One-Bit Precoding Constellation Design via Autoencoder-Based Deep Learning

This paper considers a multicasting system in which the base station has a large number of antennas with cost-effective one-bit digital-to-analog converters and aims to send a common symbol to multiple remote users. Unlike the existing literature which seeks to design the one-bit precoder for a give...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Conference record - Asilomar Conference on Signals, Systems, & Computers s. 754 - 758
Hlavní autori: Sohrabi, Foad, Yu, Wei
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 01.11.2019
Predmet:
autoencoder
Autoencoders
binary neurons
deep neural network
Digital-analog conversion
MIMO
Multicast communication
multiple-input multiple-output
Neurons
one-bit precoding
Phase shift keying
Precoding
Quadrature amplitude modulation
Symbols
Training
ISSN:2576-2303
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:This paper considers a multicasting system in which the base station has a large number of antennas with cost-effective one-bit digital-to-analog converters and aims to send a common symbol to multiple remote users. Unlike the existing literature which seeks to design the one-bit precoder for a given constellation, e.g., quadrature amplitude modulation (QAM) or phase shift keying (PSK), this paper aims to jointly design the transmit one-bit precoder and the receive constellation by leveraging the concept of autoencoder, wherein the end-to-end multicasting system is modeled using a deep neural network with the one-bit precoding constraint represented by a binary thresholding layer. To deal with the issue that such a binary layer always produces a gradient of zero, and thus prevents an effective end-to-end training when using the conventional back-propagation method, this paper uses a variant of straight-through estimator which approximates the thresholding function with a properly scaled sigmoid function in the back-propagation phase. Numerical results show that, for a fixed channel scenario, the proposed autoencoder-based constellation design is superior to the conventional QAM and PSK constellations. Using the insights obtained from fixed channel scenarios, we also propose a constellation design for varying channel scenarios and numerically show that the proposed design achieves a better performance as compared to the conventional constellations.
ISSN:2576-2303
DOI:10.1109/IEEECONF44664.2019.9048810