Abstract:We address the problem of coding for classical multiple-access channels (MACs) with the assistance of non-signaling correlations between parties. It is well-known that non-signaling assistance does not change the capacity of classical point-to-point channels. However, it was recently observed that one can construct MACs from two-player non-local games while relating the winning probability of the game to the capacity of the MAC. By considering games for which entanglement increases the winning probability, this shows that for some specific kinds of channels, entanglement between the senders can increase the capacity. We make several contributions towards understanding the capacity region for MACs with the assistance of non-signaling correlations. We develop a linear program computing the optimal success probability for coding over $n$ copies of a MAC $W$ with size growing polynomially in $n$. Solving this linear program allows us to achieve inner bounds for MACs. Applying this method to the binary adder channel, we show that using non-signaling assistance, the sum-rate $1.5425$ can be reached even with zero error, which beats the maximum sum-rate capacity of $1.5$ in the unassisted case. For noisy channels, where the zero-error non-signaling assisted capacity region is trivial, we can use concatenated codes to obtain achievable points in the capacity region. Applied to a noisy version of the binary adder channel, we show that non-signaling assistance still improves the sum-rate capacity. Complementing these achievability results, we give an outer bound on the non-signaling assisted capacity region that has the same expression as the unassisted region except that the channel inputs are not required to be independent. Finally, we show that the capacity region with non-signaling assistance shared only between each sender and the receiver independently is the same as without assistance.

Cooperative Channel Capacity Learning

Dynamic channel selection in unknown environment based on graphical game and multi-Q learning

Learning Channel Capacity with Neural Mutual Information Estimator Based on Message Importance Measure

Capacity-Approaching Autoencoders for Communications

Deep Learning for Channel Coding via Neural Mutual Information Estimation

Learning to Communicate: Channel Auto-encoders, Domain Specific Regularizers, and Attention

Multiple-Access Channel Coding with Non-Signaling Correlations

SINR-Aware Deep Reinforcement Learning for Distributed Dynamic Channel Allocation in Cognitive Interference Networks

ChannelGAN: Deep Learning-Based Channel Modeling and Generating

Do not Interfere but Cooperate: A Fully Learnable Code Design for Multi-Access Channels with Feedback

Broadcast Channels with Cooperating Decoders

Channel Agnostic End-to-End Learning based Communication Systems with Conditional GAN

Online Optimization for Learning to Communicate over Time-Correlated Channels

An End-to-end Learning Framework for Joint Compensation of Impairments in Coherent Optical Communication Systems

Compound Channel Capacities under Energy Constraints and Application

Practical Coding Schemes for Cognitive Overlay Radios

On the Multiple Access Channel with Asymmetric Noisy State Information at the Encoders

Learning Robust Representations for Communications over Noisy Channels

Online/Offline Learning to Enable Robust Beamforming: Limited Feedback Meets Deep Generative Models

Capacity Maximization in Cooperative CRNs: Joint Relay Assignment and Channel Allocation

A dynamic program to achieve capacity of multiple access channel with noiseless feedback