AI-Native Air Interface Design: End-to-End Learning for 6G Physical Layer
Dr. Sebastian Dorner, Prof. Stephan ten Brink
University of Stuttgart
Abstract
We propose an AI-native air interface where the entire physical layer — from modulation to coding to equalization — is learned end-to-end using deep neural networks. Unlike traditional block-based PHY design, our autoencoder-based system jointly optimizes all components for the actual channel conditions. Over-the-air experiments using software-defined radios demonstrate that the learned air interface outperforms 5G NR baseline by 3 dB at 10^-3 BER in frequency-selective fading channels.
AI Summary
- Complete physical layer learned end-to-end using deep autoencoders.
- Outperforms 5G NR baseline by 3 dB in real over-the-air experiments.
- Joint optimization of modulation, coding, and equalization.
- Demonstrates feasibility of AI-native air interface for 6G.
Key Findings
- 1End-to-end learning discovers non-obvious signal constellations optimal for specific channels.
- 2The system adapts to channel changes by retraining only the decoder.
- 3Latency overhead of neural network inference is within acceptable bounds for mobile.
Industry Implications
Fundamentally challenges the traditional modular PHY design approach.
Could lead to 6G systems that continuously improve through learning.
Requires new standardization approaches for AI-native air interfaces.
Read the Original Paper
Access the full paper on arXiv for complete methodology, results, and references.
Open on arXivRelated Papers
Digital Twin Networks: AI-Driven Real-Time Network Simulation for 6G
Oulu University / Ruhr University Bochum — 29 citations
AI + Network PapersIntent-Based Network Management with Large Language Models
Universidad Carlos III de Madrid — 16 citations
AI + Network PapersAI-Driven Energy Optimization in 6G Networks: A Multi-Objective Approach
Poznan University of Technology / Princeton University — 22 citations