In this paper, we propose a low decay, low bias dataset synthesis framework that models Machine Learning (ML) dataset theory using Python classes and instruction files, and whose simulation results show an 11.58% entropy decrease at classification time relative to state-of-the-art training sets. The demand for signal-domain Neural Networks (NNs) have increased significantly in recent years with respect to the classification of observed radio activity. In particular, there has been a growing interest in choosing appropriate training data in order to enhance NN performance at classification time. Developing ML based signal classifiers requires training data that captures the underlying probability distribution of real signals. To synthesize a set of training data that can capture the large variance in signal characteristics, a robust framework that can support arbitrary baseband signals and channel conditions is presented.
Physical Layer Neural Network Framework for Training Data Formation
2018-08-01
3050924 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Europäisches Patentamt | 2019
|