Network sensing tracked an unconnected drone without radar
In sum – what we know:
- Passive detection, no radar – Three commercial 5G sites in Arlington tracked a drone using signal reflections alone, with Ericsson fusing returns from all three in a multi-static configuration.
- Capex is the pitch – Reusing towers, radios, and spectrum that are already deployed and paid for changes the risk/reward math in a way most carrier enterprise plays never managed.
- Classification is unresolved – Neither company explained how the AI models tell a drone from a bird, and no false positive rates, detection probabilities, or bad-weather data have been published.
AT&T and Ericsson have built a new drone-detection system that was shown off in a demo at AT&T Stadium in Arlington recently. During the demo, “rogue” drone appeared on cue, a dashboard flagged it, a friendly drone piloted by FAA-licensed Ericsson employees flew out to deter it, and the airspace was declared safe again. The staging was obviously on rails, but the underlying idea showed off that three ordinary 5G cell sites, two of them roughly a quarter mile from the demo area, detected and tracked a small drone that was never connected to the network.
That’s the headline capability. No radar was installed and no new sensors were bolted on. The companies repurposed existing Massive MIMO radios and mid-band spectrum, added new signal-processing and AI software, and turned a slice of AT&T’s commercial network into a distributed sensing grid based on the “gNB sensing” work being standardized in 3GPP Release 19 — Integrated Sensing and Communication, or ISAC, arriving on 5G rather than waiting for 6G.
Repurposing 5G for sensing
The honest framing is that cellular networks won’t replace radar for drone detection — they’ll fill a gap radar struggles with. Traditional airborne and long-range radar systems perform poorly at low altitude, close to the ground, where terrain and clutter dominate. Cell towers sit exactly in that zone. As AT&T’s Robert Soni pointed out, the angle of illumination from a tower lets the network see lower than airborne radar typically can, and AT&T has roughly 75,000 physical sites across North America — a footprint that would be genuinely difficult for any government agency to replicate with purpose-built radar. Cellular sensing becomes one layer in a stitched-together stack that still needs other sensors, plus separate command-and-control for anyone who actually wants to take a drone down.
It’s also worth noting the distinction from what carriers can already do. Connected drones show up on networks today through their own telemetry, and ISAC as a research topic has been kicking around academia for two decades. What’s new here is passive detection. The Arlington system worked off signal reflections from an object that never touched the network — the towers transmitted standard 5G signals, and software fused the echoes from three sites into what Ericsson calls a multi-static configuration, cross-checking returns to improve accuracy and remove any single point of failure.
The performance envelope was respectable for a first public outing. Drones flying at 300 to 400 feet were detected and continuously tracked at ranges up to 6 kilometers, and the dashboard overlaid the network-derived track on the drone’s own telemetry to show they nearly matched.
The economic argument is the real pitch, though. Compared to a purpose-built radar baseline, reusing towers, radios, and spectrum that are already deployed and already paid for cuts the incremental capex of wide-area detection dramatically. Everything about the commercial case flows from that.
Reality check
The problem with carrier enterprise plays is that they have a track record, and it isn’t great. Dell’Oro analyst Stefan Pongratz is candid about it. On enterprise 5G and adjacent opportunities, he notes “it is true that they have not moved the needle much yet for the MNOs,” and says that while it’s early days for sensing, ISAC expectations remain muted.
But he also sees why this one is structurally different from, say, private 5G. “The ability to leverage the existing macro grid and minimize the incremental capex completely changes the risk/reward profile and fits better with the CSP model,” Pongratz said. The assets already exist and the carriers already own them — which is a better starting position than most enterprise ventures operators have chased over the past decade.
That doesn’t mean the sensing has to be world-class. It means it has to be good enough for the price. “Basically, the goal is not to deliver the best sensing performance – the goal is to find the sweet spot that delivers the best ROI,” Pongratz said. A dedicated counter-drone radar will outperform a repurposed cell site. It will also cost a lot more per square mile of coverage.
AT&T, to its credit, isn’t pretending this is a product yet. Soni described a phased path to commercialization rather than a wait for 6G standards to land in 2030. “We’ll offer a skateboard version of this, then we’ll offer a bicycle version of this, and then we’ll offer an automobile version of this,” he said. The skateboard customers, for the next three to five years, are mostly three-letter agencies, first responders, and critical infrastructure owners — not general commercial enterprise. Federal agencies interested in low-altitude airspace monitoring are already helping fund the R&D, which tells you something about where the near-term money is. AT&T characterizes its own internal investment as “fairly modest,” concentrated in a small technology group in Austin. In other words, nobody is betting the company on this.
Technical limitations
The least satisfying part of the demo was classification. Detecting a reflection is one thing; knowing it’s a drone rather than a bird is another, and the system leans on AI inference models layered over the signal processing to make that call and output a confidence score. How that model actually decided the non-connected object in Arlington was a drone was never made clear — neither AT&T nor Ericsson had a convincing answer during Q&A. And the question of who builds and owns those classification models going forward matters, because it’s not obvious that Ericsson and the carriers are the natural home for that AI work. My read is their role there will be limited.
The raw physics is better understood. Mid-band spectrum gives the system enough contiguous bandwidth to resolve objects down to roughly a 10cm x 10cm box, which is genuinely impressive for hardware designed to carry phone traffic. But mid-band is a deliberate compromise. Millimeter-wave sensors would deliver finer spatial resolution and lose most of their range in the bargain; low-band propagates farther but the spectrum is fragmented and the wavelengths too long to resolve small objects. Early testing confirmed mid-band as the workable middle ground rather than the ideal.
What’s missing is the data that would let anyone independently judge the system. Neither company has published false positive or false negative rates, detection probability at various ranges, or performance in bad weather or against a genuine swarm — the Arlington demo used two drones. Executives acknowledged that dense urban environments like downtown Dallas remain a real challenge, where multipath reflections and obstructed lines of sight degrade reliability. Fast-moving targets are harder too. As one executive put it, physics is physics.
And even a perfect version of this system only detects and tracks. Mitigation — jamming, capture, interdiction — has to be handed off to separately authorized entities under counter-UAS law. The carrier sells awareness, not protection. That’s a narrower product than the demo’s rogue-drone vignette implied.
Ecosystem
Ericsson frames network sensing as foundational plumbing for the “low-altitude economy” — tracking drone deliveries, feeding smart-city digital twins, eventually providing situational awareness for autonomous vehicles when onboard radar and lidar hit blind spots. Soni noted that AT&T already partners with essentially all the self-driving car companies, and that the autonomous vehicle about to run you over in Dallas or Austin is almost certainly connected to a carrier network today. That’s true, and it’s also a caution. That AV connectivity revenue has existed for years without ever showing up as a meaningful line in carrier enterprise results. Connectivity to autonomous fleets has so far been a traffic story, not a revenue story, and sensing will need a different outcome to matter.
The differentiation, if it comes, is supposed to come from openness. Against the legacy single-vendor stack model, the pitch is APIs that let third parties fuse the network’s spatial data with their own assets — camera feeds, radar, flight manifests, inventory databases, even language models for correlation. Both companies say they exposed this work early precisely to pull ecosystem requirements in before the product hardens.
That said, AT&T is clear-eyed about where open ecosystems end. The company has championed O-RAN for years, and Soni sees real value in letting startups and academics prototype on open stacks. He just doesn’t confuse that with production. “Do I really think open source finally becomes a replacement for a commercial vehicle? No,” he said. Vendor-grade hardware and software will carry the actual sensing service; open platforms feed it ideas.
Which is roughly where the whole effort sits. The technology is real, the incremental cost argument is genuinely strong, and the early customers are identifiable. But the classification question is unresolved, the performance data is undisclosed, and the revenue history of carrier enterprise ambitions counsels patience. The skateboard is here. The automobile is still a promise.
Table of Contents
