Intelligence-aided transceiver design for millimeter wave communications

Abstract

The millimeter wave (mmWave) frequency band spanning from 30 GHz to 300 GHz offers a tremendous potential frequency resource for future wireless communication systems in order to meet the ever-increasing capacity demand. However, an important challenge at mmWave frequencies is that they suffer from high propagation loss because of the oxygen absorption, rain-induced fading and foliage attenuation. In order to mitigate the propagation losses, directional transmission would be employed. Furthermore, owing to the high cost and high hardware complexity as well as power-hungry nature of analog-to-digital and digital-to-analog converters, a hybrid beamforming architecture relying on both analog beamforming and digital precoding is conceived. However, the impairments in the analog beamforming employed at the radio-frequency (RF) would limit the achievable rate of mmWave hybrid systems. Therefore, we enhance the achievable data rate of mmWave hybrid systems with the aid of diversity and beamforming schemes. More explicitly we conceive a dual-function hybrid beamforming transceiver, where both diversity and beamforming gains can be attained. This is achieved by separating the sub-arrays emerging from a full array by a sufficiently large distance so the correlation between the sub-arrays is minimum. Then to further enhance the data rate, we considered full-duplex communication at mmWave frequencies relying on hybrid beamforming, where we aimed for mitigating the self-interference (SI) by jointly designing the RF transmit and receive beamformer weights and the precoder as well as combiner matrices. The proposed solution preserves the signal's dimensionality, while mitigating the SI. We show that the proposed design is capable of reducing the SI by up to 30 dB, hence performing similarly to the hypothetical interference-free FD system.

However, we note that the aforementioned designs rely on the assumption of having perfect beam-alignment and channel impulse response knowledge. Therefore, in order to relax this assumption, we invoke machine learning tools for intelligent beam-alignment. Publications based on the location of the user, where we demonstrate that our design performs similarly to beam-sweeping based beam-alignment relying on high-complexity exhaustive beam-search. Furthermore to counteract the channel aging phenomenon, we propose learning-assisted channel prediction. In order to reduce the signalling overhead and detector complexity at the receiver, we propose deep learning assisted semi-blind detection for index modulation mmWave MIMO systems. More particularity, we propose a detector for index modulation systems operating without relying on the explicit knowledge of the CSI at the receiver. We observe by our simulations that the number of computations required for learning-assisted soft-detection is four times lower than that of the conventional soft-detector.

More information

Identifiers

ORCID for K. Satyanarayana: orcid.org/0000-0002-5411-3962

ORCID for Lajos Hanzo: orcid.org/0000-0002-2636-5214

Catalogue record

Export record