⚡ electrical engineering

Evaluation of gNB Monostatic Sensing for UAV Use Case

This paper presents an end-to-end evaluation of 5G NR gNB-based monostatic sensing for UAV detection and 3D localization under 3GPP Release 19 standards, achieving high detection probability and meter-level accuracy while releasing the 5GNRad simulator to support reproducible research.

Original authors: Steve Blandino, Neeraj Varshney, Jian Wang, Jack Chuang, Camillo Gentile, Nada Golmie

Published 2026-04-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Steve Blandino, Neeraj Varshney, Jian Wang, Jack Chuang, Camillo Gentile, Nada Golmie

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a giant, high-tech lighthouse on a hill. Traditionally, this lighthouse (a 5G cell tower) just sends out radio waves to talk to your phone, like a waiter shouting orders to a table. But what if this same lighthouse could also act like a bat or a sonar system, listening to the echoes of its own voice to "see" things flying around it, like drones?

That is exactly what this paper explores. The authors are asking: "Can our existing 5G cell towers be turned into giant, invisible radar systems to spot drones without needing extra hardware?"

Here is a breakdown of their work using simple analogies:

1. The Big Idea: The "Dual-Purpose" Lighthouse

In the past, if you wanted to track a drone, you needed a dedicated radar. It was like having a separate flashlight just for looking, and a separate megaphone just for talking.

The Problem: 5G research has been messy. Some scientists used perfect, imaginary worlds where there is no wind, no noise, and no other buildings to confuse the signal. It's like testing a car on a frictionless vacuum track—it works great in theory but fails in the real world.
The Solution: This paper builds a "real-world simulator." They took the official rules for 5G (from a group called 3GPP) and built a complete, transparent "recipe" for how a cell tower would listen to echoes. They didn't just guess; they simulated the messy reality of wind, building reflections, and the tower's own noise.

2. The Experiment: The "Echo Game"

The team set up a digital simulation based on a specific scenario: Urban Macro-Aerial Vehicle (UMa-AV).

The Setup: Imagine a cell tower in a city looking up at a sky filled with 5 small drones flying at different heights and speeds.
The Signal: The tower sends out a special "ping" (called a Positioning Reference Signal, or PRS). Think of this as the tower shouting a specific, unique code.
The Echo: The drones bounce that code back. The tower listens for the return.
The Challenge: The tower has to hear the tiny whisper of a drone echo over the roar of the city (buildings reflecting the signal) and the tower's own internal static (self-interference).

3. The Processing Chain: From Noise to Picture

The paper details a "processing chain," which is like a factory assembly line for the signal:

Cleaning the Signal: First, they filter out the "static" (like turning down the volume on a radio to hear a faint station).
The Echo Map (Range-Doppler): They create a map.
- Range tells them how far away the object is (like measuring how long it took for an echo to return).
- Doppler tells them how fast it's moving (like the change in pitch of a siren passing by).
The Detective Work (CFAR): This is the "False Alarm" filter. Imagine a security guard at a party. If they shout "Intruder!" every time a mouse squeaks, they are useless. The system uses a smart filter (CA-CFAR) that only shouts "Intruder!" if the signal is loud enough to be real, ignoring the background noise.
3D Positioning: Once a target is found, the system calculates exactly where it is in 3D space (up/down, left/right, forward/back).

4. The Results: How Good Was It?

The results were surprisingly good for a system not originally designed as a radar:

Detection: The system successfully spotted 70% of the drones.
Accuracy: When it did spot a drone, it knew where it was within about 4 to 6 meters (roughly the length of a car). That is "meter-level" accuracy.
False Alarms: It rarely cried wolf. Only about 5% of the time did it think it saw a drone when there wasn't one.

The Trade-off:
The paper notes a classic "Goldilocks" problem. If you make the system super sensitive to catch every single drone, it starts seeing ghosts (false alarms). If you make it too strict, it misses slow-moving drones. They found a "sweet spot" that works well for most situations.

5. The "Secret Sauce": The Simulator

The most important part of this paper isn't just the numbers; it's the tool they built.

They released a software package called 5GNRad.
Analogy: Imagine a chef who invents a new recipe for a cake. Instead of just saying "it tastes good," they publish the exact recipe, the brand of flour used, and the oven temperature, so anyone else can bake the same cake and taste it for themselves.
This allows other researchers to stop arguing about "who had the best simulation" and start comparing their new ideas on the same playing field.

Summary

This paper proves that 5G cell towers can double as giant, city-wide radar systems to track drones. While they aren't as perfect as dedicated military radar yet, they are good enough to be useful right now. By creating a standard, open-source way to test this, the authors have given the world a common language to improve 6G technology, making our future cities safer and smarter.

1. Problem Statement

The integration of sensing into cellular networks (Integrated Sensing and Communications, or ISAC) is a key enabler for 6G, particularly for Unmanned Aerial Vehicle (UAV) surveillance. However, current research faces significant challenges:

Fragmented Evaluations: Prior studies often rely on simplified assumptions (e.g., single-bounce targets, negligible clutter, idealized hardware) that do not reflect real-world complexities.
Lack of Reproducibility: There is no standardized, transparent end-to-end implementation to compare results across different studies.
Standardization Gaps: While 3GPP Release 19 has standardized the ISAC channel model (including target Radar Cross Section and clutter models) and evaluation scenarios, it does not prescribe the sensing processing chain (how to convert received signals into detections and estimates). This leaves a gap in how to practically implement and evaluate gNB-based sensing.

2. Methodology

The authors propose and evaluate a complete, reproducible end-to-end sensing pipeline using 5G New Radio (NR) infrastructure.

System Model:
- Waveform: Uses existing 5G NR downlink Cyclic Prefix-Orthogonal Frequency Division Multiplexing (CP-OFDM) waveforms with Positioning Reference Signals (PRS) as the sensing signal.
- Architecture: Monostatic sensing where a single gNB (Transmission-Reception Point, TRP) acts as both transmitter and receiver.
- Channel Model: Adopts the 3GPP Release 19 ISAC channel model for the Urban Macro-Aerial Vehicle (UMa-AV) scenario. This includes:
  - Sensing Targets: Modeled as scattering points with specific Radar Cross Section (RCS) characteristics (mean, angular dependency, and log-normal fluctuations).
  - Background/Clutter: A dedicated monostatic background model representing clutter multipath via virtual reference points.
  - Impairments: Includes realistic antenna patterns, array geometries, and residual self-interference (SI) modeled as additional noise.
Processing Chain (Implemented in 5GNRad Simulator):
The paper implements a baseline receiver chain that processes the received PRS signals:
1. Channel Estimation: Least-Squares (LS) estimation of the effective frequency-domain channel.
2. Range Processing: Destaggering PRS symbols and applying an Inverse Fast Fourier Transform (IFFT) to generate range profiles.
3. Clutter Suppression: Slow-time mean subtraction to remove static background components.
4. Doppler Processing: Fast Fourier Transform (FFT) across PRS occasions to generate Range-Doppler maps.
5. Detection: Cell-Averaging Constant False Alarm Rate (CA-CFAR) detector with non-maximum suppression.
6. Parameter Estimation:
  - Angle of Arrival (AoA): Uses Beamspace FFT (for Uniform Rectangular Arrays) or Bartlett beam scanning.
  - 3D Localization: Combines estimated range and AoA to calculate 3D position.
Simulation Setup:
- Scenario: Single macro site, 4 GHz carrier, 100 MHz bandwidth.
- Targets: 5 small UAVs per drop, altitudes 25–300m, speeds 0–180 km/h.
- Hardware: 8x8 Uniform Rectangular Array (URA) with dual polarization. Evaluated both Full-Digital (64 RF chains) and Hybrid (32 RF chains) architectures.

3. Key Contributions

End-to-End Baseline Implementation: The paper provides the first transparent, reproducible processing chain for 5G NR monostatic sensing that adheres to 3GPP Release 19 channel assumptions.
Open-Source Simulator (5GNRad): The authors released the simulator and associated scripts to allow the research community to reproduce results and extend the pipeline with new algorithms.
Comprehensive Performance Analysis: The study evaluates detection probability, false alarm rates, and localization errors under realistic conditions (including clutter, self-interference, and varying antenna architectures).
Architecture Trade-off Analysis: The paper quantifies the performance impact of moving from full-digital to hybrid receiver architectures and varying element spacing.

4. Key Results

The evaluation was conducted under a 3GPP UMa-AV scenario with a False Alarm Rate (FAR) constraint of 5%.

Detection Performance:
- Achieved a Probability of Detection ( $P_d$ ) > 70% with a False Alarm Rate (FAR) < 5%.
- Increasing $P_d$ beyond 0.70 significantly degrades FAR (e.g., raising $P_d$ to 0.80 increases FAR to ~20%).
- Missed detections (False Negatives) were most frequent at:
  - Low Signal-to-Noise Ratios (large ranges).
  - High elevation angles (>60°) due to antenna pattern limitations.
  - Low radial velocities (due to overlap with residual clutter).
Localization Accuracy:
- Achieved meter-level 3D localization for correctly detected targets.
- 90th-percentile errors:
  - Horizontal: ~6 meters.
  - Vertical: ~4 meters.
- Radial velocity estimation was stable with low error for most targets.
Sensitivity Analysis:
- Self-Interference (SI): Higher residual SI reduces detection probability but increases the Positive Predictive Value (fewer false alarms relative to total detections). The F1 score remained relatively stable, suggesting robustness.
- CPI Length: Longer Coherent Processing Intervals (CPI) improve detection and velocity resolution but offer diminishing returns beyond 40 ms.
- Architecture:
  - Full-Digital vs. Hybrid: Hybrid architectures preserved detection probability ( $P_d$ ) but suffered a significant drop in Positive Predictive Value ( $P_{TP}$ ) and F1 score due to reduced spatial degrees of freedom, leading to more false alarms.
  - Element Spacing: Increasing vertical spacing from $0.5\lambda$ to $0.8\lambda$ caused a modest degradation in performance.

5. Significance

This work bridges the critical gap between 3GPP standardization (which defines the channel) and practical implementation (which defines the receiver). By releasing the 5GNRad simulator and establishing a reproducible baseline, the paper:

Enables fair and objective comparison of future ISAC algorithms.
Demonstrates that 5G NR infrastructure can effectively support UAV surveillance with reliable detection and meter-level accuracy without dedicated radar hardware.
Highlights the trade-offs between hardware complexity (full-digital vs. hybrid) and sensing performance, providing guidance for future 6G network deployments.