· · · Institutional Repository

Homes, offices and vehicles are getting networked. This will enable context aware, autonomous operation of many support systems that could be controlled remotely. To achieve this there would be a large number of tiny devices -- sensors and actuators -- which are networked and they are termed generally as Internet of Things (IoT) devices. In future, they will be powered through harvested energy from the ambience to enable perennial lifetime and minimal manual maintenance. Some examples of energy sources are photovoltaic panels and piezoelectric crystals. Several challenges arise due to the nature of sources of energy. One of these challenges is that the devices (nodes) leave and re-enter networks due to fluctuating availability of harvested energy. This energy condition requires the adaptation of special means at every layer of the communication model. For example, as a result of fluctuating energy levels, the neighbor table maintained at each node changes quite often leading to complications in forming and maintaining routes. In fact initial neighbor discovery (ND) itself is a difficult task. Further, usage of directional antennas would affect the time taken to complete ND. Given the spatio-temporal variations in energy availability in harvesting environments, there are benefits of energy prediction. With the help of prediction, resource allocation within a single system and splitting of tasks between nodes in a network would be enhanced. In order to identify the various parameters that affect ND we first describe a generic analytical model of an energy harvesting device. Next, we study a network of these devices through exhaustive simulation study considering these various parameters. We demonstrate the benefits and challenges of using directional antennas for ND. We present a scheme that nodes could use to discover their neighbors during initial deployment and another scheme that could be used for subsequent discovery on re-entry into the network. We show that a dedicated ND protocol is necessary for energy harvesting networks and that directional ND is beneficial in these networks under some circumstances. Finally, we present light-weight energy prediction solutions that can be used to improve the performance of the ND process in particular.

ZigBee is a robust wireless communication standard which is based on physical and MAC layer on IEEE 802.15.4. ZigBee is also critically limited by the low data rate standard when there are many source nodes in the network. How to fairly and efficiently access the network and delivery the packets should be considered as one of the crucial issues. The simulation results illustrate that packet delivery ratio and delay are the most important measures for performance analysis in ZigBee. To obtain a better behavior of ZigBee network, we propose three methods which can significantly improve the packet delivery ratio and satisfy the requirements for fairness. Firstly, packet aggregation is introduced to ZigBee networks to aggregate data in an energy efficient manner so that network lifetime is enhanced. When there are too many nodes in the network, not all the nodes get the same chance to access the network and successfully transmit the packets. Three fairness metrics are introduced to evaluate the fairness among all the nodes. Intra-cluster and inter-cluster fairness are two proposed methods to enhance fairness based on packet aggregations. With intra-cluster fairness methods, cluster head should delay the packets and process the packets from the other node when the number of received packets from one source node is more than the average level. After intra-cluster fairness, different clusters still have fairness issue. Inter-cluster fairness should be designed to achieve absolute fairness among all the end devices. The simulations results show that our new methods could make the way of accessing the network and transmitting the packets efficient and fair as we have expected.

Cognitive Radios (CRs) address the problems of spectrum scarcity and under-utilization of the spectrum. However, realizing a CR network is neither easy nor straight-forward. The link layer in CR Ad hoc Networks is responsible for allocating suitable channels, out of the currently available channels. It is also responsible for setting up communication between nodes. Further, the spectrum efficiency should be maximized in a fair way. The problem of spectrum allocation can be modeled as a graph-theoretic problem. The selection of channels amongst the CR nodes in the network is an NP class problem. We prove this is, in fact, an NP complete problem. We propose a time-slotted system. In such a system, the schedule length needs to be kept to a minimum for higher spectrum utilization. We analyze the problem to determine conditions for an optimal allocation. We use edge coloring as a tool to analyze and propose heuristics. In ad hoc networks, distributed solutions are preferred due to the lack of infrastructure. We propose two distributed algorithms: (i) clique based, and (ii) localized heuristic algorithms. We compare the results of these heuristics with the algorithm proposed in literature. We also find the worst case bounds for these algorithms. For efficiency purposes, it is required to have a constant number of slots per frame. In such cases, producing a valid schedule is not enough since unfairness of allocation will eventually arise. To address this issue, we modify the edge coloring and clique based heuristics to produce valid fair schedules. Finally, we briefly consider the advantages of having a joint spectrum sensing-allocation scheme at the link layer. When the spectrum sensing scheme at the PHY layer is not completely reliable, a link layer scheme can help in reducing the false alarms and miss-detections. We, further constrain the system by limiting the number of channels that can be sensed within a frame. We present the spectrum utilization with this joint sensing-allocation policy.