Why Bandwidth Budgets Favor Sparse Acoustic Features Over Point Clouds
The Mars Relay Arithmetic That Forces Sparse Features
A flight concept team writing a Mars cave cartography proposal usually runs the relay math early because the answer changes every sensor decision downstream, and a sensor whose output cannot be returned through the actual link is a sensor that does not fly. The DESCANSO MER telecommunications chapter reports MER primary mission returns of about 120 Mb per rover per day. The Mars Relay Network overview describes orbiters returning data at gigabit-class aggregate volumes, but the per-rover, per-pass budget is what matters for cartography.
The DESCANSO MRO telecommunications chapter gives the Electra proximity link a range of 1 kbit/s to 2 Mbit/s depending on geometry. A 7-minute pass at 1 Mbit/s carries roughly 52 MB; at 128 kbit/s (a realistic working rate inside a pit geometry) it carries about 6.7 MB. A single LiDAR sweep of a lava tube section produces hundreds of megabytes of raw points. JPL's IPNPR note on X-band RFI adds another constraint: 11 active Mars spacecraft share the X-band channel and windows are contested. The only cartography architecture that survives this arithmetic is one that returns sparse, high-information features rather than raw geometry.
Pit geometry makes the Electra link arithmetic worse than the surface case. A rover sitting at the bottom of a Marius Hills void looks at the relay orbiter through a narrow solid angle bounded by the pit rim, which means the link sees the higher-rate end of the Electra range only during a small fraction of pass elevations. Multipath inside the pit further degrades the achievable bit rate by introducing intersymbol interference that the standard Electra demodulator was not designed for, and a full set of link budgets shows how the pass-timing arithmetic compounds across a multi-sol campaign. Working concept teams have converged on planning numbers closer to the 64-128 kbit/s bound than the 2 Mbit/s peak, and EchoQuilt's patch sizing assumes the lower number to leave margin for unfavorable geometry.
EchoQuilt's Quilt-Patch Compression vs Point Cloud Returns
EchoQuilt writes the cave into quilt patches whose deltas compress aggressively. A typical patch update is 8-32 kbit, encoding geometric features inferred from passive acoustic impulse responses rather than a raw point cloud. An hour of continuous traverse returns roughly 8 kbit of quilt patch deltas at the default compression, which fits into a fraction of a single Electra pass. By contrast, even state-of-the-art point cloud compression from MDPI Sensors on efficient large-scale point cloud geometry compression and arXiv multiscale compression work struggles to reach the kb/hr scale without sacrificing the geometric detail that made the LiDAR worth carrying.
The compression ratio works because EchoQuilt's patches encode geometric primitives, not samples. A wall is described by a plane equation and a roughness parameter; an aperture is described by a bounding curve and an estimated traversability score; a cavity is described by its modal frequencies and an inferred volume. Each primitive is a handful of floats with associated uncertainty, and the entire descriptor table for a kilometer of traverse fits inside a few tens of kilobytes. The decoder on the ground rebuilds a navigable mesh from the primitives, and the science team works with the mesh as if it had arrived as a dense scan. The fidelity gap relative to a true LiDAR scan exists, but it is bounded, measurable, and shrinks as the patch inference model is trained on more analog data.
The quilt architecture works because acoustic geometry inference is already a feature extraction step. The pipeline does not produce a point cloud and then compress it; it produces named geometric features (wall normal, bounding planes, opening apertures) tied to uncertainty envelopes and writes each as a patch descriptor. A Martian relay window therefore carries decoded science, not compressed raw data. The link budget conversation shifts from "how do we fit a point cloud in 2 Mbit/s" to "how many patches can we return per pass" - a much healthier framing for a flight review.
A companion post on feature extraction over Mars relay links describes the specific descriptor formats EchoQuilt emits and the symbol-level encoding tuned for Electra UHF demodulators.
A second architectural insight matters for concepts targeting Mars Sample Return-adjacent cave traverses. Because the patches are decoded science rather than raw data, the science team can begin interpretation during the pass rather than after a long ground-side decompression cycle. EchoQuilt's descriptors include a coarse geometric summary in the first kilobit of each patch so that even a partially received patch produces a usable bounding box and surface-normal estimate. Reviewers preparing for a sample return campaign cite this incremental usability as the deciding feature when comparing acoustic cartography to LiDAR-derived alternatives that produce no usable information until the full point cloud has arrived and been registered. The implication for operations planning is that science decisions can ride the same Electra UHF pass that carries the patches, rather than waiting for a downstream synthesis cycle that consumes another sol of mission elapsed time.
Advanced Tactics for Bandwidth-Aware Quilt Design
Three tactics raise the fidelity of a quilt returned at an 8 kb/hr budget. First, run a lossless patch descriptor log on the rover and a lossy downlink descriptor on the link. The rover stores the full-fidelity quilt locally (cheap, given modern flight-grade NAND) and downlinks only the delta between the expected patch and the measured patch. Passes that return almost-empty deltas are a feature, not a waste; they confirm the model.
Second, prioritize patches by information gain rather than by traverse order. A patch at a junction or an aperture transition carries more geometric information than a patch in a uniform corridor. EchoQuilt scores patches by information gain and sends the top-ranked patches first, so the ground team's early quilt is already the useful one. If a pass is cut short, the remaining patches degrade gracefully.
Third, treat the relay window as a scheduling constraint, not a transmission constraint. The quilt pipeline is fully asynchronous: patches are inferred, scored, and queued during the traverse; the radio ships the queue during the pass. A team that models its cartography pipeline this way sidesteps the familiar failure mode where a rover finishes a traverse with a full buffer and no way to return it before the next pass.
A fourth tactic that compounds the previous three is to exploit DTN's bundle expiry semantics for graceful old-data shedding, the same approach cave diving teams use for data-constrained mapping where rebreather noise constrains audio bandwidth. EchoQuilt tags each queued patch with a freshness score and a science priority, and when bundle expiry approaches, the lowest-priority patches are dropped before higher-value ones. This lets the rover keep operating during multi-sol comm blackouts (common when a relay orbiter is in eclipse or a contested DSN window pushes the next downlink off the schedule) without cascading failure. The patches that survive into the eventual downlink window are the highest-information ones the campaign produced, and the science team sees a quilt that prioritizes useful coverage over completeness.
JPL telecom planners working on Mars Relay Network and Electra UHF constraints have shipped similar drop-rules on heritage missions, and reviewers respond well to seeing the pattern adopted explicitly in a cartography concept.
CTA
EchoQuilt is looking for JPL Mars cave concept teams, NIAC PIs working on pit-and-cave missions, MatISSE proposers, and ESA LunaNet relay designers who need a cartography payload that lives inside an 8 kb/hr link budget without sacrificing usable geometric coverage. We provide the quilt patch descriptor format, the information-gain scheduler, and the delta encoder tuned for Electra UHF demodulators across the 64-128 kbit/s working range. Each pilot ships with a measured Mars Relay Network link-budget analysis sized for pit-bottom geometry at Marius Hills or Pavonis Mons, a DTN bundle expiry configuration tuned to Mars Sample Return-adjacent campaign schedules, and a coarse-summary descriptor module that produces a usable bounding box and surface-normal estimate from the first kilobit of each patch.
Pilot PIs shape the symbol-level Electra encoding and the information-gain priority weights that the 2027 reference release will adopt for Artemis-era LunaNet relay integration. Priority goes to teams submitting cave concept proposals in the 2026 NIAC or MatISSE cycle, JPL telecom planners working on Mars Relay Network drop-rules, and ESA LunaNet relay designers coordinating multi-agency proximity link standards. Join the Waitlist for Planetary Analog Researchers to get the link-budget emulator and reference descriptors before your next concept review.