Wireless communication interference suppression method based on channel characteristic adaptive matching

Abstract

The invention discloses a wireless communication interference suppression method based on channel characteristic adaptive matching, and relates to the technical field of wireless communication. Comprising the following steps: firstly, acquiring and preprocessing a wireless communication receiving signal, and extracting a channel feature to generate a first state feature vector; and inputting the signal into a preset interference strategy matching model, outputting a first interference identifier and a strategy parameter set, and generating and executing a first configuration instruction according to the first interference identifier and the strategy parameter set. Then, signals are collected again to generate a second state feature vector, and the performance improvement degree of the second state feature vector and the first state feature vector is calculated; if the second state feature vector is lower than the preset threshold value, judging that the strategy is invalid, inputting the second state feature vector into a correction strategy library to output a second interference identifier and a correction parameter, and executing a second configuration instruction; and meanwhile, the failure decision data pair is sent to the matching model for online incremental updating. According to the invention, the stability and reliability of a wireless communication link in a complex time-varying interference environment are improved.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, specifically to a wireless communication interference suppression method based on adaptive matching of channel characteristics. Background Technology

[0002] With the rapid development of wireless communication technology, especially in scenarios such as the Internet of Things, vehicle-to-everything (V2X) communication, and industrial wireless control, communication systems face increasingly complex and dynamically changing electromagnetic interference environments. How to effectively suppress interference and ensure the stability and transmission quality of communication links has become a core issue of continuous concern in the field of wireless communication.

[0003] In existing technologies, adaptive interference suppression methods have been extensively studied. Typical methods include fixed filter switching based on threshold detection, power control or coding adjustment based on single indicators such as signal-to-noise ratio, and interference filtering using traditional signal processing algorithms (such as adaptive beamforming and blind source separation). These methods can cope with known or slowly changing interference to a certain extent and have a certain degree of environmental adaptability. However, they are essentially still an "open-loop" or "quasi-closed-loop" design paradigm, responding according to preset rules or current measurements. They often lack a continuous and quantitative evaluation mechanism for the actual effect of the implemented suppression strategy, and cannot make closed-loop corrections to the core decision-making unit based on the evaluation results. In real dynamic environments, interference types and channel conditions may change rapidly and non-stationarily, causing previously effective suppression strategies to quickly become ineffective. The system cannot detect this failure in time, or even if it does detect it (such as a decline in communication quality), it can only passively switch to another preset strategy and cannot actively learn from failure cases. This results in the system being in an inefficient cycle of "trial and error - failure - retrial" for a long time. This results in existing technologies having inherent limitations in adaptability and long-term robustness when dealing with complex, time-varying, and non-stationary interference scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a wireless communication interference suppression method based on adaptive matching of channel features.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] This invention discloses a wireless communication interference suppression method based on adaptive matching of channel features, comprising:

[0007] The wireless communication received signal is acquired and preprocessed to obtain time-domain baseband signal data;

[0008] Channel features are extracted from the time-domain baseband signal data to generate a first state feature vector;

[0009] The first state feature vector is input into a preset interference strategy matching model, and the first interference identifier and associated strategy parameter set corresponding to the current channel interference state are output. The first configuration instruction is generated and executed based on the first interference identifier and strategy parameter set.

[0010] After a preset time following the execution of the first configuration instruction, the wireless communication received signal is reacquired and processed to generate a second state feature vector;

[0011] Calculate the performance improvement between the second state feature vector and the first state feature vector;

[0012] If the performance improvement is lower than the preset improvement threshold, the first interference strategy is determined to be ineffective, the second state feature vector is input into the preset correction strategy library, the second interference identifier and the associated correction parameter set are output, and the second configuration instruction is generated and executed.

[0013] Simultaneously, the failure decision data pair, which includes the first state feature vector, the first interference identifier, and the performance improvement degree, is sent to the interference strategy matching model for online incremental updates.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] 1. This invention transforms the interference suppression process from traditional static configuration or simple feedback into a dynamic, adaptive, and intelligent decision-making process. This scheme extracts channel state feature vectors in real time and uses these vectors to match preset interference suppression strategies, enabling the system to automatically adapt to constantly changing interference environments. This improves the accuracy and timeliness of anti-interference efforts, overcoming the shortcomings of fixed strategies in complex dynamic scenarios.

[0016] 2. The performance improvement evaluation and correction mechanism introduced in this invention provides the system with the ability to dynamically correct errors and continuously optimize. This enables the system not only to remedy single decision-making errors, but also to learn from failures and optimize future decision-making logic, thereby giving the entire interference suppression system the intelligent characteristics of self-evolution and continuous improvement.

[0017] 3. This invention, through the collaborative design of the interference strategy matching model and the correction strategy library, ensures the robustness and reliability of the system while pursuing the optimal intelligent matching solution. This prevents severe degradation of communication quality due to decision-making errors when exploring unknown interference modes and learning the model, ensuring the basic stability of the communication link and providing a decision-making basis and triggering interface for potential subsequent network-level collaborative suppression. Attached Figure Description

[0018] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0019] Figure 1 This is a flowchart of the steps of the present invention;

[0020] Figure 2 This is a flowchart of the adaptive interference suppression process of the present invention;

[0021] Figure 3 This is a flowchart illustrating the channel state tracking and strategy pre-evaluation steps of the present invention. Detailed Implementation

[0022] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0023] Application Overview:

[0024] In existing wireless communication interference suppression technologies, adaptive methods often rely on preset filter libraries or threshold judgments based on instantaneous channel states, making it difficult to balance the accuracy of suppression with dynamic environmental adaptability. When traditional methods experience rapid, non-stationary changes in the channel, single-decision interference suppression strategies are prone to quickly failing due to environmental mismatch, leading to fluctuations in communication quality. Existing systems lack a mechanism for continuous quantitative verification of the effectiveness of executed strategies. Especially in complex electromagnetic environments with abrupt changes in interference characteristics, fixed strategy libraries or static models may exhibit systematic deviations, failing to meet the anti-interference requirements of highly reliable communication links.

[0025] To address the aforementioned issues, this study discovered a strong correlation between changes in channel characteristics after policy implementation and actual suppression effectiveness (such as signal-to-noise ratio improvement). By establishing a "policy-feature-effect" feedback data model, errors in the achievable model can be identified and labeled. Further investigation revealed that the initial matching policy is highly sensitive to changes but may fail, while the highly robust correction policy, although less flexible, exhibits good stability. Therefore, a strategy of dynamically triggering policy switching and model learning based on effect verification results is proposed.

[0026] Based on the above patterns, a negative sample data package is constructed by creating the complete context (initial state, used strategy, and actual effect) when the strategy fails, and an online incremental learning mechanism is designed to perform closed-loop correction on the model. Further system simulation verification couples the effect verification stage with the model update stage in terms of timing and logic, forming an integrated adaptive enhancement system of "decision-execution-verification-learning".

[0027] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Example:

[0029] like Figure 1 As shown, the core of this invention, a wireless communication interference suppression method based on adaptive matching of channel features, lies in constructing a closed-loop adaptive interference suppression system that integrates signal acquisition, feature processing, intelligent decision-making, and dynamic learning functions. Its operation begins with physical layer signal acquisition, ultimately generating control commands and driving the system's own evolution. Specific implementation methods are described below.

[0030] The system captures wireless communication signals received in the air via a receiving antenna array. The acquired wireless communication signals are preprocessed to obtain time-domain baseband signal data. Specifically, the wireless communication signals are first down-converted to intermediate frequency (IF) or baseband analog signals by an RF front-end module containing a low-noise amplifier, mixer, and local oscillator. Subsequently, a high-speed analog-to-digital converter digitizes the analog signal at a sampling rate no less than twice the signal bandwidth to obtain a digital IF signal. The digital down-conversion module further converts this into in-phase and quadrature baseband signals, i.e., time-domain baseband signal data.

[0031] Channel feature extraction is performed on time-domain baseband signal data. First, a Fast Fourier Transform (FFT) is applied to the signal within a preset time window (e.g., 1 millisecond) to analyze its spectral distribution. Simultaneously, the average power, peak power, and power fluctuation variance of the signal during this time period are calculated. For multi-antenna systems, the phase difference between received signals from different antenna elements also needs to be calculated to estimate the signal angle of arrival. Furthermore, by performing autocorrelation operations on the signal, parameters reflecting the time-domain dispersion characteristics of the channel, such as multipath delay spread, are extracted. Finally, the calculated spectral features, power statistics, spatial features, and time delay features are normalized and vectorized in a preset order and concatenated to generate a multi-dimensional first-state feature vector representing the current instantaneous channel and interference state.

[0032] The first state feature vector is input into a preset interference strategy matching model. Through internal algorithms (such as classification, regression, or similarity matching), the model analyzes the data and outputs a first interference identifier (e.g., a classification code such as "co-channel continuous wave interference," "wideband impulse noise," or "multi-user co-channel interference") corresponding to the current channel interference state, along with a set of strategy parameters bound to it (e.g., for "co-channel continuous wave interference," the parameter set may include the target suppression frequency, notch depth, and bandwidth). Based on the first interference identifier and the set of strategy parameters, a first configuration command is generated and executed. This first configuration command is sent via the control bus to a reconfigurable component in the RF front end, such as a programmable filter, a digital beamforming weight controller, or a power amplifier bias circuit, to complete the initial adaptive configuration of the receiving link, thereby achieving preliminary interference suppression.

[0033] The interference policy matching model can be a lightweight neural network, support vector machine, or template matching-based expert system.

[0034] After a preset time (e.g., 10 milliseconds, to ensure that the suppression measures have taken effect and the channel state has stabilized) following the execution of the first configuration instruction, the system initiates the verification process, reacquires and processes the wireless communication received signal, and generates a second state feature vector. Figure 2 The diagram shown is an adaptive interference suppression flowchart of the present invention. The performance improvement is calculated by comparing the second-state feature vector with the first-state feature vector. The first-state feature vector serves as the initial evaluation benchmark for this decision. The most typical calculation method is to compare the estimated signal-to-noise ratios (SNR) of the two. The improvement is the difference between the SNR after execution and the SNR before execution, and can be selectively normalized by dividing by the noise floor.

[0035] If the improvement is higher than or equal to the threshold, the initial suppression strategy is considered effective, and the system maintains the current configuration. If the performance improvement is lower than the preset improvement threshold, the first interference strategy is deemed ineffective, and the system enters the calibration and learning process. The second state feature vector is input into the preset calibration strategy library for fast matching, and the second interference identifier and associated calibration parameter set are output. The second configuration command is generated and executed to readjust the RF link in an attempt to restore communication.

[0036] An independent correction strategy library is stored in non-volatile memory and contains a series of pre-designed, highly robust and widely applicable basic suppression strategies (such as "omnidirectional reception + time diversity" and "switching to the most conservative filtering mode").

[0037] Simultaneously, the system initiates a learning process. The complete context of this failure event—that is, the failure decision data pair containing the first state feature vector (decision basis), the first interference identifier (the decision made), and the performance improvement degree (the actual effect of the decision)—is encapsulated as a training sample. This training sample is sent to the interference policy matching model via the internal data bus. The model's learning engine (such as an online gradient descent algorithm) uses this sample, combined with possible historical failure samples, to perform small online incremental updates to the model's internal parameters. This means that the model learns from this failure experience and, in the future, when facing similar channel characteristics, reduces the probability of outputting the same failure policy or adjusts its output parameters.

[0038] This invention improves the reliability of single-instance suppression by introducing a mandatory closed loop of "post-execution verification" and "post-failure learning." The verification process can promptly detect policy failures and initiate a highly robust correction scheme, avoiding long-term communication quality degradation caused by policy mismatch. It also enhances long-term environmental adaptability by converting failure cases into training data and updating the core model online, enabling the system to accumulate experience and adaptively optimize its policy matching accuracy. This maintains and continuously improves its anti-interference performance in complex and time-varying electromagnetic environments. Furthermore, it achieves automated operation and maintenance, reducing reliance on external manual parameter adjustments or model retraining, lowering system maintenance costs, and improving deployment flexibility.

[0039] This application further proposes that, in order to construct a feature vector that accurately reflects the current channel and interference state, the system executes a series of parallel digital signal processing algorithms in the feature extraction module. The specific steps for extracting channel features from time-domain baseband signal data and generating the first-state feature vector include:

[0040] A time-domain baseband signal data segment of a preset duration is read from the data buffer. The length of this time window needs to balance time resolution and frequency resolution; a typical example is 1 millisecond, corresponding to the specific configuration of the sampling rate and signal bandwidth. A short-time Fourier transform is performed on the time-domain baseband signal data to obtain the spectral data. A Hamming window or Hanning window is typically used to reduce spectral leakage. The transformed result is the spectral data within the current time window. This spectral data reveals the energy distribution of the signal at different frequency components and is fundamental for identifying narrowband interference, broadband noise, and the signal's own spectral characteristics.

[0041] Next, the system performs subband analysis on the spectral data according to a preset subband division scheme. From the spectral data, the average power of multiple preset subbands is calculated to obtain subband power spectrum data. For example, the entire signal bandwidth can be divided into eight equal-width subbands or non-equal-width subbands based on key frequency bands. The average power of all frequency points within each subband is calculated, thus obtaining a set of subband power spectrum data. This step not only reduces data dimensionality but also highlights the differences in interference levels across different frequency bands. For example, an abnormally high average power in a certain subband may indicate strong narrowband interference in that frequency band.

[0042] Simultaneously, the system performs autocorrelation processing on the same segment of time-domain baseband signal data in parallel, calculating the correlation between the signal and its own time-shifted replicas to obtain the autocorrelation function. By detecting the amplitude and delay position of sidelobes outside the main peak of the autocorrelation function, multipath delay spread data can be extracted, for example, calculating the delay width reaching 10% or 50% of the peak energy of the autocorrelation function. This parameter quantitatively describes the time dispersion characteristics of the channel and is a key indicator for evaluating frequency-selective fading and determining the presence of severe multipath interference.

[0043] In addition, the system also simultaneously acquires or calculates Received Signal Strength Indication (RSSI) data. RSSI is typically provided by the automatic gain control loop of the RF front end or a dedicated power detection circuit, and it reflects the total power level of the received signal.

[0044] To construct a comprehensive, multi-dimensional feature representation, the feature extraction module integrates the sub-band power spectrum data, multipath delay spread data, and received signal strength indication (RSSI) data obtained from the above processing. The sub-band power spectrum data, multipath delay spread data, and RSSI data are vectorized and concatenated to generate a first-state feature vector. Specifically, each data component undergoes normalization preprocessing, such as dividing the sub-band power by the total power, dividing the delay spread value by the maximum tolerable delay, and mapping the RSSI to a preset dynamic range. Subsequently, these normalized scalars or arrays are concatenated in a predefined, fixed order to form a unified, multi-dimensional first-state feature vector. For example, a feature vector containing 8 sub-band powers, 1 delay spread value, and 1 RSSI value has a dimension of 10. This vector serves as a "digital fingerprint" of the channel state, used to drive the matching of intelligent jamming strategies.

[0045] The generated first-state feature vector integrates the signal's frequency domain (subband power spectrum), time domain (multipath delay spread), and power domain (RSSI) information, providing rich and structured input features for subsequent interference identification and strategy matching. It can more precisely and robustly characterize complex composite interference scenarios (such as co-channel superimposed multipath and partial frequency band blocking), thereby providing more accurate judgment criteria for intelligent decision-making models. This fundamentally improves the accuracy of interference type identification and classification, laying a reliable data foundation for implementing precise and effective suppression measures.

[0046] This application further proposes that the first state feature vector be input into a pre-defined interference strategy matching model, and the output be a first interference identifier corresponding to the current channel interference state and a set of associated strategy parameters. The decision-making basis of the interference strategy matching model is a pre-built interference strategy template library. Each template represents a typical interference scenario and its corresponding countermeasures, and contains three key elements: a standard feature vector. A historical average utility score , and the set of strategy parameters bound to it.

[0047] Standard eigenvectors It is obtained by collecting a large number of channel feature samples under specific interference scenarios (such as same-channel continuous wave interference, broadband jamming interference, etc.) and calculating their statistical center (such as the mean vector). It defines the "ideal" or "typical" characteristic pattern of this type of interference. Historical average utility score This is a scalar that is continuously updated during system operation. Its initial value can be set based on laboratory simulations or prior knowledge (e.g., all set to 0.5). It records the average performance improvement of the strategy represented by the template after it has been called in the past, reflecting the long-term reliability of the strategy. The strategy parameter set contains the specific hardware configuration parameters required to execute the suppression strategy, such as the filter center frequency, bandwidth, beamforming weight vector, etc.

[0048] Upon receiving the first state feature vector X from the feature extraction module, the matching process is initiated.

[0049] First, the system iterates through all K templates in the template library. From the pre-defined interference strategy template library, it reads the standard feature vector corresponding to the k-th interference strategy template. and its historical average utility score Where k = 1, 2, ..., K, and K is the total number of templates;

[0050] Calculate the first state feature vector X and each standard feature vector Weighted Euclidean distance As the feature matching degree, the formula is as follows:

[0051]

[0052] Where N is the dimension of the feature vector;

[0053] and These are the first-state feature vector X and the standard feature vector, respectively. The i-th feature component;

[0054] This is the adaptive weight for the i-th feature component, which is dynamically adjusted based on the instantaneous measurement confidence level of that feature component. For example, if the signal-to-noise ratio of a certain sub-band is too low, resulting in a large variance in its power estimation, the system will reduce the weight corresponding to the feature of that sub-band. To reduce the interference of noise on the matching results, higher weights are assigned to features that are stable in measurement and have high confidence. The confidence level can be calculated in real time from signal quality indicators (such as estimation error and number of sampling points) during the feature extraction process.

[0055] Calculate the feature matching degree Then, the system further calculates the comprehensive decision score for selecting the k-th interference strategy template. The formula is as follows:

[0056]

[0057] Here, α and β are preset weighting coefficients used to balance the importance of "current scenario matching degree" and "historical strategy reliability" in decision-making. These two coefficients can be set according to system design requirements. For example, in scenarios that pursue the highest instantaneous performance, α=0.8 and β=0.2 can be set; in scenarios that emphasize stability, α=0.5 and β=0.5 can be set.

[0058] The choice affects the overall decision score. The highest-level interference policy template is identified as the first interference identifier, and its associated policy parameter set is taken as the policy parameter set.

[0059] The training and initialization of the interference strategy matching model are completed before system deployment. A large number of "channel feature vector-optimal suppression strategy and its effectiveness" sample pairs are collected in various typical interference channel environments, either simulated or tested, to form a training dataset. Using this dataset, the initial standard feature vectors of each template are determined through optimization algorithms (such as gradient descent). and initialize its historical utility score. The model uses mean squared error as the loss function to minimize the gap between its predicted performance improvement and the expected improvement of the labeled samples. Training employs mini-batch gradient descent, with batch sizes typically set to 128 or 256. The learning rate uses a warm-up and decay strategy, initially set to 0.001. After a brief warm-up at the beginning of training, it decays exponentially after a certain number of epochs (e.g., 10 epochs) with a decay factor of 0.95 to ensure stable convergence. The number of training epochs is set according to the dataset size and model complexity, typically ranging from 100 to 300 epochs, and an early stopping strategy is used. Training terminates when the validation set loss no longer decreases for several consecutive epochs (e.g., 10 epochs) to prevent overfitting. Through this training, the model can automatically learn the key feature dimensions corresponding to different types of interference, thus naturally focusing on the features most relevant to the current interference state during matching, achieving adaptive adaptation to interference types.

[0060] Through the aforementioned technical aspects, this invention improves the robustness of instantaneous matching in complex noisy environments by dynamically focusing on high-confidence features using adaptive weights. Simultaneously, the introduction of historical utility scoring allows decision-making to draw upon long-term operational experience, avoiding the repeated selection of strategies that have repeatedly failed under similar conditions. This collectively ensures that the system can make better decisions that are both relevant to the current scenario and maintain long-term stability when facing unknown or time-varying disturbances, thereby improving the success rate of the initial suppression strategy and the overall intelligence level of the system.

[0061] This application further proposes that the interference strategy matching model be updated online through the following steps:

[0062] When an initial suppression strategy is determined to be ineffective, the system generates a complete failure decision data pair, the structure of which is (first state feature vector X, first interference identifier A, actual performance improvement degree). ),in This refers to the calculated negative value or a low positive value below a threshold. To improve update stability and avoid overfitting caused by a single sample, the update module does not immediately use this isolated sample. Instead, it merges the failure decision data pair with a preset number of historical failure data pairs stored in a circular buffer to form a mini-batch update dataset. The preset number (e.g., 8, 16, or 32) is an adjustable hyperparameter whose value needs to be determined experimentally based on the processor's computing power and the model's convergence characteristics, balancing update timeliness with gradient estimation stability.

[0063] After obtaining the small batch of updated dataset, the system initiates the model parameter tuning process. The optimization objective of this update is to minimize the gap between the model's predictive performance improvement on this batch of failed data and its measured actual performance improvement.

[0064] Specifically, for each sample (X, A, ...) in the dataset The model needs to be able to predict the performance improvement brought about by the first interference label A based on the input feature X, i.e., predict the performance improvement. .

[0065] In one specific implementation, the historical average utility score corresponding to strategy A can be used. As its predictive performance improvement The proxy or direct association. The goal is to minimize the mean squared error between the prediction strategy and the actual performance improvement of all data pairs in the mini-batch update dataset, adjusting the internal parameters of the interference strategy matching model. The update process employs the gradient descent method, using the mean squared error as the loss function L:

[0066]

[0067] Where M is the total number of samples in the current mini-batch dataset;

[0068] and These represent the improvement in predictive performance and the improvement in actual performance for the m-th sample, respectively.

[0069] The internal parameters of the interference policy matching model are fine-tuned under the drive of this loss function. After a parameter update, the batch of historical failure data can be removed from the buffer or partially retained, while new failure experience can be incorporated to prepare for the next update.

[0070] The online incremental update mechanism of this application endows the interference strategy matching model with the ability to continuously learn from failure experiences. As the system's runtime increases, the probability of the model making the same wrong decisions repeatedly decreases significantly. The model becomes more and more accurate and reliable in dealing with repeated interference scenarios in the environment or those similar to historical failure cases. This enables the system to achieve autonomous and gradual improvement in its anti-interference performance, enhancing its adaptability and robustness in long-term deployment.

[0071] This application further proposes that generating and executing the first configuration instruction based on the first interference identifier and the policy parameter set specifically includes:

[0072] The system first parses the first interference identifier. If it is a frequency domain notch filtering strategy, it extracts key parameters from the strategy parameter set. The strategy parameter set is usually generated by the model based on the current channel characteristics and contains at least two core parameters: the center frequency of the target suppression band (…). The target suppression band bandwidth (BW) and the target suppression band bandwidth (BW) are two parameters that directly define the location and range of the interference signal frequency band that needs to be filtered out in the receiving link. For example, when a strong co-channel interference with a center frequency of 2.412 GHz and a bandwidth of 20 MHz is detected, the model may output the corresponding... =2.412 GHz and BW= 20 MHz.

[0073] After obtaining the above parameters, the corresponding digital filter coefficients are calculated based on the center frequency and bandwidth data of the target suppression band. Digital filters are typically finite impulse response (FIR) or infinite impulse response (IR) filters, and their design goal is to achieve high frequency suppression. The filter exhibits high attenuation within a certain range (e.g., stopband attenuation greater than 40 dB), while maintaining a flat amplitude response and linear phase outside this range (passband) to minimize distortion of the useful signal. The filter coefficients can be calculated in real-time using pre-stored filter design algorithms, such as window function methods (e.g., Kaiser window), frequency sampling methods, or IIR filter design-based methods (e.g., bilinear transform method for Chebyshev or elliptic filters).

[0074] Based on the digital filter coefficients, a first configuration instruction for configuring the programmable filter in the receiving link is generated and executed. This first configuration instruction is a standardized hardware control instruction set, sent to the programmable filter hardware unit in the receiving link via a system bus (such as SPI, I2C, or parallel bus). The instruction directly contains the calculated filter coefficient array, as well as configuration information such as the possible filter order and operating mode. After receiving and parsing this instruction, the new filter coefficients are immediately loaded, and its internal structure is updated, thereby efficiently suppressing interference components in the received signal in real time.

[0075] Through the complete closed-loop process described above, from strategy parameter analysis and real-time filter coefficient calculation to hardware configuration, this invention achieves precise and rapid implementation of interference suppression strategies from intelligent decision-making to physical layer execution. Compared to traditional solutions using fixed filter banks, this method can dynamically generate and load tailored filter parameters based on the real-time identified interference spectrum characteristics, thereby achieving precise targeting and efficient suppression of interference frequency bands. Simultaneously, it minimizes the risk of accidental damage or residual interference to useful signals caused by mismatches in filter bandwidth and center frequency. This adaptive capability of hardware and software collaboration significantly improves the effective signal-to-noise ratio and communication quality of the system in environments with complex and dynamic spectrum interference.

[0076] like Figure 3 The diagram illustrates the channel state tracking and policy pre-evaluation steps of this invention. This application further proposes that, after generating and executing the first configuration instruction, the following channel state tracking and policy pre-evaluation steps are included, specifically:

[0077] Within a preset microsecond time slot (e.g., 5 to 20 milliseconds) after executing the first configuration instruction, it operates continuously at a rate much higher than the main loop (e.g., once every 1 millisecond), continuously acquiring and processing wireless communication received signals to generate a series of short-time state feature sequences. Each element in the short-time state feature sequence is a simplified state feature vector, which may only contain key dimensions such as signal-to-noise ratio and main interference subband power to ensure processing speed. Assume that the sequence contains L consecutive vector samples. .

[0078] Based on a short-time state feature sequence, the instantaneous gradient vector G of the channel state is calculated. This instantaneous gradient vector G is used to quantify the rate and direction of change of each key feature within a microsecond time slot. A specific calculation method is to perform linear fitting or difference calculation on the feature vectors in the sequence along the time dimension. For example, for the i-th feature dimension, its linear regression slope on that sequence can be calculated as the component of the gradient vector in that dimension. A simpler implementation can be achieved using first-order forward difference:

[0079]

[0080] in, This represents the i-th component of the t-th short-time eigenvector;

[0081] T represents the total time of the micro-slot (in seconds).

[0082] The resulting instantaneous gradient vector G reflects the instantaneous change trend of the channel in each characteristic dimension. For example, if G is a large positive number in the power dimension of a certain sub-band, it indicates that the interference intensity of that frequency band is increasing rapidly.

[0083] The system maintains a policy robustness mapping table, which uses the first interference identifier as the primary key and stores the corresponding policy robustness boundary vector B. The instantaneous gradient vector G is compared with the preset policy robustness boundary vector B corresponding to the first interference identifier. This preset policy robustness boundary vector B is a threshold vector pre-defined for each interference policy (such as "frequency domain notch filtering policy A"). It defines the upper limit of the channel change rate that the policy can effectively cope with. These boundary values ​​are determined through offline simulation, theoretical analysis, or stress test data in typical scenarios.

[0084] If the absolute value of the component of the instantaneously changing gradient vector G in at least one feature dimension i Components exceeding the policy robustness boundary vector B (Right now If the current channel is rapidly changing in a direction that exceeds the adaptability of the first interference strategy, then it is determined that the current channel is changing rapidly in a direction that exceeds the adaptability of the first interference strategy.

[0085] Once a decision is triggered, the system immediately performs two operations. First, it generates a policy switching warning command. This command, as a high-priority event, is sent to the system's main control unit, indicating that the main loop may soon need to initiate a correction process, prompting relevant modules to prepare their states in advance. Second, and more importantly, the system uses the currently calculated instantaneous gradient vector G as valuable prior knowledge and injects it into the matching calculation process of the correction policy library in advance. This means that when the main loop subsequently triggers correction due to insufficient performance improvement, the matching algorithm of the correction policy library can receive not only the second state feature vector representing the current absolute state, but also this vector G reflecting the recent trend of change. The correction policy library can use this information to prioritize candidate policies that are more robust to similar trends (such as rapid increase in interference or rapid frequency drift), or to make predictive adjustments to policy parameters (e.g., reserving a wider suppression bandwidth for interference that may drift further), thereby significantly improving the hit rate and timeliness of the correction policy.

[0086] This application, by monitoring the channel change rate in real time, can provide early warning and prepare better alternatives before interference develops to the point of causing serious degradation in communication quality. This allows the system to transition more smoothly and quickly from a potentially ineffective current strategy to a more suitable correction strategy, effectively suppressing a precipitous drop in communication quality caused by untimely strategy switching. Especially in harsh scenarios such as high-speed movement and rapid changes in interference sources, it improves the instantaneous reliability and overall stability of the communication link.

[0087] This application further proposes to calculate the performance improvement between the second state feature vector (after policy execution) and the first state feature vector (before policy execution), where the first state feature vector represents the initial state before executing the first configuration instruction, serving as a unified evaluation benchmark before and after this policy execution. Specific steps include:

[0088] The calculation of performance improvement revolves around the signal-to-noise ratio (SNR), which best comprehensively reflects the quality of the communication link. First, the system extracts the corresponding SNR estimates from the first-state feature vector and the second-state feature vector, respectively, denoted as the first SNR estimate (…). ) and the second signal-to-noise ratio estimate ( The ratio (in dB) is usually obtained by calculating the ratio of the received signal power to the real-time estimated noise power.

[0089] Calculate the difference between the second signal-to-noise ratio estimate and the first signal-to-noise ratio estimate, i.e. This original difference directly reflects the change in the signal-to-noise ratio.

[0090] However, since the ambient noise floor itself may fluctuate, and the performance improvement represented by the same magnitude of signal-to-noise ratio change is different under different environments (for example, a 3dB improvement in communication capacity under extremely low signal-to-noise ratio is different from a 3dB improvement under high signal-to-noise ratio), normalization is required to obtain a more universal and comparable evaluation index and a normalized performance improvement.

[0091] Therefore, the normalized performance improvement Calculated using the following formula:

[0092]

[0093] This is a noise floor estimate, representing the typical noise power level of the system under the current operating frequency band and configuration. Its unit is consistent with the signal-to-noise ratio estimate (e.g., dB). This value is not a fixed constant and can be obtained through methods such as: 1) long-term measurements during system initialization or quiet periods (when no signal is transmitted) and taking the statistical average; 2) calculations based on the receiver's noise figure and thermal noise formulas; 3) preset empirical values ​​based on environmental type (e.g., indoor, suburban). For example, a typical... The value may be around -100 dBm, depending on the system parameters.

[0094] The physical meaning of this formula lies in its correlation between the actual improvement in signal-to-noise ratio and the system's inherent noise level. If... and If they are on the same order of magnitude, then A value close to 1 indicates a significant improvement; if much smaller ,but The value is very small, indicating a weak improvement. The final calculated value is... It is a dimensionless scalar, which will serve as the core indicator for evaluating the effectiveness of a strategy.

[0095] The system will improve this performance. With a preset improvement threshold The comparison is used to determine whether the strategy is "successful" or "failed". A preset improvement threshold is set. The range is typically between 0.3 and 0.8; for example, it can be set to... =0.5. This means that the strategy is considered sufficiently effective only when the improvement in signal-to-noise ratio after implementation reaches more than half of the background noise power level.

[0096] By employing the aforementioned quantitative evaluation method based on the improvement in normalized signal-to-noise ratio, this invention eliminates evaluation bias caused by differences in environmental noise levels, ensuring consistency and fairness in the evaluation criteria for strategy effectiveness across different scenarios and time periods. This standardized evaluation result provides highly reliable and comparable data input for subsequent strategy failure determination, correction strategy triggering, and online model learning, ensuring that the decision-making and learning processes of the entire adaptive closed-loop system are built on a solid and consistent performance metric foundation. This effectively enhances the scientific rigor of the system optimization process and the stability of overall performance.

[0097] This application further proposes that the method for constructing and updating the correction strategy library includes:

[0098] A set of basic strategies is established, which includes a series of pre-designed, general interference suppression schemes designed to broadly address various uncertainties. Each basic strategy in the set is bound to a pre-defined, highly robust set of RF configuration parameters.

[0099] During operation, the system monitors the execution success rate of all strategies output by the interference strategy matching model (including strategies called from the correction strategy library) in real time. This monitoring is comprehensive, not only recording whether a strategy is deemed ineffective (performance improvement below a threshold), but also precisely recording the actual performance improvement (Δ) calculated after each call. For each basic strategy in the correction strategy library, the system maintains an independent sliding window counter or historical record queue to track its performance during its most recent calls.

[0100] The core logic of dynamic evolution lies in the strategy promotion mechanism. When the actual performance improvement of any basic strategy exceeds a preset success threshold in a predetermined number of calls N (between 3 and 10 times, for example, N=5 times), the success rate is determined by the strategy's promotion mechanism. At that time, the system will perform a "migration" operation.

[0101] N needs to strike a balance between avoiding accidental promotions and promptly rewarding effective strategies; a preset success threshold should be set. It is usually set to a value higher than the failure determination threshold. The value (for example, =0.7, and =0.5), the threshold is set based on a large amount of historical data or simulation analysis, and aims to screen out truly efficient strategies.

[0102] Once the promotion criteria are met, the base strategy and its current parameters are migrated from the correction strategy library to the strategy candidate set of the interference strategy matching model. The strategy candidate set is the source of strategies for matching and selection by the main decision model. This means that the template library or decision logic of the main model will be updated accordingly to incorporate this new, proven effective strategy option.

[0103] This application achieves the accumulation and solidification of knowledge from emergency experience to routine decision-making by transferring repeatedly validated strategies from the correction strategy library to the main decision model. This not only enriches the strategy option library of the main model for dealing with complex scenarios and enhances the potential optimality of its initial decision, but also enables the overall interference suppression capability of the system to continuously evolve and strengthen with the increase of running time, effectively overcoming the limitations of traditional system strategy libraries that are fixed and rigid and cannot be autonomously optimized from successful experiences.

[0104] This application further proposes that generating and executing the second configuration instruction specifically includes:

[0105] In the correction process triggered after the primary suppression strategy fails, the system first parses the second interference identifier. If the identifier is a spatial beamforming strategy, the estimated interference direction of arrival (ADI) data is read from the correction parameter set. This ADI data is usually expressed in angle form, such as the azimuth angle θ and elevation angle φ relative to the antenna array normal. This ADI prediction may come from several sources: 1) a preliminary estimate of the interference direction obtained through array signal processing algorithms (such as MUSIC, ESPRIT, or ADI estimation) during feature extraction before executing the initial strategy; 2) a typical direction preset in the correction strategy library for the type of interference represented by the identifier (such as "strong lateral interference"); and 3) a predictive adjustment made by combining the directional change trend information implicit in the instantaneously changing gradient vector G.

[0106] Based on the predicted direction of the interference wave Using the geometric data of the antenna array, calculate the array weighting vector used to form a null in the receiving beam in that direction.

[0107] Geometric data are pre-stored parameters describing the three-dimensional spatial position of each antenna element in the array. For a uniform linear array, this can be the element spacing d; for a planar array, it includes the coordinates of each element in the xy-plane. Based on this information and the wavelength λ, the system can calculate the target direction for each antenna element relative to the array reference point. Spatial phase delay.

[0108] The goal of beamforming is to synthesize a set of complex weights (i.e., the array weighting vector), applied to the received signal of each antenna element, so that the array synthesized radiation pattern is... Nulls are formed in the direction of interference. A linearly constrained minimum variance criterion is used: under the constraint that the gain in the direction of the useful signal (usually set to the desired signal direction or omnidirectional) is constant, minimizing the total power of the array output (i.e., interference plus noise power) will naturally result in a null in the direction of interference. The calculated weight vector... That is the desired result, where each They are all complex numbers, containing amplitude and phase adjustment information.

[0109] Based on the array weighting vector The system generates and executes a second configuration instruction for configuring the beamforming network in the receive link. This second configuration instruction is sent to the beamforming network hardware in the receive link via a high-speed control interface (such as JESD204B, high-speed SPI, or parallel bus). This hardware is typically a digital beamformer integrated into the RF front-end or intermediate frequency processing unit, consisting of a network of programmable complex multipliers and adders. The instruction content includes the calculated complex weight vector. After the hardware loads the new weights, it performs a weighted summation of the received signals from each antenna channel in real time, thereby physically achieving effective spatial suppression of interference in the specified direction.

[0110] By implementing the aforementioned spatial beamforming correction strategy, this application directly suppresses interference along the physical propagation path. When the primary strategy (such as frequency domain notch filtering) fails, this correction strategy serves as an efficient alternative or supplementary means, specifically creating a receiving null in the direction of incoming interference. This enhances the system's survivability and communication quality in complex multipath environments and environments with strong directional interference sources, constituting an indispensable key technical component in the closed-loop adaptive interference suppression system.

[0111] This application further proposes that, after generating and executing the second configuration instruction, it also includes:

[0112] After executing the second configuration instruction, the system waits for a second preset duration. This duration is typically slightly longer than the verification wait time after the initial suppression, for example, set to 20 to 50 milliseconds, in order to provide sufficient stabilization and activation time for complex spatial or combined correction strategies.

[0113] After a second preset time period, the system acquires and processes the wireless communication received signal for the third time, generating a third state feature vector. This third state feature vector reflects the latest channel state after the correction strategy has been implemented and the channel has stabilized after the second preset time period.

[0114] Calculate the normalized quadratic performance improvement between the third-state eigenvector and the second-state eigenvector. Extract the corresponding signal-to-noise ratio estimate, and denote it as the second signal-to-noise ratio estimate. ) and the third signal-to-noise ratio estimate ( ), second signal-to-noise ratio estimate ( This serves as the initial evaluation benchmark for the implementation of the correction strategy. The formula is as follows:

[0115]

[0116] Calculated quadratic performance improvement Similarly, with the preset improvement threshold (For example, Compare with (=0.5). If If so, it indicates that the correction strategy is effective and the system will continue to maintain the current configuration; if If the current correction strategy fails, it is determined that the current correction strategy has failed. This means that even if a highly robust correction strategy is enabled, the communication quality has not been effectively improved. The interference may be too complex (such as distributed or coordinated interference) or the environment may be too harsh, exceeding the limits of the local processing capabilities of a single device.

[0117] Once secondary correction is determined to have failed, the system will generate cooperative suppression request data. This data is a structured packet designed to report interference situations that cannot be resolved locally and request assistance to the network side (such as base stations, access points, or network controllers). Its content typically includes: 1) local device identifier; 2) current channel state summary, such as key parts of the third-state feature vector or identified interference features; 3) history of attempted suppression strategies, such as first and second interference identifiers; and 4) performance measurement results, i.e., the degree of improvement after the two attempts. and 5) Timestamp and location information (if available).

[0118] After generating the request data, the system sends the cooperative suppression request data to the network-side cooperative scheduling entity (such as a base station in an LTE / 5G network, an access point controller in a Wi-Fi network, or a cluster head node in an ad hoc network) via the device's uplink control channel or dedicated signaling interface. Upon receiving such a request from one or more nodes, the network-side entity will initiate cross-channel or cross-node joint interference suppression coordination. This may include, but is not limited to: dynamically allocating cleaner channels or frequency bands for the interfered device; coordinating adjacent base stations / access points to perform cooperative beamforming to jointly suppress interference sources in space; scheduling multiple nodes within the network to perform interference alignment or cooperative transmission; and even adjusting network-level power control or resource block allocation strategies.

[0119] By introducing the aforementioned collaborative suppression request mechanism, this application improves the survival probability and mission reliability of a single point in unpredictable and harsh electromagnetic environments. It also provides valuable real-time data input for the entire network to perform dynamic optimization, spectrum management, and interference coordination by feeding local information back to the network. This enhances the overall robustness, spectrum utilization, and quality of service of the wireless network, and elevates the anti-interference capability from "device autonomy" to "network co-governance".

[0120] The following is a specific embodiment of a wireless communication interference suppression method based on adaptive matching of channel features:

[0121] The LoRa wireless sensor network deployed in an industrial park suffered from narrowband interference from inverters, broadband noise from welding machines, and multipath reflections from metal equipment, resulting in a packet loss rate of over 30%. The LoRa signals captured by the receiving antennas of the sensor nodes were processed by the radio frequency front-end (low-noise amplifier gain of 20dB, mixer down-converted to intermediate frequency of 1MHz), and then digitized by a 12-bit ADC at a sampling rate of 250kHz (signal bandwidth of 125kHz, satisfying the Nyquist criterion) to obtain time-domain baseband signal data.

[0122] After preprocessing, the system extracts multi-dimensional channel features from the time-domain baseband signal. A 1ms Hamming window (50% overlap) is used to perform a short-time Fourier transform (STFT), dividing the signal into 8 equal-width subbands (each subband 15.625kHz). The average power of each subband is calculated to obtain the normalized subband power spectrum vector [0.12, 0.85, 0.08, 0.05, 0.03, 0.02, 0.01, 0.01]. The multipath delay is extended to 150ns through autocorrelation processing. The RSSI value of the RF front-end output is -75dBm. Finally, the signals are concatenated to form the first state feature vector X=[0.12, 0.85, 0.08, 0.05, 0.03, 0.02, 0.01, 0.01, 150ns, -75dBm].

[0123] The system calls the preset interference strategy template library, which contains template 1 (standard feature vector) corresponding to narrowband interference. =[0.05,0.90,0.03,...], Historical Utility Score =0.8), Template 2 corresponding to broadband noise ( =[0.20,0.20,0.20,...], Template 3 corresponding to multipath interference (T3=[0.10,0.10,0.10,...]) To highlight the importance of interference frequency domain characteristics, sub-band power component weights are set. =0.8, multipath delay weight =0.1, RSSI weight =0.1, calculate the weighted Euclidean distance between the first state feature vector X and template 1: Similarly, The overall decision score was then calculated using a matching degree weight of α=0.7 and a historical utility weight of β=0.3. , , Select the template with the highest score, and output the narrowband interference identifier and strategy parameters (center frequency 868.1MHz, bandwidth 10kHz).

[0124] The system calls the Kaiser window method to design an FIR filter (with a stopband attenuation of 40dB) and generates a coefficient array. Configuration commands are sent to the programmable filter via the SPI bus to filter out interference in the 868.1MHz±5kHz frequency band in real time. After 10ms, the signal is reacquired, and the second-state feature vector is extracted, showing that the power of the second sub-band has decreased to 0.15, and the SNR has increased from 8dB to 12dB. The performance improvement is calculated. ( Due to the preset improvement threshold =0.5 and If the strategy fails (interference source frequency drifts to 868.2MHz), the correction process is triggered to read the "Dynamic Frequency Domain Notch" strategy (with parameters of center frequency 868.2MHz and bandwidth 15kHz) from the correction strategy library and generate a second configuration command to adjust the filter parameters.

[0125] Finally, the system will assign the currently invalid data pair (X, Template 1, ...) to the current invalid data pair (X, Template 1, ...). =0.4) and five historical failure data points were merged into a mini-batch dataset. Using the mean squared error as the target, the standard feature vector of template 1 was adjusted (subband power components updated to 0.06, 0.88,...) and the historical utility score (reduced to 0.75) to complete the incremental model update. After applying this method, the packet loss rate of the park's sensor network decreased to below 5%, the stability of the communication link was significantly improved, the system can track interference changes in real time and automatically switch to the optimal suppression strategy, and continuously optimize the decision model through online learning, effectively solving the problem that fixed strategies cannot cope with time-varying interference, and meeting the high-reliability communication requirements of industrial parks.

[0126] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A wireless communication interference suppression method based on adaptive matching of channel features, characterized in that: include: The wireless communication received signal is acquired and preprocessed to obtain time-domain baseband signal data; Channel features are extracted from the time-domain baseband signal data to generate a first state feature vector; The first state feature vector is input into a preset interference strategy matching model, and the first interference identifier and associated strategy parameter set corresponding to the current channel interference state are output. The first configuration instruction is generated and executed based on the first interference identifier and strategy parameter set. After a preset time following the execution of the first configuration instruction, the wireless communication received signal is reacquired and processed to generate a second state feature vector; Calculate the performance improvement between the second state feature vector and the first state feature vector; If the performance improvement is lower than the preset improvement threshold, the first interference strategy is determined to be ineffective, the second state feature vector is input into the preset correction strategy library, the second interference identifier and the associated correction parameter set are output, and the second configuration instruction is generated and executed. Simultaneously, the failure decision data pair, which includes the first state feature vector, the first interference identifier, and the performance improvement degree, is sent to the interference strategy matching model for online incremental updates.

2. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: The specific steps for extracting channel features from the time-domain baseband signal data and generating a first state feature vector include: Perform a short-time Fourier transform on the time-domain baseband signal data to obtain the spectrum data; The average power of multiple preset sub-bands is calculated from the spectrum data to obtain sub-band power spectrum data; Simultaneously, autocorrelation processing is performed on the time-domain baseband signal data to obtain multipath delay spread data; The subband power spectrum data, the multipath delay spread data, and the received signal strength indication data are vectorized and concatenated to generate the first state feature vector.

3. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: The specific steps for inputting the first state feature vector into a preset interference strategy matching model and outputting the first interference identifier and associated strategy parameter set corresponding to the current channel interference state include: Obtain the first state feature vector X; Read the standard feature vector corresponding to the k-th interference strategy template from the preset interference strategy template library. and its historical average utility score Where k = 1, 2, ..., K, and K is the total number of templates; Calculate the first state feature vector X and each standard feature vector Weighted Euclidean distance As the feature matching degree, the formula is as follows: Where N is the dimension of the feature vector. and These are the first state feature vector X and the standard feature vector, respectively. The i-th feature component, The adaptive weight is the weight of the i-th feature component, which is dynamically adjusted according to the instantaneous measurement confidence of the feature component. Calculate the comprehensive decision score for selecting the k-th interference strategy template. The formula is as follows: Where α and β are preset weighting coefficients; The choice results in the comprehensive decision score. The highest interference policy template is identified as the first interference identifier, and its associated policy parameter set is identified as the policy parameter set.

4. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: The interference strategy matching model is updated online through the following steps: The failure decision data pairs are merged with a preset number of historical failure data pairs to form a small-batch update dataset. The internal parameters of the interference strategy matching model are adjusted with the goal of minimizing the mean square error between the predicted performance improvement and the actual performance improvement of all data pairs in the mini-batch update dataset.

5. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: Generating and executing the first configuration instruction based on the first interference identifier and policy parameter set specifically includes: The first interference identifier is parsed. If it is a frequency domain notch filtering strategy, the center frequency and bandwidth data of the target suppression frequency band are read from the strategy parameter set. Calculate the corresponding digital filter coefficients based on the center frequency and bandwidth data of the target suppression band; Based on the digital filter coefficients, a first configuration instruction for configuring the programmable filter in the receive link is generated and executed.

6. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: After generating and executing the first configuration instruction, the process also includes channel state tracking and policy pre-evaluation steps, specifically: Within a preset microsecond time slot after executing the first configuration instruction, wireless communication received signals are continuously acquired and processed to generate a series of short-term state feature sequences; Based on the short-time state feature sequence, the instantaneous change gradient vector G of the channel state is calculated; The instantaneous gradient vector G is compared with the preset policy robustness boundary vector B corresponding to the first interference identifier; If the component of the instantaneous gradient vector G in at least one dimension exceeds the component corresponding to the policy robustness boundary vector B, then it is determined that the current channel is rapidly changing in a direction that exceeds the adaptability of the first interference policy. Based on this determination, a strategy switching warning instruction is generated, and the instantaneously changing gradient vector G is used as prior knowledge and injected into the matching calculation process of the correction strategy library in advance.

7. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: The specific steps for calculating the performance improvement between the second state feature vector and the first state feature vector include: Extract the first signal-to-noise ratio estimate from the first state feature vector; Extract the second signal-to-noise ratio estimate from the second state feature vector; The difference between the second signal-to-noise ratio estimate and the first signal-to-noise ratio estimate is calculated, and the difference is normalized with a preset noise floor estimate to obtain the normalized performance improvement.

8. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: The method for constructing and updating the correction strategy library includes: Establish a set of basic strategies, wherein each basic strategy in the set of basic strategies is bound to a set of preset, highly robust radio frequency configuration parameters; Real-time monitoring of the execution success rate of all strategies output by the interference strategy matching model; When the actual performance improvement of any basic strategy is higher than the preset success threshold in a preset number of consecutive calls, the basic strategy and its current parameters are migrated from the correction strategy library to the strategy candidate set of the interference strategy matching model.

9. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: Generating and executing the second configuration instruction specifically includes: The second interference identifier is analyzed. If the identifier is a spatial beamforming strategy, the estimated interference direction data is read from the set of correction parameters. Based on the estimated direction of interference and the geometric data of the antenna array, calculate the array weighting vector for forming a null in the receiving beam in that direction; Based on the array weighting vector, a second configuration instruction for configuring the beamforming network in the receive link is generated and executed.

10. The wireless communication interference suppression method based on adaptive matching of channel features according to claim 1, characterized in that: After generating and executing the second configuration instruction, the following is also included: After a second preset time period, the wireless communication received signal is acquired and processed for the third time to generate a third state feature vector; Calculate the quadratic performance improvement degree between the third state feature vector and the second state feature vector; If the secondary performance improvement is still lower than the preset improvement threshold, the current correction strategy is determined to be ineffective, and collaborative suppression request data is generated. The coordinated suppression request data is sent to the network-side coordinated scheduling entity to request the initiation of joint interference suppression across channels or nodes.

Images