Core Concepts
Proposing a learning-based detection framework for uplink massive MIMO systems with one-bit ADCs, utilizing dithering to overcome under-trained likelihood functions.
Abstract
The paper introduces a learning-based approach for one-bit maximum likelihood detection in massive MIMO systems. It addresses the challenges of under-trained likelihood functions by incorporating dithering signals and adaptive learning techniques. The content is structured into sections covering system models, naive detection methods, adaptive statistical learning without CSI, and extension to channel coding. Simulation results validate the proposed methods' performance in uncoded and coded scenarios.
System Model:
Massive MIMO systems with one-bit ADCs.
Channel estimation challenges addressed through learning-based approaches.
Naive Detection:
Counting-based method for estimating likelihood probabilities.
Under-trained likelihood functions identified as a critical issue at high SNR.
Adaptive Statistical Learning:
Dither-and-learning technique introduced to prevent under-trained likelihood functions.
Adaptive dithering power update based on feedback from observations.
Extension to Channel Coding:
Introduction of frame structure for channel-coded communication frameworks.
Calculation of soft metrics using trained likelihood probabilities for LLR computation in decoding process.
Stats
The proposed method aims to address the challenges of under-trained likelihood functions by incorporating dithering signals and adaptive learning techniques.
The number of under-trained likelihood functions among 2Nr likelihood functions is evaluated for Nu = 4 users, 4-QAM, Nr = 32 antennas, and Ntr = 45 pilot signals with Rayleigh channels.
The proposed adaptive dither-and-learning (ADL) method divides the training period into Ns ∈{1, 3, 5} sub-blocks for feedback-driven updates of dithering power.