Communication Scheduling for Throughput-Fairness Balance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional MAC layer-based scheduling methods in wireless communication systems fail to achieve a balanced tradeoff between throughput and fairness, leading to suboptimal performance.
Innovation Solution
Implement a scheduling method using deep reinforcement learning (DRL) to determine scheduling policies based on system status information, leveraging a deep neural network trained on historical data to optimize resource allocation, power control, and user pairing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional MAC layer-based scheduling algorithms (e.g., proportional fair algorithm) are used to determine scheduling solutions, then the system can implement a basic tradeoff between throughput and fairness, but the communication system performance remains low due to system complexity
Solution Approach 1:
The patent replaces conventional deterministic scheduling algorithms with a deep reinforcement learning model. The DRL agent learns optimal scheduling policies through interaction with the communication environment, substituting traditional algorithmic approaches with a data-driven intelligent system that can adapt to complex system conditions and achieve better performance.
Solution Approach 2:
The patent transforms the scheduling problem from a deterministic parameter optimization task into a reinforcement learning task with state-space and action-space definitions. By changing the fundamental approach from algorithmic parameter tuning to learned policy optimization, the system achieves superior throughput-fairness tradeoff despite the added complexity of training the DRL model.
2Productivity
If deep reinforcement learning is implemented to optimize scheduling policies, then throughput and fairness balance is improved, but computational complexity and training requirements increase
Solution Approach 1:
The patent performs preliminary training of the deep reinforcement learning model offline using historical system data and simulations. This preliminary action allows the model to learn optimal scheduling policies beforehand, so that during actual operation, the system can quickly apply pre-learned policies without requiring real-time complex computations, thus improving system efficiency while managing computational complexity.
Solution Approach 2:
The patent implements a dynamic scheduling approach where the DRL model adapts to changing communication conditions by continuously learning from system feedback. The model dynamically adjusts scheduling decisions based on current system state, user channel conditions, and historical performance, enabling the system to optimize throughput and fairness in real-time while maintaining manageable computational requirements through efficient model architecture.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
According to a scheduling method and apparatus in a communication system, and a storage medium that are provided in embodiments of this application, a communication device obtains system status information, where the system status information includes network status information; obtains a scheduling policy based on the system status information and a deep neural network; and performs communication according to the scheduling policy. The deep neural network is obtained through training based on historical system status information, and the historical system status information includes system status information in all scheduling periods before a current scheduling period. Therefore, the scheduling policy obtained based on the deep neural network can meet a balancing requirement of throughput and fairness and solves a problem of low performance of an existing communication system.