Comparing Channel Assignment Results from IP Algorithm and DCA Heuristic Kavitha Chandra and Karen Daniels Center for Advanced Computation and Telecommunications University of Massachusetts Lowell Lowell, MA. 01854 Abstract This note accompanies the DCA algorithm paper [1] by Daniels et. al. This paper presents two channel assignment algorithms, one based on an Integer Programming (IP) model and the other based on a heuristic. The number of channels required to satisfy linearly increasing demand function in the randomly distributed Type v cells are presented here for reuse distances greater than 2 and 3. The results show that the new heuristic closely matches the IP solution in the number of channels required under zero blocking condition. The DCA heuristic is also applied to evaluate the blocking probabilities for demand generated by a two state Markov chain based arrival process and uniformly distributed holding times. 1 Linearly Increasing Demand The DCA heuristic is compared to the IP strategy by considering the linearly increasing demand function of Type v cells. The results of the DCA heuristic are shown in Figure 2 under the label H t and may be compared to the IP solution denoted by U ct . These results are for the case where reuse distance is greater than 2 and the threshold that satisifies this constraint is B = 27234. The number of channels found required by the heuristic is seen to be comparable to that generated by the IP algorithm. Typically, the heuristic matches the IP solution to within one channel, and deviates by at most two channels in cases (e,f,g). For case (c), the heuristic improves on the IP performance by requiring one less channel. The performance may also be compared with respect to the average channel reuse afforded by the two approaches. The heuristic reuses each channel on average 4.64 times, in comparison with the IP which reuses each channel an average of 4.5 times. The policy of reusing a channel at a location that creates a median change in residual interference was compared to two other policies where the location is selected so as to minimize or maximize the change in residual interference. The latter policy is often referred to as maximum packing. The median policy was typically the better choice, particularly in cases where the Type v cells were spread across the entire spatial grid. Although the minimum policy was often comparable to the median solution, this approach can lead to divergent solutions in cases such as (d,f), where large Type v clusters reduce the probability of finding a location that is furthest from a group of interfering cells. The maximum packing policy is in all cases the worst performing policy since the cumulative interference quickly constrains the channel reuse factor. The heuristic’s execution time is observed to increase linearly as a function of demand, which is consistent with the analysis of the algorithm [1]. The heuristic is fast, with typical running time of .002 seconds per unit of demand on a 600 MHz Compaq Alpha server. A second set of results is shown in Figure 3 where reuse distance is greater than 3 and the threshold B = 125000. Calculation of the threshold to satisfy the reuse requirements, given a channel model is described in an accompanying note.