A Low-Complexity DNN-Based DoA Estimation Method for EHF and THF Cell-Free Massive MIMO

We study the problem of direction of arrival (DoA) estimation for cell-free massive MIMO (m-MIMO) systems operating over extremely high frequency (EHF) and terahertz (THF) bands, where the wireless channel can effectively be modeled by a line-of-sight path. For this model, a low-complexity deep neur...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE Vehicular Technology Conference s. 1 - 7
Hlavní autori: Hosseini, Seyyed Saleh, Champagne, Benoit, Chang, Xiao-Wen
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 01.09.2022
Predmet:
cell-free massive MIMO
Direction-of-arrival estimation
DoA estimation
Estimation
LoS communications
neural network
Neural networks
Quantization (signal)
Training
Vehicular and wireless technologies
Wireless communication
ISSN:2577-2465
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:We study the problem of direction of arrival (DoA) estimation for cell-free massive MIMO (m-MIMO) systems operating over extremely high frequency (EHF) and terahertz (THF) bands, where the wireless channel can effectively be modeled by a line-of-sight path. For this model, a low-complexity deep neural network (DNN)-based method is proposed to estimate the DoA of a radio wave impinging on an access point (AP) equipped with an antenna array. To train the DNN, a special feature set is proposed obtained from the first superdiagonal entries of the spatial correlation matrix. This selection of features makes it possible to employ a DNN with only a few low-dimensional layers, which considerably speeds up training and processing. More importantly, it is shown that the trained DNN is robust against quantization noise in the array snapshot data. This property makes the centralized implementation of the proposed DNN-based method feasible, which is particularly well-suited for cell-free m-MIMO. Through extensive simulations, the new method is shown to achieve an estimation performance that nearly matches or exceeds that of classical bechmark methods, but with considerably reduced complexity.
ISSN:2577-2465
DOI:10.1109/VTC2022-Fall57202.2022.10012869