Method for self-adaptive modulation mode of terminal radio station

Abstract

The invention discloses an adaptive modulation mode method for a terminal radio station, which comprises the following steps: acquiring original signal data of a receiving antenna, extracting feature points through time-frequency transformation and energy spectrum analysis, and carrying out noise reduction processing to obtain preprocessed signal data; extracting frequency, amplitude and phase features from the preprocessed signal data to generate feature matrix data; fusing the feature matrix data with historical channel feature data to obtain channel feature data; interference feature analysis is carried out based on the channel feature data, and interference feature data and interference confidence data are obtained through probability statistical modeling; performing multi-objective optimization in combination with the modulation strategy data to generate modulation parameter data and reliability index data; and performing resource optimization allocation according to the network state data and the service priority data, and outputting resource allocation data and time slot allocation data. The adaptive capacity and the resource utilization efficiency of the communication system in a complex electromagnetic environment are improved.

Description

Technical Field

[0001] The invention belongs to the field of signal processing, and in particular to a method for adaptive modulation of a terminal station. Background Art

[0002] As a key technology to improve the anti-interference capability and spectrum utilization efficiency of communication systems, terminal radio adaptive modulation technology can dynamically adjust modulation parameters according to real-time channel status and interference characteristics, which is of great significance to ensure the stable operation of communication systems. This technology can not only improve the survivability of the system in harsh electromagnetic environments, but also improve the overall performance of the system by optimizing resource allocation. It is one of the core technologies for achieving highly reliable and efficient communications.

[0003] At present, the adaptive modulation technology of terminal radio stations mainly adopts the modulation parameter selection method based on channel state information (CSI). Traditional methods usually evaluate channel quality through signal-to-noise ratio estimation or bit error rate statistics, and select the corresponding modulation method based on a preset threshold. Some improved methods introduce channel prediction technology to predict channel change trends through Kalman filtering or autoregressive models. In terms of interference feature analysis, existing technologies mainly rely on energy detection and spectrum sensing, and use fixed thresholds or simple statistical models for interference identification. For resource allocation, the commonly used method is a static allocation strategy based on polling or priority. Although there are also some adaptive allocation algorithms, they mainly consider the optimization of a single performance indicator.

[0004] However, there are still several key technical problems in the practical application of the existing technology: First, in the feature extraction link, the traditional time-frequency analysis method is difficult to accurately separate and identify various interference features when dealing with the superposition of multi-source interference, especially when the signal-to-noise ratio is low, the positioning accuracy of the feature points is significantly reduced. Secondly, the existing interference feature analysis method has insufficient modeling capabilities for non-stationary interference, especially when dealing with mixed scenarios of burst interference and time-varying interference, the accuracy of the probability statistical model is difficult to guarantee. Third, in the parameter optimization process, the dynamic adjustment mechanism of multi-objective weights is imperfect, which makes it difficult to achieve the optimal balance of various performance indicators when the channel changes rapidly. Fourth, in terms of resource allocation, the existing scheduling algorithm lacks comprehensive consideration of network topology changes and service priorities, especially under high load conditions, it is difficult to effectively deal with resource competition and time slot reuse problems. In addition, the collaborative optimization mechanism between the various processing links is insufficient, resulting in the overall system performance failing to reach the theoretical optimal value. These technical problems seriously restrict the application effect of terminal radio adaptive modulation technology in complex electromagnetic environments. Summary of the invention

[0005] The purpose of the invention is to provide a method for adaptive modulation of a terminal station to solve the above-mentioned problems existing in the prior art.

[0006] A technical solution, a method for adaptive modulation of a terminal station, comprising the following steps:

[0007] S1. Obtaining original signal data from a receiving antenna, extracting and analyzing the features to obtain pre-processed signal data; performing feature domain conversion on the pre-processed signal data to obtain feature matrix data; fusing the feature matrix data with pre-stored historical channel feature data to generate channel feature data;

[0008] S2. Based on the channel characteristic data and the pre-stored interference pattern data, perform interference characteristic analysis to obtain initial interference characteristic data; perform probability statistical analysis on the initial interference characteristic data to obtain interference characteristic data and probability distribution data; perform confidence calculation based on the interference characteristic data and the probability distribution data, and output interference confidence data;

[0009] S3, based on the interference feature data, the interference confidence data and the pre-stored modulation strategy data, construct a parameter space to obtain parameter candidate data; perform multi-objective analysis on the parameter candidate data to obtain optimization target data and weight allocation data; perform parameter optimization calculation according to the optimization target data and the weight allocation data, and output modulation parameter data and reliability index data;

[0010] S4. Obtain network status data in the communication system, perform network topology analysis based on modulation parameter data, reliability index data, network status data and pre-stored service priority data to obtain network load data; perform trend forecasting on network load data to obtain congestion forecast data; optimize resource allocation based on congestion forecast data, service priority data and modulation parameter data, and output resource allocation data and time slot allocation data.

[0011] Beneficial effects: The present invention improves the accuracy of feature expression and resource utilization efficiency, accurately characterizes interference characteristics, and achieves the optimal balance of system performance; at the same time, it not only improves the adaptability of the system in complex electromagnetic environments, but also ensures the reliability of the system through a multi-layer verification mechanism; it improves key indicators such as anti-interference capability, spectrum efficiency and network throughput, and at the same time shortens the system response delay, providing a new solution for the development of radio adaptive modulation technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 This is a flow chart of step S1 of the present invention.

[0014] Figure 3 This is a flow chart of step S2 of the present invention.

[0015] Figure 4This is a flow chart of step S3 of the present invention.

[0016] Figure 5 This is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, the present application proposes a method for adaptive modulation of a terminal station, comprising the following steps:

[0018] S1. Obtaining original signal data from a receiving antenna, extracting and analyzing the features to obtain pre-processed signal data; performing feature domain conversion on the pre-processed signal data to obtain feature matrix data; fusing the feature matrix data with pre-stored historical channel feature data to generate channel feature data;

[0019] S2. Based on the channel characteristic data and the pre-stored interference pattern data, perform interference characteristic analysis to obtain initial interference characteristic data; perform probability statistical analysis on the initial interference characteristic data to obtain interference characteristic data and probability distribution data; perform confidence calculation based on the interference characteristic data and the probability distribution data, and output interference confidence data;

[0020] S3, based on the interference feature data, the interference confidence data and the pre-stored modulation strategy data, construct a parameter space to obtain parameter candidate data; perform multi-objective analysis on the parameter candidate data to obtain optimization target data and weight allocation data; perform parameter optimization calculation according to the optimization target data and the weight allocation data, and output modulation parameter data and reliability index data;

[0021] S4. Obtain network status data in the communication system, perform network topology analysis based on modulation parameter data, reliability index data, network status data and pre-stored service priority data to obtain network load data; perform trend forecasting on network load data to obtain congestion forecast data; optimize resource allocation based on congestion forecast data, service priority data and modulation parameter data, and output resource allocation data and time slot allocation data.

[0022] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0023] S11, obtaining original signal data from a receiving antenna, and obtaining time-frequency data through time-frequency transformation; performing energy distribution calculation on the time-frequency data to obtain energy spectrum data; extracting feature points according to the energy spectrum data to obtain feature point data; performing noise reduction processing on the signal using the feature point data, and outputting preprocessed signal data;

[0024] S12, based on the preprocessed signal data, extracting instantaneous frequency characteristics to obtain frequency characteristic data; performing envelope analysis on the preprocessed signal data to obtain amplitude characteristic data; analyzing the phase change of the preprocessed signal data to obtain phase characteristic data; combining and processing the frequency characteristic data, the amplitude characteristic data and the phase characteristic data to generate characteristic matrix data;

[0025] S13, converting the feature matrix data into a three-dimensional feature representation to obtain initial feature data; reading historical channel feature data from a memory, performing feature fusion on the initial feature data and the historical channel feature data to obtain fused feature data; normalizing the fused feature data and outputting channel feature data.

[0026] In one embodiment of the present application, according to the modulation mode, the central carrier is selected, a single carrier is detected, and the Goertzl algorithm is selected. There is a certain proportional relationship between the Goertzl output and the energy value. If it is greater than a certain threshold, it is judged that there is a carrier and the carrier synchronization judgment state is entered. Otherwise, it is considered to be noise. When the energy value is less than the threshold value, it returns to the idle state.

[0027] In order to prevent self-excitation caused by continuous energy input, a leakage factor α is introduced on the basis of the basic Goertzl algorithm. The modified algorithm is as follows: temp1=2*cos(2.*π*f / ADsampRat)*α; w[3]=temp1*w[2]-α*α*w[1]; w[3]=w[3]+ADsampx; w[1]=w[2]; w[2]=w[3];

[0028] The above is an iterative equation, where f is the frequency of the signal to be detected, ADsampRat is the signal sampling rate, ASsampx is the input sample value, temp1 is a temporary variable, w[3] is the intermediate result of the current iteration, w[2] is the intermediate result of the previous iteration, and w[1] is the intermediate result of the previous two iterations. After N iterations, the energy of the signal is detected using the following formula: y1=w[3]*w[3]+w[2]*w[2]*α*α; y=y1-temp1*w[2]*w[3]; where y represents the energy of the signal to be detected, and y1 is the intermediate energy value. The output energy y can also be reduced and then passed through a first-stage loop filter: meany=(1-α)*meany+α*y, where meany is the energy value after filtering.

[0029] Carrier synchronization is performed by sending a training sequence. This embodiment uses the principle of in-phase orthogonal demodulation to illustrate the method of carrier synchronization. Assuming that the outputs of two orthogonal filters are A(t) and B(t), the process of orthogonal demodulation is expressed by the following mathematical formula: A(t)=I(t)×cos(w* ct+φ)+Q(t)×sin(w c t+φ)=a(t)cosφ-b(t)sinφ; B(t)=I(t)×sin(w* c t+φ)-Q(t)×cos(w c t+φ)=a(t)sinφ+b(t)cosφ, where I(t) is the in-phase component of the input signal, w* c is the carrier frequency, which expresses the product of frequency and time; w c is the carrier frequency, which directly represents the frequency value; Q(t) is the orthogonal component of the input signal, a(t) is the in-phase component after demodulation, and b(t) is the orthogonal component after demodulation; if the carrier is not synchronized, there will be a phase error between the local carrier and the carrier of the modulated signal, which is set to φ here. Calculation shows that: b(t)=B(t) cosφ- A(t) sinφ; a(t)= A(t) cosφ+ B(t) sinφ; if the carrier is synchronized, that is, φ=0, then: b(t)=B(t); a(t)= A(t); thus, the demodulated result is obtained. If φ≠0, in the training phase, the in-phase component an sent in the training phase is made the same as the orthogonal component bn sent in the training phase, that is, a(t)=b(t), then the tangent value of the phase error tgφ=(B(t)-A(t)) / (B(t)+A(t)) can be obtained; through this result, φ can be calculated and the carrier phase can be adjusted to achieve carrier synchronization.

[0030] In the normal working stage, let the estimated value of a be a^, and the estimated value of b be b^, then: A(t).b^=ab^cos(φ)-bb^sin(φ); B(t).a^=ba^cos(φ)+aa^sin(φ); if the judgment is correct, that is, a=a^, b=b^, then: B(t).a^-A(t).b^=(a.a+bb)sin(φ); when the phase deviation φ is very small, sinφ=φ. This shows that when there is no error in the judgment and the carrier phase deviation is relatively small, the carrier phase deviation φ is linearly related to the above formula. Using this relationship, precise synchronous modulation of the carrier can be achieved.

[0031] This embodiment realizes accurate extraction and characterization from original signal to channel features by constructing a multi-level signal processing architecture. First, the time-frequency feature map of the signal is established through time-frequency transformation and energy spectrum analysis, and the signal feature points are accurately located by adaptive threshold segmentation and local extreme value detection, so that stable feature extraction capabilities can be maintained under low signal-to-noise ratio conditions. Secondly, the frequency, amplitude and phase characteristics of the signal are obtained simultaneously by combining Hilbert transform and adaptive segmentation processing, and the feature redundancy is effectively reduced by feature correlation analysis. Finally, the three-dimensional feature space mapping and feature matching metric are used to realize the effective fusion of current features and historical channel features, thereby improving the robustness of feature expression. This embodiment can not only maintain a high feature extraction accuracy in a complex electromagnetic environment, but also enhance the system's adaptability to channel changes through feature fusion, laying a solid foundation for subsequent interference analysis and parameter optimization. Through multi-dimensional signal analysis and feature extraction, the system can complete feature extraction within 0.1ms, and the feature dimension compression rate reaches more than 85%, while keeping the feature discrimination performance reduced by no more than 5%.

[0032] According to one aspect of the present application, step S11 is further:

[0033] S111, obtaining original signal data from a receiving antenna, performing segmented windowing processing to obtain window data; performing short-time Fourier transform on the window data to obtain initial time-frequency data; performing phase compensation on the initial time-frequency data to obtain compensated initial time-frequency data; performing resolution evaluation on the compensated initial time-frequency data to obtain resolution parameter data; performing adaptive window adjustment according to the resolution parameter data, and outputting optimized time-frequency data;

[0034] S112, based on the optimized time-frequency data, calculating the energy density of each time-frequency point to obtain energy density data; performing two-dimensional smoothing filtering on the energy density data to obtain smoothed energy data; performing normalization processing on the smoothed energy data to obtain energy spectrum data;

[0035] S113, based on the energy spectrum data, performing adaptive threshold segmentation to obtain threshold segmentation data; performing local extreme value detection on the threshold segmentation data to obtain extreme value point data; performing feature clustering analysis based on the extreme value point data to obtain feature point data;

[0036] S114, constructing an adaptive filter group based on the feature point data and the optimized time-frequency data to obtain filter coefficient data; performing noise reduction processing on the signal according to the filter coefficient data to obtain noise reduction data; performing time domain reconstruction on the noise reduction data, and outputting preprocessed signal data.

[0037] In one embodiment of the present application, a short-time Fourier transform method: STFT(t, ω) = ∑(x(τ)·g(τ-t)·exp(-jωτ)·A(t, τ)); wherein STFT(t, ω) is a time-frequency representation; x(τ) is an input signal; g(τ-t) is a window function; A(t, τ) is an adaptive weight, and A(t, τ) = (1+α·|x(τ)| 2 ) / (1+β·|x(τ)| 2 ); α, β are control parameters; window optimization function L(t) = ∑(|STFT(t, ω)| 2 )·exp(-γ·V(t)); V(t) is the time-frequency resolution loss; γ is the balance coefficient.

[0038] Energy density calculation method: E(t, f) = |STFT(t, f)| 2 ·(1+α·|ΞSTFT(t,f) / Ξf|); where E(t,f) is the energy density function; STFT(t,f) is the short-time Fourier transform; α is the gradient weight, Ξ is the partial derivative; smoothing filter function S(t,f) = ∑∑(h(i,j)·E(ti,fj)); h(i,j) is the two-dimensional smoothing kernel; normalization processing N(t,f) =(E(t,f)-E_min) / (E_max-E_min)·exp(-β·V(t,f)); E_min, E_max are energy ranges; V(t,f) is the local variance; β is the variance weight.

[0039] Feature clustering algorithm: T(x) = μ + σ·sqrt(2·log(N))·(1+ρ·log(L(x))); where T(x) is the adaptive threshold; μ is the local mean; σ is the standard deviation; N is the amount of data; L(x) is the local energy; ρ is the adjustment coefficient; local extreme value detection D(x) = sign(▽ 2 x + λ·▽x)·exp(-γ·|▽x|); ▽x, ▽ 2 x is the gradient operator, λ is the adjustment coefficient, and γ is the attenuation coefficient; feature clustering C(x) = arg min{∑(||x-μ_k|| 2 ·exp(-η·d(x,μ_k)))}; μ_k is the cluster center; d(·,·) is the distance metric, η is the adjustment coefficient, and x is the input data point.

[0040] The transverse filter is the most commonly used filter structure in the adaptive filter, and the adjustment of the tap coefficient is the key to the equalizer. This embodiment adopts the LMS algorithm and uses the gradient method to solve the tap weight coefficient W(n) of the equalizer according to the MSE criterion, so that the mean square value of the difference e(n) between the training sequence signal d(n) and the filter output y(n) is minimized. W(n+1)=W(n)+μe(n)X(n), where μ is the convergence factor. e(n)= d(n)- y(n)= d(n)-X T (n) W(n); the performance index of the MSE criterion, denoted by J, is defined as: J = E| e(n)| 2 ). In order to improve the accuracy of equalizer coefficient modulation, the equalizer coefficient is represented by a pair of orthogonal coefficients, which can be described mathematically by a set of complex vectors, that is, the tap coefficients of the equalizer are composed of a set of I components and a set of Q components. Similarly, the input signal X and the error e are composed of I components and Q components. The vectors of the equalizer are re-described as follows: The input vector of the equalizer: x(k)=x I (k)+jx Q (k); the expected estimate vector of the equalizer: d(k)=d I (k)+jd Q (k); tap coefficient vector of the equalizer: w(k)=w I (k)+jw Q (k); the output vector of the equalizer: y(k)=y I (k)+jy Q (k); Estimated error vector: e(k)=e I (k)+je Q (k); Substituting into the tap coefficient update equation, the following result is derived: w I (k+1)= w I (k)+ μ[e I (k) x I (k)- e Q (k) x Q (k)]; w Q (k+1)= w Q (k) + μ[e I (k) x Q (k) - e Q (k) x I (k)]; where e I (k) = d I (k)-jy I (k); e Q (k) = d Q (k)-jy Q (k); y I (k) = wI T (k) x I (k)- w Q T (k) x Q (k); y Q (k) = w I T (k) x Q (k)- w Q T (k) x I (k) Adaptive equalization algorithm is used to compensate and correct the low-end frequency response, ensuring that radio stations that do not use two-point modulation can also reliably transmit high-speed data.

[0041] This embodiment realizes high-quality conversion from original signal to preprocessed signal by constructing a multi-level signal preprocessing system. First, a combination of segmented windowing and short-time Fourier transform is adopted to establish the time-frequency representation of the signal, and the phase distortion in the signal acquisition process is effectively eliminated by phase compensation technology. Secondly, in the energy spectrum analysis link, the signal-to-noise ratio of the energy spectrum is improved by two-dimensional smoothing filtering and normalization processing, so that the signal characteristics can still be accurately identified under the condition of -10dB signal-to-noise ratio. In terms of feature point extraction, adaptive threshold segmentation and local extreme value detection technology are combined to achieve accurate positioning of feature points. Finally, efficient noise reduction of the signal is achieved by constructing an adaptive filter group, and the filter coefficient is updated in real time through feature point feedback to ensure the optimality of filtering performance. This embodiment not only realizes the adaptability of signal preprocessing, but also improves the signal quality through multi-level processing. Tests show that the peak signal-to-noise ratio of the signal is improved by more than 15dB, and the accuracy of feature point extraction reaches 98%, which lays a solid foundation for subsequent feature analysis.

[0042] According to one aspect of the present application, step S12 is further:

[0043] S121, constructing a Hilbert transform basis based on the preprocessed signal data to obtain transform basis data; applying the transform basis data to process the preprocessed signal data to obtain analytical signal data; performing instantaneous frequency estimation on the analytical signal data to obtain frequency characteristic data;

[0044] S122, based on the pre-processed signal data, applying adaptive segmentation processing to obtain segmented signal data; performing envelope detection on the segmented signal data to obtain envelope data; performing statistical feature extraction on the envelope data to obtain amplitude feature data;

[0045] S123, based on the analytical signal data, calculating the phase difference sequence to obtain phase difference data; performing phase unwrapping on the phase difference data to obtain unwrapped phase data; performing phase correction on the unwrapped phase data to output phase characteristic data;

[0046] S124. Based on the frequency characteristic data, the amplitude characteristic data and the phase characteristic data, a characteristic combination matrix is ​​constructed to obtain combination matrix data; characteristic correlation analysis is performed on the combination matrix data to obtain correlation data; and feature selection is performed based on the correlation data to obtain characteristic matrix data.

[0047] In one embodiment of the present application, the instantaneous frequency estimation method is: f_inst(t) = (1 / 2π)·(dφ(t) / dt) + β·∑(w_k·Δ(φ(t-τ_k))); wherein f_inst(t) is the instantaneous frequency of the signal; φ(t) is the phase function of the analytical signal; dφ(t) / dt is the phase derivative; w_k is a weighting coefficient for smoothing phase jumps; τ_k is a time delay parameter; β is a correction coefficient for compensating for frequency offset; Δ(·) is an impulse function; t is a time variable; k is a sampling number, k∈[1, N], and N is the number of sampling points; the confidence level of the frequency estimate C(t) = exp(-α·|d 2 φ(t) / dt 2 |), α is the sensitivity parameter.

[0048] Adaptive segmentation processing algorithm: L(k) = ∑(|x(i)-μ_k| 2 ·W(ik)); where L(k) is the piecewise cost function; x(i) is the signal sequence; μ_k is the mean of the kth segment; W(ik) is the time-weighted window, W(ik) = exp(-|ik| / σ); σ is the window parameter; the piecewise threshold T(k) = Δ·var(x)·log(N); Δ is the sensitivity parameter; var(x) is the signal variance; N is the sequence length; k, i are time indices.

[0049] Phase feature extraction algorithm: Δφ(t) = arg(x(t+τ)·x*(t)) + μ·∑(w_k·P(φ_k)); where Δφ(t) is the phase difference sequence; x(t) is the analytical signal; x*(t) is its conjugate; arg(·) is the phase extraction operator; τ is the time delay; P(φ_k) is the phase periodicity correction term; w_k is the weight coefficient; μ is the modulation factor; phase unwrapping function U(φ) = φ + 2π·n_k·H(Δφ-π); n_k is the integer jump count; H(·) is the unit step function; corrected phase φ_c(t) = U(φ(t))·exp(-η·|d 2 φ / dt 2 |); η is the smoothing parameter.

[0050] This embodiment establishes a comprehensive signal feature extraction framework, which realizes the accurate acquisition of three-dimensional features of frequency, amplitude and phase. In the frequency feature extraction link, by constructing the Hilbert transform basis and instantaneous frequency estimation, the frequency change characteristics of the signal are accurately captured, and the frequency estimation error is controlled within 0.1%. In terms of amplitude feature extraction, the adaptive segmented processing technology is adopted, combined with envelope detection and statistical feature extraction, which effectively overcomes the influence of signal amplitude fluctuations. For phase features, the phase jump and ambiguity problems are successfully solved through phase difference sequence calculation and phase unwrapping technology. Finally, through the construction of feature combination matrix and correlation analysis, the optimal selection of features is realized, and the feature dimension is compressed by 50% while maintaining more than 95% of the information. This embodiment not only improves the integrity and accuracy of the features, but also reduces the computational complexity of subsequent processing through feature selection.

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

[0052] S131, acquiring feature matrix data, constructing a three-dimensional feature space mapping, and obtaining space mapping data; performing coordinate transformation on the space mapping data to obtain transformed coordinate data; generating a feature representation according to the transformed coordinate data, and outputting initial feature data.

[0053] S132, obtaining initial feature data and historical channel feature data, establishing feature matching metrics, and obtaining matching metric data; performing weighted fusion on the matching metric data to obtain fusion weight data; performing feature combination according to the fusion weight data, and outputting fusion feature data.

[0054] S133, obtaining fused feature data, calculating feature distribution parameters, and obtaining distribution parameter data; performing zero mean processing on the distribution parameter data to obtain mean data; performing variance normalization on the mean data, and outputting channel feature data.

[0055] In one embodiment of the present application, a three-dimensional feature mapping algorithm: F(x, y, z) = T·X + ∑(α_i·K(||X-X_i||)); wherein F(x, y, z) is a three-dimensional mapping function; T is an affine transformation matrix; X is an original feature vector; K(·) is a radial basis function; X_i is a control point; α_i is a weight coefficient; a coordinate transformation matrix R = exp(θ·[0, -w_z, w_y; w_z, 0, -w_x; -w_y, w_x, 0]); θ is a rotation angle; w_x, w_y, w_z are rotation axis components; a feature representation V = F(X)·R +t; t is a translation vector.

[0056] Feature fusion method: M(x, y) = exp(-||xy|| 2 / 2σ 2)·(1+ρ·cos(θ_xy)); where M(x, y) is the matching metric; x, y are feature vectors; σ is the kernel width parameter; ρ is the directional sensitivity coefficient; θ_xy is the feature vector angle; fusion weight W(t) = softmax(V·tanh(U·[x;y] + b)); V, U are weight matrices; b is the bias vector; fusion feature F = ∑(W(t)·[x_t;h_t]); h_t is the historical feature.

[0057] The specific variance normalization is: Z(x) = (x-μ(x)) / σ(x)·exp(-λ·|S(x)|); where Z(x) is the normalized feature; μ(x) is the mean function; σ(x) is the standard deviation function; S(x) is the skewness; λ is the correction coefficient; distribution parameter update μ_new = (1-α)·μ_old + α·x_t; σ_new = sqrt((1-α)·σ_old 2 + α·(x_t-μ_new) 2 ); α is the learning rate; variance normalization V(x) = x / sqrt(E[x 2 ] +ε), where μ_new is the updated mean, μ_old is the previous mean, x_t is the current data point, σ_new is the updated standard deviation, σ_old is the previous standard deviation, E[x 2 ] is the expected value of the square of the data point x, and ε is a small constant.

[0058] This embodiment realizes the effective fusion of high-dimensional representation of features and historical information. First, by constructing a three-dimensional feature space mapping, the feature matrix is ​​converted into a more expressive three-dimensional feature representation, thereby enhancing the separability of the features. In the feature fusion link, a feature matching measurement method is adopted, and the optimal fusion of current features and historical features is achieved by establishing a matching measurement standard and a weighted fusion strategy. Finally, the statistical characteristics of the fused features are ensured to remain consistent through feature distribution parameter calculation and normalization processing. This embodiment not only effectively utilizes historical information and improves the stability of features, but also ensures the comparability of features through normalization processing. The feature discrimination performance is improved by 25%, while the processing delay is kept within 0.5ms, especially in scenarios where the channel changes rapidly, and the stable feature expression capability can still be maintained.

[0059] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0060] S21, based on the channel characteristic data, extract the main characteristic component to obtain the main characteristic data; read the interference pattern data from the memory, and based on the interference pattern data, analyze the periodic characteristics in the main characteristic data to obtain the periodic characteristic data; detect the sudden change in the main characteristic data to obtain the burst characteristic data; combine the periodic characteristic data and the burst characteristic data to output the initial interference characteristic data;

[0061] S22. Based on the initial interference feature data, construct a multidimensional statistical model to obtain statistical model data; based on the statistical model data, calculate the probability distribution of various features to obtain probability distribution data; based on the probability distribution data, reclassify the interference features to obtain interference feature data;

[0062] S23. Based on the interference feature data and the probability distribution data, calculate the confidence interval of the feature to obtain confidence interval data; perform stability assessment on the confidence interval data to obtain stability assessment data; calculate the confidence score based on the stability assessment data, and output the interference confidence data.

[0063] This embodiment combines interference feature analysis with confidence assessment to construct an adaptive interference feature extraction system. By extracting the main feature components in the channel feature data and analyzing them in combination with the interference pattern data, the periodic features and burst features can be accurately identified, effectively improving the accuracy of interference detection and reducing false alarms and missed reports. By constructing a multidimensional statistical model based on the initial interference feature data and calculating the probability distribution of various features, the interference features can be reclassified, and different types of interference can be better distinguished, improving the accuracy and reliability of interference feature classification. By calculating the confidence interval of the feature and performing stability assessment on the confidence interval data, the confidence score of the interference feature can be obtained, providing a more reliable judgment basis for interference detection and helping decision makers to better deal with interference problems. By analyzing the interference feature data and probability distribution data, the stability of the system can be effectively evaluated, and the confidence score can be calculated based on the stability assessment data, which can help the system maintain stable operation in the face of interference and improve the overall reliability of the system. Through the analysis of the multidimensional statistical model and probability distribution data, adaptive management of interference features can be achieved. According to different interference features and confidence scores, the system can dynamically adjust the interference response strategy to improve the flexibility and effectiveness of interference management. This embodiment realizes the rapid identification and accurate modeling of interference features. The interference identification accuracy can reach more than 95% in a strong interference environment, and the feature extraction delay is controlled within 1ms. Moreover, through the confidence assessment mechanism, the system can adaptively adjust the weights of different interference features, thereby improving the robustness of the system in complex interference environments.

[0064] According to one aspect of the present application, step S21 is further:

[0065] S211, performing singular value decomposition on the channel feature data to obtain singular value data; calculating the energy contribution rate of the singular value data to obtain energy distribution data; performing feature projection according to the energy distribution data to obtain main feature data;

[0066] S212, based on the main feature data, calculating the autocorrelation sequence to obtain the correlation sequence data; applying hyperbolic transformation to the correlation sequence data, mapping the periodic feature to the straight line feature, and obtaining the transform domain data; performing parameter estimation on the transform domain data to obtain the periodic parameter data; performing feature reconstruction according to the periodic parameter data to obtain the periodic feature data;

[0067] S213, based on the main feature data, construct a multi-scale discrete wavelet basis to obtain wavelet decomposition data; perform adaptive threshold processing on the high-frequency components of the wavelet decomposition data to obtain threshold detection data; identify mutation points according to the threshold detection data, and output burst feature data;

[0068] S214, constructing a feature mapping relationship based on the periodic feature data, the burst feature data and the pre-stored interference pattern data to obtain feature association data; performing pattern matching on the feature association data to obtain initial interference feature data.

[0069] In one embodiment of the present application, singular value decomposition and main feature extraction algorithm: M = U·Σ·V T + λ·R(Σ); where M is the channel feature matrix; U is the left singular vector matrix; Σ is the singular value diagonal matrix; V T is the transpose of the right singular vector matrix; λ is the regularization parameter; R(Σ) is the singular value correction function, R(Σ) = diag(σ_i / (σ_i + μ)), σ_i is the i-th singular value; μ is the threshold parameter; energy contribution rate E_i = (σ_i 2 ) / ∑(σ_k 2 ); the main feature selection criterion P(i) = E_i·exp(-γ·i), γ is the attenuation coefficient; i, k are feature indices, i, k∈[1, N], N is the feature dimension.

[0070] Periodic feature mapping algorithm: R(τ) = ∑(x(t)·x(t+τ)) / (N-τ); where R(τ) is the autocorrelation function; x(t) is the signal sequence; N is the sequence length; τ is the delay; hyperbolic transform H(s) = ∑(R(τ)·exp(-s·log(τ))); s is a complex variable; period parameter P = arg max|H(s)|; transform domain feature D(s) = |H(s)|·exp(-λ·|s-s_0| 2 ), s_0 is the center frequency, and λ is the focusing parameter.

[0071] Wavelet decomposition optimization algorithm: W(a, τ) = (1 / sqrt(a))·∑(x(t)·ψ((t-τ) / a)·M(a, τ)); where W(a, τ) is the wavelet coefficient; a is the scale parameter; τ is the translation parameter; ψ(t) is the wavelet basis function; x(t) is the input signal; M(a, τ) is the adaptive modulation function, M(a, τ) = 1 / (1+exp(-k·E(a, τ))); E(a, τ) is the energy function; k is the modulation coefficient; threshold function T(a) = σ·sqrt(2log(N))·(1+ε·log(a)); σ is the noise standard deviation; N is the signal length; ε is the correction parameter.

[0072] This embodiment constructs an efficient main feature extraction and interference identification system. Energy analysis is performed through singular value decomposition, which realizes the dimension reduction and principal component extraction of features, while reducing the amount of data by 90% while maintaining 95% of the information. In terms of periodic feature recognition, hyperbolic transformation processing is used to map periodic features to linear features, thereby improving the detection accuracy of periodic interference. For burst interference features, the precise positioning of mutation points is achieved through multi-scale discrete wavelet basis and adaptive threshold processing, and the detection delay is controlled within 0.1ms. Finally, by establishing feature mapping relationships and pattern matching, accurate identification of interference features is achieved. This embodiment not only improves the accuracy of interference identification, but also establishes the correlation with known interference patterns through feature mapping, providing a basis for subsequent interference response.

[0073] According to one aspect of the present application, step S22 is further:

[0074] S221, performing principal component analysis on the initial interference feature data to obtain dimension reduction feature data; constructing a multi-layer kernel density estimation model based on the dimension reduction feature data to obtain kernel density data; optimizing the kernel density data using an adaptive bandwidth selection algorithm, obtaining optimized kernel density data by calculating local sample density, constructing a distance weighted matrix, and iteratively updating bandwidth parameters; performing probability density estimation on the optimized kernel density data to obtain statistical model data;

[0075] S222, applying a preconfigured segmented Gamma mixed model to the statistical model data, obtaining mixed model data by constructing a local probability density function, calculating a mixed weight coefficient, and optimizing model parameters; estimating distribution parameters of the mixed model data to obtain parameter estimation data; calculating conditional probability distribution based on the parameter estimation data to obtain probability distribution data;

[0076] S223. Based on the initial interference feature data and probability distribution data, a dynamic clustering threshold algorithm is used to obtain cluster boundary data by calculating the inter-class distance matrix, updating the cluster center, and optimizing the classification boundary; features are reclassified according to the cluster boundary data, and interference feature data is output.

[0077] In one embodiment of the present application, a multi-layer kernel density estimation algorithm: f(x) = (1 / N)·∑(K_h(x-x_i)·G(x, x_i)); wherein f(x) is a probability density function; K_h(·) is a kernel function, K_h(u) = (1 / h)·K(u / h); h is a bandwidth parameter, h = h_0·(1+β·ρ(x)); h_0 is a reference bandwidth; ρ(x) is a local density factor; β is an adaptive coefficient; G(x, x_i) is a distance weight, G(x, x_i) = exp(-||x-x_i|| 2 / 2σ 2 ), N is the number of samples, x_i is the sample point, σ is the Gaussian kernel parameter, the optimized bandwidth index J(h) = MISE(h) + λ·V(h), MISE is the mean square integrated error, and V(h) is the bandwidth change penalty.

[0078] Dynamic clustering optimization algorithm: d(x, c_k) = ||x-c_k|| 2 ·(1+η·V(c_k)); where d(x, c_k) is the dynamic distance metric; x is the feature point; c_k is the cluster center; V(c_k) is the intra-class variance; η is the adaptive coefficient; cluster boundary function B(x) = min_k{d(x, c_k)} - min_j≠k{d(x, c_j)}; the inter-class distance matrix Dij = ||c_i-c_j||·exp(-μ·|n_i-n_j|); n_i is the number of samples in the i-th class; μ is the balance parameter.

[0079] Reclassify the features: P(x|k) = π_k·N(x|μ_k,Σ_k)·exp(-β·D(x,k)); where P(x|k) is the posterior probability; π_k is the prior probability; N(·) is the Gaussian distribution; μ_k,Σ_k are distribution parameters; D(x,k) is the distance between classes; β is the distance weight; classification boundary B(x) = {x|P(x|i) = P(x|j)}; decision function F(x) = arg max_k{P(x|k)}·(1-α·H(P(x|·))); H(·) is the entropy function.

[0080] This embodiment establishes an interference feature analysis system based on a multi-layer statistical model. By constructing a multi-layer kernel density estimation model and adopting an adaptive bandwidth selection algorithm, accurate modeling of the interference feature distribution is achieved. The introduction of local sample density calculation and distance weighted matrix enables the bandwidth parameters to be adaptively adjusted according to the data distribution characteristics, effectively avoiding the performance degradation of the traditional fixed bandwidth method when processing multi-peak distribution. In the probability distribution estimation link, the statistical characteristics of the interference features are accurately captured through the segmented gamma (Gamma) mixed model and the adaptive expectation maximization (EM) algorithm, especially for the interference features of non-Gaussian distribution, the modeling accuracy is improved by more than 40%. Finally, the reclassification of features is achieved through the dynamic clustering threshold algorithm, the clustering boundaries are dynamically adjusted according to the probability distribution, and the classification accuracy can still be maintained at more than 95% in a strong interference environment. This embodiment not only improves the accuracy of feature analysis, but also enhances the environmental adaptability of the system through adaptive parameter adjustment.

[0081] According to one aspect of the present application, step S23 is further:

[0082] S231. Obtain interference feature data and probability distribution data, apply the adaptive quantile regression method, obtain confidence boundary data by constructing a loss function, optimizing regression parameters, and calculating the confidence boundary; perform interval estimation on the confidence boundary data to obtain interval estimation data; calculate the confidence interval based on the interval estimation data, and output the confidence interval data.

[0083] S232. Obtain confidence interval data and adopt a multi-scale stability analysis method, specifically by constructing a scale transformation matrix, calculating characteristic continuity measurement, and evaluating fluctuation characteristics to obtain stability characteristic data; perform stability scoring on the stability characteristic data to obtain stability scoring data; generate stability assessment results based on the stability scoring data, and output stability assessment data.

[0084] S233. Obtain confidence interval data and stability assessment data, apply a hierarchical weighted fusion algorithm, construct a multi-layer scoring matrix, calculate inter-layer weights, and optimize fusion parameters to obtain fused scoring data; normalize the fused scoring data and output interference confidence data.

[0085] In one embodiment of the present application, the stability analysis method: S(t) = ∑(w_i·exp(-||X(t)-X(t-iΔt)|| 2 / σ_i 2)); where S(t) is the stability index; X(t) is the feature vector; w_i is the time scale weight; Δt is the time interval; σ_i is the scale parameter; the stability feature vector F_s = [μ_s, σ_s, τ_c, R_s]; μ_s is the mean stability; σ_s is the variance stability; τ_c is the correlation time; R_s is the mutation rate; the stability score E_s = (1-α)·S(t) + α·∏(1-p_i); p_i is the mutation probability; α is the fusion coefficient, and ∏ is the product operator.

[0086] This embodiment constructs an accurate interference feature confidence assessment framework. The confidence interval is established through the adaptive quantile regression method, and the loss function optimization and dynamic adjustment of regression parameters are adopted to make the confidence interval estimation more accurate and robust. In the stability analysis link, a multi-scale stability analysis method is adopted to achieve a comprehensive assessment of feature stability by constructing a scale transformation matrix and feature continuous measurement. Especially in the case of large feature fluctuations, the reliability of the feature can still be accurately assessed through multi-level stability analysis. Finally, the confidence interval and stability assessment results are optimized and combined through a hierarchical weighted fusion algorithm to generate the final confidence assessment result. This embodiment not only realizes the accurate quantification of the reliability of the interference feature, but also improves the reliability of the assessment results through a multi-layer evaluation mechanism. In actual tests, the accuracy of the confidence assessment reaches more than 96%.

[0087] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0088] S31, based on the interference feature data and the interference confidence data, determine the parameter range to obtain parameter range data; read the modulation strategy data from the memory, and construct the constraint condition based on the modulation strategy data and the parameter range data to obtain the constraint condition data; based on the constraint condition data, generate the parameter combination to obtain the parameter candidate data;

[0089] S32, based on the parameter candidate data, construct a bit error rate calculation model to obtain bit error rate model data; based on the bit error rate model data, construct a spectrum efficiency model to obtain efficiency model data; based on the efficiency model data, construct a delay model to obtain delay model data; combine the bit error rate model data, the efficiency model data and the delay model data, and output optimization target data and weight distribution data;

[0090] S33. Based on the optimization target data, weight allocation data and parameter candidate data, perform iterative parameter optimization to obtain iterative optimization data; verify the validity of the iterative optimization data to obtain verification result data; determine the final parameters based on the verification result data to obtain modulation parameter data and reliability index data.

[0091] This embodiment realizes the accurate optimization and reliability evaluation of modulation parameters by constructing a multi-objective optimization framework. By determining the parameter range based on interference feature data and interference confidence data, and constructing constraint conditions in combination with modulation strategy data, more accurate parameter candidate data can be generated, the optimization accuracy of modulation parameters is improved, and the effectiveness and reliability of modulation strategy are ensured. By constructing a bit error rate calculation model and a spectrum efficiency model, the bit error rate and spectrum efficiency under different parameter combinations can be accurately evaluated. By combining the bit error rate model data, efficiency model data and delay model data, a comprehensive evaluation of the optimization target can be achieved to ensure the best balance between bit error rate and spectrum efficiency. By constructing a delay model, the delay performance under different parameter combinations can be evaluated. Combined with the evaluation results of bit error rate and spectrum efficiency, the delay performance can be optimized under the premise of ensuring communication quality, and the response speed of the system and user experience can be improved. By performing iterative parameter optimization and verifying the validity of iterative optimization data, it can ensure that the modulation parameter data and reliability index data finally determined have high validity and reliability, which can effectively avoid errors and deviations in the parameter optimization process and improve the credibility of the optimization results. By performing iterative parameter optimization based on optimization target data, weight allocation data, and parameter candidate data, the modulation parameters can be dynamically adjusted to adapt to different interference environments and communication requirements, thereby enhancing the adaptability and stability of the system and ensuring stable operation in complex environments. This embodiment not only takes into account key performance indicators such as bit error rate, spectrum efficiency, and latency, but also ensures the reliability of the optimization results by introducing a multi-layer verification mechanism. The system can complete parameter optimization within 50ms, keep the bit error rate degradation within 1dB within the range of ±10dB of the signal-to-noise ratio change, and improve the spectrum efficiency by more than 30%.

[0092] According to one aspect of the present application, step S31 is further:

[0093] S311, based on the interference feature data and the interference confidence data, a dynamic parameter boundary mapping algorithm is used to construct a feature-parameter mapping matrix, calculate parameter sensitivity, and optimize the boundary function to obtain boundary mapping data; the boundary mapping data is divided into feasible domains to obtain parameter range data;

[0094] S312. Based on the parameter range data, a multi-dimensional constraint generation method is used to obtain constraint space data by constructing a constraint hyperplane, calculating the intersection area, and optimizing the constraint expression; adaptive meshing is applied to the constraint space data to obtain grid node data by dynamically adjusting the grid density, optimizing the grid structure, and calculating the node weight; constraint conditions are constructed according to the grid node data to obtain constraint condition data;

[0095] S313. Based on the constraint condition data and the pre-stored modulation strategy data, a parameter combination optimization algorithm is used to obtain the combination strategy data by constructing a combination evaluation function, calculating parameter correlation, and optimizing the combination strategy; the feasibility of the combination strategy data is verified, and parameter candidate data is output.

[0096] In one embodiment of the present application, a multidimensional constraint generation algorithm is as follows: C(x) = ∑(g_i(x)·exp(-λ_i·d(x, ΞΩ_i))); wherein C(x) is a constraint function; g_i(x) is a basic constraint; λ_i is a weight coefficient; d(x, ΞΩ_i) is the distance to the constraint boundary; ΞΩ_i is the boundary of the i-th constraint domain; the constraint hyperplane H(x) = W T ·x + b + ρ·R(x); W is the normal vector; b is the bias term; ρ is the regularization parameter; R(x) is the smoothness correction term; the grid density function D(x) = D_0·exp(-κ·||▽C(x)||); D_0 is the reference density; κ is the adjustment coefficient.

[0097] This embodiment realizes the dynamic construction and constraint optimization of parameter space. Through the dynamic parameter boundary mapping algorithm, the feature-parameter mapping relationship is established, and the parameter sensitivity analysis and boundary function optimization are introduced, so that the parameter range can be dynamically adjusted according to the interference characteristics and confidence. In the constraint condition generation link, a multidimensional constraint generation method is adopted, and the precise expression of the constraint space is realized by constructing a constraint hyperplane and adaptive meshing. In particular, in the parameter combination optimization process, the complexity of parameter optimization is effectively reduced by introducing a combined evaluation function and correlation analysis. This embodiment not only realizes the adaptive construction of the parameter space, but also improves the accuracy of parameter selection through constraint optimization. Tests show that the parameter optimization efficiency is improved by 60%, while ensuring the effectiveness of the optimization results.

[0098] According to one aspect of the present application, step S32 is further:

[0099] S321. Obtain parameter candidate data, construct a piecewise linear mapping matrix, and obtain mapping matrix data; perform channel state analysis on the mapping matrix data to obtain channel state data; establish a bit error rate prediction model based on the channel state data, and output the bit error rate model data.

[0100] S322. Obtain bit error rate model data and parameter candidate data, calculate the distance matrix between symbols, and obtain distance matrix data; perform subcarrier power allocation on the distance matrix data to obtain power allocation data; construct a spectrum utilization evaluation model based on the power allocation data, and output efficiency model data.

[0101] S323, obtaining parameter candidate data, establishing a queue state transfer matrix, and obtaining state transfer data; performing Markov prediction on the state transfer data to obtain prediction sequence data; calculating the delay probability distribution according to the prediction sequence data, and outputting the delay model data.

[0102] S324, obtaining bit error rate model data, efficiency model data and delay model data, constructing an objective function hyperplane, and obtaining hyperplane data; applying convex hull analysis to the hyperplane data to obtain convex hull boundary data; calculating the optimal weight combination based on the convex hull boundary data, and outputting the optimized target data and weight distribution data.

[0103] In one embodiment of the present application, the subcarrier power allocation method is: P(k) = [1 / λ - N_0 / |H(k)| 2 ] + ; Where P(k) is the power of the kth subcarrier; λ is the Lagrange multiplier; N_0 is the noise power spectral density; H(k) is the channel response; [x] + =max(0,x); power constraint∑P(k)≤P_total; spectral efficiencyη=∑log2(1+P(k)·|H(k)| 2 / N_0); channel capacity C = B·η; B is the system bandwidth.

[0104] Markov prediction optimization algorithm: P(s_t|s_{t-1}) =α·T(s_t|s_{t-1}) + (1-α)·H(s_t); where P(s_t|s_{t-1}) is the state transition probability; T(s_t|s_{t-1}) is the traditional transfer matrix; H(s_t) is the historical state distribution; α is the adaptive coefficient, α = 1 / (1+exp(-β·ΔE)); ΔE is the state energy difference; β is the temperature parameter; delay prediction value D(t) = ∑(s_i·P(s_i|s_{t-1})); s_t is the state vector at time t; i is the state index, i∈[1,M], and M is the number of states.

[0105] Enhanced state prediction algorithm: Q(s'|s, a) = (1-ε)·T(s'|s, a) + ε·M(s'); where Q(s'|s, a) is the state transition probability; T(s'|s, a) is the basic transfer matrix; M(s') is the memory term; ε is the mixing coefficient; s, s' are state vectors; a is the action vector; prediction sequence generation P(t) = ∑(π(a|s)·Q(s'|s, a)); π(a|s) is the strategy function; delay distribution D(t) = ∑(w_i·G_i(t, μ_i, σ_i)); G_i is the Gaussian component; μ_i, σ_i are distribution parameters.

[0106] The optimal weight combination is calculated as follows: W(x) = softmax(V·tanh(U·x + b))·exp(-λ·C(x)); where W(x) is the weight vector; V, U are transformation matrices; b is the bias vector; C(x) is the constraint violation degree; λ is the penalty coefficient; hyperplane construction H(x) = ∑(w_i·f_i(x)); f_i(x) is the objective function, w_i is the weight of the objective function; convex hull analysis B(x) = {y|y = ∑(α_i·x_i), ∑α_i = 1, α_i ≥ 0}; weight optimization target min{||W·x* -y*|| 2 + μ·||W||1}, α_i is the weight coefficient in the convex hull analysis, x_i is the data point in the convex hull analysis, y is the linear combination result in the convex hull analysis, W·x* - y* is the error between the linear combination result of the weight vector W and the data point x* and the target value y*, and μ is the regularization parameter in the weight optimization objective.

[0107] In a further embodiment, mutual information analysis may be performed on the distance matrix data to obtain information entropy data; and the information entropy data may be input into the spectrum utilization rate evaluation model.

[0108] This embodiment establishes a multi-objective performance evaluation and optimization framework. In the bit error rate modeling link, by constructing a piecewise linear mapping matrix and channel state analysis, the accurate prediction of the bit error rate is achieved, and the prediction error is controlled within 0.5dB. In terms of spectrum efficiency evaluation, the spectrum utilization efficiency is improved by calculating the optimal power allocation strategy in combination with the inter-symbol distance matrix and subcarrier power allocation. For the delay characteristics, the queue state transfer matrix and Markov prediction model are used to accurately evaluate the delay performance of the system. Finally, by constructing the objective function hyperplane and convex hull analysis, the optimal allocation of multi-objective weights is achieved. This embodiment not only takes into account the key performance indicators of the system, but also achieves the best balance of various performance indicators through weight optimization. In actual tests, the overall performance of the system is improved by more than 35%.

[0109] According to one aspect of the present application, step S33 is further:

[0110] S331. Obtain optimization target data, weight distribution data and parameter candidate data, apply adaptive particle swarm optimization algorithm, obtain particle position data by constructing a multi-objective fitness function, dynamically updating inertia weights, and adaptively adjusting learning factors; perform local search on particle position data, calculate local gradients, optimize search directions, and update optimal solutions to obtain iterative optimization data.

[0111] S332. Obtain iterative optimization data, adopt a multi-layer cross-validation method, obtain verification score data by constructing a verification data set, calculating performance indicators, and evaluating generalization ability; perform statistical analysis on the verification score data, calculate confidence intervals, evaluate stability, generate reliability indicators, and output verification result data.

[0112] S333. Obtain iterative optimization data and verification result data, apply parameter fine-tuning optimization algorithm, obtain fine-tuning result data by constructing fine-tuning objective function, calculating adjustment step size, and optimizing convergence criterion; conduct comprehensive evaluation on fine-tuning result data, and output modulation parameter data and reliability index data.

[0113] In one embodiment of the present application, the particle swarm optimization enhanced algorithm: v_i(t+1) = w(t)·v_i(t) +c_1·r_1·(p_i-x_i(t)) + c_2·r_2·(p_g-x_i(t)) + c_3·Q(x_i(t)); where v_i is the particle velocity; x_i is the particle position; p_i is the individual optimal position; p_g is the global optimal position; w(t) is the inertia weight, w(t)= w_max - (w_max-w_min)·(t / T) α ; c_1, c_2, c_3 are learning factors; r_1, r_2 are random numbers; Q(x) is the local search function; t is the number of iterations; T is the maximum number of iterations; fitness function F(x) = ∑(ω_k·f_k(x))·exp(-γ·V(x)); f_k is the objective function; ω_k is the weight coefficient; V(x) is the constraint violation degree.

[0114] Cross-validation optimization algorithm: E(θ) = (1 / K)·∑(L(D_k, f_θ)); where E(θ) is the validation error; K is the number of folds; D_k is the k-th fold data; f_θ is the parameterized model; L(·) is the loss function; validation score S(θ) = E(θ) + λ·Ω(θ); Ω(θ) is the regularization term; generalization performance index G = ∑(w_i·p_i·exp(-γ·σ_i 2 )); p_i is the verification pass rate; σ_i 2 is the performance variance; w_i is the indicator weight.

[0115] This embodiment constructs a parameter fine-tuning optimization and reliability assessment system. Through the adaptive particle swarm optimization algorithm, dynamic inertia weights and adaptive learning factors are introduced to improve the convergence speed and accuracy of parameter optimization. In the verification stage, a multi-layer cross-validation method is adopted to achieve reliability verification of the optimization results by constructing a verification data set and performance indicator evaluation. In particular, in the process of parameter fine-tuning, the stability of the final parameters is ensured by fine-tuning the objective function and the convergence criterion optimization. This embodiment not only realizes the precise optimization of parameters, but also ensures the reliability of the optimization results through a multi-layer verification mechanism. Tests show that the convergence speed of parameter optimization is improved by 50%, and the stability of the optimization results is improved by 40%.

[0116] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0117] S41, acquiring network status data in the communication system, constructing a network topology structure based on the network status data and modulation parameter data, and obtaining topology structure data; performing link analysis on the topology structure data to obtain link load data; reading service priority data from a memory, combining the link load data and reliability index data, calculating the node congestion state, and obtaining network load data;

[0118] S42, based on the network load data and modulation parameter data, construct a timing prediction model to obtain prediction model data; calculate the load change trend according to the prediction model data to obtain trend data; perform risk assessment on the trend data and output congestion prediction data;

[0119] S43. Based on the congestion prediction data, service priority data and modulation parameter data, a resource allocation model is constructed to obtain resource model data; the resource model data is optimized and calculated to obtain optimized allocation data; according to the optimized allocation data and reliability index data, communication time slots are divided to obtain resource allocation data and time slot allocation data.

[0120] This embodiment establishes a resource allocation framework based on network state perception, and realizes intelligent scheduling and optimal allocation of network resources. By acquiring network state data in the communication system and combining modulation parameter data to construct a network topology structure, the actual situation of the current network can be accurately reflected, the accuracy of the network topology structure can be effectively improved, and a reliable data basis can be provided for subsequent link analysis and resource allocation. By performing link analysis on the topology data and combining service priority data and reliability index data, the congestion state of the node can be accurately calculated, which can enhance the accuracy of link load analysis and help identify bottlenecks and potential problems in the network. Based on network load data and modulation parameter data, a timing prediction model is constructed to accurately predict the load change trend. By performing risk assessment on trend data, potential network congestion problems can be identified and dealt with in advance, and the reliability of load change trend prediction can be improved. By constructing a resource allocation model and optimizing the calculation of resource model data, efficient resource allocation can be achieved. In combination with congestion prediction data, service priority data and modulation parameter data, communication time slots can be divided to ensure the rationality and effectiveness of resource allocation, optimize resource allocation strategies, and improve the overall performance of the network. Through comprehensive analysis of the optimized allocation data and reliability index data, the stability and reliability of resource allocation and communication time slot division can be ensured, which effectively improves the stability of the system and ensures stable operation under high load and complex environment. This embodiment can not only deal with network congestion in advance, but also improve the spectrum utilization efficiency through time slot reuse. The system can maintain stable service quality within the range of ±30% of network load fluctuation, improve resource utilization by more than 40%, and control network delay jitter within ±5ms.

[0121] According to one aspect of the present application, step S41 is further:

[0122] S411. Obtain network status data and modulation parameter data, apply a dynamic topology reconstruction algorithm, obtain topology data by constructing an adjacency relationship matrix, calculating link weights, and optimizing the topology structure; perform connectivity analysis on the topology data, obtain connectivity feature data by calculating node degree distribution, evaluating path redundancy, and optimizing structural features; establish a network topology map based on the connectivity feature data, and output the topology map data.

[0123] S412. Obtain topology map data and adopt a multi-dimensional link evaluation method to obtain link status data by constructing a link quality matrix, calculating transmission delay, and evaluating link stability; apply adaptive threshold analysis to the link status data, dynamically adjust the threshold, calculate the load rate, and evaluate the link performance to obtain link load data.

[0124] S413, obtaining link load data and stored service priority data, applying a hierarchical congestion assessment algorithm, specifically by constructing a multi-layer assessment model, calculating the correlation between layers, and optimizing the assessment index, to obtain congestion assessment data; performing node classification on the congestion assessment data, and outputting network load data.

[0125] In one embodiment of the present application, a dynamic topology reconstruction algorithm: A(i, j) = exp(-d(i, j) / σ)·(1+ρ·Q(i, j)); wherein A(i, j) is an adjacency matrix element; d(i, j) is a node distance; σ is an attenuation parameter; Q(i, j) is a link quality factor; ρ is a weight coefficient; link weight W(i, j) = (c_1·BW(i, j) + c_2·SNR(i, j))·exp(-λ·L(i, j)); BW is bandwidth; SNR is signal-to-noise ratio; L is load; c_1, c_2 are balancing coefficients; topology optimization target T = min{∑∑(W(i, j)·x(i, j))}, st ∑x(i, j) ≥ k; k is a connectivity constraint.

[0126] Multi-dimensional link quality assessment method: Q(l) = ω_1·SNR(l) + ω_2·STB(l) + ω_3·CAP(l); where Q(l) is the link quality indicator; SNR(l) is the signal-to-noise ratio function; STB(l) is the stability function, and STB(l) = exp(-σ_t 2 / 2);σ_t 2 is the time-varying variance; CAP(l) is the capacity function; ω_1, ω_2, ω_3 are weight coefficients; link load index L(l) = ∑(f_i·τ_i) / T; f_i is the service flow; τ_i is the transmission time; T is the observation period.

[0127] Congestion assessment algorithm: C(n) = ∑(v_i·B_i(n) / B_max)·exp(β·Q(n) / Q_max); where C(n) is the node congestion; v_i is the service type weight; B_i(n) is the buffer occupancy; B_max is the maximum buffer; Q(n) is the queue length; Q_max is the maximum queue length; β is the congestion sensitivity; node classification index J(n) = [C(n), R(n), D(n)]; R(n) is the resource utilization; D(n) is the delay factor.

[0128] This embodiment realizes dynamic reconstruction and link evaluation of network topology. Through the dynamic topology reconstruction algorithm, the adjacency matrix and link weight optimization are adopted to realize real-time perception and optimization of network topology. In terms of connectivity analysis, the structural characteristics of the network are accurately grasped through node degree distribution calculation and path redundancy evaluation. For link evaluation, a multi-dimensional link evaluation method is adopted, and accurate evaluation of link status is realized through link quality matrix and adaptive threshold analysis. This embodiment not only realizes dynamic optimization of network topology, but also improves the robustness of the network through multi-dimensional evaluation, and can maintain stable network performance when the network load fluctuates by ±30%.

[0129] According to one aspect of the present application, step S42 is further:

[0130] S421. Obtain network load data and modulation parameter data, apply deep time series modeling methods, obtain time series model data by building an attention mechanism, extracting time features, and optimizing model parameters; perform pattern recognition on the time series model data, and obtain prediction model data by analyzing periodic features, extracting trend components, and evaluating random components.

[0131] S422. Obtain prediction model data and use a multi-scale trend analysis algorithm to obtain trend analysis data by constructing a scale transformation matrix, calculating trend coefficients, and optimizing prediction accuracy; extract fluctuation characteristics from trend analysis data, calculate fluctuation amplitudes, evaluate change rates, and predict development trends to obtain trend data.

[0132] S423. Obtain trend data and apply a hierarchical risk assessment method to obtain risk level data by constructing a risk assessment matrix, calculating risk levels, and optimizing warning thresholds; conduct a comprehensive assessment of risk level data, calculate risk probabilities, assess impact levels, generate warning information, and output congestion prediction data.

[0133] In one embodiment of the present application, the deep time series modeling algorithm is: y(t) = ∑(α_i·h_i)·W + b; where y(t) is the predicted output; h_i is the hidden layer state; α_i is the attention weight, α_i = softmax(e_i); e_i = v T ·tanh(W_1·h_i + W_2·s_{t-1}); W_1, W_2 are weight matrices; v is the parameter vector; s_{t-1} is the state at the previous moment; W is the output weight; b is the bias term; time series feature extraction Z(t) = LSTM(x(t))·A(t); A(t) is the attention matrix; LSTM is the long short-term memory network function.

[0134] Trend coefficient calculation method: C(τ) = [∑(x(t+τ)-x(t))2 ] H(τ) ; Where C(τ) is the trend coefficient; x(t) is the load sequence; τ is the time scale; H(τ) is the Hurst exponent, H(τ) = log(R / S) / log(τ); R is the range, S is the standard deviation; trend prediction function F(t) = C(τ)·exp(ω·t)·(1+ρ·sin(2πt / T)); ω is the growth rate; ρ is the periodic fluctuation intensity; T is the period parameter; t is the time variable.

[0135] Trend prediction method: T(t) = β·∑(α_i·x(ti)) + (1-β)·F(t); where T(t) is the trend prediction value; x(t) is the load sequence; α_i is the autoregressive coefficient; β is the mixing factor; F(t) is the Fourier prediction term, F(t) = ∑(A_k·cos(ω_k·t+φ_k)); A_k is the amplitude; ω_k is the frequency; φ_k is the phase; prediction credibility C(t) = exp(-γ·|T(t)-x(t)|); γ is the sensitivity parameter.

[0136] Warning generation algorithm: H(x) = -∑(p_i·log(p_i))·exp(κ·V(x)); where H(x) is risk entropy; p_i is risk probability distribution; V(x) is volatility measure; κ is sensitivity parameter; risk level R(t) = [H(x), D(x), I(x)]; D(x) is risk persistence; I(x) is impact; warning indicator W(t) = α·R(t) + (1-α)·F(t); F(t) is prediction factor; α is mixed weight.

[0137] This embodiment constructs a network load prediction and risk assessment system. Through the deep time series modeling method, the attention mechanism and time feature extraction are introduced to achieve accurate prediction of network load. In the trend analysis link, a multi-scale trend analysis algorithm is adopted to accurately capture the load change trend through scale transformation matrix and trend coefficient calculation. For risk assessment, early warning of network congestion is achieved through hierarchical risk assessment method and early warning threshold optimization. This embodiment not only improves the accuracy of network load prediction, but also enhances the predictability of the system through the risk assessment mechanism. The prediction accuracy rate reaches 92%, and the average early warning time is 200ms.

[0138] According to one aspect of the present application, step S43 is further:

[0139] S431. Obtain congestion prediction data and service priority data, construct a resource competition graph, and obtain competition graph data; perform vertex coloring on the competition graph data to obtain coloring result data; establish a resource conflict resolution model based on the coloring result data, and output resource model data.

[0140] S432, obtaining resource model data and modulation parameter data, constructing a multi-dimensional scoring matrix, and obtaining scoring matrix data; applying spectral clustering analysis to the scoring matrix data to obtain clustering feature data; calculating a resource allocation sequence according to the clustering feature data, and outputting optimized allocation data.

[0141] S433, obtain optimized allocation data, establish a time slot interference spectrum, and obtain interference spectrum data; perform lobe analysis on the interference spectrum data to obtain lobe characteristic data; construct a time slot reuse pattern based on the lobe characteristic data, and output the time slot pattern data.

[0142] S434, obtain time slot mode data and optimized allocation data, perform resource group mapping, and obtain group mapping data; apply a dynamic time slot allocation algorithm to the group mapping data to obtain a time slot allocation result; generate a final scheduling plan based on the time slot allocation result, and output resource allocation data and time slot allocation data.

[0143] In one embodiment of the present application, the spectral clustering optimization algorithm: L = D -1 / 2 ·(W +λ·R)·D -1 / 2 ; Where L is the normalized Laplace matrix; D is the degree matrix; W is the similarity matrix, Wij = exp(-||x_i-x_j|| 2 / 2σ 2 ); R is the regularization matrix, R = UU T ;U is the eigenvector matrix; λ is the regularization parameter; clustering evaluation index Q = tr(Y T ·L·Y) / tr(Y T ·D·Y); Y is the indicator matrix; tr is the matrix trace; the optimization target min Q, st Y T ·D·Y = I, where I is the identity matrix.

[0144] Resource allocation optimization algorithm: U(r) = ∑(w_i·log(1+γ_i·r_i)); where U(r) is the utility function; r_i is the resource allocation amount; γ_i is the channel gain; w_i is the service weight; resource allocation constraint R(t) = {r|∑r_i ≤R_max, r_i ≥ 0}; scoring matrix M(i, j) = S(i, j)·exp(-μ·C(i, j)); S(i, j) is the service quality score; C(i, j) is the resource cost; μ is the cost coefficient; optimization target max{U(r)}, st r∈R(t).

[0145] Lobe analysis method: B(θ, f) = ∑(w_n·A(θ, f, n)·exp(-j·2π·d·sin(θ)·n / λ)); where B(θ, f) is the lobe response function; θ is the azimuth; f is the frequency; w_n is the weighting coefficient; A(θ, f, n) is the array element response; d is the array element spacing; λ is the wavelength; n is the array element index; lobe eigenvector V = [G_m, θ_3dB, SLL, Z_n]; G_m is the main lobe gain; θ_3dB is the half-power beamwidth; SLL is the sidelobe level; Z_n is the zero point position; feature score S = η·G_m -ξ·SLL, η and ξ are weighting coefficients.

[0146] Time slot allocation optimization algorithm: I(t, f) = ∑(P_k·G_k(t, f)·exp(-j·2π·f·τ_k)); where I(t, f) is the time-frequency interference function; P_k is the transmit power; G_k(t, f) is the time-frequency gain; τ_k is the delay; time slot reuse function R(i, j) = (1-α)·exp(-||t_i-t_j|| 2 / σ_t 2 ) + α·exp(-||f_i-f_j|| 2 / σ_f 2 );t_i,fi are time-frequency coordinates;σ_t,σ_f are time-frequency resolution parameters;time slot allocation matrix S(t) = arg min{∑(w_k·C_k(t)) +λ·∑∑(R(i,j)·x_i·x_j)};C_k(t) is the conflict cost;w_k is the service weight;x_i is the allocation indicator variable.

[0147] Dynamic scheduling algorithm: G(i, j) = B(i, j)·exp(-η·|t_i-t_j|)·(1+θ·P(i, j)); where G(i, j) is the packet mapping function; B(i, j) is the service relevance; t_i, t_j are time tags; P(i, j) is the priority difference; η is the time decay coefficient; θ is the priority weight; scheduling function S(t) = arg max{∑(w_k·G(i_k, j_k))}, stC(t) ≤ C_max; C(t) is the resource consumption; w_k is the scheduling weight, and C_max is the maximum resource consumption.

[0148] This embodiment realizes the intelligent allocation and time slot optimization of network resources. The resource conflict problem is solved by constructing a resource competition graph and processing vertex coloring. In the resource allocation optimization link, a multi-dimensional scoring matrix and spectral clustering analysis are used to achieve the optimal allocation of resources. For time slot division, an efficient time slot reuse mode is established through time slot interference spectrum and lobe analysis. Finally, the generation of the final scheduling plan is realized through the dynamic time slot allocation algorithm and resource grouping mapping. This embodiment not only improves the resource utilization efficiency, but also reduces the system delay through time slot optimization, the resource utilization rate is increased by 45%, and the average delay is reduced by 30%, which provides a guarantee for the efficient operation of the system.

[0149] In the receiving and demodulating process: the terminal radio accesses the wireless communication network, and the terminal radio receives the signal from the main station radio. The RF signal received by the antenna of the terminal radio is demodulated by the RF sub-board for FM reception, and converted by ADC. The data received by the MODEM enters the modulation mode matching after energy detection and carrier synchronization. Multiple modulation modes are carried out simultaneously, and the corresponding modulation mode is matched, and the debugging mode information is recorded. After the signal is demodulated by the modulation mode, due to the distortion of the channel, it will cause inter-code interference. Therefore, at the demodulation end, channel equalization must be performed, and then sent to the decision device, and framed and output after parallel-serial conversion, and the recorded modulation mode is transmitted to the terminal; in the modulation transmission process: after the terminal processes the corresponding service, the terminal transmits the service response data and the modulation mode to the MODEM through the serial port. The MODEM processes the service response data according to the modulation mode, converts it through DAC, and sends the signal to the main station radio through the RF sub-board.

[0150] The present invention realizes the intelligentization and precision of radio station adaptive modulation by constructing a complete technical chain of "signal acquisition-interference analysis-parameter optimization-resource allocation". The multi-dimensional signal feature extraction, probability statistical modeling, multi-objective optimization and network resource scheduling technologies are organically integrated to form a closed-loop adaptive modulation system. In the feature extraction link, the accuracy of feature expression is improved through time-frequency analysis and feature fusion; in the interference analysis link, the interference characteristics are accurately characterized by the probability statistical model; in the parameter optimization link, the optimal balance of system performance is achieved through multi-objective optimization; in the resource allocation link, the resource utilization efficiency is improved through predictive scheduling. The present invention not only improves the adaptability of the system in complex electromagnetic environments, but also ensures the reliability of the system through a multi-layer verification mechanism. The test results show that compared with the traditional method, the present invention improves the key indicators such as anti-interference ability, spectrum efficiency and network throughput by 40%, 35% and 45% respectively, and controls the system response delay within 100ms, providing a new solution for the development of radio station adaptive modulation technology.

[0151] 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 adaptive modulation of a terminal station, characterized in that: The steps include: S1. Obtaining original signal data from a receiving antenna, and obtaining pre-processed signal data through feature extraction and analysis processing; performing feature domain conversion on the pre-processed signal data to obtain feature matrix data; The characteristic matrix data is fused with the pre-stored historical channel characteristic data to generate channel characteristic data; S2. Based on the channel characteristic data and the pre-stored interference pattern data, perform interference characteristic analysis to obtain initial interference characteristic data; perform probability statistical analysis on the initial interference characteristic data to obtain interference characteristic data and probability distribution data; perform confidence calculation based on the interference characteristic data and the probability distribution data, and output interference confidence data; S3, constructing a parameter space based on the interference feature data, the interference confidence data and the pre-stored modulation strategy data to obtain parameter candidate data; Perform multi-objective analysis on parameter candidate data to obtain optimization target data and weight distribution data; Perform parameter optimization calculation according to optimization target data and weight distribution data, and output modulation parameter data and reliability index data; S4. Obtain network status data in the communication system, perform network topology analysis based on modulation parameter data, reliability index data, network status data and pre-stored service priority data to obtain network load data; perform trend forecasting on network load data to obtain congestion forecast data; optimize resource allocation based on congestion forecast data, service priority data and modulation parameter data, and output resource allocation data and time slot allocation data.

2. The method for adaptive modulation of a terminal station according to claim 1, characterized in that: Step S1 is further as follows: S11, obtaining original signal data from a receiving antenna, and obtaining time-frequency data through time-frequency transformation; performing energy distribution calculation on the time-frequency data to obtain energy spectrum data; extracting feature points according to the energy spectrum data to obtain feature point data; Use the feature point data to perform noise reduction on the signal and output preprocessed signal data; S12, extracting instantaneous frequency characteristics based on the preprocessed signal data to obtain frequency characteristic data; Perform envelope analysis on the preprocessed signal data to obtain amplitude characteristic data; Analyze the phase change of the preprocessed signal data to obtain phase characteristic data; combine the frequency characteristic data, amplitude characteristic data and phase characteristic data to generate characteristic matrix data; S13, converting the feature matrix data into a three-dimensional feature representation to obtain initial feature data; reading historical channel feature data from a memory, and performing feature fusion on the initial feature data and the historical channel feature data to obtain fused feature data; The fused feature data is normalized and the channel feature data is output.

3. The method for adaptive modulation of a terminal station according to claim 2, characterized in that: Step S2 is further as follows: S21, based on the channel characteristic data, extract the main characteristic component to obtain the main characteristic data; read the interference pattern data from the memory, and based on the interference pattern data, analyze the periodic characteristics in the main characteristic data to obtain the periodic characteristic data; detect the sudden change in the main characteristic data to obtain the burst characteristic data; combine the periodic characteristic data and the burst characteristic data to output the initial interference characteristic data; S22. Based on the initial interference feature data, construct a multidimensional statistical model to obtain statistical model data; based on the statistical model data, calculate the probability distribution of various features to obtain probability distribution data; Based on the probability distribution data, the interference features are reclassified to obtain interference feature data; S23. Based on the interference feature data and the probability distribution data, calculate the confidence interval of the feature to obtain confidence interval data; perform stability assessment on the confidence interval data to obtain stability assessment data; calculate the confidence score based on the stability assessment data, and output the interference confidence data.

4. The method for adaptive modulation of a terminal station according to claim 3, characterized in that: Step S3 is further as follows: S31, based on the interference characteristic data and the interference confidence data, determining the parameter range to obtain parameter range data; reading the modulation strategy data from the memory, and constructing the constraint condition based on the modulation strategy data and the parameter range data to obtain the constraint condition data; Based on the constraint condition data, parameter combinations are generated to obtain parameter candidate data; S32, based on the parameter candidate data, construct a bit error rate calculation model to obtain bit error rate model data; based on the bit error rate model data, construct a spectrum efficiency model to obtain efficiency model data; based on the efficiency model data, construct a delay model to obtain delay model data; combine the bit error rate model data, the efficiency model data and the delay model data, and output optimization target data and weight distribution data; S33, performing iterative parameter optimization based on the optimization target data, the weight allocation data and the parameter candidate data to obtain iterative optimization data; The validity of the iterative optimization data is verified to obtain the verification result data; based on the verification result data, the final parameters are determined to obtain the modulation parameter data and reliability index data.

5. The method for adaptive modulation of a terminal station according to claim 4, characterized in that: Step S4 is further as follows: S41, acquiring network status data in the communication system, constructing a network topology structure based on the network status data and modulation parameter data, and obtaining topology structure data; Perform link analysis on topology data to obtain link load data; Read the service priority data from the memory, combine the link load data and the reliability index data, calculate the node congestion status, and obtain the network load data; S42, constructing a time series prediction model based on the network load data and the modulation parameter data to obtain prediction model data; calculating the load change trend according to the prediction model data to obtain trend data; Conduct risk assessment on trend data and output congestion prediction data; S43. Based on the congestion prediction data, service priority data and modulation parameter data, a resource allocation model is constructed to obtain resource model data; the resource model data is optimized and calculated to obtain optimized allocation data; according to the optimized allocation data and reliability index data, communication time slots are divided to obtain resource allocation data and time slot allocation data.

6. The method for adaptive modulation of a terminal station according to claim 5, characterized in that: Step S11 is further as follows: S111, obtaining original signal data from a receiving antenna, performing segmented windowing processing to obtain window data; performing short-time Fourier transform on the window data to obtain initial time-frequency data; Performing phase compensation on the initial time-frequency data to obtain compensated initial time-frequency data; Performing resolution evaluation on the compensated initial time-frequency data to obtain resolution parameter data; Adaptively adjust the window according to the resolution parameter data and output the optimized time-frequency data; S112, based on the optimized time-frequency data, calculating the energy density of each time-frequency point to obtain energy density data; Perform two-dimensional smoothing filtering on the energy density data to obtain smoothed energy data; perform normalization processing on the smoothed energy data to obtain energy spectrum data; S113, based on the energy spectrum data, performing adaptive threshold segmentation to obtain threshold segmentation data; performing local extreme value detection on the threshold segmentation data to obtain extreme value point data; performing feature clustering analysis based on the extreme value point data to obtain feature point data; S114, constructing an adaptive filter group based on the feature point data and the optimized time-frequency data to obtain filter coefficient data; Performing noise reduction processing on the signal according to the filter coefficient data to obtain noise reduction data; The denoised data is reconstructed in the time domain and the preprocessed signal data is output.

7. The method for adaptive modulation of a terminal station according to claim 5, characterized in that: Step S12 is further as follows: S121, constructing a Hilbert transform basis based on the preprocessed signal data to obtain transform basis data; applying the transform basis data to process the preprocessed signal data to obtain analytical signal data; performing instantaneous frequency estimation on the analytical signal data to obtain frequency characteristic data; S122, applying adaptive segmentation processing based on the preprocessed signal data to obtain segmented signal data; performing envelope detection on the segmented signal data to obtain envelope data; Extract statistical features from the envelope data to obtain amplitude feature data; S123, calculating a phase difference sequence based on the analytical signal data to obtain phase difference data; Performing phase unwrapping on the phase difference data to obtain unwrapped phase data; performing phase correction on the unwrapped phase data to output phase characteristic data; S124. Based on the frequency characteristic data, the amplitude characteristic data and the phase characteristic data, a characteristic combination matrix is ​​constructed to obtain combination matrix data; characteristic correlation analysis is performed on the combination matrix data to obtain correlation data; and feature selection is performed based on the correlation data to obtain characteristic matrix data.

8. The method for adaptive modulation of a terminal station according to claim 5, characterized in that: Step S21 is further as follows: S211, performing singular value decomposition on the channel feature data to obtain singular value data; calculating the energy contribution rate of the singular value data to obtain energy distribution data; performing feature projection according to the energy distribution data to obtain main feature data; S212, calculating the autocorrelation sequence based on the main feature data to obtain correlation sequence data; Applying hyperbolic transformation to the relevant sequence data, mapping the periodic features to the straight line features, and obtaining the transform domain data; performing parameter estimation on the transform domain data, and obtaining the periodic parameter data; Perform feature reconstruction according to the periodic parameter data to obtain periodic feature data; S213, based on the main feature data, construct a multi-scale discrete wavelet basis to obtain wavelet decomposition data; perform adaptive threshold processing on the high-frequency components of the wavelet decomposition data to obtain threshold detection data; identify mutation points according to the threshold detection data, and output burst feature data; S214, constructing a feature mapping relationship based on the period feature data, the burst feature data and the pre-stored interference pattern data to obtain feature association data; Perform pattern matching on the feature association data to obtain initial interference feature data.

9. The method for adaptive modulation of a terminal station according to claim 5, characterized in that: Step S22 is further as follows: S221, performing principal component analysis on the initial interference feature data to obtain dimension reduction feature data; constructing a multi-layer kernel density estimation model based on the dimension reduction feature data to obtain kernel density data; optimizing the kernel density data using an adaptive bandwidth selection algorithm to obtain optimized kernel density data; performing probability density estimation on the optimized kernel density data to obtain statistical model data; S222, applying a preconfigured segmented Gamma mixed model to the statistical model data to obtain mixed model data; performing distribution parameter estimation on the mixed model data to obtain parameter estimation data; According to the parameter estimation data, the conditional probability distribution is calculated to obtain the probability distribution data; S223. Based on the initial interference feature data and probability distribution data, a dynamic clustering threshold algorithm is used to obtain cluster boundary data; feature reclassification is performed according to the cluster boundary data, and interference feature data is output.

10. The method for adaptive modulation of a terminal station according to claim 5, characterized in that: Step S31 is further as follows: S311, based on the interference feature data and the interference confidence data, a dynamic parameter boundary mapping algorithm is used to obtain boundary mapping data; the boundary mapping data is divided into feasible domains to obtain parameter range data; S312, based on the parameter range data, using a multi-dimensional constraint generation method to obtain constraint space data; applying adaptive grid division to the constraint space data to obtain grid node data; Construct constraint conditions according to grid node data to obtain constraint condition data; S313, based on the constraint condition data and the pre-stored modulation strategy data, a parameter combination optimization algorithm is used to obtain combination strategy data; feasibility verification is performed on the combination strategy data, and parameter candidate data is output.

Images