Scheduling Policy

The objective of maximizing the total number of bits successfully transmitted across the cell per unit time. Max-Rate or Max C/I schedulers achieve this by always allocating resources to the user with the highest instantaneous Signal-to-Interference-plus-Noise Ratio (SINR).

• Mechanism: Exploits multi-user diversity by selecting users at channel peaks.
• Trade-off: Maximizes cell capacity but leads to extreme starvation for cell-edge users with poor channel conditions.
• Use Case: Best-effort traffic with no minimum bitrate guarantees.

Mechanisms that ensure equitable access to radio resources, preventing cell-edge starvation. Proportional Fair (PF) scheduling balances throughput and fairness by allocating to the user whose current channel quality is highest relative to their own historical average.

• Metric: Often measured using Jain's Fairness Index.
• Mechanism: PF scheduler uses the ratio R_instantaneous / R_average as the scheduling metric.
• Result: Provides a balance between aggregate cell throughput and per-user service consistency, giving priority to users emerging from deep fades.

The ability of a scheduler to prioritize traffic based on strict delay budgets, critical for Ultra-Reliable Low-Latency Communication (URLLC). Policies must consider Packet Delay Budget (PDB) and head-of-line packet delay.

• Mechanism: Delay-based or Largest Weighted Delay First (LWDF) schedulers assign higher weights to packets approaching their deadline.
• Challenge: Requires tight integration with the Quality of Service (QoS) Class Identifier (QCI) framework in 5G.
• Application: Essential for industrial automation, autonomous driving, and real-time control loops.

A fundamental distinction in scheduling logic. Channel-aware schedulers rely on real-time Channel State Information (CSI) feedback, such as Channel Quality Indicator (CQI) reports, to make opportunistic decisions.

• Channel-Aware: Exploits multi-user diversity (e.g., Max-Rate, Proportional Fair). Requires accurate and timely CSI feedback.
• Channel-Blind: Allocates resources cyclically without CSI, such as Round Robin (RR). Maximizes simplicity and fairness in time allocation but ignores instantaneous channel quality.
• Hybrid: Deep Reinforcement Learning (DRL) agents can learn to make decisions based on partial or delayed CSI, bridging the gap between the two paradigms.

The integration of Buffer Status Reports (BSR) from user equipment into the scheduling decision. A scheduler must know if a user has data queued to avoid wasting resource blocks on an empty buffer.

• Mechanism: Combines channel quality with queue length to prevent buffer overflow and reduce packet drops.
• Advanced Logic: Queue-aware schedulers prioritize users with larger backlogs to maintain stability, a concept formalized in Lyapunov drift-plus-penalty optimization.
• DRL Integration: The state space of a DRL agent often includes both CSI and BSR to make holistic resource allocation decisions.

The scheduling policy's impact on the power consumption of both the base station and user equipment. Modern schedulers incorporate energy metrics to meet sustainability targets without violating QoS.

• Mechanism: Energy-aware schedulers may deliberately delay non-latency-sensitive traffic to batch transmissions, allowing power amplifiers to operate in more efficient regions.
• Metric: Measured in bits per Joule. The scheduler can minimize transmissions with high Block Error Rate (BLER) that require retransmissions.
• DRL Application: A reward function can be shaped to penalize power consumption, teaching an agent to find the Pareto-optimal frontier between throughput and energy use.