Hard core dismantling the physical layer of Starlink——OFDM frame structure × beamforming × DTC actual measurement fully disclosed

Understand in one sentence:16 public documents, FCC documents, patents and measured data from around the world finally pieced together a complete picture of the Starlink physical layer. The two satellites - one connecting Dishy to broadband and one connecting mobile phones directly to text messages - use two completely different sets of physical layers. This article explains the OFDM frame structure, modulation and coding, beamforming, and DTC measured signal-to-noise ratio clearly at once.

1. Why study the physical layer of Starlink?

Since its launch in 2019, Starlink has launched more than 7,000 satellites and served 4 million+ users. But SpaceX has always been tight-lipped about the details of its physical layer - there is no white paper, no 3GPP contribution, and even the signal waveform is a private design.

As a result, at least five independent research teams around the world have used spectrum analyzers, blind signal identification, hardware timestamps, and even reverse engineering methods to dismantle Starlink's signal structure from the "outside" in the past three years. Coupled with SpaceX's own technical documents and patents submitted to the FCC, we can finally piece together a relatively complete physical layer map.

The first thing to note is that Starlink has two types of satellites:

——It’s also called “Starlink”, but the physical layer is completely different.

2. VSAT Broadband Satellite: Complete Anatomy of Private OFDM

2.1 Spectrum: 8 250 MHz channels, total downlink bandwidth 2 GHz

Starlink broadband satellites use traditional FSS (Fixed Satellite Service) frequency bands:

Source cross validation:SpaceX FCC original application (SAT-LOA-20161115-00118) + NRAO VLA radio telescope actual measurement + FCC expanded authorization document in January 2026 - the three parties' data are consistent.

2.2 OFDM frame structure: 1000 subcarriers, 287 symbols, 430.5 Mbps baseband

This is the core number of the entire Starlink physical layer, derived from cross-validation by three independent teams:

UT Austin (Humphreys 2023)The OFDM frame structure was disclosed for the first time using blind signal recognition, including the positions and partial sequence values ​​of PSS (primary synchronization sequence) and SSS (secondary synchronization sequence).

OSU (Neinavaie 2024)The "OFDM-like signal structure" was independently confirmed, and blind reception and tracking of multiple satellites were achieved using GLRT (Generalized Likelihood Ratio Test).

Karlstad University (Garcia 2024)They took another approach - instead of using a spectrum analyzer, they used a hardware timestamp network card (Intel X550T, with an accuracy of tens of nanoseconds). By analyzing the arrival time patterns of 500 million data packets,The physical layer transmission rate is deduced from the Ethernet level..

The results of the three parties are highly consistent:

what does that mean?

430.5 Mbps per 250 MHz channelPhysical layer baseband rate. Multiplied by 8 channels, the theoretical physical layer downlink capacity of a single V2 Mini is approximately 3.4 Gbps. With the addition of 16QAM/64QAM high-order modulation, it can actually be even higher.

2.3 Synchronization and Piloting: 0.66% vs 100% Beacon Secret

Kozhaya 2025 (ION Navigation)Made a shocking discovery: the previously disclosed Starlink OFDM synchronization sequence (PSS/SSS) only accounts for 10% of the complete beacon.0.66%.

By blindly estimating the received signal, Kozhaya foundFull OFDM beacon covering the entire time-frequency resource grid. Using full beacons for signal capture, you can getApproximately 18 dB additional processing gain- This means Starlink signals can be reliably detected and tracked even with low-gain, low-cost receiving antennas.

Follow-up research at UT Austin 2025 further found that Edge Pilots operate on all frames, all beams, all channels, and all satellites.completely consistent, and is a fixed 4QAM symbol. Using these frame-level predictable elements you get~48 dB processing gain.

In a word: Starlink's signal is far more "predictable" than you think - in certain scenarios, this is not a bug, but a feature.These predictable pilot and synchronization sequences are not designed for confidentiality, but to enable terminals to quickly acquire signals under extremely low signal-to-noise ratios.

2.4 Frame timing: Hardware can be used as a GPS, but software cannot

UT Austin 2025 accurately measured the relationship between Starlink frame timing and GPS time for the first time:

This means:Starlink hardwarefully equippedThe short-term stability capability of GPS-level PNT (positioning, navigation and timing) - nanosecond-level jitter is enough to support centimeter-level positioning. However, software-level frame adjustments (jumps on the order of hundreds of nanoseconds per second) and periods of high jitter (once every 15 seconds for about 15 seconds) prevent this capability from being directly used for accurate PNT.

The conclusion is clear:This is not a hardware limitation, it is a software design choice. Once SpaceX decides to "unlock" this capability, Starlink may become a global PNT system at the same time.

Additional background:There has been a lot of research exploring using Starlink signals to replace GPS navigation. At the 2024 ION conference, multiple teams demonstrated the results of using Starlink downlink OFDM signals to achieve positioning accuracy of tens of meters. The discovery of full OFDM beacons will further improve this accuracy.

3. Direct-to-Cell (DTC): running an LTE base station in the sky

3.1 System architecture: eNodeB on satellite

Physical layer of DTC system and VSAT broadband satellitecompletely different. The core difference is:

The satellite does not carry a private modem, but a complete LTE eNodeB.

This is the standard 3GPP LTE protocol stack running in space - the phone does not require any hardware or software changes, because from the phone's perspective, it is just connected to a base station flying 550 kilometers in the sky.

3.2 Spectrum: It’s not Starlink’s spectrum, it’s T-Mobile’s

DTC service uses T-Mobile's PCS G Block spectrum:

3.3 Physical layer actual measurement: Use 1 million pieces of data to tell you the truth

UPM / Weplan Analytics 2025Published the most comprehensive DTC physical layer measurement research to date. Utilizing more than 1,000 data collected in the United States between October 2024 and April 20251 millionCrowdsourced measurement data reveals the actual physical layer performance of DTC networks for the first time.

RSRP (Reference Signal Received Power) – How weak is the signal?

What is -121 dBm? This is equivalent to the signal strength at the edge of a terrestrial LTE cell. But please note that this signal is sent from a satellite 550 kilometers away and moving at 7.8 km/s - getting -121 dBm at this distance is an engineering miracle in itself.

RSRQ (reference signal reception quality) - is the signal "clean"?

RSRQ of DTC signal insteadbetter thanGround network 3 dB. This is not because the satellite signal is strong, but because there are currently very few DTC users and co-channel interference is almost non-existent.

SINR (signal-to-interference-to-noise ratio) – does it actually work?

0 dB SINR means signal power and noise + interference powersame size. This is not a wideband signal-to-noise ratio - this is a "barely decodable" level. No wonder the current DTC can only use the most robust QPSK and low bit rate encoding.

3.4 Throughput Bottleneck: Path from 4 Mbps to 12 Mbps

Based on the measured SINR data, the research team gave an accurate estimate of DTC throughput:

What is the concept of 4 Mbps per beam?

The coverage diameter of a beam on the ground is about 25-50 kilometers. All users in this area share this 4 Mbps. This is why DTC currently only opens SMS services - it is not SpaceX that limits you, but the laws of physics that limit it.

But note:The V2 Mobile satellite (expected to launch in mid-2027) aims to150 Mbps peak. This leap does not depend on a single improvement;Extended spectrum (MSS 2 GHz) + larger satellite antenna array + lower orbit (~340 km) + higher transmit powerA four-pronged approach.

4. Beams and antennas: from 192 beams to 2,048

The number of beams is the most intuitive indicator to measure satellite capacity - it determines "how many times frequency reuse can be done."

Data source: SpaceX official V3 page + FCC Gen3 constellation application (July 6, 2026)

An underrated upgrade to V3 isuplink beam. The V2 Mini has only 144 beams in the uplink and 192 in the downlink - an asymmetric design. V3 expands both uplink and downlink to 2,048 beams, which means that the satellite is no longer just a "broadcasting" device, but has become a true two-way high-density space base station.

On July 16, 2026, Starship Flight 13 launched into space carrying 20 V3 prototype satellites——This is the first time that SpaceX has launched a satellite platform natively designed for D2D into orbit.

5. Two types of satellite physical layers - explained clearly in a table

6. Three “black boxes” and one main line

Black Box #1: DTC Why not use private waveforms?

The answer isMobile phone chip does not support.

Every 4G/5G mobile phone sold in the world today only has 3GPP standard waveforms (LTE OFDMA/SC-FDMA and NR OFDM/DFT-s-OFDM) in the baseband chip. If Starlink wants to use private waveforms, Qualcomm, MediaTek, Apple, and Samsung will all need to rewrite the baseband firmware or stream new chips.

So SpaceX made a pragmatic choice:Let the satellite adapt to the mobile phone, rather than let the mobile phone adapt to the satellite.The eNodeB on the satellite simulates a terrestrial LTE base station, including Doppler pre-compensation, timing advance compensation, and even a simplified RACH process that limits random access.

The trade-off is efficiency: Private waveforms can do better in spectral efficiency. But the advantage is zero terminal adaptation - this is the fundamental reason why DTC can quickly increase in volume.

Black Box #2: How is V3’s 2,048 beams achieved?

SpaceX has not released details of the beamforming chip. But based on V3 design specifications and patents:

Self-developed beamformer chip

(Beamformer IC): V3 uses a custom silicon-based beamforming chip designed by SpaceX instead of a commercial off-the-shelf chip. This is the core enabling technology to achieve 2,048 beam density

Digital beamforming (DBF) instead of analog beamforming

: V2 has implemented some digital processing, and V3 is expected to be fully upgraded to a fully digital beamforming architecture - independent ADC/DAC + FPGA processing for each antenna unit

The area of ​​phased array antennas has increased significantly

: The physical size of V3 is approximately 2-3 times that of V2 Mini (7m × 3.5m, mass ~2,000 kg), providing the physical aperture required to achieve high-gain narrow beams

Black Box #3: What’s in the User Terminal (Dishy)?

DARKNAVY 2025Complete reverse engineering of the Rev3 (GenV2) user terminal, and some of the findings are quite surprising:

41 SSH keys, port 22 always open - this design choice shows that SpaceX has extremely high requirements for UT's remote management and diagnostic capabilities. The EAL5+ security chip provides a hardware root of trust, but the firmware's lack of encryption means nearly all code is visible once the device is physically touched.

7. Main line: Starlink physical layer is not "a technology", but a "technology matrix"

After reading the two sets of physical layers, looking back at the entire Starlink architecture, there is a clear main line:

This main line can be summarized in two sentences:

The physical layer of broadband satellites pursues "the limit of spectrum efficiency"——Private OFDM, high-density beam, laser ISL, ultra-large aperture antenna.

The physical layer of DTC satellites pursues "the limit of terminal compatibility"——Standard LTE waveform, operator spectrum reuse, and zero mobile phone modification.

The two operate in the same constellation, share the same ISL network, and are scheduled by the same operation control system - but they face completely different physical law constraints and business logic.

8. Technical variables worthy of attention in the next step

DTC upgrade from LTE to NR (5G NR NTN): 3GPP Release 17/18 has defined the NTN framework. If V2 Mobile is equipped with NR payload (rather than pure LTE), it will have more flexible numerology and higher spectrum efficiency. When the FCC authorized the Gen3 constellation in July 2026, it mentioned "advanced capabilities", implying that NR is on the road map.

Terminal ecology of MSS 2 GHz spectrum: The AWS-4 spectrum (2000-2020 MHz) that SpaceX acquired from EchoStar is not supported by any existing phones. Tim Farrar pointed out that this requires an 18-24 month adaptation cycle on the chip side - if Apple and Qualcomm do not cooperate, SpaceX may launch a "Starlink mobile phone".

Feed link for Q/V/W bands: The feed link of V3 supports Ka/E/V/W four frequency bands with a total of 60 GHz spectrum. Among them, the E-band (71-76/81-86 GHz) and the W-band are the "new frontiers" of satellite communications - atmospheric attenuation is high, but the usable bandwidth is extremely wide.

From PNT bypass to PNT main: Multiple teams have proven that Starlink signals can be used for positioning and navigation. If SpaceX removes frame adjustment jitter at the software level, Starlink could become an independent PNT system that does not rely on GPS - which would have far-reaching implications in the military and critical infrastructure fields.