Wi-SUN Unslotted CSMA/CA Backoff Adaptation Using Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing unslotted CSMA/CA method in Wi-SUN networks suffers from channel resource wastage, increased latency, and inefficient channel usage due to random backoff times, especially in dense node environments, leading to network performance degradation and energy inefficiency.
Innovation Solution
An unslotted CSMA/CA optimization device and method using reinforcement learning to adaptively optimize channel access parameters by selecting appropriate backoff times based on continuous channel clear access (CCA) measurements during backoff periods, without requiring additional packet exchange between nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If random backoff times are used in unslotted CSMA/CA, then nodes can access channels independently without coordination, but channel resource wastage increases and network performance degrades
Solution Approach 1:
The patent implements feedback mechanisms where nodes monitor channel conditions and adjust their backoff times based on observed network state. The reinforcement learning agent continuously learns from channel access outcomes and modifies backoff strategies to optimize network performance while maintaining independent access capability
Solution Approach 2:
The patent transforms static random backoff times into dynamic adaptive backoff values. The backoff duration is no longer fixed or purely random but dynamically adjusted based on real-time channel conditions, network density, and learned patterns from reinforcement learning, resolving the contradiction between independent access and network performance
2Area of stationary object
If unslotted CSMA/CA is used in dense node environments, then network coverage is extended, but packet collisions increase and channel access efficiency decreases
Solution Approach 1:
The patent applies local quality by having nodes adapt their backoff behavior based on local channel conditions and network density. Each node independently learns and adjusts its backoff strategy according to its specific environment, allowing dense networks to maintain high access efficiency while preserving extended coverage area
Solution Approach 2:
The patent dynamically changes the backoff time parameter based on network conditions. In dense environments, the system learns to select longer backoff times to reduce collisions, while in less dense areas, shorter backoff times maintain efficiency. This parameter adaptation resolves the contradiction between coverage and access efficiency
3Device complexity
If traditional unslotted CSMA/CA backoff method is used, then implementation is simple, but channel resources are wasted and latency increases
Solution Approach 1:
The patent implements self-service through autonomous reinforcement learning agents that automatically learn and optimize backoff strategies without manual configuration. The system serves itself by continuously adapting to changing network conditions, reducing both implementation complexity and access latency through automated optimization
Data Source
AI summary
An unslotted CSMA/CA optimization device includes a variable initializing unit performing variable initialization used in an algorithm for unslotted CSMA/CA optimization; an exploration and exploitation selecting unit determining exploration/exploitation using an epsilon greedy algorithm; an action selecting unit selecting an action having the best Q-value, among actions, when exploitation is selected, and randomly selecting an action when exploration is selected; a channel information collecting unit executing backoff when an action (backoff time) is selected, repeatedly executing CCA during the backoff time, and counting the number of times a channel is idle and the number of times the channel is busy; a success rewarding unit transmitting a packet when the channel is idle and rewarding success when acknowledge (Ack) is received; and a Q-table updating unit checking the received reward and updating a Q-table based on an action, a state, and a reward.