[Paper Review] Autoencoder for Interconnect's Bandwidth Relaxation in Large Scale MIMO-OFDM Processing

Messaoud Ahmed Ouameur, Daniel Massicotte|arXiv (Cornell University)|Jul 29, 2019

Radio Frequency Integrated Circuit Design17 references4 citations

TL;DR

This paper proposes using an autoencoder to compress massive MIMO-OFDM signals at the radio unit, enabling a 16-fold reduction in interconnect bandwidth by transmitting only the low-dimensional latent code. The decoder at the central unit reconstructs the signal with minimal performance loss, which is further mitigated by iterative detection, offering a practical trade-off between bandwidth efficiency and computational complexity.

ABSTRACT

Deep learning is playing an instrumental role in the design of the next generation of communication systems. In this letter, we address the massive MIMO interconnect's bandwidth constraint relaxation using autoencoders. The autoencoder is trained to learn the received signal structure so that a low dimension latent variable is transferred as opposed to the original high dimension signal. For an efficient implementation, the approach suggests to separately deploy the autoencoder components, namely the encoder and the decoder, in massive MIMO radio head and central processing units respectively. The simulation results show that one can relax the interconnect's bandwidth by a factor of up to 16 with non-substantial performance degradation of the centralized and the decentralized processing. Fortunately, such a loss can be compensated by running few extra iterations in the detection process which renders the autoencoder-iterative detection an interconnect's bandwidth and computational complexity design trade-off.

Motivation & Objective

• To address the high-bandwidth interconnect bottleneck in large-scale MIMO-OFDM systems.
• To reduce the data rate required between remote radio heads and central processing units.
• To explore the feasibility of using deep learning for signal compression in massive MIMO systems.
• To evaluate the trade-off between interconnect bandwidth reduction and system performance degradation.
• To propose a distributed implementation of autoencoder components across radio and central units.

Proposed method

• An autoencoder is trained to learn the statistical structure of the received MIMO-OFDM signal.
• The encoder is deployed at the radio unit to compress the high-dimensional signal into a low-dimensional latent code.
• The decoder is deployed at the central unit to reconstruct the original signal from the latent code.
• The system uses iterative detection to compensate for reconstruction loss.
• The autoencoder is trained end-to-end on real MIMO-OFDM signal distributions.
• The architecture separates encoder and decoder components for efficient hardware deployment.

Experimental results

Research questions

• RQ1 Can an autoencoder effectively compress MIMO-OFDM signals to reduce interconnect bandwidth?
• RQ2 What is the maximum achievable bandwidth reduction without significant performance degradation?
• RQ3 How does the reconstruction error affect detection performance in centralized and decentralized processing?
• RQ4 Can iterative detection compensate for the performance loss introduced by compression?
• RQ5 What is the trade-off between bandwidth savings and computational complexity?

Key findings

• The proposed method achieves up to a 16-fold reduction in interconnect bandwidth.
• The performance degradation due to compression is non-substantial in both centralized and decentralized detection schemes.
• The performance loss can be effectively compensated by running a few additional iterations in the detection process.
• The distributed autoencoder architecture enables practical deployment with minimal overhead.
• The approach offers a viable trade-off between interconnect bandwidth and computational complexity.
• The results demonstrate the feasibility of using deep learning for efficient signal compression in massive MIMO systems.