This work is done with my group members Selin Bayramoglu and Berkin Tan Arici as a capstone project for our Industrial Engineering degree in Bogazici University. My colleague Selin Bayramoglu presented it in German Operations Research conference in 2017. All rights reserved.
Data can be generated by generator.nlogo code. Afterwards, one can run the R code and later try on each algorithm to see the effectiveness of each algorithm. CMPL code is the optimum solution for each dataset, without the time constraint. Visualization of the process can be seen in the NetLogo code provided for the Greedy Algorithm. Other algorithms' NetLogo code is available upon request. Report is available for further understanding.
Online frequency assignment problems (FAP) arise in wireless communication domain in a variety of forms. In this project, we consider traveling radiophone users as communication agents in a terrestrial region. If two agents want to make a call and the distance between them is below a threshold they have to be assigned different frequencies, otherwise interference occurs. Given a set of available frequencies, we need to assign a frequency to each call as soon as it happens in such a way that any interference is avoided. If no frequency can be assigned to a call, we assume that the call is dropped. The objective is to minimize the number of dropped calls. The dynamic call network can be represented by a graph where vertices represent the locations of calls and edges between vertices are present whenever the related calls are close enough to each other that can cause interference. Then, the objective is to color this graph with a given number of colors in such a way that no two adjacent vertices get the same color. The problem has an online nature as vertices (and consequently edges) appear and disappear in a given time horizon. In this paper, we provide an experimental study on online FAPs. We use online algorithms from the literature, develop online algorithms and formulate IP models that provide optimal offline solutions using the full instance information. We compare the results and analyze the conditions in which some of the algorithms perform significantly better than the others.