GhostLock: Learning-Assisted Blind GPS Signal Nullification Without Victim-Side Information

arXiv:2610.20831 · cs.CR · Submitted 2026-10-14 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: Today's paper: "GhostLock: Learning-Assisted Blind GPS Signal Nullification Without Victim-Side Information".

Elias: The gist The GhostLock framework is a learning-assisted blind GPS signal nullification approach that suppresses authentic satellite signals through destructive interference using only an approximate target location estimate,

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

Paper summary: Nadia: So, GhostLock is this new method for nullifying GPS signals without needing any information from the victim receiver itself, and it uses learning to do it blind based on an approximate location estimate.

Elias: Right, it’s about suppressing those authentic satellite signals using only that approximate target location estimate and some learning.

Priya: What does that actually mean for someone who just listens to the show? Like, what's the real-world deal here?

Nadia: It means we can attack GPS in a way that doesn't require us to tap into the victim's internal workings, which is a huge step for security researchers.

Elias: From my side, it’s interesting because it tackles that big assumption that authentic signals are always available and trusted references.

Priya: But what does the paper actually show in terms of how good this learning part is? What are the numbers we're looking at?

Nadia: They reported mean absolute errors around zero point zero zero three six and root-mean-square errors near zero point zero zero three nine for the delay estimation.

Elias: Those figures suggest that the deep learning module is actually pretty accurate at predicting those relative delays from just the raw I/Q samples.

Priya: So if we take those errors, what does that imply about how reliable this blind nullification actually is in a real GPS environment?

Nadia: It implies it's feasible, establishing that blind GPS signal nullification can happen without needing any victim-side observations.

Elias: And the authors did implement it on a software-defined radio platform using GPS-SDR-SIM and GNU Radio to show it works in practice.

Priya: So, what does this actually change for someone who only listens to the show regarding GPS security?

Nadia: It motivates new detection and mitigation techniques specifically for this type of emerging GPS threat model where the attack happens before signal acquisition.

Elias: We need to figure out how cheap and easy it is for a bad actor to build a system that uses only an approximate location estimate to do this.

Conclusion: Nadia: So we're wrapping up this look at "GhostLock: Learning-Assisted Blind GPS Signal Nullification Without Victim-Side Information," and what that title really means is they’ve built a way to mess with GPS signals without needing any data from the receiver itself.

Elias: Right, it boils down to using some learning—a machine learning part—to create destructive interference on those satellite signals using only a rough guess of where the target victim is located.

Priya: What does this actually imply for us, outside of just seeing these error numbers? If you can do this blind, what does that mean for the security landscape when it comes to positioning and timing services?

Nadia: It means we have to think about defenses that don't rely on the victim receiver being a clean reference point. It opens up ways for attackers to interfere before the victim even knows they're getting a signal.

Elias: And from a cryptographic standpoint, it challenges the idea that you need prior knowledge of the channel or synchronization parameters to launch an effective physical layer attack.

Priya: The authors did show it works on an SDR platform, which is good for proving feasibility, but they also flagged that this method doesn't adapt if the signal cancellation isn't perfect later on.

Nadia: Exactly. It’s a tool for the initial suppression event, but it leaves room for other kinds of countermeasures to be developed against persistent interference.

Elias: So while GhostLock is clever because it’s blind and doesn't need victim feedback, the next big question is how much effort an attacker has to put into getting that initial location estimate right.

Khalil Ahmad, Omar Ansari, Mohsin Kamal, Asif Iqbal, Muhammad Naveed Aman

School of Computing, University of Nebraska–Lincoln · National University of Singapore

cs.CR

Submitted: 2026-10-14

Updated: 2026-10-14

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 74/100

The gist: The gist The GhostLock framework is a learning-assisted blind GPS signal nullification approach that suppresses authentic satellite signals through destructive interference using only an approximate

Key concepts

Blind Adaptive Filter (BAF)
The BAF is used on the attacker's received GPS signal to reconstruct a clean version of the signal. It works by estimating complex adaptive weights to filter out noise and interference, allowing the attacker to work with a more accurate representation of the authentic satellite waveform.
Two-Stage Learning-Based Delay Prediction
This system uses two separate neural networks (a 1D CNN) to estimate the time delay between signals. The first network finds the whole number delay, and the second refines it with a fractional component. This precise timing is crucial for aligning the attacker's signal with the victim's expected signal.
Cancellation-Aware Training Objective
The deep learning model is trained using a loss function that includes both standard delay prediction and a physical cancellation term. This term forces the network to learn how to create an anti-signal that actively cancels out the target signal, improving its ability to suppress the GPS signal.
Delay-Derived Carrier-Phase Correction
Once the exact time delay is calculated, it is converted into a specific phase rotation needed for coherent cancellation. This correction transforms the learned temporal alignment into the precise phase shift required at the victim's antenna to effectively nullify the incoming GPS signal.

Summary

The gist The GhostLock framework is a learning-assisted blind GPS signal nullification approach that suppresses authentic satellite signals through destructive interference using only an approximate target location estimate, eliminating the need for victim-side information

System Model and Assumptions

The system model illustrates the learning-assisted GSN framework which comprises a set of visible GPS satellites, a passive victim GPS receiver, an attacker GPS observation receiver, a BAF module, a learning-based synchronization module, and an SDRbased RF transmitter

The attack operates in the analog RF domain before the victim performs signal acquisition The attacker first receives authentic GPS signals at its own location and uses a BAF to reconstruct the locally observed GPS waveform with inverted phase A machine learning model then predicts the carrier-phase corrections required to transform the attackerside waveform into a cancellation waveform appropriate for the estimated victim location The physical nullification event occurs only after RF transmission, at the victim antenna and RF front end

GhostLock Framework Stages

GhostLock is a learning-assisted blind GPS signal nullification framework that operates without online victimside information It contains the following stages: 1) attacker-side GPS signal observation 2) local BAF-based waveform reconstruction; 3) ML prediction of victim-specific phase correction; 4) generation of anti-signal; and 5) analog destructive superposition at the victim antenna The BAF estimates and reconstructs the authentic waveform available at the attacker It does not adapt against an unavailable victim residual

Blind Adaptive Reconstruction

The BAF output is given by zba[n] = wH[n]u[n], where w[n] ∈ CM is the complex adaptive-weight vector, u[n] is the reference vector Using the Wiener solution, wopt = R−1uu pux, which yields the best linear reconstruction of the attacker-observed signal achievable using the selected reference space

Two-Stage Learning-Based Delay Prediction

GhostLock uses a two-stage 1-D CNN to estimate the effective relative delay from complex baseband I/Q samples Stage 1 predicts the integer-delay component from the input I/Q signal, while Stage 2 uses the same signal and the Stage 1 output to estimate the fractional component The final delay is reconstructed as db= mb + αb (50) and is subsequently used for fractional waveform alignment and the deterministic carrier-phase correction

Cancellation-Aware Training Objective

Stage 2 is trained with a composite loss that combines supervised delay regression with a physical cancellation term The cancellation loss Lc exploits a calibration receiver (available only during offline training) that provides the target waveform y(i) The weighting λc(e) is linearly annealed from 0.05 to 3.0 over the first 150 epochs, allowing the network to first learn the supervised mapping and then progressively emphasize the cancellation objective

Delay-Derived Carrier-Phase Correction

Given the final delay estimate db in samples, the delay in seconds is τb = d/f b s The associated phase rotation is ϕbc = wrap[−π,π) ϕ0 − 2πfc db/fs!, which transforms the learned temporal alignment into the phase correction required for coherent RF cancellation The model achieved mean absolute errors (MAEs) of 0.003599 and root-mean-square errors (RMSEs) of 0.003902 and 0.

Improvements for AI systems

  1. textbfBlind Acquisition-Stage Nullification Framework (GhostLock): This system can suppress authentic GPS signals through analog RF destructive interference by synthesizing a cancellation waveform based only on attacker’s locally received GPS signal, eliminating the need for victim-side observations.

  2. textbfMachine Learning Synchronization Module: This module predicts the necessary phase corrections for signal alignment, leveraging a deep learning-based synchronization module to estimate the delay and phase required for coherent GPS signal cancellation based on a coarse target location estimate.

  3. textbfBlind Adaptive Filter (BAF) Reconstruction: The BAF reconstructs the authentic waveform using only locally available GPS observations and minimizes the attacker-side mean-square reconstruction error, enabling the system to adapt without requiring online access to the victim’s raw I/Q samples.

  4. textbfTwo-Stage CNN Delay Estimator: This hierarchical network predicts temporal alignment by first estimating a full-delay estimate (Stage 1) and then refining it with a fractional component (Stage 2), allowing for sub-sample temporal synchronization of the reconstructed anti-signal.

  5. textbfCancellation-Aware Training Objective: The training objective incorporates a physical cancellation term, defined by minimizing the normalized residual power after optimal cancellation, which allows the network to learn an optimal anti-signal that maximizes destructive interference based on calibration receiver data used exclusively during offline training.

  6. textbfDeterministic Carrier-Phase Correction: The system translates learned temporal alignment into RF transmission parameters by deriving a carrier-phase correction deterministically from the final delay estimate, converting learned temporal alignment into the phase correction required for coherent RF cancellation.

Abstract

The Global Positioning System (GPS) provides critical positioning, navigation, and timing (PNT) services for numerous safety- and mission-critical applications, making it an attractive target for physical-layer attacks. While GPS spoofing and jamming have been extensively studied, GPS Signal Nullification (GSN)-a physical-layer attack that suppresses authentic satellite signals through destructive interference prior to acquisition-remains largely unexplored. Existing wireless signal nullification techniques target conventional communication systems and typically require victim-side information, such as channel estimates or synchronization parameters, making them impractical for GPS. Furthermore, many GPS spoofing defenses assume authentic satellite signals remain available as trusted references, an assumption invalidated by GSN. To address this gap, we present GhostLock, a learning-assisted blind GSN framework requiring only an approximate target location estimate. GhostLock integrates a Blind Adaptive Filter (BAF) with a deep learning-based synchronization module to generate and align an anti-signal using only the attacker's locally received GPS signal, eliminating the need for victim-side observations. We implement GhostLock on a real-time software-defined radio (SDR) platform using GPS-SDR-SIM and GNU Radio. Experimental results demonstrate successful suppression of GPS signal acquisition, establishing the feasibility of blind GSN and motivating new detection and mitigation techniques for this emerging GPS threat model.

Sources

Related papers