Evolutionary Diversity Optimisation for Sparse Directed Communication Networks

Sharlotte Gounder, Frank Neumann, Aneta Neumann

Abstract: This study proposes Evolutionary Diversity Optimisation (EDO) to Lower the Probability of Detection (LPD) in directed wireless networks. LPD communication aims to communicate between authorised parties, however minimises the probability that an intruder can detect the communication. We represent the problem as a directed graph and our objective is to minimise the area of detectability in a network whilst avoiding adversary nodes to be in the area. We utilise EDO to produce a population of solutions that are of quality i.e., that minimise the area of detectability whilst providing solutions that are diverse. To produce a solution, we find the strongly connected components by running depth-first-search (DFS) twice and extracting the edges traversed from the DFS runs. We use 3 permutation operators (insert, swap, random) to produce different solutions. We propose 2 methods for survival selection - one based on the diversity value and the other on the edge population count. We control the sparsity of the directed graphs by implementing a maximal communication range. Our results show that sparser graphs had smaller areas of detectability, however the final population was less diverse. We also found controlling the maximal communication range an effective strategy to reduce the area of detectability.