Deep MIMO Detection Using ADMM Unfolding
Refereed conference paper presented and published in conference proceedings


Times Cited
Altmetrics Information
.

Other information
AbstractThis paper presents a low-complexity deep neural network (DNN)based multiple-input-multiple-output (MIMO) detector for the BPSK and QPSK constellation cases. We employ deep unfolding, whose idea is to take insight from the structure of an iterative optimization algorithm and attempt to learn a better iterative algorithm. The structure of the network is obtained from an iterative algorithm arising from the application of ADMM to the maximum-likelihood MIMO detection problem. The number of parameters to be learnt in this new design is less than that of DetNet, a recently proposed DNN-based MIMO detector. Our numerical experiments illustrate that the new method outperforms DetNet and several existing MIMO detectors in the large-scale MIMO case. In particular, we show that for a 160×160 MIMO system, our DNN design, with 40 layers, can attain nearly optimal bit-error rate performance.
Acceptance Date14/04/2019
All Author(s) ListMan-Wai Un, Mingjie Shao, Wing-Kin Ma, P. C. Ching
Name of Conference2019 IEEE Data Science Workshop (DSW)
Start Date of Conference02/06/2019
End Date of Conference05/06/2019
Place of ConferenceMinneapolis, MN
Country/Region of ConferenceUnited States of America
Proceedings Title2019 IEEE Data Science Workshop (DSW)
Year2019
Month6
PublisherIEEE
Pages333 - 337
ISBN978-1-7281-0708-0
LanguagesEnglish-United States
KeywordsMIMO detection, deep learning, neural networks, deep unfolding, ADMM

Last updated on 2020-28-11 at 02:24