2010 10th International Conference on Hybrid Intelligent Systems
FCCTF: Fairness Congestion Control for a disTrustful wireless sensor network using Fuzzy logic Mani Zarei Amir Masoud Rahmani Razieh Farazkish Science and Research Branch, Science and Research Branch, Science and Research Branch, Islamic Azad University Islamic Azad University Islamic Azad University Tehran, Iran Tehran, Iran Tehran, Iran
[email protected] [email protected] [email protected]
Obviously existing congestion control solutions do not consider trustworthiness evaluation scheme and some diffused useless packets may worsen congestion problem. Moreover, what is missing in existing trust solutions is inattention to packet lost of determiner negotiating nodes due to buffer overflow. Our new scheme gives individual nodes the ability to efficiently estimate their neighbors trust and make appropriate decision locally on forwarding the received packets. Indeed, hop-by-hop flow control alleviates packet lost when congestion occurs. In this paper we present the improved (FCC) [1] and look at fairly flexible in-network data processing: a fair malicious node blocking for congestion control in a distrusted WSN using fuzzy logic. Our new scheme removes the traffic ratio overhead of handling the packets corresponding to corrupt and malfunctioning nodes which are interpreted as malicious nodes. This paper is organized as following: Related works are discussed in section II. Section III presents FCCTF while section IV reports simulation and results and finally section V concludes the paper.
Abstract-- One of the most important challenges in a densely wireless sensor network (WSN) with potential congestions and packet loss is dissemination of distrusted packets. In this paper we present FCCTF: Fairness Congestion Control for a disTrustful WSN using Fuzzy logic. FCCTF increases each node capability for detecting and isolating malicious nodes in order to improve packet delivery while some important packets endanger dropping due to overflowing. Indeed, FCCTF attempts to improve our previous scheme Fuzzy based trust estimation for Congestion Control in WSNs (FCC) [1]. Simulation results show that FCCTF improves packet delivery up to 18.5% more and it reduces the related packet drop of legitimate nodes 20% less than FCC.
Keyword-- WSN, malicious node, threshold trust value, fuzzy logic, Fuzzy Inference System I. INTRODUCTION A sensor network is composed of a large number of nodes, which are densely deployed inside a phenomenon. The position of nodes does not need to be engineered. This allows random deployment in inaccessible terrains meaning that sensor network protocols and algorithms must possess self-organizing capabilities. Another unique feature of sensor networks is the cooperative effort of nodes. Sensor nodes are fitted with an on-board processor. Instead of sending the raw data to the nodes responsible for the fusion, nodes use their processing power to locally carry out simple computations and transmit the required and partially processed data [2]. Indeed, the main effort of the sensor nodes is to report accurate and timely information. Thus safe and correct operation of each node is necessary. Misbehavior of nodes may destroy the function of WSN or decreases the performance. Therefore, sensor nodes are required to have a measurable belief of trust for safe interactions [1]. Resource restrictions like buffer capacity may occur as another problem in WSNs. Multi-hop communication could be beneficial, but unpredictable network situation like congestion problem may destroy these cooperative aim and some packets may drop due to overflowing. Therefore, some measures should be considered in such sensor networks. Sensor nodes are prone to malfunctioning naturally; while remote asset monitoring and control dramatically enhances operation efficiency, use of distrusted data from malfunctioning sensor nodes in critical applications could have disastrous consequences [1].
978-1-4244-7365-6/10/$26.00 ©2010 IEEE
Sara Zahirnia Islamic Azad University Hamadan, Iran
[email protected]
II. RELATED WORK Building trustworthy environment and congestion control solutions have been studied previously in WSNs with or without using fuzzy logic. Munir [3] presents a fuzzylogic based mechanism to efficiently sort out traffic and minimize the packet loss for prioritized event-driven traffic. Akkaya and Younis [4] present an energy-aware QoS routing protocol for sensor networks which can run efficiently with best-effort traffic. The proposed model classifies traffic on the basis of real-time and non-real time application data. ESRT is a transport solution developed to achieve reliable event detection in WSN with minimum energy expenditure. It includes a congestion control component that serves the dual purpose of achieving reliability and conserving energy [5]. CODA uses a combination of the present and past channel loading conditions and the current buffer occupancy, to infer accurate detection of congestion at each receiver with low cost [6]. PSFQ [7] is scalable and reliable transport protocol that deals with strict data delivery guarantees rather than desired event reliability as it is done in ESRT. RMST [8] is a transport layer paradigm designed to complement directed diffusion (DD) [9] by adding a reliable data transport service on top of it. It’s a NACK based protocol which has primarily
1
timer driven loss detection and repair mechanisms. TARA discusses the network hotspot problem and presents a topology aware resource adaptation strategy to alleviate congestion in sensor network [10]. To reduce packet loss and achieve a fair delivery ratio [11] propose a source count based hierarchical medium access control (HMAC) and weighted round robin forwarding (WRRF). These two schemes together greatly reduce media contention and thereby congestion due to collision. Evaluating the performance of fuzzy based trusted Ad hoc On-Demand Distance Vector (AODV) routing protocol in Ad hoc network with varying number of malicious nodes is proposed in [12]. Zhiying explores a distributed trust model, enabling recommendation based trust and trust-based recommendation to build reasonable trust relationship among WSN [13]. Mohammad et al. represent a new approach of calculating trust between sensor nodes based on their sensed data and the reported data from surrounding nodes [14]. It is addressing the trust issue from a continuous sensed data which is different from all other approaches addressing the issues from communications and binary point of view. A fuzzy logic based anomaly detection scheme which provides the security to the DD protocol is proposed by Chi and Cho [16]. The scheme is effective in preventing Denial-of-Service type attacks. Hull et al. examine three techniques for congestion control that spans different layers of the traditional protocol stack: hop-by-hop flow control, rate limiting source traffic when transit traffic is present, and a prioritized medium access control (MAC) protocol [17]. Even though all these mentioned studies attempt to reduce the network traffic and try to increase packet delivery, they all assume trustworthy environment and reliable interactions without considering presence of malfunctioning nodes. Malfunctioning nodes increase transit traffic and diffuse useless packets. Moreover, all preceding works that deal with trust approaches use negotiation techniques by which some communicational packets may be dropped. Zarei et al. [1] presents new approach using in-network fuzzy based processing. Each node monitors the behavior of its neighbors and estimates suitable trust values. An inferred trust value is used to show node’s legitimacy and will be compared with threshold trust value (TTV). Indeed nodes are responsible to cooperate with legitimate nodes and drop packets of malicious nodes in order to reduce network traffic. Number of triggered packets is unpredictable and is subject to change based on interactions. In our opinion using variable TTV rather than a constant value is more effective when used in WSNs with variable transit traffic.
suitable. Using dynamic TTV provides a better measure of fairness when managing congestion issues and the resulting packet loss problems. In terms of increasing or decreasing the lost packets, FCCTF does different operation. 1) Increasing Mode When the rate of received packets is more than forwarded packets, buffer overflowing gradually occurs. Through the use of fuzzy logic, FCCTF makes a predictive decision to behave more flexible when a huge amount of packets are in danger of being dropped. In our work the rate of the received packets increases in the following situations/locations:
There exist a common nodes in construction of multiple paths In nodes near the sink. When a data burst occurs due to triggering new detected phenomenon When malicious nodes in a dense regions diffuse considerable amount of invalid or duplicative packets If packet loss continues in high rate, the new scheme increases TTV based on real time parameters which are explained subsequently. Therefore, more suspicious nodes are detected as malicious and will be blocked in order to increase the chance of useful packets delivery. 2) Decreasing Mode The packet loss will not occur when the rate of received packets is less than the rate of forwarded packets. In this situation, unlike the decreasing mode of FCCTF, the TTV reduces. In the result some suspicious nodes are unblocked again cooperating with the neighbors. It is noticeable that the misbehaviors of these nodes are because of selfishness issues on forwarding invalid, duplicative or delayed packets. Indeed no catastrophic consequent occurs with unblocking these malfunctioning types. Instead, the probability of nondelivered packets due to disconnection between some regions will reduces. Blocking of nodes may raise network disconnections. Although, with unblocking some of these nodes, a considerable amount of useless packets may be transmitted to the network. In this work we consider the following assumptions
No additional negotiation is introduced between sensor nodes Packet loss may occur due to misbehavior of malicious nodes or data burst Existence of misbehaviors such as number, delay and validity ([1]). In addition in this work based on misbehavior intensity malicious nodes are classified as CB, AB and UB nodes accordingly corresponding to Catastrophic Behaving, Abnormal Behaving and Unstable Behaving. Fuzzy Inference System (FIS) is illustrated step by step in following in order to be used in the FCCTF. Fig. 1 illustrates the structure of a rule-based FIS [18].
III. FAIRNESS CONGESTION CONTROL USING FUZZY LOGIC In sensor networks corrupted nodes which are selfish to drop non-forwarded packets or malfunctioning ones that reforward duplicate packets should be detected and marked as malicious. Despite all FCC efforts, considering non-real time network traffic and choosing static and fixed TTV is not
2
A. Preparing the Inputs for FIS The utilized FIS is composed of two inputs that are explained in following independently.
high and very high values respectively. The set of linguistic values for output MFs are {LL, LM, LH, HL, HM, HH} representing low-low, low-medium, low-high, high-low, high-medium and high-high respectively.
FIS used to calculate TTV
2) Inference phase based on rules The “Inference Engine” applies a predetermined set of linguistic rules and produces the suitable TTV in fuzzy form. MFs of TTV are explained in Figure 2(c). LL and HH shows that the TTV is minimal and maximal respectively. Table 1 shows how inference engine produces TTV The first rule of Table 1 can be translated as, “If PFnet is VL and Bf is VL, the TTV is LH”.
Rule base
PFnet Fuzzification
Bf
Inference Engine
Defuzzification
TTV
Figure 1: Utilized Fuzzy Inference System for calculating TTV
Degree of membership
Degree of membership
1) First Input The ratio of received to forwarded packets is considered as first FIS input. Indeed, when the buffer is not full, the number of received packets and buffered packets are equal. In contrast, when the buffer is full, the number of received packets is far more than the number of buffered packets. For the sake of simplicity, the two factors are considered the same. Using Pnet the first input is the computed as Pnet = Pin / Pout.
Degree of membership
2) Second Input Current buffer capacity (Bf) is considered as the second input. Bf is computed as Bf = (Tp – Cp) / TP. In which Cp is number of buffered packets. Tp is buffer size, Therefore, when Cp is zero, it means that the Bf is maximal (buffer is empty) and when Cp is equal to Tp, it means that the Bf is minimal (buffer is full) and no more packets will be accepted. B. Computing TTV as FIS Output
H
VH
0.5 0 -5
-4
-3
-2
VL
1
-1
0
(a)
L
1
M
2
3
H
5
4
PFnet
VH
0.5 0
1
0
0.2
0.4
(b) 0.6
0.8
LL
LM
LH
HL
HM
5
6
7
8
9
1
Bf
HH
0.5 0
Threshold Trust Value Current capacity of buffer (Bf)
(C)
TTV
10
VL L M H VH
Incoming to outgoing packet ratio (PFnet) VL
L
M
H
VH
LH
HL
HM
HH
HH
LH
LH
HL
HM
HH
LM
LH
LH
HL
HM
LL
LM
LH
HL
HL
LL
LL
LM
LH
HL
3) Defuzzification interface The “Defuzzification interface” performs a scale mapping and converts the range of fuzzified output into corresponding non-fuzzy form. Inferred fuzzy TTVs are converted to values in range of {5, 6 ... 10}, in which “10” or “5” are non-fuzzy forms of HH or LL and show packet loss is maximum or minimal respectively. Through the use of “5”, the nodes which are not absolutely legitimate can cooperate with the other nodes and they will not be blocked until their obtained trust values are larger than TTV.
Pin Pout Pin Pout
M
Table 1: Fuzzy rule table for inferring TTV based on PFnet and Bf
1) Fuzzification interface “Fuzzification interface” can be defined as the operation that maps a crisp object to a fuzzy set, i.e., to a membership function (MF). The MFs are used to associate a grade to each linguistic term. The MFs of inputs are depicted in Figure 2(a) and (b). Equation (1) performs scale mapping and makes suitable change on minimum and maximum of universe discourse for Pnet. Thus Pnet will be converted to PFnet and related MFs are shown in Figure 2(a). Linguistic value of M illustrates that Pin and Pout are equal. H and VH show that Pin > Pout. In contrast, L and VL show Pin < Pout. MFs of Bf are depicted in Figure 2(b) as second input. VH is interpreted as if at least 70% of the buffer is empty. VL shows that at most 30% of buffer is empty.
PFnet
L
Figure 2: PFnet, Bf and TTV membership functions
The following procedure shows how these two inputs are applied to the FIS and a suitable TTV is produced as the output.
1 Pnet 1 Pnet
Pnet
VL
1
(1)
C. FCCTF operational flow chart The operation of FCCTF is summarized in Figure 3. In order to detect behavior of each neighbor, The Fuzzy Input Parameter Extraction (FIPE) processes the buffered packets and generates an analytical metrics for the number, delay
Pin Pout
In Figure 2 the set of linguistic values for inputs MFs are {VL, L, M, H, VH} representing very low, low, medium,
3
forwarded packets was either more or lower than expectation. FCC considered such nodes as legitimate, but as explained earlier FCCTF changes TTV based on congestion pressure. Indeed, in increasing mode the TTV increases and this node will be blocked, otherwise, in decreasing mode this node can cooperate with neighbors again and contribute to next routing discoveries.
If New TTV == Current TTV
FIS4 Pnet
Bf
No Update TTV
Declaration New TTV to FLC
Fuzzy Logic Controller (FLC)
Inferred Trust
Trust Verification And Drop Decision
New TTV
(TVDD)
Buffer
Buffer Management (BM)
computing Pout
Computing Pnet Reporting
Data
Order
Bf
Data
computing Pin
The concept is implemented using fuzzy logic and is investigated by artificial WSN scenario in the Network Simulator (NS-2) [15] environment. 500 nodes were deployed in a rectangular area of 600×600 m2 using DD routing protocol. All Simulation parameters are listed in Table 2. The Monte-Carlo simulating of different scenarios reported when malicious nodes are scattered in a dense region, they trigger a huge amount of useless packets and increase the probability of the buffer overflow. Based on misbehaving types such as “number”, “delay” and “validity”, suitable trust value are assigned to each neighbor. Real time transit traffic determines the appropriate TTV. Each node is considered as legitimate if the inferred trust value is equal or greater than inferred TTV. Network improvement results of FCCTF deployment, are investigated in the following. Fig. 4 shows the trust diagram for a typical node. The variable inferred trust value in predetermined intervals indicates a potential misbehavior; some undelivered packets were misconceived as duplicative and were dropped intractably or some duplicative packets re-forwarded by mistake. Through this malfunctioning the number of
Receives inferred trust from FLC And orders BM to block this node if FLC is declared as malicious
Bf
Validity
Delay
Number
Fuzzy Input Parameter Extraction (FIPE)
IV. SIMULATION AND RESULTS
Yes
Pnet
and validity of received packets from each neighbor independently [1]. Each three values are sent to the Fuzzy Logic Controller (FLC) in order to be used as the first input of FIS1, FIS2 and FIS3 accordingly. The Bf is also sent to FLC as the second input of FIS1, FIS2 and FIS3. FLC analyzes each neighbor and run these three FISs in serial form. If First FIS1 accepts the calculated trust for Number of a neighbor node’s packets, then the FIS2 will run. Afterwards, FIS3 will execute if FIS2 obtains an acceptable trust value for Delay checking. Therefore, the neighbor node is a legitimate node if FIS3 confirms it as final inference for the validity analysis. At the end of each inference if the calculated trust value is lower than the TTV, procedure will stop and the next inference system will not be executed [1]. Through the use of preceding procedure, FLC infers suitable trust value for all neighbors of the node and send them to the Trust Verification and Drop Decision (TVDD) module. TVDD receives Pnet and Bf from Buffer Management (BM) and executes the FIS4 corresponding to part A and B of section III. The new inferred TTV will be compared with previous one, if they differ, the TVDD will declare the new TTV to FLC. Moreover, the received result of inferred trust from FLC will be investigated by TVDD. TDVV will advise the BM if the node is malicious because of achieving lower trust value compared with the TTV. BM monitors the incoming and outgoing packets of the buffer and updates the information about Pin and Pout periodically. Therefore, the computed new Pnet will be reported to the TVDD. Moreover, the BM reports current buffer capacity to the FLC and TVDD after predetermined interval and finally, blocks receiving packets from the nodes which were inferred as malicious.
Pnet
To TVDD
Computing Bf in predetermined Interval
Reporting Bf to FLC and TVDD
Blocking malicious nodes based on reporting of TVDD
Figure 3. The operational Flow Chart for proposed FCCTF
Fig. 5 depicts packet loss comparison and improvement of FCCTF over FCC when treating the congestion problem. All nodes have been equipped with fuzzy system. 16% of all nodes were defined as abnormal behavior. In this scenario FCCTF operates in the increasing mode for as long as buffer overflowing exists trying to increase the TTV. Our results indicates that when TTV increases from 6 to 8, the average number of lost packet is reduced, in this situation up to 20 nodes, with an obtained trust in the range of 6 to 8, stopped working. However these 20 nodes are considered as legitimate in FCC, but they could not be trusted in FCCTF anymore. Therefore, the average number of packet loss is reduced from 15 to 12.5 packets. This results in nearly 10% improvements. With increasing the TTV to 10, further suitable result was achieved. In comparison with FCC, The average number of packet loss was reduced nearly 20%.
4
by using FCC up to 57% of CB nodes were detected and blocked. This increases the packet delivery up to 2.85% compared with non-fuzzy form. By using FCCTF, TTV is increased to 8 and finally to 10, in which up to 60% of AB and UB nodes are detected and blocked. Compared with non-fuzzy form, FCCTF improves the packet delivery up to 4% and 6% corresponding to decreasing overhead ratio of AB and UB nodes. Indeed, with increasing TTV from 6 to 10, the packet delivery is improved about 3%.
Using “10” as candidate TTV, only the fully trusted nodes could cooperate in interactions.
Average number of packet loss
Table 2. Simulation Parameters. Simulation time 1000 s Simulation area 600 x 600 m Number of node 500 Number of source 10 Number of sink 2 Frequency of operation 914 MHz Mac layer 802.11 Node placement Random Transmission range 40m Transmission power 660 mW Received power threshold 395 mW Idle power 35 mW Movement node No Maximum neighbor node 9 Number of malicious node 1% to 20% Type of malicious Number, Delay, Validity Threshold Trust Value 6-8-10 Inffered trust values by neighbors of a node
Trust Value
20 15 10 5 0 0
10
FCC-TTV=6 FCCTF-TTV=8 FCCTF-TTV=10
Interval = 0.1 (s)
50
100
150
200
250
Time (s)
8
Figure 5. Packet loss comparison between FCC and FCCTF when all nodes have been equipped with fuzzy system
6
Table 3. Packet delivery improvement based on number of defined and detected malicious nodes. 150
Scenario
200
Time (s)
Defined malicious nodes (%)
Figure 4. Evaluated Trust by neighbors of a node
Packet delivery increases if the average ratio of diffused packets is well adjusted according to the network resource restrictions such as buffer resources and communication channel limitations. Whenever congestion likelihood decreases, a greater number of packets will be delivered and packet delivery ratio increases. Through the use of FCCTF, the majority number of malicious nodes are detected and dismissed. In this case the corresponding useless packets are dropped according to reducing forced overhead ratio. For highlighting malicious nodes detection of FCCTF over FCC, we defined four scenarios. Each scenario was repeated for both 68% and 100% fuzzification corresponding to achieve low and high overhead ratio respectively [1]. Table 3 shows complete comparison between FCC and FCCTF in terms of number of defined and detected malicious nodes. Based on malfunctioning impression, corresponding trust values of CB, AB and UB nodes were considered as: lower than 6, between 6 to 8 and between 8 to 10 respectively. For instance, in first scenario 6% of all nodes were defined as suspicious in which 3% of them considered as CB, 1% as AB and 2% as UB nodes. It should be noted that such CB, AB or UB nodes may malfunction in one or more types which were explained in [1] thoroughly. In first step of first scenario, 68% of all nodes have been equipped. In this case
Detected malicious nodes (%)
CB nodes AB nodes UB nodes Total malicious nodes FCC-6 Low FCCTF-8 overhead FCCTF-10 (68%) FCC-6 High FCCTF-8 overhead FCCTF-10 (100%)
Fourth
100
Third
50
Second
0
First
4
3 1 2 6 1.6 2.2 3.4 2.8 3.8 5.8
5 1.6 3 9.6 3.2 4.22 6.14 4.8 6.4 9.2
8 2.6 4 14.6 5.6 7.4 11.4 7.8 10.2 14.4
16 4 5 25 12.8 16 20 15.6 19.8 24
In the second step of the first scenario, all the nodes have been equipped with the fuzzy system. By using FCC, 97.5% of CB nodes are detected. Due to dropping the received packets from CB nodes, 5% improvement in the packet delivery is obtained. By using FCCTF the overhead ratio of AB and UB nodes is decreased resulting in 7% and 10% improvement in the packet delivery accordingly. Figure 6 illustrates the packet delivery comparison between the FCCTF and the FCC in four scenarios. In terms of comparing these scenarios some remarkable results are achieved on packet delivery improvement: a) In each scenario using 68% or 100% fuzzification corresponding to low or high overhead ratio, the related packet delivery improvement is lower or higher respectively. b) Packet delivery improvement is dependent on malicious nodes and/or TTV increment.
5
[2]
Packet delivery improvement (%)
c) When TTV is fixed and the number of malicious nodes increases, the deference of packet delivery improvement between low and high overhead ratio of fuzzification, decreases in scenarios. This result is occurred in both FCC and FCCTF. d) Considering FCCTF with TTV=10 and low overhead ratio (FCCTF-10-Low), the packet delivery improvement is more than the condition in which FCCTF was used with TTV=8 and high overhead ratio (FCCTF-8High) in last three scenarios. e) Considering FCCTF-10-Low, results show packet delivery improvement is more than the FCC-6-High in all scenarios. 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0%
[3]
[4]
[5]
[6] FCC-6-Low FCC-6-High FCCTF-8-Low
[7]
FCCTF-8-High FCCTF-10-Low FCCTF-10-High
[8]
[9]
6% 9.6% 15.6% Defined malicious nodes (%)
25% [10]
Figure 6. Comparison of packet delivery improvement between FCC and FCCTF. The variable range of malicious nodes is considered from 3% to 25%. TTV is considered as 6, 8 and 10. Each scenario is simulated for both low and high over head ratio corresponding to 68% and 100% fuzzification respectively.
[11]
V. CONCLUSION [12]
A Fairness Congestion Control for a disTrustful wireless sensor network using Fuzzy logic (FCCTF) was proposed in this paper. As presented FCCTF is the modified form of FCC, in which the TTV decision making is based on corresponding traffic ratio of related region. Therefore TTV could change in increasing or decreasing form of traffic ratio dynamically. By increasing TTV from 6 to 8 and finally 10, more malicious nodes are detected and blocked, consequently related packets are removed and the useless packets are replaced by useful packets. Simulation results indicate that when increasing TTV to 8 or 10, the packet drop ratio is reduced to 10% or 20% respectively. This results in 7.5% and 18.5% improvement in packet delivery.
[13]
[14]
[15]
[16]
REFRENCES [1]
M. Zarei, A.M. Rahmani, A. Sasan, and M. Teshnelab. “Fuzzy based trust estimation for congestion control in wireless sensor networks,” Barcelona, IEEE Conf. Intelligent Networking and Collaborative Systems (INCOS '09), IEEE Press, pp. 233-236, doi: 10.1109/INCOS.2009.16, Nov. 2009.
[17]
I. F. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci. “Wireless sensor networks: a survey,” Elsevier North-Holland, Vol. 38, No. 4, pp. 393-422, 15 March 2002. S.A. Munir, W.B. Yu, B. Ren, and M. Ma. “Fuzzy Logic Based Congestion Estimation for QoS in Wireless Sensor Network,” Kowloon, Wireless Communications and Networking Conference (WCNC 07), IEEE. 2007, pp. 4336 – 4341. 2007 doi: 10.1109/WCNC.2007.791 K. Akkaya, and M. Younis, “An Energy-Aware QoS Routing Protocol for Wireless Sensor Networks,” Washington, DC, Distributed Computing Systems Workshops, pp. 710- 715. 2003 doi: 10.1109/ICDCSW.2003.1203636 S. Yogesh, B.A. Ozgur, and F.A Ian “ESRT Event-to-Sink Reliable Transport in wireless sensor networks,” International Symposium on Mobile Ad Hoc Networking & Computing, pp. 177–189. ACM Press, New York. 2003 Y.W. Chieh, B.E. Shane, and T.C. Andrew “CODA: Congestion Detection and Avoidance in Sensor Networks,” In: the proceedings of ACM SenSys, Los Angeles, pp. 266–279. ACM Press, New York. 2003. C.Y. Wan, A.T. Campbell, and L. Krishnamurthy, “Pump-slowly, fetch-quickly (PSFQ): a reliable transport protocol for sensor networks,” IEEE Journal of Selected Areas in Communication, Vol. 23, No. 4, pp. 862–872. 2005. R. Stann, and J. Heidemann, “RMST Reliable data transport in sensor networks,” Sensor Network Protocols and Applications. Proceedings of the First IEEE, pp. 102–112. IEEE Computer Society Press, Los Alamitos. 2003. C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and F. Silva, “Directed Diffusion for wireless sensor networking,” IEEE/ACM Transactions on, Vol. 11, No. 1, pp. 2 – 16. 2003. J. Kang, Y. Zhang, and B. Nath, “TARA Topology Aware Resource Adaptation to Alleviate Congestion in Sensor Networks,” IEEE Transaction on Parallel and Distributed Systems, Vol. 18, No. 7, pp. 919–931, 2007. Md. Mamun-Or-Rashid, M. M. Alam, Md. Abdur Razzaque, and S. H. Choong, “Reliable Event Detection and Congestion Avoidance in Wireless Sensor Networks,” High Performance Computing and Communications, Vol. 4782/2007, pp. 521-532, Springer Berlin / Heidelberg, 2007. J. Martin, L. Manickam, and S. Shanmugavel, “Fuzzy Based Trusted Ad Hoc On-demand Distance Vector Routing Protocol for MANET,” Third IEEE International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob 07), 2007. Y. Zhiying, K. Daeyoung, and D. Yoonmee, “PLUS: parameterised localised trust management-based security framework for sensor networks” International Journal of Sensor Networks, Vol. 3, No. 4, pp. 224 – 236. Vancouver, BC, 2008. M. Mohammad, and C. Subhash, “GTRSSN: Gaussian Trust and Reputation System for Sensor Networks,” Advances in Computer and Information Sciences and Engineering, Computer Science, pp. 343347, Springer Netherlands, 2008. L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A. Helmy, P. Huang, S. McCanne, K. Varadhan, Ya Xu and Haobo Yu. “Advances in Network Simulation,” IEEE Computer, Vol. 33, No. 5, pp. 59-67, 2000. S. H. Chi, and T. H. Cho, “Fuzzy Logic Anomaly Detection Scheme for Directed Diffusion Based Sensor Networks,” Fuzzy Systems and Knowledge Discovery. Springer Berlin / Heidelberg Vol. 4223/2006, pp. 725-734. 2006. B. Hull, K. Jamieson, and H. Balakrishnan, “Mitigating congestion in wireless sensor networks,” Conference On Embedded Networked Sensor Systems, ACM, pp. 134 – 147, 2004.
[18] J. M. Mendel, "Fuzzy logic systems for engineering: a tutorial," Proceedings of the IEEE, Vol. 83, No. 3, pp. 345–377, 1995.
6
