Autonomous Discovery of Wireless Communications Algorithms

As wireless communication systems grow in complexity, designing efficient algorithms—such as equalizers and receivers—becomes increasingly difficult for human engineers. This paper investigates whether Large Language Model (LLM)-driven evolutionary search can automate the discovery, implementation, and optimization of these algorithms, effectively navigating the trade-off between performance and computational complexity.

The authors introduce The AI Telco Engineer (AITE), an open-source framework that treats algorithm design as a programming problem. AITE uses a two-tier architecture: an orchestrator that proposes diverse algorithmic ideas and a pool of parallel workers that implement these ideas using agentic coding loops. Unlike traditional genetic programming, AITE leverages the reasoning capabilities of LLMs to generate, refine, and debug code. The framework evaluates candidate algorithms using a task-specific tool that measures both performance (e.g., bit error rate) and complexity (e.g., latency), ensuring that the system discovers a Pareto front of solutions rather than a single, potentially inefficient, algorithm.

The authors tested AITE on two distinct physical-layer problems. First, for an orthogonal time-frequency space (OTFS) system, AITE discovered an equalizer that outperformed the best-known baseline while reducing computational latency by a factor of 3.6. Second, for an orthogonal frequency-division multiplexing (OFDM) system operating without pilots, AITE generated the first explicit, explainable algorithms that achieved performance parity with state-of-the-art neural network-based receivers. These results suggest that agentic AI can effectively automate the discovery of next-generation communication algorithms.

This research demonstrates that LLM-driven evolutionary search has reached a capability threshold where it can produce novel, high-performance scientific results in specialized engineering domains. By automating the discovery of complex algorithms, AITE reduces the reliance on manual mathematical derivation and trial-and-error experimentation, potentially accelerating the development cycle for future wireless communication standards.