Cổng tri thức PTIT

Bài báo quốc tế

Kho tri thức

/

/

Q-Learning for Adaptive MMSE Regularization in 1-Bit Massive MIMO

Q-Learning for Adaptive MMSE Regularization in 1-Bit Massive MIMO

Đặng Ngọc Hùng

This paper develops a Q-Learning framework for adaptive MMSE regularization in massive MIMO systems with 1-bit ADCs. The receiver adjusts its regularization parameter to instantaneous channel conditions via a tabular agent operating on a 6-dimensional state and 75 discrete actions. Evaluation on 20,000 independent test samples shows consistent gains: 1.27–1.70% over the best Bussgang-MMSE baseline and up to 47% over MRC across antenna settings N ∈ {16, 32, 64, 96, 128}. Training exhibits stable learning with the positive-reward rate rising from 16.1% to 49.8% over 20,000 episodes. This gain is achieved with negligible inference overhead, adding only an O(1) table lookup to the standard O(N3) MMSE complexity.

Xuất bản trên:

Q-Learning for Adaptive MMSE Regularization in 1-Bit Massive MIMO

Ngày đăng:

DOI:


Nhà xuất bản:

Địa điểm:


Từ khoá:

Massive MIMO, 1-bit ADCs, adaptive MMSE regularization, Q-Learning, Bussgang decomposition, sum-rate.