Problem
Research questions and friction points this paper is trying to address.
Wireless physical neural networks
MIMO relay
Power amplifier nonlinearity
Activation function
Over-the-air inference
Multi-layer MIMO Relay as Deep Physical Neural Networks: Power Amplifiers as Activation Functions
🤖 AI Summary
This work addresses the high energy consumption and latency associated with nonlinear activation functions in wireless physical neural networks by proposing a deep over-the-air computing architecture based on multi-hop MIMO relaying. The approach leverages trainable complex-valued gains and biases at relay nodes and, for the first time, directly exploits the inherent hardware nonlinearity of power amplifiers as the activation function, thereby constructing an end-to-end trainable fully connected network within a cascaded structure. Combined with least-squares (LS) and singular value decomposition (SVD)-based transceiver designs, the system achieves high-accuracy over-the-air inference on image classification tasks, demonstrating that hardware-induced nonlinearity can effectively enhance model expressiveness while reducing computational overhead.
📝 Abstract
Wireless physical neural networks (WPNNs) embed neural computation directly into analog hardware, offering lower energy consumption and latency than conventional digital implementations. In this paper, we propose a deep WPNN in which nonlinear activations are realized by a multi-hop multiple-input multiple-output (MIMO) relay network, in which each relay implements a trainable complex linear gain and bias, followed by the power amplifier's intrinsic nonlinearity acting as an activation function. The cascade of multiple relays therefore realizes an over-the-air fully connected network whose parameters can be trained end-to-end. We develop two transceiver designs for different channel state information (CSI) availability scenarios: a least squares (LS)-based scheme requiring only receiver-side CSI, and a singular-value-decomposition (SVD)-based scheme requiring both transmitter-side and receiver-side CSI. Simulation results show that the proposed architecture enables accurate over-the-air inference for image classification. In particular, the results highlight the advantage of exploiting hardware nonlinearity for enhanced inference capability.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Wireless Physical Neural Networks
MIMO Relay
Power Amplifier Nonlinearity
Over-the-Air Inference
Hardware-Based Activation
🔎 Similar Papers
No similar papers found.
