Performance Analysis and Design of Mobile Ad-Hoc Networks

We focus on the performance analysis and design of a wireless ad-hoc network using a virtual-circuit or reservation based medium access layer. In a reservation based MAC network, source nodes reserve a session's link capacity end-to-end over the entire path before sending traffic over the established path. An example of a generic reservation based MAC protocol is Unifying Slot Assignment Protocol (USAP). Any reservation based medium access protocol (including USAP) uses a simple set of rules to determine the cells or timeslots available at a node to reserve link capacity along the path to the next node. Given inputs of node locations, traffic pattern between nodes and link propagation matrices, we develop models to estimate blocking probability and throughput for reservation based wireless ad-hoc networks. These models are based on extending reduced load loss network models for a wireless network. For generic USAP with multiple frequency channels, the key effect of multiuser interference on a link is modeled via reduced available link capacity where the effects of transmissions and receptions in the link neighborhood are modeled using USAP reservation rules. We compare our results with simulation and obtain good results using our extended reduced load loss network models but with reduced available link capacity distribution obtained by simulation. For the case of generic USAP using a single frequency channel, we develop models for unicast traffic using reduced load loss network models but with the sharing of the wireless medium between a node and its neighbors modeled by considering cliques of neighboring interfering links around a particular link. We compare results of this model with simulation and show good match. We also develop models to calculate source-destination throughput for the reservation MAC as used in the Joint Tactical Radio System to support both unicast and multicast traffic. These models are based on extending reduced load loss network models for wireless multicast traffic with the sharing of the wireless medium between a node and its (upto 2 hop) neighbors modeled by considering cliques of interfering nodes around a particular node. We compare results of this model with simulation and show good match with simulation. Once we have developed models to estimate throughput and blocking probabilities, we use these models to optimize total network throughput. In order to optimize total throughput, we compute throughput sensitivities of the reduced load loss network model using an implied cost formulation and use these sensitivities to choose the routing probabilities among multiple paths so that total network throughput is maximized. In any network scenario, MANETs can get disconnected into clusters. As part of the MANET design problem, we look at the problem of establishing network connectivity and satisfying required traffic capacity between disconnected clusters by placing a minimum number of advantaged high flying Aerial Platforms (APs) as relay nodes at appropriate places. We also extend the connectivity solution in order to make the network single AP survivable. The problem of providing both connectivity and required capacity between disconnected ground clusters (which contain nodes that can communicate directly with each other) is formulated as a summation-form clustering problem of the ground clusters with the APs along with inter-AP distance constraints that make the AP network connected and with complexity costs that take care of ground cluster to AP capacity constraints. The resultant clustering problem is solved using Deterministic Annealing to find (near) globally optimal solutions for the minimum number and locations of the APs to establish connectivity and provide required traffic capacity between disconnected clusters. The basic connectivity constraints are extended to include conditions that make the resultant network survivable to a single AP failure. In order to make the network single AP survivable, we extend the basic connectivity solution by adding another summation form constraint so that the AP network forms a biconnected network and also by making sure that each ground cluster is connected to atleast two APs. We establish the validity of our algorithms by comparing them with optimal exhaustive search algorithms and show that our algorithms are near-optimal for the problem of establishing connectivity between disconnected clusters