Customized K-Means Based Topology Clustering for Aeronautical Ad-hoc Networks

Bilen T., Aydemir P. J., Konu A. E., Canberk B.

IEEE 26th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD), ELECTR NETWORK, 25 - 27 October 2021 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/camad52502.2021.9617810
  • Keywords: AANETs, AANET Management, Topology Clustering, K-Means
  • Istanbul Technical University Affiliated: Yes


Aeronautical Ad-Hoc Networking (AANET) aims to provide a dynamic, feasible, and sustainable communication system above the clouds. The fundamental function of AANETs is to transfer data via clustered aircraft network. However, the unstructured and ultra-dynamic characteristics of aircraft make clustering challenging in the AANET environment. Here, these two characteristics reduce the stability of AANET clusters by decreasing the connection life of air-to-air links between aircraft. Therefore, we should obtain more organized and structured clusters consisting of durable air-to-air links to increase the Internet connectivity success efficiency of AANETs. This paper proposes a customized K-means algorithm to cluster the aircraft in a more organized and structured form. This algorithm achieves this stability by considering the position (longitude-latitude), direction, angle, and altitude information during clustering. Accordingly, we can group the aircraft having more similar flight characteristics in the same cluster. Here, we use the Haversine formula to determine the distance between aircraft by utilizing longitude-latitude and angle information. Also, we take North Atlantic as our base airspace to obtain real-time flight traffic data. After creating clusters through customized K-means, we also utilize the Learning Vector Quantization (LVQ) to map the newly arrived and unlabeled aircraft to labeled clusters. Results reveal that we can achieve the aircraft label rate higher than 85% level with the customized K-means algorithm and observe 75% reduced cluster change compared to the standard K-means algorithm.