| Summary: | In wireless sensor networks (WSNs), network lifetime (NL) maximization plays a significant role in striking a compelling compromise between maximizing the overall throughput and minimizing the energy dissipation, while extending the duration of adequate communications without battery-replacement, when the sensor nodes rely on limited energy supply. Hence, this thesis focuses on the NL maximization of battery-constrained WSNs, which is vitally important in industrial applications, where thousands of sensors may be deployed within a specific target area and the energy dissipation of each sensor node has to be minimized in order to reduce the overall cost of the applications to the industry. However, maintaining stringent quality of service (QoS) requirements under the above-mentioned NL constraints can be challenging and requires careful consideration of several conflicting design tradeoffs. Naturally, the above-mentioned energy dissipation characteristics are dependent on the entire seven-layer OSI protocol stack, where each layer contributes to the dissipated energy. Therefore, NL maximization necessitates a cross-layer operation across all these layers, where each layer has to minimize its energy dissipation without deteriorating the QoS. Hence, our objective is to maximize the NL using cross-layer design techniques in the interest of maintaining certain QoS requirements and to provide the system designer with well-informed decisions prior to embarking on hardware implementations. Hence, our approach is to investigate and to model progressively more realistic WSNs. We commence with a broad overview of the WSNs, of the design objectives and of the NL maximization techniques that have been investigated in the literature. We then provide a concise introduction to convexity, convex optimization, to the Lagrangian dual problem and to the Karush-Kuhn-Tucker (KKT) optimality conditions, which will be extensively used in our studies. Having presented the fundamentals, we formulate an initial study of the NL maximization problem based on a simple string topology in order to form a basic framework for the NL maximization of more realistic large scale networks. In this particular study, we maximize the NL in an interference-limited WSN considering a beneficial rate and power allocation scheme under both additive white Gaussian noise (AWGN) and fading channel characteristics, where we employ the KKT optimality conditions for obtaining the optimal solution to the NL maximization problem using closed-form expressions. Therefore, we were able to derive analytical expressions of the globally optimal NL for a string network operating in an interference-limited scenario, while communicating either over an AWGN or over fading channels for a given link schedule. Furthermore, the maximum NL, the energy dissipation per node, the average transmission power per link and the lifetime of all nodes in the network are obtained. We quantify how the maximum NL is reduced as a function of the fading statistics due to the poor channel conditions. Furthermore, we demonstrate that given a certain network-sum-rate, the simultaneous scheduling of weakly interfering links benefits from the associated spatial reuse by allowing each node to transmit at a lower rate, which requires a reduced transmission power and hence results in an increased NL. We also conclude that the choice of the particular scheduling scheme depends on the application, since a lower source rate favors infrequent transmissions requiring a low transmit power, while avoiding the detrimental effects of interference, when aiming for extending the NL. However, we observe that for higher source rates, a higher NL can be achieved by aggressive spatial reuse. An interesting observation is that increasing the distance between the consecutive nodes substantially reduces the NL, especially for lower source rates. However, quite surprisingly, increasing the distance between the consecutive nodes results in an improved NL for higher source rates. This is due to the reduced impact of the interferers located at a higher distance. More explicitly, even though the transmit power required has to be increased to satisfy the rate constraint, at the same time the interferers are moved a bit further away. In this particular study, the NL and source rate are considered as the QoS measure as a function of both the transmit rate and the power, where an adaptive scheme is assumed. Finally, our proposed algorithm achieved reduced complexity NL maximization compared to other techniques found in the literature. We then extend our NL maximization problem to a realistic scenario, where the parameters are selected from the practical data sheet of a National Instruments device, which is based on the IEEE 802.15.4 Standard and the energy dissipation of the signal processing operations, i.e. the energy dissipation of the transceiver circuits, is considered. Since achieving a reasonable NL at the cost of a tolerable end-to-end bit error rate (BER) for a fixed-rate system using various modulation and coding schemes (MCSs) is an important objective for the system designer considering the QoS, we strike a trade-off between the BER and the NL, which is crucial for network designers at an early design stage. Therefore, we aim for maximizing the NL for a predetermined set of target signal-to-interference-plus-noise ratio (SINR) values, which guarantees maintaining the predefined QoS of each link operating over either an AWGN channel or a Rayleigh block-fading channel, while considering or disregarding the signal processing power (SPP). We observed that especially for low target SINRs, the SPP has a dominant impact on the NL. However, for higher target SINRs the achievable NL only considering the transmit power whilst disregarding the SPP forms a benchmark for the achievable NL of the particular scenario, when the SPP is jointly considered along with the transmit power. As a further advance, a more realistic network is considered, where the same National Instruments device, which is based on the IEEE 802.15.4 Standard is used as a reference. For this realistic network, we also had to reconsider our NL definition, where we maximize the NL of a WSN relying on randomly and uniformly distributed fully connected nodes. This fully connected WSN imposes an exponentially increasing routing complexity upon increasing the number of nodes. More particularly, we focus our attention on the crosslayer optimization of the power allocation, scheduling and routing operations for the sake of NL maximization for predetermined per-link target SINR values. We use the so-called exhaustive search algorithm (ESA) as our benchmarker and conceive a near-optimal single objective genetic algorithm (SOGA) imposing a substantially reduced complexity in fully connected WSNs. We show that our NL maximization approach is powerful in terms of prolonging the NL, while striking a trade-off between the NL and the QoS requirements. Finally, we consider a multiobjective NL maximization problem, where the end-to-end delay and the energy dissipation are considered as our conflicting design objectives. More explicitly, we proposed a novel NL optimization design in order to reflect the effect of the end-to-end delay on the NL along with the aggregate energy dissipation of the same route. The distinctive aspect of this study is the simultaneous optimization of both the aggregate energy dissipation and of the end-to-end delay as a multi-objective optimization problem in order to provide the system designer with a trade-off between Pareto-optimal energyand delay-solutions. === We employ multi-objective evolutionary algorithms (MOEAs), namely the so-called non-dominated sorting based genetic algorithm-II (NSGA-II) and the multiobjective differential evolution (MODE) algorithm for obtaining the set of Pareto optimal NL-aware routes striking a trade-off between the aggregate energy dissipation and the end-to-end delay. Moreover, we characterize both the complexity and the convergence of both algorithms compared to the ESA.
|
|---|
