System and method for joint power and resource allocation using reinforcement learning
Inventors
ELSAYED, Medhat • EROL-KANTARCI, Melike
Assignees
Publication Number
US-11678272-B2
Publication Date
2023-06-13
Expiration Date
2040-10-30
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Systems and methods for joint power and resource allocation on a shared 5G channel. The method selects one of a group of grouped actions and implements this selected group of actions. The effects of the actions on the environment and/or the users are then assessed. Based on the result, a reward is allocated for the system. Multiple iterations are then executed with a view to maximizing the reward. Each of the grouped actions comprises joint power and resource allocation actions.
Core Innovation
The invention presents systems and methods for joint power and resource allocation on a shared 5G channel. It employs a method where one of a group of grouped actions is selected and executed, with each action group including joint power and resource allocation actions. The effects of these actions on the environment or users are assessed and a reward is allocated based on the assessment. This process iterates multiple times, aiming to maximize the reward, thereby optimizing resource allocation among different service categories sharing the channel.
The problem being addressed arises from the coexistence of heterogeneous service categories in 5G networks, notably Ultra-Reliable Low-Latency Communications (URLLC) and enhanced Mobile Broadband (eMBB). These categories have conflicting quality of service requirements leading to trade-offs in resource allocation, such as balancing latency, reliability, and throughput. Previous approaches have limitations in robustly balancing these key performance indicators, necessitating more efficient resource management methods.
Claims Coverage
The claims include one independent method claim and one independent system claim, each comprising inventive features focused on joint power and resource block allocation using software agents in a shared wireless channel with different service categories.
Method of joint power and resource block allocation with decentralized multi-agent selection
A method performed in a gNodeB using multiple software agents that select, without cooperation or communication of selections between agents, one action group from a predetermined group of actions for users of different service categories. Each action group includes actions relating to power allocations and/or resource block allocations. The method includes executing the selected action, assessing its effect on measurable metrics (e.g., latency, reliability, throughput), determining a reward, and iteratively repeating these steps to maximize the reward.
System for resource management with link adaptation and scheduler modules implementing decentralized reinforcement learning
A system comprising a processor with a link adaptation module allocating resources and determining measurable metrics for multiple users with different service categories, and a scheduler module that uses multiple software agents to select and execute action groups involving power and/or resource block allocations. The agents operate without cooperation, selecting from predetermined action groups, assessing effects on measurable metrics, determining rewards, and iteratively maximizing these rewards over multiple scheduling intervals.
The independent claims focus on decentralized multi-agent methods and systems for joint power and resource block allocation in a shared 5G channel environment, emphasizing iterative reward-maximizing selection of grouped actions without inter-agent cooperation to manage heterogeneous service categories.
Stated Advantages
Improves latency and reliability of URLLC users in 5G networks by addressing transmission and queuing delays through a joint power and resource allocation method.
Achieves higher throughput for eMBB users while balancing the quality of service demands between URLLC and eMBB users.
Reduces packet drop rates compared to baseline algorithms.
Enables decentralized reinforcement learning that does not require inter-agent communication, reducing overhead and complexity.
Documented Applications
Managing resource allocation on shared 5G channels among users classified under different service categories such as URLLC and eMBB.
Optimizing scheduling and resource distribution in 5G New Radio (5G-NR) networks to balance latency, reliability, and throughput requirements.
Implementing reinforcement learning (specifically multi-agent Q-learning) in general eNodeBs (gNBs) for dynamic power and resource block allocation.
Use in systems requiring joint power and resource block allocation in wireless broadband networks to improve performance metrics like latency, reliability, and throughput.
Interested in licensing this patent?