Description:
Reference #: 1707
The University of South Carolina is offering licensing opportunities for Methods for Over-the-air Computation with Modulation on Conjugate-Reciprocal Zeros.
Background:
Over-the-air computation (OAC) refers to the computation of special functions like arithmetic mean, norm, polynomial function, maximum, and majority vote (MV) by harnessing the signal superposition property of wireless multiple-access channels. With OAC, instead of acquiring information from each device independently, the transmitters’ signals are intentionally overlapped on the same time-frequency resources to realize the summation operation as part of a desired function. Hence, OAC can improve resource utilization while reducing latency when the ultimate goal of communication is computation. With more applications relying on computation over wireless networks, OAC has recently been applied to a wide range of applications such as wireless federated learning, distributed optimization, distributed localization, wireless data centers, and wireless control systems
Invention Description:
This invention is a new OAC approach based on modulation on conjugate-reciprocal zeros (MOCZ), and utilizes several methods to compute the MV function without channel state information (CSI) at the transmitters and receiver. Each transmitter maps specific data and transmits this data to the receiver. The receiver evaluates the data and detects the MV value with a low-complexity testing decoder.
Potential Applications:
Currently, over-the-air computation is particularly investigated for wireless federated learning. Google AI, Intellegens, and IBM have several projects that are considering federated learning. The idea of federated learning and over-the-air computation have already been discussed in in IEEE 802.11 Wi-Fi meetings and it is well known concepts that be applied in upcoming 6G networks. Another development is the recently formed AI-RAN Alliance, where major companies (like NVIDIA) participate. The Alliance has recently discussed deploying AI services at the network edge through RAN to increase operational efficiency and offer new services to mobile users. This invention is well aligned to AI-on-RAN concept.
Advantages and Benefits:
This invention can reduce latency a magnitude order and improve spectral efficiency for training neural networks over a wireless network. The proposed methods also do not use instantaneous channel state information at the transmitters and receiver and are thus compatible with time-varying channels and provide robustness against phase and time synchronization errors while improving the computation error rate (CER) performance.