Open Research Institute draft for review. All code and data for this article are published alongside it and the goal is for every number below to be reproducible.
https://github.com/OpenResearchInstitute/documents/tree/master/Engineering/Link_Budget/futureGEO
Why this analysis
futureGEO is an ESA-sponsored initiative, coordinated by AMSAT-DL, to define an amateur radio payload for a future geostationary satellite with a footprint covering Europe, Africa, and the Americas. A payload-ideas working document recently posted to the public futureGEO GitLab presents link budgets built around a 10.45 – 10.5 GHz (3 cm) downlink. This is the QO-100 architecture, scaled up.
The 3 cm downlink of QO-100 is an excellent choice. It sits just below commercial Ku-band, so a modified satellite-TV LNB costing a few tens of dollars became the receiver, and tens of thousands of stations followed. The general level of microwave band involvement, operations, and expertise increased in the footprint as a direct result. But, there is a problem. Hosted payloads inherit the constraints of their host, and there is no guarantee that the eventual futureGEO host platform will offer a 3 cm-compatible downlink chain. The downlink band should therefore be studied. This article puts numbers on that trade using current ITU-R propagation models, so that the community discussion can be about data.
Method
We evaluate the four space-to-Earth candidates in the IARU amateur-satellite allocation table (January 2020 revision) that appear in the posted working document:
| Band | Frequency | Allocation | Regions |
|---|---|---|---|
| 9 cm | 3.402–3.410 GHz | Secondary | Regions 2 & 3 only |
| 6 cm | 5.830–5.850 GHz | Secondary | 1, 2, 3 |
| 3 cm | 10.450–10.500 GHz | Secondary | 1, 2, 3 |
| 1.25 cm | 24.000–24.050 GHz | Primary | 1, 2, 3 |
Propagation is computed with ITU-R P.618-13 (rain attenuation, using the P.837 digital rain-rate maps), P.676 (gaseous absorption), P.840 (clouds), and P.618 tropospheric scintillation, via the open-source itur package (v0.4.0). Ten ground stations spanning the candidate footprint are evaluated against the 43°W orbital slot used in the working document’s link budgets.
There are some assumptions here. To keep results synchronized with the posted document, we take its space-segment numbers: 12.5 dBW EIRP per narrowband channel (100 W TWTA, 6 dB output backoff, 1.5 dB losses, 17 dBi antenna, 50-channel power sharing) and a 2400 Hz channel. The 17 dBi spacecraft antenna gain is footprint-defined and therefore held constant across bands, which makes per-channel EIRP band-independent in this model. The user terminal is the document’s 0.8 m dish at 65% efficiency. Receiver noise figures are our estimates of representative achievable hardware: 0.6 dB (9 cm), 0.8 dB (6 cm), 1.0 dB (3 cm, PLL LNB), 2.5 dB (1.25 cm). Clear-sky antenna temperatures 40/40/50/90 K respectively. Rain-induced sky noise is added per the P.618 convention (275 K effective medium). These receiver estimates are assumptions, not measurements, and shifting any of them by small amounts of dB does not change the conclusions below.
For a fixed dish size, free-space path loss and dish gain both scale as frequency squared and cancel out exactly. Clear-sky performance differences between bands come only from receiver noise. Availability differences come almost entirely from the atmosphere. The band trade is therefore really a trade of weather, hardware ecosystem, and regulation and is not of path loss.
Results
Clearly Stated, 43W excludes the North American West Coast
Before any propagation is computed, we are not talking about a satellite visible from North America. From 43 W, San Francisco sees the satellite at −0.4 degrees elevation and San Diego at 4.6 degrees. This is below the 5 degree practical minimum for Ku/Ka-class GEO service and below the validity floor of the ITU-R P.618 model itself. (The posted document’s San Francisco budget lists 0.4° elevation. The half-degree difference from our spherical-Earth figure is within geoid and refraction effects, and both values mean the same thing, that there is no usable link.) I
(Figure 3: elevation contours to 43°W with the ten study stations.)
Attenuation
At 99.5% average-year availability, total atmospheric attenuation (gas + cloud + rain + scintillation) is:
- 9 cm and 6 cm: 0.3 to 1.0 dB everywhere studied. Propagation is a non-issue.
- 3 cm: 0.7 to 2.2 dB which is the worst case studied (Accra, 99.9%) is 4.6 dB. Manageable with modest margin, exactly as QO-100 experience suggests.
- 1.25 cm (24 GHz): 4.6 to 13.9 dB at 99.5%, growing to 8 to 26 dB at 99.9% in wet climates (Miami, São Paulo, Accra). Rain dominates and gaseous absorption alone (2–3 dB) exceeds the total 3 cm clear-weather budget. Wild stuff!
(Figure 1: attenuation vs availability at Bochum, Miami, and Accra.)
Per-user C/N
Holding the working document’s EIRP and terminal assumptions fixed, per-user narrowband C/N at 99.5% availability:
- 9 cm: 19 to 21 dB this is the best of all bands, but unusable for Europe and Africa (no Region 1 allocation), which for this mission is disqualifying. It is included as a reference point only.
- 6 cm: 17 to 19 dB this is roughly 3 dB better than 3 cm, from lower receiver noise and negligible rain.
- 3 cm: 14 to 17 dB gives comfortable margins, consistent with the posted document’s own San Francisco-based figure of ~15.7 dB.
- 24 GHz: −2 to +8 dB which at 99.5% availability the narrowband voice link is marginal-to-closed in dry climates and actually does not close in Accra. At 99.9% it fails by 10 dB or more across the wet-climate stations. A transparent bent-pipe SSB/NB service at 24 GHz with this EIRP class doesn’t look good.
(Figure 2: C/N per user by band and station.)
Discussion
If the host offers 3 cm, take it. The propagation cost over 6 cm is about 3 dB. This is real but affordable. No other band comes close on user terminal economics: the QO-100 ecosystem of modified PLL TV LNBs, proven software, and community knowledge transfers directly. The mass-participation story of QO-100 was built on a $15 receiver.
If the host cannot offer 3 cm, 6 cm (5830–5850 MHz) is the strongest fallback on physics it has the best C/N of any Region-1-legal band and essentially weather-proof. Its open questions are not propagation but regulation and interference: the allocation is secondary, and the receive band sits adjacent to the 5.8 GHz ISM environment (Wi-Fi, drones, industrial devices), so terminal-side interference susceptibility needs measurement, not assumption. There is also no COTS LNB to modify. An open-hardware downconverter design would need to fill the role the TV LNB played for QO-100. The good news? This is a well-scoped, community-sized project, and one ORI would welcome collaborators on.
24 GHz is only viable as a modern digital band, which is an argument about payload architecture, not just spectrum. The 20+ dB availability dynamics at 24 GHz cannot be bought back with static margin at amateur EIRP levels. What closes such links in commercial Ka-band practice is adaptive coding and modulation. And, this means a regenerative payload that trades data rate against weather in real time. If futureGEO retains a 24 GHz element (the primary allocation is genuinely attractive), it strengthens the case for a regenerative on-board processor with strong modern FEC, designed and published openly so the community can build, verify, and extend both ends of the link. Conversely, a purely transparent payload architecture effectively closes down 24 GHz as a user band.
The space segment and amplifiers
A fair objection to the analysis above is that the 100 W TWTA assumption is itself a Ku-band artifact. The flight-heritage TWT ecosystem exists because of decades of Ku direct-to-home broadcasting, and it does not automatically follow the payload to other bands. If the downlink leaves 10 GHz, the amplifier question becomes very big.
One way to handle this is to invert the link budget. Instead of asking what each band delivers at 100 W, ask how much total RF power each band needs to deliver the same service as the 3 cm / 100 W baseline, at the same stations and availability. Because path loss and fixed-dish gain cancel with frequency, the answer depends primarily on receiver noise and atmospherics.
| Band | RF power for equal service (median) | (worst station) |
|---|---|---|
| 9 cm (reference only — no Region 1) | 36 W | 49 W |
| 6 cm (5830–5850 MHz) | 48 W | 62 W |
| 3 cm (baseline) | 100 W | 100 W |
| 1.25 cm (24 GHz) | ~1.8 kW | ~3.8 kW |
Losing the Ku TWT ecosystem therefore does not mean losing the downlink. The leading fallback band needs roughly half the RF power for the same user experience. Fifty to sixty watts total at 5.8 GHz is well within established solid-state practice. GaN power transistors at 5.8 GHz are commodity technology terrestrially, and space-qualified C-band GaN SSPAs are flying today. This is an engineering assessment of the technology class, not a citation of a specific qualified part for this band. SSPAs also bring properties a small hosted amateur payload should actively want. There are no high-voltage power supplies, you get graceful degradation, and you have a natural path to distributed or redundant amplifier architectures. The open item is qualification cost and schedule for a non-standard band. That’s a big deal and needs leadership and discussion.
The same inversion closes off the 24 GHz case pretty hard. Kilowatts of RF for a bent-pipe narrowband service is not a hardware-selection problem, it is physics saying no. Any 24 GHz user service must earn its availability through waveform design. This means adaptive coding and modulation on a regenerative processor instead of raw transmit power.
Bent-pipe SSB vs regenerative digital
The working document’s link budgets price one architecture. It is a transparent transponder carrying 50 by 2400 Hz analog channels, with the TWTA backed off 6 dB to control multicarrier intermodulation. That backoff is a permanent 5 dB tax (relative to roughly 1 dB single-carrier operation) and it is paid on every channel, forever. A regenerative payload refunds it. You demodulate many uplink channels on board, aggregate them into one downlink carrier, and you run the amplifier near saturation. Here we price that alternative concretely, using ORI’s open Opulent Voice scheme as the worked example: a 10 MHz uplink segment channelized into 64 by 156.25 kHz channels (MSK, K=7 r=1/2 convolutional FEC), aggregated onto a single DVB-S2 downlink carrier (8 Msym/s in the same 10 MHz slot). Air-interface constants marked as estimates in the published code should be checked against the Opulent Voice specification.
For the downlink the backoff refund becomes capacity. With the same 100 W TWTA and 17 dBi antenna as the working document, single-carrier operation yields 34.5 dBW EIRP. Selecting the best DVB-S2 MODCOD per station (ETSI EN 302 307 thresholds, 1 dB margin, 99.5% availability, 0.8 m dish) gives the following results.
- 3 cm: QPSK 2/5 to 3/5 closes everywhere studied at 6.3 to 9.5 Mbps, or roughly 200 to 300 digital voice channels at 32 kbps net, against the bent-pipe’s 50 analog channels from the same tube. Regeneration multiplies voice capacity ~4 to 6 times and adds a data service besides.
- 6 cm: QPSK 2/3 to 3/4, 10.6 to 11.9 Mbps and here the band and architecture advantages compound.
- 24 GHz: Es/N0 lands 5 to 15 dB below the lowest DVB-S2 MODCOD at 99.5% availability. Even with ACM, a 0.8 m terminal at this EIRP class does not close. 24 GHz user service would additionally require larger dishes, spot-beam gain, or substantially more power. The earlier conclusion stands, now tested against the strongest available waveform rather than against SSB.
Uplink has price tags. Digital voice is not free. Requiring Eb/N0 of 4.5 dB (K=7 r=1/2 Viterbi at 10^-5) plus 1 dB implementation margin at an estimated 27.1 kbps information rate, an Opulent Voice uplink needs C/N0 of 49.8 dBHz, versus 43.8 dBHz for usable SSB in 2400 Hz. This is a 6.0 dB delta that is independent of spacecraft G/T, since it divides out of the comparison. At the working document’s 13 cm uplink with an estimated −13 dB/K spacecraft G/T, that is about 26 dBW required user EIRP versus about 20 dBW. For instance, 3 W into a 60 cm dish, versus 1 W into the same dish for SSB. Two considerations offset the 6 dB in real terms. First, minimum shift key (MSK) is constant-envelope modulation. The transmitter is a cheap saturated PA, where SSB demands a linear amplifier. Watt for watt, the digital watt costs less. Second, what the 6 dB buys is not just the same voice channel in digital form. You get FEC-protected voice and data framing, on-board power equalization (the regenerative payload inherently solves the strong-station problem that QO-100 polices with LEILA), and a downlink whose quality is independent of the weakest uplink. Whether that trade is right for futureGEO is a community decision, but it should be made with both columns of the ledger visible, which is what this section is for. The two architectures are also not mutually exclusive. The working document’s own requirement that the regenerative section degrade gracefully to transparent operation is one we endorse and have assumed throughout.
Limitations
This study covers the downlink only. Uplink band selection involves user-side PA economics and interference geometry not modeled here. The 5.8 GHz ISM interference question is flagged but not quantified. The equal-service power inversion assumes amplifier RF watts are exchangeable across bands at the payload level. Efficiency, thermal, and qualification differences between amplifier technologies are discussed qualitatively but not costed out in detail. Receiver noise figures are representative estimates, labeled that way above. P.618 statistics describe an average year at a fixed location and say nothing about worst-month behavior, which matters for event-driven amateur operating. Elevations below 5° are reported as non-viable rather than modeled, because the propagation models are not validated there. We know that these elevations can be super exciting for DXing and we believe people should try even if the propagation model “gives up” near the horizon.
Reproducibility
The complete analysis, including the domain model, ITU-R model invocations, and figure generation, is published with this article as two Python files (band trade and architecture addendum), along with the full results table (CSV). Rerunning it requires pip install itur. We welcome and invite comment, corrections, extended station lists, alternative slots, and disagreement in the form of pull requests.
Sources: AMSAT-DL futureGEO RFEI (May 2025); “Payload ideas” working document, futureGEO GitLab (August 2026); IARU amateur-satellite allocation table (Jan 2020 rev.); ITU-R Recommendations P.618-13, P.676, P.837, P.840; itur v0.4.0.