Radio Technology Recognition Using Semi-Supervised Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based technology recognition methods for radio technologies require extensive labeled data sets, making the labeling task time-consuming and challenging, especially in unknown environments, and they lack the ability to identify unknown radio technologies without expert knowledge.
Innovation Solution
A semi-supervised neural network approach using an autoencoder and classifier, where the autoencoder is trained unsupervisedly with unlabeled data and the classifier is trained supervisedly with a limited number of labeled samples, decoupling feature extraction from classification, allowing for efficient identification of radio technologies in dynamic environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning methods are used for technology recognition, then recognition accuracy is improved, but extensive labeled data is required making the labeling task time-consuming and challenging
Solution Approach 1:
The patent segments the learning process into two distinct phases: unsupervised pre-training on unlabeled data to learn feature representations, followed by supervised fine-tuning on a small labeled dataset for classification. This segmentation allows the system to leverage the abundance of unlabeled data while minimizing the time-consuming labeling task.
Solution Approach 2:
The patent performs preliminary unsupervised learning on unlabeled data before the supervised classification stage. This preliminary action extracts meaningful features and representations from the data without requiring labels, thereby reducing the dependency on extensive labeled datasets and minimizing labeling time.
2Measurement precision
If supervised learning methods are used for technology recognition, then recognition accuracy is improved, but the labeling task becomes challenging in unknown environments
Solution Approach 1:
The patent divides the recognition system into an adaptive unsupervised feature extraction component that can learn from any environment without labels, and a supervised classification component that adapts to specific technologies. This segmentation enables the system to first adapt to unknown environments through unsupervised learning, then achieve accurate recognition with minimal labeled data.
Solution Approach 2:
The patent changes the learning paradigm from purely supervised to a hybrid unsupervised-supervised approach. By adjusting the learning parameters and stages, the system can adapt to unknown environments in the unsupervised phase and then achieve precise recognition in the supervised phase, thereby improving both adaptability and accuracy.
3Extent of automation
If state-of-the-art machine learning methods are used, then feature extraction is automated, but extensive labeled data is required
Solution Approach 1:
The patent performs preliminary unsupervised feature extraction from unlabeled data before the supervised classification stage. This preliminary automated feature extraction eliminates the need for manual feature engineering and reduces the quantity of labeled data required, as the model learns robust features from the abundant unlabeled data.
Solution Approach 2:
The patent enables the system to self-service by automatically learning features from unlabeled data without human intervention or manual labeling. The unsupervised pre-training stage allows the model to autonomously extract meaningful representations, reducing dependency on extensive labeled datasets.
4Measurement precision
If traditional domain expert methods are used, then recognition is performed with hand-crafted rules, but the process is time-consuming and requires expert knowledge
Solution Approach 1:
The patent replaces the mechanical process of manual feature extraction by domain experts with an automated neural network-based unsupervised feature extraction system. This substitution eliminates the need for expert knowledge in feature engineering while maintaining or improving recognition capability through data-driven feature learning.
Solution Approach 2:
The patent enables the system to automatically learn and extract features from unlabeled data without requiring domain experts. The unsupervised pre-training stage allows the model to self-service by autonomously identifying relevant features and patterns, thereby eliminating dependency on expert knowledge while achieving accurate recognition.
Data Source
AI summary
A computer-implemented method providing a neural network for identifying radio technologies employed in an environment. The neural network includes an autoencoder having an encoder, and a classifier. The method has the steps of sensing a radio spectrum of the environment thereby obtaining a set of data samples, labelling a subset of the data samples by a respective radio technology thereby obtaining labelled data samples, training the autoencoder in an unsupervised way by unlabelled data samples, training the classifier in a supervised way by the labelled data samples, and providing the neural network by coupling the output of an encoder network of the autoencoder to an input of the classifier.