On-Board Processing is Non-Negotiable: Latency Dominates Ground Processing

We built a Monte Carlo simulator to resolve a fundamental autonomy question for Phase 0 Prospecting Satellites: Should spectral unmixing happen on-board or on the ground? The answer is definitive—and it's not about bandwidth.

The Question

The Prospecting Satellite consensus specification identifies on-board vs ground spectral unmixing as an open question. On-board processing requires radiation-hardened compute hardware (mass/power/cost). Ground processing preserves raw data and allows algorithm updates. Which approach maximizes survey effectiveness?

We built a Monte Carlo simulator comparing both approaches across thousands of simulated asteroid encounters.

The Definitive Finding

Ground-based processing achieves only 47-50% survey efficiency compared to 100% for on-board processing.

Metric	On-board	Ground
Targets Surveyed	~7,000	~3,300
Survey Efficiency	100%	47.5%
Avg Decision Latency	1.1 hrs	7.3 hrs
Bandwidth Utilization	<1%	~97%

The 2× survey improvement is too significant to trade for ground processing advantages.

The Critical Insight: It's Latency, Not Bandwidth

We expected bandwidth to be the limiting factor—raw hyperspectral data is 10-100× larger than processed results. But the simulation reveals something surprising:

All missed opportunities are due to latency, not bandwidth:

• On-board missed by bandwidth: 0
• On-board missed by latency: 0
• Ground missed by bandwidth: 0
• Ground missed by latency: ~3,700 (in baseline configuration)

Even with 50 Mbps bandwidth (5× baseline), ground processing only improves marginally. The fundamental latency chain dominates:

1. Raw data downlink: ~2-4 hours (large hyperspectral datacubes)
2. Light travel time: ~0.3 hours (round trip at typical NEA distance)
3. Ground queue + processing: ~6 hours (DSN scheduling, compute time)
4. Command uplink: ~1 hour (targeting instructions)

Total: ~7+ hours vs typical NEA encounter windows of 2-12 hours.

Why Encounters Are Time-Limited

NEA survey opportunities are fleeting. As satellites and asteroids follow different orbits, observation windows depend on:

• Geometry: Distance, phase angle, illumination
• Velocity: Relative motion limits observation duration
• Priority: High-value targets (M-type, C-type) need follow-up observations

When spectral analysis identifies a high-value target, the satellite must decide immediately whether to extend observation or move to the next target. A 7-hour ground loop exceeds most encounter windows, causing the satellite to miss the follow-up opportunity entirely.

Scaling Analysis

We tested configurations across various constellation sizes, bandwidths, and mission durations:

Configuration	On-board	Ground	Missed
50 sats, 10 Mbps, 7 yr	7,000	3,300	+3,700
100 sats, 10 Mbps, 7 yr	14,000	6,600	+7,400
55 sats, 50 Mbps, 7 yr	7,700	3,800	+3,900
55 sats, 25 Mbps, 15 yr	16,500	8,000	+8,500

The ~50% efficiency gap persists across all configurations. More satellites, more bandwidth, longer missions—none fix the latency problem.

The Bandwidth Bonus

While not the primary driver, on-board processing delivers substantial secondary benefits:

• 97% bandwidth savings: Processed results (10 MB) vs raw data (100s MB)
• Reduced ground station costs: Fewer DSN hours required
• More communication margin: Contingency bandwidth for anomalies
• Scalability: Support larger constellations without DSN bottlenecks

Implications for Project Dyson

1. Mandate On-Board Spectral Unmixing

The 2× survey improvement justifies radiation-hardened processing hardware. Budget for space-qualified GPUs or FPGAs.

2. Implement Selective Raw Data Retention

Store high-priority observations for later downlink during low-activity periods. This enables ground reprocessing and algorithm validation without sacrificing autonomy.

3. Design for Algorithm Updates

Over-the-air updates allow improving unmixing models based on pathfinder validation. The on-board library isn't frozen forever—just during active encounters.

4. Level 3 Autonomy is Validated

The consensus specification calls for Level 3 autonomy (30-day independent operation). This simulation confirms that autonomy is not merely convenient—it's essential for effective surveying.

The Physics Behind Latency

The latency chain is fundamentally constrained by:

Speed of Light: At typical NEA survey distances (0.1-0.3 AU), light travel time adds 0.3-0.9 hours per round trip. This is irreducible.

DSN Scheduling: Ground stations serve multiple missions. Queue delays average 2-6 hours depending on demand.

Data Volume: Raw hyperspectral datacubes (256×256 pixels × 128 bands × 12-bit depth) require ~100 MB per target. At 10 Mbps, downlink alone takes 80+ seconds under ideal conditions—but real conditions include link margin, error correction, and scheduling gaps.

Try It Yourself

We've published the interactive simulator so you can explore the on-board vs ground trade-off. Adjust satellite count, bandwidth, mission duration, and encounter rates to see how survey efficiency changes.

Methodology

The simulation uses:

• Poisson encounter generation based on expected NEA survey rates
• Hyperspectral data volume modeling (configurable resolution and bands)
• Bandwidth-constrained downlink windows with overhead factor
• Latency chain modeling for ground processing pipeline
• 50 Monte Carlo runs per configuration for statistical validity

Results should be interpreted as relative comparisons between processing approaches.

What's Next

This research answers RQ-0-2 and provides definitive guidance on Prospecting Satellite autonomy architecture. Combined with constellation sizing (RQ-0-3) and fleet logistics (RQ-0-19), we're building simulation-validated specifications for Phase 0.

Remaining research priorities include:

• Specific radiation-hardened processor selection and power budget
• On-board spectral library optimization for asteroid endmembers
• Pathfinder experiment design for empirical validation

---

Research Question: RQ-0-2: On-board vs ground spectral unmixing

Interactive Tool: Spectral Analysis Simulator