Optimized Transceiver Design for Over-the-Air Distributed Computation over Cell-Free Massive MIMO Network

Abstract

Our paper presents a MapReduce-based wireless distributed computing framework designed to handle data-intensive computing on edge devices with limited storage. The framework involves three stages: Map, Shuffle, and Reduce. However, shuffling large data during the second stage can lead to performance degradation over wireless interference networks with limited spectrum bandwidth. To address these issues, we propose using over-the-air computation (AirComp) technology, which leverages interference in the multiple-access channel to compute multiple target functions reliably. This approach achieves higher computation efficiency than traditional orthogonal multi-access schemes and is more effective in combating interference. Furthermore, we employ cell-free massive MIMO technology to improve coverage and reduce the system power overhead. This technology is essential for the upcoming sixth-generation (6G) networks. We optimize the transmitting-receiving (Tx-Rx) policy to minimize the averaged computation mean squared error (MSE) while adhering to each device's power constraint. Our simulation results demonstrate that our proposed algorithm is effective and our computation framework has advantages over state-of-the-art baselines.