A method, device and medium for identifying and demodulating a non-orthogonal multicarrier shortwave signal

Abstract

This application discloses a method, device, and medium for identifying and demodulating non-orthogonal multi-carrier shortwave signals. The method includes: acquiring a target signal; extracting features from the target signal to determine multi-level feature information; inputting the multi-level feature information into a pre-set multi-model fusion classifier for signal type identification to obtain a trained multi-model fusion classifier; the multi-model fusion classifier outputs the signal type and confidence level; identifying non-orthogonal multi-carrier shortwave signals across the entire frequency band to obtain identification results; performing digital channelization processing based on the identification results to determine the raw data suitable for narrowband demodulation; calling the corresponding demodulation algorithm library according to the signal type to demodulate the raw data; and performing decision optimization based on the confidence level. This application combines multi-feature extraction with multi-model fusion to improve the accuracy and robustness of identification and demodulation.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a method, device and medium for identifying and demodulating non-orthogonal multicarrier shortwave signals. Background Technology

[0002] Shortwave communication operates in the 3-30MHz frequency band and is widely used in emergency response and military fields due to its long-distance transmission capabilities. To improve spectral efficiency and anti-interference capabilities, non-orthogonal multicarrier technologies, such as FBMC and GFDM, have been introduced into shortwave communication. However, the detection of such signals faces many challenges: on the one hand, the communication channel environment is harsh, with severe time-varying, frequency-varying, multipath, and Doppler effects, resulting in blurred signal characteristics. The inherent interference of non-orthogonal signals is coupled with channel interference, rendering traditional identification methods ineffective. On the other hand, such non-orthogonal multicarrier signals typically lack cyclic prefixes or employ non-rectangular pulse shaping, resulting in weak time-domain autocorrelation characteristics and low frequency-domain sidelobe leakage. Existing methods based on cyclic prefix detection or power spectrum analysis cannot effectively identify such signals. Under low signal-to-noise ratio and strong interference conditions, the time-frequency grid structure of non-orthogonal signals is blurred, and the blind estimation accuracy of key parameters such as subcarrier spacing and symbol rate is low, leading to subsequent demodulation failure. Summary of the Invention

[0003] To address the aforementioned issues, this application proposes a method for identifying and demodulating non-orthogonal multi-carrier shortwave signals, comprising: acquiring a target signal; extracting features from the target signal to determine multi-level feature information, wherein the multi-level feature information includes at least one or more combinations of waveform similarity-based features, correlation-based features, time-frequency domain statistical features, and nonlinear statistical features; inputting the multi-level feature information into a pre-set multi-model fusion classifier for signal type identification to obtain a trained multi-model fusion classifier; outputting the signal type and confidence level based on the multi-model fusion classifier; identifying non-orthogonal multi-carrier shortwave signals across the entire frequency band to obtain identification results; performing digital channelization processing based on the identification results to determine raw data suitable for narrowband demodulation; calling the corresponding demodulation algorithm library to demodulate the raw data according to the signal type; and performing decision optimization based on the confidence level.

[0004] In one example, the multi-level feature information is input into a pre-set multi-model fusion classifier for signal type identification to obtain a trained multi-model fusion classifier. Specifically, this includes: dividing the multi-level feature information into a training set, a validation set, and a test set; using the training set to train multiple pre-set base classifiers, including convolutional neural networks, support vector machines, random forests, and K-nearest neighbors; adjusting the hyperparameters of the multiple base classifiers according to the validation set; and fusing the decision results of the multiple base classifiers to output the final signal type and the confidence level of the judgment.

[0005] In one example, a non-orthogonal multi-carrier shortwave signal across the entire frequency band is identified to obtain an identification result. Based on this result, digital channelization processing is performed, specifically including: continuous frame processing of the baseband data across the entire frequency band and performing a Fast Fourier Transform to obtain instantaneous power spectrum and phase information; adaptive noise floor estimation of the instantaneous power spectrum to locate energy-abnormal frequency bands; multi-frame depth analysis within the energy-abnormal frequency bands, and signal confirmation and identification using a multi-model fusion classifier to obtain an identification result; and driving the corresponding digital channelizer based on the identification result to adjust the center frequency and bandwidth of the digital channelizer to match the identified signal frequency band, thereby filtering out the narrowband signal where the target signal is located and down-converting it to zero intermediate frequency.

[0006] In one example, the original data is demodulated by calling the corresponding demodulation algorithm library according to the signal type, and decision optimization is performed based on the confidence level. Specifically, this includes: determining the corresponding demodulation algorithm in a pre-set demodulation algorithm library according to the signal type, and comparing the confidence level with a pre-set threshold; if the confidence level is higher than the threshold, the corresponding demodulator is directly started; if the confidence level is lower than the threshold, a re-identification mechanism is triggered; and parameter blind estimation and demodulation are performed on the baseband data after digital channelization processing. The steps of parameter blind estimation and demodulation include pilot and subcarrier frequency blind estimation, modulation mode blind identification, distributed cooperative symbol timing synchronization, pilot-based cross-frame phase continuous compensation, signal demodulation and error verification.

[0007] In one example, blind parameter estimation and demodulation of baseband data after digital channelization processing specifically includes: blind estimation of key frequency parameters of the target signal, the objects of which include pilot frequencies and the center frequencies of all subcarriers; channelization separation based on the subcarrier frequencies to obtain time-domain data on the subcarriers, and blind identification of modulation schemes based on the time-domain data; performing distributed cooperative symbol timing synchronization to determine the optimal sampling point; performing cross-frame phase continuous compensation based on pilots to eliminate residual frequency offset and phase noise; and demodulating the signal based on the obtained accurate parameters and performing error verification.

[0008] In one example, the method further includes: performing a peak search on the power spectrum of the signal to identify frequency points with significant energy; performing multi-level correlation verification on the identified frequency points based on prior knowledge of the fixed frequency spacing between the pilot and subcarriers and between subcarriers in a non-orthogonal multicarrier signal; and when there are multiple candidate frequency point pairs, iteratively tightening the verification conditions and combining power magnitude analysis to eliminate interference, thereby determining the pilot frequency and the center frequency of all subcarriers.

[0009] In one example, the method further includes: performing channelization separation on each subcarrier to obtain time-domain data on the subcarrier; calculating the phase difference sequence between adjacent symbols of the subcarrier data and statistically analyzing the distribution of the phase difference within a preset range; and determining the modulation scheme based on the number and spacing of peaks of statistical features to determine the peak interval.

[0010] In one example, acquiring a target signal and extracting features from it specifically includes: preprocessing the target signal, the preprocessing steps of which include signal truncation, segmentation, and variable sampling normalization; extracting waveform similarity-based features from the preprocessed target signal, including Euclidean distance, dynamic time warping, derivative dynamic time warping, and weighted dynamic time warping; extracting correlation-based features from the preprocessed target signal, including autocorrelation function, partial autocorrelation function, Pearson correlation coefficient, and sliding window correlation coefficient; extracting time-frequency domain statistical features from the preprocessed target signal, including zero-crossing rate, amplitude standard deviation, skewness, kurtosis, and spectral centroid; and extracting nonlinear statistical features from the preprocessed target signal, including the mean of bispectral invariants, the standard deviation of bispectral invariants, phase entropy, normalized bispectral entropy, normalized bispectral squared entropy, and logarithmic bispectral amplitude.

[0011] On the other hand, this application also proposes an identification and demodulation device for non-orthogonal multi-carrier shortwave signals, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the identification and demodulation device for non-orthogonal multi-carrier shortwave signals to perform: the method described in any of the examples above.

[0012] On the other hand, this application also proposes a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to be the method described in any of the examples above.

[0013] This application comprehensively characterizes signal properties and improves recognition accuracy by extracting multi-level feature information, including waveform similarity, correlation, time-frequency domain statistical features, and nonlinear statistical features. Secondly, it employs a multi-model fusion classifier, integrating the advantages of multiple machine learning models. By fusing decision results through an ensemble learning strategy, it significantly improves the robustness and accuracy of signal type recognition. Furthermore, it adopts a two-stage processing strategy of "coarse detection + fine recognition," effectively reducing computational resource consumption. Simultaneously, by combining energy detection and intelligent pattern recognition, it reduces the probability of false alarms and missed alarms. Finally, pilot-based cross-frame phase continuous compensation and adaptive blind demodulation techniques ensure reliable demodulation in low signal-to-noise ratio and strong interference environments, further enhancing the system's practicality and adaptability. These features give this application significant advantages in complex shortwave channel environments. Attached Figure Description

[0014] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a method for identifying and demodulating non-orthogonal multi-carrier shortwave signals according to an embodiment of this application. Figure 2 This is a structural diagram of the multi-model fusion classifier in the embodiments of this application; Figure 3 This is a schematic diagram of the parallel processing architecture of the base classifier in an embodiment of this application; Figure 4 This is a schematic diagram of the stacked generalization fusion mechanism in the embodiments of this application; Figure 5 This is a schematic diagram of the weighted voting fusion mechanism in the embodiments of this application; Figure 6 This is a flowchart illustrating the full-band non-orthogonal multi-carrier shortwave signal detection, identification, and digital channelization process in the embodiments of this application. Figure 7 This is a schematic diagram of the adaptive blind demodulation process in an embodiment of this application; Figure 8 This is a schematic diagram of a non-orthogonal multi-carrier shortwave signal identification and demodulation device according to an embodiment of this application. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0017] like Figure 1 As shown, in order to solve the above problems, this application provides a method for identifying and demodulating non-orthogonal multi-carrier shortwave signals, the method comprising: S101. Acquire the target signal and perform feature extraction on the target signal to determine multi-level feature information. The multi-level feature information includes at least one or more combinations of waveform similarity-based features, correlation-based features, time-frequency domain statistical features, and nonlinear statistical features.

[0018] First, target signal feature information is extracted for training and modeling. The target signal is processed, including signal truncation, segmentation, and variable sampling normalization. Then, multi-level feature information capable of characterizing the intrinsic differences between different non-orthogonal multicarrier signals is extracted from the preprocessed signal. This feature information is not limited to a single type, but rather includes a combination of at least one or more of the following types.

[0019] In one embodiment, waveform similarity features include: calculating a series of dynamic time adjustment distances or editing distances between the reconnaissance signal and the target signal to overcome the effects of frequency and time axis distortion. Corresponding features include Euclidean distance, dynamic time warping, derivative dynamic time warping, and weighted dynamic time warping. These features control the "elasticity" of alignment through weights to avoid excessive time axis distortion. The choice of weighting function directly affects the calculation results and needs to be adjusted according to the application scenario. The waveform similarity feature information is shown in Table 1 below.

[0020] Table 1. Characteristics of Waveform Similarity

[0021] In one embodiment, correlation characteristics are used to capture the structural properties of the signal, such as calculating the autocorrelation function, partial autocorrelation function, Pearson coefficient, or correlation coefficient between the reconnaissance signal and the target signal. The correlation characteristics are shown in Table 2 below.

[0022] Table 2. Characteristics of Correlation

[0023] In one embodiment, the sliding window correlation coefficient dual-signal time-varying correlation analysis O(wp) non-stationary signal analysis and dynamic connection time-frequency domain statistical features include statistical features extracted from the time domain, such as zero-crossing rate and higher-order moments; and spectral features extracted from the frequency domain transformation, such as spectral centroid and spectral entropy. The time-frequency domain statistical features are shown in Table 3 below.

[0024] Table 3 Time-Frequency Domain Statistical Characteristics

[0025] In one embodiment, nonlinear statistical characteristics include the mean and standard deviation of bispectral invariants, as well as characteristic parameters such as average amplitude, phase entropy, normalized bispectral entropy, normalized bispectral squared entropy, and logarithmic bispectral amplitude. The bispectrum is a two-dimensional Fourier transform of the third-order cumulants of a signal, which can preserve the phase information of the signal and reveal second-order nonlinear interactions. The nonlinear statistical characteristics are shown in Table 4 below.

[0026] Table 4 Nonlinear Statistical Characteristics

[0027] S102. Input the multi-level feature information into a pre-set multi-model fusion classifier to identify the signal type, so as to obtain a trained multi-model fusion classifier, and output the signal type and confidence level according to the multi-model fusion classifier.

[0028] In one embodiment, such as Figure 2 As shown, the multi-level feature information extracted in the first step is used as input and fed into a multi-model fusion classifier for signal type identification, resulting in a corresponding trained multi-model fusion classifier. This multi-model fusion classifier does not rely on a single classification algorithm. Instead, it integrates the decision results of multiple machine learning models, such as convolutional neural networks, support vector machines, K-nearest neighbors, and random forests. Through ensemble learning strategies, such as weighted bidding or stacked generalization, it fuses the decisions of the base classifiers, integrating the outputs of each base classifier to finally output a signal type identification representation and the confidence level of this judgment. This confidence level is an important reliability indicator for the subsequent demodulation environment.

[0029] In one embodiment, the multi-model fusion classifier can be divided into four main stages from top to bottom: feature processing, parallel classification, decision fusion, and final output, systematically transforming the original signal into a highly reliable classification decision.

[0030] First, in the feature processing stage, the original signal undergoes cascaded processing through three modules: "feature preprocessing," "feature selection," and "feature standardization," generating a unified and standardized "multidimensional feature input vector." This vector integrates the signal's time-domain, frequency-domain, time-frequency-domain, and nonlinear features, providing a comprehensive and high-quality information foundation for subsequent parallel classification.

[0031] Subsequently, as Figure 3 As shown, the system enters the parallel classification stage. The preprocessed feature vectors are simultaneously fed into multiple heterogeneous base classifiers via "multi-path distribution," including Faster R-CNN, CNN, SVM, Random Forest, and KNN. This design fully utilizes the complementary advantages of different models: Faster R-CNN specializes in identifying spatial patterns and structures from time-frequency images, while other models (CNN, SVM, Random Forest, KNN) focus on learning discriminative rules from structured feature parameters. They work independently, producing preliminary classification results and confidence levels.

[0032] Next, in the crucial decision fusion stage, the outputs of each base classifier are aggregated into a "weighted voting / stacked generalization" layer. Here, the system does not simply adopt majority voting; instead, it corrects the probabilistic reliability of each model's output through "confidence calibration," assesses the degree of synergy between the conclusions of different models through "model consistency evaluation," and combines "feature importance analysis" to understand the decision-making basis. Finally, these calibrated and evaluated results are intelligently fused using ensemble learning strategies (such as weighted voting or stacked generalization) to form a more robust joint decision that incorporates collective wisdom.

[0033] Finally, in the output stage, the system generates a comprehensive result that includes the "signal type," the "confidence level" of the judgment, and the "reliability" based on the fusion process evaluation. This three-in-one output not only provides a clear classification conclusion but also, through quantitative confidence and reliability assessments, offers crucial decision-making quality references for subsequent processing stages such as demodulation, thereby improving the overall intelligence and robustness of the signal processing chain.

[0034] After the parallel model inference stage, the system enters the decision fusion and optimization stage. This stage employs advanced ensemble learning strategies to intelligently synthesize the independent predictions of each base classifier, generating a final decision that is more robust and accurate than any single model. The system primarily supports two core fusion mechanisms: weighted voting and stacked generalization. The operational processes of these two mechanisms are as follows: Figure 4 and Figure 5 As shown.

[0035] This mechanism assigns a dynamic or preset weight to each base classifier to quantify its historical performance or credibility in the current decision, and integrates the "opinions" of each model through weighted summation to form a comprehensive probability distribution. Its logic is clear and intuitive, computationally efficient, and it can quickly fuse multi-model decisions, flexibly reflecting the relative reliability of different models in different scenarios by adjusting the weights.

[0036] In addition, a stacked generalization fusion mechanism is provided, which is a more powerful two-layer learning framework. Its core idea is to use the prediction results of the first-layer base classifier as new "meta-features," and then train a second-layer "meta-model" to learn the optimal combination of these base model predictions, thereby achieving higher-order pattern recognition. Compared to linear weighted averaging, stacked generalization can capture and utilize the complex nonlinear interactions and complementary relationships between base model predictions, theoretically possessing stronger model combination capabilities and a higher performance ceiling, making it particularly suitable for handling signal classification tasks with complex patterns.

[0037] The design of the decision fusion layer is crucial for the system to achieve high-precision and robust recognition. By integrating one or two of the aforementioned advanced fusion mechanisms, it elevates the independent diagnoses of multiple "expert models" into a coordinated and unified "expert committee" for comprehensive judgment. This layer not only outputs accurate signal type identification results but also simultaneously provides a fusion-optimized confidence score. This confidence score is the core quantitative indicator for evaluating the reliability of the classification decision and will be seamlessly transmitted to downstream key processing stages such as demodulation and decoding. It serves as an important basis for dynamically adjusting algorithm parameters, evaluating link quality, or triggering signal re-acquisition, thereby constructing a fully intelligent and adaptive signal processing closed loop from feature extraction, model inference, decision fusion to reliable applications.

[0038] S103. Identify the non-orthogonal multi-carrier shortwave signal across the entire frequency band to obtain the identification result, and perform digital channelization processing based on the identification result to determine the raw data suitable for narrowband demodulation.

[0039] The detection, identification, and digital channelization of non-orthogonal multi-carrier shortwave signals in the entire shortwave frequency band mainly involves detecting non-orthogonal multi-carrier shortwave signals and digitally channelizing them based on the detection results, providing raw data for narrowband demodulation.

[0040] The multi-model fusion classifier trained in step one transforms from a static sample analysis tool into a core intelligent engine driving full-band signal scanning and processing. This bridges the gap between "broadband reconnaissance" and "narrowband processing," enabling the detection and identification of full-band signals. A two-stage processing strategy combining "coarse detection" and "fine identification" is employed. First, a rapid coarse scan is performed in the frequency domain to locate anomalous energy bands. Then, multi-frame in-depth analysis is conducted within these bands, using the intelligent multi-model fusion classifier for confirmation and identification. This approach avoids the enormous computational overhead of performing full-band in-depth processing, achieving a rational allocation of computing resources.

[0041] In one embodiment, such as Figure 6As shown, the continuous frame processing of full-band baseband data is performed using a half-window step. A window function (e.g., Hanning window) is added to the framed data, followed by a Fast Fourier Transform (FFT) to transform the signal from the time domain to the frequency domain, obtaining instantaneous power spectrum and phase information. After accumulating a certain length, cumulative power spectrum and multi-frame instantaneous power spectrum are obtained. The obtained instantaneous power spectrum is then subjected to adaptive noise floor estimation and compared with the instantaneous power spectrum. When the power value of one or more consecutive frequency bands continuously exceeds the detection threshold, an energy anomaly is identified, and the start and end frequencies of that frequency band are recorded and marked as "suspicious frequency bands." Within the acceptable frequency band range, the multi-frame instantaneous power spectrum, phase information, and cumulative power spectrum obtained in step one are truncated. Feature extraction is performed on the truncated data, and statistics are conducted based on waveform similarity, waveform correlation, time-frequency domain statistical features, and nonlinear features. After processing according to the input requirements of the multi-model fusion classifier, the data is fed into the classifier for inference. Sliding processing is performed on each sample point exceeding the classifier's specified length. The classifier outputs a decision, typically containing two parts: first, the classification result: "No signal," "Unknown," or "Signal A," where "No signal" indicates no signal within the current frequency band in the time range, "Unknown" indicates a signal outside the current classifier's category, and "Signal A" indicates a signal within the current classifier's category; second, the confidence level, i.e., the credibility of the previous classification result. Digital channelization is then performed based on the signal detection and recognition results. By driving a digital channelizer (usually implemented based on polyphase filtering or high-efficiency filter banks), the center frequency and bandwidth of the channelizer are adjusted to perfectly match the frequency band of the identified signal. From the broadband baseband data stream, the narrowband signal containing the target signal is precisely "filtered out" and down-converted to zero intermediate frequency, forming clean, demodulated narrowband baseband data. By employing a two-stage pipeline of "coarse detection + fine judgment," continuous deep processing across the entire frequency band is avoided, resulting in high computational resource utilization. Simultaneously, by combining energy detection and intelligent pattern recognition, the probability of false alarms (mistaking noise for a signal) and missed alarms (missing real signals) is effectively reduced, ensuring high reliability. This module not only provides the demodulation end with clean data, but also provides preliminary prior information on the signal type, enabling subsequent demodulation to be more targeted and improving the overall demodulation performance; at the same time, the boundaries of each functional segment are clear, which facilitates the implementation, debugging and upgrading of the system.

[0042] S104. Demodulate the original data by calling the corresponding demodulation algorithm library according to the signal type, and optimize the decision based on the confidence level.

[0043] During adaptive blind demodulation, the system adaptively calls a dedicated demodulation algorithm library that matches the signal type output by the identification module. This could be a filter bank demodulation algorithm for FBMC signals or a block cyclic equalization algorithm for GFDM signals. Simultaneously, the system uses the confidence level output by the identification module for decision-making: when the confidence level is higher than a set threshold, the corresponding demodulator is directly activated; when the confidence level is low, a re-identification mechanism is triggered or a more conservative joint processing strategy is adopted to ensure the system's reliability under critical conditions. Based on the digitally channelized baseband data, which contains only a single signal, blind parameter estimation and demodulation are performed according to the characteristics of non-orthogonal multicarrier signals.

[0044] In one embodiment, such as Figure 7 As shown, the multi-level verified pilot and subcarrier frequency blind estimation, after receiving the baseband data, first performs blind estimation of the key frequency parameters of the target signal in the frequency domain. Specifically, this includes: peak search of the signal's power spectrum to initially identify frequency points with significant energy. Based on prior knowledge of the fixed frequency spacing between the pilot and subcarriers, and between subcarriers, of the non-orthogonal multi-carrier signal, multi-level correlation verification is performed on the initially identified frequency points. For example, it is first verified whether there is a frequency point pair that satisfies the pilot and first subcarrier spacing, and then verified whether this frequency point pair conforms to the complete subcarrier spacing distribution with other peaks in the spectrum. When multiple candidate frequency point pairs exist, the verification conditions are iteratively tightened (e.g., increasing the number of verification interval levels) and combined with power magnitude analysis to eliminate interference, and finally robustly determine the pilot frequency and the center frequencies of all subcarriers, completing the "fingerprint" identification and location of the signal in the frequency domain.

[0045] Blind identification of modulation schemes based on the statistical distribution of phase differences involves channelizing the signal after determining the subcarrier frequency to obtain time-domain data for each subcarrier. For each subcarrier data, after initial timing synchronization (which can be coarse synchronization), the phase difference sequence between adjacent symbols is calculated. The distribution of the phase difference within the range [0, 2π) is statistically analyzed, and its statistical characteristics, particularly the number and spacing of the main peaks, are examined. The modulation scheme is determined based on these statistical characteristics: if the spacing between the main peaks of the phase difference distribution is approximately π, it is identified as BPSK modulation; if the spacing is approximately π / 2, it is identified as QPSK modulation. By utilizing the statistical characteristics of differential phase, the effects of absolute phase ambiguity and residual frequency offset are effectively overcome, achieving reliable blind identification of modulation schemes.

[0046] Pilot-based cross-frame phase continuity compensation eliminates residual frequency offset and phase noise, ensuring the continuity of differential decoding. Dynamic phase compensation is performed: Intra-frame reference acquisition: Using the pilot signal identified in step one as a clean phase reference, the phase rotation common to the current frame data is estimated and corrected. Cross-frame continuity maintenance: During differential decoding, if the calculation involves data from two adjacent frames (e.g., the first symbol of the current frame and the last symbol of the previous frame), it is ensured that all data involved in the calculation has undergone pilot-based phase compensation. This mechanism eliminates phase jumps caused by frame processing, guaranteeing the continuity and correctness of differential phase between frames, which is crucial for low bit error rate demodulation.

[0047] Signal demodulation and error verification utilize all the precise parameters obtained in the preceding steps to perform the final demodulation and verification: Parameter adaptive demodulation: Based on the modulation scheme (BPSK / QPSK) identified in step two, the corresponding differential decoding algorithm is employed; the signal is sampled at the unified sampling time determined in step three; and the stable phase data compensated in step four is used for decoding to restore the bit information carried on each subcarrier. Decoding and verification: The decoded bitstream is correctly decoded according to the subcarrier order determined in step one. Real-time error detection of the demodulated data is performed using the inherent parity bits (such as parity bits) in the signal format. The verification results are not only used to verify the correctness of the data in this frame, but their statistical values ​​(such as bit error rate) are also output as an important indicator of demodulation quality.

[0048] In one embodiment, the target signal 12PSK is taken as an example. This signal is known to consist of 12 subcarrier frequencies and 1 pilot frequency. The pilot is divided into upper and lower pilot frequencies, with a pilot frequency interval of 400Hz. The subcarrier frequency interval is 200Hz. The pilot may or may not be present. The signal with the furthest frequency interval from the pilot is the first signal, and the closest is the 12th. Odd-numbered carrier frequencies and even-numbered carrier frequencies undergo even parity checks, i.e., parity bits for channels 1, 3, 5, 7, and 9 are encoded in channel 11, and parity bits for channels 2, 4, 6, 8, and 10 are encoded in channel 12. All carrier pulses are synchronized with no time delay and are encoded sequentially. It has two modulation methods: BPSK and QPSK, both using differential coding. The initial phase offset of QPSK may be 0 or π / 4. The symbol rate of each of the 12 channels is 120Bd, and the total effective symbol rate is 1200Bd. Automatic detection, identification, and demodulation of this signal are performed across the entire shortwave frequency band.

[0049] In one embodiment, a multi-model fusion classifier capable of accurately identifying 12PSK signals is constructed. The specific implementation process is as follows: When constructing the training sample dataset, the data sources were first determined. A large number of 12PSK signal samples were generated through software simulation, while signal data in actual shortwave environments were also collected. These samples needed to cover different signal-to-noise ratio (SNR) conditions, from low to high SNR; they needed to cover various channel conditions, such as multipath, fading, and frequency offset; and they needed to include all possible configurations of 12PSK signals, such as upper pilot, lower pilot, and BPSK and QPSK modulation. Next, data preprocessing was performed, with the same processing flow applied to each sample signal, including signal detection and segmentation, windowing, and short-time Fourier transform (STFT), to generate instantaneous power spectrum and cumulative power spectrum. Finally, labeling was carried out to assign the true category to each sample, such as "12PSK signal - BPSK - upper pilot", "12PSK signal - QPSK - lower pilot", "noise", or "other type of signal", such as OFDM signal.

[0050] In the feature extraction stage, four main categories of features need to be extracted from the power spectrum data of each sample. This power spectrum data covers both single-frame power spectrum and cumulative power spectrum. The first category is waveform similarity features, obtained by calculating the DTW and DDTW distances between the sample and the ideal 12PSK signal template. The second category is correlation features, specifically calculating the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the signal. The third category belongs to time-frequency domain statistical features, requiring the calculation of indicators such as zero-crossing rate, spectral centroid, and spectral entropy. The fourth category is nonlinear statistical features, requiring the calculation of the mean of the bispectral invariants and phase entropy.

[0051] In the model training and fusion phase, the first step is to train the base classifiers. The feature dataset is divided into training, validation, and test sets. Multiple base classifiers, such as CNN, SVM, Random Forest, and KNN, are trained independently on the training set. The hyperparameters of each model are then adjusted on the validation set to ensure optimal performance. Next, ensemble learning and fusion are performed using a stacked generalization strategy. Specifically, in the first layer (base learner stage), the training set is input into all the trained base classifiers to obtain their predictions, such as class probabilities. In the second layer (meta learner stage), the predictions from the first-layer base classifiers are used as a new feature matrix. A meta classifier, such as logistic regression, is trained with the corresponding true class labels as the target. This meta classifier learns how to optimally combine the "opinions" given by the base classifiers. Finally, this fusion model, composed of multiple base classifiers and meta classifiers, is saved for subsequent online recognition.

[0052] In one embodiment, the trained model is applied to the online detection and recognition of actual shortwave full-band signals. The process is similar to... Figure 6 Correspondingly.

[0053] During full-band signal input and preprocessing, wideband baseband IQ data from a shortwave receiver is first received. The received data is then framed, a Hanning window is added, and a short-time Fourier transform (STFT) is performed to obtain the instantaneous power spectrum of each frame. Finally, the power spectra of multiple frames are accumulated to obtain the cumulative power spectrum. .

[0054] In the process of broadband coarse detection to locate suspicious frequency bands, an adaptive detection threshold is first estimated based on the background noise. Next, the instantaneous power spectrum of the current frame is... With this threshold Compare them. If there exists a continuous frequency band B whose average power satisfies... The frequency band is then marked as a "suspicious frequency band," and its start and end frequencies are recorded.

[0055] In the model-based fine identification stage, for each region marked as a "suspicious frequency band," relevant data is first extracted from its corresponding multi-frame instantaneous power spectrum and cumulative power spectrum. Then, the extracted data for that frequency band undergoes the same feature extraction process as described above, generating a feature vector. Next, this feature vector is input into a pre-trained multi-model fusion classifier, which performs inference and outputs the final identification result. This result contains two parts: first, the signal type, such as "12PSK signal," "non-12PSK signal," or "unknown signal"; and second, the confidence score, a value between 0 and 1 representing the reliability of the classification result.

[0056] In the decision-making and output phase, the system makes decisions based on the identification result and confidence level. If the identification result shows "12PSK signal" and the confidence level is higher than a preset threshold, such as 0.8, then it is determined that a target signal exists in the frequency band, and the subsequent digital channelization and adaptive demodulation process is triggered. If the confidence level is low, the system can choose to discard the result or trigger a more complex verification mechanism for further confirmation.

[0057] In one embodiment, the specific location of the pilot is searched to determine whether it is an upper or lower pilot, thereby determining the subcarrier signal numbering order and the position of each frequency point. Then, the Welch method is used to calculate the power spectral density of the signal segment, and preprocessing is performed. Preprocessing mainly includes filtering out low-power noise interference and performing frequency domain smoothing, which can be achieved through multiple smoothing operations using a rectangular window. Finally, the processed power spectral density information is obtained. In actual shortwave environments, due to various interferences such as fading and multipath propagation, the power spectral density of the signal is often quite complex.

[0058] Next, peak position detection and fixed frequency pair search are performed to find all frequency pairs that match the first carrier and pilot spacing. For each searched frequency pair, both points must be verified to determine if they are pilots, based on the spacing between the pilot and the two adjacent carriers, as well as its spacing with other peaks. If no suspected pilot frequency points satisfying the adjacent spacing condition are found, power values ​​and other indicators need to be further checked to eliminate pseudo-frequency interference. If multiple frequency pair searches are found, interference filtering must be eliminated before searching again, and verification is performed after the search is completed. For possible multiple frequency pair searches, the spacing verification condition needs to be tightened, and verification is performed sequentially. If pilot estimation fails, it indicates that the signal segment does not meet the 12P signal characteristic condition and can be discarded directly; if the signal meets the requirements, it is determined to be an upper or lower pilot based on the position of the pilot relative to the upper and lower sidebands.

[0059] In the blind modulation scheme identification process, each subcarrier is first channelized and estimated separately. For each subcarrier, the phase value at a fixed position of each symbol is selected to calculate the phase difference between adjacent symbols, avoiding symbol edges. The resulting statistical distribution of phase differences is folded and normalized to the range [0, 2π]. After smoothing, the statistical peak positions are obtained and then merged. Next, the difference between the peak positions is calculated to obtain the statistically significant phase difference distribution of the carrier. Since BPSK theoretically has only two phases, the phase difference between adjacent symbols should be π, so the number of peaks is no more than 3, taking into account the possibility that the peaks may be located at the edge of the phase range of 0 or 2π. QPSK has four phases, the phase difference between adjacent symbols is π / 2, and the number of peaks is no more than 5. Finally, if the estimation results of at least 8 out of the 12 sub-bands are consistent, the estimation result of the modulation scheme is considered reliable.

[0060] In the distributed cooperative symbol timing synchronization process, the sub-bands are first separated and channelized through mixing and filtering. Next, using different positions within a symbol as starting points, a single-symbol processing length shorter than the symbol length is selected, and the data is summed according to the symbol rate interval. During this process, the amplitude of the optimal sampling point will be higher after filtering than other sampling points. Finally, considering the optimal sampling points of each subcarrier, a unified sampling position is adopted to complete the symbol timing synchronization.

[0061] The pilot-based cross-frame phase continuity compensation process first obtains the phase of each symbol on the pilot as the reference phase value to correct the phase offset accumulated due to incomplete alignment of intermediate frequencies. For ease of subsequent use, the relevant data is normalized and stored. Next, the channelized symbol values ​​of each subcarrier are summed according to the synchronization position, and the inter-symbol difference result is obtained directly using the conjugate value. For non-first frames, the tail data from the previous frame is used, and the pilot-corrected data must be used before cross-frame data operations to compensate for the impact of frequency offset.

[0062] Signal demodulation and error verification. Depending on the demodulation method, mapping values ​​are taken for BPSK and QPSK respectively. For QPSK, a phase folding method is used to determine the initial phase, and then the decision is made based on the initial phase.

[0063] like Figure 8 As shown in the illustration, this application also provides an identification and demodulation device for non-orthogonal multi-carrier shortwave signals, comprising: At least one processor; and, A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor to enable a non-orthogonal multicarrier shortwave signal identification and demodulation device to perform the method as described in any of the embodiments above.

[0064] This application also provides a non-volatile computer storage medium storing computer-executable instructions, which are configured as described in any of the above embodiments.

[0065] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0066] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0067] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0068] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0069] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0070] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0071] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0075] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0076] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0077] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0078] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method of identification demodulation of a non-orthogonal multicarrier shortwave signal, characterized in that include: Acquire a target signal and perform feature extraction on the target signal to determine multi-level feature information, wherein the multi-level feature information includes at least one or more combinations of waveform similarity-based features, correlation-based features, time-frequency domain statistical features, and nonlinear statistical features. The multi-level feature information is input into a pre-set multi-model fusion classifier to identify the signal type, so as to obtain a trained multi-model fusion classifier. The multi-model fusion classifier outputs the signal type and confidence level. Non-orthogonal multi-carrier shortwave signals across the entire frequency band are identified to obtain identification results, and digital channelization processing is performed based on the identification results to determine the raw data suitable for narrowband demodulation. According to the signal type, the corresponding demodulation algorithm library is invoked to demodulate the original data, and decision optimization is performed based on the confidence level.

2. The method of claim 1, wherein, The multi-level feature information is input into a pre-set multi-model fusion classifier for signal type identification to obtain a trained multi-model fusion classifier, specifically including: The multi-level feature information is divided into a training set, a validation set, and a test set; The training set is used to train multiple pre-set base classifiers, including convolutional neural networks, support vector machines, random forests, and K-nearest neighbors. The hyperparameters of the multiple base classifiers are adjusted based on the validation set, and the decision results of the multiple base classifiers are fused to output the final signal type and the confidence level of the judgment.

3. The method of claim 1, wherein, The non-orthogonal multi-carrier shortwave signals across the entire frequency band are identified to obtain identification results, and digital channelization processing is performed based on the identification results, specifically including: The baseband data across the entire frequency band is processed in continuous frames and then subjected to a fast Fourier transform to obtain the instantaneous power spectrum and phase information. Adaptive noise floor estimation is performed on the instantaneous power spectrum to locate the energy anomaly frequency band; Multi-frame depth analysis was performed within the energy anomaly frequency band, and a multi-model fusion classifier was used for signal confirmation and identification to obtain the identification results; The corresponding digital channelizer is driven according to the recognition result to adjust the center frequency and bandwidth of the digital channelizer to match the frequency band of the recognized signal, thereby filtering out the narrowband signal where the target signal is located and downconverting it to zero intermediate frequency.

4. The method of claim 1, wherein, Based on the signal type, the corresponding demodulation algorithm library is invoked to demodulate the original data, and decision optimization is performed based on the confidence level, specifically including: Based on the signal type, a corresponding demodulation algorithm is determined from a pre-set demodulation algorithm library, and the confidence level is compared with a pre-set threshold. If the confidence level is higher than the threshold, the corresponding demodulator is activated directly. If the confidence level is lower than the threshold, a re-identification mechanism is triggered; Blind parameter estimation and demodulation are performed on the baseband data after digital channelization processing. The steps of blind parameter estimation and demodulation include blind estimation of pilot and subcarrier frequencies, blind identification of modulation mode, distributed cooperative symbol timing synchronization, cross-frame phase continuous compensation based on pilot, signal demodulation and error verification.

5. The method of claim 4, wherein, Blind parameter estimation and demodulation are performed on the baseband data after digital channelization, specifically including: Blind estimation is performed on the key frequency parameters of the target signal, including the pilot frequency and the center frequency of all subcarriers. Channelization separation is performed based on the subcarrier frequency to obtain time-domain data on the subcarrier, and modulation mode blind identification is performed based on the time-domain data; Perform distributed cooperative symbol timed synchronization to determine the optimal sampling point; Cross-frame phase continuity compensation is performed based on pilot signals to eliminate residual frequency offset and phase noise; The signal is demodulated based on the obtained precise parameters, and error checking is performed.

6. The method of claim 4, wherein, The method further includes: Peak search is performed on the power spectrum of the signal to identify frequency points with significant energy. Based on prior knowledge of the fixed frequency intervals between pilots and subcarriers, and between subcarriers, of non-orthogonal multicarrier signals, multi-level correlation verification is performed on the identified frequency points. When multiple candidate frequency pairs exist, the verification conditions are tightened iteratively and combined with power magnitude analysis to eliminate interference, thereby determining the pilot frequency and the center frequency of all subcarriers.

7. The method of claim 1, wherein, The method further includes: Channelization separation is performed on each subcarrier to obtain time-domain data on the subcarrier; Calculate the phase difference sequence between adjacent symbols of subcarrier data, and statistically analyze the distribution of the phase difference within a preset range; The modulation method is determined by determining the peak interval based on the number and spacing of peaks in the statistical characteristics.

8. The method of claim 1, wherein, Acquiring the target signal and performing feature extraction on the target signal specifically includes: The target signal is preprocessed, and the preprocessing steps include signal truncation, segmentation, and variable sampling normalization. Extract waveform similarity-based features from the preprocessed target signal, including Euclidean distance, dynamic time warping, derivative dynamic time warping, and weighted dynamic time warping. Extract correlation-based features from the preprocessed target signal, including autocorrelation function, partial autocorrelation function, Pearson correlation coefficient, and sliding window correlation coefficient; Extract time-frequency domain statistical features from the preprocessed target signal, including zero-crossing rate, amplitude standard deviation, skewness, kurtosis, and spectral centroid. Nonlinear statistical features are extracted from the preprocessed target signal, including the mean of bispectral invariants, the standard deviation of bispectral invariants, the phase entropy, the normalized bispectral entropy, the normalized bispectral squared entropy, and the logarithmic bispectral amplitude.

9. A non-orthogonal multicarrier short wave signal recognition demodulation apparatus, characterized by, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the non-orthogonal multicarrier shortwave signal identification and demodulation device to perform the method described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to be the method as described in any one of claims 1-8.

Images