Distributed Wireless Node Clustering via Marginal Cost Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Heterogeneous wireless networks with varying power base stations face challenges in centralized data collection for small cell clustering, as knowledge of network nodes and conditions across multiple clusters is not readily available, necessitating a distributed clustering approach without centralization.
Innovation Solution
A method for distributed clustering of wireless network nodes, where each access point determines its marginal cost of association with distinct clusters using a defined cost function, iteratively associates with the cluster minimizing this cost, and designates a cluster head based on total cluster cost, enabling autonomous and efficient cluster formation without central data collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If centralized data collection is used for small cell clustering, then clustering control can be coordinated, but network overhead and control signal traffic increase significantly
Solution Approach 1:
The patent divides the centralized clustering control into distributed cluster head selections across multiple small cells. Each small cell autonomously evaluates its suitability as a cluster head using local cost functions, segmenting the decision-making process from a centralized authority to multiple distributed nodes, thereby reducing control signal overhead while maintaining coordination
Solution Approach 2:
Small cells perform self-evaluation to determine their own suitability as cluster heads by computing cost functions based on local network conditions and parameters. This self-service mechanism eliminates the need for centralized data collection and decision-making, reducing control traffic while enabling autonomous clustering
2Loss of energy
If distributed clustering is implemented without centralization, then control signal overhead is reduced, but knowledge of network nodes and conditions across multiple clusters becomes unavailable
Solution Approach 1:
The patent implements feedback mechanisms where small cells exchange cost function evaluations and cluster head selection information with neighboring cells. This distributed feedback allows each node to make informed decisions about clustering based on local and neighbor information, compensating for the lack of centralized knowledge while maintaining effective cluster formation
Solution Approach 2:
Small cells pre-compute cost functions and evaluate their suitability as cluster heads before actual cluster formation. This preliminary action allows distributed nodes to have ready information about network conditions and node characteristics, enabling informed clustering decisions without requiring centralized data collection at the time of cluster formation
Data Source
Figure 1A~1B
Figure 2
Figure 3A
AI summary
The application relates to an algorithm that organizes access points of a wireless communication network into groups or clusters, whereby a cluster comprises a set of wireless nodes including at least one cluster head and at least one cluster member. Known algorithms rely on knowledge of network nodes and conditions over multiple clusters at a single control point. Because heterogeneous networks are unplanned, such knowledge may not be readily available. There is therefore a need to enable distributed clustering of wireless nodes without any need to centralize data collection over multiple clusters. This problem is solved in the present application as follows: Initial cluster heads CHs are determined. Through neighbor discovery, every node joins a cluster, thereby becoming a cluster member. Each member requests its CH to compute its own marginal cost which is a value of a cost function for the cluster including the requesting member minus the value of the cost function for the cluster omitting the requesting member. Once knowing its own marginal cost, the member contacts any available neighbor cluster and requests computation of the marginal cost in case it would join this cluster. This marginal cost, also referred to as neighbor marginal cost is compared to the own marginal cost and if it is lower, the member quits its current cluster and joins the neighbor cluster. These steps are performed by any member and as long as the arrangement converges to a stable solution.