**5.5. Performance with harder instances**

To conclude this analysis of the performance of the proposed methods, two more cases that are more difficult have been studied. Their main characteristics are shown in Table 4.


**Table 4.** Definition of benchmark instances A and B.

Our methods are going to be compared with the GA-based approach by Funabiki et al. (Funabiki, 2000). Using the μGA the best assignments we were able to find required 855 and 1713 channels, for problems A and B, respectively, while the method in (Funabiki, 2000) required slightly higher values, 858 and 1724 (results are shown in Table 5). On the other hand, the SQ method requires 855 and 1715 channels, respectively. Thus, in case A –which is a bit simpler than case B- both SQ and GA achieve near optimal results, while in case B –a rather more complex network- SQ requires two more channels than the GA, and 9 less than the NN method. In terms of computation times, the μGA took 11.86 and 23.76 seconds, for problems A and B, respectively; the other GA-based algorithm took about 16.73 and 32.8 seconds, respectively, and, finally, the SQ approach took 15.20 and 37.15 seconds, respectively (see Table 6). This means a reduction in time of 38−41% in favour of the proposed μGA method, while the NN and SQ approaches showed very similar execution times.


**Table 5.** Best assignments for benchmark instances A and B.


**Table 6.** Computation time for benchmark instances A and B.
