[Paper Review] Channel Agnostic End-to-End Learning based Communication Systems with Conditional GAN

Hao Ye, Geoffrey Ye Li+2 more|arXiv (Cornell University)|Jul 2, 2018

Wireless Signal Modulation Classification11 references17 citations

TL;DR

This paper proposes a channel-agnostic end-to-end communication system using a conditional generative adversarial network (cGAN) to model unknown channel effects without requiring prior knowledge of the channel state information (CSI). By conditioning the cGAN on transmitted signals and pilot symbols, the system learns the channel distribution end-to-end, achieving performance comparable to traditional methods on AWGN and Rayleigh fading channels without gradient backpropagation through the unknown channel.

ABSTRACT

In this article, we use deep neural networks (DNNs) to develop a wireless end-to-end communication system, in which DNNs are employed for all signal-related functionalities, such as encoding, decoding, modulation, and equalization. However, accurate instantaneous channel transfer function, \emph{i.e.}, the channel state information (CSI), is necessary to compute the gradient of the DNN representing. In many communication systems, the channel transfer function is hard to obtain in advance and varies with time and location. In this article, this constraint is released by developing a channel agnostic end-to-end system that does not rely on any prior information about the channel. We use a conditional generative adversarial net (GAN) to represent the channel effects, where the encoded signal of the transmitter will serve as the conditioning information. In addition, in order to deal with the time-varying channel, the received signal corresponding to the pilot data can also be added as a part of the conditioning information. From the simulation results, the proposed method is effective on additive white Gaussian noise (AWGN) and Rayleigh fading channels, which opens a new door for building data-driven communication systems.

Motivation & Objective

• To overcome the limitation of end-to-end deep learning in communication systems that require accurate instantaneous channel state information (CSI) for gradient backpropagation.
• To enable end-to-end training of transmitter and receiver DNNs without prior knowledge of the channel transfer function.
• To model the distribution of unknown channel effects using a conditional GAN, conditioned on transmitted signals and pilot data.
• To achieve robust performance on time-varying channels such as Rayleigh fading without explicit channel modeling.
• To demonstrate that a purely data-driven, channel-agnostic communication system can match traditional methods in performance.

Proposed method

• A conditional GAN is used to model the channel output distribution, with the transmitter's encoded signal as conditioning input.
• The generator learns to synthesize realistic channel outputs based on the encoded signal and, in time-varying cases, pilot symbols.
• The discriminator distinguishes between real channel outputs (from actual transmission) and synthetic outputs (from the generator).
• The transmitter and receiver DNNs are trained in a supervised manner using the end-to-end loss, while the channel generator is trained adversarially.
• Pilot symbols are incorporated as additional conditioning information to enable adaptation to time-varying channels.
• Iterative training of the transmitter, receiver, and channel generator allows end-to-end optimization without direct gradient computation through the unknown channel.

Experimental results

Research questions

• RQ1 Can a deep learning-based communication system be trained end-to-end without requiring prior knowledge of the channel state information (CSI)?
• RQ2 How effectively can a conditional GAN model the distribution of unknown channel effects in time-varying environments?
• RQ3 Can the proposed channel-agnostic system achieve performance comparable to traditional, model-based communication systems on standard channels like AWGN and Rayleigh fading?
• RQ4 What is the impact of including pilot symbols as conditioning information on the system's ability to handle time-varying channels?
• RQ5 Is it possible to train the transmitter and receiver via backpropagation when the channel is unknown, by using a learned channel generator?

Key findings

• The conditional GAN successfully models the output distribution of both AWGN and Rayleigh fading channels, producing synthetic signal constellations that closely match real channel outputs.
• On the AWGN channel, the proposed method achieves a block error rate (BLER) similar to that of a Hamming (7,4) code with maximum-likelihood decoding.
• On the Rayleigh fading channel, the system achieves BLER performance comparable to traditional model-based methods, even without explicit channel estimation.
• The inclusion of pilot symbols as conditioning information enables the model to adapt to time-varying channel conditions and improve reconstruction accuracy.
• The end-to-end system achieves competitive performance without relying on analytical channel models or prior CSI, demonstrating the feasibility of pure data-driven communication systems.
• The method enables gradient-based training of the transmitter even when the channel is unknown, by using the cGAN as a surrogate for the channel in backpropagation.