Machine Intelligence on Wireless Edge Networks

2025-06-13

AI Summary

To address power and memory bottlenecks on edge devices, this paper proposes MIWEN, an RF-based analog computing architecture that performs classification directly in the analog domain at the RF front-end. It enables wireless streaming of weights, eliminating on-chip weight storage and ADC/DAC overhead entirely. MIWEN introduces the first wireless-decoupled memory architecture; achieves the first RF-carrier-based joint encoding and computation of weights and activations; and establishes a theoretical model for effective bit resolution in RF analog computing, incorporating thermal noise. Leveraging native mixer reuse from standard transceivers and joint energy–accuracy optimization, MIWEN attains digitized DNN-comparable accuracy on MNIST while reducing energy consumption by two orders of magnitude. It supports real-time, ultra-low-power edge inference without on-chip memory.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Analogy
• Computer Vision: Learning & Optimization for CV

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Deep neural network (DNN) inference on power-constrained edge devices is bottlenecked by costly weight storage and data movement. We introduce MIWEN, a radio-frequency (RF) analog architecture that ``disaggregates'' memory by streaming weights wirelessly and performing classification in the analog front end of standard transceivers. By encoding weights and activations onto RF carriers and using native mixers as computation units, MIWEN eliminates local weight memory and the overhead of analog-to-digital and digital-to-analog conversion. We derive the effective number of bits of radio-frequency analog computation under thermal noise, quantify the energy--precision trade-off, and demonstrate digital-comparable MNIST accuracy at orders-of-magnitude lower energy, unlocking real-time inference on low-power, memory-free edge devices.

Problem

• Reducing weight storage and data movement in DNN edge inference
• Enabling memory-free edge devices via wireless weight streaming
• Achieving energy-efficient analog computation with RF carriers

Innovation

• RF analog architecture for wireless weight streaming
• Analog computation using native RF mixers
• Energy-efficient DNN inference without local memory