A Deep Learning Wireless Transceiver with Fully Learned Modulation and Synchronization

Abstract

In this paper, we present a deep learning based wireless transceiver. We describe in detail the corresponding artificial neural network architecture, the training process, and report on excessive over-the-air measurement results. We employ the end-to-end training approach with an autoencoder model that includes a channel model in the middle layers as previously proposed in the literature. In contrast to other state-of-the-art results, our architecture supports learning time synchronization without any manually designed signal processing operations. Moreover, the neural transceiver has been tested over the air with an implementation in software defined radio. Our experimental results for the implemented single antenna system demonstrate a raw bit-rate of 0.5 million bits per second. This exceeds results from comparable systems presented in the literature and suggests the feasibility of high throughput deep learning transceivers.

Blind Interactive Learning of Modulation Schemes: Multi-Agent Cooperation Without Co-Design

“…In Appendix C-A we investigated performance of the learning protocols for higher modulation orders and noticed that the difficulty of the learning task increases substantially with modulation order, and the number of preamble symbols that must be transmitted before a good scheme is learned increases exponentially. Confirming the results of others ( [9], [29]), we observed that moderate levels of noise have a regularizing effect and facilitate learning but too much noise can be detrimental to the learning process.…”

“…Since any learning communications system in the wild will be exposed to multiple SNR conditions and desired signalling rates, understanding performance variation across SNR and modulation order will be crucial. Our results indicate that moderately high training SNR leads to the best performance confirming observations by others ( [9], [29]). Appendix C-B presents experiments with Poly clone and selfalien agents demonstrating similar behavior to Neural clone and self-alien agents.…”

“…The focus of our work is to learn modulation schemes, so the primary goal of the GNU Radio implementation is to demonstrate that the learning protocols work not only in simulations but also when trained in real, physical systems. Other work such as [29] and [30] have also demonstrated that end-to-end learning of communication schemes is possible over the air in real radio systems. We plan to address other components of radio communications such as channel equalization and error correction coding in future works.…”

“…Reinforcement learning techniques have the added advantage of being implementable in software-defined radios to perform end-to-end learning over the air. To do this one must tackle the issue of time synchronization between the transmitted and received symbols as done in [29] and [30]. In [31], the general problem of synchronization in wireless networks is addressed via the use of attention models.…”

End‐to‐end waveform level receiver with deep learning

“…The timing delay issue is tackled with specially optimized frame synchronization module. Through proper pilot symbol, [15] constructs a transceiver with fully learned waveform for synchronization. The overhead of this method is relatively large, which makes it inefficient for modern communication system.…”