Relay Architecture

Almost nothing a Mars surface vehicle collects goes directly to Earth. It goes to a UHF radio, to an orbiter passing overhead for a few minutes, into that orbiter’s memory, and to the Deep Space Network hours later. The reason is mass and power, not bandwidth: a surface vehicle cannot carry the antenna gain or the transmitter power a direct-to-Earth link needs across hundreds of millions of kilometers, so a UHF link to an orbiter tens to thousands of kilometers overhead trades that cost for a scheduling problem instead, one that only exists while the orbiter is above the horizon [5]. For Spirit and Opportunity’s primary missions through September 2005, about 92 percent of total data return went through Odyssey, 5 percent through Mars Global Surveyor, and 3 percent over the X-band direct-to-Earth link [1]. Curiosity’s downlink is UHF relay only [2].

Why relay

The rover UHF return link runs at 8, 128 or 256 kbit/s to Odyssey or Mars Global Surveyor against an X-band direct-to-Earth downlink that falls to 10 bit/s in the fault configuration [1]. The link exists only while an orbiter is above the horizon, so the downlink becomes a scheduling problem rather than a bandwidth problem, and more than 60 percent of MER mission data return was planned through it. The 1997 case for relay, made before any orbiter had flown one, was the same trade stated in reverse: a lander or rover transmitting UHF to an orbiter needs on the order of 12 to 17 dB less power than the same bit rate sent direct to Earth, so the orbiter’s larger antenna and higher power absorb a cost the surface vehicle cannot carry [5]. That relay-first architecture, and its two competing pass-scheduling protocols, one polling on a fixed slot and one handshaking on request, were worked out on Mars Global Surveyor and the Mars Surveyor 98 Orbiter before Spirit and Opportunity ever landed [5].

The link-budget basis for all of this predates relay. The 1982 JPL reference volume on deep space telecommunications defines the end-to-end budget, from received power and noise spectral density through coding and the command link, against which a direct-to-Earth rover link is sized [14]. Its design control tables are pre-flight predictions for Galileo, whose high-gain antenna never deployed, so the tabulated gains were never realized [14]. A 2001 program-level paper made the relay case for the 2001 to 2014 mission set, with orbiter Earth-link capacity growing from Mars 01 to Mars Reconnaissance Orbiter to Ka-band and a planned Electra proximity payload and relay orbiter; every figure in it is a projection for missions not yet flown [13].

The measured record for MER comes from two sources that cover different periods. The DESCANSO report covering roughly one Martian year puts about 93 percent of returned data through Odyssey, 4 to 5 percent through Mars Global Surveyor and 3 percent over direct-to-Earth, out of 81 Gbit for Spirit and 71 Gbit for Opportunity, and notes that direct-to-Earth rates varied by more than a factor of fifty [12]. Its per-sol relay averages are pre-flight estimates [12]. A conference case study reports MER relay statistics only through 13 May 2004 and covers a 6 minute entry, descent and landing, with several values left as placeholders in print [11].

Pass geometry

	MER, Odyssey or MGS	Phoenix, Odyssey and MRO	Curiosity, MRO
Passes per sol per orbiter above 20 degrees elevation	1.8 average, minimum 3 per 2 sols, maximum 4
Passes per sol, both orbiters	up to 4	more than 5 average over 157 sols	2 planned, morning and afternoon
Pass duration	2 to 8 min
Pass times, local solar	MGS about 01:30 and 13:30, Odyssey about 04:30 and 16:30 at the Spirit site	forward link between 02:00 and 09:00, return between 12:00 and 19:00	afternoon pass 15:00 +/- 1 h at the equator

Sources: [1] for MER, [3] for Phoenix, [2] for Curiosity. Phoenix’s Odyssey and MRO supported 860 relay passes over the 157 sol landed mission, Odyssey providing 71 percent of the contacts; Phoenix limited itself to no more than two MRO passes per sol because MRO was still in its primary science mission [3]. Curiosity’s MRO pass start times range from 01:45 to 04:00 local mean solar time for the morning pass and 14:00 to 16:15 for the afternoon pass at 30 degrees north or south [2].

Geometry translates directly into volume. A higher maximum elevation gives both a longer pass and more time at short slant range, so the MER project preferentially scheduled high elevation passes [1]. Rover azimuth matters more than tilt, because the UHF antenna gain pattern is asymmetric while orbiter range dominates the tilt term: the same pass returning 50 Mb at a favorable azimuth can return half that if the high gain antenna assembly blocks the view. Rovers were eventually oriented deliberately in azimuth after a sol’s science to increase return, and the prediction program was later extended to handle tilt as well. On sol 278, 4 November 2004, Opportunity was tilted 31.04 degrees during the Odyssey afternoon pass, and the no-tilt against 31 degree predictions were 57.4 Mb against 41.5 Mb [1].

How much actually gets down

Mission	Planned or required	Achieved
Spirit and Opportunity	more than 60 percent of mission data return by relay	about 56 Mb/sol per rover via Odyssey, about 49 Mb/sol per rover via MGS
Phoenix	60 Mb/sol return, 30 Mb in a single afternoon overflight; 1 Mb/sol forward in a single morning overflight	more than 38 Gb over 157 sols, an average 242 Mb/sol, more than four times the requirement; 574 Mb of command data delivered, 97 percent via Odyssey
Curiosity	250 Mbit/sol prime via MRO, 100 Mbit/sol optional via Odyssey, planning assumption 125 Mbit per pass; decisional data capped at 100 Mbit/sol	modeled worst case no less than 125 Mbit/sol, average 687 Mbit/sol

Sources: [1], [3], [2]. Curiosity’s model assumes MRO performs a roll maneuver of up to 30 degrees every pass to point its UHF antenna at the rover, a 10 degree elevation mask, the rover at the equator, adaptive data rates, the best morning and evening passes, and 2 sigma return link margin [2]. Relay capacity follows a 5 or 6 sol pattern of two low volume passes followed by three or four high volume ones. The pre-landing design study for MSL’s radios had already worked from the same 125 Mb per pass, 250 Mb per sol planning numbers and the same adaptive-rate MRO analysis, so the operational figures Curiosity produced tracked what had been designed in rather than surprised the project [6].

Phoenix’s forward link exceeded its requirement in the same way: the commanding requirement was 1 Mb per sol, but once the link was characterized the uplink bundles regularly ran 1 to 2 Mb, and raising the Odyssey morning pass to a 32 kbit/s forward rate gave the science team about 12 Mb of forward capability that was rarely fully used [3].

Perseverance’s operational record after landing shows how much a single change in channel coding can move these numbers without any change to the spacecraft link budget. LDPC channel coding was demonstrated on four MAVEN overflights in April 2021, returned about 70 percent more data than convolutional coding predicted, and became the MAVEN default that August; the same 70 percent improvement held over the first 27 LDPC overflights, and a single Perseverance-MAVEN pass later returned 2340.2 Mbit, breaking both Curiosity’s 1739.3 Mbit prior record and Perseverance’s own 1373.5 Mbit record [8].

The orbiter buffer is part of the budget

Odyssey allocated 100 Mbit, 12.5 MB, of onboard memory to both MER rovers and to Beagle 2 under the pre-landing relay operations plan, later raised to 120 Mbit per rover for the primary mission [1]. What a rover can send in a pass is therefore constrained by what remains in the orbiter buffer from the previous pass, which in turn depends on the Deep Space Network coverage allocated to the orbiter [1]. When the MER return rate was raised to 256 kbit/s for some Odyssey passes, the allocation could be overflowed by a single overflight.

Prediction accuracy

Relay planning runs on predicted volumes, so the useful measurement is prediction error rather than throughput. For Phoenix and MRO over a 30 day period, with a 2 dB margin policy [3]:

Elevation mask	Average data volume delta	Passes more than 10 percent below predict
10 degrees	-0.22 percent, -0.262 Mbit	25.00 percent
15 degrees	+13.13 percent, +6.84 Mbit	3.75 percent

The Phoenix science team was shown this and chose to keep the 10 degree mask and leave the model alone, on the reasoning that the Odyssey model was conservative enough to guarantee the planned volume, with the rule that any critically needed data through MRO would be planned to a 15 degree mask instead [3]. Perseverance’s mission planning tooling runs the same prediction problem the other direction: data volume is precomputed for every overflight at every allowable rate and duration combination and then filtered against passes-per-sol, minimum and maximum volume, and latency-to-next-decisional-pass constraints, so the operational choice is which precomputed combination to fly rather than a single per-pass estimate [8].

Forward link and contingency

Relay carries commands as well as data. MSL uplinks sequences, flight software and other data through the relay orbiters, and expects to use the 03:00 MRO pass for sequence uplink more than half the time, with large non-time-critical data cleanup commands sent in that pass [2]. The Electra-Lite radio can wake the rover avionics on hearing a hail forward link from MRO, which is a contingency path independent of the rover’s own schedule.

Mars Express provided backup relay for Phoenix. It received the one-way link during part of entry, descent and landing, and a demonstration pass on the third surface sol verified relay capability at an 8 kbit/s forward and 32 kbit/s return rate [3]. Fifteen contact opportunities were booked for the critical first week, invokable at short notice by sending a direct operations request file to the Mars Express project, with the orbiter reusing a lander pointing mode tested with Opportunity in 2008 that boresights the Melacom antenna within 35 degrees of the landing site. Three of those passes were exercised. MSL’s own entry, descent and landing critical-event communications spread the same idea across four spacecraft at once: MSL’s X-band semaphore direct-to-Earth tones, MRO’s Electra recording the UHF stream open loop, Odyssey demodulating it in a bent-pipe relay, and Mars Express recording it open loop on Melacom as a second independent copy, so that landing telemetry survived the loss of any single link [6].

Governance and standards

Relay across multiple agencies’ orbiters is not just a radio compatibility problem. The CCSDS Proximity-1 Space Link Protocol standardizes the physical and link layers so a rover’s radio can hail an orbiter it was never specifically designed against, and the Mars Relay Operations Service coordinates scheduling and data handoff across the orbiters of more than one agency under bilateral service agreements; ExoMars Trace Gas Orbiter joined that framework as a relay provider in 2016 alongside the NASA orbiters already flying it [7]. A JPL strawman architecture from 2005, produced before any of the operational relay missions above had flown, had already argued for exactly this kind of standardized, multi-mission telecommunications and navigation network as the only way to sustain both human and robotic exploration traffic without a bespoke link for every mission [10].

Next-generation Mars relay

A JPL briefing proposes three relay orbiters at about 6000 km, directional X-band proximity links alongside legacy UHF, and a Ka-band trunk to Earth [15]. It states notional targets of more than 1 Tb/day for a science orbiter, against 30 Gb/day for MRO, and 200 Gb/sol for a Curiosity-class lander against about 1 Gb/sol today [15]. This is a proposal: the slide gives no link budget, contact-time model or cost, and the authors are the network’s proposers.

Lunar relay

Lunar relay is at the specification stage rather than the operational one for NASA. The LunaNet interoperability specification defines the service interfaces for lunar relay and position, navigation and timing services rather than a specific spacecraft [4]. China has already flown the alternative to a specification: Queqiao, in a halo orbit about Earth-Moon L2, carries a 4.2 m parabolic antenna, the largest ever flown in deep space at the time, giving above 44 dBi of gain and above 55 dBW EIRP at X band, and forwards commands to the Chang’e-4 lander and Yutu-2 rover simultaneously at 125 bit/s with more than 5 dB of margin. Return links reach 285 kbit/s from the rover and 555 kbit/s from the lander only with a directional dish on the surface, collapsing to 700 and 1400 bit/s on an omnidirectional antenna, because surface EIRP is about minus 1 dBW for the rover and the signal reaching Queqiao across the halo orbit’s 47,000 to 79,000 km range is below minus 136 dBm [9]. The Queqiao-2 relay serving Chang’e-6 has no open technical description in English.

Entry, descent and landing coverage

Relay is also the design basis for critical-event telemetry. After Mars Polar Lander was lost with no descent telemetry, every NASA mission has been required to cover entry, descent and landing with at least two independent links; the MSL account compares predicted and flown coverage for the 6 August 2012 landing, and its Earth-link prediction table omits the predicted percentiles, so the comparison is narrower than the abstract implies [16]. Phoenix flew the same idea with all three orbiters recording the lander’s UHF stream [18]. For the ExoMars Schiaparelli lander in 2016, open-loop UHF recording by Electra on Trace Gas Orbiter, processed on the ground, reconstructed the full 8 kb/s telemetry stream and made the landing anomaly diagnosable. The surface relay campaign that followed produced no lander data, so that half of the report is a plan rather than a performance record [17].

Orbit design, networking and operations

Orbit choice is a separate axis from the radio. A trade study for a Mars relay satellite supporting global surface networks compared elliptical orbits at the critical inclination, including a Molniya variant, for landers placed anywhere on the planet [19]. Both missions it served were never flown as described, so it is a paper study [19]. The 2021 proposal above (6000 km) is the current descendant of that question [15].

Store-and-forward networking is the other family. A simulation of Delay Tolerant Networking (Bundle Protocol over Licklider Transmission) for a one-week Mars relay scenario with two landers, two orbiters and three Deep Space Network stations used average rather than time-varying delay and omitted bit errors and finite buffers [20]. The 2008 DINET experiment ran the same protocols on the Deep Impact spacecraft at 24 million km for 27 days and is the flight evidence: store-and-forward bundling with custody transfer survived real link disruption, though two of four metrics missed their pre-declared criteria as measured [21]. The CCSDS Bundle Protocol profile standardizes the protocol for missions and contains no throughput, latency or storage figures [22]. A JPL architecture paper describes a 2030 to 2040 network built on DTN, optical plus RF links and voice and video relay, as a concept with no measurements [23]. An extension of the Deep Space Network loading simulator to hybrid RF and optical apertures, run on one notional Mars optical relay case with eight assumed ground sites, realizes 80 to 93 percent of requested downlink hours after 2027 and ignores weather [24]. Operations staffing on the network fell from 14 operators per antenna in 1980 to one per link by the 2000s through automation, using figures the paper quotes without method [25].

Surface-to-surface radio at Mars has no flown measurement basis. A survey’s ray-tracing simulations over Jezero Crater terrain find propagation terrain-dominated, and no Mars surface radio measurement exists to validate them [26].

Relay network since 2012: flown operations and planning

Flown. In the first 90 sols after the Curiosity landing the relay network, with Odyssey, MRO and Mars Express in place, returned 45.8 Gb over 339 passes, 509.28 Mb/sol on average [27]. That is one site, one season and an MRO still in staged commissioning, so it is not a steady-state figure [27]. MAVEN was converted from a science orbiter into a relay backbone: an aerobraking campaign from 11 February to 5 April 2019 used 6.8 kg of a 12.5 kg propellant budget and phased the orbit to be overhead at Mars 2020 touchdown [28]. For InSight’s 2018 landing, two MarCO CubeSats relayed entry telemetry as a bent pipe, with a 900-case Monte Carlo link simulation used to set margins; the paper reports carrier-only detection at the Green Bank and Effelsberg radio telescopes with negative margins at several milestones, and does not separate predicted from flown values [30].

Protocol. Every Mars proximity link runs on CCSDS Proximity-1. Version 3 frames carry a 5-octet header and up to 2043 data octets inside a 2048-octet maximum, with Sequence Controlled and Expedited qualities of service and hail-based session setup; the specification states no throughput [31].

Ground planning, modeled rather than flown as designed. MSL’s relay planning was rebuilt after InSight landed 600 km away and after the non sun-synchronous MAVEN and TGO joined, because their overflights drift against the planning shift; the team built an automated overflight selection tool with an ordered rule list, and the paper gives no efficiency figure before or after [32]. The earlier operations-efficiency calculation, which counts the fraction of planning cycles the ground holds data recent enough to command traverse, assumes 37 sols planned per 38 Earth days and idealized fixed pass times, so its values are ceilings [33]. A relay-pass allocation tool for the three co-located Mars Sample Return assets is a greedy model with no reported measurement [29].

Proposed. A JPL concept study sizes two small areostationary telecommunications orbiters for the 2020s; masses, powers and rates are parametric estimates with no hardware or flight data [34].

What is not established

LunaNet’s frequency plan and service catalog are the only pieces of the specification firm enough to design against; the optical link standard, allowable signal bandwidths and power levels, the frame ranging standard, and the lunar reference frame and time documents are explicitly to be determined, and the specification states no link budgets, achievable data rates, latency figures or availability numbers, so it cannot be used to size an actual lunar surface link [4]. The Mars figures above are two things at once and should not be read as one: MSL’s pre-launch design numbers are a prediction made three years before landing under an equatorial-site, best-pass assumption, while the MER and Phoenix numbers are what an all-relay or nearly-all-relay architecture measured after the fact, at one latitude, one season, and against two specific orbiters whose geometry does not generalize to other constellations [2] [6] [3] [1]. None of the held sources here give a systematic, mission-long statistic on predicted-versus-actual relay volume outside the one Phoenix/MRO month examined; Mars Exploration Rover comparisons were only made when a pass performed anomalously, so there is no residual-error distribution to compare across missions [1].