[Paper Review] Machine Learning for Predictive Deployment of UAVs with Multiple Access

Lin-Yan Lu, Zhaohui Yang+3 more|arXiv (Cornell University)|Mar 2, 2020

UAV Applications and Optimization39 references4 citations

TL;DR

This paper proposes a machine learning-driven UAV deployment framework that predicts cellular traffic using LSTM and clusters user service areas via a novel KEG algorithm (K-means and EM-based GMM). It optimizes UAV positions to minimize total transmit power, showing up to 24% power reduction versus conventional methods, with RSMA outperforming FDMA and TDMA in power efficiency.

ABSTRACT

In this paper, a machine learning based deployment framework of unmanned aerial vehicles (UAVs) is studied. In the considered model, UAVs are deployed as flying base stations (BS) to offload heavy traffic from ground BSs. Due to time-varying traffic distribution, a long short-term memory (LSTM) based prediction algorithm is introduced to predict the future cellular traffic. To predict the user service distribution, a KEG algorithm, which is a joint K-means and expectation maximization (EM) algorithm based on Gaussian mixture model (GMM), is proposed for determining the service area of each UAV. Based on the predicted traffic, the optimal UAV positions are derived and three multi-access techniques are compared so as to minimize the total transmit power. Simulation results show that the proposed method can reduce up to 24\% of the total power consumption compared to the conventional method without traffic prediction. Besides, rate splitting multiple access (RSMA) has the lower required transmit power compared to frequency domain multiple access (FDMA) and time domain multiple access (TDMA).

Motivation & Objective

• To address the challenge of dynamic, time-varying cellular traffic in UAV-assisted networks by enabling proactive UAV deployment.
• To improve energy efficiency and network capacity by minimizing total uplink transmit power through optimal UAV positioning.
• To develop a predictive framework that integrates traffic forecasting and service area clustering for intelligent UAV deployment.
• To compare the performance of multiple access techniques—RSMA, FDMA, and TDMA—under the proposed deployment strategy.
• To validate the robustness and scalability of the framework across different geographical areas, including limited and entire urban regions.

Proposed method

• Uses a Long Short-Term Memory (LSTM) network to predict future cellular traffic based on historical data.
• Proposes a KEG algorithm that combines K-means clustering and Expectation Maximization (EM) to model user service distribution using a Gaussian Mixture Model (GMM).
• Defines aerial cells as non-overlapping, fully covering regions, with UAVs positioned at cluster centroids to minimize transmit power.
• Optimizes UAV locations by solving a power minimization problem under QoS and coverage constraints.
• Evaluates three multiple access schemes—RSMA, FDMA, and TDMA—under the same deployment framework to compare their energy efficiency.
• Employs a power consumption model based on path loss and signal-to-interference-plus-noise ratio (SINR) to quantify total transmit power.

Experimental results

Research questions

• RQ1 How accurately can LSTM-based traffic prediction improve UAV deployment efficiency in dynamic urban environments?
• RQ2 To what extent does the KEG algorithm enhance the accuracy of service area partitioning compared to conventional clustering methods?
• RQ3 What is the optimal UAV placement strategy that minimizes total uplink transmit power under predicted traffic and user distribution?
• RQ4 How do different multiple access techniques (RSMA, FDMA, TDMA) compare in terms of energy efficiency under the proposed predictive deployment framework?
• RQ5 How does the framework perform across different network scales, such as limited versus entire urban areas?

Key findings

• The proposed framework reduces total transmit power by up to 24% compared to conventional non-predictive deployment methods.
• The KEG algorithm effectively partitions user distribution into distinct aerial cells, with better performance in smaller, localized areas than in large-scale urban regions.
• RSMA achieves the lowest total transmit power, reducing consumption by 35.5% compared to FDMA and 66.4% compared to TDMA.
• The scheme combining KEG clustering and UAV location optimization yields the best performance, outperforming all other configurations.
• In large-scale areas, the framework still provides measurable gains (0.47% power reduction) even when UAV count is reduced, indicating partial robustness.
• The simulation results confirm that predictive deployment using ML-based traffic forecasting and intelligent clustering significantly enhances energy efficiency and network scalability.