Inter-detector differential fuzz testing for tamper detection in gamma spectrometers

TL;DR

The paper tackles tamper detection in gamma spectrometers used for nuclear safeguards and treaty verification. It extends physical differential fuzz testing to inter-detector comparisons by anchoring a downrange detector to a trusted golden-copy baseline, despite inter-detector manufacturing variation, and demonstrates the approach on three NaI detectors from different manufacturers. Tamper detection is achieved by cross-detector comparisons using the modified reduced chi-squared metric $\chi^2/\nu$ and by analyzing how the cross-detector relationship evolves with fuzzing parameters such as pulse width $pw$, revealing attacks even when baselines are non-linear. The work highlights the need for anomaly-detection techniques to handle non-linear baselines and suggests this inter-detector fuzz testing framework can enhance real-world safeguards and treaty verification through a chain of knowledge between reference and deployed detectors.

Abstract

We extend physical differential fuzz testing as an anti-tamper method for radiation detectors [Vavrek et al., Science and Global Security 2025] to comparisons across multiple detector units. The method was previously introduced as a tamper detection method for authenticating a single radiation detector in nuclear safeguards and treaty verification scenarios, and works by randomly sampling detector configuration parameters to produce a sequence of spectra that form a baseline signature of an untampered system. At a later date, after potential tampering, the same random sequence of parameters is used to generate another series of spectra that can be compared against the baseline. Anomalies in the series of comparisons indicate changes in detector behavior, which may be due to tampering. One limitation of this original method is that once the detector has `gone downrange' and may have been tampered with, the original baseline is fixed, and a new trusted baseline can never be established if tests at new parameters are required. In this work, we extend our anti-tamper fuzz testing concept to multiple detector units, such that the downrange detector can be compared against a trusted or `golden copy' detector, even despite normal inter-detector manufacturing variations. We show using three NaI detectors that this inter-detector differential fuzz testing can detect a representative attack, even when the tested and golden copy detectors are from different manufacturers and have different performances. Here, detecting tampering requires visualizing the comparison metric vs. the parameter values and not just the sample number; moreover this baseline is non-linear and may require anomaly detection methods more complex than a simple threshold. Overall, this extension to multiple detectors improves prospects for operationalizing the technique in real-world treaty verification and safeguards contexts.

Figures (6)

• Figure 1: Top: the three NaI detectors used in the inter-detector fuzzing experiments (left to right: 269, 238, 010). Bottom: one NaI + PMT coupled to the digitizer (brown cylinder) with a ${\sim}7$ µCi Cs-137 source placed about $30$ cm to the right, all within the partial lead shield.
• Figure 2: Illustration of the calibration process for a single detector and parameter combination (detector 238, $120$ s dwell time, $1.25$ µs pulse width, $0.8$ fine gain, $800$ V bias voltage). Top left and right: a convolution filter is applied to the raw spectrum to identify peaks. Bottom left: the identified peaks are matched to known energies of the Eu-154 source, and a quadratic fit is derived from the matched channel/energy pairs. Bottom right: quadratic fit residuals.
• Figure 3: Calibrated Eu-154 spectra from the three detectors at the given hvt, fgn, and pw settings. Each spectrum is normalized by its max bin content to better highlight differences in shape among the three detectors. Bin widths differ slightly due to different calibrations. The peak centroid at $1001.644$ keV is a weighted average of the $996.29$ keV and $1004.76$ keV lines with decay fractions of $0.1048$ and $0.1801$, respectively nndc_eu154. The last bin includes overflow.
• Figure 4: Comparison of calibrated spectra collected with the same detector and parameters, with (orange, denoted 238A) and without (blue, denoted 238) the time-based attack triggering. After calibration, the spectra are interpolated to $1$-keV bins from $0$ to $1400$ keV, but, as described in Section \ref{['sec:calibration']}, due to the calibration procedure, the apparent calibrated energies typically do not correspond to true energies. As such, the Cs-137 $662$ keV peak position appears closer to $400$ keV. The $\chi^2$ weights (gray), i.e., each of the $J$ summands in Eq. \ref{['eq:rchi2']}, illustrate how much each spectrum bin contributes to the total $\chi^2/\nu$ statistic.
• Figure 5: An example of $\chi^2/\nu$ statistic distributions against fine gain (fgn, top), pulse width (pw, middle), and sample number (bottom) obtained from inter-detector fuzzing between detectors 010 and 238, with the latter subject to the time-based attack (suffix "A"). Acquisitions in which the attack actually triggered are shown in orange while non-attacked samples are shown in blue.