How the radios work

The two designs buy back the same link budget in opposite ways

Distance consumes link budget, and a satellite is a long way off. There are two ways to buy the budget back, and the two leading operators each chose a different one. Beam count, fleet size, interference management and replacement cadence all follow from that single choice (Exhibit 1).

AST’s answer is aperture: a very large antenna. Its authorisation order, adopted on 21 April 2026, describes each satellite carrying a large phased array that runs thousands of electronically steerable beams at once, sized to produce narrow primary beams with low sidelobes. The beams are shaped to fit service-area boundaries, and the order records that they can be switched off wherever they would cause interference. A bigger antenna concentrates energy into a smaller patch of ground, so more signal lands where it is wanted and less where it is not. The difficulty is flying an antenna that large and keeping it pointed.

SpaceX’s answer is proximity and numbers. The FCC’s order of 26 November 2024 authorises supplemental coverage across up to 7,500 Gen2 satellites in shells at 340, 345, 350 and 360 km. Flying lower shortens the path, and flying many satellites shortens the wait for one to be overhead. Neither lever asks anything special of the antenna, so the two approaches are opposites rather than variations on a theme. On 6 October 2026 the Commission went further in the same direction. Its order of that date authorises a separate constellation of 15,000 direct-to-device satellites in nine shells between 326 and 335 km, lower than anything in the 2024 order, with half of them to be in orbit by 7 October 2032 and all of them by 7 October 2035. The new satellites are authorised for the 2 GHz bands SpaceX is buying from EchoStar as well as the PCS G block it rents, and SpaceX says they will carry more than a hundred times the bandwidth of the current generation. The bandwidth claim is the company’s and concerns hardware that cannot be inspected; the order fixes the altitude, the bands and the dates.

The orbits show SpaceX doing what the order describes. Altitudes computed from CelesTrak two-line elements at epoch 17 September 2026 put 824 of the 10,702 Starlink objects in those element sets, or 7.7%, below 400 km, the band the order describes. The catalogue cannot say which of those objects carry direct-to-cell hardware; the only source for that is SpaceX’s prospectus of 12 June 2026, which puts its V1 Mobile fleet at approximately 650 satellites. The 824 is evidence that the low shells are being filled, and the 650 is the company’s own count of satellites serving phones. They are different measurements.

Neither design is obviously the right one; the useful observation is that each puts the cost in a different place (Exhibit 2). Aperture is expensive per satellite and awkward to launch, so AST’s is a capital and manufacturing problem, most of it incurred before the first satellite reaches orbit. Proximity is cheap per satellite and expensive in fleet size, atmospheric drag and replacement, so SpaceX’s is a continuing operating problem. We know of no public document that settles which trade works out better over a decade.

Only the service link is low and narrow

Each satellite uses three kinds of link, and the FCC authorises each in a different part of the spectrum (Exhibit 3). The service link reaches the handset on leased terrestrial spectrum. The gateway or feeder link connects the satellite to a ground station, and telemetry and command keep the satellite itself in contact.

The position of the service links explains the designs. Lower frequency buys link budget, because a signal at 700 MHz gets through walls and bends around terrain far better than one at 1.9 GHz, which is why AST’s partners lend it their low-band holdings rather than anything further up the band plan. The gateway and telemetry bands follow the same logic in reverse: they sit far higher, where bandwidth is plentiful and nothing has to reach a handset through a wall.

SpaceX has now acted on the same lesson. On 8 October 2026 it agreed to buy Grain Management’s 800 MHz portfolio, up to 14 MHz of paired spectrum at 817 to 824 and 862 to 869 MHz, and described it as a coverage layer whose job is to get the signal “through obstacles, such as walls” while the 2 GHz spectrum carries the capacity (SpaceX update; part 4 has the price and the history). The physics that makes low band worth buying has a cost, and it falls on the satellite rather than the phone. Free-space loss falls with frequency, by about 7.7 dB between 2 GHz and 820 MHz (our calculation), so the lower band closes a link the higher one cannot. An antenna’s gain for a given size falls by the same amount, however, so a satellite that wants the same beam on the ground at 800 MHz needs an aperture of about six times the area. AST made that trade when it built its arrays for the 700 MHz blocks. The 15,000 satellites the FCC authorised on 6 October are built, on SpaceX’s own description, for 2 GHz, and the 800 MHz band is not among the frequencies in their grant. The low band therefore goes to work first on the ground, in what SpaceX calls an advanced terrestrial deployment, and from orbit only once a satellite with the aperture for it exists. The relation between frequency, loss and aperture is the textbook one, and the bands and the grant are on the record; neither link has been modelled here.

Low band is narrow, and width is the capacity story. The 700 MHz blocks are 6 MHz wide and the PCS G block is a single 5 MHz pair. Whatever the constellation overhead is doing, the service is metered through a channel that narrow, shared by everyone under the beam, and neither aperture nor altitude changes it (Exhibit 4).

More satellites buy availability and revisit. They do not widen the channel.

A tower comparison shows the scale. Five megahertz is among the narrowest carriers a cellular site is built around, and a tower uses one to serve the few hundred people within reach. A satellite beam shares the same width across everyone it can see at once, which can be a city. Capacity per user then comes down to bandwidth, spectral efficiency and how tightly beams can be reused, so the beam shaping in AST’s order matters at least as much as the satellite count in SpaceX’s. A headline satellite count is an availability claim rather than a throughput claim.

The regulator sets the ceiling on transmit power

The ceiling on how hard a satellite may transmit comes from the regulator, because this spectrum is terrestrial and the binding limits concern interference into everyone else using the band on the ground, a constraint no coverage map shows. AST’s order records that it filed power flux-density and maximum EIRP density values with the FCC, and that it can adjust EIRP across operating altitudes and elevation angles to stay inside them. The link budget says how hard AST would like to transmit; the regulator says how hard it may.

On SpaceX’s side one limit stayed open for nearly two years. SpaceX asked the Commission in 2024 to waive the aggregate out-of-band power flux-density limit of −120 dBW/m²/MHz for these operations, and the Commission deferred the question, so an authorised service ran with a power question unanswered. The order of 6 October 2026 answers it. The waiver is granted, and in its place SpaceX must hold the aggregate out-of-band flux density to −110.6 dBW/m²/MHz in the 5 MHz on either side of its downlinks in the H block and the PCS G block, 9.4 dB more than the rule allows, on condition that it causes no harmful interference to terrestrial licensees and stops transmitting on notice if it does (Exhibit 5).

For engineering, the number to design to is now −110.6, and the attached condition, that operations stop on notice of harmful interference, is the one to plan around. For strategy, the two years the question stayed open show that regulatory headroom, rather than technology, may cap this service. It is one of two technical items that belong in any diligence on this market and seldom appear there; the other is the gap between authorised, catalogued and payload-carrying satellites that part 1 sets out.

An independent cross-check

One independent check is available at no cost. The SatNOGS database holds telemetry logged by volunteer ground stations. It records AST BlueBird beacons at 430.5 to 439.5 MHz (GFSK and FSK, 2.4 to 19.2 kbaud) and S-band telemetry at 2225 and 2235 MHz (BPSK, 200 kbaud). Both S-band frequencies fall inside the 2200 to 2290 MHz command-and-telemetry band in the FCC authorisation, so the paperwork and the radio agree.

Amateur observation cannot see the service payload. It does confirm that the satellites are where the paperwork says and transmit what the paperwork says, and when an unrelated source agrees with the filing we read that as a reason to trust the documents describing the service.

Terms used

• Link budget. The account of power gained and lost between a transmitter and a receiver: transmit power, antenna gains, path loss over distance and losses through walls or terrain. A link closes when enough signal remains at the receiver.
• Aperture. The effective collecting area of an antenna. A larger aperture concentrates energy into narrower beams with more gain, which is the lever AST pulls.
• Phased array. An antenna of many elements whose signals are combined electronically, so beams can be steered and shaped without moving parts.
• EIRP. Effective isotropic radiated power: the power an antenna would need if it radiated equally in all directions to produce the signal it sends in its beam.
• Power flux-density (PFD). The power arriving per unit of area, in dBW/m²/MHz. Regulators limit it to protect other users of a band. The rule’s aggregate out-of-band limit is −120 dBW/m²/MHz; SpaceX’s waiver of 6 October 2026 sets −110.6 within 5 MHz of its downlink band edges.
• 800 MHz, band 26. The 817 to 824 and 862 to 869 MHz pair SpaceX agreed to buy from Grain Management on 8 October 2026, which the standards call band 26. Low band in the same sense as AST’s 700 MHz blocks, and a 7 MHz pair, so a single channel wide.
• Service, gateway and telemetry links. The service link carries the user’s traffic between satellite and handset. The gateway or feeder link carries it between satellite and ground station. Telemetry and command keep the satellite itself in contact.
• SatNOGS. A database of transmitters and observations logged by volunteer ground stations.