A sharper signal from the archive
A preprint reports that an analysis pipeline raised the empirical signal-to-noise ratio of faint Jovian radio emissions in archival Parker Solar Probe data by an average of 1.54 ± 0.11 decibels across 50 manually curated events. Individual gains reached 3.32 dB.
The study tests whether singular value decomposition, or SVD, can reduce noise in phase-folded spectrograms used to search for faint Jovian decametric and hectometric emissions. In ordinary terms, SVD represents a large data matrix through a smaller set of patterns, then keeps only the modes that pass an information test.
Folding the archive into patterns
The archive covered 21 close-encounter intervals from October 2018 through December 2024. The analysis produced 938 plots organized by Io phase and 3,711 by Jupiter's System III longitude; 656 plots contained measurable Jovian signatures.
The pipeline folded temporal spectra into repeatable periodic arrays, creating a baseline for later noise reduction and matrix decomposition. It then used an entropy rule to choose the retained SVD modes, keeping only modes below a threshold of 3.0 nats.
What the diagnostics found
That selection was compact but not fixed across the archive. Across the 50 representative events, the retained rank averaged 9.56 ± 0.35 modes, with the value determined separately for each phase-folded spectrogram.
On the same curated sample, empirical SNR rose by 1.54 ± 0.11 dB on average, with gains as high as 3.32 dB. The reported improvement is an expected consequence of low-rank denoising, so it is not by itself independent validation of Jovian signal content.
A second diagnostic measured spatial correlation in the two-dimensional spectra. The reconstructed matrices had a lag-1 autocorrelation of 0.797 ± 0.003, compared with 0.040 ± 0.017 in the residuals, a roughly 20-fold separation. That pattern is consistent with structured content being concentrated in the reconstruction and near-white noise in what was removed, although the residuals still contained weak diffuse structure and occasional edge artifacts.
An external check used independent Stokes V spectrograms from high-rate telemetry windows. Across 65,611 phase-frequency pixels, the extracted spectrogram had a Pearson correlation of 0.783 with Stokes V. In the same comparison, the Stokes V control had lag-1 autocorrelation of 0.32 and SNR improvement of 0.64 dB, versus 1.13 dB for the primary input.
The agreement was morphological rather than a flux-calibrated measurement. The Stokes V check was limited to high-rate telemetry windows, and no confidence interval for the Pearson correlation was reported.
Timing and phase structure
The maps also showed different phase and longitude patterns by polarization. Left-hand circular polarization, or LHCP, peaked near Io phases of 100 and 260 degrees, reaching about 19 percent. Right-hand circular polarization, or RHCP, peaked near 100 and 265 degrees, reaching about 11 percent and 12 percent. LHCP longitude maxima appeared near 50 and 220 degrees in System III longitude, while the RHCP distribution was flatter.
A separate check used one simultaneous PSP-Juno HOM transient. Its cross-correlation lag was 38.24 minutes, against a predicted delay of 36.84 minutes at a PSP-Jupiter distance of 4.43 AU. The measured lag was therefore within 1.4 minutes, or 3.8 percent, of the prediction.
That result is a timing consistency check, not an inter-instrument calibration: the comparison involved one transient and was not flux-calibrated. Contamination in Juno's HOM band also limited the comparison.
A screening tool, not a final detector
For archive screening, the researchers built a second SVD from confirmed Jovian data to derive global structural modes called Eigenfaces. Each interval received a coherence score, C, based on its similarity to those modes. Confirmed intervals peaked near C = 0.09, while background peaked near 3 x 10^-3, an offset of about 30 times.
The separation was useful but incomplete. Confirmed and background intervals overlapped from roughly C = 10^-3 to 3 x 10^-2, and high-C detections clustered in six perihelion-encounter windows. The paper treats the score as an archive-enrichment screen, not a definitive interval classifier, because individual-interval separation remained imperfect.
The limits of the test
There are reasons not to read the reported averages as a complete picture of the archive. The 50-event validation set was manually curated rather than randomly or stratified sampled, so its SNR and autocorrelation statistics are not an archive-wide average. The entropy threshold is tied to the matrix dimensions used and was not tested across different spectral sizes or instruments.
The residuals retained weak diffuse non-white structure and occasional ringing at sharp emission boundaries. The Eigenface score also showed substantial overlap between confirmed and background populations and did not reliably distinguish emission morphologies. The paper presents the method as a screening tool, not a perfect detector.
The work remains a preprint. The supplied record dates its arXiv version to August 25, 2026, and labels the front matter a draft dated August 27, 2026. The paper states that PSP FIELDS/RFS data and the Python analysis pipeline are publicly available through the FIELDS archive and a Zenodo record.
Paper data and sources
Original title: A Singular Value Decomposition Framework for Jovian Radio Emissions from Parker Solar Probe
Authors: Evan Wille, Zack Li, Marc Pulupa et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-25
DOI: Not available
Original paper · Full text