We Tested a Satellite Ground Station for Six Weeks. Then Came the Antenna Decision.

Archi4 min read

satellitesspacetestingoperations

Contents
  1. Six Weeks in a Parking Lot With a Dish and a Laptop
  2. What the Passes Actually Looked Like
  3. The Number That Changed the Conversation
  4. The Decision They Made
  5. Field Notes on the Parts Nobody Warns You About
  6. Pointing Accuracy Eats Your Budget
  7. Interference Is a Neighborhood Problem
  8. The Software Was the Long Pole
  9. What We Would Test Next

Six Weeks in a Parking Lot With a Dish and a Laptop#

The ground station sits on a concrete pad behind a warehouse in a light industrial park, which is the least glamorous place in the satellite business. We had a 1.2 meter dish, a software defined radio, a laptop running a tracking pass scheduler, and a six week window to prove the whole thing could hold a link to a small satellite in low Earth orbit. The company we were working with builds payloads for agricultural imaging. They had been renting time on someone else's ground network and paying per pass. The question they wanted answered was simple on the surface: do we build our own, or do we keep renting?

By week four we had the answer to a different question. The link worked. What broke was the schedule.

What the Passes Actually Looked Like#

A LEO satellite gives you roughly eight to twelve minutes above the horizon per pass, depending on altitude and inclination. In our test, the average usable window was nine minutes and forty seconds from acquisition of signal to loss of signal. The radio locked on reliably. The modem cycled through its rates without drama. We logged bit error rates and framed the results against weather, since rain fade at our band was not a rounding error.

Here is where the testing got honest. Our antenna could only point at one thing at a time. The satellite we were tracking came over on a schedule that was not ours to set, and twice a day it crossed the sky while another pass from a different spacecraft was already in progress. We had to pick. On the days we picked wrong, the customer lost a strip of imagery they had already paid to collect.

The Number That Changed the Conversation#

The rental network gave them access to eleven antennas spread across four continents. Our single dish gave them one. Over six weeks we measured 61 attempted contacts and completed 47. Of the 14 we missed, nine were because the satellite was below our horizon, three were weather, and two were conflicts with other passes. The renters missed nothing, because they had somewhere else to point.

The company's operations lead put it plainly in a meeting: a ground station is not a receiver, it is a calendar. That reframed the build decision entirely. The hardware was never the hard part. The dish, the radio, and the mount together cost less than a year of rental, and we could assemble it in a day. The hard part was scheduling across a fleet of satellites with overlapping visibility windows, and that is a software and geography problem.

The Decision They Made#

They did not cancel the rental contract. They built the single station anyway, but they changed what it was for. The in house dish became the primary path for downlinking engineering telemetry and for testing new payload firmware before a launch. Commercial imagery kept flowing through the rented network, where capacity scaled with the number of antennas rather than the number of dishes they could bolt to a pad.

Field Notes on the Parts Nobody Warns You About#

Pointing Accuracy Eats Your Budget#

A degree of pointing error at our beamwidth was tolerable. Two degrees was not. We spent three days chasing a mount calibration issue that looked like a radio fault and was actually a slightly loose azimuth encoder. The lesson is boring and true: the mechanical chain has to be as tight as the RF chain, and nobody enjoys testing that.

Interference Is a Neighborhood Problem#

Our site was quiet for the first two weeks, then a tenant two buildings over installed equipment that raised our noise floor in the lower part of the band. We did not fully solve it. We documented it, shifted the schedule to favor passes with better elevation angles, and flagged site selection as a real risk for any expansion. A ground station's performance is partly a property of the land around it.

The Software Was the Long Pole#

Pass prediction, conflict resolution, and a queue that humans can override at 2 a.m. took more engineering hours than the entire hardware install. The team built a scheduler that ranks passes by customer priority and data value, and that ranking logic is now the most valuable thing on the site. It is also the least visible. Nobody photographs a scheduling algorithm.

What We Would Test Next#

Two dishes instead of one, which would tell us whether the conflict problem scales linearly or falls off a cliff. A second site a few hundred kilometers away, which would let a single satellite hand off between stations and double the contact time. And a full month of unattended operation, because six weeks with an engineer on site is not the same product as six months without one.

The company is still deciding on all three. Their build versus rent question turned out to be the wrong frame. The real question was which jobs belong on their own antenna and which belong on someone else's, and the answer depended on how many satellites were in the sky at once.

More articles