A method and device for realizing signal-to-noise ratio estimation of multi-scenario terminals

Abstract

The present invention discloses a method and device for realizing terminal signal-to-noise ratio estimation in multiple scenarios. The method comprises: obtaining a received signal sample sequence, platform attitude data, antenna pointing data and environmental parameter data, obtaining a feature vector through feature extraction and adaptive window processing; generating a compensation feature vector based on the feature vector and environmental compensation data; performing feature fusion in combination with platform state data to obtain a scene feature vector; performing parameter optimization on the scene feature vector and establishing a dynamic memory mechanism to generate memory state data; performing dynamic correction and robustness processing on the signal-to-noise ratio using the memory state data to obtain a corrected signal-to-noise ratio; performing quality evaluation based on a performance evaluation vector and a dynamic threshold vector, and outputting a final signal-to-noise ratio and a control parameter set. The present invention improves the accuracy and reliability of signal-to-noise ratio estimation through adaptive feature extraction, multi-dimensional compensation and dynamic memory mechanism.

Description

Technical Field

[0001] The present invention belongs to the field of signal processing, and in particular to a method and device for realizing multi-scenario terminal signal-to-noise ratio estimation. Background Art

[0002] In modern mobile communication systems, terminal devices need to maintain reliable communication performance in complex and changing environments; the signal-to-noise ratio (SNR), that is, the ratio of signal power to background noise power, is an important parameter that characterizes channel characteristics; as a key indicator for measuring communication quality, accurate estimation of the SNR is of great significance for link adaptation, power control, channel coding and modulation selection of communication systems; especially in high-dynamic scenarios such as high-speed rail, aviation, and maritime, terminal devices face drastic platform posture changes and complex electromagnetic environments, which puts higher requirements on real-time and accurate estimation of the SNR; accurate SNR estimation can not only help the system adjust communication parameters in a timely manner and optimize resource allocation, but also provide an important basis for fault diagnosis and performance evaluation of communication systems.

[0003] At present, the commonly used SNR estimation methods mainly include estimation methods based on pilot sequences, moment estimation methods and maximum likelihood estimation methods. The pilot sequence-based method estimates the SNR by measuring the received signal power at a known pilot position, which is simple to implement but occupies additional bandwidth resources. The moment estimation method uses high-order statistics of the received signal for estimation, with low computational complexity but significantly degrades performance under low SNR conditions. The maximum likelihood estimation method has the best theoretical performance, but has high computational complexity and slow convergence speed. In practical applications, these methods often use fixed signal processing windows and preset feature extraction parameters, and are mainly designed for static or low-dynamic scenarios, lacking the ability to adapt to complex scenarios.

[0004] However, the existing technical solutions have the following technical problems in practical applications: first, the traditional fixed window processing method cannot be dynamically adjusted according to the platform motion state, which may lead to signal truncation or redundancy during intense motion, affecting the accuracy of feature extraction; second, the existing feature extraction methods often consider the time domain, frequency domain or phase features separately, lack of effective feature fusion mechanism, and it is difficult to fully reflect the signal characteristics; third, the influence of environmental parameters and platform posture on signal characteristics is not fully compensated, especially when environmental parameters such as temperature and air pressure change rapidly, the estimation results will have large deviations; fourth, the traditional method is too simple in evaluating the reliability of features, and no complete reliability evaluation system has been established, which may lead to the output of unreliable estimation results in complex scenarios; fifth, the existing technology generally lacks an effective mechanism for utilizing historical data, and fails to improve the temporal consistency of estimation by establishing dynamic memory; finally, there is a lack of multi-dimensional evaluation of system performance and parameter adaptive optimization mechanism, which makes it difficult to ensure stable estimation performance in different scenarios; these technical problems seriously restrict the application effect of signal-to-noise ratio estimation in complex scenarios. Summary of the invention

[0005] The purpose of the invention is to propose a method and device for realizing signal-to-noise ratio estimation of multi-scenario terminals to solve the above-mentioned problems existing in the existing technology.

[0006] Technical solution: A method for estimating the signal-to-noise ratio of a terminal in multiple scenarios, comprising the following steps:

[0007] S1. Obtain the received signal sample sequence, platform attitude data, antenna pointing data and environmental parameter data, obtain the feature vector through feature extraction and adaptive window processing; calculate the environmental compensation data based on the environmental parameter data; generate the compensation feature vector based on the feature vector and the environmental compensation data; obtain the fused feature vector through feature fusion processing based on the compensation feature vector; perform reliability evaluation and optimization based on the fused feature vector, and output the optimized feature sequence and reliability index;

[0008] S2. Perform frequency domain analysis on the optimized feature sequence and reliability index to obtain energy distribution features; construct a noise benchmark based on the energy distribution features to obtain a noise feature vector; perform cross-validation based on the noise feature vector, environmental compensation data and compensation feature vector to generate a credibility index; calculate a preliminary signal-to-noise ratio based on the energy distribution features and the noise feature vector; optimize the preliminary signal-to-noise ratio based on the credibility index, and output the optimized signal-to-noise ratio and reliability matrix;

[0009] S3, constructing a scene feature vector based on the optimized signal-to-noise ratio, the reliability matrix, and the pre-stored platform state data; optimizing parameters based on the scene feature vector to obtain an optimized parameter set; constructing a dynamic memory mechanism based on the optimized parameter set to generate memory state data; dynamically correcting the optimized signal-to-noise ratio based on the memory state data to obtain a corrected signal-to-noise ratio; performing robustness processing on the corrected signal-to-noise ratio to output a robust signal-to-noise ratio;

[0010] S4. Based on the robust signal-to-noise ratio, scene feature vector, optimized parameter set and memory state data, a performance evaluation vector is constructed; based on the performance evaluation vector, a dynamic threshold vector is generated; quality evaluation is performed on the performance evaluation vector and the dynamic threshold vector to obtain a quality evaluation value; based on the quality evaluation value, the robust signal-to-noise ratio is optimized to generate a final signal-to-noise ratio; at the same time, based on the quality evaluation value, parameter feedback control is performed to output a control parameter set.

[0011] A device for realizing multi-scenario terminal signal-to-noise ratio estimation, comprising:

[0012] at least one processor; and,

[0013] a memory communicatively connected to at least one of the processors; wherein,

[0014] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the multi-scenario terminal signal-to-noise ratio estimation method.

[0015] Beneficial effects: The present invention ensures the complete capture of original signal characteristics, effectively eliminates the interference of external factors, improves the reliability and stability of estimation results, realizes continuous optimization of system performance, and can adapt to changes in signal characteristics in different scenarios; while ensuring estimation accuracy, it maintains a high computational efficiency, achieves high accuracy, high reliability and strong adaptability of signal-to-noise ratio estimation results, and can meet the needs of various complex application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 This is a flow chart of step S1 of the present invention.

[0018] Figure 3 This is a flow chart of step S2 of the present invention.

[0019] Figure 4 This is a flow chart of step S3 of the present invention.

[0020] Figure 5 This is a flow chart of step S4 of the present invention.

[0021] Figure 6This is a flow chart of signal-to-noise ratio estimation algorithm selection according to an embodiment of the present invention.

[0022] Figure 7 This is a performance comparison chart of the M2M4 and maximum likelihood estimation algorithms of an embodiment of the present invention.

[0023] Figure 8 This is a performance comparison chart of the maximum likelihood estimation and maximum likelihood estimation equalization algorithm in an embodiment of the present invention.

[0024] Fig. 9 This is a performance comparison chart of the M2M4 and M2M4 balancing algorithms of an embodiment of the present invention.

[0025] Fig.10 4 is a comparison chart of various signal-to-noise ratio estimation algorithms under high signal-to-noise ratio according to an embodiment of the present invention.

[0026] Fig.11 4 is a comparison chart of various signal-to-noise ratio estimation algorithms under low signal-to-noise ratio according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] like Figure 1 As shown, the present invention proposes a method for realizing signal-to-noise ratio estimation of a terminal in multiple scenarios, comprising the following steps:

[0028] S1. Obtain the received signal sample sequence, platform attitude data, antenna pointing data and environmental parameter data, obtain the feature vector through feature extraction and adaptive window processing; calculate the environmental compensation data based on the environmental parameter data; generate the compensation feature vector based on the feature vector and the environmental compensation data; obtain the fused feature vector through feature fusion processing based on the compensation feature vector; perform reliability evaluation and optimization based on the fused feature vector, and output the optimized feature sequence and reliability index;

[0029] S2. Perform frequency domain analysis on the optimized feature sequence and reliability index to obtain energy distribution features; construct a noise benchmark based on the energy distribution features to obtain a noise feature vector; perform cross-validation based on the noise feature vector, environmental compensation data and compensation feature vector to generate a credibility index; calculate a preliminary signal-to-noise ratio based on the energy distribution features and the noise feature vector; optimize the preliminary signal-to-noise ratio based on the credibility index, and output the optimized signal-to-noise ratio and reliability matrix;

[0030] S3, constructing a scene feature vector based on the optimized signal-to-noise ratio, the reliability matrix, and the pre-stored platform state data; optimizing parameters based on the scene feature vector to obtain an optimized parameter set; constructing a dynamic memory mechanism based on the optimized parameter set to generate memory state data; dynamically correcting the optimized signal-to-noise ratio based on the memory state data to obtain a corrected signal-to-noise ratio; performing robustness processing on the corrected signal-to-noise ratio to output a robust signal-to-noise ratio;

[0031] S4. Based on the robust signal-to-noise ratio, scene feature vector, optimized parameter set and memory state data, a performance evaluation vector is constructed; based on the performance evaluation vector, a dynamic threshold vector is generated; quality evaluation is performed on the performance evaluation vector and the dynamic threshold vector to obtain a quality evaluation value; based on the quality evaluation value, the robust signal-to-noise ratio is optimized to generate a final signal-to-noise ratio; at the same time, based on the quality evaluation value, parameter feedback control is performed to output a control parameter set.

[0032] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0033] S11, receiving a signal sample sequence collected by a receiving terminal device, including real and imaginary data; obtaining platform attitude data from an attitude sensor, including roll angle, pitch angle, and yaw angle; obtaining antenna pointing data from an antenna control unit, including azimuth and pitch angle; obtaining environmental parameter data from an environmental sensor, including temperature and air pressure values;

[0034] S12, obtaining the current platform speed value and attitude change rate, and calculating the dynamic window length based on the pre-stored basic window length parameter, combined with the platform speed value and attitude change rate; segmenting the received signal sample sequence based on the dynamic window length to obtain a signal segment sequence; extracting the time domain energy distribution, phase jump, envelope fluctuation and signal mutation characteristics of the signal segment sequence respectively, and generating a feature vector;

[0035] S13, based on the environmental parameter data, the environmental compensation coefficient is calculated by the preset environmental impact mapping parameters; at the same time, based on the platform attitude data and the antenna pointing data, the attitude compensation coefficient is calculated by the preset attitude mapping parameters; the characteristic vector is multiplied by the environmental compensation coefficient and the attitude compensation coefficient to obtain a compensation characteristic vector;

[0036] S14, reading preset feature quality assessment parameters, performing quality assessment on each feature component in the compensated feature vector, and obtaining a feature quality index; based on the feature quality index, calculating a weight coefficient of each feature component, and obtaining a feature weight vector; performing a weighted sum operation on the compensated feature vector and the feature weight vector, and obtaining a fused feature vector;

[0037] S15, read the preset stability evaluation parameters, time-varying characteristic evaluation parameters and discrete degree evaluation parameters, perform reliability evaluation on the fused feature vector, and obtain a reliability index; based on the preset optimization coefficient, optimize the fused feature vector and the reliability index, and finally output the optimized feature sequence.

[0038] In one embodiment of the present application, the basic calculation of the dynamic window length is performed: L(t) = L0 × α(v) × β(θ) × γ(ρ); where L(t) is the dynamic window length at time t; L0 is the preset basic window length; α(v) is the speed influence factor; β(θ) is the posture change influence factor; γ(ρ) is the signal feature influence factor. Calculate the speed influence factor: α(v) = 1 + k1 × tanh(v(t) / v0); where v(t) is the platform speed at time t; v0 is the reference speed threshold; k1 is the speed adjustment coefficient; tanh is the hyperbolic tangent function. Calculate the attitude change influence factor: β(θ) = exp(-λ × ||Ω(t)||2 / Ω0); where Ω(t) = [ω_roll(t), ω_pitch(t), ω_yaw(t)] is the attitude angular velocity vector; ||Ω(t)||2 is the 2-norm of the attitude angular velocity vector; Ω0 is the attitude change rate reference value; λ is the attitude influence adjustment coefficient. Calculate the signal feature influence factor: γ(ρ) = 1 + μ × (ρ(t) – ρ*) / σ_ρ; where ρ(t) is the feature point density at time t; ρ* is the historical average feature point density; σ_ρ is the standard deviation of the feature point density; μ is the density adjustment coefficient. Optimize the comprehensive window length: L_opt(t) = min{L_max, max{L_min, L(t) × η(t)}}; where L_opt(t) is the final optimized window length; L_max is the maximum allowable window length; L_min is the minimum allowable window length; η(t) is the adaptive correction factor; Calculate the adaptive correction factor: η(t) = 1 + Δ × SNR_est(t) / SNR_ref; where SNR_est(t) is the estimated signal-to-noise ratio at time t; SNR_ref is the reference signal-to-noise ratio threshold; Δ is the correction coefficient; Make a window update decision: ΔL(t) = |L_opt(t) - L_opt(t-1)| > ε × L_opt(t-1); where ΔL(t) is the window length change decision result; ε is the change threshold coefficient; L_opt(t-1) is the optimized window length at the previous moment; Calculate the feature point density: ρ(t) = N_f(t) / (L_opt(t) × Δt); where N_f(t) is the number of feature points in the window at time t; Δt is the sampling time interval.Calculate the multi-dimensional environmental compensation mechanism: E(t) = K × T(t) × P(t) × H(t); where E(t) is the environmental compensation coefficient at time t; K is the preset basic compensation coefficient; T(t) is the temperature influence factor; P(t) is the air pressure influence factor; H(t) is the humidity influence factor; calculate the temperature influence factor: T(t) = exp(-α1× |T_m(t) - T_ref| / T_std); where T_m(t) is the measured temperature value at time t; T_ref is the reference temperature value; T_std is the temperature standard deviation; α1 is the temperature adjustment coefficient; calculate the air pressure influence factor: P(t) = 1 + β1× (P_m(t) - P_ref). 2 / P_var; where P_m(t) is the measured air pressure value at time t; P_ref is the reference air pressure value; P_var is the air pressure variance; β1 is the air pressure adjustment coefficient. Calculate the coupling effect of environmental parameters: C(t) = ∑(w_i × φ_i(t)) + ∑∑(k_ij ×φ_i(t) × φ_j(t)); where φ_i(t) is the normalized value of the i-th environmental parameter; w_i is the single factor weight coefficient; k_ij is the interaction coupling coefficient; perform dynamic memory state update calculation: M(t) = α × M(t-1) + (1-α) ×S(t); where M(t) is the memory state at time t; M(t-1) is the memory state at the previous moment; S(t) is the current state vector; α is the forgetting factor; calculate the state credibility: R(t) = exp(-λ × ||S(t) - M(t)||2 / σ_m); where ||S(t) - M(t)||2 is the Euclidean distance between the state vector and the memory state; σ_m is the standard deviation of the memory state; λ is the credibility attenuation coefficient; perform reliability matrix update: Q(t) = β × Q(t-1) + (1-β) × (v(t) × v(t)T); where Q(t) is the reliability matrix at time t; v(t) is the verification feature vector; β is the smoothing coefficient; calculate performance evaluation index: P(t) =w1 × S_t + w2 × A_t + w3× C_t + w4 × U_t; where S_t is the stability index; A_t is the adaptability index; C_t is the convergence index; U_t is the consistency index; w1, w2, w3, w4 are weight coefficients; calculate quality evaluation value: Q_v(t) = ∑(λ_i × q_i(t)) × exp(-μ × σ_q(t)); where q_i(t) is the i-th mass component; λ_i is the mass component weight; σ_q(t) is the standard deviation of the mass component; μ is the fluctuation penalty factor; perform signal-to-noise ratio optimization calculation: SNR_opt(t) = SNR_est(t) × (1 + γ × Q_v(t)); where SNR_opt(t) is the optimized signal-to-noise ratio; SNR_est(t) is the estimated signal-to-noise ratio; γ is the optimization gain coefficient; Q_v(t) is the quality assessment value; build a prediction model: Y(t+1) = ∑(a_i × Y(ti)) + ∑(b_j × X(tj)) + ε(t); where Y(t) is the predicted target sequence; X(t) is the input feature sequence; a_i, b_j are model coefficients; ε(t) is the prediction error term.

[0039] In another embodiment of the present application, the process of obtaining the compensated feature vector is specifically as follows: constructing an environment compensation coefficient matrix (23 dimensions): env_comp_matrix = diag(w1, w2, ..., w23), where wi is the environment compensation weight of the i-th dimension feature. Constructing a posture compensation coefficient matrix (23 dimensions): pose_comp_matrix = diag(p1, p2, ..., p23), where pi is the posture compensation weight of the i-th dimension feature; generating a compensated feature vector: compensated_feature = feature_vector × env_comp_matrix × pose_comp_matrix; where feature_vector represents a feature vector.

[0040] This embodiment can dynamically adjust the analysis window length according to the platform posture change, avoiding the signal truncation or redundancy problem caused by the traditional fixed window length; the multi-scale coefficient decomposition method is used to extract the signal features, which can not only capture the time domain, frequency domain and phase characteristics of the signal, but also reflect the transient change characteristics of the signal through the mutation feature; by introducing the environment compensation coefficient and the posture compensation coefficient to correct the feature vector, the influence of the environment change and the platform posture change on the feature extraction is effectively eliminated; at the same time, the feature quality evaluation and weight adaptive adjustment mechanism are adopted, so that the final fused feature vector can highlight the main features of the signal and suppress the influence of secondary or unreliable features; in particular, through the hierarchical fusion and quality evaluation of features, this embodiment can adapt to the changes in signal features in different scenarios, and improve the robustness and reliability of feature extraction; in addition, through reliability evaluation and optimization processing, the temporal consistency and numerical stability of the output feature sequence are guaranteed, providing high-quality feature input for subsequent signal-to-noise ratio estimation.

[0041] According to one aspect of the present application, step S12 is further:

[0042] S121, obtaining the current platform speed value and attitude change rate, and calculating the speed influence factor based on the platform speed value; performing exponential smoothing on the attitude change rate to obtain a smoothed attitude change rate; performing a nonlinear combination operation on the speed influence factor and the smoothed attitude change rate based on a pre-stored basic window length parameter to generate an initial window length value;

[0043] S122, performing wavelet packet decomposition on the received signal sample sequence to obtain a multi-scale coefficient matrix; performing sparsity analysis on the multi-scale coefficient matrix based on a preset energy threshold to obtain a signal feature point sequence; calculating the local mutation degree based on the signal feature point sequence to generate a mutation weight coefficient; performing weighted calculation on the initial window length value and the mutation weight coefficient to obtain a dynamic window length;

[0044] S123, based on the dynamic window length, segmenting the received signal sample sequence to obtain a signal segmentation matrix; based on the signal feature point sequence, optimizing the segmentation boundaries of the signal segmentation matrix to generate an optimized segmentation matrix; performing density statistics on the signal feature point sequence in each segment to obtain a mutation feature sequence;

[0045] S124, performing Hilbert transform on each signal segment in the optimized segmented matrix to obtain an analytical signal matrix; extracting instantaneous amplitude and phase information from the analytical signal matrix to obtain an amplitude sequence and a phase sequence; performing high-order moment calculation on the amplitude sequence based on preset statistical parameters to obtain a time domain eigenvector;

[0046] S125. Perform wavelet transform on the optimized segmented matrix to extract energy distribution characteristics and obtain energy feature vector; perform unwrapping and statistical analysis on the phase sequence to obtain phase feature vector; generate mutation feature vector based on the statistical characteristics of the phase feature vector and the mutation feature sequence; combine the time domain feature vector, energy feature vector, phase feature vector and mutation feature vector into a feature vector.

[0047] In one embodiment of the present application, the wavelet packet decomposition feature extraction calculation is performed: W(j, k) = ∑(x[n] × ψ_j, k[n]); wherein W(j, k) is the wavelet packet coefficient of the j-layer k-node; x[n] is the input signal sequence; ψ_j, k[n] is the wavelet packet basis function; n is the sampling point number; the node energy is calculated: E(j, k) = ∑|W(j, k)| 2; Where E(j, k) is the energy value of the k node in the j layer; W(j, k) is the wavelet packet coefficient of the corresponding node; Calculate the energy ratio: P(j, k) = E(j, k) / E_total; Where P(j, k) is the energy proportion of the k node in the j layer; E(j, k) is the node energy value; E_total is the total energy value; Calculate the information entropy feature: H(j) = -∑(P(j, k) × log2(P(j, k))); Where H(j) is the information entropy of the j layer; P(j, k) is the node energy proportion; k is the node number of this layer; Calculate the feature vector compensation mechanism: F_c(t) = F(t) ○ (w_e × E_c(t) +w_p × P_c(t)); where F_c(t) is the compensated eigenvector; F(t) is the original eigenvector; E_c(t) is the environmental compensation coefficient; P_c(t) is the attitude compensation coefficient; w_e, w_p are weight coefficients; ○ is the Hadamard product; calculate the environmental compensation coefficient: E_c(t) = α × exp(-β × ||ΔE(t)||2) + (1-α) × E_c(t-1); where ΔE(t) is the environmental parameter change vector; α is the update coefficient; β is the sensitivity coefficient; E_c(t-1) is the environmental compensation coefficient at the previous moment; calculate the attitude compensation coefficient: P_c(t) = K_p × exp(-λ × (||θ(t)||2 + μ × ||ω(t)||2)); where θ(t) is the attitude angle vector; ω(t) is the angular velocity vector; K_p, λ, μ are adjustment coefficients; perform feature quality evaluation calculation: Q(f)= w1 × S(f) + w2 × C(f) + w3 × R(f); where Q(f) is the quality evaluation value of feature f; S(f) is the stability index; C(f) is the consistency index; R(f) is the reliability index; w1, w2, w3 are weight coefficients; calculate the stability index: S(f) = exp(-α × σ_f / μ_f); where σ_f is the standard deviation of the eigenvalue; μ_f is the mean of the eigenvalue; α is the stability adjustment coefficient; calculate the consistency index: C(f) = 1 - |ρ(f(t), f(t-1)) - ρ_ref| / ρ_ref; where ρ(f(t), f(t-1)) is the correlation coefficient between the current eigenvalue and the eigenvalue at the previous moment; ρ_ref is the reference correlation coefficient; calculate the reliability index: R(f) = exp(-β × D(f)); where D(f) is the Mahalanobis distance between the eigenvalue and the predicted value; β is the reliability adjustment coefficient; update the feature weight: W(t) = η × W(t-1) + (1-η) × Q(t); where W(t) is the feature weight vector at time t; Q(t) is the feature quality assessment vector; η is the smoothing coefficient;Calculate the correlation compensation coefficient: γ(i, j) = exp(-λ × |corr(f_i, f_j)|); where corr(f_i, f_j) is the correlation coefficient between features f_i and f_j; λ is the correlation penalty coefficient; Calculate the comprehensive feature quality: Q_total = ∑(W(i) × Q(f_i) × ∏γ(i, j)); where W(i) is the feature weight; Q(f_i) is the single feature quality; γ(i, j) is the correlation compensation coefficient. ;

[0048] In another embodiment of the present application, the feature vector is constructed and dimensioned as follows: construct a time domain feature vector (8 dimensions): time_features = [mean amplitude mean_amplitude, standard deviation of amplitude std_amplitude, skewness, kurtosis, peak factor peak_factor, pulse factor pulse_factor, margin factor margin_factor, waveform factor waveform_factor]; construct an energy feature vector (6 dimensions): energy_features = [total energy total_energy, low frequency band energy ratio band_energy_1, mid-frequency band energy ratio band_energy_2, high frequency band energy ratio band_energy_3, energy entropy energy_entropy, energy spread energy_spread]; construct a phase feature vector (5 dimensions): phase_features = [mean phase mean_phase, phase variance phase_variance, phase continuity phase_continuity, phase difference mean phase_diff_mean, phase difference variance phase_diff_var]; construct a mutation feature vector (4 dimensions): mutation_features = [mutation number mutation_count, mutation strength mutation_strength, mutation interval mutation_interval, mutation trend mutation_trend]; Combine the final feature vector (23 dimensions): feature_vector = [time_features, energy_features, phase_features, mutation_features].

[0049] This embodiment solves the adaptability problem of traditional fixed windows in complex scenarios by dynamically adjusting the window length in combination with the platform speed and attitude change rate; wavelet packet decomposition is used to realize multi-scale analysis of signals, which can not only capture the energy distribution characteristics of different frequency bands, but also identify the local characteristics and mutation characteristics of signals; the feature point extraction method based on energy threshold effectively avoids the influence of noise interference on feature extraction; through density analysis of signal feature point sequence, the time-varying characteristics of the signal are accurately reflected; in particular, the adaptive window processing and multi-scale feature extraction mechanism implemented in this embodiment enable the system to maintain stable feature extraction performance in different scenarios, providing a reliable data basis for subsequent feature fusion and signal-to-noise ratio estimation; through the dynamically adjusted processing window and multi-level feature extraction strategy, it not only ensures the accurate capture of signal features, but also improves the computational efficiency of the algorithm.

[0050] In another embodiment of the present application, the signal feature point sequence acquisition process of step S12 can also be: calculate the root mean square value of the attitude change rate based on the roll angular rate data, the pitch angular rate data and the yaw angular rate data to obtain the attitude change rate data; calculate the dynamic window length value according to the preset basic window length parameter and the attitude change rate data; use the dynamic window length value to segment the received signal sample sequence to obtain a signal segmentation sequence; perform wavelet packet decomposition on the signal segmentation sequence to obtain a multi-scale coefficient matrix; according to a preset energy threshold, perform statistics on the energy distribution of the multi-scale coefficient matrix to obtain a signal feature point sequence.

[0051] According to one aspect of the present application, step S13 is further:

[0052] S131, performing wavelet decomposition on the environmental parameter data to obtain environmental characteristic coefficients of different scales; performing LOESS smoothing on the environmental characteristic coefficients based on preset local regression parameters to obtain smoothed environmental coefficients; calculating the first-order and second-order differences of the smoothed environmental coefficients to obtain environmental change rate data; performing Fourier transform on the smoothed environmental coefficients and the environmental change rate data to obtain environmental spectrum data;

[0053] S132, based on preset neural network parameters, nonlinearly mapping the environmental spectrum data to obtain environmental mapping data; performing principal component analysis on the environmental mapping data, selecting principal components whose cumulative contribution rate exceeds a preset threshold value, and obtaining environmental principal component data; based on the environmental principal component data, calculating the Mahalanobis distance to obtain environmental anomaly data; based on the environmental anomaly data, weighting the environmental principal component data to obtain an environmental compensation coefficient;

[0054] S133, performing Kalman filtering on the platform attitude data and the antenna pointing data to obtain filtered attitude data and filtered pointing data; based on preset quaternion parameters, converting the filtered attitude data into quaternion representation to obtain attitude quaternion; calculating the angular velocity and angular acceleration of the attitude quaternion to obtain attitude dynamic data; constructing a state observation equation based on the filtered pointing data and the attitude dynamic data, and obtaining attitude prediction data through unscented Kalman filtering;

[0055] S134, based on the preset fuzzy rule parameters, fuzzy reasoning is performed on the posture prediction data to obtain a posture evaluation value; based on the posture evaluation value, a compensation weight is calculated to obtain posture weight data; and the posture prediction data and the posture weight data are weightedly combined to obtain a posture compensation coefficient;

[0056] S135. Group the feature vectors according to preset dimensions to obtain group feature data; based on the environment compensation coefficient and the posture compensation coefficient, compensate each group of the corresponding group feature data to obtain group compensation data; perform nonlinear optimization on the group compensation data to obtain optimized compensation data; reorganize the optimized compensation data to generate a compensation feature vector.

[0057] This embodiment extracts multi-scale environmental features by performing wavelet decomposition, LOESS smoothing and Fourier transform on environmental parameter data, and extracts the main features and abnormality data of the environment through neural network and principal component analysis, ensuring the accuracy of the environmental compensation coefficient, thereby improving the accuracy of signal-to-noise ratio estimation. Kalman filtering and unscented Kalman filtering are used to process platform attitude data and antenna pointing data to obtain more stable and reliable attitude prediction data; attitude compensation coefficients are generated through fuzzy reasoning and weighted combination, further enhancing the robustness of the system under different attitudes and antenna pointing. By constructing the state observation equation, the signal-to-noise ratio estimation model can be dynamically adjusted according to real-time environmental and attitude changes. This dynamic adaptability enables the system to maintain high performance in variable practical application scenarios. By grouping the feature vectors according to preset dimensions and compensating and nonlinearly optimizing the grouped feature data, the calculation efficiency is improved while ensuring accuracy; the reorganization process of optimizing the compensation data ensures that the finally generated compensation feature vector can be efficiently used for subsequent signal-to-noise ratio estimation. By using statistical methods such as cross-validation, principal component analysis and Mahalanobis distance, the environment and posture data are evaluated and weighted to ensure the reliability of the compensation coefficient and signal-to-noise ratio estimation results; through quality assessment and parameter feedback control, the accuracy and reliability of the final signal-to-noise ratio are further optimized.

[0058] In one embodiment of the present application, step S13 can also: calculate the statistical moments of each scale for the signal feature point sequence, including the mean, standard deviation, skewness and kurtosis, to obtain time domain feature data; based on the preset frequency band division parameters, calculate the energy proportion of the multi-scale coefficient matrix in each frequency band to obtain energy feature data; perform Hilbert transform on the signal segmented sequence and extract phase information to obtain phase feature data; calculate mutation feature parameters according to the distribution density of the signal feature point sequence to obtain mutation feature data; combine the time domain feature data, energy feature data, phase feature data and mutation feature data in a preset order to generate a 23-dimensional feature vector; use pre-stored environment mapping parameters to calculate a 23-dimensional environment compensation coefficient based on the environment parameter data; use pre-stored attitude mapping parameters to calculate a 23-dimensional attitude compensation coefficient based on the platform attitude data; multiply each dimensional component of the feature vector by the corresponding environment compensation coefficient and attitude compensation coefficient to obtain a compensated feature vector.

[0059] According to one aspect of the present application, step S14 is further:

[0060] S141, reading preset feature quality assessment parameters from a memory, performing singular spectrum analysis on an input compensation feature vector to obtain feature spectrum coefficients; calculating entropy values ​​of the feature spectrum coefficients to obtain spectrum entropy data; decomposing the compensation feature vector using preset wavelet packet parameters to obtain a wavelet coefficient matrix; calculating energy distribution based on the wavelet coefficient matrix to obtain energy distribution data;

[0061] S142, using preset deep autoencoder parameters, encoding and reconstructing the compensated feature vector to obtain reconstructed feature data; calculating the mean square error between the original feature and the reconstructed feature to obtain reconstruction error data; constructing a quality assessment index based on the reconstruction error data and spectral entropy data to obtain a feature quality index;

[0062] S143, performing fuzzy C-means clustering on the feature quality index to obtain cluster center data; calculating the membership of each feature to the cluster center to obtain the feature membership; calculating the feature importance based on the feature membership and energy distribution data to obtain a feature weight vector;

[0063] S144, using preset Copula function parameters, performing correlation analysis on each dimension of the compensated feature vector to obtain a feature correlation matrix; performing graph analysis based on the feature correlation matrix to obtain feature connectivity; combining the feature weight vector with the feature connectivity to generate an optimized weight vector;

[0064] S145, performing weighted combination of the compensated feature vector and the optimized weight vector to obtain an initial fusion vector; performing sparse optimization on the initial fusion vector to obtain sparse feature data; performing dimensionality reduction processing based on the sparse feature data to generate a fusion feature vector.

[0065] This embodiment can effectively evaluate the quality of features and ensure the reliability and stability of input features through singular spectrum analysis and spectral entropy calculation; using wavelet packet decomposition technology, it can refine the frequency domain information of features and calculate energy distribution data, which helps to capture local changes and important information of features; encoding and reconstructing features through deep autoencoders, calculating reconstruction error data, and further evaluating the fidelity and effectiveness of features; clustering analysis is performed on feature quality indicators to obtain cluster center data and feature membership, which can effectively distinguish and classify features and improve the accuracy of feature selection; using Copula functions for correlation analysis, combined with graph analysis to obtain feature connectivity, it can reveal the intrinsic relationship between features and optimize feature weight vectors; through sparse optimization and dimensionality reduction processing, a fused feature vector is generated to reduce redundant information and improve the representativeness and computational efficiency of features. This embodiment improves the quality and application effect of feature vectors through multi-level and multi-angle feature processing and optimization technologies, and provides a solid foundation for subsequent data analysis.

[0066] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0067] S21, reading preset wavelet transform parameters, performing wavelet transform processing on the optimized feature sequence to obtain time-frequency distribution data; performing energy integration operation on the time-frequency distribution data based on preset integral parameters to generate energy distribution features;

[0068] S22, read the preset reference noise evaluation parameters, perform local minimum value statistics on the energy distribution characteristics, and obtain the noise mean; calculate the standard deviation of the energy distribution characteristics to obtain the noise variance; and synthesize the noise mean, noise variance, and their first-order derivatives and second-order derivatives into a noise feature vector;

[0069] S23, performing a nonlinear mapping operation on the environment compensation coefficient and the attitude compensation coefficient in step S13 and the noise feature vector to obtain a verification vector; performing a weighted calculation on the verification vector based on a preset dynamic weight parameter to generate a credibility index;

[0070] S24, based on the preset weighting coefficient, weighted summing the energy distribution characteristics to obtain the total signal energy; extracting the noise level value from the noise feature vector, calculating the ratio of the total signal energy to the noise level value, and obtaining a preliminary signal-to-noise ratio through logarithmic operation; combining the credibility index, the reliability index and the verification vector into a reliability matrix;

[0071] S25, reading preset optimization weight parameters, performing weighted mapping operations on various components of the reliability matrix to obtain optimization coefficients; multiplying the preliminary signal-to-noise ratio by the optimization coefficients to generate an optimized signal-to-noise ratio.

[0072] In one embodiment of the present application, the wavelet transform time-frequency distribution is calculated: TF(a, b) = |∫x(t) ×ψ*((tb) / a)dt| 2 ; Where TF(a, b) is the time-frequency distribution value; x(t) is the input signal; ψ* is the conjugate of the wavelet basis function; a is the scale parameter; b is the translation parameter; calculate the energy integral: E(f) =∫|TF(a, b)| 2 dadb; where E(f) is the energy value of frequency band f; TF(a, b) is the time-frequency distribution; the integration range is determined by the frequency band; local minimum statistical calculation: N_min(t) ={x(i) | x(i) < x(i±k), k∈[1,K]}; where N_min(t) is the local minimum set in the window at time t; x(i) is the sample sequence; K is the search range; estimate the noise variance: σ 2 _n(t) = median{|x(i) - x(i-1)| 2} / 0.6745²; where σ 2 _n(t) is the noise variance estimate; x(i) is the sample sequence; median is the median operator; generate the verification vector: V(t) = Φ(E(t)) × Φ(P(t)); where V(t) is the verification vector; Φ( ) is the nonlinear mapping function; E(t) is the environment compensation data; P(t) is the posture compensation data; calculate the dynamic weight: W(t) = softmax(V(t) T Q(t)); where W(t) is the weight vector; V(t) is the verification vector; Q(t) is the quality assessment matrix; calculate the reliability matrix element: R_ij(t) = exp(-λ||f_i(t) - f_j(t)||2 / σ_f); where R_ij(t) is the reliability matrix element; f_i(t), f_j(t) are the eigenvectors; σ_f is the eigenstandard deviation; optimize the weight coefficient: W_opt(t) = argmin_w{||R(t)w - y||2 2+ α||w||1}; where W_opt(t) is the optimized weight coefficient; R(t) is the reliability matrix; α is the regularization parameter; calculate the environmental entropy value: H_e(t) = -∑p_i(t)log(p_i(t)); where H_e(t) is the environmental entropy value; p_i(t) is the probability distribution of the i-th environmental parameter; estimate the joint distribution of parameters: P(x, y) = ∑w_k × N(μ_k, Σ_k); where P(x, y) is the joint probability density; w_k is the mixed weight; N(μ_k, Σ_k) is the Gaussian distribution; optimize the objective function: L(θ) = ||Y - f(X;θ)||2 2 + λ1||θ||1 + λ2||θ||2 2 ; Where L(θ) is the objective function; θ is the optimization parameter; λ1, λ2 are regularization coefficients; perform gradient update: θ_t = θ_{t-1} - ηgrad L(θ_{t-1}); where θ_t is the parameter value at time t; η is the learning rate; grad L is the objective function gradient; build a state update model: S(t) = F× S(t-1) + G × u(t) + w(t); where S(t) is the state vector; F is the state transfer matrix; G is the control matrix; u(t) is the control input; w(t) is the process noise; perform historical information screening: I_s(t) = {S(i) | D(S(i), S(t))< ε, tT ≤ i < t}; where I_s(t) is the filtered historical information set; D( ) is the distance function; ε is the screening threshold; T is the time window length; build an adaptive gain matrix: K(t) = P(t)H T (HPH T + R) -1 ; Where K(t) is the gain matrix; P(t) is the prediction error covariance; H is the observation matrix; R is the observation noise covariance; Calculate the performance index: PI(t)= [S_t, A_t, C_t, U_t] T; Where PI(t) is the performance index vector; S_t, A_t, C_t, U_t are stability, adaptability, convergence, and consistency indicators respectively; Perform dynamic threshold update: Th(t) = Th(t-1) + α(Th_target - Th(t-1)) + β(PI(t) - PI(t-1)); Where Th(t) is the dynamic threshold; Th_target is the target threshold; α, β are update coefficients; Construct quality assessment function: Q(x) = tanh(w1x + b1) × sigmoid(w2x + b2); Where Q(x) is the quality assessment value; w1, w2, b1, b2 are mapping parameters; Construct performance prediction model: Y(t+k) = LSTM(X(t), X(t-1), ..., X(tn)); where Y(t+k) is the k-step prediction value; X(t) is the input sequence; LSTM is a long short-term memory network model; calculate the warning threshold: Th_w(t) = μ_Y(t) + k × σ_Y(t); where Th_w(t) is the warning threshold; μ_Y(t) is the prediction mean; σ_Y(t) is the prediction standard deviation; k is the warning coefficient.

[0073] This embodiment uses wavelet transform to process the optimized feature sequence, which can not only obtain the energy distribution characteristics of the signal at different scales, but also capture the local characteristics of the signal through time-frequency joint analysis; a noise baseline is established by performing local minimum statistics on the energy distribution characteristics, thereby avoiding the limitations of the traditional fixed threshold method; a verification mechanism based on nonlinear mapping is introduced, and cross-validation is performed in combination with environmental compensation data and attitude compensation data, thereby effectively improving the accuracy of noise estimation; a verification vector is weighted and calculated through dynamic weight parameters, so that the credibility index can accurately reflect the reliability of the current estimation result; in particular, a hierarchical weighting and optimization strategy is used to correct the preliminary signal-to-noise ratio, which not only takes into account the time-varying characteristics of the signal energy distribution, but also integrates the influence of the environment and platform status, thereby improving the adaptability of the signal-to-noise ratio estimation; in addition, a reliability matrix is ​​constructed to comprehensively evaluate the estimation results, thereby ensuring the temporal consistency and numerical stability of the output signal-to-noise ratio.

[0074] According to one aspect of the present application, step S21 is further:

[0075] S211, based on preset empirical mode decomposition parameters, perform EMD decomposition on the optimized feature sequence to obtain an intrinsic mode function; calculate the Hilbert spectrum of the intrinsic mode function to obtain time-frequency spectrum data; adaptively segment the time-frequency spectrum data to obtain segmented spectrum data;

[0076] S212, performing energy integration on the segmented spectrum data in different frequency bands to obtain frequency band energy data; calculating the coefficient of variation of the frequency band energy data to obtain the energy coefficient of variation; stratifying the energy coefficient of variation based on a preset adaptive threshold parameter to obtain energy hierarchical data;

[0077] S213, based on the intrinsic mode function, calculating the instantaneous frequency to obtain an instantaneous frequency sequence; performing time-frequency clustering analysis on the instantaneous frequency sequence to obtain frequency clustering data; combining the frequency clustering data and the energy hierarchy data to generate a time-frequency distribution matrix;

[0078] S214, based on preset wavelet basis parameters, performing wavelet transform on the time-frequency distribution matrix to obtain wavelet coefficient data; calculating the local energy of the wavelet coefficient data to obtain a local energy matrix; based on the local energy matrix, extracting energy distribution features to obtain an energy feature sequence;

[0079] S215. Perform kernel density estimation on the energy feature sequence to obtain energy density data; combine the energy density data and the time-frequency distribution matrix to obtain energy pattern data through tensor decomposition; reconstruct the energy pattern data to generate energy distribution features.

[0080] This embodiment can accurately extract the intrinsic mode function and time-frequency spectrum data of the signal through EMD decomposition and Hilbert spectrum calculation, and provide high-resolution time-frequency analysis results; adaptive segmentation and band energy integration of time-frequency spectrum data can capture energy changes in different frequency bands, calculate energy variation coefficients, and provide detailed energy level data; calculate instantaneous frequency based on the intrinsic mode function, and perform time-frequency clustering analysis to generate a time-frequency distribution matrix to reveal the clustering characteristics of the signal at different times and frequencies; use wavelet transform to process the time-frequency distribution matrix, calculate the local energy of wavelet coefficient data, extract energy distribution characteristics, and provide more detailed energy analysis; perform kernel density estimation on the energy feature sequence, combine the time-frequency distribution matrix, obtain energy mode data through tensor decomposition, reconstruct and generate energy distribution characteristics, and improve the expression ability and analysis accuracy of the features. This embodiment improves the time-frequency resolution and energy distribution characteristics of the feature sequence through multi-level and multi-angle time-frequency analysis and energy feature extraction technology.

[0081] According to one aspect of the present application, step S22 is further:

[0082] S221, reading preset reference noise evaluation parameters from a memory, performing wavelet denoising on input energy distribution characteristics to obtain denoised energy data; calculating the difference between the denoised energy data and the original data to obtain an initial noise sequence; using preset high-order statistical parameters, performing cumulative analysis on the initial noise sequence to obtain noise statistics;

[0083] S222, constructing a probability distribution model based on the noise statistic to obtain noise distribution data; performing probability density estimation using preset kernel function parameters to obtain a density estimation value; performing bootstrap sampling on the density estimation value to obtain a sampling statistic; calculating a confidence interval based on the sampling statistic to obtain noise interval data;

[0084] S223, using preset mixed Gaussian model parameters, performing cluster analysis on the noise distribution data to obtain noise cluster data; calculating the inter-class distance and intra-class distance of the noise cluster data to obtain a cluster evaluation value; performing noise component analysis based on the cluster evaluation value to obtain noise component data;

[0085] S224, performing principal component analysis on the noise component data to obtain a noise principal component; calculating a Mahalanobis distance based on the noise principal component and the noise interval data to obtain a noise anomaly; and weighting the principal component in combination with the noise anomaly to obtain weighted noise data;

[0086] S225. Combine the weighted noise data, the mean and variance, the first-order derivative and the second-order derivative of the noise distribution data to generate a noise feature vector.

[0087] This embodiment uses wavelet denoising technology to effectively remove noise from energy distribution characteristics, obtain purer denoised energy data, calculate the difference between denoised energy data and original data, generate an initial noise sequence, and lay the foundation for subsequent noise analysis; use high-order statistical parameters to perform cumulative analysis on the initial noise sequence, build a probability distribution model of noise, and obtain noise distribution data; use kernel functions to perform probability density estimation and bootstrap sampling, calculate confidence intervals, provide noise interval data, and enhance the accuracy of noise assessment; perform noise component analysis based on clustering evaluation values, identify different noise components, and provide detailed noise component data; weight the principal components in combination with the noise anomaly to obtain weighted noise data, combine the weighted noise data with the mean, variance, first-order derivative, and second-order derivative of the noise distribution data, generate a noise feature vector, and provide a comprehensive description of noise characteristics. This embodiment improves the quality and reliability of energy distribution characteristics through multi-level and multi-angle noise processing and analysis technology.

[0088] According to one aspect of the present application, step S24 is further:

[0089] S241, reading a preset energy weighting coefficient from a memory; performing local energy calculations of different scales on the input energy distribution characteristics to obtain a local energy matrix; performing weighted summation on the local energy matrix using the preset weighting coefficient to obtain a total signal energy value;

[0090] S242, extracting noise mean and variance information from the input noise feature vector; performing exponential smoothing on the noise mean to obtain a smoothed noise baseline; dividing the signal total energy value by the smoothed noise baseline and performing a logarithmic operation to obtain a preliminary signal-to-noise ratio;

[0091] S243, combining the input credibility index and the reliability index of step S1 to obtain an initial reliability vector; performing a nonlinear transformation on the verification vector to obtain a verification feature vector; calculating the weight of each dimension based on the initial reliability vector and the verification feature vector to generate a weight coefficient matrix;

[0092] S244, combining the initial reliability vector, the verification feature vector and the weight coefficient matrix into a reliability matrix; using the reliability matrix to perform weighted correction on the preliminary signal-to-noise ratio to generate a corrected preliminary signal-to-noise ratio.

[0093] In one embodiment of the present application, the process of constructing and applying the reliability matrix is: constructing a standardized reliability matrix (4×4): R = [time domain reliability index [R11, R12, R13, R14], frequency domain reliability index [R21, R22, R23, R24], phase reliability index [R31, R32, R33, R34], mutation reliability index [R41, R42, R43, R44]]; wherein the diagonal elements Rii represent the self-reliability of each type of feature; the non-diagonal elements Rij represent the cross-correlation reliability between features; calculate the reliability weighting coefficient: reliability_weights = eigenvector(R), corresponding to the eigenvector with the largest eigenvalue.

[0094] In another embodiment of the present application, step S24 can also be: using a preset weighting coefficient sequence, weighted summing the energy distribution characteristics to obtain the total energy value of the signal; performing logarithmic operation on the total energy value of the signal to obtain the logarithmic value of the total energy; extracting mean and variance information from the noise feature vector to obtain a noise reference value; subtracting the logarithmic value of the total energy from the noise reference value to obtain a preliminary signal-to-noise ratio; using a preset weighting matrix, weighted combining the credibility index and the reliability index to obtain a 4×4 dimensional initial reliability matrix; performing a nonlinear transformation on the verification vector to obtain verification feature data; calculating the eigenvalue decomposition based on the initial reliability matrix and the verification feature data to obtain a weight coefficient matrix; combining the initial reliability matrix, the verification feature data and the weight coefficient matrix into a reliability matrix.

[0095] This embodiment uses a weighted summation method to process the energy distribution characteristics, which effectively highlights the main energy distribution characteristics of the signal; establishes a noise baseline through statistical analysis of noise feature vectors, avoiding the limitations of the traditional fixed threshold method; introduces a weighted combination mechanism of credibility indicators and reliability indicators, and realizes accurate evaluation of the reliability of the estimation results; constructs a weight coefficient matrix through nonlinear transformation and eigenvalue decomposition methods, ensuring the optimization performance of the feature fusion process; in particular, the multi-level evaluation and weighted optimization mechanism established in this embodiment enables the system to accurately reflect the reliability of the current signal-to-noise ratio estimation, and improves the estimation accuracy through adaptive weight adjustment; by constructing a high-dimensional reliability matrix, not only a comprehensive evaluation of the estimation results is achieved, but also a reliable basis is provided for subsequent parameter optimization.

[0096] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0097] S31, based on the preset entropy value calculation parameters, perform entropy value calculation on the environmental parameter data to obtain the environmental entropy value; obtain the platform state data from the platform sensor, including speed data, acceleration data and attitude change rate data; combine the platform state data and the environmental entropy value into a scene feature vector;

[0098] S32, optimizing objective function parameters, optimizing signal-to-noise ratio, reliability matrix and scene feature vector based on preset parameters, and calculating objective function value; optimizing objective function value based on preset gradient update parameters, and obtaining an optimized parameter set, including signal processing, feature extraction and correction parameters;

[0099] S33, combining the optimized signal-to-noise ratio, the reliability matrix, the scene feature vector and the optimized parameter set to form a current state vector; based on a preset memory update parameter, performing a selective update calculation on the current state vector and pre-stored historical state data to generate memory state data;

[0100] S34, calculating the adaptive gain value based on the preset correction gain parameter and the memory state data; calculating the observation matrix based on the reliability matrix; performing correction calculation on the optimized signal-to-noise ratio, the adaptive gain value and the observation matrix to obtain the corrected signal-to-noise ratio;

[0101] S35. Based on the preset anomaly detection parameters, the mean and standard deviation of the corrected signal-to-noise ratio are calculated to obtain a calculation result; based on the calculation result, outlier detection is performed to obtain an anomaly index; and the corrected signal-to-noise ratio and the anomaly index are weighted to generate a robust signal-to-noise ratio.

[0102] This implementation achieves accurate modeling of complex scenes by constructing scene feature vectors and comprehensively considering the platform motion state, environmental parameter changes and historical state information; optimizes the objective function so that the optimization parameter set can be adaptively adjusted to adapt to scene changes; introduces a dynamic memory mechanism to retain valid historical information through a selective update strategy, thereby avoiding the error divergence problem that may be caused by the accumulation of historical data; calculates the adaptive gain value based on the memory state data, and constructs the observation matrix in combination with the reliability matrix, so that the correction process can fully utilize the statistical characteristics of historical data; performs robust processing on the correction results through the anomaly detection mechanism, effectively suppressing the influence of sudden interference or outliers; in particular, the dynamic memory and adaptive correction mechanism established in this embodiment enables the system to maintain the stability and reliability of the estimation results while maintaining rapid response capabilities.

[0103] According to one aspect of the present application, step S31 is further:

[0104] S311, acquiring speed data from the platform sensor, performing multi-point sampling, and obtaining a speed sampling sequence; performing Kalman filtering on the speed sampling sequence to obtain a filtered speed value; calculating the standard deviation of the speed sampling sequence to obtain a speed fluctuation value; combining the filtered speed value and the speed fluctuation value into a speed feature vector;

[0105] S312, acquiring acceleration data from the platform sensor, sampling, and obtaining an acceleration sampling sequence; performing median filtering on the acceleration sampling sequence to obtain a filtered acceleration value; performing mutation detection on the acceleration sampling sequence based on a preset threshold parameter to obtain an acceleration mutation mark; combining the filtered acceleration value and the acceleration mutation mark into an acceleration feature vector;

[0106] S313, acquiring attitude change rate data from the platform sensor; performing exponential smoothing on the attitude change rate data to obtain a smooth attitude change rate; calculating the first-order difference of the attitude change rate data to obtain an attitude acceleration value; performing motion pattern analysis based on the smooth attitude change rate and the attitude acceleration value to generate an attitude feature vector;

[0107] S314, based on the preset entropy value calculation parameters, perform entropy value calculation on the environmental parameter data to obtain the environmental entropy value; perform time series correlation analysis on the environmental parameter data to obtain environmental change characteristics; combine the environmental entropy value and the environmental change characteristics into an environmental feature vector;

[0108] S315, performing feature fusion on the velocity feature vector, the acceleration feature vector, the posture feature vector and the environment feature vector to generate a scene feature vector.

[0109] In one embodiment of the present application, environmental parameters such as temperature and air pressure are obtained from environmental sensors, and environmental features are extracted based on information entropy theory. The specific steps are as follows: normalize and preprocess the obtained environmental parameters: normalize the temperature parameter T: T_norm = (T - T_min) / (T_max - T_min); normalize the air pressure parameter P: P_norm = (P - P_min) / (P_max - P_min); where T_min, T_max, P_min, P_max are pre-calibrated parameter ranges. Construct a joint probability distribution of multidimensional environmental parameters, and use the kernel density estimation (KDE) method to estimate the joint probability density function of temperature and pressure: P(T, P) = (1 / nh 2 )∑K((T-Ti) / h, (P-Pi) / h), where K is a two-dimensional Gaussian kernel function and h is a bandwidth parameter, which is adaptively determined by the Silverman criterion. Calculate the Shannon entropy and conditional entropy: Shannon entropy: H(T, P) = -∑∑P(T, P)log(P(T, P)); conditional entropy: H(T|P) = H(T, P) - H(P); H(P|T) = H(T, P) - H(T). Calculate the mutual information of environmental parameters: I(T;P) = H(T) + H(P) - H(T,P); construct the environmental entropy value feature vector: ENV_entropy = [H(T,P), H(T|P), H(P|T), I(T;P)]; calculate the time-varying characteristics of environmental parameters based on the sliding time window: use the exponentially weighted moving average (EWMA) method to calculate the time-varying trend of each entropy value, and the window length W is adaptively adjusted according to the rate of environmental change; generate the final environmental feature vector: environmental feature vector = [ENV_entropy, EWMA(ENV_entropy), dENV_entropy / dt], where dENV_entropy / dt is the time derivative of the entropy value, which characterizes the rate of environmental change.

[0110] This embodiment processes the velocity and acceleration data through Kalman filtering and median filtering, which effectively suppresses the random fluctuation of sensor data; accurately represents the platform motion state based on standard deviation calculation and mutation detection; analyzes the posture change rate data through smoothing and differential operations, and accurately captures the dynamic characteristics of the platform posture; uses reliability weight vectors for feature fusion to ensure the representativeness of the scene feature vector; in particular, the multi-dimensional motion feature extraction and weighted fusion mechanism established in this embodiment enables the system to accurately describe the dynamic characteristics of the current scene, and provides reliable scene information for subsequent parameter optimization; through multi-level data processing and feature fusion strategies, both the accuracy of feature extraction is guaranteed and a high processing efficiency is maintained.

[0111] According to one aspect of the present application, the reliability matrix application method is as follows: read the reliability matrix R, extract the reliability weight vector reliability_weights, and obtain the weight coefficient for weighting the scene feature vector. Perform feature fusion and reliability weighting: basic scene feature vector = [speed feature vector, acceleration feature vector, posture feature vector, environment feature vector]; reliability weighted scene feature vector = basic scene feature vector × diag(reliability_weights).

[0112] In one embodiment of the present application, step S31 can also be: obtaining a reliability weight vector according to the reliability matrix; obtaining speed sampling data from the platform sensor and performing Kalman filtering to obtain a filtered speed value; calculating the standard deviation of the speed sampling data to obtain a speed fluctuation value; combining the filtered speed value and the speed fluctuation value into a speed feature vector; obtaining acceleration sampling data from the sensor and performing median filtering to obtain a filtered acceleration value; detecting the acceleration sampling data according to a preset mutation detection threshold to obtain an acceleration mutation mark; combining the filtered acceleration value and the acceleration mutation mark into an acceleration feature vector; smoothing and differentially operating the posture change rate data to obtain smoothed posture data and posture change data, respectively; combining the smoothed posture data and posture change data into a posture feature vector; weightedly combining the speed feature vector, the acceleration feature vector, the posture feature vector and the reliability weight vector to obtain a scene feature vector.

[0113] In another embodiment of the present application, step S314 may also be: acquiring temperature data and air pressure data from an environmental sensor, subtracting a preset minimum temperature value from the temperature data, and dividing the result by a preset temperature variation range to obtain normalized temperature data; subtracting a preset minimum air pressure value from the air pressure data, and dividing the result by a preset air pressure variation range to obtain normalized air pressure data; using a preset bandwidth parameter, performing Gaussian kernel density estimation on the normalized temperature data and the normalized air pressure data to obtain joint probability density data; performing logarithmic operations on the joint probability density data and summing the data to obtain a basic entropy value. data; solving the edge probability density according to the joint probability density data and calculating the logarithm to obtain conditional entropy data; calculating mutual information based on the basic entropy data and the conditional entropy data to obtain mutual information data; combining the basic entropy data, the conditional entropy data and the mutual information data into an entropy feature vector; using a preset exponential weighting coefficient, performing an exponential weighted average operation on the entropy feature vector to obtain a time-varying entropy vector; calculating the time derivative of the entropy feature vector to obtain an entropy change vector; combining the entropy feature vector, the time-varying entropy vector and the entropy change vector into an environment feature vector.

[0114] According to one aspect of the present application, step S34 is further:

[0115] S341, read the preset correction gain parameter from the memory; perform time series decomposition on the input memory state data to obtain a historical trend sequence and a fluctuation sequence; perform autoregressive processing on the historical trend sequence using preset recursive parameters to obtain a trend prediction value; perform periodic analysis on the fluctuation sequence to generate a fluctuation feature vector;

[0116] S342, calculating the adaptive gain value according to the memory state data and the preset correction gain parameter; extracting the reliability index of each dimension from the input reliability matrix; performing principal component analysis on the reliability index of each dimension to obtain the main influencing factors; constructing the covariance matrix based on the main influencing factors to generate the observation noise matrix;

[0117] S343, performing weighted combination of the adaptive gain value and the fluctuation characteristic vector to obtain a basic gain matrix; using the observation noise matrix to correct the basic gain matrix to generate an adaptive gain matrix; performing weighted combination of the main influencing factors to generate a comprehensive correction coefficient;

[0118] S344. Calculate the deviation between the input optimized signal-to-noise ratio and the trend prediction value to obtain an estimated deviation value; use an adaptive gain matrix to correct the estimated deviation value to obtain a correction increment value; multiply the correction increment value by the comprehensive correction coefficient to obtain a final correction amount; perform algebraic operations on the optimized signal-to-noise ratio and the final correction amount to generate a corrected signal-to-noise ratio.

[0119] This embodiment achieves accurate grasp of signal change trends by performing time series decomposition and trend prediction on memory state data; uses principal component analysis to process the reliability matrix and effectively extracts the main influencing factors; based on the correction mechanism of adaptive gain value and observation noise matrix, it ensures the accuracy and stability of the correction process; through multi-level gain adjustment and correction calculation, it achieves accurate correction of the signal-to-noise ratio; in particular, the adaptive correction mechanism established in this embodiment not only takes into account the statistical characteristics of historical data, but also integrates the reliability information of current observations, thereby improving the accuracy of the correction results; by establishing a complete memory update and correction optimization mechanism, it effectively balances the system's rapid response capability and estimation stability.

[0120] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0121] S41, calculating the first-order derivative and the second-order derivative of the robust signal-to-noise ratio to obtain a stability index; calculating the scene change rate based on the scene feature vector to obtain an adaptability index; calculating the parameter change rate based on the optimized parameter set to obtain a convergence index; calculating the similarity between the previous and next states based on the memory state data to obtain a consistency index; combining the stability index, adaptability index, convergence index and consistency index into a performance evaluation vector;

[0122] S42, calculating the influence factor based on the scene feature vector and the reliability matrix; performing weighted calculation on the preset basic threshold parameter and the influence factor to obtain a dynamic threshold vector; and adaptively adjusting the dynamic threshold vector based on the change rate of the performance evaluation vector to obtain an adjusted dynamic threshold vector;

[0123] S43, based on the preset quality evaluation parameters, performing nonlinear mapping operation on the corresponding components of the performance evaluation vector and the adjusted dynamic threshold vector to obtain the evaluation value of each component; based on the preset weight parameters, performing weighted summation on the evaluation value of each component to obtain the quality evaluation value; performing weighted calculation on the robust signal-to-noise ratio and the quality evaluation value to generate a final signal-to-noise ratio;

[0124] S44, constructing a prediction model based on the performance evaluation vector, the scene feature vector and the memory state data; performing prediction calculation based on the prediction model to obtain a prediction performance vector; comparing the prediction performance vector with the adjusted dynamic threshold vector to generate a warning signal;

[0125] S45. Based on the early warning signal, calculate the deviation between the quality assessment value and the preset target quality value to obtain the assessment deviation; based on the assessment deviation and its integral value and derivative value, calculate the control quantity to obtain the control signal; based on the control signal, update the optimization parameter set to generate the control parameter set.

[0126] In one embodiment of the present application, a multi-dimensional performance evaluation index is calculated: P(t) = [P_s(t), P_a(t), P_r(t), P_c(t)]; wherein P(t) is a comprehensive performance vector; P_s(t) is a stability index; P_a(t) is an accuracy index; P_r(t) is a reliability index; and P_c(t) is a consistency index. The stability index is calculated as follows: P_s(t) = exp(-α× σ_x(t) / μ_x(t)) × (1 - β × |dx(t) / dt| / v_ref); wherein σ_x(t) is the standard deviation of the state quantity; μ_x(t) is the mean of the state quantity; dx(t) / dt is the rate of change; v_ref is the reference rate of change; α and β are weight coefficients; the accuracy index is calculated as follows: P_a(t) = 1 / (1 + γ × ||x(t) - x_ref(t)||2); where x(t) is the current state; x_ref(t) is the reference state; γ is the scale factor; calculate the reliability index: P_r(t) = exp(-λ1 × D_m(t)) × (1– λ2× H(t)); where D_m(t) is the Mahalanobis distance; H(t) is the information entropy; λ1, λ2 are adjustment coefficients; calculate the consistency index: P_c(t) = ρ(x(t), x(t-τ)) × exp(-μ × |σ(t) - σ(t-τ)|); where ρ( ) is the correlation coefficient; τ is the time delay; σ(t) is the fluctuation measure; μ is the penalty factor; perform index fusion optimization: W_opt(t) = argmin_w{||w T × P(t) - y_target||2 2 + α × ||w||1 + β × w T × Σ × w}; where W_opt(t) is the optimization weight vector; P(t) is the performance indicator vector; Σ is the indicator covariance matrix; α, β are regularization parameters; adaptive weight update: W(t) = η × W(t-1) + (1-η) × W_opt(t) + ξ(t); where W(t) is the final weight vector; η is the smoothing factor; ξ(t) is the adaptive adjustment item; reliability matrix update calculation: R(t) = λ × R(t-1)+ (1-λ) × (v(t) × v(t) T + Δ(t)); where R(t) is the reliability matrix; v(t) is the eigenvector; Δ(t) is the perturbation matrix; λ is the forgetting factor; the generated perturbation matrix is: Δ(t) = U(t) × S(t) × U(t) T; Where U(t) is an orthogonal matrix; S(t) is a diagonal scaling matrix; eigenvalue decomposition is applied: R(t) = Q(t) × Λ(t) × Q(t) T ; where Q(t) is the eigenvector matrix; Λ(t) is the eigenvalue diagonal matrix; construct the eigenvalue screening criterion: λ_k(t) = {λ_i(t) | λ_i(t) > ε × tr(Λ(t)) / n}; where λ_k(t) is the set of retained eigenvalues; tr( ) is the matrix trace; n is the matrix dimension; ε is the screening coefficient; perform matrix reconstruction: R_opt(t) = Q_k(t) × Λ_k(t) × Q_k(t) T ; R_opt(t) is the reconstructed reliability matrix; Q_k(t) is the retained eigenvector; Λ_k(t) is the retained eigenvalue; construct stability constraint: ||R_opt(t) - R_opt(t-1)||_F ≤ Δ × ||R_opt(t-1)||_F; || ||_F is the Frobenius norm; Δ is the stability threshold; calculate the comprehensive score: Score(t) = tr(R_opt(t)) × exp(-μ × cond(R_opt(t))); cond( ) is the matrix condition number; μ is the balance coefficient; construct an adaptive adjustment mechanism: α(t) = α0× exp(-β × ||grad Score(t)||2); α(t) is the adaptive adjustment coefficient; α0 is the initial coefficient; grad Score(t) is the score gradient.

[0127] This embodiment constructs a performance evaluation vector to comprehensively evaluate the system performance from four dimensions: stability, adaptability, convergence and consistency. The impact factor is calculated based on the scenario feature vector and the reliability matrix to achieve dynamic adjustment of the threshold parameters and avoid the limitations brought by the fixed threshold. Nonlinear mapping and weighted combination methods are used for quality evaluation so that the evaluation results can accurately reflect the reliability of the current estimate. Performance indicators are predicted by establishing a prediction model, and an early warning mechanism for system performance is implemented in combination with dynamic thresholds. Parameter feedback control is performed based on the quality evaluation results to form a closed-loop optimization mechanism, which not only improves the system's adaptability, but also ensures the long-term stability of the estimation results. In particular, the multi-dimensional evaluation and closed-loop control mechanism implemented in this embodiment enables the system to promptly detect and correct performance degradation and maintain the optimal working state.

[0128] According to one aspect of the present application, step S43 is further:

[0129] S431, reading preset quality evaluation parameters from a memory; normalizing the input performance evaluation vector to obtain a normalized performance vector; normalizing the input dynamic threshold vector to obtain a standardized threshold vector; calculating the deviations of each dimension between the normalized performance vector and the standardized threshold vector to generate a performance deviation vector;

[0130] S432, performing component projection on the performance deviation vector to obtain an orthogonal deviation matrix; performing eigendecomposition on the orthogonal deviation matrix using a preset weight parameter to obtain an eigenvalue sequence; calculating the contribution of each dimension based on the eigenvalue sequence to generate a weight distribution vector; performing weighted combination of the weight distribution vector and the performance deviation vector to obtain a weighted deviation value;

[0131] S433, performing time series analysis on the input robust signal-to-noise ratio to obtain a signal-to-noise ratio trend sequence; calculating local statistical features of the signal-to-noise ratio trend sequence to generate a statistical feature vector; using the statistical feature vector to correct the weighted deviation value to obtain a corrected deviation value; calculating an evaluation score based on the corrected deviation value to generate a quality evaluation value;

[0132] S434, read the preset quality gain parameter from the memory; calculate the quality gain coefficient according to the quality evaluation value; perform nonlinear mapping on the statistical feature vector to obtain the feature mapping vector; combine the quality gain coefficient with the feature mapping vector to obtain the quality correction coefficient; perform weighted calculation on the input robust signal-to-noise ratio and the quality correction coefficient to generate the final signal-to-noise ratio.

[0133] This embodiment realizes standardized evaluation of system performance by normalizing and orthogonalizing performance evaluation vectors and dynamic threshold vectors; accurately reflects the importance of each performance indicator based on eigenvalue decomposition and weight distribution calculation; accurately describes the signal-to-noise ratio variation characteristics through time series analysis and statistical feature extraction; optimizes processing using quality gain parameters and feature mapping mechanism to ensure the accuracy of the final output; in particular, the multi-dimensional evaluation and optimization mechanism established in this embodiment enables the system to accurately evaluate the current working status and maintain optimal performance through adaptive parameter adjustment; and effectively improves the reliability and stability of the system by establishing a complete quality evaluation and optimization process.

[0134] like Figure 6 As shown, according to one aspect of the present application, a method for implementing signal-to-noise ratio estimation of a terminal in multiple scenarios collects the current information of the terminal, makes a judgment based on the current state of the terminal, and selects a signal-to-noise ratio estimation method suitable for the current state. Specifically, the demodulator makes a comprehensive judgment by receiving the state information sent by the host computer and the current estimated signal-to-noise ratio, and then selects the current optimal signal-to-noise ratio estimation method; if the conditions change, the signal-to-noise ratio estimation method also changes accordingly to achieve the optimal estimation in various scenarios.

[0135] According to the advantages and disadvantages of several SNR estimation algorithms, the SNR estimation algorithm is selected according to the state information of the current scene and the SNR in different scenarios. When the SNR is relatively high, the estimation performance of the M2M4 algorithm is better than the maximum likelihood estimation. The performance comparison results are as follows: Figure 7 Therefore, in high signal-to-noise ratio scenarios, the M2M4 estimation algorithm should be selected, and in low signal-to-noise ratio scenarios, the maximum likelihood estimation algorithm should be selected.

[0136] When the antenna pointing does not change, and the terminal only receives signals from one main station, the noise level basically does not change. Therefore, when the terminal is stationary or in uniform linear motion, the M2M4 algorithm or the maximum likelihood estimation algorithm can be optimized. The improved algorithms are the M2M4 equalization algorithm and the maximum likelihood estimation equalization algorithm. The core idea of ​​the equalization algorithm is to smooth the noise (the prerequisite is that the mean and variance of the noise are fixed at this time), and to equalize the noise level of N frame signals to make the noise level smoother. After equalization, the estimated signal-to-noise ratio is also more stable. Figure 8 and Fig. 9 As shown. However, the equalization algorithm is a statistic of noise within a certain period of time. If the platform where the terminal is located has a large maneuver or other operation, the antenna direction changes greatly. The equalization algorithm needs a long time to recover to the current correct noise level. This is a slow-changing process. Therefore, the equalization algorithm can only be applied to static or uniform linear motion.

[0137] In one embodiment of the present application, when the device is powered on, the terminal first captures the GPS signal to determine the current longitude and latitude, and obtains the beam information based on the longitude and latitude. In the satellite scanning stage, the demodulator uses the maximum likelihood estimation algorithm to scan the satellite beam direction. At this time, the estimated signal-to-noise ratio is an independent signal-to-noise ratio for each frame of data. The demodulator feeds the estimated signal-to-noise ratio back to the antenna, and the antenna searches for the maximum value of the signal-to-noise ratio (i.e., the satellite beam direction) during the scan based on the fed-back signal-to-noise ratio.

[0138] When the antenna locks onto the satellite, the antenna informs the demodulator that it is currently in the satellite tracking state and sends azimuth and off-axis angle information to the demodulator at regular intervals. The demodulator counts the changes in off-axis angle and azimuth within a time slice. When the change exceeds the set threshold, it determines that the current terminal is in a non-stationary and non-uniform linear motion state. At this time, the antenna needs the signal-to-noise ratio information fed back by the demodulator in real time for tracking. During the tracking process, large maneuvers may occur, causing the pointing direction to shift and causing the noise level to change. Therefore, in this state, the maximum likelihood estimation algorithm is required to independently estimate the signal-to-noise ratio of each frame of data.

[0139] When the antenna is stationary or in uniform linear motion, the antenna pointing direction does not change significantly. Under a fixed pointing direction, the noise level will basically not change, so the demodulator counts the signal-to-noise ratio information in a time slice. If the signal-to-noise ratio is greater than the set threshold, it is considered to be in a high signal-to-noise ratio state, and the demodulator uses the M2M4 equalization algorithm for signal-to-noise ratio; if the signal-to-noise ratio is less than the set threshold, it is considered to be in a low signal-to-noise ratio state, and the demodulator uses the maximum likelihood estimation equalization algorithm for signal-to-noise ratio estimation.

[0140] This embodiment improves the signal-to-noise ratio estimation method for vehicle-mounted and airborne satellite communication systems, and can effectively and accurately estimate the current signal-to-noise ratio in various scenarios. The estimation performance is as follows: Fig.10 and Fig.11 As shown in the figure, while improving the system performance, the reliability and stability of the system are guaranteed, which has great engineering application value.

[0141] According to one aspect of the present application, a device for implementing multi-scenario terminal signal-to-noise ratio estimation includes:

[0142] at least one processor; and,

[0143] a memory communicatively connected to at least one of the processors; wherein,

[0144] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the multi-scenario terminal signal-to-noise ratio estimation method described in any of the above embodiments.

[0145] The present invention has established a complete set of multi-scenario terminal signal-to-noise ratio estimation methods. Through the organic combination of feature extraction, noise modeling, dynamic correction and performance evaluation, it has achieved accurate estimation of the signal-to-noise ratio in complex scenarios; the adaptive window processing and multi-scale feature extraction technology are adopted to ensure the complete capture of the original signal features; the interference of external factors is effectively eliminated through the environmental compensation and posture compensation mechanisms; the dynamic memory mechanism and adaptive correction strategy are introduced to improve the reliability and stability of the estimation results; a multi-dimensional performance evaluation and closed-loop control mechanism is established to achieve continuous optimization of system performance. The present invention can adapt to the changes in signal characteristics in different scenarios. Through dynamic feature extraction, adaptive parameter adjustment and multi-level optimization strategies, it maintains a high computing efficiency while ensuring the estimation accuracy; through layered data processing and optimization mechanisms, the high precision, high reliability and strong adaptability of the signal-to-noise ratio estimation results are achieved, which can meet the needs of a variety of complex application scenarios.

[0146] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for realizing signal-to-noise ratio estimation of a terminal in multiple scenarios, characterized in that: The steps include: S1. Obtain the received signal sample sequence, platform attitude data, antenna pointing data and environmental parameter data, and obtain a feature vector through feature extraction and adaptive window processing; calculate and obtain environmental compensation data based on the environmental parameter data; generate a compensation feature vector based on the feature vector and the environmental compensation data; obtain a fused feature vector based on the compensated feature vector through feature fusion processing; Fuse feature vectors, conduct reliability assessment and optimization, and output optimized feature sequences and reliability indicators; S2. Perform frequency domain analysis on the optimized feature sequence and reliability index to obtain energy distribution characteristics; based on the energy distribution characteristics, construct a noise benchmark to obtain a noise feature vector; Based on the noise feature vector, environmental compensation data and compensation feature vector, cross-validation is performed to generate a credibility index; based on the energy distribution characteristics and noise feature vector, a preliminary signal-to-noise ratio is calculated; based on the credibility index, the preliminary signal-to-noise ratio is optimized, and the optimized signal-to-noise ratio and reliability matrix are output; S3, constructing a scene feature vector based on the optimized signal-to-noise ratio, the reliability matrix, and the pre-stored platform status data; Based on the scene feature vector, parameter optimization is performed to obtain an optimized parameter set; based on the optimized parameter set, a dynamic memory mechanism is constructed to generate memory state data; Based on the memory state data, the optimized signal-to-noise ratio is dynamically corrected to obtain the corrected signal-to-noise ratio; the corrected signal-to-noise ratio is robustly processed to output the robust signal-to-noise ratio; S4. Based on the robust signal-to-noise ratio, the scene feature vector, the optimized parameter set and the memory state data, a performance evaluation vector is constructed; based on the performance evaluation vector, a dynamic threshold vector is generated; and quality evaluation is performed on the performance evaluation vector and the dynamic threshold vector to obtain a quality evaluation value; Based on the quality evaluation value, the robust signal-to-noise ratio is optimized to generate the final signal-to-noise ratio; at the same time, based on the quality evaluation value, parameter feedback control is performed to output a control parameter set.

2. The method for implementing multi-scenario terminal signal-to-noise ratio estimation according to claim 1, characterized in that: Step S1 is further as follows: S11, receiving a signal sample sequence collected by a receiving terminal device, including real and imaginary data; obtaining platform attitude data from an attitude sensor, including roll angle, pitch angle, and yaw angle; obtaining antenna pointing data from an antenna control unit, including azimuth and pitch angle; obtaining environmental parameter data from an environmental sensor, including temperature and air pressure values; S12, obtaining the current platform speed value and attitude change rate, and calculating the dynamic window length based on the pre-stored basic window length parameter, combined with the platform speed value and attitude change rate; segmenting the received signal sample sequence based on the dynamic window length to obtain a signal segment sequence; extracting the time domain energy distribution, phase jump, envelope fluctuation and signal mutation characteristics of the signal segment sequence respectively, and generating a feature vector; S13, based on the environmental parameter data, the environmental compensation coefficient is calculated by the preset environmental impact mapping parameters; at the same time, based on the platform attitude data and the antenna pointing data, the attitude compensation coefficient is calculated by the preset attitude mapping parameters; the characteristic vector is multiplied by the environmental compensation coefficient and the attitude compensation coefficient to obtain a compensation characteristic vector; S14, reading preset feature quality assessment parameters, performing quality assessment on each feature component in the compensated feature vector, and obtaining a feature quality index; Based on the feature quality index, the weight coefficient of each feature component is calculated to obtain a feature weight vector; the compensation feature vector and the feature weight vector are weighted summed to obtain a fused feature vector; S15, read the preset stability evaluation parameters, time-varying characteristic evaluation parameters and discrete degree evaluation parameters, perform reliability evaluation on the fused feature vector, and obtain a reliability index; based on the preset optimization coefficient, optimize the fused feature vector and the reliability index, and finally output the optimized feature sequence.

3. The method for implementing multi-scenario terminal signal-to-noise ratio estimation according to claim 2, characterized in that: Step S2 is further as follows: S21, reading preset wavelet transform parameters, performing wavelet transform processing on the optimized feature sequence to obtain time-frequency distribution data; performing energy integration operation on the time-frequency distribution data based on preset integral parameters to generate energy distribution features; S22, read the preset reference noise evaluation parameters, perform local minimum value statistics on the energy distribution characteristics, and obtain the noise mean; calculate the standard deviation of the energy distribution characteristics to obtain the noise variance; and synthesize the noise mean, noise variance, and their first-order derivatives and second-order derivatives into a noise feature vector; S23, performing a nonlinear mapping operation on the environment compensation coefficient and the attitude compensation coefficient in step S13 and the noise feature vector to obtain a verification vector; performing a weighted calculation on the verification vector based on a preset dynamic weight parameter to generate a credibility index; S24, based on the preset weighting coefficient, weighted summing the energy distribution characteristics to obtain the total signal energy; extracting the noise level value from the noise feature vector, calculating the ratio of the total signal energy to the noise level value, and obtaining a preliminary signal-to-noise ratio through logarithmic operation; combining the credibility index, the reliability index and the verification vector into a reliability matrix; S25, reading the preset optimization weight parameter, performing weighted mapping operation on each component of the reliability matrix to obtain the optimization coefficient; multiplying the preliminary signal-to-noise ratio by the optimization coefficient to generate the optimized signal-to-noise ratio.

4. The method for implementing multi-scenario terminal signal-to-noise ratio estimation according to claim 3, characterized in that: Step S3 is further as follows: S31, based on the preset entropy value calculation parameters, perform entropy value calculation on the environmental parameter data to obtain the environmental entropy value; obtain the platform state data from the platform sensor, including speed data, acceleration data and attitude change rate data; Combine the platform state data and the environmental entropy value into a scene feature vector; S32, optimizing objective function parameters, optimizing signal-to-noise ratio, reliability matrix and scene feature vector based on preset parameters, and calculating objective function value; optimizing objective function value based on preset gradient update parameters, and obtaining an optimized parameter set, including signal processing, feature extraction and correction parameters; S33, combining the optimized signal-to-noise ratio, the reliability matrix, the scene feature vector and the optimized parameter set to form a current state vector; based on a preset memory update parameter, performing a selective update calculation on the current state vector and pre-stored historical state data to generate memory state data; S34, calculating the adaptive gain value based on the preset correction gain parameter and the memory state data; calculating the observation matrix based on the reliability matrix; performing correction calculation on the optimized signal-to-noise ratio, the adaptive gain value and the observation matrix to obtain the corrected signal-to-noise ratio; S35. Based on the preset anomaly detection parameters, the mean and standard deviation of the corrected signal-to-noise ratio are calculated to obtain a calculation result; based on the calculation result, an outlier detection is performed to obtain an anomaly index; and the corrected signal-to-noise ratio and the anomaly index are weighted to generate a robust signal-to-noise ratio.

5. The method for realizing multi-scenario terminal signal-to-noise ratio estimation according to claim 4, characterized in that: Step S4 is further as follows: S41, calculating the first-order derivative and the second-order derivative of the robust signal-to-noise ratio to obtain a stability index; Based on the scene feature vector, the scene change rate is calculated to obtain the adaptability index; based on the optimized parameter set, the parameter change rate is calculated to obtain the convergence index; Based on the memory state data, the similarity between the previous and next states is calculated to obtain the consistency index; the stability index, adaptability index, convergence index and consistency index are combined into a performance evaluation vector; S42, calculating the influence factor based on the scene feature vector and the reliability matrix; performing weighted calculation on the preset basic threshold parameter and the influence factor to obtain a dynamic threshold vector; and adaptively adjusting the dynamic threshold vector based on the change rate of the performance evaluation vector to obtain an adjusted dynamic threshold vector; S43, based on the preset quality evaluation parameters, performing nonlinear mapping operation on the corresponding components of the performance evaluation vector and the adjusted dynamic threshold vector to obtain the evaluation value of each component; based on the preset weight parameters, performing weighted summation on the evaluation value of each component to obtain the quality evaluation value; performing weighted calculation on the robust signal-to-noise ratio and the quality evaluation value to generate a final signal-to-noise ratio; S44, constructing a prediction model based on the performance evaluation vector, the scene feature vector and the memory state data; performing prediction calculation based on the prediction model to obtain a prediction performance vector; comparing the prediction performance vector with the adjusted dynamic threshold vector to generate a warning signal; S45. Based on the early warning signal, calculate the deviation between the quality assessment value and the preset target quality value to obtain an assessment deviation; Based on the evaluation deviation and its integral value and derivative value, the control quantity is calculated to obtain the control signal; Based on the control signal, the optimization parameter set is updated to generate a control parameter set.

6. The method for realizing multi-scenario terminal signal-to-noise ratio estimation according to claim 5, characterized in that: Step S12 is further as follows: S121, obtaining the current platform speed value and attitude change rate, and calculating the speed influence factor based on the platform speed value; performing exponential smoothing on the attitude change rate to obtain a smoothed attitude change rate; performing a nonlinear combination operation on the speed influence factor and the smoothed attitude change rate based on a pre-stored basic window length parameter to generate an initial window length value; S122, performing wavelet packet decomposition on the received signal sample sequence to obtain a multi-scale coefficient matrix; performing sparsity analysis on the multi-scale coefficient matrix based on a preset energy threshold to obtain a signal feature point sequence; calculating the local mutation degree based on the signal feature point sequence to generate a mutation weight coefficient; performing weighted calculation on the initial window length value and the mutation weight coefficient to obtain a dynamic window length; S123, based on the dynamic window length, segmenting the received signal sample sequence to obtain a signal segmentation matrix; based on the signal feature point sequence, optimizing the segmentation boundaries of the signal segmentation matrix to generate an optimized segmentation matrix; performing density statistics on the signal feature point sequence in each segment to obtain a mutation feature sequence; S124, performing Hilbert transform on each signal segment in the optimized segmented matrix to obtain an analytical signal matrix; extracting instantaneous amplitude and phase information from the analytical signal matrix to obtain an amplitude sequence and a phase sequence; performing high-order moment calculation on the amplitude sequence based on preset statistical parameters to obtain a time domain eigenvector; S125, performing wavelet transform on the optimized segmented matrix, extracting energy distribution characteristics, and obtaining an energy characteristic vector; The phase sequence is unwrapped and statistically analyzed to obtain a phase feature vector; based on the statistical characteristics of the phase feature vector and the mutation feature sequence, a mutation feature vector is generated; and the time domain feature vector, the energy feature vector, the phase feature vector and the mutation feature vector are combined into a feature vector.

7. The method for realizing multi-scenario terminal signal-to-noise ratio estimation according to claim 5, characterized in that: Step S13 is further as follows: S131, performing wavelet decomposition on the environmental parameter data to obtain environmental characteristic coefficients of different scales; performing LOESS smoothing on the environmental characteristic coefficients based on preset local regression parameters to obtain smoothed environmental coefficients; calculating the first-order and second-order differences of the smoothed environmental coefficients to obtain environmental change rate data; performing Fourier transform on the smoothed environmental coefficients and the environmental change rate data to obtain environmental spectrum data; S132, based on the preset neural network parameters, nonlinearly mapping the environmental spectrum data to obtain environmental mapping data; performing principal component analysis on the environmental mapping data, selecting the principal component whose cumulative contribution rate exceeds the preset threshold value, to obtain environmental principal component data; based on the environmental principal component data, calculating the Mahalanobis distance to obtain environmental anomaly data; Based on the environmental anomaly data, the environmental principal component data is weighted to obtain the environmental compensation coefficient; S133, performing Kalman filtering on the platform attitude data and the antenna pointing data to obtain filtered attitude data and filtered pointing data; based on preset quaternion parameters, converting the filtered attitude data into quaternion representation to obtain an attitude quaternion; calculating the angular velocity and angular acceleration of the attitude quaternion to obtain attitude dynamic data; Based on the filtered pointing data and attitude dynamic data, the state observation equation is constructed, and the attitude prediction data is obtained through the unscented Kalman filter; S134, based on the preset fuzzy rule parameters, fuzzy reasoning is performed on the posture prediction data to obtain a posture evaluation value; based on the posture evaluation value, a compensation weight is calculated to obtain posture weight data; and the posture prediction data and the posture weight data are weightedly combined to obtain a posture compensation coefficient; S135. Group the feature vectors according to preset dimensions to obtain group feature data; based on the environment compensation coefficient and the posture compensation coefficient, compensate each group of the corresponding group feature data to obtain group compensation data; perform nonlinear optimization on the group compensation data to obtain optimized compensation data; reorganize the optimized compensation data to generate a compensation feature vector.

8. The method for implementing multi-scenario terminal signal-to-noise ratio estimation according to claim 5, characterized in that: Step S21 is further as follows: S211, based on preset empirical mode decomposition parameters, perform EMD decomposition on the optimized feature sequence to obtain an intrinsic mode function; calculate the Hilbert spectrum of the intrinsic mode function to obtain time-frequency spectrum data; adaptively segment the time-frequency spectrum data to obtain segmented spectrum data; S212, performing energy integration on the segmented spectrum data in different frequency bands to obtain frequency band energy data; Calculate the coefficient of variation of frequency band energy data to obtain the energy coefficient of variation; stratify the energy coefficient of variation based on a preset adaptive threshold parameter to obtain energy hierarchical data; S213, based on the intrinsic mode function, calculating the instantaneous frequency to obtain an instantaneous frequency sequence; performing time-frequency clustering analysis on the instantaneous frequency sequence to obtain frequency clustering data; combining the frequency clustering data and the energy hierarchy data to generate a time-frequency distribution matrix; S214, based on preset wavelet basis parameters, performing wavelet transform on the time-frequency distribution matrix to obtain wavelet coefficient data; Calculate the local energy of the wavelet coefficient data to obtain a local energy matrix; based on the local energy matrix, extract the energy distribution characteristics to obtain an energy feature sequence; S215, performing kernel density estimation on the energy feature sequence to obtain energy density data; Combining energy density data and time-frequency distribution matrix, energy pattern data is obtained through tensor decomposition; the energy pattern data is reconstructed to generate energy distribution characteristics.

9. The method for realizing multi-scenario terminal signal-to-noise ratio estimation according to claim 5, characterized in that: Step S31 is further as follows: S311, acquiring speed data from the platform sensor, performing multi-point sampling, and obtaining a speed sampling sequence; performing Kalman filtering on the speed sampling sequence to obtain a filtered speed value; calculating the standard deviation of the speed sampling sequence to obtain a speed fluctuation value; combining the filtered speed value and the speed fluctuation value into a speed feature vector; S312, acquiring acceleration data from the platform sensor, sampling, and obtaining an acceleration sampling sequence; performing median filtering on the acceleration sampling sequence to obtain a filtered acceleration value; Based on the preset threshold parameters, mutation detection is performed on the acceleration sampling sequence to obtain an acceleration mutation mark; the filtered acceleration value and the acceleration mutation mark are combined into an acceleration feature vector; S313, acquiring attitude change rate data from the platform sensor; performing exponential smoothing on the attitude change rate data to obtain a smooth attitude change rate; calculating the first-order difference of the attitude change rate data to obtain an attitude acceleration value; performing motion pattern analysis based on the smooth attitude change rate and the attitude acceleration value to generate an attitude feature vector; S314, based on the preset entropy value calculation parameters, perform entropy value calculation on the environmental parameter data to obtain the environmental entropy value; perform time series correlation analysis on the environmental parameter data to obtain environmental change characteristics; combine the environmental entropy value and the environmental change characteristics into an environmental feature vector; S315, performing feature fusion on the velocity feature vector, the acceleration feature vector, the posture feature vector and the environment feature vector to generate a scene feature vector.

10. A device for realizing signal-to-noise ratio estimation of a terminal in multiple scenarios, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the multi-scenario terminal signal-to-noise ratio estimation method described in any one of claims 1 to 9.

Images