LEO satellite broadband user terminals (Starlink, Kuiper, OneWeb class)

Deep-dive: LEO satellite broadband user terminals — Starlink, Kuiper, OneWeb class

Research date: 2026-04-12

Target tier: deep-dive

Thesis under test: The user terminal is the binding constraint on LEO broadband economics. The operator that can commoditize a phased-array terminal—through cost, power, software-defined beamforming, and regulatory compliance—captures the margin pool, not the satellite operator alone.

Executive summary

A LEO broadband user terminal is not a passive dish; it is a software-defined phased-array radio that must continuously track fast-moving satellites, manage Doppler and handover, operate in unshielded outdoor environments, and meet regulatory EIRP/spectral-mask requirements.

For Starlink, the terminal has been commercially deployed at scale. A peer-reviewed measurement study, “A First Look at Starlink Performance” (ACM IMC 2022), measured real-world performance using active Starlink terminals and found median download throughput on the order of 100 Mbps and median latency in the tens of milliseconds, well below the theoretical link capacity of a clean 250 MHz-wide Ku/Ka channel [S2]. Kuiper has disclosed a family of terminals with a committed production-cost target under $400 for its standard unit [S5]. OneWeb has positioned itself primarily through enterprise, maritime, aviation, and government distribution partners rather than residential direct sales [S10].

Company filings add a risk layer that marketing disclosures lack. Amazon’s FY2023 Form 10-K states that Project Kuiper will require significant additional capital expenditures and that Amazon may fail to launch services on its planned timeline or at competitive prices [S4]. That filing does not repeat the “under $400” terminal figure as an unconditional commercial commitment [S5].

Source tier and evidence structure

This note uses:

• Tier 1 regulator: FCC Order and Authorization, FCC 18-38 [S1]; FCC Gen2 partial approval, FCC 22-91 [S7].
• Tier 1 peer-reviewed: ACM IMC 2022 study, DOI 10.1145/3517745.3561416 [S2].
• Tier 1 granted patent: U.S. Patent No. 10,735,075 B2 [S3].
• Tier 1 company regulatory filing: Amazon FY2023 Form 10-K [S4].
• Tier 2 self-reported company disclosures: Starlink residential service page [S6], Amazon Kuiper terminal announcement [S5], OneWeb technology page [S10], Amazon Kuiper first launch announcement [S9].

Every load-bearing technical claim is cited to one of those sources. Where a parameter is assumed for the worked calculation, it is flagged as assumed or derived, and no performance conclusion is inferred from it.

System architecture of a LEO user terminal

A LEO user terminal has five main blocks:

1. Phased-array antenna aperture: an array of radiating elements, each with gain/phase control, used to steer a beam electronically.
2. RF front-end / beamformer RFIC: low-noise amplifiers, phase shifters, attenuators, and combiners, often integrated into silicon beamformer ICs.
3. Modem / baseband SoC: waveform modulation, forward error correction, timing and Doppler correction, and satellite handover logic.
4. Power supply and thermal system: handles outdoor temperature extremes, condensation, and the substantial DC power draw of the active array.
5. Software and network control: satellite ephemeris, beam pointing, network entry, and interference mitigation.

The fundamental architectural change versus a GEO dish is that LEO terminals must track a satellite that moves across the sky every few minutes. Starlink commercially addressed this with a motorized, self-aligning early terminal and later moved toward fixed rectangular phased-array units. The granted SpaceX patent record [S3] discloses phased-array antenna structure; this is a recorded claim, not proof of commercial performance or component cost.

Kuiper’s disclosed terminal lineup follows the same phased-array logic. Amazon describes its standard terminal as an 11-inch-square phased-array unit using Ka-band, with the compact terminal smaller and intended for residential customers [S5]. OneWeb terminals likewise use electronically steered flat panels or mechanical tracking units from Hughes, Intellian, Kymeta, and Satcube; OneWeb’s public materials list enterprise, mobility, maritime, aviation, and government use cases [S10].

Phased-array RF front end and beamforming

A phased array steers its beam by deliberately shifting the phase of each element’s signal. For a receive array, the signal from each element is phase-shifted before combination so that signals arriving from a chosen direction add coherently. The same concept works in transmit.

The terminal must operate at frequencies where LEO constellations are authorized. For Starlink, the FCC order in FCC 18-38 authorized operation in Ku/Ka bands including 10.7–12.7 GHz downlink and 14.0–14.5 GHz uplink, among others [S1]. These frequencies place the array elements close together; at 12 GHz, free-space wavelength is approximately 2.5 cm. Maintaining a usable beam requires many elements across a panel, and each element requires a phase/gain control. The patent record [S3] and the Kuiper terminal disclosure [S5] both describe active phased-array apertures.

For a rectangular aperture with physical area \(A\), efficiency \(\eta\), and wavelength \(\lambda\), the directive gain is approximately:

\[

G = 10 \log_{10}\left(\frac{4 \pi A \eta}{\lambda^2}\right)

\]

If the terminal has an effective aperture area of 0.15 m², efficiency of 0.5, and operates at 12 GHz where \(\lambda \approx 0.024983\) m, then:

\[

G = 10 \log_{10}\left(\frac{4 \pi (0.15)(0.5)}{(0.024983)^2}\right) \approx 31.8 \ \text{dBi}

\]

That gain is high enough to support a terrestrial-satellite link with a satellite EIRP of tens of dBW. It is also the reason the terminal is expensive: the RF front end must repeat amplification/phase control across many elements, and the antenna must be calibrated.

The worked calculation in Section 7 uses this gain-derived G/T value as an input and does not claim that any specific commercial terminal achieves exactly these numbers.

Modem, baseband, and network software

The modem must compensate for two LEO-specific effects:

1. Doppler shift: at Ku/Ka band, a satellite moving at orbital velocity produces frequency shift that changes rapidly as the satellite passes.
2. Handover: a constellation with many satellites requires deterministic or network-controlled handover every few minutes.

The IMC 2022 study [S2] is useful here because it observed sustained TCP and UDP behavior on real terminals. The study reported median round-trip latencies in the tens of milliseconds, with throughput variation and lower performance in obstructed or congested conditions. Critically, the study did not validate any vendor’s claimed “theoretical” link rates; it measured application-visible performance on operational hardware.

Regulatory authorization also constrains the modem. FCC 18-38 imposed conditions on SpaceX including power flux-density limits, elevation angle limits, and coordination obligations [S1]. The FCC’s Gen2 partial approval, FCC 22-91, again conditioned operations on interference mitigation and coordination [S7]. These are not mere paperwork; they constrain terminal transmit power, beam selection, and uplink scheduling.

Power, thermal, and environmental constraints

Active phased arrays consume significant power. The IMC 2022 study reported active Starlink terminal power draw in the range of roughly 50–75 W under test conditions; this is an independent measurement but was taken on one hardware generation and should not be assumed identical across all Starlink versions [S2]. Kuiper has not published an independently measured power draw for its production terminal. Any figure for Kuiper or OneWeb terminal power must be marked “needs primary source validation.”

Thermal design is also a differentiator. A consumer terminal must survive direct sunlight, rain, snow, ice, and temperature cycling. Because each element in an active aperture dissipates heat, thermal spreading and cooling are design constraints. The granted patent record [S3] shows a phased-array packaging approach, but patent drawings do not independently prove outdoor reliability.

Starlink’s user-facing documentation [S6] describes obstruction and environmental constraints in operational terms, and the IMC study [S2] confirms that obstruction and satellite visibility affect throughput and outage behavior.

Worked link budget calculation

This calculation is illustrative, not a validation of any operator’s proprietary link budget. Inputs marked “assumed” are not sourced from a vendor publication; if these assumptions are wrong, the final capacity estimate changes accordingly.

Inputs

Free-space path loss

\[

\text{FSPL} = 20 \log_{10}\left(\frac{4 \pi d}{\lambda}\right)

\]

At \(40^\circ\) elevation:

\[

d = \frac{550}{\sin 40^\circ} = 855.6 \ \text{km}

\]

\[

\text{FSPL} = 20 \log_{10}\left(\frac{4\pi (855{,}600)}{0.024983}\right) \approx 172.7 \ \text{dB}

\]

At \(25^\circ\) elevation:

\[

d = \frac{550}{\sin 25^\circ} = 1{,}301.5 \ \text{km}

\]

\[

\text{FSPL} = 20 \log_{10}\left(\frac{4\pi (1{,}301{,}500)}{0.024983}\right) \approx 176.3 \ \text{dB}

\]

The path loss difference between 40° and 25° is about 3.6 dB.

SNR estimate

Using log-domain link budget:

\[

\text{SNR} = \text{EIRP} - \text{FSPL} - L_{\text{other}} + G/T + 228.6 - 10 \log_{10}(B)

\]

At \(40^\circ\):

\[

35 - 172.7 - 2 + 10 + 228.6 - 83.98 \approx 14.9 \ \text{dB}

\]

At \(25^\circ\):

\[

35 - 176.3 - 2 + 10 + 228.6 - 83.98 \approx 11.3 \ \text{dB}

\]

Shannon-bound capacity

\[

C = B \log_2(1 + 10^{\text{SNR}/10})

\]

At \(40^\circ\), SNR ≈ 14.9 dB:

\[

C = 250 \times 10^6 \log_2(1 + 30.9) \approx 1.24 \ \text{Gbps}

\]

At \(25^\circ\), SNR ≈ 11.3 dB:

\[

C = 250 \times 10^6 \log_2(1 + 13.5) \approx 0.96 \ \text{Gbps}

\]

Sensitivity

• If terminal G/T degrades by 3 dB, SNR falls by 3 dB; at 40° the Shannon bound drops to about 1.0 Gbps.
• If satellite EIRP is 3 dB lower than assumed, the same drop occurs.
• Real-world application throughput is far below the Shannon bound because of multiple user sharing, framing, scheduling, FEC overhead, and protocol effects. The IMC 2022 measurement [S2] reported median download throughput around 100 Mbps—about one order of magnitude below the clean 250 MHz single-user bound above. That gap is normal for shared satellite access, but it is the central contradiction between “theoretical capability” and “delivered consumer experience.”

Parameter table

IP landscape and patent evidence

The patent record establishes claimed technical territory, not operational performance.

For this note, I reviewed the granted U.S. patent record at:

https://patents.google.com/patent/US10735075B2/en

U.S. Patent No. 10,735,075 B2 is a granted patent record associated with a phased-array antenna implementation. Grant date: 2020-08-04. Accessed 2026-04-12 [S3].

I do not use this patent as evidence that any Starlink terminal reaches a particular throughput, cost, or reliability. A granted patent is evidence of filed and examined claims; it can support a moat only if the claims are practiced, detected, and enforceable. The patent record [S3] is useful here for the same reason all terminal patents are useful: it maps a design space that competitors must engineer around.

Kuiper’s disclosed terminal family [S5] does not itself disclose patent numbers, but Amazon, like SpaceX, has broad patent filings in satellite communications. For this note, I am not citing an Amazon terminal patent because I have not verified a specific granted Amazon patent number resolving to the Kuiper terminal design.

Tier discipline: A patent is not a peer-reviewed performance paper. The IMC 2022 study [S2] is the stronger source for “does this work in practice.” The FCC order [S1] is the stronger source for “is this legal to operate.”

Peer-reviewed performance evidence

The IMC 2022 study [S2] is the highest-quality independent evidence used here. It is a peer-reviewed ACM conference paper, DOI 10.1145/3517745.3561416, published October 2022.

Relevant findings relevant to user terminals:

• It measured real Starlink terminals, not lab emulators.
• Median download throughput was on the order of 100 Mbps, with difference across regions and test locations.
• Latency was in the tens of milliseconds, showing that LEO terminal latency is network/geometry dominated in good conditions.
• Throughput and outage behavior degraded with obstructions or poor sky-view geometry.
• The study did not measure Kuiper or OneWeb terminals because those constellations were not generally available at the time.

This paper is stronger than a company marketing page because it is independent, but it has limits: it is a snapshot of early Starlink hardware, subject to changes in software and capacity.

Contradictions and source-bound claims

11.1 Amazon marketing vs Amazon 10-K

Amazon’s March 2023 announcement states that the Kuiper standard customer terminal is designed to cost under $400 to produce [S5]. Amazon’s FY2023 Form 10-K describes Project Kuiper as requiring significant additional capital expenditures and notes that Amazon may not complete the project on its planned timeline or at competitive prices [S4]. These are not direct numerical contradictions, but the 10-K does not warrant the recurring under-$400 production cost as a realized commercial outcome. The terminal cost claim remains self-reported [S5].

11.2 Theoretical link capacity vs measured user experience

The link budget in Section 7 yields a Shannon capacity above 1 Gbps for a single 250 MHz carrier under the assumed conditions. The independent IMC study [S2] measured median download throughput about 10× lower. This is not proof of vendor failure; it is the expected effect of shared capacity, scheduling, modulation/coding selection, and TCP behavior. But it is a critical contradiction between “terminal theoretically supports X” and “user sees Y.”

11.3 Patent record vs certification

The granted patent [S3] describes a phased-array design. It is not an FCC equipment authorization. FCC 18-38 [S1] authorized the constellation, while individual user terminals must still meet equipment authorization and EIRP limits. Patent claims and regulatory certification are separate evidentiary regimes.

Unit economics / business model

How operators monetize

• Starlink: sells a bundled terminal plus recurring broadband subscription. Starlink’s residential page currently lists residential hardware and monthly service; as of this research date, U.S. residential service is listed from $120/month and hardware from $499 [S6]. This is self-reported Tier 2 pricing and can change.
• Kuiper: plans to monetize through Amazon’s consumer and enterprise channels. Amazon has disclosed a standard terminal production-cost target under $400 and has positioned the service as part of Amazon’s broader ecosystem [S5].
• OneWeb: monetizes through distribution partners and service providers rather than direct-to-consumer. Its public product positioning is enterprise, government, maritime, aviation, and backhaul [S10].

Cost structure

The structural cost driver is the active phased-array RF front end: the antenna aperture, beamformer RFICs, multi-layer PCB, and calibration/test. This is an inference from the architecture [S3][S5][S1], not a disclosed bill of materials.

Margins

None of the terminal vendors publicly discloses terminal unit gross margin. Consumer satellite broadband has historically used subsidized hardware to acquire a service subscriber, meaning terminal gross margin can be negative at first. That statement is an inference based on common telecommunications practice and is not a sourced figure; treat terminal-level margin claims as “needs primary source validation.”

Who captures value

If terminal cost falls below a threshold, operator captures service margin. If terminal cost remains high, component suppliers and module makers capture margin. A company that can integrate beamformer, modem, and software in-house—and pass regulatory certification—captures more terminal value. The patent record [S3] and cost disclosure [S5] are consistent with integration, not disintermediated chip sales.

Disconfirming evidence

The thesis would be wrong or weakened if:

• Independent, multi-region measurements show median Starlink/Kuiper/OneWeb terminal download throughput below 10 Mbps outside congestion hotspots, contradicting the broad “low-latency broadband” equivalence.
• A credible teardown and bill-of-materials analysis shows production cost above the retail hardware price at planned scale, breaking the Kuiper under-$400 cost assumption [S5].
• FCC equipment-authorization records show Kuiper or OneWeb consumer terminals lagging satellite launch progress by years, indicating terminal commercialization is the bottleneck.
• Amazon’s 10-K later discloses terminal manufacturing delays or per-unit losses, contradicting its terminal cost narrative.
• The granted patent record [S3] is invalidated, narrowed, or shown not to be practiced, reducing the terminal IP moat from a patent perspective.
• Independent testing shows the Starlink terminal power draw [S2] cannot be maintained in warm climates without throttling, undermining outdoor terminal reliability.