การประยุกต์ใช้งานฟัซซีนิวรอลเนตเวิร์กสำหรับหน้าที่ควบคุมพารามิเตอร์การใช้ในโครงข่ายเอทีเอ็ม / เอมอัชนา นิรันตสุขรัตน์ = An application of fuzzy neurel networnk and usage parameter control function an ATM networks / Aimaschana Niruntasukrat
This thesis proposes a novel traffic policer for UPC function in ATM Networks. The architecture of this policer is based on the Token Leaky Bucket mechanism using a fuzzy neural network, which dynamically updates the token generation rate. With the inherit of formalized approximate reasoning offered by fuzzy logic, the control process would be smooth and flexible, and would make selectivity improvement possible. With the inherits of mapping capability and fast parallel calculation offered by a neural network, the complexity in inference process of fuzzy logic is eliminated by replacing it with the neural network. The performance evaluation of the proposed policer is done with a packet voice source model and is compared to a conventional method (unbuffered leaky bucket) and the fuzzy logic window. In selectivity aspect, we compare cell loss ratio caused by policers with ideal violation probability in terms of Root of Sum Square Error (RSSE). With the proposed mechanism, unbuffered leaky bucket and fuzzy logic window, the RSSEs obtained are 0.011550, 1.023511 and 0.161944, respectively. The comparison of the responsive behavior in terms of the average number of cells from the violating source allowed by the policer to pass into the network before detecting the first cell is made. The proposed scheme takes action after 359 cells, while the unbuffered LB and fuzzy logic window do so after 309 and 731 cells, respectively. From the results obtained, the fuzzy neural network approach performs well in both aspects. Despite the shower rejection time compared to the unbuffered LB, the percentage of cells detected by the proposed policer grows at the hightest rate in long term comparison. Moreover, the fuzzy neural network approach yields the highest performance in selectivity among other mechanisms.
