Source-linked AI summary
Direct Satellite-to-Device Communications: From Cooperative Task Offloading to Non-Cooperative Access Monitoring
Sai Huang, Wanli Ni, Ke Lv, Pengcheng Zhang, Yurui Zheng, Menghan Zhang, Zihui Gong, Zhiyong Feng
TL;DR
DS2D must overcome dynamic satellite channels, limited satellite computing resources, and unauthorized access threats. The paper unifies cooperative task offloading with non-cooperative signal monitoring using channel-aware reinforcement learning and Transformer-based detection and AMC models. Its simulations report reduced latency, strong signal detection, and improved low-SNR modulation-classification accuracy.
Problem
DS2D deployment is challenged by severe Doppler and fast fading, constrained satellite computing resources, and unauthorized access that threatens spectrum security.
Method
The framework combines LCEADC channel estimation and Doppler compensation with D3QN association optimization, plus SAM-based detection and CNN-Transformer AMC for non-cooperative scenarios.
Results
The simulations report up to 225% latency reduction, 13.9% F1-score gain with false alarms reduced to 4.5% for LTE detection, and 70.44% low-SNR AMC accuracy versus 61.15% for TLDNN.
Takeaways & Limitations
A unified DS2D system can jointly support cooperative computing and satellite-based monitoring of unauthorized transmissions under challenging channel and SNR conditions.
Abstract
from arXiv · showhide
Direct satellite-to-device (DS2D) communication is emerging as a transformative paradigm for extending ubiquitous connectivity and edge computing capabilities to remote and underserved regions within 6G non-terrestrial networks. However, practical deployment faces dual critical challenges: i) dynamic satellite channel conditions (e.g., severe Doppler shifts, fast fading) and constrained satellite computing resources in cooperative scenarios; and ii) unauthorized satellite access introduces significant spectrum security threats in non-cooperative scenarios. To address these challenges, we propose a versatile DS2D system that supports cooperative task offloading and non-cooperative access monitoring. For cooperative DS2D communications, we integrate a channel estimation module with a dueling double deep Q-network (D3QN) to dynamically optimize task offloading strategy. For non-cooperative DS2D communications, we propose Transformer-based models to enable blind signal detection and automatic modulation classification (AMC). Simulation results show that: 1) The D3QN algorithm reduces average latency by up to 225\% compared to static association policies. 2) Our signal detection model achieves an average presence detection probability of 90.5\% for DS2D signals. 3) The proposed AMC algorithm achieves superior performance across different signal-to-noise ratios (SNRs), with a 9.4\% accuracy gain in low-SNR regimes compared to existing methods.
I. INTRODUCTION
DS2D addresses terrestrial coverage limitations but must manage difficult satellite channels, constrained computing resources, and unauthorized access. This work unifies cooperative task offloading with non-cooperative access monitoring.
- DS2D extends connectivity toward remote and underserved regions where terrestrial infrastructure is geographically limited, costly, or vulnerable to disasters.
- Dynamic satellite conditions, including weak links, severe Doppler shifts, fast fading, and intermittent visibility, complicate DS2D service delivery.
- Existing research largely addresses cooperative and non-cooperative DS2D challenges separately, motivating a unified framework.
- 225% latency reduction is achieved by a DRL algorithm combining channel estimation, adaptive Doppler compensation, and user-association decisions against static association policies.
- The non-cooperative pipeline combines SAM-based signal detection with a CNN-Transformer dual-stream AMC model, reaching over 88% detection probability and 91.6% high-SNR recognition accuracy.
II. DS2D COMMUNICATIONS: KEY TECHNOLOGIES AND APPLICATION SCENARIOS
The proposed DS2D architecture combines cooperative service delivery and non-cooperative spectrum monitoring across physical, network, and application layers.
- The physical layer includes LEO satellites with multibeam antennas, onboard computing units, and RF monitoring receivers.
- The network layer manages user association, channel estimation, task offloading, and unauthorized-transmission identification.
- The application layer supports services including disaster-area emergency communications, maritime connectivity, and aerial-platform support.
1) Satellite Beamforming and Beam Management:
DS2D beam management must sustain weak-signal access despite rapidly varying satellite channels, while multi-satellite coverage enables selection, handover, and cooperative access.
- Limited UE transmit power and antenna gain make link budget central to DS2D availability, motivating multibeam coverage and beamforming for array gain.
- Doppler from high relative LEO–UE velocity creates rapidly varying carrier-frequency offsets that can disrupt synchronization, random access, and uplink transmission.
- The system uses lightweight closed-loop Doppler mitigation to reduce the resulting inter-carrier interference in multicarrier links.
- Overlapping coverage from multiple LEO satellites supports intelligent satellite selection, seamless handover, and cooperative access.
4) Satellite-Terrestrial Cooperative Computing:
Satellite-terrestrial cooperative computing lets DS2D serve diverse environments while balancing connectivity benefits against security risks from unauthorized access.
- DS2D enables resource-constrained devices to offload tasks to LEO satellites, terrestrial servers, or cloud infrastructure through satellite backhaul.
- Cooperative satellites support connectivity in urban, highway, remote, disaster-response, airborne, and maritime scenarios.
- Non-cooperative satellites and unauthorized access channels introduce security risks alongside these legitimate application scenarios.
- DS2D provides immediate wide-area communication where terrestrial base stations are prohibitively expensive, infeasible, damaged, or unavailable.
- Direct satellite links support sustained maritime connectivity and time-critical services for aerial platforms beyond terrestrial coverage.
III. COMPUTING TASK OFFLOADING VIA COOPERATIVE DS2D COMMUNICATIONS
Cooperative DS2D task offloading addresses constrained satellite resources and dynamic channel conditions by jointly modeling access-point selection, channel estimation, and latency-aware deep reinforcement learning.
- LEO satellites provide computing services for remote or underserved devices but have limited onboard computational resources and energy budgets.
- Each UE selects either a LEO satellite or terrestrial BS for task execution based on channel conditions, node loads, and service requirements.The selected node executes the task, returns the result, and supplies latency feedback for the next decision epoch.
- The binary association factor assigns each user exactly one service node per time slot, preventing redundant resource consumption.
- LCEADC-D3QN combines lightweight channel estimation with adaptive Doppler compensation and a dueling double deep Q-network to minimize average end-to-end latency.
C. Simulation Results
The cooperative offloading results show increasing latency with system load, while non-cooperative monitoring targets unauthorized DS2D access under challenging noise and channel conditions.
- Average latency grows monotonically with the number of UEs under all evaluated association schemes.The trend reflects increasing system load.
- LCEADC-D3QN reduces average latency by up to 225% relative to all-BS, all-satellite, and random association baselines.Dynamic access-point selection uses real-time channel conditions, node loads, and service demands.
- All-satellite association incurs high propagation and processing latency because of long satellite-ground paths and limited onboard computing capacity.
- All-BS association benefits from shorter propagation delays but becomes suboptimal as overload creates queuing congestion and processing bottlenecks.
- Unauthorized DS2D access creates coexistence between legitimate signals and interference from non-cooperative satellite systems, complicating spectrum monitoring.
- Energy detection is noise-sensitive at low SNRs, whereas cyclostationary detection requires long observations and degrades under Doppler spread and time-varying fading.
1) System Model:
The proposed signal-detection system converts received DS2D signals into time-frequency spectrograms and performs automatic, category-aware semantic segmentation with an adapted SAM architecture.
- Hybrid radio-frequency signals are transformed into two-dimensional time-frequency spectrograms using the short-time Fourier transform.
- Signal detection is formulated as multi-class semantic segmentation, with pixel-level masks labeling spectrogram regions by category.
- The SAM-based model freezes the pre-trained ViT backbone and adds lightweight adapters for spectrogram-domain adaptation.
- Automatic prompt generation removes SAM’s dependence on user-provided points or boxes.
- A redesigned mask decoder supports end-to-end multi-class pixel-level segmentation through learnable category representations and improved feature upsampling.
3) Simulation Results:
Non-cooperative DS2D monitoring evaluates detection on LTE and 5G NR signals and AMC under low-SNR satellite-access conditions, with proposed models outperforming baseline approaches.
- The detection method improves LTE IoU by 21.7% and F1-score by 13.9% compared with Baseline 2.
- For LTE signals, the proposed detector reduces false alarm rate from 9.2% to 4.5%.
- The proposed detector consistently outperforms baselines across IoU, F1-score, and false alarm rate for 5G NR signals.All methods perform better on 5G NR than LTE, reflecting richer time-frequency structure and greater feature separability.
- Onboard AMC faces extremely low SNRs because monitoring satellites primarily intercept transmissions through antenna sidelobes.
1) System Model:
The monitoring system addresses unauthorized satellite links under inherently low-SNR conditions and uses adaptive denoising with dual-stream attention for modulation recognition. Experiments evaluate ten modulation types and report robust, low-latency performance.
- The monitoring satellite receives unauthorized-link signals mainly through antenna sidelobe leakage, producing inherently low SNRs.
- AVDSDN combines adaptive VMD denoising, dual-stream feature extraction, and cross-stream attention fusion for modulation recognition.
- Ten modulation types are evaluated, including PSK, PAM, and QAM formats.The set includes BPSK, QPSK, 8PSK, 16PSK, 4PAM, 8PAM, 32QAM, 64QAM, 128QAM, and 256QAM.
- 70.44% average recognition accuracy is achieved from -20 dB to 0 dB SNR, exceeding TLDNN’s 61.15%.
- Approximately 5-6 ms end-to-end processing latency per signal segment is achieved with INT8 quantization and GPU-accelerated VMD denoising.The passage reports negligible accuracy loss from quantization and identifies the implementation as practical for resource-constrained satellite platforms.
A. Integrated Satellite Communication, Sensing, Positioning, and Navigation
The paper outlines future DS2D directions spanning integrated communication and sensing, distributed learning, and reconfigurable satellite networking. These directions target broader functionality, privacy preservation, reduced backhaul, and lower lifecycle costs.
- A. Integrated Satellite Communication, Sensing, Positioning, and Navigation: Unified waveforms may support data transmission, environmental sensing, high-accuracy positioning, and navigation in remote or GPS-denied regions.The proposed integration leverages LEO satellites’ global coverage and mobility for situational awareness applications.
- Federated learning can preserve privacy and reduce backhaul load by exchanging model parameters or knowledge over satellite links.Future DS2D FL research must address intermittent connectivity and heterogeneous conditions.
- In-orbit software reconfiguration can update communication protocols, modulation schemes, and signal processing algorithms without replacing satellite hardware.
- Network function virtualization could dynamically instantiate emergency alert broadcasting, IoT data relay, and spectrum monitoring services on shared satellite platforms.The passage calls for robust over-the-air updates, fault-tolerant execution, and standardized cross-vendor interfaces.
D. Lightweight Security Protocols for DS2D Communications
DS2D security is constrained by broadcast transmissions and limited handset resources, motivating lightweight cryptography. The paper also identifies unresolved analytical gaps in rapidly varying channels and satellite edge-computing trade-offs.
- D. Lightweight Security Protocols for DS2D Communications: Broadcast satellite transmissions and limited handset resources make conventional cryptographic protocols costly in latency and energy.
- D. Lightweight Security Protocols for DS2D Communications: Future security work should develop lightweight primitives, including post-quantum secure symmetric ciphers and compact digital signatures.
- Information-theoretic limits remain poorly understood under LEO Doppler spreads, intermittent connectivity, and non-stationary fading.
- Analytical frameworks are lacking for latency, reliability, and energy-efficiency trade-offs in satellite edge-computing systems.
- The proposed framework combines cooperative task offloading with non-cooperative access monitoring using D3QN and Transformer-based models.The conclusion reports up to 225% latency reduction and 90.5% detection probability for unauthorized access attempts.
- Unified DS2D deployment requires advanced resource orchestration, cross-domain learning, standardized interfaces, and collaboration among industry, academia, and regulators.