This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
Resource-Aware and Link Quality Based Routing Metric for Wireless Sensor and Actor Networks V. Cagri Gungor∗ , Chellury Sastry† , Zhen Song‡ and Ryan Integlia§ ∗ Georgia
Institute of Technology, Atlanta, GA, 30332, USA Email:
[email protected] † Siemens Corporate Research, Princeton, NJ, 08540, USA Email:
[email protected] ‡ Utah State University, Logan, UT, 84322, USA Email:
[email protected] § Rutgers University, New Brunswick, NJ, 08901, USA Email:
[email protected]
Abstract—This paper presents a resource-aware and link quality based (RLQ) routing metric to address energy limitations, link quality variations, and node heterogeneities in wireless sensor and actor networks (WSANs). The RLQ metric is a combined link cost metric, which is based on both energy efficiency and link quality statistics. The primary objective of the proposed metric is to adapt to varying wireless channel conditions, while exploiting the heterogeneous capabilities in WSANs. Different from most of the existing simulation based studies, this research effort is guided by extensive field experiments of link quality dynamics at various locations over a long period of time using recent sensor platforms, which realistically addresses the real-world wireless communication challenges in WSANs. Performance evaluations, via test-bed experiments, show that the RLQ routing metric achieves high performance in terms of packet reception rate, network throughput and network lifetime.
I. I NTRODUCTION Wireless Sensor and Actor Networks (WSANs) are characterized by the collective effort of heterogenous nodes called sensors and actors [1]. Typically in WSANs, battery-powered sensor nodes collect information from the physical world and deliver their measurements to a central controller (sink node), which determines the event features and sends action commands to resource-rich actor nodes to initiate the actions upon the sensed phenomenon. The collaborative operation of energy constrained sensors and resource-rich actors brings significant advantages over traditional sensing, including improved accuracy; high aggregate intelligence via parallel processing; and expanded spatial coverage of the environment. The realization of these potential gains, however, directly depends on energy efficient and reliable communication capabilities of the deployed sensor/actor network. Recent experimental studies [2], [3], [4] and [5] have shown that in real sensor network deployments, wireless link quality varies over space and time, deviating to a large extent from the idealized unit disc graph models used in network simulation tools. Based on these empirical studies and measurements, it is also found that the coverage area of sensor radios is neither circular nor convex, and packet losses due to fading
and obstacles are common at a wide range of distances and keep varying over time. These studies provide valuable and solid foundations for several sensor network protocols [6], [7] and have guided design decisions and tradeoffs for a wide range of sensor network applications [8]. Although these early studies made many important observations for the problems of reliable data transmission in wireless sensor networks, the challenges of integrating battery-powered sensors with resource-rich actor nodes are yet to be efficiently studied and addressed. First of all, all these experimental studies do not take node heterogeneity into account when making routing decisions; they assume that all nodes are identical in capabilities. This assumption clearly leads to waste of valuable network resources in heterogenous sensor networks, especially in WSANs. Second, since all these studies were conducted, the design space of sensor platforms and their radio hardware have advanced significantly. Recently, many sensor platforms, including Tmote Sky [9] and MicaZ [10], have gravitated towards an international sensor network standard (IEEE 802.15.4 [11]) and even a single radio chip (CC2420 [12]), which provides an additional radio hardware link quality indicator (LQI) to several network services. This newer technology differs significantly from earlier radios and thus, these recent 802.15.4 based sensor platforms may behave differently compared to earlier sensor platforms [13], [14]. Consequently, all these new advances in sensor radio hardware as well as link quality variations and node heterogeneities in WSANs call for new empirical measurements on recent sensor platforms and the design of resource-aware protocols for WSANs. In this paper, we present a resource-aware and link quality based (RLQ) routing metric for WSANs to address the challenges mentioned above as well as to exploit the unique features of WSANs. The RLQ routing metric is a combined link cost metric, which is based on both energy efficiency and link quality statistics. The primary objective of the RLQ metric is to adapt to varying wireless channel conditions, while exploiting the heterogeneous capabilities in WSANs. To accomplish this
1-4244-0353-7/07/$25.00 ©2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
objective, the proposed metric biases the use of resource-rich actor nodes over energy-constrained sensor nodes for packet forwarding and processing in the network. Specifically, the proposed link cost metric captures expected energy cost to transmit, receive and retransmit a packet, while considering the residual energy levels of the sensor nodes. Also, different from most of the existing simulation based studies, our research effort is guided by extensive field experiments of link quality dynamics at various locations over a long period of time using recent sensor network platforms, which realistically addresses the real-world wireless communication challenges in WSANs. In summary, we make the following contributions in this paper: •
•
•
We reveal the spatiotemporal impacts on wireless communication and identify the relationship between dynamics of link quality and radio hardware measurements based on several link quality measurements on recent sensor network platforms. We present a resource-aware link cost metric, which simultaneously considers the energy consumption and link quality statistics as well as the node heterogeneities in the network. An extensive set of test-bed experiments show that the proposed metric balances the energy expenditure and network load across available paths, accounting for energy drain of individual nodes. We consider realistic battery models in our evaluations, which takes non-linear relationship between battery voltage and the remaining energy level into account during the operation of the network. We also make our experimental data publicly available [15], which can help developing realistic link quality and battery models for simulation-based studies.
The remainder of this paper is organized as follows. In Section II, we present the network architecture and describe the design principles and functionalities of the proposed routing metric in detail. Performance evaluation and simulation results are presented in Section III. Finally, the paper is concluded in Section IV. II. L INK Q UALITY M EASUREMENTS IN WSAN S In the following sections, we first describe the network architecture and characteristics of WSANs and then based on these characteristics, we discuss the main design components of the proposed resource-aware and link quality based routing metric in detail. We also present a case study in order to gain more insight regarding the link quality estimation in WSANs. A. Network Architecture Wireless Sensor and Actor Networks (WSANs) are composed of heterogeneous nodes referred to as sensors, controllers and actors. Sensors are low-cost, low-power, multifunctional devices that communicate un-tethered in short distances. Controllers collect and process sensor data and send action commands to the actors in order to perform appropriate actions upon the environment. Hence, in a typical WSAN architecture, controllers and actors are resource-rich devices
(a) Fig. 1.
(b)
Experimental sites: (a) in a corridor, (b) in an office environment.
equipped with better processing capabilities, high transmission power and longer battery life time (or line powered). In WSANs, a large number of sensor nodes, i.e., on the order of hundreds, are deployed in a target area to perform a collaborative sensing task. Such a dense deployment is usually not necessary for controllers and actors, since they are sophisticated devices with higher capabilities that can act on large areas. B. Target Application In this study, we focus on indoor wireless sensor/actor network applications, such as advanced building automation systems. In such an integrated system, several sensor nodes monitor the ambient conditions of the indoor environment to determine when to start or stop heaters and chillers, modulate air dampers, activate pumps for freeze protection. After the sensors detect an event occurring in an indoor building environment, the event data is distributively processed and transmitted to the terminal equipment controllers, which gather, process, and eventually reconstruct the event data and communicate with actors to initiate the actions upon the environment. Due to ever increasing installation and maintenance costs, energy efficient and reliable configuration of such systems significantly reduces the operational expenses. Although these systems bring significant advantages over traditional sensing and facilitate fine-grained monitoring and control of indoor environments within a limited budget, in a case study made by Siemens Building Technologies [16], it is observed that wireless link quality varies over space and time and it has a significant impact on network performance, including network throughput, network lifetime and resource utilization. Therefore, the design of reliable and energy efficient communication protocols is of great importance to provide several economic and operational benefits. This motivated us to design of resource-aware and link quality based routing metrics for WSAN. Note also that although our research effort is motivated by the challenges of building automation applications, the wireless link quality variations and energy limitations are common in several WSAN applications [1] and thus other real-world applications can benefit from our experimental observations and findings.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
PRR vs. LQI
C. Experimental Setup and Results Packet Reception Rate (PRR)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Real Fitted
0 50
60
70
80
90
100
110
LQI
(a) PRR vs. RSSI 1
Packet Reception Rate (PRR)
In WSANs, rapid variations in the wireless channel precludes an efficient mechanism for knowing instantaneous link quality at the time of transmission, thus making it difficult to estimate the instantaneous value of the link quality. Moreover, in bandwidth limited and battery-operated sensor networks, there is a tradeoff between keeping the communication overhead and energy expense at a minimum, which call for wireless channel measurements with high period, and obtaining a reliable estimate of link quality, which requires frequent channel measurements. Striking a good balance in this tradeoff requires a good understanding of the behavior of wireless channel quality during the operation of the network. This motivates us to explore whether it is possible to obtain a good estimate of the link quality based on only a few radio hardware measurements. In this study, we first focused on how to characterize and measure link quality in WSANs. We have conducted experiments with Tmote Sky nodes. Tmote Sky nodes use CC2420 radio component, which are apart from being more advanced than older radios, and supports the IEEE 802.15.4, an emerging wireless sensor network standard [11]. Specifically, CC2420 operates in 2.4 GHz ISM band with a nominal data rate of 250 kbps, a much higher rate than older radios. In our experiments, to measure the link quality during the operation of the network, two useful radio hardware link quality metrics were used: i) link quality indicator (LQI) and ii) received signal strength indicator (RSSI). More specifically, RSSI is the estimate of the signal power and is calculated over 8 symbol periods, while LQI can be viewed as chip error rate and is calculated over 8 bits following the start frame delimiter. LQI values are usually between 110 and 50, and correspond to maximum and minimum quality frames respectively. In link quality measurements, we use a pair of Tmote Sky nodes in an indoor environment, one as the sender and the other as the receiver. In Fig. 1, we present our experimental sites in an indoor environment. We vary the distance from the receiver to the sender from 1 m to 30 m, in steps of 1 m. At each distance, the transmitter sends 100 data packets with a rate of 2 packets per second. We deliberately chose a low rate to avoid any potential interference, so that the effect of unreliable links can be isolated from that of congestion. In Figs. 2 (a) and (b), we present our preliminary experiment results to elaborate the relationship between packet reception rate (PRR) and link quality metrics. Here, packet reception rate represents the ratio of the number of successful packets to the total number of packets transmitted over a certain number of transmissions. We make two observations from these figures. First, in Fig. 2 (a), we observe that there is a clear trend that PRR increases, when average LQI measurements increase. Thus, there is a strong correlation between the average LQI values and the PRR at the receiver. Statistical analysis of our empirical measurements shows that the Pearson’ s correlation coefficient is around 0.80 between these two variables. Note that we also observed some inconsistencies in Fig. 2 (a),
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 −100
−95
−90
−85
−80
−75
−70
−65
RSSI (dbm)
(b) Fig. 2.
Test-bed measurements: a) PRR vs. LQI b) PRR vs. RSSI.
especially when the received signal is weak. These inconsistencies clarify why the Pearson’ s coefficient is not 1.0. However, the observed correlation is still interesting, since LQI is calculated only from the packets that are received, whereas the packet reception rate also considers those packets that are dropped. This correlation implies that average LQI is a good measurable indicator of the packet reception rate. In Fig. 2 (a), we also fitted a curve to the average LQI vs. PRR and observed although the curve fits the data quite good, there are still a few outliers, which can be caused by environmental changes and interference from 802.11 networks in the deployment field. Second, we have observed that there is a smaller correlation between RSSI and the packet reception rate as shown in Fig. 2 (b). The Pearson’ s correlation coefficient is only 0.55 between the packet reception rate and the RSSI values. Specifically, it is found that when the signal is weak (especially when it is around sensitivity threshold (−94dbm)), even though there is a considerable variation in the packet reception rate, RSSI does not provide useful information to predict packet reception rate. On the other hand, when the signal is higher than sensitivity threshold, RSSI is a promising link quality estimator, since it shows small variance compared to LQI measurements. This observation is also consistent with the results in the related
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
literature [14]. Therefore, to minimize the estimation error and link measurement costs, one can use LQI measurements as a link quality metric as long as its variances are factored out. D. Resource Aware and Link Quality Based Routing Metric In WSANs, the wireless link quality between pairs of nodes vary during the lifetime of a network based on distance, transmit power, radio interference and environmental factors (such as obstructions and people in the sensor network field attenuating radio signals) [3]. Even if the locations of nodes in the network are fixed and each node is configured with an identical transmit power, node inter-connectivity changes during the life-time of the network. Note also that in WSANs, the energy limitations of the sensors exacerbate the challenge of reliable wireless communication. In order to alleviate these drawbacks, we present a resourceaware and link quality based (RLQ) routing metric, which is based on both energy efficiency and link quality statistics. RLQ routing metric adapts to varying wireless channel conditions, while exploiting the heterogeneous capabilities in the network. Specifically, it captures expected energy cost to transmit, receive and retransmit a packet, while considering the residual energy levels of the sensor nodes. Moreover, for nodes that have high energy resources, e.g., actor nodes, transmission and reception of packets have negligible energy cost, which is also reflected in the proposed link cost metric. Note that the actors in WSANs have longer network lifetime compared to sensor nodes, since the order of magnitude of the energy required for actions is much higher than that required for sensing and communication. To calculate link cost, let’s assume each node uses CSMA/CA MAC protocol with DATA/ACK exchange, which is supported by IEEE 802.15.4. Then, the energy cost (Clink ) for a reliable transmission of a data packet over a single hop can be typically calculated as follows: Clink = ηtx αtx + ηrx αrx
(1)
where ηtx and ηrx represent normalized energy cost for the transmitter and receiver, respectively. The variables αtx and αrx are 1 for battery-powered sensor nodes and 0 for linepowered actor nodes. In addition, normalized energy costs ηtx and ηrx are calculated as follows: ηtx
y Etx−res = [(Ctx−data + Crx−ack )Elink ] 1 + (1 − ) Etx−init (2) x
y Erx−res ) 1 + (1 − Erx−init (3) where Ctx and Crx represent the energy consumption during transmission and reception, respectively. Also, Etx−init and Erx−res are the initial and remaining energy of the transmitter, and Erx−init and Erx−res are the initial and remaining energy of the receiver, respectively. Here, Elink represents x
ηrx = [(Crx−data + Ctx−ack )Elink ]
the expected number of transmissions, which is calculated as follows: Elink =
K
i (1 − P RR)i P RR
(4)
i=0
where P RR represents the packet reception rate and K is the maximum number of re-transmissions before the packet is ignored. To calculate P RR of a link, we utilize link quality indicator (LQI) reported by the physical layer of IEEE 802.15.4 [11]. This way, the nodes dynamically adapt to changing wireless network conditions and select the paths with high quality links. Note also that with the use of normalized energy cost in the proposed metric, when the sensors have plenty of residual energy, e.g., at the beginning of the network deployment, the energy consumption term in equations (2) and (3) is emphasized, while if the residual energy of a node becomes lower, then the residual energy term is more emphasized. Thus, we try to balance the energy expenditure and network load across available paths, accounting for energy drain of individual nodes. It is also important to note that in equations (2) and (3), the variables x and y are the weighting factors that can be adjusted to find the minimum energy path or the path with nodes having the most energy or combination of above. For example, if x=y=0, then the shortest cost path is minimum hop path and if x = 1 and y = 0, then the shortest path is the minimum total energy consumption path. Thus, these weighting factors provide flexibility to the user based on the application-specific requirements. Note that in energy cost calculation of a link, if both transmitter and receiver are not battery-powered, the equation (1) becomes equal to zero. In order to avoid a link energy cost of zero, we also take the maximum of the calculated link cost and a small constant for these cases. Overall, with the use of proposed link cost metric, we can choose paths that contain as few batterypowered data transmissions and receptions as possible and thus utilizes resource-rich nodes in the deployment field in order to maximize the network lifetime. III. P ERFORMANCE E VALUATION In order to gain more insight into link quality variations and energy limitations in WSANs, we first investigate the effects of link quality indicator (LQI) on the overall network performance. In the first set of experiments, all the nodes in the network are battery-powered. Then, we evaluate the impact of energy heterogeneity in the network, where some nodes (actor nodes) are line-powered and other nodes are battery-powered. Furthermore, we compared the performance of different routing metrics on a physical test-bed, including 21 Tmote Sky nodes. The experiments were carried out in a large office floor with obstructions (people also move during the day) and 802.11b networks to mimic the realistic operating network conditions. Note that some of 802.15.4 frequencies overlap with 802.11b frequencies [17] increasing the effects of external
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
90
90
80
70
60
50
40
30
20
Shortest_Path LQI_Instant LQI_MovAvg RLQ
10
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
600
500
80
70
60
50
40
30
20
Shortest_Path LQI_Instant LQI_MovAvg RLQ
10
0
1
Increase in Network Lifetime (%)
Cumulative percentage of node pairs
100
Cumulative percentage of node pairs
100
0
0.1
0.2
0.3
0.4
0.5
0.6
Throughput (packet/s)
Packet Reception Rate
(a)
(b)
0.7
0.8
0.9
400
300
200
100
1
0
LQI_Inst
LQI_MovAvg
RLQ
RLQ_Act(15%) RLQ_Act(20%) RLQ_Act(25%)
(c)
Fig. 3. Performance results: (a) CDF vs. PRR, (b) CDF vs. Throughput, and (c) The effect of line powered actor nodes on network lifetime. Here, RLQ Act(x%) represents RLQ routing when x% of nodes in the network are line powered actor nodes.
TABLE I N ETWORK PARAMETERS USED IN THE EXPERIMENTS
Area of sensor field Number of nodes Packet length Buffer size Re-transmission threshold (K) Traffic type Transmission power Weighting factors (x, y)
20x30 m2 21 30 bytes 64 packets 5 CBR -25 dbm (1,1)
interference on link quality. The parameters used in our performance measurements are listed in Table I. To communicate sensor data to the sink node, we employed a CSMA/CA MAC protocol with DATA/ACK exchange supported by 802.15.4. In our experiments, we also consider energy heterogeneity in the network and non-ideal battery behaviors. In the evaluations, we investigate the following performance metrics: •
•
•
Throughput is the number of unique packets received at the sink node divided by the interval between the start and the end of the experiment. Packet Reception Rate is the ratio between the total number of unique packets received at the sink node and the total number of packets generated by all the sensor nodes. Network Lifetime is defined as the smallest time that it takes for at least one node in the network to drain its energy beyond the point where it can function normally. We also normalize network lifetime to network throughput to avoid giving unfair advantage to less reliable networks.
In our experiments, each sensor node sends packets towards a sink node with a rate of 1 packet per second. We deliberately chose a low rate to avoid any potential interference and network congestion, so that the effect of unreliable links can be isolated from that of network congestion. Multiple trials
were used in all experiments and the results are the average of these trials. The duration of each experiment trial was at least 1500 seconds. For each experiment, we logged every packet transmission and reception at each sensor node and at the sink node, respectively. This fairly detailed logging helps us to visualize the performance of different routing metrics in several ways. In performance evaluations, we have used multihop LQI routing algorithm [18] in TinyOS, because the code for the the Tmote Sky platform was available. Using multihop LQI routing algorithm, we have also implemented four different routing algorithms: i) shortest path routing algorithm, which we call Shortest P ath; ii) multihop routing algorithm using instantaneous LQI measurements, which we call LQI Instant; iii) multihop routing algorithm using moving average of LQI measurements, which we call LQI M ovAvg; iv) multihop routing algorithm using the proposed RLQ metric, which we call RLQ. In Shortest P ath routing algorithm, each sensor node uses a hop count metric as the sole link cost metric and prefers a short path over (potentially) very poor radio links rather than a longer path over high-quality links. Specifically, when link quality varies significantly, it leads to low network throughput, since it causes limited bandwidth to be consumed by retransmissions. In LQI Instant and LQI M ovAvg routing algorithms, each sensor node maintains a recent history of the LQI measurements to their neighbors and uses link quality estimations to select the parent with the lowest cost path to the sink node. The only difference between LQI Instant and LQI M ovAvg is that LQI M ovAvg uses moving average of LQI measurements with a window size of fifty in order to extract the variances of LQI measurements from the data. In RLQ routing algorithm, the routing metric captures expected energy cost to transmit, receive and retransmit a packet, while considering the residual energy levels of the sensor nodes. This metric takes unreliable wireless links into account, while accounting for residual energy of the sensors in order to extend the network lifetime.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.
Figs. 3 (a) and (b) show the cumulative distributive function (CDF) of the packet reception rate and throughput performance of the routing metrics under comparison for the set of twenty paths, respectively. As shown in Figs. 3 (a) and (b), LQI M ovAvg leads to the highest packet reception rate and average network throughput over all paths, with a median of 60% and 108bps. Compared to LQI M ovAvg, LQI Instant and Shortest P ath are noticeably worse, with a median packet reception rate of 30% and 10% and with a median average network throughput of 48bps and 26bps, respectively. It is also important to note that the network performance of the RLQ routing metric is slightly lower than that of LQI M ovAvg. This is because when the residual energy of a node along the high quality path becomes lower, RLQ metric changes the path to improve the network lifetime. This way, it aims to balance the energy expenditure and network load across available paths, accounting for both link quality variations and energy drain of individual nodes. Overall, these test-bed results show that the routing selection metric has a large impact on overall network performance. The LQI M ovAvg and RLQ routing metric provide high network performance using a simple model mapping from average LQI measurements to packet reception rate. Note that in these experiments, all the nodes in the network are batterypowered. In the second set of experiments, we also investigated the effect of routing metrics and energy heterogeneity on network lifetime. Fig. 3 (c) shows the normalized network lifetime performance of different routing metrics under comparison. Here, we present percentage of network lifetime increase compared to the case when shortest path routing algorithm is used in the network. As shown in Fig. 3 (c), when there is no linepowered actor nodes in the network, the proposed RLQ metric achieves the best performance compared to LQI M ovAvg and LQI Instant routing metrics. For example, we obtain that the average network lifetime increase achieved by RLQ metric is 25% and 15% higher than that of LQI Instant and LQI M ovAvg, respectively. When line powered actor nodes are included in the network (RLQ Act(x%) cases), it is observed that network lifetime increases significantly, since RLQ routing metric biases the use of resource-rich actor nodes over energy-constrained sensor nodes for packet forwarding and processing. For example, when 20% of nodes in the network are line powered, we obtain that the average network lifetime increases 40% compared to the case when RLQ metric is used. In our experiments, we have also observed that the selection of the position of the line-powered actor nodes affects the overall network performance. Therefore, in addition to resource-aware and link quality based routing metrics, optimal deployment strategies should be developed to fully utilize the potential of network heterogeneity. IV. C ONCLUSION Wireless sensor and actor network protocol design is a challenging task due to severe limitations of energy resources and network bandwidth as well as wireless link quality variations
and node heterogeneities in the network. In this paper, to address these challenges, we present a resource-aware and link quality based (RLQ) routing metric for WSANs. The RLQ routing metric is a combined link cost metric, which is based on both energy efficiency and link quality statistics. Based on extensive empirical measurements and test-bed experiments, we also found that there exists a strong correlation between the average LQI measurements and packet reception rates. Comparative performance evaluations via test-bed experiments show that the RLQ routing metric achieves high performance in terms of packet reception rate, network throughput and network lifetime. Future work includes investigating the impact of different heterogenous resources, such as transmission power, network bandwidth and processing power, on overall network performance and optimal placement of these resources in the network. We also plan to make detailed statistical analysis of our link quality measurements. R EFERENCES [1] I. F. Akyildiz and I. H. Kasimoglu, “Wireless Sensor and Actor Networks: Research Challenges,” Ad Hoc Networks (Elsevier), vol. 2, no. 4, pp. 351–367, October 2004. [2] D. Lal and et al., “Measurement and Characterization of Link Quality Metrics in Energy Constrained Wireless Sensor Networks,” in Proc. of IEEE GLOBECOM 2003, San Francisco, USA, December 2003. [3] J. Zhao and R. Govindan, “Understanding Packet Delivery Performance in Dense Wireless Sensor Networks,” in Proc. of ACM SENSYS 2003, CA, USA, November 2003. [4] D. Son, B. Krishnamachari, and J. Heidemann, “Experimental Analysis of Concurrent Packet Transmissions in Low-Power Wireless Networks,” in Proc. of ACM SENSYS 2006, November 2006. [5] G. Zhou, T. He, J. Stankovic, and T. Abdelzaher, “RID: Radio Interference Detection in Wireless Sensor Networks,” in Proc. of IEEE INFOCOM 2005, Miami, USA, March 2005. [6] K. Seada and et al., “Energy-Efficient Forwarding Strategies for Geographic Routing in Lossy Wireless Sensor Networks,” in Proc. of ACM SENSYS 2004, November 2004. [7] A. Woo, T. Tong, and D. E. Culler, “Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks,” in Proc. of ACM SENSYS 2003, CA, USA, November 2003. [8] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless Sensor Networks: A Survey,” Computer Networks (Elsevier), vol. 38, no. 4, pp. 393–422, Mar. 2002. [9] Moteiv Corporation, Tmote Sky datasheet. [Online]. Available: www.moteiv.com/products/docs/tmote-sky-datasheet.pdf [10] Crossbow Technology, MicaZ datasheet. [Online]. Available: www.xbow.com/Products/Product pdf files/Wireless pdf/MICAz Datasheet.pdf [11] “Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LRWPANs),” IEEE 802.15.4 Standard, Oct. 2003. [12] CC2420 Radio. [Online]. Available: http://www.chipcon.com [13] J. Polastre, R. Szewczyk, and D. Culler, “Telos: Enabling Ultra-low Power Wireless Research,” in Proc. of IEEE ISPN 2005, April 2005. [14] K. Srinivasan and P. Levis, “RSSI is Under Appreciated,” in Proc. of EMNET 2006, May 2006. [15] Experiment measurements at Siemens Sensor Network and RFID Lab. [Online]. Available: http://grumpy.usu.edu/∼zhensong/ ExperimentMeasurements/ [16] T. Kevan, “Case Study: When Renovation Includes Building Automation,” in Proc. of Sensors Magazine, May 2006. [17] I. Howitt and J. Gutierrez, “IEEE 802.15.4 Low Rate-Wireless Personal Area Network Coexistence Issues,” in Proc. of IEEE WCNC 2003, March 2003. [18] TinyOS MultiHopLQI routing algorithm. [Online]. Available: www. tinyos.net/tinyos-1.x/tos/lib/MultiHopLQI/
