New AI-Powered Approach Enables Massive Uncoordinated Devices to Share Wireless Spectrum

A novel wireless communication framework harnesses large pretrained models to enable efficient spectrum sharing among vast numbers of uncoordinated devices. The approach, called token-domain multiple access (ToDMA), treats digital information as compact semantic tokens rather than traditional bits, allowing receivers to reconstruct messages even when transmissions collide.

How Token-Based Communication Works#

Instead of sending raw data, each message is first converted into a sequence of discrete tokens—compact units that capture semantic meaning. These tokens are then assigned to shared modulation codewords, which both encodes the data and exposes its underlying structure to the receiver. This token-centric design creates an opportunity for intelligent recovery: when multiple devices transmit simultaneously, the receiver can still identify which tokens were sent and work backward to reconstruct the full message.

Handling Collisions with Context#

The heart of the system lies in managing situations where multiple devices transmit tokens at the same time. When collisions occur, the receiver first uses compressed sensing techniques to detect which tokens are present and measure the radio channel conditions associated with each. However, some tokens may become unassigned due to overlapping signals, creating gaps in the reconstructed sequences.

Rather than simply accepting these gaps, ToDMA fills them using candidate-restricted masked-token prediction—a technique borrowed from large language models. The pretrained contextual model examines the surrounding tokens and predicts what the missing entries should be, much like autocomplete in text applications. This leverages the inherent structure and dependencies within natural sequences to mitigate collision effects.

Performance and Scalability#

Testing on both image and text transmission tasks shows that ToDMA simultaneously reduces latency while maintaining high-quality token recovery and semantic reconstruction. By enabling massive uncoordinated access—thousands of devices can transmit without coordination—the system offers clear advantages for Internet-of-Things scenarios and other applications where coordinating access is impractical.

The research demonstrates how semantic communication principles, grounded in modern large models, can overcome fundamental limitations of conventional multiple access schemes.