Key points are not available for this paper at this time.
This work explores the use of low-dimensional image approximations based on Proper Orthogonal Decomposition (POD) for data compression in order to enable real-time data downlink via low-rate telemetry. Two approaches are considered: one based on image projection and a second one based on interpolation via the Discrete Empirical Interpolation Method (DEIM). Both approaches are examined in the context of the upcoming ISS experiment “Effect of Marangoni Convection on Heat Transfer in Phase Change Materials” (MarPCM), which investigates the potential of thermocapillary-driven melting to enhance the performance of PCM devices. The experiment will provide real-time telemetry data at 1 Hz and high-resolution images, which can be downlinked in near real-time from the ISS with a latency of tens of minutes ( Porter et al., 2023 ). The first technique consists of projecting the experimental images onto a low-dimensional space formed by the leading POD modes — obtained here via Singular Value Decomposition (SVD) — of a representative image database, then downlinking the projection amplitudes as real-time telemetry. This approach requires building the image database, computing its SVD, and selecting the number of orthogonal modes to be used for projection. Once the projection amplitudes are received on ground, they can be used to (linearly) reconstruct the associated image(s) and thus track the experiment in real time. The second technique, using DEIM, relies on a limited set of pixel values whose location is selected to ensure that the interpolation problem is well-conditioned. Choosing between projection or interpolation involves a trade-off between computational cost and reconstruction quality. In both cases, the usual downlink of uncompressed images can be used to update the database and improve the quality of the reconstruction if needed.
Sánchez et al. (Mon,) studied this question.