Deep Learning Fusion for Automatic Modulation Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic modulation classification methods face challenges in distinguishing Wide Band Frequency Modulation (WBFM) signals from Double Sideband-Amplitude Modulation (DSB-AM) signals and Quadrature Amplitude Modulation (16 QAM) signals from 64 QAM signals, particularly due to limitations in the RML 2016.10a data set, resulting in low identification accuracy.
Innovation Solution
An automatic modulation classification method based on deep learning network fusion, involving data enhancement, multi-channel feature fusion using LSTM and FPN networks, and fractional order Fourier transformations, to improve signal feature extraction and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional maximum likelihood-based recognition method is used, then probability distribution can be obtained, but classification accuracy is insufficient for similar modulation modes
Solution Approach 1:
The patent transforms the 2D constellation diagram into a 3D representation by adding time as the third dimension. This dimensional expansion allows the network to capture temporal dynamics and evolution of signal patterns, enabling better distinction between similar modulation modes like WBFM and DSB-AM, as well as 16 QAM and 64 QAM.
Solution Approach 2:
The patent employs a nested architecture where a CNN network is embedded within an LSTM network. The CNN extracts spatial features from constellated signals, and these features are then fed into the LSTM for temporal sequence analysis. This nested structure enables hierarchical feature extraction, improving classification accuracy by combining local pattern recognition with global temporal understanding.
2Measurement precision
If feature extraction-based recognition method is used, then optimal Bayesian estimation can be achieved, but algorithm complexity becomes excessive and parameter estimation dependency increases
Solution Approach 1:
The patent replaces traditional mechanical feature extraction and parameter estimation processes with a deep learning system. Instead of manually designing features and performing complex parameter estimation, the CNN-LSTM network automatically learns optimal features from raw constellated signal data, significantly reducing algorithmic complexity while maintaining or improving estimation accuracy.
Solution Approach 2:
The deep learning network performs self-service by automatically extracting features and performing classification without manual intervention. The network learns optimal feature representations and decision boundaries during training, eliminating the need for expert-designed features and reducing dependency on parameter estimation techniques.
3Extent of automation
If existing deep learning methods are used with RML 2016.10a data set, then identification can be performed, but distinction between WBFM and DSB-AM signals remains difficult
Solution Approach 1:
The patent introduces dynamic temporal analysis by processing constellated signals as sequences rather than static patterns. The LSTM component captures the evolution of signal characteristics over time, enabling the system to distinguish between WBFM and DSB-AM signals based on their temporal dynamics, which static analysis cannot detect.
4Extent of automation
If existing deep learning methods are used with RML 2016.10a data set, then identification can be performed, but distinction between 16 QAM and 64 QAM signals remains difficult
Solution Approach 1:
By adding the temporal dimension through sequence processing, the network can distinguish between 16 QAM and 64 QAM signals based on their temporal evolution patterns. The CNN-LSTM architecture captures both the spatial distribution of constellation points and their temporal transitions, providing additional discriminatory information that resolves the confusion between these similar modulation modes.
Data Source
AI summary
The present invention discloses an automatic modulation classification method based on deep learning network fusion, comprising: acquiring a WBFM sample signal within a data set RML 2016.10a, and selecting a proper threshold γ to separate a WBFM signal during a silence period; expanding a new WBFM signal to 1000 by adopting a data enhancement method, and expanding an original data set; dividing the data set expanded in the step S2 into a training set, a verification set and a test set; respectively calculating amplitude, phase and a fractional order Fourier transformation result for data in the step S3; building a multi-channel feature fusion network model composed of an LSTM network and an FPN network; performing network model training, after the end of training, inputting verification set data into a trained network model for verification, and calculating prediction accuracy; and performing parameter fine adjustment on the network model through said test set, improving prediction precision, and taking a final model as an automatic modulation classification model. The present invention enables the improvement to the average classification accuracy rate of communication signals.