Performance of Predictive Indoor mmWave Networks with Dynamic Blockers

TL;DR

A novel beam recovery procedure that leverages Machine Learning (ML) tools to predict the starting and finishing of blockage events, which erases the delay introduced by current 5G New Radio (5G-NR) procedures when switching to an alternative serving base station and beam, and then re-establish the primary connection after the blocker has moved away.

Abstract

In this paper, we consider millimeter Wave (mmWave) technology to provide reliable wireless network service within factories where links may experience rapid and temporary fluctuations of the received signal power due to dynamic blockers, such as humans and robots, moving in the environment. We propose a novel beam recovery procedure that leverages Machine Learning (ML) tools to predict the starting and finishing of blockage events. This erases the delay introduced by current 5G New Radio (5G-NR) procedures when switching to an alternative serving base station and beam, and then re-establish the primary connection after the blocker has moved away. Firstly, we generate synthetic data using a detailed system-level simulator that integrates the most recent 3GPP 3D Indoor channel models and the geometric blockage Model-B. Then, we use the generated data to train offline a set of beam-specific Deep Neural Network (DNN) models that provide predictions about the beams' blockage states. Finally, we deploy the DNN models online into the system-level simulator to evaluate the benefits of the proposed solution. Our prediction-based beam recovery procedure guarantee higher signal level stability and up to $82\%$ data rate improvement with respect detection-based methods when blockers move at speed of $2$ m/s.

Figures (10)

• Figure 1: Layout of the indoor mmWave network.
• Figure 2: (a) Top-down view of the area in the x-y plane illuminated by the Tx beams of the BS-3. (b) Structure of the UPA formed by $M_{Tx}^{\rm{H}} \times M_{Tx}^{\rm{V}}$ antennas and placed in the local coordinates system $(x',y',z')$. The Tx beam has azimuth $\theta_{Tx}$ defined between the x' axis and the Tx beam projection on the x'-y' plane and elevation $\phi_{Tx}$ defined between the z' axis and the Tx beam direction.
• Figure 3: Beam switching operations of the method based on detection according to the temporal variations of the SNR values for primary and backup beams.
• Figure 4: Key steps of the procedure used to obtain the beam state predictions.
• Figure 5: Two examples of input-output vectors forming the dataset for the beam depicted with the line pattern (magenta) on the left side of Figs \ref{['fig:dataMappingInputNonBlocked']} and \ref{['fig:dataMappingInputBlocked']}. In Fig. \ref{['fig:dataMappingInputNonBlocked']}, the input data are taken at a time $t_1-\eta$ from the ${\mathrm{SNR}}$ measurements and correspond to the non-blocked state at the time $t_1$ in Fig. \ref{['fig:dataMappingOutput']} . Conversely, in Fig. \ref{['fig:dataMappingInputBlocked']}, the input data are taken at a time $t_2-\eta$ and correspond to the blocked state at the time $t_2$ in Fig. \ref{['fig:dataMappingOutput']}.