Bayesian and Deterministic Neural Network approaches to Faraday Cup calibration and plasma parameter estimation
Lidiya Ahmed, Michael L Stevens, Kristoff Paulson, Anthony W Case, Samuel T. Badman · 2025
Read Bayesian and Deterministic Neural Network approaches to Faraday Cup calibration and plasma parameter estimation by Lidiya Ahmed, Michael L Stevens, Kristoff Paulson, Anthony W Case, Samuel T. Badman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-20946.txt.
Introduction
Lidiya Ahmed leads four co-authors on calibrating Faraday cup plasma detectors by learning from a well calibrated sister spacecraft in a similar orbit. Ground truth is prepared from the reference craft by aligning time series with dynamic time warping, a pattern matching method.
A neural network then learns to reproduce that ground truth from the target craft raw measurements, even when those measurements are poorly calibrated or taken at low level. Because training targets come from the reference, calibration errors in the target stream do not poison the result. The scheme offers a path to rescue plasma data from aging instruments whose settings have drifted.
Catalog record
| CREATOR | Lidiya Ahmed, Michael L Stevens, Kristoff Paulson, Anthony W Case, Samuel T. Badman |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-20946
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian and Deterministic Neural Network approaches to Faraday Cup calibration and plasma parameter estimation Lidiya Ahmed, Michael L Stevens, Kristoff Paulson, Anthony W Case, Samuel T. Badman Abstract We describe a novel scheme for analyzing particle detector measurements when a well-calibrated, similarly instrumented spacecraft is present in a similar orbit. To prepare ground truth from measurements provided by a reference spacecraft, the method uses dynamic time warping (DTW)–a technique often used for pattern-matching in time series data. An artificial neural network (ANN) is created and trained to reproduce this ground truth from measurements at the target spacecraft. Unlike previous approaches, this procedure is insensitive to calibration errors in the target data stream, as the neural network may be trained from poorly calibrated particle spectra or even directly from low-level data in engineering units. We demonstrate a proof-of-concept by training an ANN to estimate solar wind proton densities, temperatures, and speeds from the DSCOVR PlasMag Faraday Cup, using the Wind Solar Wind Experiment as a reference. We present both deterministic and Bayesian neural network approaches. Applications for Parker Solar Probe, HelioSwarm, and other missions are discussed. I Introduction Machine learning has been applied productively in different areas of solar physics research, such as sunspot classification ( Chola & Benifa 2022 ) , solar flare forecasting ( Nishizuka et al. 2017 ) , solar radio burst detection ( Zhang et al. 2021 ) , geomagnetic storm prediction ( Domico 2022 ) , and thermal particle spectrum analysis ( Vech, D. et al. 2021 ) . The focus of this study lies in solar wind bulk flow analysis, using in situ measurements of velocity distribution functions (VDFs) by particle detectors such as Faraday Cups (FCs) ( Vech, D. et al. 2021 ) . FCs are simple detectors that measure the flux of charged particles over finite ranges of energy-per-charge. In typical space research applications, FCs are configured to resolve the energy distributions of the most significant ion species over tens of graduations. Examples include the Solar Probe Cup (SPC) on board Parker Solar Probe ( Case et al. 2020 ) , the Deep Space Climate Observatory (DSCOVR) PlasMag cup ( Loto’aniu et al. 2022 ) , the Wind Solar Wind Experiment(SWE) cup ( Ogilvie et al. 1995 ) , and many others. In this study, Solar wind proton measurements from the DSCOVR PlasMag cup and Wind SWE cup are used. Both of these spacecraft orbit near the Lagrange 1(L1) point which is located 1.5 million kilometers from the Earth in the direction towards the Sun. The DSCOVR PlasMag cup closely resembles the Wind SWE Cups with the distinction being that DSCOVR’s circular collector plate is divided into three independent 120° sectors ( Loto’aniu et al. 2022 ) , whereas the Wind SWE cups each have two semi-circular collector plates. The segmentation helps to capture deflections of solar wind and understand the geometry of the plasma flow. The DSCOVR and Wind FCs were also built in the same lab with nearly identical mechanical and electronics design (apart from the segmentation mentioned above), and they have similar energy range, measurement cadence, and operating modes. These instruments are calibrated in space and on the ground through careful procedures that finely characterize the sensitivity of collecting elements, the measurement circuit, and the high voltage element that provides energy discrimination. These calibrations must be monitored and updated over time with limited on-board calibration systems, through comparison with other spacecraft instruments, and with physics-based tests. Plasma parameters like density ( n ), thermal speed ( w ) or Temperature ( T ), and speed ( v ) are estimated by fitting the observations to a forward model of a velocity distribution function, such as a Maxwellian, using nonlinear least-squares fitting or some similar regression approach. For large data sets, nonlinear fitting can be time-consuming and computationally demanding. It may also fail to converge or improperly converge, for example when presented with unexpected distribution shapes or noisy data. This in turn can prompt ad hoc approaches and contingencies that are similarly difficult to implement with large data sets. This project aims to bypass the current modeling, calibration, and fitting process by implementing a robust and time-efficient approach. Artificial Neural Networks(ANNs) can help with this process by minimizing the time and effort involved in data processing. Vech, D. et al. 2021 previously showed that the nonlinear regression approach can be bypassed by a well-trained ANN, increasing computational efficiency in the process. Lacking ground truth, however, the Convolutional Neural Network(CNN) in that study was trained using simulated data only for a particular range of solar wind parameters, and perfect calibration was assumed when analyzing SPC data. We take this a step further, showing that an entire data analysis pipeline segment might be similarly imitated by a well-trained ANN. We show that the raw data numbers from the instrument can be used to train an ANN without being converted to physical units. Among other advantages, this technique avoids assumptions such as perfect calibration or a particular form for the ion distribution function, making it more robust. Repeating anomalies or non-ideal instrument behaviors that are opaque to the operator might also, in principle, be learned by the ANN and mitigated. In this paper, we train a neural network directly with low level data, bypassing the methods employed in typical space plasma instrument pipelines. As mentioned above, Wind SWE is used as our ground truth for training the ANN model as it provides well-calibrated and extremely well-vetted thermal solar wind ion data sets from a contemporary spacecraft in nearly the same orbit as DSCOVR. To optimize the model, we account for the time-shift difference between the two spacecraft using DTW. The DTW technique is implemented in python using the respective magnetic field measurements from both DSCOVR and Wind as time-matching keys. Application as a procedure for future missions such as HelioSwarm is discussed. Table 1: Data sources and variables for DSCOVR and Wind . Data for each span from 2017/01/01 to 2019/06/27. data type data source variables DSCOVR PlasMag FC (level 0) fc0 ( Rowland 2016 ) current spectra (engineering units) DSCOVR PlasMag FC (science) DSCOVR_H1_FC ( Kasper et al. 2021 ) V_GSE, THERMAL_SPD, Np Wind SWE WI_H1_SWE ( Ogilvie et al. 2021a ) Proton_W_nonlin, Proton_V_nonlin, Proton_Np_nonlin Wind MFI WI_H0_MFI ( Koval et al. 2021 ) BX, BY, BZ DSCOVR PlasMag MAG DSCOVR_H0_MAG ( Koval 2021 ) B1GSE II Data The DSCOVR science data and Solar Wind measurements from Wind SWE used in this study were retrieved from NASA’s Coordinated Data Analysis Web (NASA-CDAWeb) portal. The DSCOVR magnetic field data are from the data source "DSCOVR-H0-MAG", and the variable "B1GSE" in those files contains the field vector in the Geocentric solar ecliptic (GSE) coordinate system as a function of time while the magnetic field vector components for Wind are from "WI_H0_MFI". The uncalibrated DSCOVR data is the faraday cup level 0 "fc0" data, which is current spectra (current as a function of instrument applied voltage) in engineering units ( Rowland 2016 ) . The data source for Wind SWE, is "WI_H1_SWE" and the plasma variables in it produced by forward-modeling an anisotropic Maxwellian and fitting to the Wind SWE Faraday Cup measurements are: proton thermal