| dc.description.abstract | Mobile and wireless networks represent the next trend in networking because of their usefulness in assisting an emerging mobile workforce in a growing information?oriented society. Advances in networking technologies, wireless communications, and portable computing devices-along with the reduction in the physical size of computers-have led to the rapid development of mobile communication infrastructure and caused a new paradigm of computing and support systems. Hence, mobile and wireless networks present many challenges to application, hardware, software, and network designers and implementers.
The new challenges in designing supporting systems for mobile networks include resource management, mobility management, power conservation, and security. The increasing use of small portable computers, wireless networks, and satellites to support computing has stimulated the design of mobile networks for mobile computing.
The most important characteristics in mobile networks that bring several issues toward design and research activities are the requirements to share a limited bandwidth due to several wireless problems such as fading, Doppler effect, etc., and user mobility. The combination of dynamic network state due to mobility and limited network resources has several important implications for network control issues encountered in mobile networks. Some of the major issues are resource management, mobility management, security, and mobile computing.
Most problems in mobile networks are real?time based, requiring fast computation, real?time optimal solutions, and adaptability to the network?traffic situation to achieve desired goals. Some of them, such as optimal path finding, optimal resource allocation, and call management, are NP?hard. Hence, heuristic methods are often used to provide rapid and near?optimal solutions. Neural networks provide a novel and potentially powerful alternative approach to solving such problems. They are intrinsically parallel, with much potential for hardware implementation, and adaptable to the system since they can be retrained in real time using the latest measurements.
Neural networks have emerged as powerful and distributed computing architectures equipped with significant learning abilities. In summary, neural networks help represent highly nonlinear and multivariable relationships with ease. Some computations performed by neural networks include classification, regression, and constrained function optimization.
Several research teams have suggested that developments in neural?network technology might provide capabilities well suited to solving challenging and outstanding control problems in communication systems, particularly in high?speed and mobile networks. These proposals use two classes of neural networks: feed?forward multilayer neural networks and optimization neural networks.
Feed?forward neural networks have been proposed to provide adaptation to the changing conditions encountered in operational communication networks. On the other hand, optimization neural networks have been proposed to offer sub?microsecond solutions to some intractable optimization problems in mobile networks, such as routing, packet switching, and dynamic resource allocation.
The potential benefits of neural networks include:
� efficient adaptive control through the use of adaptive learning
� real?time solutions to optimization problems due to massive parallelism
� high degree of robustness
Admission Control
In mobile networks, traffic fluctuation is unpredictable due to the mobility of mobile hosts and the varying resource requirements of multimedia applications. It is also challenging to handle handoff and new applications when they arrive in bulk (bursty traffic) at a high rate due to micro/pico?cellular environments and increasing user density.
To maintain service guarantees for running applications, it is essential to keep traffic within network capacity. This thesis proposes an Admission Control (AC) mechanism in a mobile cellular environment that supports real?time and non?real?time application traffic. Each multimedia application type has its own distinct acceptable ranges of QoS requirements (packet loss, delay, jitter, etc.).
We propose a Linear Programming Resource Reduction (LP?RR) principle for admission control while maintaining QoS guarantees for existing applications. Recurrent Neural Networks (RNNs) are used to solve this linear programming problem and take online admission?control decisions for handoff and new applications.
Location Management
Location management is the process by which the current location of a mobile host (MH) is determined. It consists of:
� mobile tracking � keeping track of the MH抯 current location
� mobile locating � finding the MH抯 location to deliver an incoming call
One way to find an MH抯 location is by broadcasting paging messages in every cell in the geographical area. However, this consumes extremely high channel bandwidth. Another method requires MHs to send location?update signals, but this generates high signalling traffic.
Several attempts have been made to reduce traffic during location updates and paging. One approach predicts a mobile host抯 future location using its movement behavior and traffic characteristics. If the location can be predicted accurately, explicit location updates and extensive paging can be minimized.
We propose a prediction?based location management method using a Multi?layer Neural Network (MNN). The method predicts future MH locations based on training the MNN with historical movement patterns. The performance is compared against an existing prediction method.
Connectivity Management
Mobile computing in wireless networks is a fast?growing environment offering anytime?anywhere information services. Due to mobility, heavy information exchange, and computational requirements, mobile computing must maintain continuity of data transfer among participating devices.
Wireless limitations and bandwidth constraints lead to frequent disconnections, posing challenges in maintaining user?to?user connectivity. We propose a Neural Network?based connectivity management system that maintains MH status information at the base station to handle disconnections due to handoffs and interruptions. Neural networks are trained on status information to make intelligent connectivity decisions.
Reliable Multicast Routing
Mobile networks handling multicast services (video?on?demand, news?on?demand, etc.) require reliable point?to?multipoint communication. Mobility causes routes to change frequently, creating challenges for multicast routing.
We propose a neural?network?based multicast routing algorithm that constructs a reliable multicast tree connecting group members. A mobile network is divided into clusters of nodes using adjacency relations. A central cluster is computed based on its proximity to group members. A multicast tree passing through this central cluster is constructed.
� Kohonen抯 Self?Organizing Map (KNN) is used for clustering
� Hopfield Neural Networks (HNNs) are used to construct an optimal multicast tree
This reduces recomputation time when mobile hosts move.
Simulation and Results
All proposed techniques were simulated individually on a Linux platform using a Pentium III (500 MHz) system. Programs were written in C/C++.
Key results:
� Proposed AC scheme achieves higher acceptance and lower rejection rates, with fair resource allocation.
� Location management scheme shows high accuracy for uniform and regular movement patterns.
� Connectivity management exhibits high acceptance rates and effective resource utilization.
� Multicast routing shows low computation time and minimal number of links, even with host mobility.
Overall, neural networks provide efficient, real?time, intelligent solutions to multiple mobile?network challenges. | |
