Path Loss Does Not Tell You Which Microwave Band Wins

The study that prompted this article is Open Research Institute's Which Downlink? An Open Trade Study of Candidate Bands for futureGEO, published on 8 September 2026 with the byline “michelle” and explicitly presented as a draft for review. Microwave satellite payloads are not my specialist area. My interest here is in whether the comparisons hold together: what stays constant, where the noise enters, and what the numbers justify.

I read the report, both Python models and their CSV results, then reran the models using itur 0.4.0. All five CSV outputs reproduced byte-for-byte. I also checked representative results independently from received power density, effective aperture and thermal noise. That establishes reproducibility of this calculation; it does not turn estimated receiver figures into measurements or a candidate payload into qualified hardware.

The source edition matters. The code examined here is ORI commit f40cf7be0db36f9a20ed538d6f17484f20754de0. futureGEO itself is a feasibility and mission-definition effort: AMSAT-DL's workshop description sets out several candidate architectures. This article evaluates a trade study, not a selected flight design.

The cancellation is physical, but the comparison must be named

A dish collects power from the incident flux. For a dish of diameter D and aperture efficiency η, its effective collecting area is A_e = ηπD²/4. Holding this area constant is a sensible comparison when the user's constraint is an 80 cm reflector on the house.

These are linear quantities, with polarization matched, correct pointing and compatible antenna reference planes. The receive gain increases as frequency squared, cancelling the same dependence in free-space path loss. At fixed directional EIRP, distance and effective receive aperture, free-space received power is frequency-independent. Atmosphere, polarization mismatch and other losses still multiply that result.

This is the same gain–aperture relationship I use elsewhere. My preference for a dipole reference in dBi or dBd? does not change the calculation: Friis uses compatible linear gains, conventionally converted from dBi. A reference-label change cannot create or remove a link advantage.

ORI fixes spacecraft gain at 17 dBi as a footprint assumption, rather than holding a spacecraft reflector's diameter fixed. That is a legitimate trade-study boundary. Increasing transmit aperture gain at 24 GHz is a different service comparison if it replaces broad coverage with spot beams. As in my EIRP discussion of KJ6ER Primer 2, extra directional intensity must be associated with the directions receiving it.

What the published calculation actually produces

The models use 3.406, 5.840, 10.489 and 24.048 GHz, a GEO slot at 43°W, a 0.8 m receive dish at 65% efficiency, 12.5 dBW EIRP per analog channel and a 2,400 Hz noise bandwidth. Ten stations are listed; two are excluded by the code's elevation cutoff, leaving eight evaluated locations at each availability.

The following ranges come from the reproduced CSV, not from rounding the prose summary. “99.5%” refers to average-year propagation statistics at each location. Attenuation is the modeled atmospheric loss at that percentile, not an independently established service margin.

Thus the suggested 0.3–1 dB description is a loose envelope for 5.8 GHz across the two availability cases, not its exact 99.5% range. The 8–26 dB range at 24 GHz describes all eight evaluated stations at 99.9%; it should not be relabeled as a range drawn only from the wet-climate subset. Also, gaseous attenuation at 24 GHz spans 1.34–3.27 dB in these rows, rather than uniformly 2–3 dB.

Here is a reproducible spot check across three climates. The temperatures include the model's receiver and atmospheric-emission terms.

For the independent check I used P_r = EIRP × A_e/(4πR²) × 10^−A/10 and N = kT_sysB. Recomputing 12 station/band cases from the CSV's rounded atmospheric contributions agrees within 0.08 dB with its rounded C/N values. This is a separate algebra check; the atmospheric maps and their empirical validity remain shared inputs.

Receiver noise is capable of moving the result

The code estimates noise figures of 0.6, 0.8, 1.0 and 2.5 dB and clear-sky antenna temperatures of 40, 40, 50 and 90 K, respectively. Receiver temperature follows T_rx = 290(10^NF/10 − 1). A noise-figure change in decibels is therefore not the same numerical change in the link's C/N when the antenna is looking at a cold sky.

I changed only the 5.840 GHz noise figure from 0.8 to 1.8 dB. At Bochum, Miami and Accra the modeled C/N falls by 2.39, 2.42 and 2.35 dB. That consumes much of its original advantage over 10.489 GHz. It does not erase the much greater atmospheric penalty at 24 GHz, but it challenges any suggestion that small receiver changes cannot materially affect the closer band comparison. Feed loss before the first amplifier also attenuates the signal and introduces thermal noise; it belongs in the same reference-plane budget.

The sky-temperature accounting needs a clearer definition. The implementation adds 275[1 − 10^{−(A_gas+A_rain+A_cloud)/10}] K to the stated clear-sky antenna temperature. P.618-13, §3, equation 63 gives atmospheric sky emission plus the attenuated cosmic background. If the 40/50/90 K baselines already contain clear-sky atmospheric emission, part may be counted again; if they represent separately defined spillover and other contributions, the interpretation differs. The documentation does not resolve that decomposition, so I would call it an ambiguity, not a proven arithmetic bug.

A physically specified model would separate receiver noise, feed loss, ground spillover, background brightness and atmospheric emission, all referred to the same plane and weighted by the antenna pattern. That also explains why equal physical diameter alone does not guarantee equal G/T, the receive gain divided by system noise temperature.

Weather, pointing and coverage need their own boundaries

The atmospheric combination is A_gas + √[(A_rain + A_cloud)² + A_scint²], following P.618's prescribed statistical combination rather than simply adding all four terms. The code excludes scintillation from its emission-noise term. The inspected library defaults also matter: polarization tilt is 45° (the package's circular-polarization convention), and scintillation aperture efficiency defaults to 0.5 although gain uses 0.65. Making the latter explicit changes attenuation by less than 0.001 dB in my 12 checked cases, but leaving it implicit makes the experiment harder to specify.

The rerun used P.618-13, P.676-12, P.837-7 and P.840-7. ITU lists P.618-14 as the in-force edition; P.618-13 is superseded. Reproducing an older implementation is useful, but calling it the current recommendation is inaccurate. Neither edition makes an average-year percentile a guarantee for a particular weekend, worst month, obstructed site or simultaneous set of users. An average-year unavailability of 0.5% corresponds to about 44 hours per 365-day year; 0.1% is about 8.8 hours, with no promise that outages are evenly spread.

The spherical geometry yields about −0.4° for San Francisco and 4.6° for San Diego. The code excludes both below 5°. That is a defensible service-screening choice; it is not a universal lower validity boundary of P.618. The recommendation explicitly includes a rain-path formula below 5° and separate low-elevation scintillation/multipath treatments. Refraction, terrain and local noise matter near the horizon. Nor should these two exclusions be expanded into “no North American coverage”: Chicago, New York and Miami remain in the calculation.

At fixed diameter, beamwidth decreases roughly as λ/D. Wind deflection, mount errors, satellite position variation, feed alignment and reflector surface accuracy therefore become more expensive at the higher band. A loss of aperture efficiency from 0.65 to 0.50 costs 1.14 dB of signal gain alone. Polarization mismatch needs a separate coupling term; the library's rain-polarization setting does not include terminal misalignment or establish cross-polar isolation.

Spectrum is a separate filter. The January 2020 IARU summary supports the study's broad Region 1 exclusion at 3.4 GHz, secondary 5.830–5.850 GHz downlink and 10.450–10.500 GHz amateur-satellite entries, and primary 24.000–24.050 GHz entry. Its 3.4 GHz provision is 3.400–3.410 GHz in Regions 2 and 3 under footnote 5.282, broader than the deck's candidate slice starting at 3.402 GHz. These historical allocation statements are not current national authorizations or a satellite coordination decision. Terminal interference at 5.8 GHz also requires site-specific evidence; the thermal-noise budget does not model it.

The equal-service watts inherit the backoff assumption

The analog baseline starts with a 100 W traveling-wave-tube amplifier, then applies 6 dB output backoff, 1.5 dB transmit loss, 17 dBi gain and equal sharing among 50 channels:

The second expression reproduces the power inversion, including its use of C/N rounded to 0.1 dB. The median results are 36 W, 48 W, 100 W and 1,764 W; the maxima are 49 W, 62 W, 100 W and 3,802 W. Each station is matched to its own 10 GHz baseline at 99.5%, not to one common quality threshold. A median across eight selected stations is not a population-weighted coverage guarantee.

The word “watts” needs particular care here. With 6 dB backoff, the original 100 W amplifier delivers about 25.1 W operating RF before the 1.5 dB loss. Therefore, retaining the same backoff makes the 48 W and 1,764 W entries approximately saturation-power equivalents, with about 12 W and 443 W operating RF. Calling the latter 1.8 kW of delivered downlink RF blurs the reference plane. The power ratio remains useful; the hardware interpretation must include waveform and backoff.

A traveling-wave tube and a solid-state amplifier need not offer the same usable output, efficiency or linearity at the same nominal rating. Linearization can reduce the backoff needed for a given distortion limit; it cannot be assumed to make every multicarrier waveform a saturated carrier. Intermodulation, spectral regrowth, AM/AM and AM/PM behavior, filters, predistortion stability and temperature all belong in the comparison.

This connects directly to my power-budget article and PA operating-envelope discussion: dB describes a ratio, while watts and duty cycle determine much of the thermal burden. A continuously occupied digital downlink does not inherit a voice transmitter's intermittent-duty cooling assumption. Processor power, DC conversion, heat rejection, radiation tolerance and qualification remain outside this RF-only inversion.

Regeneration is valuable, but the channel counts compare different services

The architecture script changes to one DVB-S2 downlink carrier, 8 Msymbol/s, with 20% roll-off and 1 dB assumed output backoff. Its 34.5 dBW EIRP is 5 dB above the analog transponder's total EIRP. The operating RF rises from 25.1 to 79.4 W. This is an assumed single-carrier waveform and amplifier-operating advantage, not a gain caused by regeneration alone. A shaped digital waveform still needs an actual nonlinear-amplifier assessment.

The C/N bookkeeping independently checks out:

For example, Bochum's approximately 14.8 dB analog C/N becomes about 1.5 dB E_s/N₀. The script selects QPSK 2/5 with its threshold plus a 1 dB margin. Its 0.789 bit/symbol gives 6.312 Mbit/s; dividing by an assumed 32,000 bit/s per voice stream gives 197 equivalents. Across the 10 GHz rows the outputs are 197–297 such equivalents.

Three boundaries stop that number becoming a proof of four-to-six times as many simultaneous users:

• Bandwidth changes. Fifty 2,400 Hz analog channels sum to 120 kHz before guards. The digital carrier occupies approximately 9.6 MHz: an 80-fold ratio to that sum, not a same-bandwidth comparison. The AMSAT-DL payload-ideas deck (internal modification date 27 August 2026, slides 6 and 9) separately sketches a 300 kHz narrowband transponder and a 9 MHz wideband transponder; the addendum's nominal 10 MHz slot is another declared design assumption.
• The uplink has 64 channels. The script declares OPV_CHANNELS = 64 but does not cap its reported downlink equivalents at 64. A downlink able to carry 197 stream equivalents does not create 197 independent uplink demodulators. The stated channelizer admits at most 64 simultaneous single-channel uplinks before additional access or processing arrangements.
• Quality and overhead differ. Usable SSB at an assumed C/N and compressed digital voice at a chosen bitrate are not an equal-audio-quality test. Codec settings, framing, signaling, packet losses and acceptable delay must be part of a service definition.

The OPV v1.1 implementation draft specifies 40 ms frames and a reference implementation with 134 bytes before channel encoding, comprising a 12-byte OPV header and 122-byte payload. Those correspond to 26.8 and 24.4 kbit/s respectively, before interpreting the encapsulated protocol overhead. This helps distinguish the addendum's estimated 27.1 kbit/s information rate and 32 kbit/s downlink share from an exact codec bitrate or demonstrated end-to-end throughput.

Per-station MODCOD selection also does not assign every receiver its own full downlink capacity simultaneously. A shared broadcast stream must use a mode its intended audience can decode; adaptive time-sliced streams require a schedule and capacity accounting. Uplink decoding and re-encoding can prevent analog noise from being forwarded continuously, but cannot reconstruct packets already lost on the uplink. Power equalization after decoding likewise cannot protect an overloaded front end from a strong incoming station.

ACM does not require an onboard regenerative processor

Adaptive coding and modulation changes robustness and throughput with reception conditions. The DVB implementation guide A171-1, §4.4 and Figure 15 explicitly describes ACM through a transparent satellite, with the modulator at a gateway. That directly qualifies the report's inference that ACM implies regeneration.

Onboard regeneration can still be a compelling way to collect independent amateur uplinks into one downlink without a ground aggregation hop. The engineering case is access control, decoding, multiplexing and service flexibility. A gateway-based alternative has a different routing and latency budget. Neither architecture can manufacture link energy once its most robust supported waveform fails.

The script evaluates only one 8 Msymbol/s rate and a finite DVB-S2 MODCOD list. Its 24 GHz rows fail that configuration at 99.5%, but this is not proof that every transparent 24 GHz service is impossible. Reducing symbol rate by a factor of ten raises E_s/N₀ by 10 dB at unchanged C/N₀, at the price of approximately one-tenth the data rate and possible implementation constraints. Larger effective receive area or narrower spacecraft beams are other changed assumptions. Each purchases a different service; none is a free recovery of the original one.

The uplink check gives required C/N₀ values of 43.80 dBHz for the assumed SSB criterion and 49.83 dBHz for the addendum's OPV criterion. Their 6.03 dB difference follows from the stated thresholds and rates, not a measured equal-quality comparison. It implies about four times the EIRP under otherwise identical conditions; the report's illustrative 3 W versus 1 W is only 4.77 dB. Its −13 dB/K spacecraft G/T is also an estimate; the inspected 13 cm deck budget uses −12 dB/K. Absolute user-power figures must retain that distinction.

A single ground-to-GEO-to-ground passage already takes roughly a quarter of a second in propagation. Framing, interleaving, decoding and queueing add delay; a round trip is roughly twice the propagation contribution. An adaptive scheduler also needs feedback timing, rain-fade response and a policy for congested voice/data traffic. These are service requirements, not details supplied by a downlink C/N table.

What I would carry into the next trade

I accept the fixed-aperture argument and the reproduced result that 24 GHz is much harder for this broad-coverage, 0.8 m, high-availability service. I would keep 5.8 GHz as a serious candidate on modeled propagation and receiver-noise grounds, while treating its terminal noise and interference performance as requirements to establish. The 10 GHz comparison remains credible where an existing receiver ecosystem and host compatibility outweigh a few modeled decibels.

The next useful comparison should keep coverage, availability, simultaneous users, audio quality, delay and occupied bandwidth explicit, then compare usable RF output and total payload power. Regeneration deserves consideration on those terms. Calling a downlink bitrate quotient a user count, or calling ACM inherently regenerative, overstates what this otherwise useful open study establishes.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.