Multi-UAV Tracking Evaluation Using 5G Uplink Signals on an O-RAN ISAC Simulation Testbed

Imagine you are trying to spot a tiny, fast-moving drone in a crowded city using only the Wi-Fi signals already bouncing around the neighborhood. You don't have a special radar gun; you have to listen to the "echoes" of the regular data signals that phones and drones use to talk to cell towers. This is the world of Integrated Sensing and Communication (ISAC). Think of it like a bat that doesn't just use sound to navigate but also uses that same sound to chat with other bats. In the future, our 5G networks might do exactly this: use their existing radio waves to detect objects like drones without needing extra, expensive hardware. But there's a catch. If you have one drone, it's easy to hear. If you have three drones flying close together, their echoes might get mixed up, making it impossible to tell who is who. This is the puzzle scientists are trying to solve: how do we turn a communication network into a reliable "drone police" system that can track multiple targets at once without getting confused?

This paper takes a giant step toward answering that question by building a digital "sandbox" to test a new way of tracking multiple drones. The researchers created a virtual city using a simulation testbed called O-RAN, which is like a modular, open-source version of a cell network. Instead of using real radios, they used a computer to simulate the signals, the drones, and the environment. They set up a scenario with three different drones flying at different heights and speeds, all trying to be detected by a single cell tower equipped with a special 8-antenna array. The goal wasn't just to see if the tower could hear the drones (detection), but to see if it could follow them individually over time (tracking) and hand that information off to a command center, just like a real security system would need to do.

Here is what they found, and it's a bit of a twist. First, the system was actually really good at hearing the drones. All three were detected most of the time. However, the system struggled to keep them as separate identities. Imagine trying to follow three friends in a crowded mall. You can see all three of them, but every time they walk close together, you lose track of who is who, and you might accidentally tag two friends with the same ID or give one friend two different names. In this simulation, 73.7% of the time, two drones were so close in their signal "fingerprint" that the system couldn't tell them apart immediately. The biggest problem wasn't that the system was too weak to hear the drones (sensitivity); the problem was that the signals were fighting for attention (contention).

The researchers also discovered a clever trick with the antenna setup. By using a flat, grid-like array of antennas (a "planar array") instead of just a line, the system could measure the height of the drones. This didn't help the system see the drones any further away, but it was a lifesaver for telling them apart. When two drones were at the same distance and speed but at different heights, the height measurement helped the system decide which track belonged to which drone about 58% of the time. It was like having a 3D map instead of a flat one; it didn't make the map bigger, but it helped sort out the clutter.

However, the paper also warns us about a "ghost" in the machine. The simulation ran on a computer that was too slow to keep up with real-time speed. It was like watching a video of a race in slow motion. Because the computer was slow, the system thought the drones had disappeared for longer periods than they actually had, causing the tracking to break down. When the researchers fixed the timing and looked at the data as if it were happening in real time, the tracking performance improved significantly, but it still wasn't perfect. The system could detect the drones, but it still struggled to keep a single, unbroken ID on each one, especially for the drone flying the lowest and closest to the ground.

Finally, the team showed how to package this information to send it to a real-world security command center. They built a translator that takes the raw tracking data and formats it into a standard language (called SAPIENT) that other systems can understand. They even added a "confidence score" to each report, telling the command center how sure the system is about what it's seeing. But they also found a hard truth: to make the system more accurate at keeping track of multiple drones, they had to sacrifice something else. If they tried to force the system to track more drones by loosening the rules, it started mixing up their identities, giving the same drone multiple names. The researchers decided that keeping the identity correct was more important than catching every single moment, so they chose to be stricter.

In short, this paper shows that using 5G signals to track multiple drones is possible and promising, but it's not a magic bullet yet. The system can hear the drones, but it gets confused when they get too close. The solution isn't just better hearing; it's smarter sorting. The researchers proved that while the technology works in a simulation, the real challenge is managing the "traffic jam" of signals and keeping the identities of the drones straight, all while ensuring the computer processing the data is fast enough to keep up with the real world. It's a solid step forward, but the journey from a digital simulation to a real-life drone defense system still has some hurdles to clear.