Fuzzy connection admission control for ATM networks based on possibility distribution of cell loss ratio

This paper proposes a connection admission control (CAC) method for asynchronous transfer mode (ATM) networks based on possibility distribution of cell loss ratio (CLR). The possibility distribution is estimated in a fuzzy inference scheme by using observed data of CLR. This method makes possible secure CAC, thereby guaranteeing the allowed CLR. In this paper, first, a fuzzy inference method is proposed, based on a weighted average of fuzzy sets, in order to estimate possibility distribution of CLR. In contrast to conventional methods, the proposed inference method can avoid estimating excessively large values of CLR. Second, the learning algorithm is considered for tuning fuzzy rules for inference. In this, energy functions are derived so as to efficiently achieve higher multiplexing gain by applying them to CAC. Because the upper bound of CLR can easily be obtained from the possibility distribution by using this algorithm, CAC can be performed guaranteeing the allowed CLR. The simulation studies show that the proposed method can well extract the upper bound of CLR from the observed data. The proposed method also makes possible self-compensation in real time for the case where the estimated CLR is smaller than the observed CLR. It preserves the guarantee of the CLR as much as possible in operation of ATM switches. Third, a CAC method which uses the fuzzy inference mentioned above is proposed. In the area with no observed CLR data, fuzzy rules are automatically generated from the fuzzy rules already tuned by the learning algorithm with the existing observed CLR data. Such areas exist because of the absence of experience in connections. This method can guarantee the allowed CLR in the CAC and attains higher multiplex gain as much as possible. The simulation studies show its feasibility. Finally, this paper concludes with some brief discussions.