Figure 2.3: The architecture of network optimisation in dynamic WSNs.
This chapter briefly discusses network optimisation theories related to this thesis. Figure 2.3 summarises the architecture of network optimisations in dynamic WSNs.
Chapter 3
A Practical Solar Powered WSN
Recent advances in solar harvesting technologies pave the way for sustainable environmental- monitoring applications in the emerging field of solar powered wireless sensor networks (SP- WSNs). The complexities associated with the low-resource, high-dynamic, and vulnerable sensor nodes operating in potentially unattended or hostile environments require a high degree of self-management and automation. This chapter presents AutoSP-WSN, a novel distributed framework that achieves sustainable data collection while also optimising end-to-end network performance for SP-WSNs. Initially, we present an energy-aware support component that provides reliable energy monitoring and prediction. This drives the power management com- ponent, which is adaptive to time-varying solar power, avoiding battery exhaustion as well as maximising the per-node utility. Finally, to demonstrate the key design issues of the network protocols, we propose two adaptive network protocols, a routing protocol SP-BCP and a rate control scheme PEA-DLEX. Through extensive experiments on a real SP-WSN platform and hardware-driven simulations, we show that the proposed schemes achieve substantial improve- ments over previous work, in terms of reliability, sustainable operation, and network utility.
The experimental work in this chapter demonstrates the practical issues of dynamic low-power and low-capacity WSNs. The algorithm design principles such as the adaptiveness of dynamic network conditions, and the tradeoffs between optimality and complexity are also highlighted. Such practical principles also guide the algorithm design in later chapters.
Pload Power consumed by the load (MicaZ mote).
Psolar Power generated from solar panel.
Pbattery Battery recharging power.
Iload The electric current of the load.
Isolar The electric current of the solar panel.
Vsolar The voltage across the solar panel.
η1 Energy translation efficiency from solar panel to load. η2 Energy translation efficiency from solar panel to battery. η3 Energy translation efficiency from battery to load. M The number of prediction intervals in a day. L The number of slots in a prediction interval. T The duration of a slot.
Pre(i, l, d) The reference solar power at the ith slot in lth at the dth day.
Psolar
real (i, l, d) The real solar power at the ith slot in lth at the dth day.
αre Weighting parameter for computing the reference solar power vector.
αwv Weighting parameter for strong weather volatility.
wv(d) Overall weather condition level of the dth day.
wv0(d) The number of peaks and troughs of the solar profile in the dth day. wv1(d) Strong weather volatility of the dth day.
F IT Predefined threshold to defined strong weather volatility.
αwv Weighting parameter for strong and weak weather volatilities.
Pactive Load power when a sensor node is active. Pidle Load power when a sensor node is idle. Dmax
x (l) The maximum feasible duty cycle of node x at prediction interval l.
P Si
x The solar power of node x at slot i.
Eleak Battery energy leakage in a slot.
Bmax The maximal amount of energy that a battery can store.
Dmin User defined minimum duty cycle. ϕ A constraint parameter for S-type LPM. ϕi
x(d) A constraint parameter of node x at slot i in the dth day for D-type LPM.
Emin The minimum energy consumption in a slot.
i0(d) The first slot of daytime in the dth day. i1(d) The first slot of night in the dth day.
BEN O The minimal energy that a sensor node should store at i1(d) to ensure ENO. DRx The remaining active durations of node x.
DPx,y The duration required for transmitting a packet from node x to node y.
Qx The data queue backlog at node x.
wx,y the backpressure weight of wireless link (x,y).
STx A subtree rooted at node x, excluding x.
C wireless link capacity. rx The sensing rate of node x.
R R = r1, r2, ... , rx, a sensing rate allocation of all nodes.
AFx The actual date forwarding rate of node x.
λx The maximum allowed data rate of x.
Bx(i) The difference between actual and expected battery levels of node x in slot i.
DBT Predefined threshold of battery level difference to trigger RD or RI events.
λx The difference between current and last updated λx.
3.1. Introduction 25
SP-WSNs Solar Powered Wireless Sensor Networks AutoSP-WSN Autonomic SP-WSN
BCP Backpressure Collection Protocol SP-BCP Solar Powered BCP
DLEX A Distributed max-min rate allocation algorithm defined in [54] PEA-DLEX Prediction Error Adaptive DLEX.
LPM Local Power Management
ENO Energy Neutral Operation
EA Energy Awareness
WC-EWMA Weather-Conditioned Exponential Weighted Moving Average D-type Dynamic type
S-type Static type
MAC Medium Access Control
CTP Collection Tree Protocol MCU MicoController Unit
RD Rate Decrease
RI Rate Increase
Table 3.2: Summary of abbreviations used in Chapter 3.
3.1
Introduction
Environmental monitoring is one of the most important applications for Wireless Sensor Net- works (WSNs) [187]. The majority of sensor nodes are currently powered by batteries that require manual replacement when they are depleted. However, in many outdoor deployments it can be difficult or even impossible to physically access the sensor nodes. As a result, the bounded lifetime of WSNs has become a restriction that impacts their use in long-term envi- ronmental monitoring applications.
The development of new photovoltaic harvesting techniques, such as those found in [14,141], are demonstrating that solar power is indeed a viable first step towards autonomous WSNs [171]. However, due to the limited size of micro solar panels, harvested solar energy remains scarce. To make best use of this resource, one needs to know how the node uses power, and to understand the dynamics of energy generation. However, hardware power usage is complex and the nature of solar power is highly dynamic. Further complications arise because of the heterogeneous spatial harvesting capabilities across different nodes in a sensing space, due to shading or cloud
the sink
wireless channel
(a) A sunny day
(b) A cloudy day
7LPH 3R Z HU P : 7LPH 3R Z H U P :solar powered sensor nodes
time-varying solar profile
heterogeneous harvesting opportunities
Figure 3.1: Illustration of a typical multi-hop SP-WSN.
coverage: see Figure 3.1. These have significant impacts on the design of reliable multi-hop solar-powered WSNs (SP-WSNs). For instance, a routing protocol should adaptively select and dynamically adjust the end-to-end path to avoid the time-varying routing hot-spots (i.e. relay nodes with low harvesting power).
This chapter therefore aims to provide sustainable data collection capability, while also max- imising end-to-end delivery performance (e.g. network goodput) for SP-WSNs. To this end, the following issues should be considered. Firstly, the internal power system of sensor nodes should be realistically modelled (e.g. solar power and battery recharging/discharging process) to pro- vide reliable energy awareness. Secondly, the capacity of self-management for each individual node is required to ensure sustainable operation while optimising its local long-term utilities (this is its power management function). Thirdly, through adapting to the time-varying and heterogeneous distributed solar harvesting opportunities, distributed and responsive networking protocols (e.g. routing) should be derived for SP-WSNs to achieve sustainable data collection and optimal global end-to-end performance.
Current studies on energy harvesting WSNs, per-node power management schemes [81, 108, 119, 151], and network-wide protocols [51, 53, 103, 190, 192], address the above issues separately.
3.1. Introduction 27
Furthermore, most are theoretical work, and are therefore likely to perform poorly or even fail in real-world SP-WSNs. In this chapter, we present a systematic study covering both individual components (i.e. energy awareness, power management, and network protocols) and the SP-WSN as a whole. The contributions of this chapter are summarised as follows:
1. We develop and implement AutoSP-WSN which is, to our knowledge, the first distributed framework for practical SP-WSNs. The overarching goal of the AutoSP-WSN is to opti- mise the usage of solar power as well as the end-to-end network performance, and more importantly, to achieve so-called Energy Neutral Operation (ENO) [81], i.e. to guarantee that no node will run out of energy at any point in real-world deployments.
2. We propose an efficient and reliable Energy Awareness (EA) support component that pro- vides realistic hardware-driven power models as well as reliable real-time energy monitor- ing. To further support EA, we also propose Weather-Conditioned Exponential Weighted Moving Average (WC-EWMA), a lightweight weather-aware prediction algorithm that forecasts future solar energy with a high degree of accuracy.
3. Based on the reliable EA support, we develop a Local Power Management (LPM) com- ponent that achieves both ENO and long-term per-node utility optimisation. Compared with directly solving a linear programming problem at runtime, our LPM remains optimal while achieving a much lower overhead. In addition, the LPM can automatically tune its parameters to support different types of network protocols.
4. To demonstrate the key issues of designing network protocols in practical SP-WSNs (e.g. the requirement of self-protection capacity to avoid network failure caused by solar power prediction error), we propose two adaptive network protocols: a routing protocol called SP-BCP, and a rate control protocol called PEA-DLEX, for sustainable multi-hop data collections.
5. Through extensive evaluations on both our own developed SP-WSN platform and the Tossim simulator [97], we show that both the proposed individual schemes and the entire AutoSP-WSN achieve substantial improvements over previous approaches.
All symbols and abbreviations used in this chapter are summarised in Tables 3.1 and 3.2 respectively. The remainder of this chapter is organised as follows. The next section presents the related work. The overall AutoSP-WSN architecture is described in Section 3.3. Section 3.4 discusses the hardware and software support for energy awareness. The energy prediction, WC- EWMA, is described in Section 3.5. Section 3.6 presents the details of the LPM component. Two solar-aware network protocols, SP-BCP and PEA-DLEX are proposed in Sections 3.7 and 3.8 respectively. Evaluations in Section 3.9 show the performance and effectiveness of AutoSP-WSN. Finally, we summarise this chapter in Section 3.10.