Environment and Intent Specific Adaptive PHY via Deep Learning

Abstract

Approved for Public ReleaseThe design of error-correcting codes for reliable transmission in noisy channels is a fundamental building block of physical layer communications. Traditional code anddecoder designs are optimized for fixed environments such as AWGN or fading channels. However, real-world scenarios are dynamic, with changing factors like fading processes, interference patterns, and spectrum availability. Additionally, evolving communication needs influenced by application and command layer decisions, such as varying data rates and latency, present further challenges. Current PHY layer adaptations to these changes are slow and inefficient, often hampered by inflexible encoding and decoding designs and the separation between different layers.In this project, we apply data-driven learning to adapt the channel decoders (and encoders) to varying environmental conditions and the evolving intent of upper commandand application layers. Specifically, our approach involves leveraging machine learning to enhance PHY adaptivity through low-dimensional parameterization andfine-tuning using data from various channel conditions and upper-layer needs, aiming to design a PHY layer that is closely aligned with the environment and command intentand demonstrate the proposed PHY in an over-the-air environment. The challenge lies in creating efficient, lightweight, and data-adaptive PHY schemes. This proposalexpands upon our recent successin developing efficient and reliable data-driven PHY using deep learning techniques.

Document Details

Document Type
DoD Grant Award
Publication Date
Nov 09, 2024
Source ID
N000142412605

Entities

People

  • Pramod Viswanath

Organizations

  • Office of Naval Research
  • Trustees of Princeton University
  • United States Navy

Tags

Readers

  • Defense Acquisition Program Management
  • Neural Network Machine Learning.
  • Radio communications and signal processing.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks