6g high dynamic adaptive receiving method driven by visual-electromagnetic combined perception

Abstract

The application discloses a 6G high-dynamic self-adaptive receiving method driven by visual-electromagnetic combined perception, comprising the following steps: visual-electromagnetic combined perception and state transfer tensor generation; observation noise modulation based on interference cancellation ratio; directional process noise remodeling based on array orientation matching; joint demodulation decoding and decoding lattice topology entropy extraction; closed-loop particle swarm optimization based on physical boundary constraint. The application connects spatial geometry perception, bottom-layer electromagnetic waveform observation, channel estimation, demodulation decoding reliability feedback and receiving parameter closed-loop optimization into a unified receiving process, and can improve channel tracking stability and data receiving reliability in a high-dynamic strong interference scene.

Description

Technical Field

[0001] This invention relates to the field of wireless communication physical layer receiving technology, and more specifically to a 6G high dynamic adaptive receiving method driven by visual-electromagnetic joint sensing. Background Technology

[0002] In 6G high-dynamic and high-interference scenarios, the communicating parties or environmental scatterers may be in a state of high-speed relative motion. The channel will be affected by factors such as Doppler frequency shift, rapid phase rotation, sudden blockage, multipath structure changes, and colored interference, exhibiting obvious non-stationary time-varying characteristics. For physical layer receivers, these changes will directly affect the channel estimation, equalization, demodulation, and decoding performance, and in severe cases, lead to channel tracking loss and decreased decoding reliability.

[0003] Existing high dynamic range (HVR) reception methods typically rely primarily on posterior estimation of the received RF signal itself, such as adjusting filter parameters based on signal residuals, pilot errors, or innovative sequences. While these methods can adapt to channel variations to some extent, they still suffer from the following limitations: 1. There is a lack of effective integration between spatial geometric sensing and electromagnetic waveform observation. Traditional receivers mainly estimate frequency offset, channel gain, and noise statistics from the received waveform, making it difficult to utilize the three-dimensional spatial motion information of communication nodes, obstructions, and scatterers in a timely manner. When the direct path is blocked or the main reflection path changes, pure electromagnetic posterior estimation suffers from observation lag, making it difficult to provide predictive information to the channel tracking module before the obstruction occurs.

[0004] 2. Existing adaptive Kalman channel tracking methods mostly employ scalar or global matrix scaling to adjust process noise and observation noise. This approach cannot distinguish between specific affected physical propagation paths or spatial dimensions. When a physical path deteriorates due to obstruction, changes in reflective surfaces, or interference, the overall amplified process noise may simultaneously reduce the estimation stability of unaffected dimensions, thus affecting channel tracking accuracy.

[0005] 3. Existing receiver parameter optimization methods typically treat Kalman noise parameters, adaptive filtering parameters, or decoding-related parameters as independent mathematical variables for searching. They lack search boundaries constrained by physical motion limits and closed-loop feedback from the underlying decoder's reliability. Therefore, the optimization process may result in parameter combinations that do not conform to actual motion patterns or fail to effectively improve receiver reliability. Summary of the Invention

[0006] The present invention provides a visual-electromagnetic joint sensing driven 6G high dynamic adaptive reception method that can improve the reliable reception capability under high dynamic non-stationary channels, and can at least solve one of the above-mentioned technical problems.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A 6G high-dynamic adaptive reception method driven by visual-electromagnetic joint sensing includes the following steps: S1. The receiver processes the baseband radio frequency signal, uses a visual sensor to acquire the three-dimensional spatial geometric information of communication nodes, obstructions and scatterers, classifies and identifies direct, reflected scattering and diffraction propagation mechanisms, and performs confidence-weighted fusion with the electromagnetic frequency offset estimate obtained from OFDM cyclic prefix autocorrelation in the received baseband radio frequency signal, and outputs a state transition tensor for channel tracking. S2. Perform adaptive interference suppression on the received baseband radio frequency signal, extract the interference cancellation ratio, and output the observation noise penalty coefficient to reduce the weight of the contaminated observation signal based on the abrupt change characteristics of the interference cancellation ratio. S3. During Kalman channel tracking, an array steering vector is constructed based on the visually predicted obstructed physical path and matched with the eigenvector of the process noise covariance matrix. Only the eigenvalues ​​of the matching dimension are directionally expanded. At the same time, the observation noise matrix is ​​modulated using the observation noise penalty coefficient, and the channel state estimate is output. S4. Based on the channel state estimate, perform joint demodulation and Viterbi decoding to recover the service bit stream, and calculate the decoding lattice topology entropy from the path metrics of the surviving path and the contention path. S5. With the goal of minimizing the topological entropy of the decoding lattice, and combined with the physical manifold boundary measured by the visual sensor, the observation noise penalty control factor and the process noise directional expansion ratio are updated online through a particle swarm optimization algorithm with a boundary repulsion term, and fed back to S2 and S3 respectively, so as to realize the online update of the observation noise modulation and process noise directional reshaping parameters.

[0008] Furthermore, S1 employs a propagation mechanism to perceive and construct the state transition of a visual-electromagnetic confidence-gated Lie group, including: S1.1. For the non-stationary random process caused by the high-speed movement of both the mobile base station and the target node, the complex equivalent baseband physical model of the received signal is constructed as follows:

[0009] in, For the first The received signal at each sampling time, For multipath channel complex gain, For the existence of the first Transmitting a complex equivalent baseband signal with a sampling delay. It is additive complex white Gaussian noise. For true frequency offset, The sampling period is For sampling time index, For sampling delay index, This represents the total sampling delay. The three-dimensional spatial geometry topology of the communication environment is acquired using a visual sensor, and the physical propagation mechanisms of direct path, reflected scattering path and diffraction path are classified and identified. The spatial geometric features of each multipath component are extracted respectively. S1.2. Based on the direct path, reflected scattering path, diffraction path, and shadow occlusion state identified by the visual sensor, a propagation mechanism confidence gating factor is set for each candidate physical path, where: For unobstructed direct rays, improve their Doppler prediction contribution; For the blocked direct path, reduce its Doppler prediction contribution and output the direction of arrival of the blocked direct path to S3; For reflected scattering paths and diffraction paths, the reliability of their participation in prediction is determined based on the geometric stability of the reflecting surface, scattering body, or obstacle edge. Thus, visual spatial geometric information is first transformed into path-by-path propagation confidence, path-by-path occlusion state, and path-by-path wave direction of arrival, and then participates in the prediction of visual prior Doppler frequency shift features. S1.3. Based on the confidence gating results of the path-by-path propagation mechanism described in S1.2, a multi-mechanism joint visual prior Doppler frequency shift feature prediction model is constructed to address the multipath superposition and shadow occlusion effects in high dynamic environments:

[0010]

[0011] in, This represents the visual prior Doppler frequency shift prediction value at the nth sampling time. The total number of valid physical multipaths identified by the visual perception module. For effective physical path indexing, , The power attenuation weight for the r-th path is... The binary shadow occlusion state indicator for the r-th path. As the confidence gating factor for the r-th path propagation mechanism, At the speed of light, For carrier frequency, and Let r be the spatial relative velocity vector of the r-th radius and the unit vector of the direction of arrival, respectively. Represents the dot product of spatial vectors. Let r be the confidence level for the geometric stability of the r-th path. Confidence level for the type of propagation mechanism of the r-th path; S1.4. For the received radio frequency signal, the prior value of the frequency offset of the underlying electromagnetic waveform is estimated by utilizing the autocorrelation characteristics of the cyclic prefix in the OFDM symbol:

[0012] in, The prior value of frequency offset of the underlying electromagnetic waveform is obtained based on the OFDM cyclic prefix autocorrelation characteristics estimation. For the number of points in the Fast Fourier Transform, This represents the complex conjugate sample value of the received signal after one symbol period; To integrate the underlying electromagnetic waveform estimate with the visual prior Doppler frequency shift characteristics in S1.3, a confidence weighting coefficient is calculated based on the variance of the estimation errors of the two:

[0013] in: The confidence weighting coefficients for the visual prior Doppler frequency shift features are... The confidence weighting coefficient for the frequency offset estimate of the underlying electromagnetic waveform; The variance representing the visual perception estimation error is obtained from the covariance matrix of the optical flow tracking reprojection error of the target feature points by the visual sensor, and the calculation formula is:

[0014] In the formula, The velocity error variance generated when estimating relative velocity vectors using a vision sensor. The variance of the angle error generated when estimating the direction of arrival angle for a visual sensor; This represents the frequency offset error variance estimated from the underlying electromagnetic waveform, obtained by extracting the most recent... The sliding time variance of the OFDM cyclic prefix autocorrelation peak within one symbol period is calculated in real time, and the calculation formula is:

[0015] In the formula, The length of the cyclic prefix of the OFDM symbol. This represents the instantaneous signal-to-noise ratio of the received signal within the current observation window. High-precision equivalent instantaneous dynamic frequency offset characteristics are output through weighted calculation. The equivalent instantaneous dynamic frequency offset characteristic The expression comprehensively characterizes carrier frequency offset, the Doppler effect of relative motion between the two ends, and phase jitter caused by channel non-stationarity, and is as follows:

[0016] in, The visual prior Doppler frequency shift prediction value in S1.3 is adopted; S1.5, Construct based on The augmented state vector of a Lie group manifold The equivalent instantaneous dynamic frequency offset characteristics This is converted into the Doppler phase evolution rate and compared with the three-dimensional spatial angular velocity vector extracted by the visual sensor. and linear velocity vector Together they constitute the elements of the Lie algebra. And generate state transition tensors through exponential mapping. This achieves a nonlinear conversion from a three-dimensional geometric trajectory to an electromagnetic phase, expressed as:

[0017] in, Let be the augmented state vector at the k-th state update time. It is a three-dimensional rotation matrix. It is a three-dimensional translation vector. It is an electromagnetic phase state quantity. This is the adjoint matrix form corresponding to the Lie algebra elements. This is the time step for state updates.

[0018] Furthermore, S2 employs interference cancellation ratio-driven observation noise nonlinear modulation, including: S2.1 Introduce a variable step-size normalized minimum mean square error adaptive filter based on a tapped delay line structure. Let the first step be... The input signal delay vector at each adaptive filtering iteration is The filter tap weight vector is The predicted estimates for multi-source complex interference are:

[0019] in, This represents the transpose operation of the weight vector. This is the index of the discrete iteration time of the adaptive filter; Real received sample values ​​sampled in real time at the antenna end Compared with the predicted value The residual calculation error signal is used to obtain the purified expected signal component after removing interference: ; S2.2 To adapt to non-stationary disturbances in highly dynamic environments, the weight vector step size factor is... An exponential adaptive update mechanism based on error energy is adopted, and its expression is:

[0020] in, For maximum step size constraint, This is the step-size adaptive sensitivity control coefficient, used to adjust the degree of influence of error energy on the weight vector step-size factor; Update the filter weight vector using dynamic step size:

[0021] in, This is the index for the discrete iteration time of the adaptive filter. For the first The updated filter tap weight vector at each iteration time. To prevent positive small constants with a denominator of zero, It is the complex conjugate of the error signal; S2.3. Using an exponentially weighted moving average mechanism, simultaneously predict the interference power and the cleaned signal power statistically, and define the interference cancellation ratio. for:

[0022] in, The smoothing forgetting factor, used to control the proportion of historical energy data, has a value range of [value range missing]. , For the historical sampling index within the sliding statistics window, For the i-th historical sample data in the th... The exponentially decaying weights corresponding to each adaptive filtering iteration time. Let i be the predicted interference signal at the i-th sampling time. This is the purification error signal obtained after interference cancellation at the i-th sampling time; S2.4 When a sudden strong electromagnetic pulse or colored interference causes filter tracking lag, the interference cancellation ratio... This will produce a descent abrupt change, constructing a nonlinear adaptive mapping function, and outputting an observation noise penalty coefficient for S3 channel tracking observation updates. :

[0023] in, This is the steady-state reference threshold under undisturbed conditions. It is a nonlinear penalty control factor, and in S5, it is optimized and updated online at a slow scale using the particle swarm optimization algorithm. when When below the safety threshold, It is amplified exponentially to reduce the weight of contaminated electromagnetic observations in the Kalman update.

[0024] Furthermore, S3 employs array-guided matching process noise physical-directed dilation channel tracking, including: S3.1 Establish a Kalman recursive state-space model and use the state transition tensor output by S1. Perform prior prediction of channel states:

[0025] in, For the first The prior prediction estimate vector of the channel state at time step. This is the posterior state estimation vector from the previous time step; For the initial process noise covariance matrix Perform eigenvalue decomposition:

[0026] in, The eigenvector matrix of the process noise covariance matrix. , It is a diagonal matrix of eigenvalues. ; S3.2. Based on the obstructed physical path predicted in S1, extract its physical direction of arrival relative to the receiving antenna array. Let the angle of arrival of this obstruction path be... The receiver is configured with array element spacing. , number of array elements A uniform linear array; Based on the phase deflection law of electromagnetic wave propagation in space, this physical direction of arrival is mapped to the physical steering vector of the receiving antenna array. :

[0027] in, For radio frequency carrier wavelength, The element spacing of the receiving array. Indicates matrix transpose. The imaginary unit; The guide vector The array response characteristics of the blocked physical path in electromagnetic space were accurately characterized; S3.3, in the eigenvector matrix In the middle, the index of the target feature vector that is most correlated with the occluded physical path is locked by the spatial inner product norm. :

[0028] in, Eigenvector matrix The z-th eigenvector in For feature vector index; Only for target feature vectors Corresponding eigenvalues Apply dynamic expansion ratio Reshape the eigenvalue diagonal matrix ,make The remaining non-target feature values ​​remain unchanged; Combining eigenvector matrix Reconstruct the process noise covariance matrix at the current time step:

[0029] Combined with the reconstructed process noise covariance matrix Calculate the prediction error covariance matrix:

[0030] in, For the first The prior prediction error covariance matrix of the channel state at time step. Let be the posterior error covariance matrix of the previous time step. Represents the conjugate transpose of a matrix; S3.4, Observation noise penalty coefficient received from S2 output The variance of the basic measurement noise is modulated to generate the observation noise matrix at the k-th state update time. :

[0031] in, It is the identity matrix; S3.5, Combine the observation signal matrix at the k-th state update time. Dynamically calculate adaptive Kalman gain :

[0032] in, This represents the matrix inversion operation; Using Kalman gain Combined with actual received baseband radio frequency signals The residuals from the predicted observed signals are used to complete the posterior update of the symbol-level channel state:

[0033] in, To use the received signal at the k-th observation time to complete the updated channel state posterior estimation vector; The posterior error covariance matrix is ​​updated synchronously to provide high-precision initial values ​​for tracking in the next symbol period, forming a closed-loop iteration:

[0034] in, This is for updating the completed posterior error covariance matrix.

[0035] Furthermore, S4 employs decoding lattice topological entropy to drive joint demodulation decoding reliability feedback, including: S4.1 Construct a search lattice graph based on the soft-output Viterbi algorithm, and receive the channel state posterior estimation vector at the k-th state update time output by S3. With the One decoded received symbol Define the branch metric as:

[0036] in, and These represent the previous state and the current state in the lattice diagram, respectively. Indicates the first The theoretical transmitted signal on the state transition branch at each decoding and receiving symbol moment. For the time index of the decoded received symbol; Recursively accumulate data using the add-compare-select operation to reach the current state. Path metric:

[0037] in, This is an index for the received symbol time in the Viterbi decoding process. , This represents the total number of received symbols in the current decoded frame. Indicates the first The cumulative path metric value for each decoded received symbol reaching the current state at each time step. This represents the cumulative path metric value from the time the previously decoded received symbol arrived at the previous state. This represents the current branch metric. This represents the set of predecessor states of the current state; S4.2 In the backtracking optimization phase of Viterbi decoding, for a length of... Each decoding moment within the backtracking window Extract the surviving path metric to reach the optimal state node. metric for the abandoned competing path ; Introducing a scaling factor constrained by the fundamental noise The path metric is converted into a probability confidence score of local state transition using the Softmax function. :

[0038] in, This is an index for path types; surv indicates a surviving path, and comp indicates a contested path. Indicates the first Decoding time path The corresponding cumulative path metric, For the backtracking window length, This serves as an index for the decoding time within the backtracking window; Calculate the Shannon information entropy for the local probability distribution and throughout the backtracking window. Calculate the mean value internally and output the decoded lattice topological entropy. :

[0039] When high-dynamic, strong interference degrades electromagnetic observations, the difference in metrics between surviving and competing paths narrows, the local probability distribution tends to become more uniform, and the decoding lattice topological entropy... Consequently, the topological entropy of the decoding lattice increases, therefore, the topological entropy of the decoding lattice increases. It can be used to quantify the decoding decision uncertainty under the current channel estimation results and receiver parameter configuration, and serve as the fitness feedback quantity of the closed-loop optimization module.

[0040] Furthermore, S5 employs physical manifold boundary constraints for closed-loop heterogeneous particle swarm optimization, including: S5.1, Construct a system that includes dynamic expansion ratios With nonlinear penalty control factor Hyperparameter optimization vector ; S5.2. Transform the maximum physical motion limit extracted in S1 into the rigid physical manifold boundary of the parameter search space. Construct the boundary potential function for this boundary. :

[0041] in, For the search space dimension, For the search space dimension index, hyperparameter vector The position coordinates in the q-th dimension and These are the upper and lower physical limits of this dimension, respectively; When the particle's position approaches the boundary of the physical manifold, the boundary potential function Generate a repulsive gradient inward along the boundary normal and add it to the particle swarm velocity update equation. ,in The potential energy repulsion gain coefficient; S5.3, Receive the decoded lattice topology entropy output from S4 Minimizing the desired topological entropy is used as the fitness function for particle swarm optimization:

[0042] in, Represents the hyperparameter vector The corresponding fitness function value, Represents the mathematical expectation operation; S5.4, drive the particle swarm to perform intergenerational state updates, the first The particle in the first The speed update formula for the second optimization iteration is:

[0043] in, For particle indexing, Index of the number of iterations for particle swarm optimization Let g be the optimization velocity of the g-th particle. Let be the hyperparameter position vector of the g-th particle. This represents the optimal position of the historical individual of the g-th particle. The optimal position globally. For inertial weights, and Individual cognitive learning factors and social learning factors, and A random number uniformly distributed within the interval [0,1]. The potential energy repulsion gain coefficient defined in S5.2; At the same time, the updated velocity is used to drive the particles to perform position jumps:

[0044] in, For the g-th particle in the th... The hyperparameter position vector after the optimization iteration is completed; S5.5. Through velocity and position iteration guided by the physical manifold, the decoding lattice topological entropy is finally locked. Achieving the minimum global optimal hyperparameter set It then sends the closed-loop data down to the underlying physical layer, and updates the process noise covariance matrix in S3 and the observation noise penalty coefficient in S2 in real time.

[0045] The beneficial effects of this invention are reflected in: This invention does not rely solely on a single type of sensor information. Instead, it uses the spatial topology, motion trends, and occlusion states obtained from the visual side, along with the frequency offset, interference residuals, and reception reliability obtained from the electromagnetic side, as the basis for receiver state updates. Through the joint fusion of visual and electromagnetic multidimensional features, it utilizes visual information to provide spatial priors that precede the occurrence of RF residuals, and uses electromagnetic information to correct visual estimation errors and reflect the actual degree of link degradation. Combining a dual adaptive mechanism of "fast-scale mapping compensation and slow-scale closed-loop optimization," it enables symbol-level channel prediction, noise matrix adjustment, and hyperparameter updates across reception cycles to work together. This aims to overcome the problems of single situational awareness dimension, weak resistance to non-stationary colored interference, fixed Kalman tracking parameters, and lack of low-level optimization closed loop in existing technologies. As a result, it achieves improved channel trajectory stability and autonomous evolution of system performance under extreme high-dynamic and strong interference environments. Attached Figure Description

[0046] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0047] Figure 1 This is a schematic diagram of the overall process of the method according to an embodiment of the present invention.

[0048] Figure 2 This is a comparison chart of channel tracking performance in all spatial dimensions under high dynamic occlusion and sudden interference scenarios according to an embodiment of the present invention, showing the total channel normalized mean square error.

[0049] Figure 3 This is a comparison chart of channel tracking performance under high dynamic occlusion and sudden interference scenarios according to an embodiment of the present invention, excluding the spatial dimension corresponding to the occluded physical path, and only counting the normalized mean square error of the other non-occluded dimensions.

[0050] Figure 4 This is a comparison chart of received bit error rate under high dynamic occlusion and sudden interference scenarios according to an embodiment of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] It should be noted that the meaning of "and / or" throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, "multiple" refers to two or more. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0053] See Figure 1 This invention provides a 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing, comprising the following steps: S1: Visual-electromagnetic joint perception and state transition tensor generation.

[0054] This step takes the currently received complex equivalent baseband signal, OFDM cyclic prefix observations, and spatial information of the target, obstruction, and scatterer obtained by the visual sensor as input, and outputs equivalent instantaneous dynamic frequency offset characteristics, the direction of the obstructed physical path, and a state transition tensor for channel prediction. The receiver performs normalization preprocessing on the baseband RF signal, uses the visual sensor to acquire three-dimensional spatial geometric information of the communication node, obstruction, and scatterer, and transforms the visible obstacle boundary, main scattering region, node motion trajectory, and relative velocity relationship into physical priors that can be used by the baseband algorithm; identifies direct, reflected, and diffracted propagation mechanisms, where the direct path mainly reflects the direct relative motion between communication nodes, the reflected or scattered path reflects the change in the direction and amplitude of electromagnetic wave propagation by the environmental surface, and the diffracted path is used to describe the weak propagation components that may still exist at the edge of the obstacle; obtains the relative motion vector and possible obstruction direction of each physical path; and simultaneously utilizes the OFDM cyclic prefix observations in the received RF signal. The autocorrelation characteristics of the DM symbol cyclic prefix are used to estimate the electromagnetic frequency offset prior value. This electromagnetic frequency offset prior value retains the influence of residual carrier offset and sampling clock deviation in the actual RF link, and is used to constrain the visual inference results from deviating excessively from the actual received waveform. Confidence weights are calculated based on the visual estimation error and the electromagnetic frequency offset estimation error. The Doppler features obtained from the visual inference are weighted and fused with the electromagnetic frequency offset features to construct an augmented state vector containing spatial motion and electromagnetic phase evolution. A state transition tensor is generated through Lie algebra exponential mapping, so that changes in spatial motion can enter the Kalman prediction stage in the form of phase evolution rate, and the output is sent to the channel tracking module as a priori prediction input.

[0055] S2: Observation noise modulation based on interference cancellation ratio.

[0056] This step takes the current baseband RF signal and the energy statistics before and after interference suppression as input, and outputs an observation noise penalty coefficient to describe the reliability of the current observation sample. This coefficient does not directly modify the channel state, but reduces the proportion of abnormal observations in the Kalman gain by modulating the observation noise matrix. Adaptive interference suppression is performed on the baseband RF signal. Without changing the service signal frame structure and pilot structure, the energy change before and after interference suppression is statistically analyzed to obtain the interference cancellation ratio, which characterizes the degree of deterioration of the current electromagnetic environment. The lower the interference cancellation ratio, the more difficult it is for the filter to separate the effective signal from the received waveform, and the higher the possibility that the current observation is contaminated by sudden interference, colored interference, or multi-source superposition interference. When the interference cancellation ratio decreases relative to the steady-state reference threshold, an observation noise penalty coefficient is generated through a nonlinear mapping function and output to the channel tracking module to reduce the weight of contaminated observation signals in the Kalman update.

[0057] S3: Noise reshaping of the orientation process based on array-guided matching.

[0058] This step, located in the channel estimation and channel tracking stage, maps the spatial propagation judgment of S1 and the interference reliability judgment of S2 simultaneously to the Kalman filtering process. The channel tracking module receives the state transition tensor output by S1 and completes the prior prediction of the channel state. This prior prediction is used to provide the channel evolution trend for the next symbol period before the pilot observation fully reflects sudden obstruction or fast Doppler changes. Before the Kalman observation update, a corresponding receiving array steering vector is constructed based on the obstructed physical path direction output by S1, converting the visually identified affected propagation direction into the electromagnetic spatial response direction in the antenna array. The process noise covariance matrix is ​​decomposed into eigenvalues, and the target eigenvector with the highest correlation to the array steering vector is found in the eigenvector matrix. Only the eigenvalue corresponding to the target eigenvector is amplified according to the dynamic expansion ratio, while other eigenvalues ​​remain unchanged, thereby avoiding the disturbance of the unobstructed dimension when amplifying the process noise in the traditional overall process noise, thus reconstructing the process noise covariance matrix. At the same time, the observation noise penalty coefficient output by S2 is used to modulate the observation noise matrix to complete the Kalman filtering update and output the channel state estimate. This channel state estimate is then used as a direct input to equalization, soft-decision demodulation, and decoding to determine the quality of subsequent service bit recovery.

[0059] S4: Joint demodulation and decoding with topological entropy extraction of the decoding lattice.

[0060] This step is part of the receiver demodulation and decoding process. On one hand, it recovers the service bits; on the other hand, it extracts reliability indicators reflecting the difficulty of reception from the decoder's internal state. The joint demodulation and decoding module receives the channel state estimate output from S3 and the current baseband RF signal. The channel state estimate is used to compensate for multipath fading and phase rotation, while the current baseband RF signal is used to generate the soft-decision likelihood input. Based on the channel state estimate, equalization, soft-decision demodulation, and Viterbi decoding are performed to recover the service bitstream. The service bitstream is the final received output, and path contention information during the decoding process is preserved. Path metrics for surviving and contention paths are extracted within the Viterbi decoding backtracking window. The path metrics are converted into local path probabilities using the Softmax function, and the average information entropy within the backtracking window is calculated to obtain the decoding lattice topology entropy, which serves as feedback for the decoding decision uncertainty under the current receiving parameters. This feedback differs from simple error statistics; it does not require knowledge of the actual transmitted bits and can be directly used to evaluate the current parameter combination during online reception.

[0061] S5: Closed-loop particle swarm optimization based on physical boundary constraints.

[0062] This step belongs to the slow-scale receiver parameter optimization stage, used to correct the front-end noise modulation and channel tracking parameters based on the decoding reliability feedback over a period of time. The closed-loop optimization module receives the decoding lattice topology entropy output by S4 and uses minimizing this entropy value as the fitness target; when the decoding lattice topology entropy decreases, it indicates that the distinction between surviving paths and competing paths has improved, and the current receiver parameters are more conducive to reliable decoding; the maximum relative velocity, maximum angular velocity, or maximum Doppler change range measured by the visual sensor in S1 is mapped to the physical boundary of the particle swarm search space; this boundary makes the search space consistent with the upper limit of the actual node velocity, angular velocity, and Doppler change, avoiding physically unrealizable parameter combinations; a repulsion term generated by the physical boundary potential energy function is added to the particle swarm velocity update equation, so that particles are constrained to the interior of the feasible region when approaching the boundary; the particle swarm algorithm outputs the optimal nonlinear penalty control factor and dynamic expansion ratio, and feeds them back to S2 and S3 respectively, wherein the nonlinear penalty control factor is fed back to the observation noise modulation, and the dynamic expansion ratio is fed back to the process noise directional reshaping, realizing the online update of the observation noise modulation and process noise directional reshaping parameters.

[0063] This method organizes the modules in the order of front-end sensing, channel tracking, joint demodulation and decoding, and closed-loop parameter optimization. Specifically: S1 outputs the state transition tensor and the direction of the obstructed physical path as the spatial prior input of S3; S2 outputs the observation noise penalty coefficient as the observation reliability input of S3; S3 outputs the channel state estimate as the input of equalization, soft demodulation, and decoding of S4; S4 outputs the decoding lattice topology entropy, which is then used as the fitness feedback of S5, thereby forming a complete input-output link for reliable reception.

[0064] In this embodiment, S1 is used to further illustrate the correspondence between visual geometric quantities, electromagnetic frequency offset quantities, and state transition tensors. It mainly employs a propagation mechanism to perceive and construct the state transition of a visual-electromagnetic confidence-gated Lie group, including: S1.1. For the non-stationary random process caused by the high-speed movement of both the mobile base station and the target node, the complex equivalent baseband physical model of the received signal is constructed as follows:

[0065] in, For the first The received signal at each sampling time, For multipath channel complex gain, For the existence of the first Transmitting a complex equivalent baseband signal with a sampling delay. It is additive complex white Gaussian noise. For true frequency offset, The sampling period is For sampling time index, For sampling delay index, This represents the total sampling delay. The three-dimensional spatial geometry topology of the communication environment is acquired using a visual sensor, and the physical propagation mechanisms of direct path, reflected scattering path and diffraction path are classified and identified. The spatial geometric features of each multipath component are extracted respectively. S1.2. Based on the direct path, reflected scattering path, diffraction path, and shadow occlusion state identified by the visual sensor, a propagation mechanism confidence gating factor is set for each candidate physical path, where: For unobstructed direct rays, improve their Doppler prediction contribution; For the blocked direct path, reduce its Doppler prediction contribution and output the direction of arrival of the blocked direct path to S3; For reflected scattering paths and diffraction paths, the reliability of their participation in prediction is determined based on the geometric stability of the reflecting surface, scattering body, or obstacle edge. Thus, visual spatial geometric information is first transformed into path-by-path propagation confidence, path-by-path occlusion state, and path-by-path wave direction of arrival, and then participates in the prediction of visual prior Doppler frequency shift features. S1.3. Based on the confidence gating results of the path-by-path propagation mechanism described in S1.2, a multi-mechanism joint visual prior Doppler frequency shift feature prediction model is constructed to address the multipath superposition and shadow occlusion effects in high dynamic environments:

[0066]

[0067] in, This represents the visual prior Doppler frequency shift prediction value at the nth sampling time. The total number of valid physical multipaths identified by the visual perception module. For effective physical path indexing, , The power attenuation weight for the r-th path is... The binary shadow occlusion state indicator for the r-th path. As the confidence gating factor for the r-th path propagation mechanism, At the speed of light, For carrier frequency, and Let r be the spatial relative velocity vector of the r-th radius and the unit vector of the direction of arrival, respectively. Represents the dot product of spatial vectors. Let r be the confidence level for the geometric stability of the r-th path. Confidence level for the type of propagation mechanism of the r-th path; S1.4. For the received radio frequency signal, the prior value of the frequency offset of the underlying electromagnetic waveform is estimated by utilizing the autocorrelation characteristics of the cyclic prefix in the OFDM symbol:

[0068] in, The prior value of frequency offset of the underlying electromagnetic waveform is obtained based on the OFDM cyclic prefix autocorrelation characteristics estimation. For the number of points in the Fast Fourier Transform, This represents the complex conjugate sample value of the received signal after one symbol period; To integrate the underlying electromagnetic waveform estimate with the visual prior Doppler frequency shift characteristics in S1.3, a confidence weighting coefficient is calculated based on the variance of the estimation errors of the two:

[0069] in: The confidence weighting coefficients for the visual prior Doppler frequency shift features are... The confidence weighting coefficient for the frequency offset estimate of the underlying electromagnetic waveform; The variance representing the visual perception estimation error is obtained from the covariance matrix of the optical flow tracking reprojection error of the target feature points by the visual sensor, and the calculation formula is:

[0070] In the formula, The velocity error variance generated when estimating relative velocity vectors using a vision sensor. The variance of the angle error generated when estimating the direction of arrival angle for a visual sensor; This represents the frequency offset error variance estimated from the underlying electromagnetic waveform, obtained by extracting the most recent... The sliding time variance of the OFDM cyclic prefix autocorrelation peak within one symbol period is calculated in real time, and the calculation formula is:

[0071] In the formula, The length of the cyclic prefix of the OFDM symbol. This represents the instantaneous signal-to-noise ratio of the received signal within the current observation window. High-precision equivalent instantaneous dynamic frequency offset characteristics are output through weighted calculation. The equivalent instantaneous dynamic frequency offset characteristic The expression comprehensively characterizes carrier frequency offset, the Doppler effect of relative motion between the two ends, and phase jitter caused by channel non-stationarity, and is as follows:

[0072] in, The visual prior Doppler frequency shift prediction value in S1.3 is adopted; S1.5, Construct based on The augmented state vector of a Lie group manifold The equivalent instantaneous dynamic frequency offset characteristics This is converted into the Doppler phase evolution rate and compared with the three-dimensional spatial angular velocity vector extracted by the visual sensor. and linear velocity vector Together they constitute the elements of the Lie algebra. And generate state transition tensors through exponential mapping. This achieves a nonlinear conversion from a three-dimensional geometric trajectory to an electromagnetic phase, expressed as:

[0073] in, Let be the augmented state vector at the k-th state update time. It is a three-dimensional rotation matrix. It is a three-dimensional translation vector. It is an electromagnetic phase state quantity. This is the adjoint matrix form corresponding to the Lie algebra elements. This refers to the time step for state updates; In this embodiment, S2 is used to explain how strong interference is first measured in the front-end observation and then transformed into the modulation basis of the Kalman observation noise matrix. It mainly employs interference cancellation ratio to drive nonlinear modulation of the observation noise, including: S2.1 Introduce a variable step-size normalized minimum mean square error adaptive filter based on a tapped delay line structure. Let the first step be... The input signal delay vector at each adaptive filtering iteration is The filter tap weight vector is The predicted estimates for multi-source complex interference are:

[0074] in, This represents the transpose operation of the weight vector. This is the index of the discrete iteration time of the adaptive filter; Real received sample values ​​sampled in real time at the antenna end Compared with the predicted value The residual calculation error signal is used to obtain the purified expected signal component after removing interference:

[0075] S2.2 To adapt to non-stationary disturbances in highly dynamic environments, the weight vector step size factor is... An exponential adaptive update mechanism based on error energy is adopted, and its expression is:

[0076] in, For maximum step size constraint, This is the step-size adaptive sensitivity control coefficient, used to adjust the degree of influence of error energy on the weight vector step-size factor; Update the filter weight vector using dynamic step size:

[0077] in, This is the index for the discrete iteration time of the adaptive filter. For the first The updated filter tap weight vector at each iteration time. To prevent positive small constants with a denominator of zero, It is the complex conjugate of the error signal; S2.3. Using an exponentially weighted moving average mechanism, simultaneously predict the interference power and the cleaned signal power statistically, and define the interference cancellation ratio. for:

[0078] in, The smoothing forgetting factor, used to control the proportion of historical energy data, has a value range of [value range missing]. , For the historical sampling index within the sliding statistics window, For the i-th historical sample data in the th... The exponentially decaying weights corresponding to each adaptive filtering iteration time. Let i be the predicted interference signal at the i-th sampling time. This is the purification error signal obtained after interference cancellation at the i-th sampling time; S2.4 When a sudden strong electromagnetic pulse or colored interference causes filter tracking lag, the interference cancellation ratio... This will produce a descent abrupt change, constructing a nonlinear adaptive mapping function, and outputting an observation noise penalty coefficient for S3 channel tracking observation updates. :

[0079] in, This is the steady-state reference threshold under undisturbed conditions. It is a nonlinear penalty control factor, and in S5, it is optimized and updated online at a slow scale using the particle swarm optimization algorithm. when When below the safety threshold, It is amplified exponentially to reduce the weight of contaminated electromagnetic observations in the Kalman update.

[0080] In this embodiment, S3 is used to illustrate that the improvement of Kalman filtering is not simply increasing the noise, but selectively adjusting the dimension of process noise characteristics according to the physical propagation direction. This mainly employs array-guided matching process noise physical-directional expansion channel tracking, including: S3.1 Establish a Kalman recursive state-space model and use the state transition tensor output by S1. Perform prior prediction of channel states:

[0081] in, For the first The prior prediction estimate vector of the channel state at time step. This is the posterior state estimation vector from the previous time step; For the initial process noise covariance matrix Perform eigenvalue decomposition:

[0082] in, The eigenvector matrix of the process noise covariance matrix. , It is a diagonal matrix of eigenvalues. ; S3.2. Based on the obstructed physical path predicted in S1, extract its physical direction of arrival relative to the receiving antenna array. Let the angle of arrival of this obstruction path be... The receiver is configured with array element spacing. , number of array elements A uniform linear array; Based on the phase deflection law of electromagnetic wave propagation in space, this physical direction of arrival is mapped to the physical steering vector of the receiving antenna array. :

[0083] in, For radio frequency carrier wavelength, The element spacing of the receiving array. Indicates matrix transpose. The imaginary unit; The guide vector The array response characteristics of the blocked physical path in electromagnetic space were accurately characterized; S3.3, in the eigenvector matrix In the middle, the index of the target feature vector that is most correlated with the occluded physical path is locked by the spatial inner product norm. :

[0084] in, Eigenvector matrix The z-th eigenvector in For feature vector index; Only for target feature vectors Corresponding eigenvalues Apply dynamic expansion ratio Reshape the eigenvalue diagonal matrix ,make The remaining non-target feature values ​​remain unchanged; Combining eigenvector matrix Reconstruct the process noise covariance matrix at the current time step:

[0085] Combined with the reconstructed process noise covariance matrix Calculate the prediction error covariance matrix:

[0086] in, For the first The prior prediction error covariance matrix of the channel state at time step. Let be the posterior error covariance matrix of the previous time step. Represents the conjugate transpose of a matrix; S3.4 Receive the observation noise penalty coefficient output by S2 The variance of the basic measurement noise is modulated to generate the observation noise matrix at the k-th state update time. :

[0087] in, It is the identity matrix; S3.5, Combine the observation signal matrix at the k-th state update time. Dynamically calculate adaptive Kalman gain :

[0088] in, This represents the matrix inversion operation; Using Kalman gain Combined with actual received baseband radio frequency signals The residuals from the predicted observed signals are used to complete the posterior update of the symbol-level channel state:

[0089] in, To use the received signal at the k-th observation time to complete the updated channel state posterior estimation vector; The posterior error covariance matrix is ​​updated synchronously to provide high-precision initial values ​​for tracking in the next symbol period, forming a closed-loop iteration:

[0090] in, This is for updating the completed posterior error covariance matrix.

[0091] In this embodiment, S4 is used to illustrate that the optimization objective of this method comes from the actual decoding process, rather than isolated filtering errors or empirical parameters. It mainly employs decoding lattice topological entropy to drive joint demodulation and decoding reliability feedback, including: S4.1 Construct a search lattice graph based on the soft-output Viterbi algorithm, and receive the channel state posterior estimation vector at the k-th state update time output by S3. With the One decoded received symbol Define the branch metric as:

[0092] in, and These represent the previous state and the current state in the lattice diagram, respectively. Indicates the first The theoretical transmitted signal on the state transition branch at each decoding and receiving symbol moment. For the time index of the decoded received symbol; Recursively accumulate data using the add-compare-select operation to reach the current state. Path metric:

[0093] in, This is an index for the received symbol time in the Viterbi decoding process. , This represents the total number of received symbols in the current decoded frame. Indicates the first The cumulative path metric value for each decoded received symbol reaching the current state at each time step. This represents the cumulative path metric value from the time the previously decoded received symbol arrived at the previous state. This represents the current branch metric. This represents the set of predecessor states of the current state; S4.2 In the backtracking optimization phase of Viterbi decoding, for a length of... Each decoding moment within the backtracking window Extract the surviving path metric to reach the optimal state node. metric for the abandoned competing path ; Introducing a scaling factor constrained by the fundamental noise The path metric is converted into a probability confidence score of local state transition using the Softmax function. :

[0094] in, This is an index for path types; surv indicates a surviving path, and comp indicates a contested path. Indicates the first Decoding time path The corresponding cumulative path metric, For the backtracking window length, This serves as an index for the decoding time within the backtracking window; Calculate the Shannon information entropy for the local probability distribution and throughout the backtracking window. Calculate the mean value internally and output the decoded lattice topological entropy. :

[0095] When high-dynamic, strong interference degrades electromagnetic observations, the difference in metrics between surviving and competing paths narrows, the local probability distribution tends to become more uniform, and the decoding lattice topological entropy... Consequently, the topological entropy of the decoding lattice increases, therefore, the topological entropy of the decoding lattice increases. It can be used to quantify the decoding decision uncertainty under the current channel estimation results and receiver parameter configuration, and serve as the fitness feedback quantity of the closed-loop optimization module.

[0096] In this embodiment, S5 is used to explain that particle swarm optimization is not a general black-box search, but a closed-loop optimization of receiver parameters constrained by both visual physical boundaries and decoding reliability. It mainly employs physical manifold boundary-constrained closed-loop heterogeneous particle swarm optimization, including: S5.1, Construct a system that includes dynamic expansion ratios With nonlinear penalty control factor Hyperparameter optimization vector ; S5.2. Transform the maximum physical motion limit extracted in S1 into the rigid physical manifold boundary of the parameter search space. Construct the boundary potential function for this boundary. :

[0097] in, For the search space dimension, For the search space dimension index, hyperparameter vector The position coordinates in the q-th dimension and These are the upper and lower physical limits of this dimension, respectively; When the particle's position approaches the boundary of the physical manifold, the boundary potential function Generate a repulsive gradient inward along the boundary normal and add it to the particle swarm velocity update equation. ,in The potential energy repulsion gain coefficient; S5.3 Receive the decoded lattice topology entropy output from S4 Minimizing the desired topological entropy is used as the fitness function for particle swarm optimization:

[0098] in, Represents the hyperparameter vector The corresponding fitness function value, Represents the mathematical expectation operation; S5.4, drive the particle swarm to perform intergenerational state updates, the first The particle in the first The speed update formula for the second optimization iteration is:

[0099] in, For particle indexing, Index of the number of iterations for particle swarm optimization Let g be the optimization velocity of the g-th particle. Let be the hyperparameter position vector of the g-th particle. This represents the optimal position of the historical individual of the g-th particle. The optimal position globally. For inertial weights, and Individual cognitive learning factors and social learning factors, and A random number uniformly distributed within the interval [0,1]. The potential energy repulsion gain coefficient defined in S5.2; At the same time, the updated velocity is used to drive the particles to perform position jumps:

[0100] in, For the g-th particle in the th... The hyperparameter position vector after the optimization iteration is completed; S5.5. Through velocity and position iteration guided by the physical manifold, the decoding lattice topological entropy is finally locked. Achieving the minimum global optimal hyperparameter set It then sends the closed-loop data down to the underlying physical layer, and updates the process noise covariance matrix in S3 and the observation noise penalty coefficient in S2 in real time.

[0101] To verify the receiving performance of this invention under conditions of high dynamic occlusion and sudden interference, a time-varying channel simulation scenario based on multipath spatial modes was constructed. The simulation carrier frequency was set to 28 GHz, the channel state update time was 100 μs, and a four-dimensional equivalent array channel model was used at the receiver. The Doppler frequency shifts corresponding to each spatial propagation mode were set to 2200 Hz, -1450 Hz, 760 Hz, and 260 Hz, respectively. During the simulation, one of the physical paths was shaded, and spatial colored sudden interference was superimposed on some of the occluded intervals.

[0102] The simulation setup includes three comparison methods: The first is the traditional fixed-parameter Kalman method, which uses electromagnetic frequency offset estimation to construct the state transition matrix while keeping the process noise matrix and observation noise matrix fixed. The second is the overall noise expansion adaptive Kalman method, which uses electromagnetic frequency offset to construct the state transition matrix, uniformly expands all feature dimensions of the process noise matrix when a channel anomaly is detected, and modulates the observation noise using the same interference cancellation ratio as this invention, but does not use visual-electromagnetic frequency offset fusion and occlusion direction matching. The third is the visual-electromagnetic joint sensing orientation adaptive method described in this invention, which uses visual-electromagnetic fusion frequency offset to construct the state transition tensor, directionally expands the process noise matrix according to the occluded physical path direction of the visual output, and modulates the observation noise matrix according to the interference cancellation ratio.

[0103] Figure 2 The figure shows the total channel normalized mean square error across all spatial dimensions, used to reflect the overall channel tracking performance of different methods; Figure 3 The figure shows the normalized mean square error (MSE) for the spatial dimensions excluding those corresponding to obstructed physical paths, and only calculating the MSE for the remaining unobstructed dimensions. This is used to analyze the impact of process noise expansion on the normal channel dimensions. The lower the MSE, the closer the channel state estimation result is to the actual channel state, and the better the channel tracking performance.

[0104] Depend on Figures 2-3 As can be seen, when the channel state is relatively stable, all methods can maintain continuous tracking, and the channel tracking error of the method of this invention remains at a low level overall. When the physical path is shadowed, the traditional fixed-parameter Kalman method experiences a significant increase in tracking error because the state model cannot reflect path changes in a timely manner. The overall noise dilation method improves the response to channel changes by increasing the process uncertainty of all state dimensions, but compared with the method of this invention, its estimation error of the total channel and non-shadow dimensions is still relatively large. In the burst interference range, the tracking errors of all methods increase to varying degrees, while the error increase of the method of this invention is relatively small.

[0105] This invention determines the target feature dimension with the highest correlation to the occluded physical path direction obtained from visual perception within the eigenvectors of the process noise covariance matrix. Directional dilation is applied only to this feature dimension, while other feature dimensions maintain their original update state. Simultaneously, the interference cancellation ratio is used to adjust the weight of contaminated observations in the Kalman update. From the overall trend, the method of this invention shows lower total channel tracking error in the occluded and burst interference intervals compared to the two contrasting methods, and the error variation in the unoccluded spatial dimension is also more stable. These results demonstrate that the combined application of visual-electromagnetic fusion state prediction, directional dilation of process noise, and adaptive modulation of observation noise can enhance the tracking capability of occluded paths and reduce the impact of overall noise dilation on other normal channel dimensions.

[0106] Figure 4 The figure shows the receive bit error rate (BER) for the three methods when the signal-to-noise ratio (SNR) ranges from 0 dB to 18 dB. The BER is used to evaluate the performance of different channel tracking methods in recovering service data after equalization and symbol decision at the receiver. The lower the BER, the higher the reception reliability.

[0107] Depend on Figure 4 As can be seen, the bit error rate (BER) of all three methods generally decreases with increasing signal-to-noise ratio (SNR). The traditional fixed-parameter Kalman method, unable to adaptively adjust noise parameters according to highly dynamic channel changes and sudden interference, exhibits a relatively high BER under all SNR conditions. The overall noise dilation method enhances the response to channel changes by increasing the uncertainty of process noise, thus achieving a lower BER than the traditional fixed-parameter Kalman method. However, overall dilation also alters the filter update state in the non-occluded dimension.

[0108] The method of this invention generally achieves a lower bit error rate (BER) than the two comparative methods under various signal-to-noise ratio (SNR) conditions. Under higher SNR conditions, as the influence of thermal noise gradually weakens, residual burst interference and high-dynamic channel estimation errors become the main factors affecting reception performance. The BER curve exhibits local fluctuations at certain SNR positions, but this does not affect its overall downward trend. These results demonstrate that this invention, through visual-electromagnetic frequency offset fusion, process noise directional reshaping, and observation noise nonlinear modulation, improves the consistency between channel estimation results and actual channel states, and reduces the BER probability during equalization and symbol decision processes.

[0109] Simulation results demonstrate that this invention can map spatial motion, propagation path, and obstruction direction information obtained through visual perception to the receiver's state prediction and process noise update processes, and adaptively handle sudden interference by combining underlying electromagnetic waveform observations. Under conditions of high dynamics, multipath obstruction, and sudden interference, this invention can simultaneously maintain the tracking capability of the obstructed path and the estimation stability of the non-obstructed dimension, thereby improving the channel tracking accuracy and service data reception reliability of the physical layer receiver.

[0110] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the 6G high dynamic adaptive reception method driven by the above-described visual-electromagnetic joint sensing.

[0111] This invention also provides a computer device, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the 6G high dynamic adaptive reception method driven by the above-described visual-electromagnetic joint sensing.

[0112] This invention also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the steps of the 6G high dynamic adaptive reception method driven by the above-described visual-electromagnetic joint sensing.

[0113] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above-described visual-electromagnetic joint sensing driven 6G high dynamic adaptive reception method.

[0114] It should be noted that those skilled in the art will understand that all or part of the steps implemented in the embodiments of the present invention can be implemented entirely or partially by software, hardware, firmware, or any combination thereof. When implemented in hardware, it can be implemented entirely or partially by purchasing standard parts or modifications. When implemented in software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state disks (SSDs)).

[0115] In summary, this invention addresses the problems of unstable and unreliable reception capabilities under existing high-dynamic non-stationary channels by providing a 6G high-dynamic adaptive reception method driven by visual-electromagnetic joint sensing. This method collaboratively designs spatial geometric information obtained visually, electromagnetic waveform estimation, channel tracking, demodulation and decoding reliability assessment, and closed-loop optimization of reception parameters. This allows the physical propagation laws to be mapped into the receiver algorithm parameter update process, thereby improving reliable reception capabilities under high-dynamic non-stationary channels.

[0116] It should be understood that the examples and embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various modifications or changes based on them. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing, characterized in that, Includes the following steps: S1. The receiver processes the baseband radio frequency signal, uses a visual sensor to acquire the three-dimensional spatial geometric information of communication nodes, obstructions and scatterers, classifies and identifies direct, reflected scattering and diffraction propagation mechanisms, and performs confidence-weighted fusion with the electromagnetic frequency offset estimate obtained from OFDM cyclic prefix autocorrelation in the received baseband radio frequency signal, and outputs a state transition tensor for channel tracking. S2. Perform adaptive interference suppression on the received baseband radio frequency signal, extract the interference cancellation ratio, and output the observation noise penalty coefficient to reduce the weight of the contaminated observation signal based on the abrupt change characteristics of the interference cancellation ratio. S3. During Kalman channel tracking, an array steering vector is constructed based on the visually predicted obstructed physical path and matched with the eigenvector of the process noise covariance matrix. Only the eigenvalues ​​of the matching dimension are directionally expanded. At the same time, the observation noise matrix is ​​modulated using the observation noise penalty coefficient, and the channel state estimate is output. S4. Based on the channel state estimate, perform joint demodulation and Viterbi decoding to recover the service bit stream, and calculate the decoding lattice topology entropy from the path metrics of the surviving path and the contention path. S5. With the goal of minimizing the topological entropy of the decoding lattice, and combined with the physical manifold boundary measured by the visual sensor, the observation noise penalty control factor and the process noise directional expansion ratio are updated online through a particle swarm optimization algorithm with a boundary repulsion term, and fed back to S2 and S3 respectively, so as to realize the online update of the observation noise modulation and process noise directional reshaping parameters.

2. The 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing as described in claim 1, characterized in that, The S1 employs a propagation mechanism to perceive and construct the state transition of a visual-electromagnetic confidence-gated Lie group, including: S1.

1. For the non-stationary random process caused by the high-speed movement of both the mobile base station and the target node, the complex equivalent baseband physical model of the received signal is constructed as follows: in, For the first The received signal at each sampling time, For multipath channel complex gain, For the existence of the first Transmitting a complex equivalent baseband signal with a sampling delay. It is additive complex white Gaussian noise. For true frequency offset, The sampling period is For sampling time index, For sampling delay index, This represents the total sampling delay. The three-dimensional spatial geometry topology of the communication environment is acquired using a visual sensor, and the physical propagation mechanisms of direct path, reflected scattering path and diffraction path are classified and identified. The spatial geometric features of each multipath component are extracted respectively. S1.

2. Based on the direct path, reflected scattering path, diffraction path, and shadow occlusion state identified by the visual sensor, a propagation mechanism confidence gating factor is set for each candidate physical path, where: For unobstructed direct rays, improve their Doppler prediction contribution; For the blocked direct path, reduce its Doppler prediction contribution and output the direction of arrival of the blocked direct path to S3; For reflected scattering paths and diffraction paths, the reliability of their participation in prediction is determined based on the geometric stability of the reflecting surface, scattering body, or obstacle edge. Thus, visual spatial geometric information is first transformed into path-by-path propagation confidence, path-by-path occlusion state, and path-by-path wave direction of arrival, and then participates in the prediction of visual prior Doppler frequency shift features. S1.

3. Based on the confidence gating results of the path-by-path propagation mechanism described in S1.2, a multi-mechanism joint visual prior Doppler frequency shift feature prediction model is constructed to address the multipath superposition and shadow occlusion effects in high dynamic environments: in, This represents the visual prior Doppler frequency shift prediction value at the nth sampling time. The total number of valid physical multipaths identified by the visual perception module. For effective physical path indexing, , The power attenuation weight for the r-th path is... The binary shadow occlusion state indicator for the r-th path. As the confidence gating factor for the r-th path propagation mechanism, At the speed of light, For carrier frequency, and Let r be the spatial relative velocity vector of the r-th radius and the unit vector of the direction of arrival, respectively. Represents the inner product of spatial vectors. Let r be the confidence level for the geometric stability of the r-th path. Confidence level for the type of propagation mechanism of the r-th path; S1.

4. For the received radio frequency signal, the prior value of the frequency offset of the underlying electromagnetic waveform is estimated by utilizing the autocorrelation characteristics of the cyclic prefix in the OFDM symbol: in, The prior value of frequency offset of the underlying electromagnetic waveform is obtained based on the OFDM cyclic prefix autocorrelation characteristics estimation. For the number of points in the Fast Fourier Transform, This represents the complex conjugate sample value of the received signal after one symbol period; To integrate the underlying electromagnetic waveform estimate with the visual prior Doppler frequency shift characteristics in S1.3, a confidence weighting coefficient is calculated based on the variance of the estimation errors of the two: in: The confidence weighting coefficients for the visual prior Doppler frequency shift features are... The confidence weighting coefficient for the frequency offset estimate of the underlying electromagnetic waveform; The variance representing the visual perception estimation error is obtained from the covariance matrix of the optical flow tracking reprojection error of the target feature points by the visual sensor, and the calculation formula is: In the formula, The velocity error variance generated when estimating relative velocity vectors using a vision sensor. The variance of the angle error generated when estimating the direction of arrival angle for a visual sensor; This represents the frequency offset error variance estimated from the underlying electromagnetic waveform, obtained by extracting the most recent... The sliding time variance of the OFDM cyclic prefix autocorrelation peak within one symbol period is calculated in real time, and the calculation formula is: In the formula, The length of the cyclic prefix of the OFDM symbol. This represents the instantaneous signal-to-noise ratio of the received signal within the current observation window. High-precision equivalent instantaneous dynamic frequency offset characteristics are output through weighted calculation. The equivalent instantaneous dynamic frequency offset characteristic The expression comprehensively characterizes carrier frequency offset, the Doppler effect of relative motion between the two ends, and phase jitter caused by channel non-stationarity, and is as follows: in, The visual prior Doppler frequency shift prediction value in S1.3 is adopted; S1.5, Construct based on The augmented state vector of a Lie group manifold The equivalent instantaneous dynamic frequency offset characteristics This is converted into the Doppler phase evolution rate and compared with the three-dimensional spatial angular velocity vector extracted by the visual sensor. and linear velocity vector Together they constitute the elements of the Lie algebra. And generate state transition tensors through exponential mapping. This achieves a nonlinear conversion from a three-dimensional geometric trajectory to an electromagnetic phase, expressed as: in, Let be the augmented state vector at the k-th state update time. It is a three-dimensional rotation matrix. It is a three-dimensional translation vector. It is an electromagnetic phase state quantity. This is the adjoint matrix form corresponding to the Lie algebra elements. This is the time step for state updates.

3. The 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing as described in claim 1, characterized in that, The S2 employs interference cancellation ratio-driven observation noise nonlinear modulation, including: S2.1 Introduce a variable step-size normalized minimum mean square error adaptive filter based on a tapped delay line structure. Let the first step be... The input signal delay vector at each adaptive filtering iteration is The filter tap weight vector is The predicted estimates for multi-source complex interference are: in, This represents the transpose operation of the weight vector. This is the index of the discrete iteration time of the adaptive filter; Real received sample values ​​sampled in real time at the antenna end Compared with the predicted value The residual calculation error signal is used to obtain the purified expected signal component after removing interference: ； S2.2 To adapt to non-stationary disturbances in highly dynamic environments, the weight vector step size factor is... An exponential adaptive update mechanism based on error energy is adopted, and its expression is: in, For maximum step size constraint, This is the step-size adaptive sensitivity control coefficient, used to adjust the degree of influence of error energy on the weight vector step-size factor; Update the filter weight vector using dynamic step size: in, This is the index for the discrete iteration time of the adaptive filter. For the first The updated filter tap weight vector at each iteration time. To prevent positive small constants with a denominator of zero, It is the complex conjugate of the error signal; S2.

3. Using an exponentially weighted moving average mechanism, simultaneously predict the interference power and the cleaned signal power statistically, and define the interference cancellation ratio. for: in, The smoothing forgetting factor, used to control the proportion of historical energy data, has a value range of [value range missing]. , For the historical sampling index within the sliding statistics window, For the i-th historical sample data in the th... The exponentially decaying weights corresponding to each adaptive filtering iteration time. Let i be the predicted interference signal at the i-th sampling time. The purified error signal is obtained after interference cancellation at the i-th sampling time. S2.4 When a sudden strong electromagnetic pulse or colored interference causes filter tracking lag, the interference cancellation ratio... This will produce a descent abrupt change, constructing a nonlinear adaptive mapping function, and outputting an observation noise penalty coefficient for S3 channel tracking observation updates. : in, This is the steady-state reference threshold under undisturbed conditions. It is a nonlinear penalty control factor, and in S5, it is optimized and updated online at a slow scale using the particle swarm optimization algorithm. when When below the safety threshold, It is amplified exponentially to reduce the weight of contaminated electromagnetic observations in the Kalman update.

4. The 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing as described in claim 1, characterized in that, The S3 employs array-guided matching process noise-physical directional dilatation channel tracking, including: S3.1 Establish a Kalman recursive state-space model and use the state transition tensor output by S1. Perform prior prediction of channel states: in, For the first The prior prediction estimate vector of the channel state at time step. This is the posterior state estimation vector from the previous time step; For the initial process noise covariance matrix Perform eigenvalue decomposition: in, The eigenvector matrix of the process noise covariance matrix is... , It is a diagonal matrix of eigenvalues. ; S3.

2. Based on the obstructed physical path predicted in S1, extract its physical direction of arrival relative to the receiving antenna array. Let the angle of arrival of this obstruction path be... The receiver is configured with array element spacing. , number of array elements A uniform linear array; Based on the phase deflection law of electromagnetic wave propagation in space, this physical direction of arrival is mapped to the physical steering vector of the receiving antenna array. : in, For radio frequency carrier wavelength, The element spacing of the receiving array. Indicates matrix transpose. The imaginary unit; The guide vector The array response characteristics of the blocked physical path in electromagnetic space were accurately characterized; S3.3, in the eigenvector matrix In the middle, the index of the target feature vector that is most correlated with the occluded physical path is locked by the spatial inner product norm. : in, Eigenvector matrix The z-th eigenvector in For feature vector index; Only for target feature vectors Corresponding eigenvalues Apply dynamic expansion ratio Reshape the eigenvalue diagonal matrix ,make The remaining non-target feature values ​​remain unchanged; Combining eigenvector matrix Reconstruct the process noise covariance matrix at the current time step: Combined with the reconstructed process noise covariance matrix Calculate the prediction error covariance matrix: in, For the first The prior prediction error covariance matrix of the channel state at time step. Let be the posterior error covariance matrix of the previous time step. Represents the conjugate transpose of a matrix; S3.4, Observation noise penalty coefficient received from S2 output The variance of the basic measurement noise is modulated to generate the observation noise matrix at the k-th state update time. : in, It is the identity matrix; S3.5, Combine the observation signal matrix at the k-th state update time. Dynamically calculate adaptive Kalman gain : in, This represents the matrix inversion operation; Using Kalman gain Combined with actual received baseband radio frequency signals The residuals from the predicted observed signals are used to complete the posterior update of the symbol-level channel state: in, To use the received signal at the k-th observation time to complete the updated channel state posterior estimation vector; The posterior error covariance matrix is ​​updated synchronously to provide high-precision initial values ​​for tracking in the next symbol period, forming a closed-loop iteration: in, This is for updating the completed posterior error covariance matrix.

5. The 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing as described in claim 1, characterized in that, The S4 employs decoding lattice topological entropy to drive joint demodulation decoding reliability feedback, including: S4.1 Construct a search lattice graph based on the soft-output Viterbi algorithm, and receive the channel state posterior estimation vector at the k-th state update time output by S3. With the One decoded received symbol Define the branch metric as: in, and These represent the previous state and the current state in the lattice diagram, respectively. Indicates the first The theoretical transmitted signal on the state transition branch at each decoding and receiving symbol moment. For the decoding and receiving symbol timing index; Recursively accumulate data using the add-compare-select operation to reach the current state. Path metric: in, This is an index for the received symbol time in the Viterbi decoding process. , This represents the total number of received symbols in the current decoded frame. Indicates the first The cumulative path metric value for each decoded received symbol reaching the current state at each time step. This represents the cumulative path metric value from the time the previously decoded received symbol arrived at the previous state. This represents the current branch metric. This represents the set of predecessor states of the current state; S4.2 In the backtracking optimization phase of Viterbi decoding, for a length of... Each decoding moment within the backtracking window Extract the surviving path metric to reach the optimal state node. metric for the abandoned competing path ; Introducing a scaling factor constrained by the fundamental noise The path metric is converted into a probability confidence score of local state transition using the Softmax function. : in, This is an index for path types; surv indicates a surviving path, and comp indicates a contested path. Indicates the first Decoding time path The corresponding cumulative path metric, For the backtracking window length, This serves as an index for the decoding time within the backtracking window; Calculate the Shannon information entropy for the local probability distribution and throughout the backtracking window. Calculate the mean value internally and output the decoded lattice topological entropy. : When high-dynamic, strong interference degrades electromagnetic observations, the difference in metrics between surviving and competing paths narrows, the local probability distribution tends to become more uniform, and the decoding lattice topological entropy... Consequently, the topological entropy of the decoding lattice increases, therefore, the topological entropy of the decoding lattice increases. It can be used to quantify the decoding decision uncertainty under the current channel estimation results and receiver parameter configuration, and serve as the fitness feedback quantity of the closed-loop optimization module.

6. The 6G high dynamic adaptive reception method driven by visual-electromagnetic joint sensing as described in claim 1, characterized in that, The S5 method employs physical manifold boundary constraints for closed-loop heterogeneous particle swarm optimization, including: S5.1, Construct a system that includes dynamic expansion ratios With nonlinear penalty control factor Hyperparameter optimization vector ; S5.

2. Transform the maximum physical motion limit extracted in S1 into the rigid physical manifold boundary of the parameter search space. Construct the boundary potential function for this boundary. : in, For the search space dimension, For indexing the search space dimensions, hyperparameter vector The position coordinates in the q-th dimension and These are the upper and lower physical limits of this dimension, respectively; When the particle's position approaches the boundary of the physical manifold, the boundary potential function Generate a repulsive gradient inward along the boundary normal and add it to the particle swarm velocity update equation. ,in The potential energy repulsion gain coefficient; S5.3, Receive the decoded lattice topology entropy output from S4 Minimizing the desired topological entropy is used as the fitness function for particle swarm optimization: in, Represents the hyperparameter vector The corresponding fitness function value, Represents the mathematical expectation operation; S5.4, drive the particle swarm to perform intergenerational state updates, the first The particle in the first The speed update formula for the second optimization iteration is: in, For particle indexing, Index of the number of iterations for particle swarm optimization Let g be the optimization velocity of the g-th particle. Let be the hyperparameter position vector of the g-th particle. This represents the optimal position of the historical individual of the g-th particle. To be the globally optimal position For inertial weights, and Individual cognitive learning factors and social learning factors, and A random number uniformly distributed within the interval [0,1]. The potential energy repulsion gain coefficient defined in S5.2; At the same time, the updated velocity is used to drive the particles to perform position jumps: in, For the g-th particle in the th... The hyperparameter position vector after the optimization iteration is completed; S5.

5. Through velocity and position iteration guided by the physical manifold, the decoding lattice topological entropy is finally locked. Achieving the minimum global optimal hyperparameter set It then sends the closed-loop data down to the underlying physical layer, and updates the process noise covariance matrix in S3 and the observation noise penalty coefficient in S2 in real time.

Images