Methods for Reliable Over-the-Air Computation with Pulses for Distributed Learning


Reference #: 01541

The University of South Carolina is offering licensing opportunities for Methods for Reliable Over-the-Air Computation with Pulses for Distributed Learning


Federated edge learning (FEEL) is an implementation of federated learning (FL) over a wireless network to train a model by using the local data at the edge devices (EDs) without uploading them to an edge server (ES). Within each iteration of FEEL, a substantial number of parameters (e.g., model parameters or model updates) from each ED needs to be transmitted to the ES for aggregation. Thus, the communication aspect of FEEL is one of the major bottlenecks.  

Invention Description:

One of the promising solutions to this issue is to perform the aggregation by utilizing the signal-superposition property of a wireless multiple access channel, i.e., over-the-air computation (AirComp). However, an AirComp scheme often requires channel state information (CSI) at either the EDs or ES to maintain coherent superposition of the signals from EDs, which can cause a non-negligible overhead and unreliable aggregation in a mobile wireless network. In this work, we address this issue with a new AirComp method.

Potential Applications: 

This invention addresses the communication latency problem of training an artificial intelligence model over a wireless network. It reduces the latency with over-the-air computation when there are many users. The invention does not use the channel information (e.g., channel frequency response) needed for wireless communications at the edge devices (e.g., a user) or edge server (e.g., a base station).

Advantages and Benefits:

1) The proposed scheme does not need a channel inversion at the EDs. From this aspect, it is compatible with time-varying channels or mobile networks including drones, cars, or satellites. 2) It does not lose the gradient information due to the truncation. 3) The proposed scheme reduces PMEPR as it uses pulses. 5) The PMEPR can be adjusted based on the resources in time, i.e., offer flexibility. 4) it also does not require CSIs at the ES or multiple antennas for over-the-air computation.

Patent Information:
For Information, Contact:
Technology Commercialization
University of South Carolina
Alphan Sahin
Everette Bryson
Safi Shams Muhtasimul Hoque
Distributed learning
federated edge learning
orthogonal frequency division multiplexing
over-the-air computation
peak-to-mean envelope power ratio
pulse-position modulation
© 2024. All Rights Reserved. Powered by Inteum