal waveform to suppress the out-of-band radiation component of the initial radio frequency signal waveform and generate a predistortion corrected radio frequency output signal.

[0104] In base station radio frequency (RF) signal processing, due to the nonlinear characteristics of devices such as power amplifiers, the initial RF signal waveform may generate out-of-band radiation components. These out-of-band radiation components can interfere with signals in other frequency bands, affecting the performance of the entire communication system. The nonlinear predistortion compensation coefficient in real-time modulation parameters is used to compensate for this nonlinearity. Through specific predistortion correction algorithms and circuit structures, the initial RF signal waveform is predistorted according to the nonlinear predistortion compensation coefficient. For example, if the nonlinear predistortion compensation coefficient indicates that a certain frequency band of the signal needs to be attenuated to suppress out-of-band radiation, then the initial RF signal waveform can be processed accordingly in that frequency band, thereby suppressing the out-of-band radiation components of the initial RF signal waveform and generating a predistorted RF output signal. This RF output signal has better spectral characteristics and can reduce interference with signals in other frequency bands.

[0105] In step S1435, the time-domain waveform characteristics and frequency-domain energy distribution characteristics of the RF output signal are synchronously fed back to the error compensation module of the target RF link, and error correlation analysis is performed with the RF feedback signal waveform of the baseband signal dataset of the next transmission cycle to update the real-time modulation parameters.

[0106] In the closed-loop feedback control mechanism of the base station, after the time-domain waveform characteristics and frequency-domain energy distribution characteristics of the RF output signal are synchronously fed back to the error compensation module of the target RF link, the error compensation module performs error correlation analysis with the RF feedback signal waveform of the baseband signal dataset in the next transmission cycle. For example, information such as the change of signal amplitude over time and phase fluctuations in the time-domain waveform characteristics, and the energy magnitude of different frequency bands in the frequency-domain energy distribution characteristics, are compared and analyzed with the corresponding information in the RF feedback signal waveform of the next transmission cycle. Through this error correlation analysis, the trend of signal changes and the source of error during transmission can be discovered. Based on these analysis results, the real-time modulation parameters are updated so that the signal can be modulated and optimized more accurately in the next transmission cycle, thereby continuously improving the quality of signal transmission and reducing problems such as phase noise and amplitude distortion.

[0107] In one possible implementation, the closed-loop feedback control link is implemented by:

[0108] Step S210: Real-time acquisition of waveform sampling data of the radio frequency output signal, and calculation of the instantaneous error between the implemented waveform sampling data and the ideal reference waveform.

[0109] The radio frequency (RF) output signal carries the information that the base station wants to transmit to the mobile terminal. Its waveform sampling data contains key characteristics such as amplitude and phase of the signal at different times. To measure the accuracy of signal transmission, it is necessary to calculate the instantaneous error between these waveform sampling data and the ideal reference waveform. The ideal reference waveform is set based on theoretically perfect signal transmission; it represents the waveform that the RF output signal should have under conditions of no interference or distortion. Calculating the instantaneous error involves, for example, at each sampling moment, using specific mathematical formulas to calculate the amplitude difference and phase difference between the waveform sampling data and the ideal reference waveform at that moment. These differences combined constitute the instantaneous error. This instantaneous error reflects the degree to which the RF output signal deviates from the ideal state at each instant.

[0110] Step S220: When the instantaneous error exceeds the dynamic threshold, a fast compensation mechanism is triggered. The correction parameters of the most recent N frames are extracted from the target modulation parameter set and a weighted average calculation is performed to generate emergency compensation parameters.

[0111] A dynamic threshold is a limit value dynamically set based on system performance requirements and actual operating conditions. In the signal transmission environment of a base station, due to various factors such as environmental interference and equipment aging, instantaneous errors may fluctuate within a certain range. When the instantaneous error exceeds this dynamic threshold, it indicates that the signal deviation has reached a level requiring immediate adjustment. At this time, a fast compensation mechanism is triggered, extracting the correction parameters of the most recent N frames from the target modulation parameter set and performing a weighted average calculation to generate emergency compensation parameters. The target modulation parameter set contains frame-by-frame correction parameters after multi-level vector modulation compensation for each signal frame. These correction parameters are based on previous signal analysis and adjustments. The correction parameters of the most recent N frames are extracted; for example, assuming N is 5, the correction parameters of the most recent 5 signal frames are selected. When performing a weighted average calculation on these correction parameters, different weights may be assigned based on the importance of each signal frame or its relevance to the current situation. For example, signal frames closer to the current time may have their correction parameters assigned a higher weight because they better reflect the current signal trend. The emergency compensation parameters obtained through this weighted average calculation can quickly compensate for the current signal deviation.

[0112] Step S230: Insert the emergency compensation parameter into the head of the queue of the real-time modulation parameter, and load it first into the vector modulator for waveform reconstruction.

[0113] In the signal processing flow of a base station, real-time modulation parameters are a queue of parameters used to control the vector modulator's modulation of the baseband signal. Inserting the emergency compensation parameter at the head of the queue ensures that the vector modulator receives it first and prioritizes its use for waveform reconstruction. Upon receiving the emergency compensation parameter, the vector modulator quickly adjusts the baseband signal to be transmitted based on the phase and amplitude adjustment information contained within it. For example, if the emergency compensation parameter includes a phase correction value and an amplitude adjustment ratio, the vector modulator immediately changes the phase and amplitude of the baseband signal accordingly, thereby reconstructing a radio frequency output signal waveform closer to the ideal state, reducing the impact of instantaneous errors on signal transmission.

[0114] In one possible implementation, the method further includes:

[0115] Step S310: Perform spectrum monitoring on the radio frequency output signal to extract the energy value and harmonic distribution characteristics of out-of-band interference frequency points.

[0116] When a base station transmits data to a mobile terminal, spectrum monitoring of the radio frequency (RF) output signal is performed to ensure signal quality in the frequency domain. The spectrum monitoring process extracts the energy values ​​and harmonic distribution characteristics of out-of-band interference frequencies in the RF output signal. The spectrum of the RF output signal contains the energy distribution of the signal at different frequencies. Out-of-band interference frequencies refer to frequencies outside the normal signal band that exhibit abnormal energy due to various reasons (such as harmonics generated by nonlinear devices, external interference sources, etc.). Spectrum monitoring equipment or algorithms can accurately detect the energy values ​​of these out-of-band interference frequencies. For example, if the detected energy value at a specific frequency is higher than the normal background noise level, that frequency is identified as an out-of-band interference frequency, and its energy value is recorded. Simultaneously, harmonic distribution characteristics are extracted. Harmonics are frequency components that are integer multiples of the original signal frequency, generated by the signal passing through nonlinear devices (such as power amplifiers). By analyzing the distribution of these harmonics, including the energy magnitude and phase relationship of different harmonics, a comprehensive understanding of the frequency domain characteristics of the RF output signal can be obtained.

[0117] Step S320: If the energy value at any frequency point exceeds the threshold, then generate parameter configuration instructions for the notch filter based on the harmonic distribution characteristics.

[0118] A threshold value is a pre-defined energy limit used to determine whether out-of-band interference is severe enough to require intervention. When the energy value of an out-of-band interference frequency exceeds this threshold, it indicates that the interference at that frequency may significantly impact signal transmission. In this case, parameter configuration instructions for the notch filter are generated based on the harmonic distribution characteristics. For example, if the harmonic distribution characteristics show strong harmonic energy and complex phase relationships at and around an interference frequency, these factors need to be considered when generating the notch filter parameter configuration instructions. The center frequency and bandwidth of the interference frequency whose energy value exceeds the threshold in the harmonic distribution characteristics are determined and used as the initial stopband parameters of the notch filter. The center frequency is the core frequency of the interference frequency, and the bandwidth represents the range of affected frequencies at and around the interference frequency.

[0119] In one possible implementation, step S320 includes:

[0120] Step S321: Determine the center frequency and bandwidth of the interference frequency points whose energy values ​​exceed the threshold value in the harmonic distribution characteristics, and use the center frequency and bandwidth as the initial stopband parameters of the notch filter.

[0121] Under the analysis of spectrum monitoring equipment or algorithms, the spectrum of the RF output signal is analyzed in detail. For each frequency point, there is a corresponding energy value measurement result. When the energy value of a frequency point exceeds a preset threshold, that frequency point is determined to be an interference frequency point. For example, assuming communication is conducted near the 2.4GHz band, and an abnormally high energy value is detected at the 2.45GHz frequency point, exceeding the threshold, then 2.45GHz is the center frequency of this interference frequency point. Simultaneously, the bandwidth of this interference frequency point needs to be determined, as the bandwidth reflects the affected frequency range of the interference frequency point and its surrounding area. This can be determined by analyzing the energy distribution of the interference frequency point on the frequency axis. Assuming that the energy is significantly interfered with in the range of 2.43GHz to 2.47GHz, then 0.04GHz is the bandwidth of this interference frequency point. Using the center frequency of 2.45GHz and the bandwidth of 0.04GHz as the initial stopband parameters of the notch filter means that the notch filter is initially set to suppress interference within this frequency range.

[0122] Step S322: Based on the amplitude ratio and phase difference between the primary and secondary harmonics in the harmonic distribution characteristics, calculate the stopband attenuation slope and phase compensation factor of the notch filter to generate stopband suppression enhancement parameters.

[0123] In the harmonic distribution of RF output signals, the primary and secondary harmonics are of great significance. For example, suppose the primary harmonic has a frequency of f1 and an amplitude of A1, the secondary harmonic has a frequency of f2 and an amplitude of A2, and there is a phase difference θ between them. Using a specific mathematical algorithm, the stopband attenuation slope and phase compensation factor of the notch filter are calculated based on the amplitude ratio A2 / A1 and the phase difference θ. If the value of A2 / A1 is large, it indicates that the energy of the secondary harmonic is relatively stronger than that of the primary harmonic, and a steeper stopband attenuation slope may be needed to suppress the secondary harmonic. The specific calculations may involve complex formulas based on RF signal processing theory and filter design principles. Simultaneously, the phase difference θ also affects the calculation of the phase compensation factor, because different phase relationships during filtering can lead to signal distortion or changes, requiring adjustment through the phase compensation factor. The stopband attenuation slope and phase compensation factor obtained through such calculations constitute the stopband suppression enhancement parameter, which can further optimize the notch filter's interference suppression effect.

[0124] Step S323: Based on the initial stopband parameters and the stopband suppression enhancement parameters, perform dynamic offset compensation on the cutoff frequency boundary of the notch filter to generate the cutoff frequency range after transition band optimization.

[0125] In notch filter design, the initial stopband parameters (center frequency and bandwidth) and stopband suppression enhancement parameters (stopband attenuation slope and phase compensation factor) jointly influence the determination of the cutoff frequency boundary. For example, based on an initial center frequency of 2.45 GHz and a bandwidth of 0.04 GHz, along with the calculated stopband attenuation slope and phase compensation factor, the cutoff frequency boundary of the notch filter is adjusted. If the stopband attenuation slope is large, the cutoff frequency boundary may need to be appropriately shifted towards the interference frequency to ensure more effective interference suppression within the stopband. Simultaneously, considering the optimization of the transition band (the frequency band transitioning from the passband to the stopband), this shift is dynamic and considers multiple factors. For instance, to avoid excessive signal fluctuations or distortion within the transition band, the shift of the cutoff frequency boundary needs to be finely adjusted based on factors such as the phase compensation factor in the stopband suppression enhancement parameters. Through such dynamic shift compensation, a cutoff frequency range optimized for the transition band is ultimately generated, which better meets the requirements of the notch filter in suppressing interference and maintaining signal quality.

[0126] Step S324: Determine the passband ripple coefficient and filter order of the notch filter based on the cutoff frequency range and the preset maximum group delay constraint.

[0127] In base station signal transmission, after determining the cutoff frequency range, the passband ripple coefficient and filter order need to be determined based on the preset maximum group delay constraint. The maximum group delay constraint ensures that the signal does not experience excessive delay when passing through the notch filter, as excessive delay may affect signal synchronization and transmission quality. For example, assuming the cutoff frequency range is 2.43GHz - 2.47GHz, the preset maximum group delay is T. The passband ripple coefficient reflects the fluctuation of the signal amplitude within the passband (i.e., the frequency band for normal signal transmission) of the notch filter. Based on the cutoff frequency range and the maximum group delay constraint, the passband ripple coefficient is determined using a specific filter design algorithm. If the cutoff frequency range is wide, it may be necessary to appropriately reduce the passband ripple coefficient to reduce the impact on the signal within the passband. Meanwhile, the filter order is related to the filter's complexity and filtering performance. A higher filter order usually provides better filtering results, but it also increases computational complexity and potential signal delay. Under the premise of satisfying the maximum group delay constraint, an appropriate filter order is determined based on factors such as the cutoff frequency range and the passband ripple coefficient. For example, if a more precise filtering effect is required and the cutoff frequency range allows, a higher filter order may be selected, but it is important to ensure that the group delay does not exceed the preset value.

[0128] Step S325: By iteratively adjusting the combination of the passband ripple coefficient and the filter order, the out-of-band suppression depth of the notch filter reaches the energy suppression requirement of the subharmonics in the harmonic distribution characteristics.

[0129] In determining the parameters of a notch filter, it may not be possible to determine the passband ripple coefficient and filter order that meet the subharmonic energy suppression requirements in one step. For example, first set an initial passband ripple coefficient and filter order, and then calculate the out-of-band suppression depth of the notch filter at this point. Assume the initial passband ripple coefficient is Rp1, the filter order is N1, and the calculated out-of-band suppression depth is D1. If D1 is less than the energy suppression requirement of the subharmonics in the harmonic distribution characteristics, the passband ripple coefficient and filter order need to be adjusted based on the calculation results. For example, increase the filter order to N2, and simultaneously adjust the passband ripple coefficient to Rp2, then recalculate the out-of-band suppression depth to obtain D2. If D2 still does not meet the requirements, continue adjusting, and iterate repeatedly until the out-of-band suppression depth meets the energy suppression requirement of the subharmonics in the harmonic distribution characteristics. This process requires precise calculations and a deep understanding of the filter performance to ensure that the notch filter can effectively suppress out-of-band interference.

[0130] Step S326: The final determined cutoff frequency range, stopband attenuation slope, phase compensation factor, passband ripple coefficient, and filter order are encapsulated into the parameter configuration instruction.

[0131] In detail, this parameter configuration instruction contains all the parameter information required for the notch filter to perform effective filtering. In the base station's signal processing system, these parameters are encapsulated into an instruction to be accurately passed to the subsequent filtering unit of the vector modulator. For example, this parameter configuration instruction may be encapsulated in a specific data format, where each parameter is encoded in a prescribed order and format, ensuring that the subsequent filtering unit can correctly parse and configure the notch filter according to these parameters.

[0132] Step S330: The parameter configuration command is sent to the post-stage filtering unit of the vector modulator, and an inverse compensation component is superimposed on the modulation parameters of the next signal frame to cancel out-of-band interference.

[0133] In this embodiment, after receiving the parameter configuration command, the subsequent filtering unit configures the notch filter according to these parameters to effectively suppress out-of-band interference. Simultaneously, an inverse compensation component is superimposed on the modulation parameters of the next signal frame to cancel out-of-band interference. During signal modulation at the base station, the modulation parameters of the next signal frame are adjusted based on the filtering effect of the notch filter. The superposition of the inverse compensation component is to further compensate for the potential impact of out-of-band interference on the signal. For example, if out-of-band interference causes a change in the amplitude or phase of the signal at a certain frequency, by superimposing the inverse compensation component, the amplitude and phase of the signal at that frequency can be adjusted back to a near-normal state, thereby improving the signal transmission quality and reducing the impact of out-of-band interference on signal transmission.

[0134] In one possible implementation, the method further includes:

[0135] Step S410: Calculate the channel quality index of the target radio frequency link based on the accumulated error statistics in the closed-loop feedback control link.

[0136] During operation, the closed-loop feedback control link continuously collects and accumulates various error information, such as statistics on error indicators like phase offset, amplitude distortion, and spectral leakage factor. These error statistics reflect the accuracy and stability of signal transmission in the radio frequency link. A specific algorithm combines these error statistics to calculate the channel quality index. For example, different weights might be assigned based on the importance of different error indicators to signal quality, and then the statistics of each error indicator are weighted and summed or subjected to other mathematical operations to obtain the channel quality index. This channel quality index comprehensively reflects the communication quality of the target radio frequency link.

[0137] Step S420: When the channel quality index is lower than the first threshold, the modulation order is reduced and the error correction coding redundancy is increased based on the first optimization strategy; and when the channel quality index is higher than the second threshold, the modulation order is increased based on the second optimization strategy and compressed sensing technology is used to optimize the spectrum efficiency.

[0138] In scenarios where a base station transmits data to a mobile terminal, the first threshold is a pre-defined boundary used to determine whether the channel quality is poor. When the channel quality index falls below this first threshold, it indicates that the channel quality has deteriorated to the point where measures need to be taken to improve signal transmission reliability. At this point, the modulation order is reduced based on the first optimization strategy. The modulation order is related to the signal transmission rate and anti-interference capability; a higher modulation order can achieve a higher data transmission rate, but it is more susceptible to interference when the channel quality is poor. Reducing the modulation order means sacrificing some data transmission rate in exchange for stronger anti-interference capability. For example, if a higher modulation order modulation method like 64QAM (64-level quadrature amplitude modulation) was originally used, it might be reduced to 16QAM. Simultaneously, error correction coding redundancy is increased. Error correction coding is coded information added to detect and correct errors during signal transmission. Increasing redundancy means adding more error correction coding information to the signal, so that even under poor channel quality conditions, the mobile terminal can correct errors in the received signal using this error correction coding information. For example, increasing the original error correction coding rate from 1 / 2 to 3 / 4 increases the redundancy information in the signal.

[0139] The second threshold is also a pre-set limit used to determine whether the channel quality is good. When the channel quality index is higher than this second threshold, it indicates that the channel quality is good, and this advantage can be used to improve data transmission rate and spectral efficiency. Based on the second optimization strategy, the modulation order is increased, for example from 16QAM to 64QAM or even higher, thereby improving the data transmission rate. Simultaneously, compressed sensing technology is used to optimize spectral efficiency. In base station radio frequency transmission, spectrum resources are limited. Compressed sensing technology can utilize the sparsity of the signal to improve spectral efficiency without increasing bandwidth. For example, by processing the baseband signal using compressed sensing algorithms, the signal distribution on the spectrum becomes more compact, thus enabling the transmission of more data within the same spectrum range.

[0140] Step S430: The adjusted modulation strategy parameters are encapsulated into control commands, and the configuration parameters of the vector modulator are updated in real time through the closed-loop feedback control link.

[0141] In the base station's signal processing flow, adjusted modulation strategy parameters (such as modulation order and error correction coding redundancy) need to be accurately transmitted to the vector modulator to update its configuration. These parameters are encapsulated into control commands according to a specific data format and protocol. These control commands contain all the information required for the vector modulator to be correctly configured. Through a closed-loop feedback control link, these control commands are sent in real-time to the vector modulator's configuration interface. Upon receiving the control command, the vector modulator updates its own configuration according to the parameters in the command. For example, if the control command includes information to increase the modulation order from 16QAM to 64QAM, the vector modulator will adjust its internal modulation modules accordingly to modulate the baseband signal according to the new modulation order. Similarly, if it includes information to adjust the error correction coding redundancy, the error correction coding modules will also be updated accordingly. In this way, by updating the vector modulator's configuration parameters in real time, the signal transmission strategy can be adjusted promptly according to changes in channel quality, improving the performance and efficiency of the entire wireless communication system.

[0142] Figure 2 The diagram illustrates exemplary hardware and software components of a communication service system 100 that can implement the inventive ideas of the present invention, according to some embodiments of the invention. For example, a processor 120 may be used in the communication service system 100 and to perform the functions of the present invention.

[0143] The communication service system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the vector modulation-based baseband radio frequency signal data feedback processing method of the present invention. Although only one server is shown in this invention, for convenience, the functions described in this invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0144] For example, the communication service system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the communication service system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The communication service system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0145] For ease of explanation, only one processor is described in the communication service system 100. However, it should be noted that the communication service system 100 of the present invention may also include multiple processors, and therefore the steps performed by one processor as described in the present invention may also be performed jointly or individually by multiple processors. For example, if the processor of the communication service system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0146] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-described baseband radio frequency signal data feedback processing method based on vector modulation is implemented.

[0147] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for baseband radio frequency signal data feedback processing based on vector modulation, characterized in that, The method includes: The receiver receives a baseband signal dataset transmitted by the target radio frequency link within a preset frequency band. The baseband signal dataset contains the original modulation parameters of multiple signal frames arranged in a time sequence, as well as the radio frequency feedback signal waveform corresponding to each signal frame. The corresponding feedback error index is calculated based on the RF feedback signal waveform. The feedback error index consists of phase offset, amplitude distortion and spectral leakage factor, and is used to quantify the modulation error of the signal frame during transmission. Based on the feedback error index, the time-frequency domain joint analysis is performed on the RF feedback signal waveform of each signal frame to extract the error feature vector of each signal frame. The error feature vector is then input into the pre-trained error prediction model for cross-frame correlation learning to generate a dynamic adjustment parameter sequence containing time-series dependencies. Based on the dynamically adjusted parameter sequence and the original modulation parameters of the multiple signal frames, a closed-loop feedback control link is constructed. The step of constructing a closed-loop feedback control link based on the dynamically adjusted parameter sequence and the original modulation parameters of the multiple signal frames includes: Based on the phase compensation weight and amplitude compensation ratio in the dynamically adjusted parameter sequence, multi-level vector modulation compensation is performed on the phase component and amplitude component in the original modulation parameters to generate a target modulation parameter set containing frame-by-frame correction parameters. Each correction parameter contains a phase compensation curve and amplitude compensation coefficient matrix that are dynamically adjusted according to the error characteristics of adjacent signal frames. The correction parameters in the target modulation parameter set are jointly encoded with the corresponding radio frequency feedback signal waveform to generate a parameter optimization sample set. The parameter optimization sample set is then iteratively optimized through a feedback control network to output real-time modulation parameters that meet the preset error convergence conditions. The feedback control network uses a gradient descent-based weight update mechanism to perform nonlinear correction on the phase compensation curve and amplitude compensation coefficient matrix. The real-time modulation parameters are loaded into the parameter configuration interface of the vector modulator, which drives the vector modulator to perform multi-carrier quadrature modulation and waveform reconstruction on the baseband signal according to the real-time modulation parameters, generating an RF output signal that suppresses phase noise and amplitude distortion, and synchronously feeding back the waveform characteristics of the RF output signal to the error compensation module of the target RF link to form a closed-loop feedback control link. The step of loading the real-time modulation parameters into the parameter configuration interface of the vector modulator, driving the vector modulator to perform multi-carrier quadrature modulation and waveform reconstruction on the baseband signal according to the real-time modulation parameters, and generating an RF output signal that suppresses phase noise and amplitude distortion includes: The real-time modulation parameters are parsed into multiple subcarrier modulation parameter groups according to the frequency band division rules of orthogonal subcarriers, and the bandwidth and center frequency are adapted and mapped for each subcarrier modulation parameter group according to the preset subcarrier frequency band allocation strategy. The quadrature modulation unit of the vector modulator generates corresponding baseband IQ modulation signals based on the phase compensation curve and amplitude compensation coefficient matrix in each subcarrier modulation parameter group, and orthogonally superimposes the baseband IQ modulation signals of all subcarriers to generate a multi-carrier baseband quadrature modulation signal. The multi-carrier baseband quadrature modulation signal is processed by digital-to-analog conversion to generate an analog baseband waveform, and the analog baseband waveform is up-converted to a preset radio frequency carrier frequency by a mixer to generate an initial radio frequency signal waveform. Based on the nonlinear predistortion compensation coefficient in the real-time modulation parameters, the initial radio frequency signal waveform is subjected to predistortion correction processing to suppress the out-of-band radiation component of the initial radio frequency signal waveform and generate a predistortion corrected radio frequency output signal. In addition, the time-domain waveform characteristics and frequency-domain energy distribution characteristics of the RF output signal are synchronously fed back to the error compensation module of the target RF link, and error correlation analysis is performed with the RF feedback signal waveform of the baseband signal dataset of the next transmission cycle to update the real-time modulation parameters. The method further includes: The frequency spectrum of the radio frequency output signal is monitored to extract the energy value and harmonic distribution characteristics of out-of-band interference frequency points; If the energy value at any frequency point exceeds the threshold, then a parameter configuration instruction for the notch filter is generated based on the harmonic distribution characteristics. The parameter configuration command is sent to the post-stage filtering unit of the vector modulator, and an inverse compensation component is superimposed on the modulation parameters of the next signal frame to cancel out-of-band interference. The parameter configuration instructions for generating the notch filter based on the harmonic distribution characteristics include: The center frequency and bandwidth of the interference frequency points whose energy values ​​exceed the threshold value in the harmonic distribution characteristics are determined, and the center frequency and bandwidth are used as the initial stopband parameters of the notch filter. Based on the amplitude ratio and phase difference between the primary and secondary harmonics in the harmonic distribution characteristics, the stopband attenuation slope and phase compensation factor of the notch filter are calculated to generate stopband suppression enhancement parameters. Based on the initial stopband parameters and the stopband suppression enhancement parameters, dynamic offset compensation is performed on the cutoff frequency boundary of the notch filter to generate the cutoff frequency range after transition band optimization. Based on the cutoff frequency range and the preset maximum group delay constraint, the passband ripple coefficient and filter order of the notch filter are determined. By iteratively adjusting the combination of the passband ripple coefficient and the filter order, the out-of-band suppression depth of the notch filter reaches the energy suppression requirement of the subharmonics in the harmonic distribution characteristics. The final determined cutoff frequency range, stopband attenuation slope, phase compensation factor, passband ripple coefficient, and filter order are encapsulated into the parameter configuration command.

2. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The step of calculating the corresponding feedback error index based on the radio frequency feedback signal waveform includes: The radio frequency feedback signal waveform is quadrature demodulated to generate a baseband in-phase component sequence and a baseband quadrature component sequence; Based on the baseband in-phase component sequence and the baseband quadrature component sequence, calculate the instantaneous phase sequence and instantaneous amplitude sequence of each signal frame; The instantaneous phase sequence is compared with a preset reference phase sequence point by point to generate a phase difference sequence. The phase difference sequence is then averaged using a sliding window to obtain the phase offset of each signal frame. The instantaneous amplitude sequence is normalized and the ratio of the instantaneous amplitude sequence to the preset reference amplitude sequence is calculated to generate an amplitude difference sequence. Statistical variance analysis is then performed on the amplitude difference sequence to obtain the amplitude distortion of each signal frame. A windowed Fourier transform is performed on the RF feedback signal waveform to extract the ratio of main lobe energy to side lobe energy, and the spectral leakage factor of each signal frame is calculated based on the ratio. The phase offset, amplitude distortion, and spectral leakage factor are fused to generate the feedback error index; The step of performing time-frequency domain joint analysis on the RF feedback signal waveform of each signal frame based on the feedback error index, and extracting the error feature vector of each signal frame, includes: Based on the difference between the phase offset and the preset reference phase trajectory, a phase error variation curve is generated, and the amplitude distortion is normalized by a sliding window to obtain an amplitude error distribution histogram. The phase error variation curve, the amplitude error distribution histogram, and the spectral leakage factor are spliced ​​together using multi-channel features to generate an initial error feature vector. The initial error feature vector is enhanced locally by using a convolutional neural network to extract the coupling relationship between high-frequency error components and low-frequency error components, thereby generating the error feature vector containing multi-scale error features. The step of performing local feature enhancement on the initial error feature vector using a convolutional neural network, extracting the coupling relationship between high-frequency error components and low-frequency error components, and generating the error feature vector containing multi-scale error features includes: The initial error feature vector is subjected to multi-channel convolution processing. Convolution kernels of different scales are used to perform parallel convolution operations on the phase error change curve, amplitude error distribution histogram and spectral leakage factor in the initial error feature vector to generate a first feature map containing different frequency response features. The first feature map is subjected to spatial pyramid pooling to extract the maximum values ​​of local regions at different spatial levels, thereby generating a second feature map with multi-resolution features. The second feature map is separated into high-frequency and low-frequency error components based on the frequency domain decomposition filter bank, generating high-frequency and low-frequency feature sub-maps. The high-frequency feature sub-map and the low-frequency feature sub-map are input into the cross-fusion layer, and a coupled feature map of high-frequency error components and low-frequency error components is generated by element-wise multiplication and channel attention weighting. The coupled feature map is subjected to depthwise separable convolution processing, and the convolution kernel is slid along the time axis and frequency axis respectively to integrate multi-scale contextual dependencies and generate a multi-scale fused feature map. The multi-scale fused feature map is residually concatenated with the initial error feature vector, and after normalization by an activation function, the error feature vector containing multi-scale error features is output.

3. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The step of inputting the error feature vector into a pre-trained error prediction model for cross-frame correlation learning to generate a dynamically adjusted parameter sequence containing temporal dependencies includes: The error feature vectors are input into the Long Short-Term Memory network in the order of signal frames to predict the phase shift trend and amplitude distortion trend of the next signal frame and generate preliminary adjustment parameters. Obtain the set of error feature vectors of the target radio frequency link within the historical transmission period, and calculate the correlation weight between the current error feature vector and the historical error feature vector through an attention mechanism; The historical adjustment parameters are weighted and fused according to the correlation weight to generate the historically dependent compensation correction amount, and the preliminary adjustment parameters are superimposed with the compensation correction amount to generate the dynamic adjustment parameter sequence.

4. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The step involves performing multi-level vector modulation compensation on the phase and amplitude components of the original modulation parameters based on the phase compensation weights and amplitude compensation ratios in the dynamically adjusted parameter sequence, generating a target modulation parameter set containing frame-by-frame correction parameters, including: The phase compensation weight and amplitude compensation ratio in the dynamically adjusted parameter sequence are decomposed into primary compensation parameters, intermediate compensation parameters and advanced compensation parameters according to a preset compensation level. The primary compensation parameters include the linear compensation coefficient for phase shift within the signal frame and the mean correction amount for amplitude distortion. The intermediate compensation parameters include the gradient suppression coefficient for phase jump between adjacent signal frames and the boundary constraint value for amplitude fluctuation range. The advanced compensation parameters include the phase equalization factor based on the time-frequency distribution matrix and the amplitude distortion compensation weight. Based on the primary compensation parameters, preliminary linear compensation is performed on the phase component and amplitude component in the original modulation parameters of the current signal frame to obtain primary correction parameters, wherein the primary correction parameters include the average offset correction value of the phase component and the normalized scaling factor of the amplitude component. The primary correction parameters are input into the nonlinear correction module corresponding to the intermediate compensation parameters. The instantaneous jump of the phase component is smoothed and filtered according to the gradient suppression coefficient. The extreme values ​​of the amplitude component are truncated and corrected in combination with the boundary constraint values ​​to generate intermediate correction parameters. The intermediate correction parameters and the advanced compensation parameters are input into the time-frequency domain joint optimization module. The phase component frequency domain distribution is adjusted for energy balance according to the phase equalization factor. The amplitude component time domain fluctuation is suppressed by the amplitude distortion compensation weight to generate the advanced correction parameters. The primary correction parameters, intermediate correction parameters, and advanced correction parameters are superimposed and integrated in the order of signal frames to generate a target modulation parameter set containing frame-by-frame correction parameters. The correction parameters for each signal frame include the three-level compensation superposition result of the phase component and the multi-level constraint fusion value of the amplitude component.

5. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The step of performing multiple rounds of iterative optimization on the parameter optimization sample set through a feedback control network to output real-time modulation parameters that satisfy a preset error convergence condition includes: The parameter optimization sample set is divided into a training set and a validation set, and the samples in the training set are feature-mapped through the fully connected layer of the feedback control network to output intermediate optimization parameters. Calculate the mean square error between the intermediate optimization parameters and the corresponding RF feedback signal waveforms in the validation set, and update the weight parameters of the feedback control network based on the mean square error using the backpropagation algorithm. When the rate of decrease of the mean square error is less than a preset threshold during a series of iterations with a preset number of rounds, the iteration is terminated and the real-time modulation parameters are output; otherwise, the number of hidden layer nodes is increased and the parameters are remapped.

6. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The implementation process of the closed-loop feedback control link includes: The waveform sampling data of the radio frequency output signal is acquired in real time, and the instantaneous error between the real-time waveform sampling data and the ideal reference waveform is calculated. When the instantaneous error exceeds the dynamic threshold, a fast compensation mechanism is triggered. The correction parameters of the most recent N frames are extracted from the target modulation parameter set and a weighted average is calculated to generate emergency compensation parameters. The emergency compensation parameter is inserted at the head of the queue of the real-time modulation parameter and is preferentially loaded into the vector modulator for waveform reconstruction.

7. The baseband radio frequency signal data feedback processing method based on vector modulation according to claim 1, characterized in that, The method further includes: The channel quality index of the target radio frequency link is calculated based on the accumulated error statistics in the closed-loop feedback control link. When the channel quality index is lower than the first threshold, the modulation order is reduced and the error correction coding redundancy is increased based on the first optimization strategy; and when the channel quality index is higher than the second threshold, the modulation order is increased and compressed sensing technology is used to optimize the spectrum efficiency based on the second optimization strategy. The adjusted modulation strategy parameters are encapsulated into control commands, and the configuration parameters of the vector modulator are updated in real time through the closed-loop feedback control link.

Images