Time-Scale-Adaptable Spectrum Sharing

• Time-scale-adaptable spectrum sharing is a framework that manages spectrum resources over nested intervals from sub-millisecond to minutes by integrating dynamic sensing, inference, and control.
• It employs modular RAN designs like O-RAN with non-real-time, near-real-time, and real-time controllers to optimize resource allocation and interference management using AI and graph-theoretic methods.
• The framework enhances spectrum utilization and regulatory compliance by balancing trade-offs between responsiveness, complexity, and signaling overhead through hierarchical control.

A time-scale-adaptable spectrum sharing framework is a network control paradigm that coordinates spectrum access, allocation, and resource management over multiple nested time scales—ranging from sub-millisecond to minutes or longer—by dynamically sensing, inferring, and reacting to spectrum usage across regulatory, architectural, and protocol layers. These frameworks are realized in advanced radio access network (RAN) designs, especially within disaggregated architectures like O-RAN, and are implemented via integrated algorithmic, sensing, policy, and AI-driven modules able to exploit traffic, interference, and incumbent-user variability. By separating control functions and resource assignment policies across physical, MAC, and network layers and employing an explicit notion of adaptable timescales, such frameworks enable robust coexistence with prioritized or heterogeneous users, maximize spectrum utilization, and provide explicit trade-offs between responsiveness, complexity, and signaling overhead.

1. Multi-Timescale Architectural Paradigms

Time-scale-adaptable frameworks instantiate modular control architectures that map principal RAN functions (e.g., policy generation, spectrum optimization, real-time scheduling) onto distinct time layers, each with specific control granularity and bounded reaction intervals. The canonical O-RAN disaggregation employs:

• Non-Real-Time RIC (rApp control, non-RT, ΔT ≳ 1–15 min): Executes global traffic prediction, static or semi-static policy selection, and long-term resource reservation (e.g., numerology, frequency reuse, fairness scheme) (Giannopoulos et al., 20 Jan 2026, Rasti et al., 19 Feb 2025).
• Near-Real-Time RIC (xApp control, near-RT, 1 ms ≤ T ≤ 1 s): Performs per-slot or per-second interference sensing, PRB conflict or graph-coloring-based assignment, and short-term optimization exploiting recent telemetry and mobility statistics (Giannopoulos et al., 20 Jan 2026, Baldesi et al., 2022).
• Real-Time (RT) Control (dApp/dedicated logic, T ≈ 1–10 ms): Enacts sub-frame or sub-millisecond scheduling, spectrum sensing, or waveform adaptation using direct access to I/Q streams, physical control channels, or user-plane measurements (Gangula et al., 2024, Baldesi et al., 2022).
• Marketplace or Broker Layer: In inter-operator scenarios, an authorized broker manages trading, settlement, and dispatching of spectrum via a multi-timescale API (Rasti et al., 19 Feb 2025).

This layered approach is exemplified in state-of-the-art O-RAN deployments, hybrid satellite-terrestrial coordination, LSA dynamic management, and opportunistic MAC/controller decomposition (Giannopoulos et al., 20 Jan 2026, Wang et al., 26 Jan 2026, Ponomarenko-Timofeev et al., 2015, Teng et al., 2016).

2. Spectrum Sensing, Inference, and Adaptation Across Time Scales

Adaptive spectrum sharing requires continuous (or periodic) sensing and inference of spectral occupancy, typically by:

• Energy Detection or ML-based Techniques: Leveraging PAPR, cyclostationary, or deep learning classifiers directly on I/Q samples to infer incumbent or co-channel presence, with configurable sensing window (T_sense) and detection interval (T_decide), e.g., sub-1 ms to hundreds of ms (Gangula et al., 2024, Baldesi et al., 2022).
• Graph-Theoretic Conflict Modeling: Building and updating user-interference conflict graphs, PRB compatibility, and satisfaction constraints on the order of 1 ms–1 s, crucial for distributed or cluster-wide resource assignment (Giannopoulos et al., 20 Jan 2026).
• Waveform and Beamspace Adaptation: Dynamically recomputing space-time waveform codes and beamforming vectors based on short-term interference autocorrelation, with reassignment interval adapted to interference variability (T_update) (Nouri et al., 8 Apr 2025).
• Channel-State Aggregation: In satellite-terrestrial frameworks, employing only statistical CSI on coarse intervals (e.g., 10 s), with joint scheduling and power control remaining constant over such times (Wang et al., 26 Jan 2026).

The reaction and sensing intervals are selectable according to target protection levels, coexistence policy demands, or the agility of incumbents.

3. Optimization and Resource Allocation Mechanisms

Allocation mechanisms must operate efficiently and stably across distinct time domains:

• Centralized Slow-Timescale Optimization: Long-term partitioning (minutes–hours) of spectrum or power masks for entire cells or clusters, e.g., convex utility maximization or assignment, often with fairness or prediction-driven policy (Giannopoulos et al., 20 Jan 2026, Ponomarenko-Timofeev et al., 2015, Teng et al., 2016, Rasti et al., 19 Feb 2025).
• Fast-Timescale Distributed Scheduling: Per-slot (ms-level) resource allocation, e.g., opportunistic MAC, MPF scheduling, or PRB barring, re-optimized reactively using current interference, queue, or mobility statistics (Gangula et al., 2024, Giannopoulos et al., 20 Jan 2026, Teng et al., 2016).
• Joint Mixed Integer Programming (MIP): Hybrid satellite-terrestrial cases require per-T (seconds–tens of seconds) joint link assignment and power allocation under statistical CSI, solved via hierarchical clustering, assignment, and convex approximation (Wang et al., 26 Jan 2026).
• Marketplace Mechanisms: AI-driven demand estimation, spectrum valuation, and price setting are embedded into centralized or decentralized auction-based brokers, computed at multiple time layers, subject to fairness, reservation, or cost constraints (Rasti et al., 19 Feb 2025).

4. Performance Metrics and Experimental Results

Time-scale-adaptable frameworks have been validated via over-the-air, emulator, and large-scale simulation using metrics such as:

• Detection Latency: Sub-ms to 1 ms between sensing and reconfiguration (Gangula et al., 2024, Baldesi et al., 2022).
• Throughput and Capacity: UE throughputs decrease proportionally to PRB occupation by incumbents with rapid recovery post-vacancy; network sum-rate gains of >15–23% vs. non-adaptive baselines in multi-layer sharing (Gangula et al., 2024, Wang et al., 26 Jan 2026).
• Fairness (Jain’s Index): Achieves ≥85–97% fairness in service share under advanced graph-coloring/MPF scheduling (Giannopoulos et al., 20 Jan 2026).
• Interference Constraint Violation Rate: Dynamic LSA and limit-power policies consistently enforce interference caps to protect prioritised incumbents, outperforming ignore/shutdown baselines (Ponomarenko-Timofeev et al., 2015).
• Sensing/Signaling Overhead: Overheads are kept ≤0.35% via sparsified sensing; complex clustering and scheduling heuristics maintain practical complexity for typical deployment sizes (Gangula et al., 2024, Wang et al., 26 Jan 2026).

5. Design Trade-offs and Practical Considerations

Key practical design aspects include:

• Control Interval Selection: Tuning the interval for sensing, allocation, and control balances reactivity (smaller T) against signaling, estimation error, and computational load. For slow-varying channels or interference, longer T reduces overhead without significant performance loss; for dynamic traffic environments, reduced T yields closer-to-optimal coordination (Wang et al., 26 Jan 2026, Rasti et al., 19 Feb 2025).
• Interface Compliance and Modularity: Explicit separation between non-RT/near-RT/RT functions (O-RAN A1/E2/D) provides pluggable and updatable module support—machine-learning, meta-policy, and real-time reaction logic can be rapidly deployed or adjusted (Giannopoulos et al., 20 Jan 2026, Baldesi et al., 2022, Rasti et al., 19 Feb 2025).
• Regulatory/Standardization Alignment: LSA and O-RAN frameworks provide formalized hooks for regulatory protection and operator guarantees, e.g., via low-latency triggering, reserved static slices, or prescribed auction rules (Ponomarenko-Timofeev et al., 2015, Rasti et al., 19 Feb 2025).
• Scalability and Robustness: Hierarchical control possesses inherent scalability, with each layer handling only its necessary granularity; broker and policy center modules manage spectrum trading seamlessly even across heterogeneous RAN domains (Rasti et al., 19 Feb 2025).

6. AI-Driven and Data-Driven Enhancements

Recent advances employ AI for enhanced forecasting, inference, and policy learning:

• Forecasting and Policy Adaptation: Integration of discriminative (LSTM, Transformer), generative (GAN, VAE, Diffusion), and reinforcement learning models enables anticipatory spectrum estimation, rapid auction clearing, and adaptive scheduling at multiple time scales, significantly reducing allocation lag and increasing efficiency (Rasti et al., 19 Feb 2025).
• Meta-Control: Non-RT controllers can monitor operational confidence and dynamically retune key control parameters (e.g., T, threshold α) to maintain target accuracy and system utility (Baldesi et al., 2022).
• Collaborative Learning and Marketplace Dynamics: Federated and distributed learning facilitates private, cross-operator improvement of models; pricing mechanisms adapt over multiple time and spatial hierarchies to optimize for market efficiency and operator objectives (Rasti et al., 19 Feb 2025).

7. Open Challenges and Future Research

Key avenues for research and development include:

• Ultra-Low-Latency Signaling: Achieving sub-ms end-to-end actuation across distributed RAN and regulatory entities (Ponomarenko-Timofeev et al., 2015).
• Granularity Optimization: Learning optimal time-spatial sharing granularity dynamically, balancing allocation fidelity against signaling cost (Rasti et al., 19 Feb 2025).
• Advanced Incumbent Detection and Evasion: ML-based classifiers for highly dynamic or low-SNR environments; fine-grained time-frequency avoidance; and integration with external incumbent databases (Gangula et al., 2024, Baldesi et al., 2022).
• Decentralized Ledgers and Security: Exploring blockchain or DLT for spectrum transaction security, transparency, and smart-contract enforcement (Rasti et al., 19 Feb 2025).
• Market Mechanism Robustness: Designing long-term incentive-compatible auctions, efficient spectrum supply and demand clearing over highly variable conditions (Rasti et al., 19 Feb 2025).

Time-scale-adaptable spectrum sharing frameworks thus mark a transition towards fine-grained, responsive, and optimally coordinated resource management in heterogeneous wireless networks, leveraging modular architectures, adaptive inference, and learning-driven optimization to maximize both efficiency and protection under complex coexistence scenarios (Gangula et al., 2024, Giannopoulos et al., 20 Jan 2026, Nouri et al., 8 Apr 2025, Ponomarenko-Timofeev et al., 2015, Baldesi et al., 2022, Wang et al., 26 Jan 2026, Teng et al., 2016, Rasti et al., 19 Feb 2025).