Methods for Long-Range Federated Edge Learning with Chirp-Based Over-the-Air Computation

Reference #: 01574

The University of South Carolina is offering licensing opportunities for Methods for Long-Range Federated Edge Learning with Chirp-Based Over-the-Air Computation

Background:

Federated edge learning (FEEL) deploys federated learning (FL) over a wireless network, in which many edge devices (EDs) participate in training using locally accessible data and an edge server (ES) aggregates the local decisions without accessing the data at the EDs. With FEEL, the communication load can be a challenge as a significant number of model parameters/gradients/updates need to be exchanged between the ES and the EDs over the wireless channel. The conventional orthogonal multiple access techniques require more spectral resources as the number of EDs grows. Hence, the spectral congestion problem limits the scalability of FEEL.

Invention Description:

This invention reduces the communication latency of training an artificial intelligence model over a wireless network with a low peak-to-mean envelope power ratio (PMEPR) over-the-air computation scheme. The innovation allows for a wider cell coverage due to the low PMEPR.

Potential Applications:

The technology could be useful for artificial intelligence technologies over wireless or sensor networks, 5G and beyond, 6G wireless standardization, IEEE 802.11 Wi-Fi.

Advantages and Benefits:

The proposed scheme utilizes low PMEPR circularly-shifted chirps to transmit gradient information. This makes the scheme suitable for long-range (LoRa) applications. Also, the scheme does not rely on the availability of the channel state information (CSI) to operate.

Patent Information: