Academia.eduAcademia.edu

Figure 25 – uploaded by John Stankovic

See full PDFdownloadDownload figure
Fig. 27. Energy Consumption vs. Num of Required Report 8. RELATED WORK Real-time protocols play an important role to guarantee the effectiveness of the in- teractions between wireless sensor networks and the physical world. RAP [Lu et al. 2002] uses a novel velocity monotonic scheduling to prioritize the real-time traf- fic and enforce such prioritization through a differentiated MAC Layer. Woo and Culler [Woo and Culler 2001] propose an adaptive rate control scheme to achieve fairness among the nodes with different distances to a base station. Huang [Huang et al. 2003] et al. propose the Mobicast protocol to provide just-in-time information dissemination to nodes in a mobile delivery zone. Given the complete knowledge of traffic pattern, Li [Li et al. 2005] proposes a SLF message scheduling algorithm to exploit spatial channel reuse, so that deadline misses can be reduced. The Lightning protocol [Wang et al. 2004] localizes the acoustic source with a bounded delay regardless of the node density. Carley [Carley et al. 2003] designs a periodic message scheduler to provide a contention-free predicable medium access control. Somasundara [Somasundara et al. 2004] proposes a mobile agent scheduling algo- rithm to collect the buffered sensor data, before the buffer overflow occurs at the sensor nodes. Vasudevan [Vasudevan et al. 2003] proposes an application-specific compression for time delay estimation in sensor networks, and He [He et al. 2004] adaptively performs application independent data aggregation in a time sensitive manner. Nam [Nam et al. 2005] proposes time-parameterized sensing task model for real-time tracking. Eswaran [Eswaran et al. 2005] designs and implements a reservation-based real-time operating system, with multi-hop networking support for use in wireless sensor networks. Yang [Yang and Vaidya 2004] proposes a wakeup scheme that assists balancing energy saving and end-to-end delay. Felemban [Felem-

Figure 27 Energy Consumption vs. Num of Required Report 8. RELATED WORK Real-time protocols play an important role to guarantee the effectiveness of the in- teractions between wireless sensor networks and the physical world. RAP [Lu et al. 2002] uses a novel velocity monotonic scheduling to prioritize the real-time traf- fic and enforce such prioritization through a differentiated MAC Layer. Woo and Culler [Woo and Culler 2001] propose an adaptive rate control scheme to achieve fairness among the nodes with different distances to a base station. Huang [Huang et al. 2003] et al. propose the Mobicast protocol to provide just-in-time information dissemination to nodes in a mobile delivery zone. Given the complete knowledge of traffic pattern, Li [Li et al. 2005] proposes a SLF message scheduling algorithm to exploit spatial channel reuse, so that deadline misses can be reduced. The Lightning protocol [Wang et al. 2004] localizes the acoustic source with a bounded delay regardless of the node density. Carley [Carley et al. 2003] designs a periodic message scheduler to provide a contention-free predicable medium access control. Somasundara [Somasundara et al. 2004] proposes a mobile agent scheduling algo- rithm to collect the buffered sensor data, before the buffer overflow occurs at the sensor nodes. Vasudevan [Vasudevan et al. 2003] proposes an application-specific compression for time delay estimation in sensor networks, and He [He et al. 2004] adaptively performs application independent data aggregation in a time sensitive manner. Nam [Nam et al. 2005] proposes time-parameterized sensing task model for real-time tracking. Eswaran [Eswaran et al. 2005] designs and implements a reservation-based real-time operating system, with multi-hop networking support for use in wireless sensor networks. Yang [Yang and Vaidya 2004] proposes a wakeup scheme that assists balancing energy saving and end-to-end delay. Felemban [Felem-

Related Figures (24)

Fig. 1. The Delay Breakdown in Tracking Operation
Fig. 1. The Delay Breakdown in Tracking Operation
Fig. 2. Detection Probability for fast targets shown in the Figure 2(a). The size of the area is 295R-L +7 - SR?/2, where SR is the Sensing Range. For a single node located at (x,y) in this area, the probability that the node detects the target P(x,y) is SDC + I(x, y)/(P- TS) if SDC + l(a,y)/(P + TS) < 1, where I(x, y) is the overlapping length of the node’s sensing range and the target’s trace, and T'S is the Target Speed. If we consider all possible locations in this area, we can get P, in Equation 1 by integrating and normalizing P(x,y) over the area. We note that when (x,y) is in the circle (area A), as shown in Figure 2(b), [(z,y) = SR? — y?+L—«. When (z, y) is in area B, l(a, y) = 2. SR? — y?. Propc. MSRe=yP4h-2)\ 7.1 prong 2VSRe=¥? a,
Fig. 2. Detection Probability for fast targets shown in the Figure 2(a). The size of the area is 295R-L +7 - SR?/2, where SR is the Sensing Range. For a single node located at (x,y) in this area, the probability that the node detects the target P(x,y) is SDC + I(x, y)/(P- TS) if SDC + l(a,y)/(P + TS) < 1, where I(x, y) is the overlapping length of the node’s sensing range and the target’s trace, and T'S is the Target Speed. If we consider all possible locations in this area, we can get P, in Equation 1 by integrating and normalizing P(x,y) over the area. We note that when (x,y) is in the circle (area A), as shown in Figure 2(b), [(z,y) = SR? — y?+L—«. When (z, y) is in area B, l(a, y) = 2. SR? — y?. Propc. MSRe=yP4h-2)\ 7.1 prong 2VSRe=¥? a,
Fig. 3. Detection Probability for slower targets We note that P, calculated by Equation 1 is valid only when the target speed is faster than or equal to 2SR/(P — P-SDC). We define this as a fast target. a target with a speed slower than 25R/(P —P-+-SDC), which we define as a s target, it may happen that for a node located at (x,y), the corresponding I(x is greater than (P — P- SDC)- TS, so that SDC + I(x, y)/(P-TS) > 1 the node at this location, the SDC + l(a, y)/(P- TS). slow targets. We define variab 4 probability that it detects the target is 1 instead herefore, we need to revise the result in Equation 1 ea such that SDC + 2a/(TS': P) = 1. In Figure 3, For low +Y) For of for it can be easily proven that if a node appears inside area C bounded by the dashed arc and lines, the probability t hat it detects the target is 1. Note that the distance between the dashed line and the target trace is VSR? —a?. The dashed arc is centered at (ZL — 2a,0) and its circle with radius SR centered at (L,0) into area A’ and area B’. The detecti radius is SR. The rest of the area is divided by the 10n probabilities for nodes in area AM and area B’ have the same forms as those for the fast target case, correspondingly. Then we have nodes in area A and area B in
Fig. 3. Detection Probability for slower targets We note that P, calculated by Equation 1 is valid only when the target speed is faster than or equal to 2SR/(P — P-SDC). We define this as a fast target. a target with a speed slower than 25R/(P —P-+-SDC), which we define as a s target, it may happen that for a node located at (x,y), the corresponding I(x is greater than (P — P- SDC)- TS, so that SDC + I(x, y)/(P-TS) > 1 the node at this location, the SDC + l(a, y)/(P- TS). slow targets. We define variab 4 probability that it detects the target is 1 instead herefore, we need to revise the result in Equation 1 ea such that SDC + 2a/(TS': P) = 1. In Figure 3, For low +Y) For of for it can be easily proven that if a node appears inside area C bounded by the dashed arc and lines, the probability t hat it detects the target is 1. Note that the distance between the dashed line and the target trace is VSR? —a?. The dashed arc is centered at (ZL — 2a,0) and its circle with radius SR centered at (L,0) into area A’ and area B’. The detecti radius is SR. The rest of the area is divided by the 10n probabilities for nodes in area AM and area B’ have the same forms as those for the fast target case, correspondingly. Then we have nodes in area A and area B in
Fig. 4. Initial Delay vs. SDC
Fig. 4. Initial Delay vs. SDC
Fig. 5. Detection Confidence vs. Detection Delay According to Equation 6, the sub-deadline Dgetection can be decided by choosing a desired P(t > Daetection) value. In addition, we desire to know how a detection algorithm performs under a given sub-deadline Dyetection. We define the Detection Confidence (DC), as the confidence on the target detection, i.e., 100% DC indicates this sensor has no doubt about the existence of the target. Normally, the longer Daetection is, the more information about target signature a sensing algorithm can obtain, and therefore, it can achieve a higher detection confidence DC. Such relationship depends on the type of sensors. n order to quantitatively analyze the relation between DC and Dagetection aS a case study, we performed experiments on XSM motes with the magnetic sensing algorithm detecting a moving vehicle in an outdoor environment. We approximate the sensing range as 7 meters around the sensor node, according to empirical data. Figure 5 plots the relation between the detection confidence and the detection delay, based on the experiments. As we can see from the figure, DC does not have a linear relation to Ddetection. Based on experimental measurements, we use a polynomial to characterize DC versus Detection. Figure 5 shows a series of polynomials of different orders that fit the points representing the relation between the detection confidence and the detection delay. The plotting indicates that the polynomials of an order higher than 5 are fairly close to each other and fit the points well. Hence,
Fig. 5. Detection Confidence vs. Detection Delay According to Equation 6, the sub-deadline Dgetection can be decided by choosing a desired P(t > Daetection) value. In addition, we desire to know how a detection algorithm performs under a given sub-deadline Dyetection. We define the Detection Confidence (DC), as the confidence on the target detection, i.e., 100% DC indicates this sensor has no doubt about the existence of the target. Normally, the longer Daetection is, the more information about target signature a sensing algorithm can obtain, and therefore, it can achieve a higher detection confidence DC. Such relationship depends on the type of sensors. n order to quantitatively analyze the relation between DC and Dagetection aS a case study, we performed experiments on XSM motes with the magnetic sensing algorithm detecting a moving vehicle in an outdoor environment. We approximate the sensing range as 7 meters around the sensor node, according to empirical data. Figure 5 plots the relation between the detection confidence and the detection delay, based on the experiments. As we can see from the figure, DC does not have a linear relation to Ddetection. Based on experimental measurements, we use a polynomial to characterize DC versus Detection. Figure 5 shows a series of polynomials of different orders that fit the points representing the relation between the detection confidence and the detection delay. The plotting indicates that the polynomials of an order higher than 5 are fairly close to each other and fit the points well. Hence,
time for a radio module to detect the existence of the radio signal. For example, the CC2420 radio transceiver takes at least 2ms (8 symbol periods, as specified by 802.15.4 [IEEE ]) to access the radio activity. Based on TP and CCA, we can get the Non-Sentry Duty Cycle (NSDC) as ccs. At the sentry side, once a sentry detects a target, it broadcasts a radio message with a long preamble. This long preamble is guaranteed to be sensed by neighboring non-sentry nodes as long as this preamble has a length equal to or longer than the toggle period of non-sentry nodes. The worst case wake-up delay WC peiay equals TP. In other words, the sub- deadline Dywakeup can be ensured trivially in our design by setting T’7P = Dwakeup. Let the power consumption for an active node during a unit of time be E, the energy consumption for a non-sentry node is £ coe Since the amount of time tc check the radio activity (CCA) is constant for a specific radio hardware, the length of the toggle period determines the energy consumption rate in non-sentry nodes. In general, a long toggle period T’'P leads to a low energy consumption, however, it also leads to a long delay in waking up the non-sentry nodes. Figure 7 shows such a tradeoff, using the CC1000 radio transceiver for MICA2/XSM motes as an example. As shown in Figure 7, a sub-deadline of 200ms lead to a 99% energy saving for the non-sentry nodes.
time for a radio module to detect the existence of the radio signal. For example, the CC2420 radio transceiver takes at least 2ms (8 symbol periods, as specified by 802.15.4 [IEEE ]) to access the radio activity. Based on TP and CCA, we can get the Non-Sentry Duty Cycle (NSDC) as ccs. At the sentry side, once a sentry detects a target, it broadcasts a radio message with a long preamble. This long preamble is guaranteed to be sensed by neighboring non-sentry nodes as long as this preamble has a length equal to or longer than the toggle period of non-sentry nodes. The worst case wake-up delay WC peiay equals TP. In other words, the sub- deadline Dywakeup can be ensured trivially in our design by setting T’7P = Dwakeup. Let the power consumption for an active node during a unit of time be E, the energy consumption for a non-sentry node is £ coe Since the amount of time tc check the radio activity (CCA) is constant for a specific radio hardware, the length of the toggle period determines the energy consumption rate in non-sentry nodes. In general, a long toggle period T’'P leads to a low energy consumption, however, it also leads to a long delay in waking up the non-sentry nodes. Figure 7 shows such a tradeoff, using the CC1000 radio transceiver for MICA2/XSM motes as an example. As shown in Figure 7, a sub-deadline of 200ms lead to a 99% energy saving for the non-sentry nodes.
Fig. 7. Wakeup Delay Vs. Non-Sentry Energy Saving
Fig. 7. Wakeup Delay Vs. Non-Sentry Energy Saving
Specifically, we organize the nodes in the We use a semi-dynamic leader election [Luo vicinity of a target into one group. et al. 2005] to minimize the delay. Nodes that detect the target become the group members, which, upon detection, immediately report their own locations and sensing data to a leader. The leader then averages the locations of members as t he estimates of the target positions, and sends such estimates to a base station. To filter out the sporadic false alarms of individual nodes, we introduce a configurable parameter, DOA (Degree of Ag- gregation), which forces the leader to withho d reports to a base station until the number of received member reports reaches DOA. To achieve a high confidence in target detection, one should set a high DOA value (e.g., 4). On the other hand, a higher DOA value inevitably introduces a onger group aggregation delay since the leader waits longer to expect more member reports. This tradeoff allows us to choose appropriate DOA to meet the sub-deadline Daggregation: Fig. 8. The Detection Areas Before and After Movement
Specifically, we organize the nodes in the We use a semi-dynamic leader election [Luo vicinity of a target into one group. et al. 2005] to minimize the delay. Nodes that detect the target become the group members, which, upon detection, immediately report their own locations and sensing data to a leader. The leader then averages the locations of members as t he estimates of the target positions, and sends such estimates to a base station. To filter out the sporadic false alarms of individual nodes, we introduce a configurable parameter, DOA (Degree of Ag- gregation), which forces the leader to withho d reports to a base station until the number of received member reports reaches DOA. To achieve a high confidence in target detection, one should set a high DOA value (e.g., 4). On the other hand, a higher DOA value inevitably introduces a onger group aggregation delay since the leader waits longer to expect more member reports. This tradeoff allows us to choose appropriate DOA to meet the sub-deadline Daggregation: Fig. 8. The Detection Areas Before and After Movement
under different target speeds. Figure 9 gives a more concrete design space by depicting the group aggregation delay for varied DOA values and target speeds when the sensing range is 10m, the node density is 1 per 100 m?. We note that this result is consistent with the results obtained from the large-scale simulations presented in Section 7. Fig. 9. Minimal Group Aggregation Delay for Varying DOA and Target Speed 4.5 Communication Delay and Its Tradeoff
under different target speeds. Figure 9 gives a more concrete design space by depicting the group aggregation delay for varied DOA values and target speeds when the sensing range is 10m, the node density is 1 per 100 m?. We note that this result is consistent with the results obtained from the large-scale simulations presented in Section 7. Fig. 9. Minimal Group Aggregation Delay for Varying DOA and Target Speed 4.5 Communication Delay and Its Tradeoff
Fig. 10. Random Deployment Performance We first consider the random deployment. We derive the probability in question as follows. Consider an arbitrary point Q under question. Since each base can serve a radius of L, once a base station is deployed, we know that the point Q has a probability of mie of being located inside this base’s service radius. Therefore, once N bases are deployed, the coverage probability for an arbitrary point @ is 1-(1- nL?\N . Notice that this derivation does not take into account the boundary Ss effect. This approximation is valid when S$ >> 7L?, as verified in our experiment.
Fig. 10. Random Deployment Performance We first consider the random deployment. We derive the probability in question as follows. Consider an arbitrary point Q under question. Since each base can serve a radius of L, once a base station is deployed, we know that the point Q has a probability of mie of being located inside this base’s service radius. Therefore, once N bases are deployed, the coverage probability for an arbitrary point @ is 1-(1- nL?\N . Notice that this derivation does not take into account the boundary Ss effect. This approximation is valid when S$ >> 7L?, as verified in our experiment.
Fig. 11. Regular Deployment Strategies We focus on two types of grid strategies, square based and hexagon based. These strategies are shown in Figure 11.
Fig. 11. Regular Deployment Strategies We focus on two types of grid strategies, square based and hexagon based. These strategies are shown in Figure 11.
Fig. 12. Deployment Site
Fig. 12. Deployment Site
Fig. 13. Various Delays Measurements from Field Test
Fig. 13. Various Delays Measurements from Field Test
Fig. 15. Energy Consumption vs. Target Speed
Fig. 15. Energy Consumption vs. Target Speed
Fig. 16. Delays under Varying Detection Delay
Fig. 16. Delays under Varying Detection Delay
relative small, this system parameter does not noticeably affect the overall energy consumption, as shown in Figure 17.
relative small, this system parameter does not noticeably affect the overall energy consumption, as shown in Figure 17.
Fig. 18. Delays vs. Sentry Duty Cycle 7.4 Performance vs. Sentry Duty Cycle Fig. 17. Energy Consumption vs. Detection Delay
Fig. 18. Delays vs. Sentry Duty Cycle 7.4 Performance vs. Sentry Duty Cycle Fig. 17. Energy Consumption vs. Detection Delay
linearly with the SDC, which indicates that an efficient sentry scheduling algorithn is beneficial. Fig. 20. Delays vs. Non-Sentry Duty Cycle
linearly with the SDC, which indicates that an efficient sentry scheduling algorithn is beneficial. Fig. 20. Delays vs. Non-Sentry Duty Cycle
Fig. 21. Energy Consumption vs. Non-Sentry Duty Cycle
Fig. 21. Energy Consumption vs. Non-Sentry Duty Cycle
Fig. 19. Energy Consumption vs. Sentry Duty Cycle
Fig. 19. Energy Consumption vs. Sentry Duty Cycle
shown in Figure 21. This is because the non-sentry nodes are by far the majority, so a duty-cycle increase for the non-sentry nodes leads to a quick increase in the total energy. This result indicates that it is beneficial to increase the wake-up delay, when possible, in exchange for the energy saving. 7.6 Performance vs. DOA
shown in Figure 21. This is because the non-sentry nodes are by far the majority, so a duty-cycle increase for the non-sentry nodes leads to a quick increase in the total energy. This result indicates that it is beneficial to increase the wake-up delay, when possible, in exchange for the energy saving. 7.6 Performance vs. DOA
Fig. 25. Energy Consumption vs. Sensing Range
Fig. 25. Energy Consumption vs. Sensing Range
Fig. 26. Delays vs. Num of Required Reports
Fig. 26. Delays vs. Num of Required Reports
Related topics:
Computer ScienceDesignReal Time ComputingReal Time SystemsComputer communication networksReal time and Embedded SystemsLarge Scale Simulation
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved