Toward a numerical deshaker for PFS

Frédéric Schmidt, Irina Shatalina, Matthieu Kowalski, Nicolas Gac, Bortolino Saggin, M. GiurannaPlanetary 2013 · DOI · arXiv

Introduction

Since 2004, the Planetary Fourier Spectrometer (PFS) aboard the Mars Express (MEx) spacecraft has provided high-resolution infrared data of the Martian atmosphere. However, the instrument faces a significant challenge: mechanical microvibrations from spacecraft components, such as reaction wheels and the Inertia Measurement Unit, interfere with the spectrometer’s internal pendulum. These vibrations manifest as "ghosts"—spurious spectral features that can mimic real chemical signatures, complicating the analysis of Martian atmospheric composition.

The Challenge of Mechanical Vibrations

Traditionally, researchers have mitigated this noise by "stacking" or averaging approximately ten spectra. While this cancels out much of the interference, it comes at a high cost to spatial resolution. Stacking smears a 7 km observation spot into a 70 km track, making it impossible to study localized phenomena like trace gas plumes or mineral deposits.

Figure 1: Martian spectrum #106 from Orbit 0032, illustrating the presence of spectral ghosts and various vibration components (VC) that interfere with the true signal.

As seen in Figure 1, these vibrations create distinct artifacts across the spectrum, particularly around the CO₂ bands and the reference laser line. Identifying and removing these without losing spatial detail is the primary goal of the "numerical deshaker."

Methodology: Semi-Blind Deconvolution

The proposed solution is a semi-blind deconvolution algorithm designed to "deshake" individual spectra. The researchers modeled the observed signal as a convolution of the "true" Martian spectrum with a sparse "vibration kernel," plus additive noise.

Because both the true spectrum and the kernel are unknown, the team employed an iterative optimization strategy using the Fast Iterative/Thresholding Algorithm (FISTA). The "secret sauce" of this approach is Sparsity Regularization ($L_1$ norm). Since mechanical vibrations occur at specific, discrete frequencies rather than as random noise, the vibration kernel is expected to be sparse—mostly zeros with a few sharp spikes. By enforcing this sparsity, the algorithm can pinpoint and remove interference with high precision while maintaining the physical plausibility of the resulting Martian spectrum through smoothness constraints.

Results: Clarity and Resolution

The "deshaker" has demonstrated remarkable efficacy in cleaning flight data. In test cases from the first 310 orbits of the mission, the algorithm removed approximately 85% of the ghost energy.

Figure 3: Comparison of original and filtered intensity for Martian spectrum #10, showing the successful suppression of high-frequency artifacts and noise.

As illustrated in Figure 3, the filtered signal is significantly cleaner, successfully removing artificial features (such as those previously mistaken for atmospheric constituents around 2700 cm⁻¹) while preserving real CO₂ absorption bands. Most importantly, the method achieves a signal-to-noise ratio on a single spectrum equivalent to stacking 11 spectra. This effectively improves the usable spatial resolution of the PFS dataset by a full order of magnitude.

Conclusion and Future Work

The numerical deshaker transforms a hardware limitation into a solvable signal-processing problem. By leveraging the sparse nature of mechanical vibrations, researchers can now access high-resolution Martian data that was previously obscured by noise. Future work will focus on fully automating the selection of regularization parameters to process the entire PFS archive of over 6,000 orbits and developing validation methods for later mission phases where the internal reference laser was no longer active.

Summary updated

BibTeX
@inproceedings{schmidt2013toward,
  author = {Frédéric Schmidt and Irina Shatalina and Matthieu Kowalski and Nicolas Gac and Bortolino Saggin and M. Giuranna},
  title = {{Toward a numerical deshaker for PFS}},
  booktitle = {Planetary},
  year = {2013},
  doi = {10.1016/j.pss.2013.11.012},
}
Figure