[Paper Review] Machine Learning for Predictive On-Demand Deployment of UAVs for Wireless Communications

Qianqian Zhang, Mohammad Mozaffari+3 more|arXiv (Cornell University)|Apr 30, 2018

UAV Applications and OptimizationEngineering12 references19 citations

TL;DR

This paper proposes a machine learning framework using Gaussian Mixture Models (GMM) and Weighted Expectation Maximization (WEM) to predict cellular traffic hotspots and enable predictive, on-demand deployment of UAVs as aerial base stations. The system optimizes UAV service area partitioning and location to minimize total power consumption (transmission + mobility), achieving over 20% improvement in power efficiency compared to non-predictive deployment.

ABSTRACT

In this paper, a novel machine learning (ML) framework is proposed for enabling a predictive, efficient deployment of unmanned aerial vehicles (UAVs), acting as aerial base stations (BSs), to provide on-demand wireless service to cellular users. In order to have a comprehensive analysis of cellular traffic, an ML framework based on a Gaussian mixture model (GMM) and a weighted expectation maximization (WEM) algorithm is introduced to predict the potential network congestion. Then, the optimal deployment of UAVs is studied to minimize the transmit power needed to satisfy the communication demand of users in the downlink, while also minimizing the power needed for UAV mobility, based on the predicted cellular traffic. To this end, first, the optimal partition of service areas of each UAV is derived, based on a fairness principle. Next, the optimal location of each UAV that minimizes the total power consumption is derived. Simulation results show that the proposed ML approach can reduce the required downlink transmit power and improve the power efficiency by over 20%, compared with an optimal deployment of UAVs with no ML prediction.

Motivation & Objective

• Address the challenge of on-demand, energy-efficient wireless coverage in cellular hotspots caused by temporary traffic surges.
• Overcome limitations of static or non-predictive UAV deployment by enabling proactive, data-driven UAV positioning.
• Minimize total power consumption in UAV-assisted networks by jointly optimizing downlink transmission and UAV mobility energy.
• Develop a scalable, fair service area partitioning strategy for UAVs based on predicted user demand.
• Demonstrate the performance gain of predictive ML-based UAV deployment over non-predictive optimal deployment.

Proposed method

• Employ a Gaussian Mixture Model (GMM) with Weighted Expectation Maximization (WEM) to predict spatiotemporal patterns of cellular traffic and identify potential congestion zones.
• Formulate a power minimization problem that jointly optimizes UAV transmit power and mobility power based on predicted traffic.
• Derive optimal UAV service area partitions using a fairness-based principle to balance user load across UAVs.
• Compute optimal UAV locations via closed-form expressions derived from minimizing the sum of transmit and mobility power, using weighted integrals over user distribution.
• Use a gradient-based algorithm to solve the non-convex optimization problem for joint transmit and mobility power minimization.
• Validate the framework using simulations with realistic parameters: 5 GHz band, 10 MHz bandwidth, and 0.1 J/m mobility cost.

Experimental results

Research questions

• RQ1 Can machine learning techniques improve the energy efficiency of UAV-assisted cellular networks by enabling predictive deployment?
• RQ2 How does predictive traffic forecasting via GMM and WEM impact the required downlink transmit power and UAV mobility energy?
• RQ3 What is the optimal service area partitioning strategy for UAVs that balances fairness and power efficiency?
• RQ4 How do UAV location and number of UAVs affect total network power consumption under predicted traffic?
• RQ5 To what extent does ML-based prediction outperform non-predictive optimal deployment in terms of power efficiency?

Key findings

• The proposed ML-based predictive deployment reduces required downlink transmit power by 20.68% to 24.40% compared to non-predictive optimal deployment as UAV count increases from 9 to 36.
• Power efficiency improves by up to 22.34% with the ML-based approach, due to better alignment of UAV deployment with predicted traffic hotspots.
• As the number of UAVs increases, total transmit power and average power per UAV decrease due to reduced path loss from smaller service areas.
• Despite lower transmit power with more UAVs, power efficiency decreases due to increased mobility energy from longer travel distances to remote hotspots.
• The WEM-based GMM effectively captures complex, multi-modal traffic patterns, enabling accurate hotspot prediction for proactive UAV deployment.
• The derived optimal UAV location formulas (Equations 20a and 20b) provide a closed-form solution for minimizing transmit power based on user density and path loss.