Performance of IEEE 802.11 based Wireless Sensor Networks in ...

5 downloads 911 Views 132KB Size Report
sor network performance up to order of magnitudes with respect to the network ... in order to sense and monitor environmental conditions in wild habitat. Another ...
Performance of IEEE 802.11 based Wireless Sensor Networks in Noisy Environments Tamer Nadeem

Ashok Agrawala

Department of Computer Science University of Maryland College Park, MD 20742 [email protected]

Department of Computer Science University of Maryland College Park, MD 20742 [email protected]

Abstract Wireless sensor networks have recently gained a lot of attention from the research community. Wireless sensor networks are often partitioned into clusters, each managed by a cluster head (gateway). In this paper, we study the performance of the IEEE 802.11 based wireless sensor networks in noisy environments. We show that using the standard binary exponential backoff (BEB) mechanism in noisy clustered sensor networks results in a poor throughput performance due to its inability of differentiating between the causes of unsuccessful packet transmissions and verify the analytical model using the ns-2 simulator. We show the enhanced BEB mechanism in [1] enhances clustered sensor network performance up to order of magnitudes with respect to the network error rates (noise level). We also study the noise effect on the fairness of IEEE 802.11 and show that our proposed mechanism maintains the channel fairness between the competing sensors.

1

Introduction

Wireless sensor networks (WSN) are an emerging field of research which combines many challenges of modern computer science, wireless communication and mobile computing. WSN popularity has increased dramatically due to their potential for some civil and military applications. Each sensor is capable of sensing surrounding conditions such as temperature, or detecting the presence of certain objects. Sensor networks have several civil and military application. For example, sensors could be deployed in order to sense and monitor environmental conditions in wild habitat. Another application is where sensors can perform surveillance missions by detecting moving targets or objects. WSN are deployed in large number of unattended nodes and hence the underlying network architecture has become one of the challenging areas in wireless sensor network research [11, 9, 12, 16]. A common network architecture

Base Station

Network cluster

Cluster head

Data sensor

Relay sensor

Figure 1: Clustered architecture of sensor network

of deploying the sensor nodes is to employ network clustering. Such cluster-based architecture assigns for each cluster a cluster head (gateway) in which can reach all the nodes in the cluster in one hop. Also, each gateway is responsible to interact and collaborate with other gateways in the networks, in addition to the interaction with the network base node, to execute the required missions as shown in Figure 1. Hereafter we use the terms ”sensor” and ”node” interchangeably. Like in all shared-medium networks, medium access control (MAC) is an important technique that enables the successful operation of the network. Current MAC design for WSN can be broadly divided into contention-based and TDMA protocols. Most of the contention based MAC

protocols proposed in the literature follow the operational model of carrier sense multiple access (CSMA) [10], incorporating handshaking signals and a back-off mechanism to reduce the probability of collisions. In CSMA, a node listens to the channel before transmitting. If it detects a busy channel, it delays access and retries later. The widely used standardized IEEE 802.11 distributed coordination function (DCF) [2] is an example of the contention-based protocol. One problem with the IEEE 802.11 standards is that the standards developed with no energy minimization mechanisms that is necessarily for WSN. Many mechanisms based on the 802.11 standards have been proposed to solve the problem of the energy minimization [7, 17, 6]. Although in this paper we consider IEEE 802.11 DCF as underneath MAC protocol for the wireless sensor networks, the analysis and mechanisms developed in this paper can easily be applied to any similar contention based protocol. The IEEE 802.11 DCF protocol works in two different modes: infrastructure mode and ad hoc mode. In infrastructure mode communication between modes must go through a central node, while in ad hoc mode nodes communicate directly with each other. Infrastructure mode fits the communication model of the sensor network clusters in which sensors with a cluster communicate directly with their cluster head. On the other hand, the communications between gateways follow the ad hoc mode. In this paper we focus on communication within WSN clusters and we assume IEEE 802.11 DCF infrastructure mode is the underneath MAC protocol for the cluster’s nodes. In IEEE 802 DCF mechanism, the binary exponential backoff (BEB) mechanism is used for resolving packet collisions that occur as the uncoordinated nodes contend for the channel. To ensure packet transmission reliability, MAC acknowledgment (ACK) frames are used to indicate the correct reception of the data packets. When a node does not receive a corresponding ACK frame, it assumes the packet has been dropped due to a collision, and invokes the BEB mechanism for retransmission. We refer to such mechanism in this paper as naiveBEB . Applying naiveBEB mechanism in WSN clusters that suffers from errors due to the noise in the wireless channels, results in unpredictable delays and poor throughput performance because it always assumes that the packet corruptions are due to collisions only. In this paper, we study the performance of the IEEE 802.11 MAC for WSN clusters in noisy environments. Show how smartBEB proposed in [1], that capable of differentiating between different types of corruptions that cause unsuccessful transmissions, enhances WSN clusters performance up to order of magnitudes with respect to the network error rates (noise level) especially with clus-

ter of moderate number of sensors (e.g., tens of sensors) and small packet size which is the case in WSN. Moreover, sensors are located at different distances from the gateway and with the presence of noise, data packets from each sensor experience different bit error rate and hence affect the sensor’s fairness. We study the noise effect on the fairness of WSN cluster and show that smartBEB mechanism maintains the channel fairness between the cluster’s sensors. By enhancing cluster throughput and maintaining channel access fairness within the cluster’s sensors, the cluster’s quality of service is enhanced and consequently the WSN information assurance is supported.

2 Related Work One of the issues in the analysis of the IEEE 802.11 protocol is to devise an analytical model which can predict the collision probability and its effect on the performance metrics. Paper [5] analyzes the throughput and fairness issues of the DCF function and paper [4] gives the theoretical throughput limit of 802.11 based on a p-persistent variant. However, none of these captures the effect of the Contention Window(CW) and binary slotted exponential back-off procedure used by DCF in 802.11. Paper [3] uses Markov process to analyze the saturation throughput of 802.11 and show that the Markov analysis works well. The model is extended in [18] to consider the frame retransmission limits. While these studies use the stochastic analysis, TC model [15] uses the mathematical approximations with average values. The models mentioned so far assume ideal channel conditions, where packet error does not occur. Qiao and Choi [13, 14] assume additive white Gaussian noise channel (AWGN) and calculate packet error probability, then derive the goodput performance of PHY/MAC protocol analytically. However they assume that there are only two nodes (one sender and one receiver) therefore no collisions occur. In our model we consider both packet errors and the collisions among nodes. To our knowledge, neither of the previous works addressed the effect of environment noises of the network performance, nor the fairness between nodes suffering from different noise values. A node can reduce its packet error rates by reducing its data transmission rate in the noisy environments. However, in additions to the anomaly of using different data transmission rates in a wireless network[8], the total goodput of the node could be lowered with the reduction of its data rate. For example, assume a node experiences a 30% packet error rates with transmission rate of 11Mpbs. By reducing the transmission rate to 2Mbps it would eliminate the packet error rates, however the total goodput is decreased by 50% which is more than the experienced packet error rate with original transmission rate.

(1-pd) / W 0

1 0,0

1 0,1

0,2

... ...

are the average time the channel is sensed busy because of a successful transmission, failure (corrupted) transmission, or a collision respectively. The Pid , Ptr , and Pcl are the probability a time slot idle where no node is transmitting, has transmission of only one node with probability of pe of corrupting the packet, or has a collision (cl) because two or more nodes are transmitting in the same time respectively. Such probabilities are calculated as follow:

(1-pd) / W 0 1 0,W 0-2

0,W 0-1

...

...

pd / W 1

...

...

...

...

i-1,0

pd / W i

1

i,0

1 i,1

i,2

... ...

pd / W i 1 i,Wi-2

i,Wi-1

...

...

Pid Ptr Pcl

pd / W i+1

...

...

...

pd / W m 1 m,0

1 m,1

m,2

... ... ... ...

pd / W m 1 m,W m-2

m,W m-1 pd / W m

pd / W m

Figure 2: Markov Chain model for the backoff window in noisy environments

3

Performance of IEEE 802.11 DCF in Noisy Environment

In [1], we analytically studied the performance of the IEEE 802.11 standard BEB mechanism (naiveBEB ) in a noisy environments. Our model is based on the one proposed by [3] and we use the same assumption for our analysis. Let m is the maximum backoff stage, W0 is the initial contention window size of the BEB mechanism, and Wi = 2i W0 where i ∈ (0, m) is called backoff stage. We model the bi-dimensional process s(t), b(t) as discretetime Markov chain and show it in Figure 2 where s(t) is the stochastic process representing the backoff stage (0, . . . , m) of the node at time t, and b(t) be the stochastic process representing the back-off window size for a given node at slot time t1 . Unlike paper [3], pd is used to captures the effect of the packet error rate, pe , in the model in addition to the collision probability pc as follow: pd = pc + pe − pc pe

(1)

We define the saturation goodput of the WSN cluster as: G = =

= (1 − τ )n = nτ (1 − τ )n−1 = 1 − (1 − τ )n−1 (1 − τ + nτ )

E[successfully transmitted payload bytes in a slot time] E[length of a slot time] (1 − pe )Ptr E[S] Pid δ + (1 − pe )Ptr Ts + pe Ptr Tf + Pcl Tc

(2)

where E[S] is the average packet length, and δ is the duration of an empty (idle) slot time. The Ts , Tf , and Tc 1 The slot time refers to the time interval between two consecutive backoff time counter decrements. This value is fixed (δ) in case of idle medium, or variable that includes a packet transmission when medium is busy.

where τ is the probability that a node transmits in a randomly chosen slot time is and is calculated as follow: Pm τ = b i=0 i,0 2(1 − 2pd )

=

(1 − 2pd )(W0 + 1) + pd W0 (1 − (2pd )m )

(3)

where bi,k is the stationary probability for state s(t)=i,b(t)=k, i ∈ (0, m) and k ∈ (0, Wi − 1). In steady state, pd is expressed as: pd

=

1 − (1 − pe )(1 − τ )n−1

(4)

Equations 3 and 4 represent a nonlinear system in two unknowns τ and pd (pc ) which can be solved using numerical techniques. We validated our model by comparing the analytical results with the results from ns-2 simulator as shown in [1]. All the parameters used in analytical model and our simulations follow the parameters of DSSS [2], and are summarized in Table 5. Note that PHY header, RTS frame, and CTS frame are sent at the basic access rate. Figure 3 plots the pc and pd values. In this figure we observe an interesting behavior in which pd increases with pe while pc decreases with the increase in pe . This indicates that the increase in pe has the same effect as that reducing the number of nodes. Specifically, with increasing pe we can increase the number of active nodes to utilize the additional number of idle slots introduced by naiveBEB mechanism while maintaining the original conditional collision probability (pc ) when pe = 0. Consequently, values of the original Pid and Ptr are maintained that utilizes the WSN cluster saturation goodput. The additional number of nodes could be calculated as follow: nadditional = 1 +

ln (1 − pd ) − ln (1 − pe ) − n0 ln (1 − τ )

(5)

where n0 is the original number of nodes, τ is calculated using Equation 3, and pd is calculated using Equation 1. Note that pc value is fixed with different values of pe and is calculated by solving Equations 3 and 4 for the case when n = n0 and pe = 0.

1

0.6

0.4

0

0 0.2

Comments

PHY header

24 octets

PHY layer overhead

MAC header

28 octets

MAC layer overhead

ACK

38 octets

ACK frame + PHY header

RTS

44 octets

RTS frame + PHY header

CTS

38 octets

CTS frame + PHY header

100

50

0.1

Value

150

0.2

0

Parameter Additional stations

200

# of Stations

0.8

Probability

250

Pd (Analytical) Pd (Simulation) Pc (Analytical) Pc (Simulation)

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

Figure 3: The pc and pd values in noisy WSN cluster.

0.8

0

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

0.8

Figure 4: The additional nodes to original 20 nodes.

Slot time

20 µs

idle slot time (δ)

SIFS

10 µs

SIFS time

DIFS

50 µs

SIFS + 2 * δ

aCWmin

31

minimum contention window

m

5

backoff levels

Figure 5: MAC and PHY system parameter.

4

Enhanced IEEE 802.11 MAC (smartBEB )

To overcome the inefficiency of the current IEEE 802.11 BEB mechanism (naiveBEB ), we proposed in [1] the smartBEB mechanism to enhance the IEEE 802.11 with a capability to differentiate between different causes for packet corruptions. In case a packet is dropped because of collision corruption, the IEEE 802.11 standard BEB mechanism is followed and the contention window (CW ) is doubled. If the cause of dropping a packet is noise (error) corruption, smartBEB handles the transmission as successful one and resets the CW to W0 . In addition, smartBEB handles the retransmission of the dropped packet as a new packet transmission. We proposed mechanisms to implement smartBEB in RTS/CTS access mode as well as the basic access mode of the IEEE 802.11 DCF. Defining the percentage of the goodput enhancement of smartBEB over naiveBEB as: ∇G =

´−G G × 100 G

(6)

Figure 6 shows analytical results of the ∇G for different configuration of data rates, number of nodes, and access modes in noisy WSN cluster. smartBEB mechanism enhances the system goodput significantly because it limits the contention window size that reduces the number of unnecessary idle time slots. Figure 7 shows the ns-2 simulation results for a WSN cluster with 10 sensors in addition to the gateway transmitting data at rate of 22Mbps using the basic access method (no RTC/CTS). From the figure, smartBEB enhances the network performance significantly to order of magnitudes especially for small data packets which is a common case for most sensor networks. We also experimented with smartBEB in a hierarchal WSN cluster in which some sensors act as a relay to another sensors as in Figure 1. In this scenario, we used two levels cluster in which the number of hops between sensors collecting data nd the gateway is two hops through relay

sensors. We experimented with different error configurations. We assumed error happens: 1) only at the gateway, 2) only at relay sensors, or 3) at both the gateway and relay sensors. Figure 8 shows shows the ns-2 simulation results for a WSN cluster with 10 relay sensors in addition to the gateway using 22Mbps as the data rate. For each relay sensor there is 10 sensors that collect data and send them to the gateway through the relay using basic access method. As shown, smartBEB overcome the inefficiency of naiveBEB and enhances the cluster performance.

5 802.11 fairness in Noisy environments In this section we briefly study the IEEE 802.11 fairness when different nodes experience different error rates in the noisy WSN environments. We extended the Markov model in Section 3 to represent different classes of nodes. In this paper, we describe the extension for two classes only while the extension for more than two classes is straight forward. Each node in the first class of n1 nodes experiences packet drop rate pd1 which consists of error rate pe1 and the collision rate of pc1 . On the other hand, each node in the other class of n2 nodes experiences drop rate pd2 that is a function of pe2 and pc2 . Similar to Equations 3 and 4, we get: τ1

=

τ2

=

pd1 pd2

= =

2(1 − 2pd1 ) (1 − 2pd1 )(W0 + 1) + pd1 W0 (1 − (2pd1 )m ) 2(1 − 2pd2 ) (1 − 2pd2 )(W0 + 1) + pd2 W0 (1 − (2pd2 )m ) 1 − (1 − pe1 )(1 − τ1 )n1 −1 (1 − τ2 )n2 1 − (1 − pe2 )(1 − τ2 )n2 −1 (1 − τ1 )n1

(7)

where τ1 and τ2 are the probabilities of transmitting in a randomly chosen slot time for nodes in the first class and the second class respectively. Similar, Equations in 3 are extended to: Pid

= (1 − τ1 )n1 (1 − τ2 )n2

60 50 40 30 20

120

50 Bytes 100 Bytes 200 Bytes 500 Bytes 1000 Bytes

250

Goodput Enhancement

Goodput Enhancement

300

5n, 22Mbps, basic, 500B 5n, 22Mbps, rts/cts, 500B 5n, 22Mbps, basic, 1000B 5n, 22Mbps, rts/cts, 1000B 5n, 11Mbps, rts/cts, 1000B 10n, 22Mbps, rts/cts, 1000B 10n, 11Mbps, rts/cts, 1000B 20n, 11Mbps, rts/cts, 1000B

70

200

150

100

50

10 0 0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

Figure 6: Analytical Goodput enhancement of smartBEB . Ptr1

=

Ptr2 Pcl

= =

n2 τ2 (1 − τ2

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

(8)

1 − Pid − Ptr1 − Ptr2

(1 − pe1 )Ptr1 S %

and

G2 =

40

0.7

0.8

0

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

0.8

Figure 8: Goodput enhancement for hierarchal WSN cluster .

6 Conclusion and Future Work

)n1

where Ptr1 is the probability that the time slot has a single transmission of a node belongs to first class, and Ptr2 is the probability that the time slot has a single transmission form the second class. The goodput G1 and G2 of the first class and the second class are expressed as: G1 =

0.6

Figure 7: Goodput enhancement using basic access method.

− τ1

60

0 0

n1 τ1 (1 − τ2 )n1 −1 (1 − τ2 )n2 )n2 −1 (1

80

20

0 0

Reception error at Gateway Reception error at relay sensors Reception error at both

100

Goodput Enhancement

80

(1 − pe2 )Ptr2 S %

% = (Pid δ + (1 − pe1 )Ptr1 Ts + pe1 Ptr1 Tf + (1 − pe2 )Ptr2 Ts + pe2 Ptr2 Tf + Pf l Tc ) and S, δ, Ts , Tf ,

where

Tc are the same as defined in Section 3. As an example, we consider the configuration of a WSN cluster consists of 10 active nodes, in addition to the gateway, where n1 = 5 nodes form the first class that do not experience any error rate (pe = 0), and the rest of the nodes n2 = 5 form the second class that experience same error rates where 0 ≤ pe2 ≤ 0.8. Figure 9 shows the corresponding τ values. In case of using naiveBEB , the cluster is in favor of the nodes belonging to the first class and assign them more probability to access the medium. While smartBEB guarantees that both classes will have equal probability (fair share) to access the medium. Figure 10 plots the analytical total goodput in addition to the goodputs of the individual classes assuming 11Mbps data rate and packet size of 1000 bytes and using RTS/CTS access mode. Total naiveBEB total goodput is higher than the smartBEB total goodput because naiveBEB favors the nodes with lower error rates which results in more successful transmissions. On the other hand, smartBEB maintains the fairness between nodes that decreases the number of successful transmission. Figure 11 shows the G1 /G2 ratio with ns-2 simulation. With naiveBEB , the goodput of the first class reaches hundreds times the goodput of the second class of nodes. Using smartBEB , the goodput ratio is corresponding to the error rates.

In this paper, we analyzed the clustered wireless sensor network performance in noisy environments. We showed how the standard BEB of IEEE 802.11 degrades significantly the network performance in such environments analytically and by simulation. We proposed an enhanced BEB, smartBEB , that enhances the WSN clusters performance by order of magnitudes in noisy WSN clusters. We showed how to implement the smartBEB in basic access mode and in the RTS/CTS access mode with minimal modification requirement to the IEEE 802.11 standard. Further, we studied the effect of the noises on the cluster fairness and showed how smartBEB guarantees the fairness by forfeiting the cluster goodput and hence increase the cluster’s quality of service. As future work, we are examining different fairness criteria based on pricing and performance models that allow the system to choose the optimum model for the wireless sensor network requirements.

Acknowledgments This work was supported in part by the Maryland Information and Network Dynamics (MIND) Laboratory, its founding partner Fujitsu Laboratories of America, and by the Department of Defense through a University of Maryland Institute for Advanced Computer Studies (UMIACS) contract.

References [1] T. Nadeem, and A. Agrawala, “IEEE 802.11 DCF Enhancements for Noisy Environments,” In Proceedings of 15th IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC 2004), Barcelona, Spain, September 5-8, 2004 [2] “IEEE 802.11b, Part ll: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: High-Speed Physical Layer Extension in the 2.4GHz Band,” supplement to IEEE 802.11 Standard, September 1999.

1.2e+07

tau1 & tau2 (smart_beb) tau1 (naive_beb) tau2 (naive_beb)

1e+07

Goodput (bits/sec)

0.05

tau

0.04

0.03

0.02

0.01

1000

G1 + G2 (smart_beb) G1 (smart_beb) G2 (smart_beb) G1 + G2 (naive_beb) G1 (naive_beb) G2 (naive_beb)

8e+06

G1/G2 (smart_beb, analytical) G1/G2 (smart_beb, simulation) G1/G2 (naive_beb, analytical) G1/G2 (naive_beb, simulation)

100

G1/G2

0.06

6e+06

4e+06

10

2e+06

0

0 0

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

Figure 9: The τ values of the two classes.

0.8

1 0

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

Figure 10: The Goodput of the two classes.

[3] G. Bianchi, “Performance Anaylsis of the IEEE 802.11 Distributed Coordination Function,” IEEE Journal on selected areas in communications, 18, March 2000. [4] F. Cali, M. Conti, and E. Gregori, “Dynamic Tuning of the IEEE 802.11 Protocol to Achieve a Theoretical Throughput Limit,” IEEE/ACM Transactions on Networking, 8(6), December 2000. [5] H. Chhaya and S. Gupta, “Performance modeling of asynchronous data transfer methods of IEEE 802.11 MAC protocol,” Wireless Networks, 3:217–234, 1997. [6] N. V. E. Jung, “An energy efficient mac protocol for wireless lans,” In Proceedings of IEEE INFOCOM’02, New York City, NY, June 2002. [7] B. L. E. Poon, “Smartnode: Achieving 802.11 mac interoperability in power-efficient sensor networks with dynamic range adjustments,” In Proceedings of the IEEE International Conference on Distributed Computing Systems (ICDCS), Providence, Rhode Island, May 2003. [8] M. H. A. D. G. Sabbatel, F. Rousseau, “Performance Anomaly of 802.11b,” In Proceedings of IEEE INFOCOM, San Francisco, California, March 2003. [9] V. A. G. P. K. Sohrabi, J. Gao, “Protocols for selforganization of a wireless sensor network,” IEEE Personal Communications, pages 16–27, October 2000. [10] F. T. L. Kleinrock, “Packet switching in radio channels: Part I carrier sense multiple access modes and their throughput delay characteristics,” IEEE Transactions on Communications, 23(12), 1975. [11] R. K. L. Subramanian, “An Architecture for Building Self Configurable Systems,” In IEEE/ACM Workshop on Mobile Ad Hoc Networking and Computing, August 2000.

0.8

0

0.1

0.2

0.3 0.4 0.5 Packet Error Rate (Pe)

0.6

0.7

0.8

Figure 11: G1/G2 ratio of the two classes.

[12] K. A. M. Younis, M. Youssef, “Energy-Aware Routing in Cluster-Based Sensor Networks,” In Proceedings of the 10th IEEE/ACM International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS2002), Fort Worth, Texas, October 2002. [13] D. Qiao and S. Choi, “Goodput enhancement of IEEE 802.11a wireless LAN via link adaptation,” In Proceedings of IEEE International Conference on Communications (ICC’2001), 2001. [14] D. Qiao, S. Choi, and K. G. Shin, “Goodput Analysis and Link Adaptation for IEEE 802.11a Wireless LANs,” IEEE Transaction On Mobile Computing, 1(4), 2002. [15] Y. Tay and K. Chua, “A Capacity analysis for the IEEE 802.11 MAC protocol,” Wireless Networks, pages 159–171, January 2001. [16] A. W. A. C. W. Heinzelman, A. Sinha, “Energy Scalable Algorithms and Protocols for Wireless Microsensor Networks,” In Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP ’00), June 2000. [17] D. E. W. Ye, J. Heidemann, “An energy-efficient mac protocol for wireless sensor networks,” In Proceedings of the IEEE INFOCOM’02, New York City, NY, June 2002. [18] H. Wu, Y. Peng, K. Long, S. Cheng, and J. Ma, “Performance of Reliable Transport Protocol over IEEE 802.11 Wireless LAN: Analysis and Enhancement,” In Proceedings of IEEE INFOCOM, Nevada, USA, June 2002.