Direct-to-Cell: A First Look into Starlink's Direct Satellite-to-Device Radio Access Network through Crowdsourced Measurements

Low Earth Orbit (LEO) satellite mega-constellations have emerged as a viable access solution for broadband connectivity in underserved areas. In 2024, Starlink, in partnership with T-Mobile, began beta testing an SMS-only Supplemental Coverage from Space (SCS) service. This marks the first large-scale deployment of Direct Satellite-to-Device (DS2D) communications, allowing unmodified smartphones to connect directly to spaceborne base stations. This paper presents the first measurement study of deployed DS2D technologies. Using crowdsourced mobile network data from the U.S. between October 2024 and July 2025, we provide evidence-based insights into the capabilities, limitations, and future evolution of DS2D technologies for extending mobile connectivity. We find a strong correlation between the number of satellites deployed, the number of unique cell identifiers measured, and the volume of measurements, concentrated in accessible areas with poor terrestrial network coverage, such as national parks and sparsely populated counties. Stable physical-layer measurements were observed throughout the period, with a 24-dB lower median RSRP and a 3-dB higher RSRQ compared to terrestrial networks, reflecting the SMS-only usage of the DS2D network during this period. Based on the SINR measurements collected, we estimate the expected performance of the announced DS2D mobile data service to be around 3 Mbps per beam in outdoor conditions. We also discuss strategies to expand this capacity up to 18 Mbps in the future, depending on key regulatory and business decisions, including allowable out-of-band emissions, permitted number of satellites, and availability of spectrum and orbital resources.

💡 Research Summary

This paper presents the first large‑scale measurement study of a commercially deployed Direct Satellite‑to‑Device (DS2D) network, focusing on Starlink’s “Supplemental Coverage from Space” (SCS) service that was beta‑tested in the United States from October 2024 through July 2025 in partnership with T‑Mobile. Using millions of crowdsourced LTE measurements collected from Android devices, the authors quantify how the growing Starlink constellation translates into observable cellular‑style cells, assess the physical‑layer radio performance, and project the capacity of the forthcoming data‑enabled version of the service.

The dataset, supplied by Weplan Analytics, contains standard 3GPP radio‑access parameters (RSRP, RSRQ, SINR, EARFCN, ECI) together with PLMN identifiers. The authors isolate Starlink measurements (MCC = 310, MNC = 830 or 210) and compare them with T‑Mobile terrestrial measurements (MNC = 260). They also compile an external time series of the number of Starlink satellites equipped for DS2D operation, overlay T‑Mobile coverage maps, and use U.S. county, national‑park, and census data to compute a spatial “SCS share” metric – the proportion of all measurements in a region that belong to Starlink.

Key empirical findings include: (1) a strong linear correlation between the cumulative number of DS2D‑capable satellites (400 → 710 over the study period) and the count of unique LTE cell identifiers (ECIs) observed, confirming that each satellite effectively creates a new “cell” visible to user equipment; (2) physical‑layer statistics showing a median RSRP of roughly –107 dBm, which is 24 dB lower than typical terrestrial LTE, and a median RSRQ about 3 dB higher, reflecting the narrow, high‑gain beams used by the satellites; (3) a SINR distribution centered near 5 dB, with only a small tail above 10 dB, consistent with the service’s current SMS‑only operation that limits transmit power and bandwidth.

To estimate the performance of the planned data service, the authors adopt a modified Shannon capacity model for LTE introduced in prior work. Using a bandwidth‑efficiency factor s = 0.57, fitting coefficients a = 0.9, b = 1.25, and a modulation ceiling m = 4.22 bps/Hz (corresponding to 256‑QAM), the spectral efficiency η is computed as η = min{0.51 · log₂(1 + SINR/1.25), 4.22}. Plugging the measured SINR distribution yields an average η of about 0.9 bps/Hz. With the current 5 MHz of PCS‑G band allocated for the beta, this translates to roughly 3 Mbps per beam in outdoor conditions. The authors argue that if Starlink secures additional spectrum (e.g., AWS‑4 and PCS‑H blocks) and expands the number of beams, the per‑beam bandwidth could rise to 20 MHz, pushing the theoretical peak to around 18 Mbps.

Regulatory and business analysis highlights that the FCC initially imposed strict out‑of‑band emission (OOBE) limits, later relaxed by 9 dB after industry push‑back. The authors note that further spectrum acquisitions and a larger constellation (potentially >1,000 DS2D‑capable satellites) would enable higher beam reuse and greater aggregate capacity, making the 18 Mbps target realistic. Spatial analysis of SCS share shows the service is most prevalent in national parks and sparsely populated counties where terrestrial coverage is weak; in those areas, Starlink accounts for over 30 % of all observed measurements, underscoring its role as a genuine supplemental coverage provider.

In conclusion, the paper delivers a data‑driven characterization of a live DS2D network, documenting its radio‑access parameters, the relationship between satellite deployment and observable cellular cells, and realistic performance projections for future data services. The work provides a valuable empirical foundation for researchers, network planners, and policymakers as they consider the integration of LEO‑based non‑terrestrial networks into the broader mobile ecosystem. Future research directions suggested include indoor and high‑mobility performance, multi‑beam spatial multiplexing, and alignment with upcoming 5G NR‑NTN standards.