中國機械工程學刊第三十卷第一期第59~65頁(民國九十八年)
Journal of the Chinese Society of Mechanical Engineers, Vol.30, No.1, pp.59~65 (2009)
A Novel Energy-Efficient Adaptive Routing
Protocol for Wireless Sensor Network
Chia-Pang Chen*, Cheng-Long Chuang**, Chwan-Lu Tseng***,
En-Cheng Yang****, Maw-Yang Liu***** and Joe-Air Jiang******
Keywords：wireless sensor network, energy-efficient,
routing algorithm.
ABSTRACT
Energy efficiency is one of major issues which must be improved to prolong the lifetime of wireless sensor networks (WSNs). Considering the maintenance of Quality of Service (QoS) and energy consumption, we proposed a novel energy-efficient adaptive routing protocol (EEARP) for WSNs. EEARP is composed by a gradient-based topology generator (GBTG) and a novel energy-efficient adaptive routing algorithm (EEARA). After the sensor nodes were deployed broadly, the topology is constructed by GBTG, and then the information of resulted links among the nodes is provided to EEARA. In order to minimize power consumption and transmission delay, the EEARA with load-balance capability is utilized for selecting efficient path toward the destination sink (the node that is the data storage center). Two simulation scenarios were considered in this study. The simulation results show that the proposed protocol can balance the burden of all nodes effectively and significantly reduce the overall energy consumption.
INTRODUCTION
Nowadays, wireless sensor network (WSN) is one of the most important topics in the field of computer network. There were many applications using WSN as their fundamental communication platform, such as traffic control (Wang and Howitt, 2005), industrial security (Wang et al., 2008), agricultural monitoring (Pierce and Elliott, 2008), military information integration (Diamond et al., 2007; Salmanian, 2003), etc. A WSN platform usually is formed by numerous wireless sensor devices (also referred wireless sensor nodes). Each sensor node consists of micro-sensors, low-power microprocessor, and miniature radio communication device. The task of the sensor nodes in WSNs is to gather information about physical or environmental conditions, and transmit the gathered data to a predetermined node (i.e., sink) to collect all data measured throughout the entire WSN.
One of the major issues in WSNs is to develop an energy-efficient routing protocol and to minimize the energy consumption, since WSN suffers from several restrictions including limited energy supply, small memory, and insufficient communication capability. Several routing algorithms and protocols have been proposed for preserving energy and prolonging the lifetime in some literatures (Tang et al., 2007; Mamun-or-Rashid et al., 2007). The routing algorithms can be classified into three types: flat routing, hierarchical routing, and adaptive routing. Flat routing (e.g. flooding approach) is simple and easy to implement without extra cost for topology maintenance and packet routing. However, it is not an energy-efficient approach, and is not suitable for WSN. To solve the problem, a directed diffusion (DD) approach was proposed (Intanagonwiwat et al., 2003). The DD approach has several strategies, which are on-demand data querying by sink, data aggregation, and path setup mechanisms, for energy saving. Some gradient-based routing (GBR) approaches have also been applied to route packets in WSN (Han et al., 2004; Verbist et al., 2006). The measured data were transmitted according to the direction of descending gradient back to the sink. Besides, in order to
Paper Received July, 2008. Revised October, 2008. Accepted January, 2009. Author for Correspondence: Joe-Air Jiang.
* Ph.D. student, Department of Bio-Industrial Mechatronics
Engineering, National Taiwan University, Taipei 106, Taiwan.
** Ph.D. student, Department of Bio-Industrial Mechatronics
Engineering, National Taiwan University, Taipei 106, Taiwan.
*** Associate Professor, Department of Electrical Engineering,
National Taipei University of Technology, Taipei 106, Taiwan.
**** Associate Professor, Department of Entomology, National
Taiwan University, Taipei 106, Taiwan.
***** Assistant Professor, Department of Electrical engineering,
National Ilan University, Ilan 260, Taiwan.
****** Professor, Department of Bio-Industrial Mechatronics
Engineering, National Taiwan University, Taipei 106, Taiwan.
J. CSME Vol.30, No.1 (2009) minimize the energy consumption some hierarchical
protocols such as LEACH (Heinzelman et al., 2000) and PEGASIS (Lindsey and Raghavendra, 2002) were proposed to form clusters with cluster heads.
In this study, a novel energy-efficient adaptive routing protocol in network layer is proposed to initialize a given WSN and route packets to its sink. Due to everlastingly changing environment, our proposed routing algorithm can conquer the load-balancing problem mentioned above. Generally, each packet has multiple paths to reach its destination node. The task of our proposed routing algorithm is to select the best path with minimum energy consumption to route the packet without generating any hot-spot. The hot-spot is a node which is fully utilized, and it causes serious transmission delay for packets. This is usually happened on the nodes adjacent to the sink. Evidently, selecting the best path to forward packets to destination is very important for WSNs. Nevertheless, path selection is not needed only in WSNs but also in other types of engineering applications, e.g. Chen et al. (2008) proposed artificial immune system to determine the optimal path for the pile setting out. They have the common objective for improving the performance.
The rest of this paper is organized as follows: The second section explains the sensor network model adopted in this work. A novel energy-efficient adaptive routing algorithm for WSN is presented in the third section. Simulation results of applying the proposed approach to a WSN sample are demonstrated in the fourth section. Conclusions are given in the last section.
PRELIMINARIES
To develop the proposed EEARP, we consider a sensor network model similar to that used in the work of Schurgers and Srivastava (2001). There are some reasonable assumptions in this sensor network model: z All sensor nodes are homogeneous and
stationary. Each node has a nonrenewable energy budget. The functions of the node will shut down and cease to operate when its on-board energy supply is exhausted.
z A sink in this network can aggregate all sensing data from sensor nodes, which sense the environment periodically.
z When some nodes detect the events, the sensing data are transmitted toward the sink by the intermediate relay nodes.
ENERGY EFFICIENT ADAPTIVE
ROUTING PROTOCOL
Gradient-Based Topology Generator (GBTG)The main purpose of WSN is to measure and
collect physical or environmental information to the base station (or sink) in a large-scale area. Wireless sensor nodes are deployed to the positions of interest. After these sensor nodes have been deployed, a network topology must be constructed to form the data links into an accessible network. Gradient-based routing approaches (GBRAs) were usually utilized to help constructing the topology of a WSN (Han et al., 2004; Verbist et al., 2006; Cheng and Jia, 2005). We adopted part of the gradient flooding approach to set gradient value of each sensor subject to a predetermined sink (Han et al., 2004). Therefore, we can delaminate all of the nodes in WSN by broadcasting the initiation (INIT) packets, which are issued from sink. The INIT packet contains a gradient value defaulted as 0, and every node has an infinite gradient value initially. As a node receives the INIT packet first, it sets its gradient value for the received gradient value plus 1 and then broadcasts the updated INIT packet with the new gradient value. The process continues until all nodes receive the INIT packet at least once. As long as the node receives any duplicate INIT packet, the redundant packet must be discarded without being further broadcasted. The links in WSN can be made according to the pseudo code as follows:
Initialize total Nodes
A sink (regarded as a Node) starts broadcasting the INIT packet.
While each Node M receives the INIT packet from
Node N
If M has received the INIT packet previously
{
Discard the INIT packet and not broadcast.
Break
}Else{
Extract the gradient value n from the received INIT packet.
Set n = n + 1 as the gradient value of M. The link EMN between M and N is made.
M broadcasts the INIT packet with its gradient value.
}
getNeighborLink(M);
/* The link EMF can be established between M and the
neighboring Node F which has the same n. F∈{the set of all the neighboring nodes of node M} */ selectBestNeighbor(M);
/* Select the best link from all the EMF according to
received signal strength indication (RSSI) */
EndWhile
Energy-Efficient Adaptive Routing Algorithm (EEARA)
The primary objective of this study is to maintain the data packet flow in the wireless sensor network unobstructed. The definition of the sensor network model is given as follows.
C.P. Chen et al.: A Novel Energy-Efficient Adaptive Routing Protocol for Wireless Sensor Network.
composed of Γ and Λ, where Γ is the set of nodes and
Λ is the set of links. Due to the feature of multi-hop
transmission, it could have many paths from source node s to destination node d. Therefore, we let Π(s,d)
denote the set of all possible paths starting from s to d. According to these definitions, it is known that Π(s,d)
is the subset of Λ, which is generated by the proposed GBTG. Let π represent a generic path, and πi(s, d)
represents i-th path in a journey from source node s to destination node d. Also, we let Φ(π) be a generic cost function associated to a designated path π. Φ(π) can be the delay time θ(π) for a packet transfers through a path π, or the number of hops ε(π), even a hybrid function combined both of them. For two or more paths possessed the same cost, we consider them as Φ-equivalent. As the above-given definitions, the given network N(Γ, Λ) in fact is a graphical model.
In this study, we present a parallel routing algorithm in which each node is an independent router. Each node exchanges power conditions with its neighboring nodes via network packets, and estimates an optimal route for a given packet by the information at that time. Since each node in the WSN functions as a router that works independently, several connection matrices are defined and stored in each node to help performing the energy efficient routing function:
Connection matrix is denoted as Ts = [ts,d],
where Ts is an array that registers all outgoing links
of node s, and ts,d determines whether node s has an
outgoing link connected to node d or not, 1 represents connected, 0 represents disconnected. Let Bs,d denotes
the bandwidth utility ratio of link from node s to node
d. ps,df that determines the probability table for
deciding the next node of a packet transferring from node s to its final destination node df,. ps,df,j
corresponds to the probability for the packet transferred to node j. Note that Σj(ps,df,j) = 1. Also,
suppose that Efull is the initial energy on each node
and Ej is the remaining energy on node j. With the
above-given notations, the weight of choosing node j as the next node for transferring the packet while the packet is queued at node s is
1 2 3
, , full
( , , ) s j [ , f, ]C [1 s j]C [ j/ ]C ,
w s j df =t ⋅ ps d j ⋅ −B ⋅ E E (1)
where C1, C2, and C3 are weighting factors that
regulate the importance of probability matrix (ps,df),
bandwidth utility ratio (Bs,j), and remaining energy
ratio (Ej/Efull) during routing process, respectively.
Equation (1) functions like a firing strength for the routing algorithm to send a packet from node s to node j. The values of C1, C2, and C3 are set to 1 in
most cases. Supervisors can obtain desired performance by adjusting these parameters. For example, if load-balancing on each link is more important than other factors, C2 should be set to a
value larger than 1. Also, if power preservation on
each node is the highest concern, C3 should be set to a
value larger than 1. There is no systematic approach to obtain the optimal values of these parameters since it involves too many human factors (such as the aforementioned performance of interest). Hence, generally, we set the values of C1, C2, and C3 to 1.
The probability of choosing node j as next node of the packet can be defined as follow
, ( , ,( , , )). h s j w s j d fw s h d f σ ∀∑ =
(2)
The load-balance feature of the proposed EEARA can be exploited using Equation (2). Eq. (2) is a normalized firing strength to send a packet from node
s to node j. This equation contains the terms of
bandwidth utility ratio (Bs,j) and remaining energy
ratio (Ej/Efull). The higher bandwidth utility ratio it is,
the lower visibility of ps,df,j will be. Likewise, the
lower remaining energy on node j, the lower visibility of ps,df,j will be. Therefore, the proposed EEARA is
capable of preventing from sending more packets through hot-spot or low energy nodes. In our investigation, once the packets reaches to its final destination node through a path πi(s, d), the cost
function of the path can be defined as follow
Φ( (πi s, d)) θ( (= πi s, d)). (3)
The possibility of selecting a path πi(s, d) to transfer a
packet from node s to node j is updated by the following equation if ( , ) ( ) Φ( ( )) , , , 0 otherwise f i i Q s j s, d s, j s d j p ⎧ ⊂ ⎪ = ⎨ ⎪⎩ Δ π π (4)
where Q, an awarding amount, is constant and can be any value. Therefore, we can update the probability matrix ps,df according to the following equation
, f, , f, , f, , ( , , ),
s d j s d j s d j f
p = ⋅ρ p + Δp ∀ s d j (5) where ρ is the learning rate. In order to keep the probability matrix ps,df normalized, every time after
ps,df has been updated by Equation (5). We have to
perform normalization on ps,df using
, , , , , , . f f f s d j s d j s d h h p p p ∀ = ∑ (6)
These parameter updating procedures will be executed once a packet reaches its destination node.
J. CSME Vol.30, No.1 (2009) 0 50 100 150 200 250 0 50 100 150 200 A B D E U Meters Me te rs
Fig. 1. Topology of the WSN constructed by the proposed GBTG algorithm.
SIMULATION RESULTS
Network Properties and InitializationTo compare the performance of our proposed EEARP with previous study, we employ a WSN sample deployed by Schurgers and Srivastava (2001), which is illustrated in Figure. 1. This WSN contains 41 nodes at the same positions, and the user node (U, regarded as sink) is remarked in red color.
The constitution of the WSN is described as below:
(1) Node size = 41, (2) Packet size = 128 bits,
(3) Absolute position of user node = (195, 10), (4) Power consumption for sending a packet = 0.0054
mJ(Schurgers and Srivastava, 2001), and
(5) Initial power of energy source on each node = 0.76 mJ (Schurgers and Srivastava, 2001).
In this WSN, each node is able to send about 140 packets using the limited energy. The network topology constructed by using the proposed topology generator is depicted in Fig. 1. Before we test the performance of the proposed EEARA, several network parameters are predetermined in follows: (1) Maximum transmission speed of links: 250 kbps; (2) Size of data packet: 128 bits;
(3) Packet generation delay: 10 μs; (4) Routing delay: 0.1 ms; and (5) C1 = C2 = C3 = 1.
The initial probability matrices (ps,df) (mentioned
in the third Section) on all nodes are optimized by using prior computer-based routing simulation. To initialize the probability matrices, only the topology generated by the proposed GBTG algorithm is required, we do not need any precise information about the physical position of nodes deployed in the WSN. Therefore, the remaining energy ratio (Ej/Efull)
in Eq. (1) is also ignored since we only need to initialize the probability matrix (ps,df) in prior. In the
computer-based routing simulation, 1800 packets are
generated by each node within 1 second (simulation time), and the target of all of the packets is the user node (i.e., U in Fig. 1). The initial probability matrices (ps,df) of all nodes can achieve its steady
state in 0.25 seconds (simulation time). The initialized probability matrices (ps,df) are used in the
simulation scenarios described in the following subsections.
Simulation – Scenario 1
In this scenario, nodes A and B in Fig. 1 are chosen to measure the environmental parameters. The measurement data are packed into packets, and then sent to the user node at a regular time interval. Nodes A and B simultaneously send new packets to the user node in every 0.119 seconds. Therefore, each node generates 100 packets in 11.8 seconds. Since we gave each node 0.76 mJ of initial energy, no node has drained all power of its energy source. Hence, connectivity of the entire WSN is still fully functional. For comparison, five network traffic-spreading techniques: standard, stochastic, energy based, stochastic energy based, and stream based (Schurgers and Srivastava, 2001), are applied to the same WSN. Note that the topologies of the WSN in these comparison cases are generated by (Schurgers and Srivastava, 2001) except for the one generated by the proposed GBTG.
The detailed information about the five network traffic spreading techniques and the topologies of the WSN used in these comparison cases are referred to the reference (Schurgers and Srivastava, 2001). Besides, we utilize the root mean square ERMS as the
indicator that depicts the averaged energy consumed by each node (the lower value, the better). The evaluations of ERMS for all of the techniques versus
simulation time are shown in Figure. 2. The shown results are averaged from 500 repeat simulations. It is clear that the proposed algorithm works similarly to the other five network traffic spreading techniques in the beginning 6 seconds. When the remaining energy
2 4 6 8 10 12 0 0.05 0.1 0.15 0.2 0.25 Time (second) ERM S (m J) Standard Stochastic Energy based Stochastic energy based Stream based The proposed algorithm
C.P. Chen et al.: A Novel Energy-Efficient Adaptive Routing Protocol for Wireless Sensor Network.
of some nodes in the WSN drops, the energy-efficient feature of the proposed EEARA starts working. The packets generated by node A and B are routed to other nodes that have more remaining energies. Please note that we used ERMS versus simulation time
to demonstrate the performances yielded by the proposed EEARA and other algorithms because it is a more realistic way to show the status of the network varying with time. We supposed that it is more realistic since the simulation is conducted in a dynamic manner. For this reason, the packet generated at a simulation time may affect the path of other packets generated at different times. In addition, the same packet generated by A or B at the same time in replicated simulations may be routed to the user node via different paths. Thus, the conducted experiment is an accumulative system in which the packets generated at time t may affect other packets generated at time t + k or t – k.
After all nodes become inactive, which means that all packets have reached to user node, the histogram of energy consumptions of all nodes is estimated and depicted in Figure. 3. Except for two nodes (node A and B) with 71% energy consuming, our proposed adaptive routing algorithm is able to spread the network traffic, and preserves the energies on the rest of nodes. For example, only four nodes (all of them are connected with node A or B) consumed about 35% of its available energy, the rest 35 nodes still have abundant energy preserved. This is very important because this would also extend the lifetime of the entire network.
Simulation – Scenario 2
Similar to scenario 1, but node D and E are set to be the source nodes in this case. Nodes D and E generate one hundred packets in 11.8 seconds. After nodes D and E become inactivated, the histogram of energy consumptions of all nodes is depicted in Figure. 4. We can see that the power consumption of some nodes increased about 10% comparing with scenario 1. The reason is that nodes D and E are close to each other, and the nodes around the source nodes need to relay more packets comparing with scenario 1. This is because the shortest path for packets generated by nodes D and E pass through the same node. Although there are some alternative paths can be used for the packets to be transmitted to the user node, nodes D and E still share those alternative paths as well. If the adaptive routing algorithm choose to route the packets via more undesirable paths, that would significantly increase the delay time and makes these paths becoming unattractive. Thus, large amount of packets still been routed through shorter paths. If we extend the simulation time, those undesirable paths would become more attractive when some bottleneck nodes have low remaining energies.
In both aforementioned simulation scenarios, if we keep generating new packets from source nodes A
0~10 10~20 20~30 30~40 40~50 50~60 60~70 70~80 80~90 0 5 10 15 20 25 Energy used (%) 90~100
Fig. 3. Histogram of energy consumptions for scenario 1
Fig. 4. Histogram of energy consumptions for scenario 2
and B in scenario 1 (or nodes C and D in scenario 2), the first bottleneck nodes appeared are the source nodes themselves because they consumed all of its available energies. Such circumstances could happen if we extend the duration of simulation to 16.75 seconds. For example, once the source nodes are shut down, no newly generated packets could be added into the network. Hence, all nodes become inactive, and the entire network becomes silent because no traffic on the network.
CONCLUSIONS
A WSN platform is formed by a number of sensor nodes that equipped with wireless communication modules. The issues of interest in WSN are mostly surrounded around two topics, which are (1) to prolong the lifetime of the network, and (2) to deliver packets to its destinations with power- and time-efficiency. However, it is hardly to simultaneously solve both problems because the trade-off between the benefits of both issues. Previous studies on WSN have been many but they were mainly focused on one problem only.
In this study, a novel energy-efficient adaptive routing algorithm for WSN, named EEARP, was proposed. The goals of EEARP can be divided into two parts: (1) GBTG for automatically constructing a network topology for a given number of WSN nodes; and (2) EEARA for routing packets to its destination nodes with minimized delay time, and balance the power usage of all nodes in the entire network.
A simulated WSN platform that consists of 41 nodes is employed to validate the function of GBTG,
J. CSME Vol.30, No.1 (2009) and to evaluate the performance of the proposed
EEARA. The configuration of WSN is defined based on specification of real sensor nodes. Two kinds of scenarios were simulated. The first scenario is that two distanced nodes generate 100 packets in 11.8 seconds, and the second scenario is that two neighboring nodes send 100 packets to sink node in the same time period in the first scenario. The experimental simulations demonstrate that these goals can be achieved by the proposed EEARP. The GBTG is able to generate a nearly balanced network topology without constructing too many redundant links. This is important because redundant links may increase the complexity of network management, and leads to higher power consumption. Also, the EEARA is able to reduce the overall energy consumption. By avoiding large amount of packets been routed to the same relaying nodes, most of nodes can preserve most of their energies. Thus, the life time of the network can be prolonged. The simulations conducted in this study are to investigate the problem of how does the EEARA equalize the traffic over the network in order to prolong the lifetime of WSN. If we keep generating new packets from source nodes A and B (or C and D), the first bottleneck nodes appeared are the source nodes themselves. More comprehensive simulations, such as bottleneck nodes appeared due to large amount of packets generated by super nodes equipped with high-capacity batteries, will be studied in the future.
ACKNOWLEDGMENT
This work was supported in part by the National Science Council, Taiwan, under for financially contracts no.: NSC 95-2218-E-002-073 and NSC 96-2218-E-002-015. The authors would also like to thank the Council of Agriculture of the Executive Yuan, Taiwan, for their financial supporting under contract no.: 97AS-9.1.1-FD-Z1 (3).
REFERENCES
Akyildiz, F., and Su, W., Sankarasubramaniam, Y., and Cayirci, E., “Wireless sensor networks: a survey,” Computer Networks, 38(4), 393-422 (2002).
Chen, S.S. and Lee, K.C., “Path Planning for the Automatic Pile Setting out,” Journal of the
Chinese Society of Mechanical Engineers, 29(1),
83-88 (2008).
Cheng, H., and Jia, X., “An energy efficient routing algorithm for wireless sensor networks,” Proc. of WCNM 2005, Volume 2, 905-910 (2005). Diamond, S.M., and Ceruti, M.G., “Application of
Wireless Sensor Network to Military Information Integration,” Proc. of Industrial
Informatics, 2007 5th IEEE International
Conference on, Volume 1, 317-322 (2007). Han, K.H., Ko, Y.B., and Kim, J.H., “A Novel
Gradient Approach for Efficient Data Dissemination in Wireless Sensor Networks,”
Proc. of VETECF’04, Volume 4, 2979-2983
(2004).
Heinzelman, W., Chandrakasan, A., and Balakrishnan, H., “Energy-Efficient Communication Protocols for Wireless Microsensor Networks (LEACH),”
Proc. of the 33rd Hawaii International Conference on Systems Science, Volume 8,
3005-3014 (2000).
Intanagonwiwat, C., Govindan, R., Estrin, D., Heidemann, J., and Silva, F., “Directed Diffusion for Wireless Sensor Networking,”
IEEE/ACM Transactions on Networking, 11(1),
2-16 (2003).
Lindsey, S. and Raghavendra, C., “PEGASIS: Power-Efficient Gathering in Sensor Information Systems,” Proc. of IEEE Aerospace Conference, Volume 3, 1125-1130 (2002).
Mamun-or-Rashid, Md., Mahbub Alam, M., and Hong, C.S., “Energy Conserving Passive Clustering for Efficient Routing in Wireless Sensor Network,”
Proc. of the 9th International Conference on Advanced Communication Technology, Volume
2, 982-986 (2007).
Pierce, F.J. and Elliott, T.V., “Regional and on-farm wireless sensor networks for agricultural systems in Eastern Washington,” Computers and
Electronics in Agriculture, Volume 61, Issue
1, pp. 32-43 (2008)
Salmanian, M., Military Wireless Network Information
Operation Scenarios, RDC-OTTAWA-TM-
2003-241, Defence R&D Canada-Ottawa (2003).
Schurgers, C. and Srivastava, M. B., “Energy efficient routing in wireless sensor networks,” Proc. of
MILCOM 2001, Volume 1, 357-361 (2001).
Tang, B., Wang, Y., and Zhou, M., “Energy-Balanced Cluster Range Control algorithm for Wireless sensor networks,” Proc. of ICWAPR 2007, Volume 1, 1-6 (2007).
Verbist, F., Festjens, N., Steenhaut, K., and Nowe, A., “Hop count discovery protocol for gradient based routing in wireless sensor networks,” Communications and Electronics, Proc. of ICCE
'06, 102-105 (2006).
Wang, J. and Howitt, I. “Optimal traffic distribution in minimum energy wireless sensor networks,”
Proc. of GLOBECOM '05. IEEE, Volume 6, pp.
3274-3278 (2005)
Wang, Y., Wang, X. Xie, B., Wang, D., and Agrawal, D.P., “Intrusion Detection in Homogeneous and Heterogeneous Wireless Sensor Networks,”
IEEE Transactions on Mobile Computing,
C.P. Chen et al.: A Novel Energy-Efficient Adaptive Routing Protocol for Wireless Sensor Network.
NOMENCLATURE
EMN The link between node M and node N
F The set of all the neighboring nodes of one node
Γ The set of nodes
Λ The set of links
Π(s,d) The set of all possible paths starting from
node s to node d
πi(s, d) The i-th path in a journey from source
node s to destination node d
Φ(π) A generic cost function associated to a designated path π
θ(π) A cost function with respect to delay time ε(π) A cost function with respect to hop count
N(Γ, Λ) A wireless sensor network consists of
Γand Λ
Ts An array that registers all outgoing links
of node s
ts,d A parameter determines whether node s
has an outgoing link connected to node d or not
Bs,d The bandwidth utility ratio of link from
node s to node d
σs, j Normalized firing strength to send a
packet from node s to node j
ps,df The probability table for deciding the next
node of a packet transferring from node s to its final destination node df
ps,df,j The probability for the packet transferred
to node j with destination node df
Efull The initial energy on each node
Ej The remaining energy on node j
C1, C2, C3 The weighting factor regulate the
importance of probability matrix (ps,df),
bandwidth utility ratio (Bs,d ), and
remaining energy ratio (Ej/Efull).
Q A constant awarding amount
ρ A learning rate
