Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals

arXiv:2607.11955 · eess.SP, cs.SY, eess.SY · Submitted 2026-07-12 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals".

Dev: Low-altitude uncrewed aerial vehicles (UAVs) pose growing risks to airspace safety, security, and privacy.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So we're looking at the paper titled "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals," and the authors, Wenyu Huang, Nuria Gonzalez-Prelcic, Vishnu Ratnam, Murat Bayraktar, and Charlie Jianzhong Zhang <ref:2607.11955#pg0,Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink>. Rosa is a field roboticist who's curious if this actually works outside of a controlled lab setting and for how long.

Dev: I'm looking at the title and authors now; it points to a specific approach using multiple user equipments for passive detection, which sounds really interesting from an engineering standpoint concerning the hardware requirements.

Taro: From an autonomy researcher's view, the emphasis on exploiting uplink signals from standard 5G New Radio instead of downlink measurements is what catches my attention; it suggests a different kind of sensing capability entirely <ref:2607.11955#pg0>.

Rosa: I mean, if we can use existing 5G infrastructure to passively sense low-altitude UAVs without needing dedicated radar hardware, that opens up some possibilities for distributed sensing in urban areas <ref:2607.11955#pg0>.

Dev: Exactly; the core idea seems to be using standard Sounding Reference Signal pilots transmitted by multiple UEs as the input for the base station to observe echoes from a UAV.

Taro: That distributed nature is what makes it compelling because it leverages existing communication channels, which is crucial when you're trying to deploy autonomous systems in real-world environments where new sensors are expensive or impractical.

Rosa: And what does this mean practically for deployment? Can we actually expect a stable localization result once the UAV starts moving and the environment gets cluttered?

Dev: The paper suggests they tackle that by proposing a framework that uses LOS-referenced synchronization to handle timing and frequency impairments from each UE independently.

Taro: That synchronization part is key because if you can maintain sub-nanosecond accuracy, it allows for tracking dynamic targets in unpredictable urban settings where things move fast.

Rosa: Sub-nanosecond accuracy sounds incredibly demanding for a deployed system; how does the paper handle that precision when there are residual impairments from each user equipment?

Dev: They address those impairments by proposing a four-step LOS-referenced synchronization scheme, which reuses timing advance commands and conjugate products to remove residuals without needing extra signaling.

Taro: So they're trying to clean up the inherent noise of the uplink signal before attempting any detection, which is a necessary step for reliable autonomy.

Rosa: And what about the overall implication? Does this framework move us away from traditional sensing methods that rely on dedicated hardware?

Dev: It aims to provide a solution for distributed FR1 sensing architectures, specifically focusing on urban single-cell scenarios where multiple UEs transmit SRS pilots to the base station.

The paper's summary: Rosa: To summarize what we've seen so far in "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals," the authors are proposing a method where multiple ground user equipments transmit standard Sounding Reference Signal pilots to a base station, and the base station then receives echoes from a small UAV acting as a scatterer of these signals <ref:2607.11955#pg0,Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink>.

Dev: Essentially, they build the system around observing how those SRS pilots are reflected by the UAV in relation to the different UEs, which allows for obtaining multiple bistatic views at the base station.

Taro: The system model they describe involves an urban single-cell setup where static and dynamic scatterers are present, including ground vehicles and pedestrians moving as user equipments.

Rosa: They're focusing on how the state of the UAV is embedded in these multiple uplink channel estimates observed at the base station, which is what feeds into their detection process.

Dev: Their methodology involves a detailed process: first they estimate LOS delay and direction using a TA command, then construct a "LOS reference" channel, and then use an adjacent-occasion conjugate product to remove unknown phase terms.

Taro: The paper highlights that this approach results in detection rates nearly four times higher than what you would get from a traditional detect-then-fuse baseline setup.

Rosa: That's a significant factor; improving the detection rate by that much suggests the fusion technique is quite effective at isolating the target echo from the background noise.

Dev: They further refine this by employing two filters: one to remove static components by subtracting a slow-time sample mean, and another using a soft spatial projector to exploit elevation information.

Taro: The geometry-coupled bistatic fusion step is where they combine evidence across all the user equipments in the shared three-dimensional state space to find the UAV's position.

Rosa: So, in simple terms, they're taking noisy uplink data from several users, cleaning up the noise using synchronization and filtering techniques specific to 5G NR signals, and then fusing those views geometrically to pinpoint a low-altitude UAV’s location <ref:2607.11955#pg0>.

Dev: The paper demonstrates that this fusion process yields a median three dee position error of about four point eight four meters in a cluttered urban scene under these conditions <ref:2607.11955#pg2>.

Taro: That level of error is respectable for passive sensing in an uncontrolled environment, especially considering the complexity of the propagation path they're dealing with, like static scatterers and moving ground UEs.

Rosa: It’s impressive that they managed to achieve that median error while operating within a single-cell uplink scenario where signal quality can fluctuate quite a bit.

The paper's improvements: Dev: Moving on to the specific enhancements proposed in "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals," the primary improvement centers on solving the inherent three classes of channel estimate impairments: residual timing offset, common frequency reference, and common complex scalar distortion <ref:2607.11955#pg0,Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink>.

Rosa: The key methodological advance here is that they introduce a LOS-referenced synchronization scheme that manages these impairments by reusing existing TA commands and an adjacent-occasion conjugate product to clean the channel estimates without requiring any additional signaling from the UEs.

Taro: That's smart because it keeps the system efficient; you don't add more complex signaling overhead just to fix synchronization issues, which is a big win for low-power sensing.

Dev: Beyond that, they address clutter suppression by employing two distinct filters: one filter specifically removes static components by subtracting the slow-time sample mean, and another filter exploits elevation by applying a soft spatial projector using a steering dictionary spanning an elevation range of

θlow, θhigh: .

Rosa: I like how they tackle clutter from different angles; removing static elements at every delay bin simultaneously is one thing, but using the spatial projector to exploit elevation for the residual channel is another layer of refinement.

Taro: This two-pronged filtering approach seems essential for handling a cluttered urban scene because it addresses both temporal and spatial noise sources effectively.

Dev: The final significant improvement lies in their geometry-coupled bistatic fusion, where they calculate a candidate UAV state and then beamform the residual channel toward that direction to extract a delay-Doppler response.

Rosa: That final step, calculating the log contrast k and normalizing it by the range zeta k(theta), before fusing them using a trimmed mean T(theta) that discards the weakest UE.

Taro: That trimming step is what allows them to reveal those echoes that are "too weak to be detected from any single UE view," which is crucial for overcoming clutter masking or substitution effects.

Dev: So, in essence, they improve detection by making the evidence more robust through sophisticated synchronization, targeted clutter removal, and a fusion strategy that specifically looks for weak signals across multiple perspectives.

Rosa: It’s clear that the improvements aren't just about adding more data; it’s about intelligently processing the existing signals to extract meaningful target information from a very noisy uplink environment.

Conclusion: Dev: So, wrapping up our discussion on "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals," the paper demonstrates that by combining LOS-referenced synchronization, specific clutter suppression filters, and geometry-coupled bistatic fusion, they can achieve reliable passive localization using standard 5G NR uplink signals <ref:2607.11955#pg0,Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink>.

Rosa: It seems like the main implication is that we can create a system capable of detecting low-altitude UAVs without needing specialized radar hardware by utilizing ubiquitous cellular networks for sensing.

Taro: The ability to provide a median three dee position error of four point eight four meters in urban settings, even with clutter, suggests this framework could be useful for rapidly assessing airspace safety when dedicated surveillance is unavailable <ref:2607.11955#pg2>.

Dev: And considering the operational requirements, the latency and loop rate are critical factors; they need to maintain high fidelity under real-world signal variations to be viable for actual autonomous flight control systems.

Rosa: I think the practical impact is that this makes low-cost, passive surveillance a reality by leveraging existing cellular infrastructure for security monitoring of restricted airspace.

Taro: If we look at the future work, the paper doesn't explicitly detail what comes next, but it leaves room for expanding the sensing modalities beyond just SRS pilots to see if other signals can be used.

Dev: I agree; they focused heavily on this specific uplink framework for now, and extending it to handle different propagation environments will be a natural progression.

Rosa: So, in short, "Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink Signals" shows a viable path toward using existing communication signals for passive sensing of aerial threats <ref:2607.11955#pg0,Fuse-then-Detect for Passive UAV Localization Using Multi-UE 5G Uplink>.

Department of Electrical and Computer Engineering, University of California San Diego, USA · Standards and Mobility Innovation Lab, Samsung Research America, USA

eess.SP, cs.SY, eess.SY

Submitted: 2026-07-12

Updated: 2026-10-05

Comments: This work is accepted to the 2026 IEEE Integrated Sensing and Communication Conference (ISAC)

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 75/100

The gist: Low-altitude uncrewed aerial vehicles (UAVs) pose growing risks to airspace safety, security, and privacy.

Key concepts

LOS-Referenced Synchronization: This technique uses an existing timing advance command and an adjacent occasion conjugate product to remove timing offsets, frequency references, and amplitude distortions in each UE's channel estimate without needing extra signaling. It refines the channel by projecting it onto an estimated Line-of-Sight direction.
Clutter Suppression Filters: Two filters are used to clean up the received signal. The first removes static clutter by subtracting the time-invariant sample mean across all delay bins. The second filter uses a soft spatial projector based on a steering dictionary to suppress residual clutter based on elevation angle.
Geometry-Coupled Bistatic Fusion: This method combines data from multiple UEs in 3D space. It predicts the expected excess delay for a candidate UAV position and then beamforms the residual channel toward that direction. Evidence is fused using a trimmed mean, discarding weak measurements to reveal echoes invisible to any single UE view.

Terminology

User Equipment — 9× in this paper
Accuracy — 5× in this paper · explained in 1 episode
Steering — 4× in this paper · explained in 2 episodes
Consistency — 2× in this paper · explained in 3 episodes
Alignment — 1× in this paper · explained in 1 episode
Aperture — 1× in this paper · explained in 1 episode
Calibration — 1× in this paper · explained in 7 episodes
Clustering — 1× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
Failure Modes — 1× in this paper · explained in 3 episodes
Fidelity — 1× in this paper · explained in 2 episodes
G NR Sounding Reference Signal — 1× in this paper
G New Radio — 1× in this paper
Physical Constraints — 1× in this paper · explained in 7 episodes
Quantization — 1× in this paper · explained in 12 episodes

Summary

Low-altitude uncrewed aerial vehicles (UAVs) pose growing risks to airspace safety, security, and privacy. This paper presents the first uplink framework for passive UAV detection and localization by exploiting standard 5G New Radio (NR) uplink signals transmitted by multiple user equipments (UEs).

The gist: The proposed framework achieves sub-nanosecond synchronization and a 4.84 m median 3D position error in a cluttered urban scene using a LOS-referenced synchronization scheme and geometry-coupled bistatic fusion.

System Model

The scenario involves an urban single-cell uplink where multiple ground UEs transmit standard 5G NR Sounding Reference Signal (SRS) pilots to a base station (BS), which then receives the UAV echoes. The full observation tensor, denoted as Hb, serves as the input to the sensing algorithm. The propagation environment includes static scatterers and dynamic scatterers, with ground UEs moving as pedestrians.

Uplink Receiver Processing and Residual Impairments

The BS performs standard 5G NR uplink synchronization before exporting the channel estimates. These estimates carry three classes of impairments:

Residual timing offset, described by equation (1): ∆tk[s] = τk,est[s] + τk,q[s] + τk,clk[s] + δtsync k[s].
Common Frequency Reference, where the observed Doppler is expressed as ν obs kl [s] = ¯ν kl[s] − ν ref k[s], as per equation (2).
Common complex scalar βk[s], which summarizes amplitude and phase distortion, including AGC variation and oscillator phase noise.

LOS-Referenced Synchronization

To address the independent timing offset, frequency reference, and amplitude scalar corruption in each UE's channel estimate, a LOS-referenced synchronization scheme is proposed that reuses the existing timing advance (TA) command and an adjacent-occasion conjugate product to remove the residuals without additional signaling. This four-step process involves:

Estimating LOS delay and direction using the TA command TTA k, which is refined by accumulating power over SRS occasions.
Constructing a LOS reference channel gk[n, s] by projecting the channel onto this estimated direction.
Removing unknown phase terms (LOS path phase, unknown LOS path phase, and LOS slow-time Doppler) using the adjacent-occasion conjugate product Qk[n, s], which exposes only the inter-occasion increments ηk[s].
Estimating timing increments ηˆk[s] and accumulating them to produce synchronized channel corrections Tˆk[s] and ΦˆP k[s].

Clutter Suppression

The framework employs two filters to reduce clutter effects:

The first filter removes the static component by subtracting the slow-time sample mean He e k [n, s] = He k[n, s] − 1/S sum over s' of He k[n, s']. This suppresses the time-invariant component at every delay bin simultaneously.
The second filter exploits elevation by applying a soft spatial projector F to the antenna axis of the clutter-suppressed residual channel He(res) k [n, s], which is defined using a steering dictionary Alow spanning the elevation range [θlow, θhigh].

Geometry-Coupled Bistatic Fusion

The detection and localization rely on combining evidence across multiple UEs in a shared three-dimensional state space. The process involves:

Defining a candidate UAV state θ = (az, el, ρ) and calculating the predicted LOS-referenced bistatic excess delay τ⋆ k(θ).
Beamforming the residual channel toward the candidate direction uˆ(az, el) to obtain Xk[n, s; θ].
Extracting a delay-Doppler response Pk(τ, ν; θ) and power map.
Calculating a log contrast Λk based on this response, which is normalized by the range ζ k(θ) = Λk(θ) − medρ Λk medρ Λk − medρ Λk.
Fusing the evidence using a trimmed mean T(θ) = 1/(K-1) sum over k=1 to K of ζk, which discards the weakest UE, and applying a minimum support constraint Kmin. This fusion reveals UAV echoes that are too weak to be detected from any single UE view.

UAV Position Estimation

After cluster validation, the BS-side UAV direction uˆ and LOS-referenced excess delay ∆ˆτ k for each supporting UE are obtained. The UAV position is then constrained to lie on the ray q̂ = pBS + ρuˆ.

Improvements for AI systems

Here are specific improvements to AI systems based on the proposed Uplink Sensing, however, introduces new challenges framework:

Enhance Autonomous Air Traffic Management (UTM) and Airspace Safety Systems:
Improve UAV Detection and Tracking Reliability in Cluttered Urban Environments:
Develop Robust Low-Cost Passive Surveillance for Critical Infrastructure Security:

Enhanced Autonomous Air Traffic Management (UTM) and Airspace Safety Systems:

By implementing the proposed framework, UTM systems can achieve sub-nanosecond synchronization between ground UEs and the base station to accurately resolve UAV delay and Doppler signatures. This enables autonomous systems to perform real-time, passive detection of low-altitude UAVs using existing 5G infrastructure without deploying dedicated radar hardware. The improved system can:

Perform high-precision, distributed localization of UAVs by fusing bistatic evidence from multiple user equipments (UEs), even when the echo is weaker than clutter.
Track the trajectory and velocity of detected UAVs with a median 3D position error of 4.84 meters, significantly improving safety margins for manned aircraft operating in dense airspace.
Provide robust localization in scenarios where monostatic sensing fails due to urban obstructions or signal masking by strong static scatterers.

Improved UAV Detection and Tracking Reliability in Cluttered Urban Environments:

The proposed LOS-referenced synchronization scheme mitigates residual timing, frequency, and amplitude impairments inherent in uplink sensing. This allows AI-driven detection algorithms to process noisy 5G Uplink Sounding Reference Signals (SRS) with high fidelity. The improved AI system can:

Achieve a detection rate nearly four times higher than traditional detect-then-fuse baselines by exploiting normalized contrast and geometry-coupled fusion, effectively isolating the weak UAV echo from strong urban clutter.
Reduce false alarms caused by static reflectors or dynamic clutter substitution (clutter masking/substitution) by enforcing physical bistatic consistency across multiple UE views.
Produce highly reliable detection results in complex urban scenes at Frequency Range 1 (FR1), where conventional methods struggle due to the low transmit power of UEs relative to BS downlink signals.

Development of Robust Low-Cost Passive Surveillance for Critical Infrastructure Security:

Leveraging standard 5G NR uplink resources, this framework allows for the creation of a cost-effective, passive surveillance system that exploits ubiquitous cellular networks for security monitoring. The improved AI system can:

Continuously monitor restricted airspace or critical infrastructure (e.g., power plants, sensitive sites) by passively sensing unauthorized UAV presence using only the communication signals from nearby pedestrian UEs.
Implement a detection and localization pipeline that is compatible with existing cellular hardware, minimizing deployment costs and complexity associated with dedicated radar systems.
Provide actionable intelligence on low-altitude aerial activity by accurately localizing detected threats within a 10m error radius, supporting rapid response coordination for security agencies.