Call Admission Control with Neural network in ATM networks

• Main Authors: Yu-Fen Chung, 鍾玉芬
• Other Authors: 黃蓮池
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/84887464996173855331

碩士 === 義守大學 === 資訊工程學系 === 89 === Asynchronous Transfer Mode (ATM) netowrks is an essential technology for integrating multimedia services in high-speed networks and recommended by International Telecommunications Union (ITU) for broadband integrated services digital networks (B-ISDN). It provides different quality of services (QoS) for different types of traffic sources with widely varying traffic characteristics. In order to guarantee the QoS requirements and to achieve high network utilization, it is necessary to implement a call admission controller. In this thesis, we investigate a CAC algorithm with neural network first. Owing to the self-learning capacity, the neural networks can be trained to fit the uncertainty of the traffic source. However, increasing the inputs result in increasing the complexity of the neural network. We think that different types of inputs can be processed separately, such as network status and traffic characteristics. Therefore, we increase the inputs of neural networks to promote its ability. Secondly, we propose a CAC scheme with two algorithms, named B&WCAC. In many literatures, they always apply only one algorithm to implement CAC. In ATM networks, we think it can adopt more than one algorithm to implement CAC scheme according to different types of traffic sources. We expect that it can be more precise and efficient for CAC by dividing traffic sources and algorithms into two types, black and white.