GNSS jamming and spoofing against drones: what the 2026 research says and what to test in your fleet

By Kike Gandia · Drones & Defense · 12 min read

A drone that loses GNSS does not just lose its map: it loses the reference it uses to correct heading, hold altitude and return home. This year IEEE Communications Surveys & Tutorials published an open-access survey (Zeng et al., vol. 28, 2026) reviewing five years of research on GNSS jamming and spoofing against drones. We read it the way an offensive team would: which attacks work, which defenses fit in a real drone and what can be tested on a fleet.

1 W

is enough to disrupt GNSS reception (within a 30 km radius)

+2 dB

is all a spoofer needs to dominate the correlation peak (over the authentic signal)

1,600+

GNSS interference cases reported in the Baltic (first half of 2024, civil aviation)

−160 dB·W

GNSS signal power at the antenna (below the thermal noise level)

Figures quoted by the survey (pp. 5913-5916).

Why a drone falls when GNSS lies

The GNSS signal travels more than 20,000 km to reach the antenna and arrives below the thermal noise level. That weakness is what makes attacking it so cheap. The survey sums up a drone’s vulnerability on three fronts:

Signal Open and very weak

Civil signals have a public structure and minimal power. The receiver cannot tell an authentic signal from a fake one with the same format.

Hardware Poorly adaptable receiver

Automatic gain control (AGC) and tracking-loop bandwidth are limited: under strong jamming the front end saturates and works in the nonlinear region.

Platform No room to defend

Payload and onboard computing make antenna arrays or complex algorithms hard to fit.

The three baseline vulnerabilities of GNSS on a UAV.

Then there is the attacker: low-cost software-defined radio platforms such as the HackRF One can generate interference capable of severely disrupting commercial GNSS receivers. And regulation is moving: in 2024 ICAO published Annex 10 edition 10, with GNSS safety provisions for UAV operations and technical criteria for navigation integrity alerts.

Jamming: drowning out the signal

Jamming floods the receiver with power to bury the real signal. Because that signal arrives so weak, even a small jammer is effective. The survey classifies them by their shape in the spectrum:

The damage is measured in carrier-to-noise density (C/N0). According to the survey, raising the jamming-to-noise ratio from 55 to 92 dB·Hz can degrade C/N0 by 15 dB. In a cited test (Borio et al.), at jamming-to-signal ratios of 65 and 70 dB a u-blox 5H receiver gave valid positioning only 16% and 14% of the time. On a drone with RTK the effect multiplies: the reference station can also stop computing fixed solutions, and the loss of carrier-phase continuity degrades attitude determination, which is critical for flight control.

Spoofing: deceiving the receiver

Spoofing does not block the signal: it replaces it with a fake one the receiver accepts. It is easier to hide and, according to the survey, its low cost and stealth make it an especially serious threat.

Generated Synthesized signal

GNSS-looking signals are built from known structures. It requires knowing the structure of the authentic signal.

Meaconing Delayed rebroadcast

The real signal is captured and re-emitted with a delay and more power. It can target even encrypted military signals without decrypting them.

Asynchronous Forces a loss of lock

High-power fake signals make the receiver lose the real signal and lock onto the fake one.

Synchronous Aligned with the real one

The fake signal matches the authentic one in time and frequency and is gradually amplified until it takes over the tracking loops.

Spoofing types by source and synchronization strategy (table V).

Two results worth keeping in mind. In an experiment cited by the survey, controlled delay adjustments let attackers hijack several UAVs at once inside a given area. And with RTK, progressive-phase spoofing deviates the drone’s heading, which can end in unauthorized control.

What’s new: attacks that adapt, hide and coordinate

The most useful part of the survey for a security team is its review of emerging threats. Its conclusion: the danger is no longer just that the drone fails, but that someone takes control away from it.

Adaptive Choosing the moment

Launching spoofing just as a target leaves a tunnel, where it depends on GNSS, raised the success rate from under 40% to about 75% (Chang et al.). With a GPS simulator that perceived the drones’ positions, another study made two drones misread their position and collide at a predetermined point (He et al.).

Covert Staying under the threshold

A progressive attack deviated heading by −1.13° over 670 s without triggering alarms (Gao et al.). On tightly coupled GNSS/INS systems, a position error better than 1.01 m in 3D was achieved with a 39% higher success rate (Geng et al.).

Multidimensional Swarms and learning

A multi-antenna array mimicking more than four satellites lures an entire formation toward a false destination without altering internal relative positions, so no consistency check fires (Ceccato et al.). With reinforcement learning, the attacker learns when to act without knowing the target’s model (Ma et al.).

Three families of emerging threats, with results from the studies the survey cites (pp. 5916-5918).

Another study (Jung et al.) went beyond the signal: by tampering with the relative position and yaw angle fed to a reinforcement-learning navigation system, it made drones collide with obstacles. The threat is no longer just the radio layer but the navigation logic that consumes it.

Two kinds of drone, two different defenses

The survey separates two worlds that almost everyone mixes up. An inspection, agriculture or logistics operator is almost always in the one on the left.

Commercial and consumer Cost and weight rule

• Typical threat: narrowband jamming and meaconing
• Defense: single-antenna signal processing and low-cost MEMS-INS
• Return home after jamming using sensor data
• GNSS threat simulators before the mission
• Data-link authentication; in swarms, cooperative fusion and consistency checks

Tactical and military Performance and reliability rule

• Typical threat: broadband jamming, generated, adaptive and covert spoofing
• Defense: array antennas (as on the US “Gray Eagle” UAV) and multisensor fusion
• Physical shielding of the antenna against terrestrial interference
• Cryptographic authentication: M-code, and Galileo OS-NMA on the E1 band

Threats and countermeasures by drone type (table VI and p. 5918).

Detection: what fits in a drone and what doesn’t

For jamming, the cheapest approach is to watch AGC and C/N0: it is lightweight, but it produces false positives when power varies for natural reasons. Time-frequency analysis gives more detail at the cost of more computing, and arrays estimate the direction of arrival to separate the jammer from the satellites. With machine learning, a multilayer perceptron reached 98.9% detection with a 0.28% false alarm rate (Alkhatib et al.), and a sliding-window detector reached 99.97% detection on u-blox M8T receivers at 20 µs per sample (Sormayli et al.).

For spoofing, the picture is less comfortable: almost every method has an Achilles’ heel.

Mitigation: from detecting to recovering position

600 m → 12 m

maximum error under spoofing with INS-aided tracking (INSAT, TEXBAT dataset: −98%)

23.6 m → 1.03 m

3D error (RMS) in a tightly coupled RTK/INS system (adaptive robust Kalman filter (Hao))

9.78 m → 2.77 m

pseudorange error with multicorrelator processing (illustrative figure in the survey)

Three spoofing-mitigation results collected in the survey (pp. 5934-5936).

Against jamming, the adaptive notch filter is the economical option for narrowband; time-frequency transforms work for chirps and pulses; and arrays with beamforming (PI, MVDR) achieve the greatest suppression at the cost of hardware. AI also separates signal from interference: a deep convolutional network reached 36 dB of suppression (Sun et al.). Against spoofing, the survey covers multicorrelators, double-difference measurement correction (valid when receivers are at least 20 m apart), arrays, reinforcement learning and, above all, tight GNSS/INS coupling, which uses the INS’s short-term stability to correct GNSS.

What the survey recommends by drone. For small drones with very tight size, weight and power budgets, the notch filter is the default against narrowband jamming; if the threat is more complex, time-frequency transforms offer a good balance. Deep learning is generally infeasible unless heavily optimized. For spoofing on small drones, supervised learning offers a good balance; medium drones can afford arrays and multisensor fusion.

Its overall assessment is honest: signal processing and arrays are mature techniques; AI improves detection, but in mitigation it is still exploratory.

The four challenges that remain open

1. 1 — Onboard resources
   Payload, computing and battery leave no room for multi-antenna arrays or large networks. It is the bottleneck behind everything else.
2. 2 — Lack of flexibility
   The classic “detect, then suppress” chain adds latency against dynamic interference, and AI depends on training data: outside it, it degrades.
3. 3 — Few UAV-specific solutions
   Most are generic adaptations focused on jamming or on spoofing, not on hybrid attacks.
4. 4 — Swarms
   Information fusion adds latency, inter-drone communication is a target and, if critical nodes fall, redundancy is not enough.

The survey’s challenges (section VI): they chain together, because the first limits the rest.

In response, the survey proposes three layers: reception (multi-frequency, multi-constellation, low-Earth-orbit satellite support and compact arrays), algorithms (lightweight architectures such as distilled Transformers) and system (GNSS-centered multisensor fusion, swarm collaboration and intelligent decision-making).

What this means for your fleet

The paper leaves a lesson that is not about the lab: almost nobody tests what their drone does when GNSS lies. Five questions to start with:

1. What does the drone do when it loses GNSS? Hold, land or return home. You test it, you don’t assume it.
2. What does it do when GNSS is present but false? This is the hard case: no loss of signal fires.
3. Does it cross-check GNSS against INS, vision or another sensor? And at what discrepancy does it warn?
4. Does it rely on unauthenticated RTK corrections or a data link? Those are a separate entry point.
5. What does the ground operator see? An alert, or just a plausible position.

How to test without committing a crime

Broadcasting false or jamming GNSS signals outdoors is generally prohibited (the survey itself cites the FCC ban on jamming devices and EASA rules). Responsible testing happens in a controlled environment: with a constellation simulator and conducted signal or in a shielded chamber, with written authorization from the drone’s owner and a closed scope. The same applies to any radiated test: check with legal counsel first.

We hack drones and we hack with drones

A drone is an IoT device with propellers: firmware, SoC, radio link, mobile app, cloud and debug ports. That is why it falls squarely within our IoT and hardware pentesting. We approach it from both sides:

Against the drone The drone as a target

Firmware and debug ports (UART, JTAG, SWD), control and video link, app and cloud, and its behavior when GNSS lies.

With the drone The drone as a tool

Aerial reconnaissance of facilities and a platform for red team exercises when physical access is not an option.

More technical detail in “Hacking drones (II)”, linked at the end of the article.

Frequently asked questions

Can a cheap jammer disable a commercial drone?

According to the survey, 1 W is enough to disrupt GNSS reception within a 30 km radius, and a low-cost SDR such as the HackRF One can generate interference capable of disrupting commercial receivers. How much it degrades depends on the receiver: in the Borio et al. test, a u-blox 5H gave valid positioning only 16% and 14% of the time.

Which is worse, jamming or spoofing?

Jamming is cruder: the drone usually knows it has lost the signal. Spoofing is the one that can lead the drone where the attacker wants, and covert variants are designed not to trigger alarms.

Does AI solve the problem?

It helps a lot with detection, but for mitigation it is still exploratory. It depends on representative training data and, on small drones, deep learning is generally infeasible unless heavily optimized.

Can my fleet’s resilience be tested?

Yes, with authorization and in a controlled environment. Tell us which drones and which mission you have and we will define the scope.

Related service

IoT and hardware pentesting

Topics

#Drones #GNSS #GPS-spoofing #Jamming #UAV #IoT-Hacking #Hardware-Hacking #IEEE #Research

Keep reading

• Drone security: critical threats and how to protect airspace in 2025
• Hacking Drones III: Drone Warfare – Active Defense and Operational Resilience
• Hacking Drones (II): Anatomy of the Aerial Vector

Sources

• Zeng, Lu, Zhao, Xiao, Ni, Han and Li, “GNSS Jamming and Spoofing Threats in UAV Navigation: Countermeasure Status and Challenges”, IEEE Communications Surveys & Tutorials, vol. 28, 2026 (open access, CC BY 4.0)

Tell us which drones or fleet you have