Direct Satellite-to-Device (DS2D)

• Direct Satellite-to-Device (DS2D) is a wireless architecture enabling standard user devices to connect directly with LEO satellite networks without relying on terrestrial infrastructure.
• It employs virtualized satellite resource pools, predictive beamforming, and advanced interference management to optimize connectivity and support robust, high-speed data services.
• DS2D addresses challenges like rapid handovers, spectrum regulation, and economic scalability while enhancing IoT connectivity and non-orthogonal access strategies.

Direct Satellite-to-Device (DS2D) describes a class of wireless architectures and protocols enabling unmodified user equipment—namely, standard handsets and IoT devices—to communicate directly with non-terrestrial networks, most frequently Low Earth Orbit (LEO) satellite constellations, without reliance on terrestrial cellular infrastructure. By virtualizing satellite assets into dynamic resource pools and integrating advanced interference management, predictive beamforming, and cross-band uplink strategies, DS2D targets seamless global service extension, robust coverage, and continuity of broadband-grade mobile access in environments where cellular networks are unavailable or impractical. The field encompasses foundational work in multi-constellation management, regulatory spectrum frameworks, hardware architectures, and algorithmic innovations for link adaptation, handover, and uplink robustness.

1. System Architecture and Constellation Management

Current DS2D networks are built upon multi-orbit satellite constellations, each satellite featuring phased-array antennas capable of highly directional beamforming and multi-beam operation, and integrating sophisticated onboard radio access network (RAN) and, increasingly, 5G core functions (Wang et al., 1 Jul 2025, Elgueta, 13 Jul 2025, Aijaz et al., 6 Oct 2025). Modern system models treat the entire fleet—across operator boundaries—as a shared infrastructure pool, 𝒮, with sub-constellations (SCs) formed on demand for spatial service regions Ωₖ according to user density and traffic requirements (Wang et al., 1 Jul 2025). Each satellite sᵢ in 𝒮 is parameterized by orbital altitude hᵢ, antenna gain Gᵢ, and connection capacity Cᵢ (number of simultaneous links).

Service orchestration in advanced DS2D follows the “Constellation as a Service” (CaaS) paradigm: given ground UE demands and a partition of the globe into regions {Ω₁,…,Ω_R}, the system dynamically forms S_r⊆𝒮 for each region r, solving

$$\max_{S_r \subseteq \mathcal{S}} R_r(S_r) - \lambda H_r(S_r)$$

with $$R_r(S_r)=\sum_{d\in\mathcal{D} \cap \Omega_r} \log_2(1+\mathrm{SINR}_d(S_r))$$ (aggregate service rate) and $H_r(S_r)$ the expected handover count, while enforcing per-satellite capacity, instantaneous coverage, and cardinality constraints (Wang et al., 1 Jul 2025).

The space segment incorporates electronically steered phased arrays (64×64 to 256×256 elements) generating 1,000+ simultaneous beams, with per-beam EIRP of 40–60 dBm, and <1 ms re-steering agility for mobile users (Elgueta, 13 Jul 2025). Core network entities—UPF, AMF/SMF, AUSF, UDM—are increasingly shifted to orbit for reduced RTT and seamless cross-beam/session management.

2. Key Radio and Link Layer Innovations

DS2D introduces foundational changes in both downlink and uplink:

Downlink Beamforming and Predictive Management. Predictive beamforming employs Generative AI (e.g., Transformer-based sequence models) trained on historical multi-constellation Channel State Information (CSI) traces to forecast future CSI and recommend optimal beamforming vectors wᵢ,ₖ. The optimal per-device, per-satellite beamformer solves

$$w^{\ast}_{i,d} = \arg \max_{w} E[\mathrm{SINR}_{i,d}(w)] - \gamma \|w-w^{\text{pred}}_{i,d}\|^2_2$$

where $w^{\text{pred}}_{i,d}$ is the GenAI prediction. This enables satellites to adapt beam patterns with high reactivity to atmospheric changes and orbital dynamics (Wang et al., 1 Jul 2025).

Uplink Enhancement: The elevation-aware Supplementary Uplink (SUL) framework exploits dual-band operation, assigning standard PUL (e.g., 30 GHz/Ka) or SUL (e.g., 1.6 GHz/L) access depending on elevation angle θ and link margin estimation. DS2D UEs execute a carrier selection algorithm based on predicted link margins $M_f(\theta)$ per carrier, with hysteresis Δₕ to avoid frequent carrier toggling (Shrestha et al., 23 Feb 2026). At low-elevation (<30°–40°), SUL extends coverage to θ≈10°, preserving link reliability even at large slant ranges (up to ≈1,930 km for h=600 km, θ=10°).

Multiple Access and Non-Orthogonal Techniques: DS2D uplinks leverage both NOMA (power/frequency-domain) and aggressive non-orthogonal transmissions to maximize spectral efficiency despite the absence of traditional spatial multiplexing in LoS channels. In the NL-COMM-Sat framework, up to K>2 UEs transmit over the same frequency-time resource, with a single-antenna receiver using non-linear multi-user detection (sphere search, turbo MAP) to decode. At SNR=4 dB, four-user NL-COMM-Sat achieves ~1.6 b/s/Hz spectral efficiency, doubling traditional OMA/NOMA (Nikitopoulos et al., 27 Apr 2026).

3. Handover, Interference Management, and Mobility

DS2D networks contend with high satellite mobility (dwell <5 min per cell), frequent handovers, and dense beam overlaps. CaaS and related frameworks address this through pre-configured, cost-optimized handover paths (modeled as shortest paths in region-specific Directed Acyclic Graphs G_r), minimizing signaling overhead and handover frequency (Wang et al., 1 Jul 2025). An example: average HO frequency drops >50%, signaling overhead by ≈45%, and per-UE coverage remains >99% versus ≈95% for static or non-predictive schemes.

Interference-aware handover schemes further adapt the elevation angle threshold (EAT) dynamically to balance satellite visibility and co-channel interference. Lowering θₜₕ increases visibility and thus handover intervals, but at the expense of increased CCI; raising θₜₕ reduces CCI, but increases HO rates and possible coverage gaps. Numerical optimization (via grid search over θₜₕ) yields up to 40% reduction in packet loss and 3–6 dB average SINR gain (Li et al., 28 Feb 2026).

Robust online detection of terrestrial interference is enabled by hypothesis-testing frameworks (Rayleigh-vs-Rice), leveraging CUSUM or GLR-based detectors for known/unknown interference direction, achieving millisecond-grade detection with sub-percent FAR (Liu et al., 15 Jun 2025).

4. Spectrum Allocation, Regulatory Context, and Coexistence

Spectrum regulation for DS2D is shaped by ITU/WRC and FCC policies that aim to maximize flexible access while enforcing coexistence and passive-service protection (Rappaport et al., 23 Jun 2025). Key bands include:

• L-/S-band (~1–4 GHz): globally harmonized, but suitable only for low-rate DS2D (SMS, IoT).
• PCS G-block (1.91–1.95 GHz): currently used for commercial DS2D (e.g., Starlink Direct-to-Cell).
• Upper-midband (7–24 GHz, esp. 7.125–13.25 GHz): optimal tradeoff between available bandwidth, path loss, and manageable atmospheric absorption.
• Ka and Q/V bands (18–24 GHz): support hundreds of MHz to multi-GHz channels, with higher capacity but increased atmospheric and rain attenuation.

Interference to terrestrial and passive (EESS/RAS) services is limited by ITU/FCC emission masks and EPFD criteria, enforced using directional beams, dynamic power control, and spatial/temporal orthogonality in frequency reuse.

A summary table of relevant spectrum bands is provided below:

Frequency Band	DS2D Use Case	Bandwidth
L/S-band (1–4 GHz)	IoT/safety/low-rate	≤40 MHz
PCS G-block	SMS, basic DS2D	2×5 MHz paired
7–13 GHz (FR3)	Broadband DS2D	>1 GHz contiguous
18–24 GHz	High-capacity DS2D	2–4 GHz per band

5. Economic Feasibility and Scaling

Recent techno-economic analysis frameworks demonstrate that DS2D can achieve per-user costs and returns on investment (ROI) comparable to terrestrial mobile networks when properly architected (Aijaz et al., 6 Oct 2025). Regenerative and 3D Open RAN architectures reduce operational and capital expenditures by ~16–76% compared to legacy bent-pipe designs. High-beam-count payloads, dense orbital shells, and ground-core software disaggregation (Open RAN) are essential for achieving positive ROI within 5 years, and per-user costs below $12/month—on par with mass-market terrestrial services.

Key economic levers include:

• Satellite count and per-node beamforming capability
• Launch cost reductions with heavy-lift vehicles
• Bandwidth per user/region divided by active population density
• OPEX–CAPEX tradeoffs for space-based vs. ground-based core functions

6. IoT and Non-Orthogonal Access in DS2D

IoT applications in DS2D are addressed using beacon-assisted synchronization (e.g., BU-LoRaWAN) and non-orthogonal multiple access. BU-LoRaWAN leverages LoRaWAN Class B beacons for transmission window alignment and queue-based collision avoidance, achieving PDR increases (up to +20%) and collision reductions (≈25–35%) versus standard LoRaWAN (Mojamed, 2024). Uplink NOMA approaches with power-domain separation and successive interference cancellation (SIC) enable dense IoT operation; controlled transmit power (CTP) strategies outperform fixed power and classical ALOHA by 29–101% in throughput and >35% in energy efficiency as device count scales (Tondo et al., 2024).

NL-COMM-Sat further advances uplink capacity by exploiting aggressive user overloading and non-linear ML detection, providing up to 2× higher spectral efficiency than OMA/NOMA in single-antenna regimes with realistic Doppler (Nikitopoulos et al., 27 Apr 2026).

7. Open Research Challenges and Future Directions

Future DS2D systems are expected to incorporate multi-constellation orchestration beyond LEO (MEO/GEO fusion), fine-grained spectrum sharing with terrestrial networks, integration of 6G enablers such as digital twins and reconfigurable intelligent surfaces, and AI-driven resource scheduling (Wang et al., 1 Jul 2025). Link budget closure for handhelds, Doppler and phase synchronization in multi-satellite MIMO (Bakhsh et al., 2024), robust CSI prediction under satellite-group mobility, and real-time cross-domain spectrum scheduling remain active research frontiers.

A 15-year roadmap from fall-back-grade coverage to urban-grade, orbital-native overlays has been articulated, targeting 1,024+ beam satellites, >5 b/s/Hz spectral efficiency, per-user 50–100 Mbps rates, and full decoupling from terrestrial cores by 2040 (Elgueta, 13 Jul 2025).

---

References:

(Wang et al., 1 Jul 2025, Shrestha et al., 23 Feb 2026, Li et al., 28 Feb 2026, Liu et al., 15 Jun 2025, Elgueta, 13 Jul 2025, Aijaz et al., 6 Oct 2025, Bakhsh et al., 2024, Mojamed, 2024, Tondo et al., 2024, Nikitopoulos et al., 27 Apr 2026, Rappaport et al., 23 Jun 2025)