Radio frequency broadband signal receiving and processing method based on interference mode recognition

Abstract

The invention relates to the technical field of signal processing, and discloses a radio frequency broadband signal receiving and processing method based on interference pattern recognition, which aims to solve the problems of high power consumption, low recognition precision and no collaborative optimization among domains in the existing interference processing scheme for radio frequency broadband signal receiving in the prior art. Simulation domain interference pre-recognition is achieved through the memristor cross array, fine recognition is combined with a digital domain, low-power-consumption sampling is achieved in cooperation with an event-driven ADC, physical front anti-interference is completed by means of a programmable metasurface, and electromagnetic energy can be collected to achieve energy self-consistency; the recognition precision is improved through space-time frequency three-dimensional reconstruction and dynamic dictionary learning, a memristor primary function can be updated online to adapt to unknown interference, a collaborative optimization mechanism across radio frequency, analog and digital domains is constructed, the interference suppression ratio and recognition efficiency are greatly improved, and system performance self-optimization and environment self-adaption are achieved.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a radio frequency broadband signal receiving and processing method based on interference pattern recognition. Background Technology

[0002] With the rapid development of wireless communication technology, radio frequency spectrum resources are becoming increasingly congested. Various communication, radar, and navigation signals coexist with man-made interference and environmental electromagnetic noise in the same frequency band, posing complex electromagnetic environment challenges to radio frequency broadband signal receiving systems. At the same time, the upgrading of electronic countermeasures technology has led to increasingly diversified forms of interference, covering various types such as single-tone interference, narrowband interference, broadband interference, and pulse interference. Moreover, parameters such as the power, frequency, and direction of arrival of the interference signal exhibit dynamic changes, placing extremely high demands on the anti-interference capabilities of radio frequency signal receivers.

[0003] The aforementioned and existing related technologies often suffer from the following drawbacks: Current interference handling methods for broadband RF signal reception mostly adopt the traditional architecture of full-bandwidth digitization after down-conversion. This means the RF front-end directly digitizes the signal across the entire bandwidth using a high-speed ADC, and then performs interference identification and suppression in the digital domain. This architecture has several technical limitations: First, high-speed full-bandwidth sampling results in extremely high hardware power consumption and data processing volume, making it particularly unsuitable for portable, low-power RF receiving devices. Second, digital domain interference identification is mostly based on single-dimensional signal feature analysis, lacking the ability to jointly process space, time, and frequency. This leads to low identification accuracy in low signal-to-noise ratio and multi-interference scenarios, and poor adaptability to unknown interference patterns. Third, the RF, analog, and digital domains of interference handling are independent of each other, lacking a collaborative optimization mechanism. The front-end hardware cannot dynamically adjust its operating state based on the interference identification results, resulting in low interference suppression ratio, high signal distortion, limited input dynamic range, and a tendency for ADC saturation. Summary of the Invention

[0004] The technical problem to be solved by this invention is that existing interference processing schemes for radio frequency broadband signal reception have drawbacks such as high power consumption, low recognition accuracy, and lack of inter-domain collaborative optimization. To address this, we propose a radio frequency broadband signal reception processing method based on interference pattern recognition.

[0005] To achieve the above objectives, this application adopts the following technical solution: a radio frequency broadband signal receiving and processing method based on interference pattern recognition, comprising the following steps: S1: Radio frequency front-end receiver, which captures the radio frequency broadband signal through the receiving antenna, and obtains the intermediate frequency analog signal through low noise amplification and down-conversion processing; S2: Analog domain interference pre-identification: The intermediate frequency analog signal is directly input to the memristor cross array. The interference mode basis function characterized by the conductance of the memristors in the array is multiplied and added in parallel in the analog domain with the input signal. The analog current characterizing the interference mode matching degree is output, and the analog current is converted into a digital pulse sequence. The pulse characteristics of the digital pulse sequence are used to identify the preliminary type and intensity of the interference. S3: Digitization and preprocessing: The signal processed by the memristor cross array is sampled and digitized to obtain a digital signal, and then down-converted, filtered and synchronized to obtain a regular baseband signal. S4: Fine feature extraction and recognition: Perform transform domain analysis on the regular baseband signal, and combine it with the preliminary results identified by the digital pulse sequence to extract fine features. Then, complete the final identification and parameter estimation of the interference mode through a classification algorithm. S5: Adaptive interference suppression. Based on the finally identified interference pattern, a corresponding suppression strategy is generated to eliminate interference in the signal in the digital domain. S6: Signal recovery and output, compensates and equalizes the suppressed signal, and recovers and outputs the target signal.

[0006] Preferably, in the analog domain interference pre-identification step, the operation of the memristor cross array specifically includes: The row and column lines of the memristor cross array are connected by memristors, and the intermediate frequency analog signal is applied to the column lines; The conductance of the memristor is pre-programmed as a basis function for a variety of typical interference modes; According to Ohm's law and Kirchhoff's law, the current flowing into each row line is the result of the analog multiplication and addition of the input signal and each basis function; The analog current is converted into a sparse pulse sequence via a spiking neuron circuit, wherein the pulse frequency is encoded with matching strength and the pulse timing is encoded with pattern category confidence.

[0007] Preferably, the digitization and preprocessing steps use an event-driven ADC for sampling, and its sampling rate and start time are triggered and controlled by the digital pulse sequence. High-precision sampling and subsequent processing are initiated only when the pulse sequence indicates the presence of potential interference; otherwise, a low-power monitoring state is maintained.

[0008] Preferably, the fine feature extraction and recognition step adopts a spatiotemporal-frequency three-dimensional joint processing framework, specifically as follows: The regularized baseband signal of the multi-antenna channel is constructed as a three-dimensional tensor of space, time, and frequency; A compressed sensing algorithm based on structured sparse Bayesian learning is used to reconstruct the complete spatiotemporal spectrum from the random subsampled data of the three-dimensional tensor, which is less than 5%. By utilizing a dynamic dictionary learning algorithm, the regularization parameters and sparse constraints of dictionary atoms in the reconstruction algorithm are adaptively adjusted based on the prior information provided by the digital pulse sequence, in order to focus on the interference mode region.

[0009] Preferably, in the RF front-end receiving step, the signal passes through a programmable metasurface before being input to the memristor cross array. The method further includes: Based on the preliminary interference identification results provided by the digital pulse sequence, the phase response of each unit of the programmable metasurface is dynamically adjusted; The metasurface forms a deep null in the direction of the incoming interference wave, and the physical level interference suppression is used as a pre-processing step to reduce the input dynamic range requirements of the subsequent memristor array and digital processing module.

[0010] Preferably, the method is driven by harvesting ambient electromagnetic energy, specifically including: Energy in the radio frequency broadband signal is collected by a broadband rectified antenna; The collected energy is used to power the memristor cross array and analog processing circuitry; The system's power scheduler adaptively switches the power supply of subsequent digital processing modules on and off according to the density of the digital pulse sequence, thereby achieving energy-consistent interference processing.

[0011] Preferably, the fine feature extraction and recognition step employs a biomimetic auditory pulse neural network model, specifically as follows: The regularized baseband signal is decomposed into subband signals using a cochlear filter bank model; Each sub-band signal, along with the digital pulse sequence, is input into a pulse neural network. The spiking neural network learns the spatiotemporal correlation between interference pulses and subband signal features through pulse temporal dependence plasticity rules, and completes the classification.

[0012] Preferably, based on the interference mode and its parameters finally identified in step S4, an optimized configuration instruction is dynamically generated and fed back to the RF front-end and memristor cross array for real-time adjustment of their hardware operating status. The specific process of dynamically adjusting the memristor cross array includes: When the interference mode identified by S4 is an unknown new mode, the corresponding normalized basis function vector is calculated and generated based on its spatiotemporal characteristic vector. The basis function vector is mapped to a specific voltage pulse sequence; The voltage pulse sequence is applied to a preset redundant row of the memristor cross array to change the conductance value of the memristor in that row, so as to write the new interference mode basis function into the hardware online; Update the system's base function-row index mapping table to associate the new interference pattern with the redundant row.

[0013] Preferably, the specific process of dynamically adjusting the RF front end includes controlling the programmable metasurface: Based on the interference wave direction and center frequency parameters provided by S4, calculate the required phase offset for each unit of the metasurface; The phase offset is quantized into a digital phase control word; The phase control word is sent to the driving circuit of the programmable metasurface via the control bus, so that the metasurface forms an adaptive deep null in the direction of the interference wave.

[0014] Preferably, the specific process of dynamically adjusting the radio frequency front end also includes: Based on the interference power level parameters provided by S4, the analog voltage is output through the digital-to-analog converter, and the gain bias point of the low-noise amplifier is dynamically adjusted to prevent the subsequent analog-to-digital converter from saturating. Based on the interference bandwidth parameters provided by S4, the bandwidth of the variable bandwidth filter is controlled through the serial peripheral interface to pre-filter the interference.

[0015] The technical effects and advantages of this invention are as follows: In this invention, analog domain interference pre-identification is achieved through a memristor cross array, combined with digital domain fine identification. An event-driven ADC is used to achieve low-power sampling. A programmable metasurface is used to complete physical pre-interference suppression and can also collect electromagnetic energy to achieve energy self-consistency. The recognition accuracy is improved through spatiotemporal-frequency three-dimensional reconstruction and dynamic dictionary learning. The memristor basis function can be updated online to adapt to unknown interference. A collaborative optimization mechanism across the radio frequency, analog, and digital domains is constructed, which greatly improves the interference suppression ratio and recognition efficiency, and realizes system performance self-optimization and environmental adaptation. Attached Figure Description

[0016] 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: Figure 1 This is a flowchart of the radio frequency broadband signal receiving and processing method based on interference pattern recognition according to the present invention. Detailed Implementation

[0017] 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.

[0018] Reference Figure 1As shown, the present invention provides a technical solution: a radio frequency broadband signal receiving and processing method based on interference pattern recognition, comprising the following steps: S1: Radio frequency front-end receiver, which captures the radio frequency broadband signal through the receiving antenna, and obtains the intermediate frequency analog signal through low noise amplification and down-conversion processing; S2: Analog domain interference pre-identification: The intermediate frequency analog signal is directly input to the memristor cross array. The interference mode basis function characterized by the conductance of the memristors in the array is multiplied and added in parallel in the analog domain with the input signal. The analog current characterizing the interference mode matching degree is output, and the analog current is converted into a digital pulse sequence. The pulse characteristics of the digital pulse sequence are used to identify the preliminary type and intensity of the interference. S3: Digitization and preprocessing: The signal processed by the memristor cross array is sampled and digitized to obtain a digital signal, and then down-converted, filtered and synchronized to obtain a regular baseband signal. S4: Fine feature extraction and recognition: Perform transform domain analysis on the regular baseband signal, and combine it with the preliminary results identified by the digital pulse sequence to extract fine features. Then, complete the final identification and parameter estimation of the interference mode through a classification algorithm. S5: Adaptive interference suppression. Based on the finally identified interference pattern, a corresponding suppression strategy is generated to eliminate interference in the signal in the digital domain. S6: Signal recovery and output, compensates and equalizes the suppressed signal, and recovers and outputs the target signal.

[0019] In the analog domain interference pre-identification step, the operation of the memristor cross array specifically includes: The row and column lines of the memristor cross array are connected by memristors, and the intermediate frequency analog signal is applied to the column lines; The conductance of the memristor is pre-programmed as a basis function for a variety of typical interference modes; According to Ohm's law and Kirchhoff's law, the current flowing into each row line is the result of the analog multiplication and addition of the input signal and each basis function; The analog current is converted into a sparse pulse sequence via a spiking neuron circuit, where the pulse frequency is encoded with matching strength and the pulse timing is encoded with pattern category confidence.

[0020] The digitization and preprocessing steps use an event-driven ADC for sampling. Its sampling rate and start time are triggered and controlled by the digital pulse sequence. High-precision sampling and subsequent processing are initiated only when the pulse sequence indicates the presence of potential interference; otherwise, a low-power monitoring state is maintained.

[0021] The refined feature extraction and recognition steps employ a spatiotemporal-frequency three-dimensional joint processing framework, specifically as follows: The regularized baseband signal of the multi-antenna channel is constructed as a three-dimensional tensor of space, time, and frequency; A compressed sensing algorithm based on structured sparse Bayesian learning is used to reconstruct the complete spatiotemporal spectrum from the random subsampled data of the three-dimensional tensor, which is less than 5%. By utilizing a dynamic dictionary learning algorithm, based on the prior information provided by the digital pulse sequence (including the preliminary type and intensity confidence of the interference), the regularization parameters and sparse constraints of the dictionary atoms in the reconstruction algorithm are adaptively adjusted to focus on the interference mode region.

[0022] In the RF front-end receiving step, before the signal is input to the memristor cross array, it first passes through a programmable metasurface. The method further includes: Based on the preliminary interference identification results provided by the digital pulse sequence, the phase response of each unit of the programmable metasurface is dynamically adjusted; The metasurface forms a deep null in the direction of the incoming interference wave, and the physical level interference suppression is used as a pre-processing step to reduce the input dynamic range requirements of the subsequent memristor array and digital processing module.

[0023] The method is driven by harvesting electromagnetic energy from the environment, and specifically includes: The energy in the radio frequency broadband signal (especially the energy of the interference signal) is collected by a broadband rectified antenna. The collected energy is used to power the memristor cross array and analog processing circuitry; The system's power scheduler adaptively switches the power supply of subsequent digital processing modules on and off according to the density of the digital pulse sequence, achieving energy-consistent interference processing.

[0024] The refined feature extraction and recognition steps employ a biomimetic auditory pulse neural network model, specifically as follows: The regularized baseband signal is decomposed into subband signals using a cochlear filter bank model; Each sub-band signal, along with the digital pulse sequence, is input into a pulse neural network. Spiking neural networks learn the spatiotemporal correlation between interference pulses and subband signal features through pulse temporal dependence plasticity rules, and complete the classification.

[0025] The complete spatiotemporal spectrum is reconstructed from the random subsampled data of the three-dimensional tensor (less than 5%) using a compressed sensing algorithm, specifically as follows: First, a mathematical model for compressed sensing is established, assuming the complete spatiotemporal three-dimensional spectrum signal to be reconstructed is a vector. (flatten a three-dimensional tensor into a one-dimensional vector); The signal is in a certain transform base (e.g., Fourier basis, wavelet basis, or learned dictionary) is K-sparse. ),Right now ,in It is a sparse coefficient vector; Observation A result after random subsampling ,in ; The observation model is: ; in, It is a random measurement matrix (e.g., a Gaussian random matrix, a Bernoulli matrix); It is observation noise, modeled as having a mean of 0 and a variance of . Complex Gaussian white noise, i.e.: ; It is an equivalent perception matrix.

[0026] The specific calculations are as follows: First, establish a probability model: Likelihood function: Given a noise model, the conditional distribution of the observed data y is: ; Sparse prior: for coefficients To promote sparsity, a hierarchical Gaussian prior is set up: ; in , These are hyperparameters that control sparsity; Hyperprior: a hyperparameter and noise accuracy Setting conjugate priors typically involves using uninformative priors: ; ; Where a, b, c, and d are small constants, for example... This indicates a lack of prior knowledge.

[0027] Evidence maximization or variational inference is solved iteratively using the EM algorithm: E-step: Fixed hyperparameters and Calculate the sparsity coefficient The posterior probability distribution is a Gaussian distribution: ; The posterior mean and covariance are: ; ; For the current iteration, the sparsity coefficients The best estimate, i.e., the reconstructed spectrum; M-step: Fixed posterior distribution Update hyperparameters and To maximize expectations; renew : ; in It is the posterior mean The One element, It is the posterior covariance The One diagonal element; renew : ; Iteration and Convergence: Repeat E-step and M-step until... and If the change is less than a certain threshold or the maximum number of iterations is reached, during the iteration process, most of the It will tend towards infinity, that is Thus forcing the corresponding posterior mean The coefficients tend to 0, and eventually only a few coefficients that match the observed data are retained, thus achieving automatic correlation determination, i.e., sparse reconstruction; After iterative convergence, the posterior mean That is, the reconstructed sparse coefficients The final reconstructed spatiotemporal frequency spectrum is .

[0028] Structured sparse Bayesian learning, by introducing a Bayesian probabilistic framework, provides a prior probability distribution for the sparsity assumption of the signal. It can automatically learn the correlation and noise level of each component in the signal, thereby enabling: Achieve higher reconstruction accuracy: especially at low signal-to-noise ratios and ultra-low sampling rates, its performance is significantly better than traditional methods based on L1 norm regularization (such as LASSO); Provides uncertainty measurement: It can output the confidence level of the reconstruction results, which is crucial for subsequent interference identification and decision-making.

[0029] The dynamic dictionary learning algorithm is as follows: Prior information extraction: Extract structured information from digital pulse sequences to form a prior vector P; A vector whose components represent the confidence level of the presence of different types of interference; (encoded by pulse frequency) A vector indicating which frequency sub-bands are more likely to experience interference; (derived from the pulse timing and sub-band mapping relationship). A vector indicating which spatial directions are more likely to be disturbed; Dynamically adjusting parameters based on structured sparse Bayesian learning: The prior P is mapped through a function. This involves transforming the hyperparameters into new initial values ​​or constraints based on structured sparse Bayesian learning, primarily adjusting the variance of the sparse prior. ; Method 1: Initialization: Before iterating based on structured sparse Bayesian learning, initialize with prior information. Instead of setting all of them to 1 or a constant; ; in It is a monotonically increasing function, for example In other words, in dimensions where prior information indicates potential interference ( (Value is too large), its coefficient is allowed. It has a larger variance, while in other dimensions it is given a very small initial variance, which prompts it to converge to 0 more quickly; Method 2: Constrain M-step, update in M-step The formula incorporates prior dependencies to prevent the algorithm from deviating too far from the prior indication region; ; in As a penalty item, To balance the parameters, when When I was very young, It is very large, thus strongly suppressing The growth of forces the coefficient to be 0; Dynamically adjust the dictionary : Maintain a dictionary database Each dictionary is optimized for different types of interference, including single-tone interference, narrowband interference, broadband interference, and impulse interference. Based on the initial identification of the interference type using the impulse sequence, the most suitable dictionary is selected from the database. Used for this reconstruction based on structured sparse Bayesian learning; With the current dictionary Starting with prior information P, an online dictionary learning algorithm is used to adjust the dictionary atoms with the current observation data y, so that it can better sparsely represent the current perturbation instance. The optimization objective is: ; in It is a regular expression term used to ensure that the updated dictionary does not conflict with prior information P and retains its physical meaning.

[0030] The specific calculations are as follows: Input: Observational data y, measurement matrix Prior information P (from memristor array); Dynamic initialization: Initialize the hyperparameters based on structured sparse Bayesian learning according to P. and initializing the dictionary ; Dynamic Iteration Based on Structured Sparse Bayesian Learning: Run EM iteration based on structured sparse Bayesian learning, but in the M-step, The update is constrained by P; Output: Reconstructed spectrum The result of this reconstruction not only depends on the data y, but also incorporates the prior P provided by the analog domain pre-identification, thereby achieving a deep integration of analog and digital processing and achieving the effect of adaptively adjusting the reconstruction algorithm.

[0031] Structured sparse Bayesian learning provides a Bayesian reconstruction framework. Dynamic dictionary learning uses prior information to inject domain knowledge and attention mechanisms into this framework, enabling it to efficiently and accurately complete the task of extracting interference features even under extreme conditions of ultra-low sampling rates.

[0032] Based on the interference modes and parameters finally identified in step S4, optimized configuration instructions are dynamically generated and fed back to the RF front-end and memristor cross array for real-time adjustment of their hardware operating status. The specific process of dynamically adjusting the memristor cross array includes: When the interference mode identified by S4 is an unknown new mode, the corresponding normalized basis function vector is calculated and generated based on its spatiotemporal characteristic vector. The basis function vector is mapped to a specific voltage pulse sequence; The voltage pulse sequence is applied to a preset redundant row of the memristor cross array to change the conductance value of the memristor in that row, so as to write the new interference mode basis function into the hardware online; Update the system's base function-row index mapping table to associate the new interference pattern with the redundant row; The specific process of dynamically adjusting the RF front end includes the control of the programmable metasurface: Based on the interference direction of arrival (DOA) and center frequency parameters provided by S4, calculate the required phase offset for each unit of the metasurface; The phase offset is quantized into a digital phase control word; The phase control word is sent to the driving circuit of the programmable metasurface via the control bus, so that the metasurface forms an adaptive deep null in the direction of the interference wave.

[0033] The specific process of dynamically adjusting the RF front end also includes: Based on the interference power level parameters provided by S4, the analog voltage is output through the digital-to-analog converter (DAC), and the gain bias point of the low-noise amplifier (LNA) is dynamically adjusted to prevent the subsequent analog-to-digital converter (ADC) from saturating. Based on the interference bandwidth parameters provided by S4, the bandwidth of the variable bandwidth filter is controlled via the serial peripheral interface (SPI) to pre-filter the interference.

[0034] The amplitude, width, and number of voltage pulse sequences are calculated based on a pre-stored memristor conductance-voltage pulse response model, which is described as follows: ; in Let be the change in electrical conductivity, k be the material constant, and A be the pulse amplitude. The pulse width; Based on the interference modes and parameters finally identified in step S4, optimized configuration instructions are dynamically generated and fed back to the RF front-end and memristor cross array for real-time adjustment of their hardware operating state. This is executed by an independent adaptive optimization engine module. The input of this module includes the interference mode identifier and interference parameter vector from S4, and its output is a control bus signal to the memristor array, programmable metasurface, and RF front-end circuit. The dynamic adjustment strategy is based on a predefined optimization strategy table and interference-action mapping model. The strategy table takes the interference mode, intensity, and frequency as input and the adjustment amount and direction of the hardware configuration parameters as output.

[0035] By leveraging the precise identification results in the digital domain, the front-end hardware is dynamically optimized: the interference basis function is adaptively evolved through online reprogramming of the memristor array, enabling the system to cope with unknown interference; by real-time control of the metasurface to form precise pattern nulls and adjusting RF circuit parameters, collaborative optimization across the analog, RF, and digital domains is achieved, thereby systematically improving the interference rejection ratio, reducing signal distortion, expanding the processing dynamic range, and realizing on-demand energy management based on the interference threat level, ultimately achieving a dual improvement in performance self-optimization and environmental adaptation.

[0036] Working Principle: First, the RF front-end captures a broadband RF signal, which is amplified and down-converted to obtain an intermediate frequency (IF) analog signal. This signal is then passed through a programmable metasurface for pre-processing physical interference suppression before being input into a memristor cross array. Pre-programmed interference mode basis functions in the array are used to perform parallel multiplication and addition operations in the analog domain. The output analog current is converted into a digital pulse sequence, achieving preliminary interference identification. Subsequently, the digital pulse sequence triggers an event-driven ADC for sampling. The signal is digitized and pre-processed to obtain a regularized baseband signal. A three-dimensional space-time-frequency tensor is then constructed. Combined with pre-identified prior information, structured sparse Bayesian learning and dynamic dictionary learning are used to complete spectrum reconstruction and fine feature extraction. A classification algorithm is then used to achieve final interference mode identification and parameter estimation. Based on the identification results, adaptive interference suppression is performed in the digital domain, and the target output signal is restored after signal compensation and equalization. Meanwhile, the system is powered by electromagnetic energy collected by a broadband rectifier antenna. It can also dynamically adjust the parameters of the programmable metasurface and RF front-end circuit through an adaptive optimization engine based on the fine identification results, and update the interference basis function of the memristor array online. This enables cross-domain collaborative optimization of the RF, analog, and digital domains, achieving low power consumption, high precision interference processing, and system environment adaptation.

[0037] 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 radio frequency broadband signal receiving and processing method based on interference pattern recognition, characterized in that, Includes the following steps: S1: Radio frequency front-end receiver, which captures the radio frequency broadband signal through the receiving antenna, and obtains the intermediate frequency analog signal through low noise amplification and down-conversion processing; S2: Analog domain interference pre-identification: The intermediate frequency analog signal is directly input to the memristor cross array. The interference mode basis function characterized by the conductance of the memristors in the array is multiplied and added in parallel in the analog domain with the input signal. The analog current characterizing the interference mode matching degree is output, and the analog current is converted into a digital pulse sequence. The pulse characteristics of the digital pulse sequence are used to identify the preliminary type and intensity of the interference. S3: Digitization and preprocessing: The signal processed by the memristor cross array is sampled and digitized to obtain a digital signal, and then down-converted, filtered and synchronized to obtain a regular baseband signal. S4: Fine feature extraction and recognition: Perform transform domain analysis on the regular baseband signal, and combine it with the preliminary results identified by the digital pulse sequence to extract fine features. Then, complete the final identification and parameter estimation of the interference mode through a classification algorithm. S5: Adaptive interference suppression. Based on the finally identified interference pattern, a corresponding suppression strategy is generated to eliminate interference in the signal in the digital domain. S6: Signal recovery and output, compensates and equalizes the suppressed signal, and recovers and outputs the target signal.

2. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 1, characterized in that: In the analog domain interference pre-identification step, the operation of the memristor cross array specifically includes: The row and column lines of the memristor cross array are connected by memristors, and the intermediate frequency analog signal is applied to the column lines; The conductance of the memristor is pre-programmed as a basis function for a variety of typical interference modes; According to Ohm's law and Kirchhoff's law, the current flowing into each row line is the result of the analog multiplication and addition of the input signal and each basis function; The analog current is converted into a sparse pulse sequence via a spiking neuron circuit, wherein the pulse frequency is encoded with matching strength and the pulse timing is encoded with pattern category confidence.

3. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 2, characterized in that: The digitization and preprocessing steps use an event-driven ADC for sampling. Its sampling rate and start time are triggered and controlled by the digital pulse sequence. High-precision sampling and subsequent processing are initiated only when the pulse sequence indicates the presence of potential interference; otherwise, a low-power monitoring state is maintained.

4. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 3, characterized in that: The refined feature extraction and recognition steps employ a spatiotemporal-frequency three-dimensional joint processing framework, specifically as follows: The regularized baseband signal of the multi-antenna channel is constructed as a three-dimensional tensor of space, time, and frequency; A compressed sensing algorithm based on structured sparse Bayesian learning is used to reconstruct the complete spatiotemporal spectrum from the random subsampled data of the three-dimensional tensor, which is less than 5%. By utilizing a dynamic dictionary learning algorithm, the regularization parameters and sparse constraints of dictionary atoms in the reconstruction algorithm are adaptively adjusted based on the prior information provided by the digital pulse sequence, in order to focus on the interference mode region.

5. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 4, characterized in that: In the radio frequency front-end receiving step, before the signal is input to the memristor cross array, it first passes through a programmable metasurface. The method further includes: Based on the preliminary interference identification results provided by the digital pulse sequence, the phase response of each unit of the programmable metasurface is dynamically adjusted; The metasurface forms a deep null in the direction of the incoming interference wave, and the physical level interference suppression is used as a pre-processing step to reduce the input dynamic range requirements of the subsequent memristor array and digital processing module.

6. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 5, characterized in that: The method is driven by harvesting ambient electromagnetic energy and specifically includes: Energy in the radio frequency broadband signal is collected by a broadband rectified antenna; The collected energy is used to power the memristor cross array and analog processing circuitry; The system's power scheduler adaptively switches the power supply of subsequent digital processing modules on and off according to the density of the digital pulse sequence, thereby achieving energy-consistent interference processing.

7. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 6, characterized in that: The refined feature extraction and recognition steps employ a biomimetic auditory pulse neural network model, specifically as follows: The regularized baseband signal is decomposed into subband signals using a cochlear filter bank model; Each sub-band signal, along with the digital pulse sequence, is input into a pulse neural network. The spiking neural network learns the spatiotemporal correlation between interference pulses and subband signal features through pulse temporal dependence plasticity rules, and completes the classification.

8. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 7, characterized in that: Based on the interference modes and parameters finally identified in step S4, optimized configuration instructions are dynamically generated and fed back to the RF front-end and memristor cross array for real-time adjustment of their hardware operating status. The specific process of dynamically adjusting the memristor cross array includes: When the interference mode identified by S4 is an unknown new mode, the corresponding normalized basis function vector is calculated and generated based on its spatiotemporal characteristic vector. The basis function vector is mapped to a specific voltage pulse sequence; The voltage pulse sequence is applied to a preset redundant row of the memristor cross array to change the conductance value of the memristor in that row, so as to write the new interference mode basis function into the hardware online; Update the system's base function-row index mapping table to associate the new interference pattern with the redundant row.

9. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 8, characterized in that: The specific process of dynamically adjusting the RF front end includes the control of the programmable metasurface: Based on the interference wave direction and center frequency parameters provided by S4, calculate the required phase offset for each unit of the metasurface; The phase offset is quantized into a digital phase control word; The phase control word is sent to the driving circuit of the programmable metasurface via the control bus, so that the metasurface forms an adaptive deep null in the direction of the interference wave.

10. The radio frequency broadband signal receiving and processing method based on interference pattern recognition according to claim 9, characterized in that: The specific process of dynamically adjusting the radio frequency front end also includes: Based on the interference power level parameters provided by S4, the analog voltage is output through the digital-to-analog converter, and the gain bias point of the low-noise amplifier is dynamically adjusted to prevent the subsequent analog-to-digital converter from saturating. Based on the interference bandwidth parameters provided by S4, the bandwidth of the variable bandwidth filter is controlled through the serial peripheral interface to pre-filter the interference.

Images