Illinois Tech researchers use machine learning to solve drone base station deployment

The idea of using drones as mobile cell towers isn’t new. We’ve covered it in disaster response scenarios, from drones creating temporary mobile networks for mountain rescue crews in Wales to Nokia’s AI-powered drone network for Swiss emergency services. But a fundamental problem has persisted: how do you figure out exactly how many drones to deploy, and where to put them, without wasting energy or leaving users without coverage?

A team at Illinois Institute of Technology (Illinois Tech) now has an answer, and it involves training a machine learning algorithm to make those deployment decisions in near-real-time.

• The Development: Illinois Tech Professor of Electrical and Computer Engineering Yu Cheng and his students have published a paper in IEEE Transactions on Vehicular Technology describing a machine learning method that jointly optimizes both the number and placement of UAV base stations to supplement fixed cell towers.
• The “So What?”: The algorithm can compute optimal drone deployment fast enough to adjust network coverage over the course of a single day, not just weeks or months.
• The Source: Illinois Institute of Technology news release, published February 10, 2026.

Fixed cell towers waste resources that drones can reclaim

Today’s network coverage relies on fixed base stations, large towers that consume significant power and provide coverage to a set geographic zone. These towers are expensive, immobile, and built to handle the maximum expected capacity for their area. If demand drops at 2 a.m. on a Tuesday, that excess capacity just sits idle. If demand spikes during a major event, the tower can’t flex to meet it.

Deploying drones as supplemental base stations changes this equation. A UAV positioned on a rooftop or lamppost can extend network bandwidth to areas where demand is high and retract when it’s not needed. The technology to do this already exists. The hard part is the math.

“How many UAVs should be deployed?” Cheng asked. “If the number is too large, we unnecessarily waste energy, and if it’s too small, we won’t have enough spectrum to support users, so we want to find the optimal number.”

The machine learning approach solves what traditional methods cannot

The Illinois Tech approach uses machine learning to jointly consider two variables at once: the number of drones deployed and their physical locations. This is a combinatorial optimization problem, and traditional methods for solving it rely on approximations that take a long time to compute. That’s fine for static planning. It’s useless for real-time deployment decisions.

Cheng’s team combined two machine learning techniques to build an algorithm that arrives at the optimal solution quickly. The algorithm was trained on results from those slower traditional methods, learning to replicate their accuracy at a fraction of the processing time.

“Our method can be a very powerful solution for this category of combinatorial optimization problems,” Cheng said.

Because the computation is fast, network providers could adjust drone-based coverage throughout the day in response to shifting demand. Think of it as dynamic load balancing, but for wireless network infrastructure instead of web servers.

The research fits a broader push toward drone-based network infrastructure

This work doesn’t exist in isolation. The FCC has been moving to support drone communications for years now, proposing new spectrum rules for drone operations in 2023 and formally adopting rules for the 5030-5091 MHz band in 2024. Shenzhen committed $1.7 billion to drone infrastructure, including over 8,000 5G-Advanced base stations for low-altitude drone operations. Israel has been testing drone connectivity on 5G networks.

The missing piece in most of these efforts is the optimization layer. You can build the spectrum rules and the 5G towers, but figuring out where to actually place mobile aerial nodes in real-time is an entirely different problem. That’s what the Illinois Tech research addresses.

Cheng expects some version of drone-based network coverage to become a reality within five years. “It’s similar to having UAV deliver things. Right now, it’s being done on a small scale and is still in the experimental stage, but theoretically, everything is doable,” he said.

The research contributors include Oluwaseun T. Ajayi (M.S. ECE ’24, Ph.D. EE Candidate) and Suyang Wang (M.S. EE ’17, Ph.D. EE ’24).

DroneXL’s Take

We’ve been covering the “drones as cell towers” concept since at least 2022, when we reported on the Welsh mountain rescue network project. What’s changed is that we’re finally seeing the software catch up to the hardware. The spectrum is being allocated. The 5G networks are being built. The drones can fly. But without smart deployment algorithms, you’re just throwing expensive UAVs at the problem and hoping for the best.

What catches my attention about this research is the speed. Being able to recompute optimal drone placement throughout a single day turns this from a static infrastructure play into something genuinely responsive. Imagine a concert venue, a natural disaster zone, or a rural area during harvest season where connectivity demands spike and fade on short timescales. Fixed towers can’t handle that. A fleet of drone base stations, guided by this kind of algorithm, could.

The five-year timeline Cheng mentions feels about right. With telecom companies like Nokia already running drone-as-a-service networks in Switzerland and the FCC continuing to open up spectrum for drone communications, the regulatory groundwork is being laid. Expect the first commercial pilot programs for drone-based network supplementation to appear in disaster response and large-scale event coverage by 2028. The IoT explosion will only accelerate demand from there.

Editorial Note: AI tools were used to assist with research and archive retrieval for this article. All reporting, analysis, and editorial perspectives are by Haye Kesteloo.