Implementation of Deep Learning-based Automatic Modulation Classifier on FPGA SDR Platform
Zhi-Ling Tang,Si-Min Li,Li-Juan Yu
DOI: https://doi.org/10.3390/electronics7070122
IF: 2.9
2018-07-19
Electronics
Abstract:Intelligent radios collect information by sensing signals within the radio spectrum, and the automatic modulation recognition (AMR) of signals is one of their most challenging tasks. Although the result of a modulation classification based on a deep neural network is better, the training of the neural network requires complicated calculations and expensive hardware. Therefore, in this paper, we propose a master–slave AMR architecture using the reconfigurability of field-programmable gate arrays (FPGAs). First, we discuss the method of building AMR, by using a stack convolution autoencoder (CAE), and analyze the principles of training and classification. Then, on the basis of the radiofrequency network-on-chip architecture, the constraint conditions of AMR in FPGA are proposed from the aspects of computing optimization and memory access optimization. The experimental results not only demonstrated that AMR-based CAEs worked correctly, but also showed that AMR based on neural networks could be implemented on FPGAs, with the potential for dynamic spectrum allocation and cognitive radio systems.
engineering, electrical & electronic,computer science, information systems,physics, applied