Topics Discussed Regarding Energy Efficiency

Published Date: 02 Nov 2017

Sensor node is battery operated and so is energy constrained affecting the lifetime of the whole network. Their inexpensive nature and ad hoc method of deployment provides many serious energy and computational constraints (Archana and Vijay Anand, 2004). All the aspects like entire architecture, protocols for communication, algorithms, circuits and sensing has to be energy efficient. Continuous research is being done to exploit this area in detail. Several techniques can be applied at design time to reduce the power consumption (static approaches). Lifetime of sensor nodes depend greatly on the power consumption in each sensor node. Energy constraint in wireless sensor networks affects the whole network lifetime and connectivity. Energy consumption is the most important factor to determine the life of a sensor network because usually sensor nodes are driven by battery and have very low energy resources. Reduce the energy is more complicated in sensor networks because it involves not only the reduction in energy consumption but also prolonging of the life of the network, as much as possible. Efficient energy management should be incorporated in all levels of system hierarchy from hardware to software architecture and from operating system to the communication protocols. All system components critically affect the energy dissipation depending on the application involved (Xiao-Hui Lin and Yu-Kwong Kwok, 2006). So energy awareness must be involved in every level of the system design and operation, to maintain the network connectivity and lifetime (Eugene et al., 2001, Carle and Ryl, 2004, Hong et al., 2001, Raghunathan et al., 2002 and Ye et al., 2002).

Sensor nodes exist in an unattended environment. So highly efficient power management is essential to improve the lifetime (Chuan et al., 2006). System lifetime can be very much extended by applying energy efficient techniques to all levels of system hierarchy (Eugene et al., 2001). Much research has been done to have a significant decrease in energy consumption in various aspects of hardware design, data processing, network protocols, and operating system. Based on the system computation aspects, the research efforts prove the following results. Supply voltage can be actively and adaptively adjusted, in

Dynamic Voltage Scaling (DVS), in conjunction with the clock frequency, in response to the CPU utilization (Sinha and Chandrakasan, 2001). Different keys of varying length can be used at the application layer, by allowing a tradeoff between the expended computation energy and security (Yuang and Qu, 2002). By the proper design of the operating system for sensors, the different components of the node can be made to enter various states (idle, sleep, active), to save energy according to the environmental variations, at the expense of some degree of system performance degradation (Schurgers et al., 2002).

The major energy consumers in wireless sensor networks are the sensing unit, the computation unit and the communication unit. Dynamic Modulation Scaling (DMS), similar to DVS, is proposed by Sinha and Chandrakasan (2001) and Schurgers et al., (2001). According to the number of queued packets in the system, DMS can adaptively change the modulation level, to lower the overall energy consumption, while bounding packet delay at an acceptable level. DMS is combined with packet fair queuing algorithm and this result in an energy efficient packet scheduling protocol similar to NTP (Network Time Protocol). It first organizes the wireless sensor networks to form a hierarchical structure. Along every edge of this tree like structure network, synchronizing algorithm based on a two way message exchange is performed by taking the root node as the reference node. This leads to a simple implementation. Every node must synchronize with the parent node, by pair wise message transmission similar to NTP. Lot of traffic

overhead will be incurred (Quing et al., 2005).

The resource availability constraints of the wireless sensor networks impose specific requirements on the protocol design for time synchronization, which is essential for the self configuration feature of the wireless sensor networks. To realize real-time event management and event monitoring in distributed networks, time synchronization is highly essential. Reliability of data transmission should be reinforced, according to the

fluctuation in link quality with respect to time. This can be achieved by increasing the transmission power level or by adding FEC (Forward Error Correction) to the raw data. The first method leads to the rapid depletion of sensor energy and produce interference to wireless transmission at the terminals (Xiao-Hui Lin and Yu-Kwong Kwok, 2006). Using the second method, as the channel quality changes with time, the amount of error protection incorporated should also vary with the instantaneous channel condition, to make sure that Bit Error Rate (BER) rises above the required level. So a more amount of error protection redundancy in the transmitted packet occurs for poorer wireless links and vice versa (Kwok and Lau, 2002). This is due to the reason that more error protection bits are needed in a poor wireless link, in order to provide a reliable transmission. The following aspects should be considered, while discussing the extra energy dissipation incurred to combat the extra energy consumption.

1. Considering the computation point of view, due to the packet redundancy more energy is expended for encoding and decoding data at the two communication sides. This decreases the battery life.

2. Length of every frame increases on including error protection. So an extra energy is needed for message communication. For the same transmission rate, all the radio circuits have to be ON for a longer duration. More energy is consumed. This makes the design of energy resource management schemes very much challenging (Cianca et al., 2002).

To have scalability and Energy efficiency in a sensor network, cluster based hierarchy is preferred as the ideal solution (Zhou et al., 2004) and (Zhao and Govindan, 2003). Data collected by the sensors in close proximity is highly correlated. Communication between each sensor and end user is both energy as well as bandwidth consuming. So the data should be processed locally to get rid of data redundancy. The whole network is divided into different clusters. One sensor node is elected as the cluster head to perform local information filtering, aggregation and data fusion for all the sensors in its cluster. Traffic is routed among cluster heads. Thus the network management gets simplified and also decreases the energy needed for communicating useful data to the end user. Different methods for organization of cluster based networks are discussed by Banbyopadhyay and

Coyle (2003), Hac (2003), Shen et al., (2000) and Akyildig et al., (2002).

2.1 TOPICS DISCUSSED REGARDING ENERGY EFFICIENCY

The main issue to be dealt with, in the design of wireless sensor networks is to reduce the energy consumption, so as to prolong the life of the network. Research efforts in the past which resulted in bringing forth many techniques to tackle this problem, based on various aspects, are discussed below.

2.1.1 Clustering concept

Clustering can localize the route set up inside clusters and reduce the size of the routing table maintained inside a cluster. It can conserve the communication bandwidth, can stabilize the network topology, and can implement the optimized management strategies to enhance the network operation so as to prolong the network lifetime of the sensors (Younis et al., 2003). Cluster heads can effectively schedule the activities in the cluster so that its nodes can switch to low power sleep modes most of the time to reduce energy consumption. Similar packets from multiple nodes may be aggregated. So the number of transmissions gets reduced. Data aggregation combines the data from different sources by using various functions like suppression (for eliminating duplicates), minimum, maximum and average (Krishnamachari et al., 2002). Computation is energy efficient compared to communication. So aggregation can produce good energy savings. In the self organizing systems, sensor nodes are scattered randomly (Sohrabi et al., 2000, Heinzelman et al., 2002, Younis et al., 2004 and Doina, 2009). In terms of Energy efficiency and performance, the cluster head positioning is very crucial. Optimal clustering always leads to an energy efficient network operation. Cluster heads are picked from the deployed sensors in the network of homogeneous sensor nodes (Heinzelman et al., 2002, Lindsey and Raghavendra, 2002). Cluster heads are carefully tasked to avoid the energy from being depleted away unnecessarily. Communication range and proximity to the base station are some important issues to be considered. If the sensor communication ranges do not reach the base station, multi hop routes have to be used. Inter cluster head connectivity is an important factor affecting the clustering schemes (Banerjee and Khuller, 2001 and Banbyopadhyay and Coyle, 2003).

(a) Load balancing

Sensors should be evenly distributed among the clusters, where the cluster head performs data processing and intra cluster management duties (Gupta and Younis, 2003). Load balancing is a critical issue in wireless sensor networks where the cluster heads are picked from the currently available sensors (Younis and Fahmy, 2004). For extending the network lifetime, equal sized clusters are important. This prevents the exhaustion of energy of a subset of cluster heads at a high rate and prevents their premature failure.

(b) Fault tolerance

This is to avoid the loss of important data. To recover from cluster head failure re-clustering of the network is needed. But during this process, resource burden occurs on the nodes. Backup cluster heads are also assigned to recover from failure. Neighboring cluster heads can adapt sensors in the failing clusters, if the nodes have sufficient radio range (Gupta and Younis, 2003). Rotating the role of cluster heads among all the nodes in the cluster can be a means of attaining fault tolerance (Heinzelman et al., 2002).

(c) Increased connectivity and reduced delay

In large networks, to decrease the energy needed for communication, inter cluster head connectivity is needed. This becomes a critical requirement unless the cluster heads have long haul communication capability.

(d) Maximum network life

Network lifetime is of major concern especially in bad environments. If cluster heads are richer in resources, the energy for intra cluster communication can be minimized (Younis et al., 2003). Otherwise cluster heads should be placed very close to their sensors (Ilker et al., 2004 and Hou et al., 2005). If cluster heads are regular sensors, lifetime can be increased only by limiting their load. Combined clustering and route setup can be together considered for maximizing the network lifetime (Dasgupta et al., 2003). Adaptive clustering can be used to increase the network life (Khanna et al., 2006 and Moscibroda and Wattenhofer, 2005). Low Energy Adaptive Clustering Hierarchy (LEACH) is proposed by Heinzelman et al., (2002). It forms clusters based on the received signal strength and uses cluster head nodes as the routers to the base station. All data processing is done local to the cluster. Distributed algorithm is used by nodes to make autonomous decisions without using a centralized control. Initially a node decides to be the cluster head and it broadcasts its decision to others. Each non cluster head node now determines its suitable cluster, by choosing the cluster head that can be reached using the least communication energy. Role of being the cluster head can be rotated periodically among the nodes of the cluster, in order to balance the load. Rotation is performed by making each node to choose a random number between 0 and 1. A node becomes the cluster head for the current rotation if this number is less than a Threshold value. The cluster heads are assumed to have sufficient communication range so as to reach the base station directly. Different clustering strategies and clustering algorithms have been discussed by Ameer and Mohamed (2007). The different clustering schemes are classified according to their objectives, the desired cluster properties and the clustering process.

2.1.4 Time synchronization aspect

Time synchronization plays a key role to meet the real time requirements and to improve the multiplexing efficiency. Performance limitation of time synchronization for wireless sensor networks in terms of synchronization accuracy is discussed by Quing et al., (2005). The sources of synchronization accuracy are identified and the mathematical models to analyze the time synchronization schemes are proposed here. The light weight protocols proposed are capable of suppressing the communication overheads and approaching the performance limit. This is based on the idea that there always exists a synchronization error correlation between nodes receiving the same sequence of time synchronized packets.

Theoretical analysis is validated by the simulation result Time synchronization is essential in distributed networks to realize real time event management and event monitoring. Redundant information in the events reported at the same time from multiple sensors can be removed to save energy using synchronization clocks. Synchronization clocks can be used to activate the sleeping sensor nodes at the scheduled time and make use of Time Division Multiple Access (TDMA) schemes to improve the overall throughput of wireless sensor networks. In standalone computer applications, precise clock board/ radio clock that receives time reference transmitted from radio stations administered by National Institute of Standards and Technology (NIST) can be used to improve the accuracy of computer time. Global Positioning System (GPS) can be used to

synchronize hardware clocks with satellites. Both the above methods are costly. For networked computers, Network Time Protocol (NTP) is used to synchronize computer clocks in a hierarchical way. But its heavy weight implementation cannot be supported by the sensor nodes. Post factor is a simple method discussed by Elson and Estrin (2001) to synchronize clocks in a local neighborhood of sensor nodes. Nodes are initially unsynchronized. When a stimulus arrives, each node records the receiving time using its local clock. Immediately afterwards, a beacon covering the whole area, broadcasts a synchronizing signal to all the nodes in the neighborhood. With respect to the time reference, receiving nodes correct their stimulus timestamps. Communication range of the beacon is the crucial limit in this algorithm. RBM derived from Post factor, is proposed by Elson et al., (2002). It keeps the time of the neighboring nodes synchronized. One node periodically broadcasts reference beacons without explicit time stamps, to its neighbors. Receivers use beacon arrival time as reference, to compare their local clocks by exchanging the beacon receiving time. Thus all the nodes know about the clock offset among each other. Large energy is consumed due to the large number of packet transmissions.

Tiny sync and Mini sync (Sichitiu and Veerarittiphan, 2003) are proposed to keep global time in wireless sensor networks by synchronizing any two nodes in the whole network. A pair of nodes use bidirectional time stamped packet transmissions to estimate the clock offset between them thus making two nodes synchronous. To get synchronized, every pair of nodes should perform two way message exchanges. So a large communication overhead is incurred due to the large traffic.

Another time synchronizing protocol to maintain global time is the Time Sync Protocol for Sensor Networks (TPSN) proposed by Ganeriwal et al., (2003). An idea similar to TPSN is suggested by Quing et al., (2005), but the communication overhead is reduced considerably because it requires only some specific adjuster nodes to do the two way message exchange. Here the time synchronizing algorithm requires the client to follow the server. A sequence of reference packets with timestamps are sent by a node to the receiver. The four delays encountered in the message transmission path are process delay, access delay, propagation delay and receive delay. These delays affect the accuracy of the system algorithm. In the LESSAR algorithm, proposed by Quing et al., (2005), a global time is maintained in wireless sensor networks by organizing the whole network system into levels. Level discovery is performed initially when the network is deployed. Sink which collects information from all nodes forms the root and is assigned level 0. It broadcasts level discovery packet to its neighbors. Nodes receiving the packets are assigned level 1 and broadcast the level discovery packet to the other nodes. One node may as a result, receive many packets but it accepts only the one with the lowest level as its ancestor and takes its value +1 as its own level. Thus broadcasting continues. All the sensor nodes are connected in this hierarchical network topology. When a new node enters, it broadcasts the level request packet to enquire the current level values of its neighbors. From the responses obtained, it selects the smallest one + 1 as its level. On node failure, its children notice this, when its timer of observing keep alive message expires. These nodes broadcast level request packet and redo the level discovery process again. In LESSAR, nodes are synchronized level by level. Each node believes that the clocks in its upper level are accurate than its local clock and synchronize with them. It only accepts time sync packets from the upper level and drops all others from the lower levels. So the whole wireless sensor network follows the clock of the sink. This will be synchronized by GPS/NTP. This method has very low resource consumption and computation complexity.

2.1.5 Power management schemes

To deal with the energy management problem, different power management schemes are discussed here. The most important constraint in all wireless sensor networks is the Energy efficiency problem since they are equipped with limited power sources. So an efficient power management should be adopted. Research is conducted using static approaches to attain power management by making the nodes which are not currently being utilized to go to low power states but this should be decided earlier, in a fixed time schedule and not at run time.

Dynamic Power Management (DPM) is widely used in wireless sensor networks. During run time, dynamic techniques can further improve the reduction in power consumption by selectively shutting down the hardware components. After designing a system, additional power savings can be obtained by Dynamic Power Management. Protocols and algorithms have to be tuned for an application. Embedded operating systems and software become a critical requirement of such networks. Different DPM schemes have been proposed to decrease the power consumption in sensor nodes and battery powered embedded systems (Chuan Lin et al., 2006, Sinha and Chandrakasan, 2001, Chung et al., 1999, A.Z. et al., 2003, IBM and Monta Vista Software, 2002). Sensor node is shut down if no events occur (Benini and Micheli, 1997). Major consumer of energy in a wireless sensor network is the energy communication circuits. So communication should be performed only when needed. DPM should always consider when a node should go to sleep/idle state and how long it should remain there. Sensor nodes communicate using short data packets which have more dominance of startup energy (Akyildig et al., 2002). External events represent the interaction between the sensor node and the environment (Rodrigo et al., 2005). So it must also be considered while reducing the power consumption. Application constraints in different DPM models are described by Chung et al. (1999) and A. Z. et al., (2003). So DPM involves shutting down the sensor node during no event and waking them up when needed. So good energy saving is achieved. But sensors communicate using short data packets. So there is more dominance of startup

energy. Therefore DPM should be carefully implemented. Operation in energy saving mode becomes energy efficient only if the time spent in that mode is greater than a decided Threshold. The common DPM policies are the Predictive policy and the Stochastic policy.

Predictive policy:

Predictive policy involves turning OFF the system components if the idle time is greater than or equal to the Timeout Threshold. The assumption is that it may remain idle for a long time. Idle time is predicted by Raghunathan et al., (2004) using the exponential average method. Operating system based direct management techniques are proposed by Sinha and Chandrakasan (2001).

Stochastic policy:

Stochastic policy is given by Benini et al., (1999). System is provided with a service provider, a service requester (both represented by Markov processes), a power manager and a request queue. The power manager represents the device state of operation by issuing proper commands to the service provider. Energy efficient DPM is proposed by Chuan et al., (2006). It uses a modified sleep state policy combined with Optimal Geographical Density Control (OGDC) (Zhang and Hou, 2004), so as to keep a minimum number of sensor nodes in the active mode. So the network lifetime is prolonged. Power aware sensor model is proposed which describes the power consumption in different levels of node sleep states. There can be many sleep states for a node with many components. Every node has a latency to transition to that state. Every sleep state is characterized by power consumption and latency overhead. If a node is in a deeper sleep state, lesser power is consumed and more latency is needed to awaken it. DPM should consider the energy consumption needed for awakening the node back to the active state and how long it remains idle. Saved energy should always be greater than the expended transition energy. Simulation results show that DPM combined with OGDC prolong the network lifetime than with only DPM. The energy and extra time needed to awaken the node are not considered by Sinha and Chandrakasan (2004). In deep sleep state, the sensor cannot detect any event or receive message from the remaining nodes. In clustering protocol, the cluster head should never enter the sleep state. The possible ways to avoid event missing are not considered by Chuan et al., (2006). Another problem with OGDC is that each node should have its positional information property of the wireless channel is taken into account by Xiao-Hui and Yu-Kwong (2006). This had been neglected in most existing energy saving schemes. Neglecting the effects of varying channel quality, leads to the loss of precious battery resources which in turn leads to the depletion of sensor energy. In effect, at a later instant of time many nodes become dysfunctional which ultimately leads to the partitioning of the network. A Channel Adaptive energy management protocol is proposed here to consider the time varying property of the wireless link. Each node can intelligently access the wireless medium according to the current link quality and the predicted traffic load to produce efficient utilization of energy. Results indicate a 40% increase in energy savings compared to the other protocols without channel adaptation. Quality of a wireless link is a time varying function. So the management of energy resources is crucial to prolong the network lifetime. Energy aware packet scheduling schemes for sensor networks are proposed in a channel fluctuating environment. During the situations of poor channel quality, the packets get buffered until the channel quality recovers to the required Threshold. A network system is proposed in which each sensor can decide the state of the communication equipment (idle/active/sleep) with respect to the current channel condition. A fair scheduling and queuing algorithm is designed, in order to avoid the communication Latency and buffer overflow. Thus an optimum balance between the Energy efficiency and Fairness is attained. CAEM (Channel Adaptive approach to Energy Management) is a cluster based hierarchy in which the nodes are assumed to be static or of low mobility. Adaptive Physical layer design ABICM, proposed by Y. K. Kwok et al., (2002), was adopted in which variable throughput modulator and channel coding are used. When Channel State Information (CSI) is available at the transmitter, it does burst by burst throughput adaptation with respect to the CSI (Cianca et al., 2002), i.e., when CSI indicates a very good quality channel, the transmitter performs high order modulation and appropriate error protection to protect the packet transmission. In CAEM, real time monitoring of the change in the CSI of the wireless link is done for all the sensor nodes. Simplicity of the traffic mode (from sensor to sink) leads to the simplification of the design for MAC layer management. Here sensor nodes are equipped with two radios: a tone radio and a data radio, working at different frequencies. If no data is to be transmitted, both radios are turned OFF. If a sensor has packets to send, it turns ON the tone radio and senses the channel whether it is free or not. If sensed negative, (i.e. receives other than idle tones from the channel head) it keeps monitoring the tone channel. If it senses the data channel to be free, (i.e. receives idle tone pulses), it measures the received tone signal strength and further checks whether it is above the required SNR measurement. If not, it continues monitoring the tone channel; otherwise it backs off for a random period of time. After back off time, the sensor checks whether the channel is free and whether the quality requirement is satisfied. If both are found positive, the sensor turns ON the data radio and transmits the buffered packets. If either of the two cases is not found positive, the sensor returns to the sensing state and again monitors the channel. During collision, the channel head sends collision tone pulses and notifies all the sensor nodes. During data packet transmission, the sensor node should keep its tone radio ON and on receiving collision tone pulses; it stops packet transmission by turning OFF the data radio and returns to the sensing state. In CAEM, CSMA/CD is used to detect collision thus reducing the energy wasted in packet collisions. Simulations prove that the behavior of the wireless channel can influence the energy consumption.

Life of a sensor network is highly influenced by the power consumption at each sensor node. Longer network lifetime results when efficient power energy efficient communication process at the hardware and system levels (Perillo and Heinzelman, 2003) and sensor node operating system (Hill et al., 2000). Dynamic Voltage Scaling (DVS) is proposed by Sinha and Chandrakasan (2001) and Calhoun and Chandrakasan (2004). Dynamic Voltage and Frequency scaling is suggested in IBM and Monta Vista Software (2002). Sleep state and Active power management has been dealt by Sinha and Chandrakasan (2001) and Brock and Rajamani (2003). Sentry based power management is proposed by Hui et al., (2003). DPM shuts down components when not used and wakes them up when necessary (Lu et al., 2000).

Sensing coverage is a very important issue in wireless sensor networks. Several centralized and distributed algorithms are proposed for coverage sensing by Ye et al., (2002), Li et al., (2003), Slijepcevic and Potkonjak (2001) and Tian and Georgana (2002). The problem of finding a maximal number of covers in a sensor network is addressed by Slijepcevic and Potkonjak (2001). Different communication models exist in a sensor network. They are direct transmission to base station, multi hop and clustering. In direct transmission, the sensed packets should be transmitted from the sensed node to the base station in a single hop. This single hop transmission is costly and if nodes are far away from the sink, a large amount of transmission energy is required, leading to enhanced energy consumption. In multi hop transmission, transmission occurs over a short communication radius, leading to an energy saving. Data aggregation will lead to enhanced energy saving. Clustering is considered more energy efficient than the other transmission schemes. Communication energy is expensive compared to computation energy. In clustering, the cluster head aggregates and transmits the data to the base station management techniques are adopted. Different methods are discussed to design an energy efficient communication process at the hardware and system levels (Perillo and Heinzelman, 2003) and sensor node operating system (Hill et al., 2000). Dynamic Voltage Scaling (DVS) is proposed by Sinha and Chandrakasan (2001) and Calhoun and Chandrakasan (2004). Dynamic Voltage and Frequency scaling is suggested in IBM and Monta Vista Software (2002). Sleep state and Active power management has been dealt by Sinha and Chandrakasan (2001) and Brock and Rajamani (2003). Sentry based power management is proposed by Hui et al., (2003). DPM shuts down components when not used and wakes them up when necessary (Lu et al., 2000).

Sensing coverage is a very important issue in wireless sensor networks. Several centralized and distributed algorithms are proposed for coverage sensing by Ye et al., (2002), Li et al., (2003), Slijepcevic and Potkonjak (2001) and Tian and Georgana (2002). The problem of finding a maximal number of covers in a sensor network is addressed by Slijepcevic and Potkonjak (2001). Different communication models exist in a sensor network. They are direct transmission to base station, multi hop and clustering. In direct transmission, the sensed packets should be transmitted from the sensed node to the base station in a single hop. This single hop transmission is costly and if nodes are far away from the sink, a large amount of transmission energy is required, leading to enhanced energy consumption. In multi hop transmission, transmission occurs over a short communication radius, leading to an energy saving. Data aggregation will lead to enhanced energy saving. Clustering is considered more energy efficient than the other transmission schemes. Communication energy is expensive compared to computation energy. In clustering, the cluster head aggregates and transmits the data to the base station.

2.2 DYNAMIC POWER MANAGEMENT SCHEMES

Dynamic Power Management is widely used in wireless sensor networks to deal with the Energy efficiency problem. DPM schemes are classified as follows.

2.2.1 Based on Topology management aspect

Radio is the main energy consumer in a sensor node (Raghunathan et al., 2002, Sohrabi et al., 2000 and Estrin and Govindan, 1999). To reduce the energy, the best way is to turn the radio OFF (Raghunathan et al., 2002). Topology management schemes coordinate which nodes turn the radio OFF and when, so as to maintain traffic forwarding satisfactorily and with minimum energy consumption. When an event occurs, the data should be forwarded to the sink. So the network now transitions to the active transfer state. Separate paging channel is proposed to do this wakeup (Guo et al., 2001). This assumes the listen mode of the paging radio to be of ultra low power. An algorithm that uses low duty cycle radio is proposed by McGlynn and Borbesh (2001). Channel access and node wakeup are integrated together in SMAC (Ye et al., 2002). A new topology management scheme STEM is proposed by Curt et al., (2002). Energy consumption in the monitoring state is reduced to a bare minimum along with ensuring a sufficient latency for transitioning to the transfer state. STEM offers trading of energy for Latency. It is found to be more energy efficient than SMAC while assuming a timely transitioning to the transfer state. In the monitoring state, when there is no traffic to forward, only node’s sensors and some preprocessing circuits are ON. This is sleep state. When a possible event is detected, it goes to ON state. On receiving a wakeup message, it turns ON the primary radio which deals with regular data transmission. Multi hop routing is adopted. In the monitoring state, instead of the complete asleep state, it goes to low power listen mode. The initiator node now polls the node and thus wakes up the

target node. Now the links between the two nodes are activated. Data gets transferred using the MAC protocol.

2.2.2 Based on the real characteristics of electronic components

Dynamic Power Management for handheld devices which integrates predictive shutdown and non stationary stochastic concepts is proposed by Hung Cheng Shih and Kuochen (2006). Electronic components can have several inactive and active states. The real characteristics of the electronic components are discussed in three situations. Transitioning of states is to save energy. System can transition from a high power consumption state to a low power consumption state if the time in that state is long enough to compensate for the extra energy consumed by the state transition. Time in low power state should be calculated in advance. Adaptive Hybrid Dynamic Power Management (AH-DPM) is compared with Oracle algorithm (optimal since it is aware of all requests issued), Adaptive Timeout algorithms (Douglis et al., 1995) and Predictive shutdown algorithm (Hac, 2003). The energy consumed in AH-DPM is proved to be very close to the Oracle algorithm. Compared with the classical DPM algorithms, power consumption of AH-DPM is found to be 17.45% less than Adaptive Timeout algorithm (Douglis et al., 1995) and 20.93% less than predictive shutdown algorithm (Hac, 2003). State Transition Matrix in AH-DPM has to be trained. Maximum Likelihood Estimation

is used to train the state transition matrix. The time complexity and space complexity are found to be higher.

2.2.3 Operating system directed power management aspect

The success in Dynamic Power Management involves implementing the correct policy for sleep state transitioning. In operating system directed DPM suggested by Sinha and Chandrakasan (2001), a power aware sensor node model describes the power consumption in different levels of sleep states. Nodes in deep sleep state may have minimum power consumption but a higher energy cost to awaken. This model is quite similar to the system power model in APCI (Advanced Configuration and Power Interface) (Banbyopadhyay and Coyle, 2003). Energy efficient DPM in wireless sensor networks is suggested in Curt et al., (2002) where the modified sleep policy developed by Sinha and Chandrakasan (2001) is combined with OGDC (Zhang and Hou, 2004), to keep a minimum number of nodes in active mode. The state of computation should also be considered when a system turns components ON/OFF to reduce energy. Another issue considered is density

control which ensures only a subset of nodes to be in active mode fulfilling coverage and connectivity. The system models and algorithms developed by Sinha and Chandrakasan (2001) are modified, considering the battery status. New energy efficient DPM combined with OGDC is proposed (Zhang and Hou, 2004). All sleep states of a node may not be useful (Zuquim et al., 2003). The dynamics of sensing coverage versus time is simulated and it is proved that OGDC provides over 95% coverage. DPM with OGDC is found to prolong the network lifetime than that with only DPM. Latency is not analyzed and the problem of avoiding event missing in node sleep states has not been addressed here.

2.2.4 Application driven approach

Several DPM approaches have contributed to decreasing the energy consumption but few consider the application constraints. An application driven power management approach is proposed by Rodrigo et al., (2005). The work by A.Z. et al., (2003) influenced the development of this work. In this, the state of an embedded program in power state machine formulation is included to adapt the QoS in communication intensive devices so as to ensure low power consumption in an embedded system. Theory of Hybrid Automata from Henzinger (1996) is adopted for a fire detection scenario. Application driven DPM is compared with Ideal DPM & Naïve model where no DPM is used. In Naive, the total energy spent is constant. Higher the fire probability, higher is the power consumption for

Application driven DPM & Ideal DPM models. The similarity in performance proved for Ideal DPM & Application Driven DPM proves the better performance of Application Driven DPM. Multi hop operation is not considered.

2.2.5 Link Quality Aspect

As stated earlier in section 2.1.5, to realize the energy saving in a wireless scenario, the time varying property of wireless channels should be considered because it may lead to the loss of battery resources, if channel conditions are poor. So each sensor node in the network should intelligently access the current link quality and the predicted traffic load, so as to improve the energy utilization. A Channel Adaptive energy management protocol is proposed by Xiao-Hui Lin and Yu-Kwong (2006). Each node in the network can intelligently access the current link quality and predicted traffic load to improve the energy utilization. Cluster based hierarchy is assumed and a fair scheduling and queuing algorithm is designed to avoid buffer overflow and communication Latency. An adaptive

Physical layer design ABICM proposed by Kwok and Lau (2002) is adopted. CSMA/CD is used to detect collision. Results prove a 40% increase in energy savings, compared to the other protocols.

DUTY CYCLE APPROACH ON ENERGY CONSERVATION

IN WIRELESS SENSOR NETWORKS

The main energy conserving method is to put the radio transceiver in low power mode when communication is not needed. So there is a constant switching between sleep and active states. This is duty cycling. Duty cycle represents the fraction of time the nodes are active. The sleep/ wake up times should be coordinated since sensor nodes should

always process and send the data to the sink. A sleep/ wake up scheduling algorithm should accompany any duty cycling scheme which is a distributed algorithm based on which sensor nodes decide when to transit from active to sleep state.

Duty cycling can be attained by two different approaches. In the 1st set up, a minimum subset of nodes is adaptively selected to remain active so as to maintain connectivity thus resulting in less energy consumption. The other nodes can go to the sleep state. Finding the correct set of nodes is Topology control. It manages to prolong the network life time (Ganeshan et al., 2004, Mainwaring et al., 2002, Warrier et al., 2007). Nodes selected for the above scheme need not maintain their radio ON continuously. So they keep switching between sleep and wake up states forming power management.

Efficient sharing of the communication resources among the sensor nodes is the prime concern in the design of an efficient MAC protocol. Power consumption in MAC protocol in wireless sensor networks is due to collision, overhearing transmission from neighbors, control packet transmission and idle listening (Bahareh and Hooman, 2008). MAC protocols are broadly classified as schedule based and contention based, based on the resource sharing mechanism. Schedule based schemes owing to its ability of power conservation satisfy the requirements of wireless sensor networks. Contention based schemes implement periodic ON and OFF timing of radio using a contention window. This leads to poor performance due to high contention because of high overhead in resolving contention and collision but is simple and has good bandwidth utilization (Ioannis et al., 2008 and Hull et al., 2004). Improved power consumption and higher end to end delay is obtained by implementing low duty cycle schemes (Ioannis et al.,

2008 and Edgar and Callaway, 2003) Power management protocols These are sleep/ wake up protocols and do not depend on topology or connecting aspects.

(A) Independent sleep/ wake up schemes on top of MAC This is classified as On demand schemes and Scheduled schemes by Armstrong (2005).

(1) On demand schemes A node should wake up to receive a packet from a neighboring node to reduce energy consumption. In a fire detection scenario (Rodrigo et al., 2005),

sensor nodes are in a monitoring state most of the time. Nodes transit to transfer state on receiving an event, thus ensuring reduced energy consumption, in the monitoring state. Two different channels, a data channel for data communication and a wake up channel for waking up nodes are used. Most schemes propose to have two different radios thus reducing the wake up latency but the cost factor is high. Sparse Topology and Energy Management protocol (STEM) is proposed by Schurgers et al., (2002).

Usually topology control protocols are combined with STEM to have a reduction in energy consumption. Combination of STEM-B and GAF is also suggested by Schurgers et al., (2002), to reduce the energy consumption in a sensor network without topology control or power management. Pipelined Tone Wake up (PTW) provides a tradeoff between energy saving and wake up latency (Yang and Vaidya, 2004). It also uses two hannels like STEM, but tone detection is done in

the sender unlike STEM. Senders send a wake up tone on detecting an event and receivers wake up periodically. The performance of the PTW scheme is much

better in terms of energy consumption and latency than STEM (Yang and Vaidya, 2004). STEM and PTW use an asynchronized sleep/ wake up to enable a duty cycle on wake up radio. In different approaches (Guo et al., 2001, Rabaey et al., 2002, Nosovich and Todd, 2000 and Shih et al., 2002), the low power is continuously in standby mode and on receiving a signal, data radio is woken up. The transmission range of wake up radio is smaller than the data radio but not negligible, in the low power mode. To get rid of these problems, Radio Triggered Power Management scheme is proposed by Gu and Stankovic (2005). Energy in wake up message is used to trigger the activation of the sensor node. The limitation of the scheme is the limitation on the maximum distance from which the wake up message can be sent.

(2) Scheduled scheme

All neighbors should wake up at the same time. Usually all nodes wake up periodically to check for any communication and then return to sleep. The advantage of this scheme is that when a sensor node is awake all nodes will be awake. This facilitates messages to be transmitted to all neighbors (Armstrong, 2005). But synchronization is essential. A detailed research to achieve a good synchronization effect is the prime need. A fully synchronized pattern is proposed by Keshavarzian et al., (2006). All the nodes in the network wake up at the same time according to a periodic pattern. Such fully synchronized protocols are proposed in SMAC (Ye et al., 2004) and Time out MAC (TMAC) (Dam and Langendoen, 2003). In staggered wake up pattern (Keshavarzian et al., 2006), nodes at different levels wake up at the same time. Staggered wake up pattern is used in DMAC (Lu et al., 2004). In the staggering scheme, the nodes at different levels of the data gathering tree wake up at different times. So during active time only a subset of the nodes is active resulting in a lesser number of collisions. But nodes at the same level wake up at the same time producing small amount of collisions. The fixed active/wake up approach accounts for the limited flexibility. Active period should be less.

Adaptive and low latency staggered scheme is proposed by Anastasi et al., (2006). The length of the active period is set in accordance with the network activity resulting in reduced energy consumption. A new approach derived from on demand TDMA, FPS (Flexible Power Scheduling) is proposed by Hohlt et al., (2004). Slots are arranged to form a periodic cycle. A node keeps its radio ON only if it has something to receive/ transmit. But this scheme also suffers from lack of flexibility.

(B) MAC protocols with low duty cycle

The energy issues in wireless sensor networks can be dealt with considering the MAC protocols. This is usually by having a low duty cycle MAC scheme so as to achieve an efficient power management. Energy conservation is attained by turning OFF radios when not needed thus trading off network delay for energy conservation (Suh and Ko, 2005 and Shah and Khannad, 2005). Most of the existing MAC protocols fall in two categories as TDMA based and Contention based (Li De-liang Peng Fei, 2009).

(1) Schedule based

Schedule based protocols divide time into several time slots using TDMA and channel access is done on a slot by slot basis (Ching et al., 2004, Rhee et al., 2005 and Chen and Khokhar, 2004). Nodes turn ON their radio during their own slots which hence result in low energy consumption. Clustering schemes can also reduce energy consumption since each node need not directly communicate with the sink. This is used in LEACH (Heizelman et al., 2000) and energy aware TDMA based MAC schemes.

Traffic Adaptive Medium Access protocol (TRAMA) proposed by

Rajendran et al., (2003) is an energy efficient TDMA based protocol for wireless sensor networks. The time is divided into random access period for slot reservation and scheduled access period for slot access. The two hop neighborhood information from nodes establishes the collision free schedule. A slot is assigned to a node using the election procedure. Good Energy efficiency is achieved in Light weight MAC (LMAC) [Van Hoesel and Havinga (2004)] and it considers the Physical layer properties. The

radio state transactions and protocol overhead are reduced. But specifying the fixed time length of the frame prior to deployment is the major drawback. In order to rectify these effects, Adaptive Information centric LMAC is proposed (Chatterjea et al., 2004).

(2) Contention based MAC

MAC protocols used in wireless sensor networks are contention based. They are categorized as synchronous MAC schemes, asynchronous MAC schemes and hybrid MAC schemes.

(i) Synchronous scheme

The most popular synchronous MAC scheme is BMAC (Polastre et al., 2004 and Chaari and Kamoun, 2010). It has a very energy efficient channel access mechanism by implementing a back off scheme, channel estimation, optional ACK and for a low duty cycle an asynchronous sleep/wake up scheme based on periodic listening. Fixed wake up time and no need of synchronization are the features.

SMAC (Edgar and Callaway, 2003, Ye et al., 2004 and Wei Ye et al., 2002) is one of the well known energy efficient protocols for wireless sensor networks. It is a contention based random access protocol with a preset listen/sleep cycle and uses a synchronized sleep mechanism. For the purpose of announcement and synchronization for the subsequent data transmission, synchronizing (SYN) and Ready To Send / Clear To Send (RTS/CTS) control packets are sent during the listen period based on the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism. Any two nodes exchanging RTS/CTS packets in the listen period require being in the active state and entering the data transmission phase without entering the sleep state. To avoid the energy wastage due to idle listening, all the other nodes enter the sleep state. The duration of a listen period is always fixed in SMAC. This results in redundant energy wastage.

Traffic aware Energy Efficient MAC (TEEM) (Suh and Ko, 2005 and Chaari and Kamoun, 2010) reduces the power consumed in the listening period of SMAC. Listening period is shortened by rearranging the structure of the listening period to save power. But in SMAC and TEEM, all nodes have to wake up to sense the signals during the contention period. DSMAC (Lin et al., 2004) and UMAC (Yang et al., 2005) help to decrease the Latency for delay sensitive applications compared to SMAC.

SMAC does not work well when the traffic load fluctuates. FTMAC (Dam and Langendoen, 2003 and Shah and Khannad, 2005) resolves the problem of SMAC by following an aggressive power conserving policy. It is a variation of SMAC with an adaptive length of the active state by a fine time out. Nodes can go to an early sleep state resulting in an augmented Latency and a lesser Throughput. Data gathering MAC (DMAC) (Lu et al., 2004) a different protocol using an adaptive duty cycle, gives a low node to sink latency by staggering the wake up times of the nodes in a converge-cast tree. DMAC outperforms SMAC in having a lower Latency and higher Energy efficiency.

(ii) Asynchronous scheme

Distributed mediation device (DMD) (Edgar and Callaway, 2003) causes neighbor nodes not to have the same schedule as each other. This results in an increase in transmission Latency and power consumption. MDMD (Chen and Khokhar, 2004, Shah and Khannad, 2005 and Ching-Wen Chen et al., 2008) rectifies all the problems existing in DMD.

(iii) Hybrid MAC protocols

A new hybrid MAC scheme (ZMAC) (Rhee et al., 2005) is used for sensor networks to merge the strengths of TDMA and CSMA. Its adaptability to the level of contention in the network is the major aspect of ZMAC. Under low contention, it acts like CSMA and under high contention like TDMA. It is robust to the topology changes and time synchronization failures regularly occurring in wireless sensor networks.