Abstract

Handoff is the procedure providing the connection to the backbone network while a mobile terminal is moving across the boundaries of coverage of two wireless points of connection. The complexity of the handoff decision process has led to the examination of a number of traditional and pattern recognition handoff decision algorithms for wireless networks. Traditional algorithms use a received signal strength measurement and an optional threshold, hysteresis, or a dwell timer to determine the handoff decision. Degradation of the signal level, however, is a random process, and simple decision mechanisms result in a ping–pong effect whereby several consecutive handoffs degrade the service provided by the network. Consequently, more complex pattern recognition algorithms are needed to decide on the optimal time for handoff. In these algorithms, the handoff decision receives off line training to create a reference database of possible handoff locations in an environment with an associated handoff "fingerprint" at those locations. This dissertation introduces newly designed neural network and adaptive network based fuzzy inference system (ANFIS) pattern recognition algorithms. To select appropriate algorithms for a specific wireless network, we need to create an analytical framework for performance evaluation. The design of a framework for comparative performance evaluation of different handoff algorithms is a complex problem as different networks have different performance evaluation criteria.

This dissertation divides wireless networks into three categories according to their topology and wireless service application: traditional cellular phone networks, heterogeneous wireless data networks, and rate adaptive wireless data networks. For each category of wireless networks we define a performance evaluation scenario and using Monte Carlo simulations, Monte Carlo calculations, and direct mathematical analysis we analyze the effects of different handoff decision algorithms. The Manhattan micro-cellular scenario is used for traditional cellular phone networks. Using Monte Carlo simulations on this scenario, the performance of traditional and our neural network and ANFIS handoff decision algorithms are compared. A moving-in moving-out performance evaluation scenario for heterogeneous wireless data networks is defined to characterize intertechnology roaming between two networks with substantially different data rates. We use Monte Carlo calculations to define the optimum handoff location for a mobile terminal in this scenario. Using Monte Carlo simulations and the optimal handoff location, we perform comparative performance evaluation of newly introduced asymmetric traditional and pattern recognition algorithms designed for intertechnology handoff. Finally, we introduce two performance evaluation scenarios for rate adaptive wireless networks to characterize user mobility in rate adaptive networks with random and grid deployments. For the first scenario we provide mathematical analysis for the effects of handoff using relative power to calculate the average throughput observed by the mobile terminal for different distances between the two wireless points of connection. For the second scenario designed for grid deployment we present a comparative performance analysis using Monte Carlo calculations for four handoff decision algorithms.