Robust Hybrid Beamforming with Quantized Deep Neural Networks
Yükleniyor...
Dosyalar
Tarih
2019
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
IEEE Computer Society
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Hybrid beamforming is integral to massive multiple-input multiple-output (MIMO) communications in reducing the training overhead and hardware cost associated with large antenna arrays. Prior works have employed optimization and greedy search to jointly estimate the precoder and combiner weights. High computational complexity of these methods apart, their performance strongly relies on accurate channel information. In this paper, we propose a computationally efficient, deep learning approach that also provides robust performance against the deviations in the channel characteristics. Further, we employ a convolutional neural network with quantized weights (Q-CNN) so that it is deployable in mobile devices that have less memory resources and low overhead requirements. We show that the proposed Q-CNN, saved in at least 6 bits, yields superior performance over conventional massive MIMO hybrid beamforming. © 2019 IEEE.
Açıklama
29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019 -- 13 October 2019 through 16 October 2019 -- 155874
Anahtar Kelimeler
Convolutional neural networks; deep learning; hybrid beamforming; massive MIMO; quantization
Kaynak
IEEE International Workshop on Machine Learning for Signal Processing, MLSP
WoS Q Değeri
Scopus Q Değeri
N/A
Cilt
2019-October
