One-Dimensional Convolution Neural Network for Real-Time Signal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying electromagnetic signals are limited by the need for prior knowledge and high computational complexity, making real-time identification of unknown signals challenging.
Innovation Solution
A machine learning architecture using a signal-analyzing neural-network with one-dimensional convolutional layers and non-linear activation functions is employed to classify parameter-varying signals, allowing for real-time identification without prior knowledge of signal presence or type, leveraging the efficiency of one-dimensional convolutions to reduce computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual or semi-automated observation methods with filter banks or cyclostationary processing are used to identify electromagnetic signals, then signal identification can be performed with existing technology, but the computational complexity increases and real-time identification becomes difficult
Solution Approach 1:
The patent replaces traditional mechanical signal processing methods (filter banks, cyclostationary processing) with a neural network-based system. The neural network automatically learns signal features through training, eliminating the need for manual feature engineering and complex computational searches, thereby reducing computational complexity while maintaining identification accuracy
Solution Approach 2:
The patent transforms the signal identification problem by changing the approach from parameter-based searching (guessing signal characteristics) to pattern-based recognition. The neural network processes raw signal data and automatically extracts features, changing the fundamental parameters of how identification is performed and enabling real-time operation
2Adaptability or versatility
If traditional signal identification methods are used, then prior knowledge of signal presence or type can guide the search, but the ability to identify unknown signals without prior knowledge is limited
Solution Approach 1:
The patent applies preliminary action by training the neural network offline with diverse signal data before deployment. This pre-training enables the network to recognize various signal types without requiring prior knowledge during actual identification, allowing rapid classification of unknown signals in real-time without exhaustive searching
Solution Approach 2:
The neural network is designed with universal applicability to handle multiple signal types and unknown signals simultaneously. Through comprehensive training, a single network model can identify various electromagnetic signals without requiring separate specialized algorithms for each signal type, enhancing adaptability while reducing identification time
3Measurement precision
If complex signal processing algorithms are used to achieve high accuracy in signal classification, then classification precision improves, but power consumption increases
Solution Approach 1:
The patent segments the signal processing task into two phases: offline training and online inference. The computationally intensive training phase is performed separately, while the deployed neural network performs lightweight inference with minimal power consumption. This segmentation allows high accuracy to be achieved during training while maintaining low power consumption during actual signal classification
Solution Approach 2:
The patent creates a trained neural network model that copies the learned knowledge from the training phase. Once trained, the model can be deployed with minimal computational resources, as it applies the learned patterns directly to new signals without requiring the same level of computational power as the training process, thereby reducing power consumption while maintaining classification accuracy
Data Source
AI summary
An output module cooperates with a machine learning architecture to analyze parameter-varying signals. The signal-analyzing neural-network contains at least a one-dimensional-convolution layer to apply a series of i) a one-dimensional convolutional-based operation on the data of the parameter-varying signals ii) followed by a non-linear activation function on the data of the parameter-varying signals, under analysis, with multiple representations of the parameter-varying signals. Each representation of the parameter-varying signal is analyzed in a different domain in order to produce a classification of an entity into a specific category of an object corresponding to identifying features of the time-varying signals. Branches of the signal-analyzing neural-network are constructed to apply at least two or more successive layers of the one-dimensional-convolution layer followed by the non-linear activation function layer to values of time and frequency features of the parameter-varying signals, under analysis.