SensorNet Paper Reviews by mty3
From Ece
- Return to the paper reading schedule or the Sensor Networks homepage.
All reviews present on this page conducted by Matthew Young (mty3 AT msstate DOT edu)
Reviews
January 22, 2007
Connecting the Physical World with Pervasive Networks
D. Estrin, D. Culler, K. Pister, and G. Sukhatme, Connecting the Physical World with Pervasive Networks , IEEE Pervasive Computing, pp. 59-69, January-March 2002.
This paper presents challenges and opportunities for employing pervasive, yet invisible, sensor-rich computing systems into to the physical world. The author address issues such as size, runtime-self organization, limited access, and sampling. The author cited that the recent ability to implement the radio or optical transceiver into the same package as the processor has allowed for advances in size as well as power. In addition to the previously stated challenges, the author offered that location information an increasingly important aspect. The author states, “Localization is a main part of registration between the virtual and physical worlds.” While the author offered main challenges that senor networks face, the one challenge that seemed to be most prevalent was self-organization. Because of the variety of application that sensor networks can be applied, the ability for the network to spatially confront is key to its success. While the author provided vast amounts of challenges that sensor nets face, the one challenge that was not addressed was security. In addition to presenting the aspect of securing data, another challenge that sensor networks face is sensor drift. Sensor drift along with security are major issues that need to be addressed in order to help the advancement of sensor networks. Back to top
Sensor Networks for Emergency Response: Challenges and Opportunities
K. Lorincz, D. Malan, T.R.F. Fulford-Jones, A. Nawoj, A. Clavel, V. Shnayder, G. Mainland, S. Moulton, and M. Welsh, Sensor Networks for Emergency Response: Challenges and Opportunities, In IEEE Pervasive Computing, Special Issue on Pervasive Computing for First Response, Oct-Dec 2004.
This paper focused on a solution for capturing, processing, and communicating critical data to first responders in the event of a disaster. Examples to the critical data that was presented were heart rate, blood oxygen saturation, and location-tracking.
The paper suggested the use of a developed software infrastructure that integrates sensor nodes along with other wireless device to facilitate an ad hoc network used in a disaster response setting. This software infrastructure has been tagged CodeBlue. Additionally, the authors developed a radio frequency based localization system, MoteTrack, which allows rescuers to determine their location within a building and track patients. The use of CodeBlue would allow first responders, such as EMTs and firefighters, to placing sensing node on patients and have data delivered to a PDA where it’s sorted based on the severity of the patients conditions. The MoteTrack system uses a tree based message routing system along with know reference signatures within the building to give an estimated location of patients and/or rescue personnel within a building.
The paper did a good job presenting data about the vital sign monitors and the location tracking system. In particular the paper explored different methods of security that could be used with the vital sign meter. The authors weighted storage needed for using public keys, as well as the time required to generate security key for different encryption methods.
While the paper did a good job of explaning the operation of the MoteTrack system, there was no information presented in regard to methods for ensuring transmission security. Additionally there was no mention of how the data transmitted from the beacon nodes of the MoteTrack system was interpreted.
I feel that the vital sign monitor would benefit more from a tree based routing schema due the large amount of data that will be broadcast using an ad hoc network. Allowing a set number of sensor to communication with one another and report to a higher network component could possible deduce some latency associatied with the messages traveling through the ad hoc network. Back to top
Habitat Monitoring with Sensor Networks
R. Szewczyk, E. Osterweil, J. Polastre, M. Hamilton, A. Mainwaring, and D. Estrin, Application Driven Systems Research: Habitat Monitoring with Sensor Networks, Communications of the ACM, Special Issue on Sensor Networks. June 2004.
This paper focused on a solution for gathering data directed at climate and habitat utilization. Previously this data was extrapolated using a few or even individual measurements that were then extrapolated in order to have a somewhat full representation. The authors proposed the use of densely deployed sensor networks that could sample climatic and habitat changes in order to provide a finer-grain of information to be used to accurately model and predict future changes. The paper did a good job exploring different methods of duty cycling the sensor nodes. Additionally the paper mentioned ways that the nodes could perform heath monitoring. One such method discussed was the use of a sensor internal to some nodes that could measure a change in relative humidity which would indicate that the external shell of the node had been compromised. One weakness that I found in the paper was a lack of information regarding routing methods after nodes were removed from the network. The initial configuration of the network was stated to be a tree based network routing topology. Removing a parent node in the tree would cause that parent’s children to loss communication with node higher up in the tree. In the latter portion of the paper a statement was made regarding possible advances that could be made. The statement was, “individual nodes should be designed to sense changes and trigger subsequent processing both locally and in neighboring nodes.” This could be using the tree based topology with an individual node notifying its parent to initiate processing on neighboring nodes. Back to top
Towards a Sensor Network Architecture: Lowering the Waistline
D. Culler, P. Dutta, C.T. Ee, R. Fonseca, J. Hui, P. Levis, J. Polastre, S. Shenker, I. Stoica, G. Tolle, and J. Zhao, Towards a sensor network architecture: Lowering the waistline. Proceedings of Hot Topics in Operating Systems (HotOS '05), 2005
This paper addresses the current lack of architecture for sensor networks. Current sensor networks are designed such that an individual set of components are able to work together, but are unable to interoperate with other systems. The authors propose a communication architecture, called the Sensor-net Protocol (SP), that will be a foundation that sensor network application may be built. The authors proposed the design of SP to target issues such as discovery, topology management, naming, and many others. This paper provided a fair amount of challenges that the SP architecture faces. It does a good job of identifying some of the main concerns such as address-free and name-based networks, as well as cross-layer services. Although there were a great deal of concerns address there was not sufficient information provided on ways that SP would overcome some of the concerns. One example is the concern of address-free and name-based network design. The paper presented the reason for concerns but did not present a means by which the SP architecture would handle the concern. One change that I could see to be beneficial to the architecture is a standard security layer, which is built above or within the data link layer. Because of the variety of application that can be addressed using sensor networks having a standardized security layer would alleviate some inconsistencies across senor application. Back to top
January 29, 2006
Just-In-Time Sensor Networks
G. Yee, B. Shucker, J. Dunn, A Sheth, R Han, "Just-In-Time Sensor Networks", IEEE Workshop on Embedded Networked Sensors (EmNets) 2006, pp. 6-10.
This paper discussed the idea of run-time deployment of nodes in a sensor network. A process by which nodes could be deployed in waves as real-time data is analyzed or nodes could be deployed in order to reduce congestion and optimize redundancy. The first process was illustrated using a search and rescue operation following a mudslide. The paper states the use of three different waves of nodes. The first wave gives a broad coverage of the disaster area, the second wave is used to concentrate more node in a area based on run-time data, while the final wave would consist of a higher cost seismic sensors to provided more detailed data and call rescuers.
In addition to providing real-time deployment, the paper examined over-the-air reprogramming and load balancing of routes. By using what is called a data ferry for a means of delivering large payloads, such as new programs, nodes that were not concerned with the information were not affected.
This paper provided very little technical backing to the concepts that were presented. There was no relevant information that explained how the JIT deployer would calculated and position to the coordinates of interest. Additionally there was little information provided about how message traffic was handled, and what form of network topology would be used.
This paper would benefit greatly from a more existence explanation of the JIT system. In particularly, more information needs to be directed on how the deployer decides where the nodes should be placed. Also there should be additional information provided regarding the placement of nodes for reducing congestion and optimizing redundancy. Back to top
MANTIS OS: An Embedded Multithreaded Operating System for Wireless Micro Sensor Platforms
S. Bhatti, J. Carlson, H. Dai, J. Deng, J. Rose, A. Sheth, B. Shucker, C. Gruenwald, A. Torgerson, R. Han, "MANTIS OS: An Embedded Multithreaded Operating System for Wireless Micro Sensor Platforms, " ACM/Kluwer Mobile Networks & Applications (MONET), Special Issue on Wireless Sensor Networks, vol. 10, no. 4, August 2005
This paper proposes the use of a multithreaded preemptively time-sliced operating system approach for the sensor networks. The proposed operating system differs from our currently used operating system in that it is developed using a power-efficient scheduler and allows for implementation of sensor specific features such as dynamic reprogramming.
The MANTIS OS (MOS) conducts threading mush like POSIX threads in UNIX. MOS has a priority based thread scheduler that operates in a round robin fashion within the priority level. Power management is achieved in the MOS system by the use of different sleep stages based on the current activity of the processor. For example, if there is not a thread to execute the processor will enter a deep sleep in order to conserve power. If the system is suspended on I/O it will enter a moderate idle sleep mode. MOS also allows for threads that were in the ready queue prior to a sleep to have a high priority after the processor awakes.
The paper presented the idea of dynamic reprogramming but failed to discuss the ways that MOS would implement the concept. The paper stated that the MOS system would allow two ways for reprogramming the system. The first was to physically connect the node to a serial communication. The second is an over-the-air deployment. While the first approach was exampled in detail, little information about the second approach was provided.
While the paper presented vast amounts of information on the scheduler, power management concepts, and in-situ configuration via a remote shell access there was no consideration placed on the concept of data security. As sensor networks become more popular, a greater concern will be placed on securing data transmitted throughout the network. Back to top
The Emergence of Networking Abstractions and Techniques in TinyOS
P. Levis, S. Madden, D. Gay, J. Polastre, R. Szewczyk, A. Woo, E. Brewer and D. Culler, The Emergence of Networking Abstractions and Techniques in TinyOS, USENIX/ACM Symposium on Networked Systems Design and Implementation (NSDI). 2004.
This paper give a discussion of current networking techniques and abstracts that appear in TinyOS used by different research groups. The paper surveyed CVS trees for six research group and found that network abstraction fall into one of four categories, general, specialized, in flux, or absences. In addition to abstractions, the paper identified four system design techniques used in the abstractions. These techniques include, cross-layer control, static resource allocation, snooping, and scheduled communication.
The paper gives an overview of TinyOS, major developments in hardware platforms, and gives three sensor network applications, which use TinyOS. The paper also describes in some detail single-hop communication, multi-hop communication, and systems services that were would in the OS implementation surveyed.
The paper presents a well-defined set of abstractions and techniques used in the development of sensor networks. Although most senor network applications developed currently are specific to one application the need for a higher level abstraction will become more important as sensor nets become prevalent. Back to top
A Dynamic Operating System for Sensor Networks
C.-C. Han, R.K. Rengaswamy, R. Shea, E. Kohler and M. Srivastava. A Dynamic Operating System for Sensor Networks, International Conference on Mobile Systems, Applications, and Services (MobiSys), 2005.
This paper presents a new concept operation system for use in sensor network application. Unlike other sensor network application, such as TinyOS and Mante Bombilla virtual machine, SOS takes a more dynamic approach to design. SOS uses dynamically loaded modules and a common kernel, which implements messaging, dynamic memory, and module loading and unloading.
The paper presents a high level, in-depth explains of how SOS achieves its dynamic approach to an operating system. In particular the paper discuss module structure, interaction, and insertion and removal, as well as potential failure modes that may occur during operation. The paper also provides an evaluation of SOS against TinyOS and Mante Bombilla. The evolution was conducted in two separate areas, CPU overhead and code updates. After evaluation, the authors found that SOS preformed between TinyOS and Mante in both tests. The major difference was noted between SOS and TinyOS when it came to code update. SOS consumed around 100 times less energy than TinyOS.
While the paper present an argument for the use of a dynamic operation systems there is still portions of the concept that need to be more detailed. Before the idea of dynamically allocated memory becomes widely observed in sensor network research efforts will need to be focused on methods to ensure memory integrity and alternative methods to failure handling that was mentioned in this paper. While the concept of a dynamic sensor operating system has its advantages, the failure to correctly ensure the integrity of the allocated memory may become the pitfall to such a system. Back to top
February 12, 2007
On calculating power-aware connected dominating sets for efficient routing in ad hoc wireless networks
Jie Wu, Ming Gao and Ivan Stojmenovic, On calculating power-aware connected dominating sets for efficient routing in ad hoc wireless networks, submitted for publication, 2000.
This paper focused on a extension of a distributed algorithm for calculating a connected dominating set an ad-hoc network seen here [1]. The extension of the algorithm made selections based on node degree and the current energy level of each host. The paper outlined the algorithms presented in [1], as well as presenting the conditions that vertex (node) must meet in the revised algorithm.
The focus of the revised algorithm was to produce a dominating-set route that contains an optimal number of gateway nodes, while preserving power consumption within the network. This process was completed in two distinct rule sets. The authors refer to these sets of rules as Node-Degree-Based and Energy-Level-Based rules. The first is an attempt to reduce the size of a connected dominating set that is generated during the marking process. The second rule sets attempts to reduce the size of a connected dominating set generated during marking, while at the same time making decisions that will prolong the life of the network.
The authors presented an evaluation of the revised algorithm and compared the number of dominating sets generated and the total number so of message intervals completed before a percentage of node depletion. The results found during simulation showed that the rule sets imposed provided for a smaller number of dominating sets as well as prolonging the life of the network.
One analysis that I felt was omitted from the paper was the amount of time and memory that such a routing concept would entail. As sensor networks become more popular, more emphasis will need to be placed on the amount of storage and processing power that such a routing protocol will require.
References:[1] J. Wu and H.Li, “On calculating connected dominating set for dfficient routing in ad hoc wireless networks,” in Proc. Of the 3rd Int’l Workshop on Discrete Algorithms and Methods for Moblile Computing and Commun., 199, pp. 7-14.
Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks
A. Woo, T. Tony, and D. Culler, Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks, ACM Conference on Embedded Networked Sensor Systems (SenSys), November 2003.
An emerging problem in sensor networks is the reliable and self-configuration of links in multihop wireless networks. This paper presents an evaluation of link quality characteristics, neighborhood management, and routing in dense sensor networks.
The paper evaluated many different processes for estimating the link quality. The authors also introduced a new estimator, window mean with exponentially weighted moving average (WMEWMA). This process computes an average success rate over a specified time period and then smoothens the average with an EWMA. Of the estimators that were tested, they found that the newly developed WMEWMA preformed the best. Next the paper evaluated policies of neighborhood management. The policy found to be most efficient, FREQUENCY, was able to maintain a subset of good neighbors over a fixed table size. Even at nodes densities of 200 FREQUENCY was about to maintain a neighbor list of containing around 25 nodes when the table size was set to 40. Final the paper evaluated different from of routing protocols including Shortest Path (SP), Broadcast, and Minimum Transmission (MT). During evaluation the paper showed that MT preformed above 50% reliable even at high node densities.
One aspect that I felt the paper failed to evaluate is how costly these methods are with respect to power. The main purpose of sensor networks is to sample data from phenomena that may not be readily accessible to humans. Meaning that node will have to operate for long periods of time with little to no human intervention.
Highly dynamic Destination-Sequenced Distance-Vector routing (DSDV) for mobile computers
C.E. Perkins and P. Bhagwat, Highly dynamic Destination-Sequenced Distance-Vector routing (DSDV) for mobile computers , SIGCOMM Symposium on Communications Architectures and Protocols (SIGCOMM), September 1994.
This paper presented a new routing protocol that looks to reduce some of the problems present in today protocols for ad-hoc networks. The protocol they proposed, Destination-Sequenced Distance Vector (DSDV) Protocol, is designed as an attempt to reduce the amount of loop-back paths created in an ad-hoc sensor network.
DSDV uses routing tables stored on each station to list all available destinations and the corresponding cost or number of hops. This method is similar to the Distributed Bellman-Ford algorithm except the DSDV protocol adds other entries into the routing table to help alleviate short and long-lived communication loops. The DSDV protocol includes an attribute in the routing table to distinguish if a certain link in the routing table is a stable link. Also, the protocol periodically makes decisions on stale nodes that have not sent a communication in a set amount of time. Nodes that have not communicated for a set amount of time will be removed from the neighborhood table and replaced by nodes that are not in the table and have communicated more often. If a node detects that a link has been broken the node assigns a infinite weight to the communication edges so that no messages will be routed the node to the lost destination.
The aspect that the author did not cover is the amount of memory that this routing protocol will require. Although the concept of the routing protocol seem fesiable there is no way to judge the usefulness of the protocol without knowing the memory footprint it requires.
Self-configuring localization systems: Design and Experimental Evaluation
Nirupama Bulusu, John Heidemann, Deborah Estrin, and Tommy Tran. Self-configuring localization systems: Design and Experimental Evaluation. ACM Transactions on Embedded Computing Systems, 3 (1 ), pp. 24-60, February, 2004.
This paper presents two algorithms focused on self-configuration for medium and dense localization systems. The algorithms that were used instituted a form of beacon localization.
The first algorithm discussed in the paper was the HEAP algorithm. The main goal of HEAP is to discover places to add a few new beacons to maximize improvements in localization, rather than to completely redeploy the entire beacon network. HEAP uses distributed in-network processing to select placement sites. The second algorithm discussed was STROBE, which stands for Selectively TuRning Off Beacons. The process of the algorithm is just as it sounds. It selectively turns of beacon node in the localization system based on the current number of active beacon nodes. The focus of this algorithm is the reduce the amount of inter-network interference induced by collision from multiple beacons in a small area.
The paper presented simulation results collected from running each of the algorithms. In addition to the simulation results the paper presented real-world evaluation data for the HEAP algorithm using the Berkeley Rene mote. The evaluation results for the HEAP systems confirmed that it can benefit a real deployment localization system. While the paper presented an evaluation for the STROBE system the evaluation wasn’t based on a real deployment of the system. The evaluation was based on an emulation of the system. The author did a good job explaining the considerations needing to be accounted for in developing a self-configuring localization system. The STROBE system was unable to be evaluated in a real world setting due to the fact that it would have been difficult to measure the depletion of energy on beacon nodes at a fix point in time. Also because of the power cycling, the energy depilation would have to be measured of a large period of time. The one change that I would suggest is that the emulation of the STROBE algorithm be reran so that it accounted for amount of energy consumption due to information processing.
February 26, 2007
The nesC Language: A Holistic Approach to Networked Embedded Systems
D. Gay, P. Levis, R. von Behren, M. Welsh, E. Brewer, and D. Culler, The nesC Language: A Holistic Approach to Networked Embedded Systems, ACM Conference on Programming Language Design and Implementation (PLDI), June 2003.
This paper discussed the nesC programming language, a programming language used for networked embedded systems. The focus of this paper was how the nesC language coupled with the nesC complier could help reduce code size and eliminate potential bugs in embedded system software.
nesC is a components based programming language that allows programmers to “wire” to different components of the system using interface declarations. The process of using components allows that nesC compiler to remove unused components at compile time. The removal allows nesC to achieve a much lower code footprint. In addition to removing unused components, the nesC compiler detects potential race conditions. Even though the detection of race conditions is not full proof this feature of the compiler helps point out potential problems before testing begins. The paper states two methods that programmers can use to help alleviate race conditions. The first is to convert all sharing to tasks so the OS scheduler will execute the code at an idle time, or second to use an atomic section to update the shared state. An atomic section is a code sequence in nesC will run automatically.
In the paper, the authors state that nesC prohibits the use of function points and dynamic memory allocation. Later in the paper, the authors state that the use of a function pointer only current results in a compile time warning. If the nesC language prohibits the use of function pointers then why does the use of a function pointer result in a warning and not an error?
EnviroSuite: An Environmentally Immersive Programming Framework for Sensor Networks
L. Luo, T.F. Abdelzaher, T. He, and J.A. Stankovic. EnviroSuite: An Environmentally Immersive Programming Framework for Sensor Networks. ACM Transaction on Embedded Computing System (TECS), 2006.
This paper introduces a software framework for embedded sensor networks called EnviroSuite. EnviroSuite is an object-based programming language that is structured around the elements within a sensing environment rather than the regions, neighborhoods, or sensor groups of a sensor network.
EnviroSuite’s ability to be structure around the environment allow for a smaller number of sensed object to exist within the network. This means that the network is able to maintain a low amount of data replication for systems that are used for tracking moving objects throughout the network. The paper explains the process taken by EnviroSuite as it applies to a tracking system. The core of EnviroSuite is built on object types and object id’s. As an object enters a network, the nodes of the network begin to determine a leader node based on distance from the object and node power level. All nodes in the network that are tracking the object report to the “leader” node. Data is aggregated as it follows through the network based on aggregation parameters specified by the programmer. Once at the leader the data is transferred to a base station. The dynamic functionality is written in a mix of in EnviroSuite primitives and nesC. The EnviroSuite program is then ran through a compiler (EIPLC), which converts the EnviroSuite primitives into nesC language. The compiled nesC is used to run on with TinyOS, and can then be install on supported sensor nodes. The one draw back to the EnviroSuite framework is the increase in code size after compilation. The paper states that a EnviroSuite code size of 5.9KB will compile to a code size of 111.0KB of nesC code.
The one improvement that I would suggest is that the author evaluate the compilation in an attempt to optimize the output from the compiler. All though the high-level interface of the EnviroSuite is attractive, the rate of growth of the compilation code is a major disadvantage.
Mate: a Virtual Machine for Tiny Networked Sensors
P. Levis, N. Patel, S. Shenker, and D. Culler, Mate: a Virtual Machine for Tiny Networked Sensors, International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), Dec 2002.
This paper introduced the Mate virtual machine architecture. Mate is described as a tiny communication-centric virtual machine designed for sensor networks. The basis of Mate is that it allows sensor node programs to be reprogrammed with a smaller overhead compared to conventional binary reprogramming methods.
The key aspect to Mate is that it uses smaller “capsules” to package programs. Since Mate is a bytecode-interoperated architecture, each capsule sent can contain 24 instructions. The virtual machine based programming strategy allows Mate to only reprogram the application layer instead of the entire OS. The paper outlined an evaluation of the Mate VM versus a native architecture running on TinyOS. This evaluation looked at the instruction use rate (instruction per second) and energy required to operate each of these architectures. In the evaluation of IPS, the authors found that the Mate architecture operated at a cost of 33.5:1 for a logical and on two words and 1.03:1 for a packet send. The authors attributed this increase in IPS to the fact that each Mate instruction is executed as a separate TinyOS task, which requires an enqueue and dequeue operation. This accounted for an extra two instruction calls per one Mate instruction. During the energy evaluation, the authors found that a Mate application that runs for less that five days before being reinstalled is more beneficial than that of the native image. After six days of execution the CPU overhead required to operate Mate become greater than the cost of installing a binary version.
While the Mate system is most beneficially for systems that are being reprogrammed every few days, I feel that the Mate architecture would benefit from a mixed installation approach. By a mixed installation, I mean that during initial configuration the Mate system uses the VM based architecture update scheme. After the sensor program has been tuned to the correct parameters the Mate architecture would allow for full binary image to be installed. This would blend the trade offs of rapid reprogramming with prolonged execution.
Active Sensor Networks
P. Levis, D. Gay, and D. Culler, Active Sensor Networks, USENIX/ACM Symposium on Networked Systems Design and Implementation (NSDI), 2005.
In this paper, the author introduced an extension of prior work on the Mate Virtual Machine. The extension, called application specific virtual machine (ASVM), provides three abstractions: handlers, operations, and capsules. Handlers are defined as code routines that run in response to system events, operations are the units of execution functionality, and capsules are the units used for code propagation.
ASVM has a thread execution model and a stack-based architecture. ASVM is separated into two classes: templates and extensions. Templates are included into all ASVMs and extensions are application-specific components that define particular ASVM. The template class included in all ASVMs is broken down into three different parts: the scheduler, concurrency manager, and the capsule store. The scheduler is used to execute threads in a round robin fashion. The concurrency manager determines which threads are runnable, ensuring race conditions and deadlock-free execution. The capsule store is responsible for storage and loading, code propagation, and notification. The authors evaluated each of the three parts of the template VM. Of the parts evaluated the capsule store’s code propagation seemed to achieve to the majority of focus. The evaluation of code propagation was carried out using a comparison between a native regions or tracking application versus the RegionsVM. From the data collected the overall code size of the RegionsVM was more than twice the size of the native application, but RegionsVM only required 71B of data to be retasked while the native program had to transmit 19KB.
I feel that while the authors have improved on some things from the early Mate VM system they still have work to be done to better incorporate the VM code with the underlying OS. One point that was stated but not covered in the paper was the amount of time that a node requires to feel it is up to date.
March 5, 2007
VigilNet: An Integrated Sensor Network System for Energy-Efficient Surveillance
T. He, et al. VigilNet: An Integrated Sensor Network System for Energy-Efficient Surveillance. ACM Transactions on Sensor Networks, 2(1): 1-38, February 2006.
This paper presented the first developed sensor network that we have looked at thus far. The paper addressed issues mentioned in previous works as well as defining problems that were not taken into account in simulations of prior works.
The system evaluated in this paper, called VigilNet, is a surveillance and tracking systems that is intended to be used for military operations where individual human surveillance would be to dangerous. The system was evaluated using 70 MICA2 motes placed in a grassy area along the side of a road. The paper outlined many approaches that the system took to improve the energy efficiency of the motes. One such approach was the method the system used to alternate non-sentry nodes from a sleep to an awake stage. Two methods were evaluated in the paper, a proactive control and a reactive control. In the first approach, the sentry (leader) will send out a sleep beacon periodically and non-sentry nodes will remain awake until they hear this beacon. The beacon contains a duration at which the nodes are to remain asleep. After this time expires the node wake and remain awake until they receive another sleep beacon. In the second approach there is no beacon sent out to place nodes in a sleep state. Nodes sleep and wake based on values specified in the program. Non-sentry nodes break out of the sleep-wake cycle when they receive a wake beacon from a sentry. This beacon is sent out when the sentry detects an activities/object that needs to be tracked. The first approach improves tracking efficiency because all nodes are wake at approximately the same time, but the sleep beacon increased the amount of network congestion. The latter approach does not congest the network and is stealthier, which is need in order not to be detected.
While the paper brought light to inconsistence of sensors and communication of real sensor networks there was not sufficient evidence provided on the constraints of the VigilNet system. During the evaluation section of the paper, the authors mentioned the effect of false positives and false negatives as vehicle speed increases. There is never a mention for the expected optimal sensing speed of the network. Additionally when the authors evaluated the power consumption of the motes they stated that the motes could operate on 2AA batteries with a capacity of 2200mA/h. This is not a correct justification due to the fact that the output voltage of the batteries used is proportional to the capacity of charge remain on the battery. As the motes being to consume power for the batteries the voltage output is going to begin to drop. At some point in time, before the batteries loss complete charge, the output voltage of the batteries is going to be lower than the minimum expected input voltage for operating the motes. So the effective life of the motes is derived from some fraction of the batteries total capacity.
Design and Deployment of Industrial Sensor Networks: Experiences from a Semiconductor Plant and the North Sea
L. Krishnamurthy, R. Adler, P. Buonadonna, J. Chhabra, M. Flanigan, N. Kushalnagar, L. Nachman, and M. Yarvis, Design and Deployment of Industrial Sensor Networks: Experiences from a Semiconductor Plant and the North Sea, ACM Conference on Embedded Networked Sensor Systems (SenSys), November 2005.
This paper introduced a predictive maintenance sensor network system that was deployed in two separate locations. The first location was a semiconductor fabrication plant (FabApp), while the second was onboard an oil tanker stationed in the North Sea. Each of these deployments presented different requirements that had to be meet in order for the sensor network to be affect. For the FabApp deployment, a major concern was the security of the information being propagated through the network. In the tanker deployment, the major concern was the inconsistence of available communication paths. Due to the location of the sensor network communication from gateways may be disrupted due to the closing of watertight doors.
Both deployments sensed vibration levels of selected machine in order to help predict potential machinery failure. The systems consisted of similar components including sensing nodes, cluster heads, gateways, and root node. The FabApp deployment used a bridge node, serially connected to the root node, to exchange information between the sensor network root and the enterprise intranet. The serial connection helped allivate an potential security threat from communicating directly from the sensor root to the intranet. In the tanker deployment the network root was used as the network gateway. This was done because of the isolated area in which the sensor network was deployment did not produce an security concerns.
Through out the paper the authors continually compared two different hardware platforms, the MICA2 motes and the Intel motes. Although they never mentioned the difference in software between the two motes. In particular, the authors never analyzed the difference in the motes’ OS efficiency.
Vineyard Computing: Sensor Networks in Agricultural Production
J. Burrell, T. Brooke, and R. Beckwith, Vineyard Computing: Sensor Networks in Agricultural Production, IEEE Pervasive Computing, pp. 38-45, 2004.
This paper presents an analysis for the use of sensor networks in an agricultural environment. The specific agricultural environment targeted for this analysis was an Oregon vineyard and winery.
The paper started by explaining their method of researching the application. The authors used a method of ethnographic research where they studied the roles that all workers took in the vineyard, as well as looking at the specific information and interaction needs for the different workers. From the research, the authors concluded that some elements of the sensor network could be best achieved by the use of proactive computing. Proactive computing is a method by which data is collected about a certain phenomena, the data is then analyzed and a conclusion to react or not is made with no human interaction. An example that the authors use in the paper is detecting the presence of pests, such as birds, and performing a preventative action to deter them. Proactive computing is best used in situations where a response to some condition must be made in a time frame that is not humanly possible. In addition, the authors proposed the use of data mules to transport data rather than using a structured form of network communications. The reason that the authors choose this form of data transport is because the data collected from sensor network, like temperature, humidity, and moisture content, is not needed immediately. This data has to be processed by the vineyard manager to determine the best time to harvest the grapes. Because to the continued presence of potential data mules the sensor network could be put into place with little to no effect on the day-to-day routine of workers.
While this paper presented a different approach to researching a potential application for a sensor network deployment, there was little mentioned about result from the decisions obtained from the research. The authors presented the justification of the design and had solid evidence to back up their decisions, but there was no mention of the performance of an actually system. I think that the paper would have been more creditable if there was evidence that the actual decisions performed as expected.
Wireless Mobile Ad-hoc Sensor Networks for Very Large Scale Cattle Monitoring
M. Radenkovic and B. Wietrzyk, "Wireless Mobile Ad-hoc Sensor Networks for Very Large Scale Cattle Monitoring", Proc. of Sixth Int'l Workshop on Applications and Services in Wireless Networks (ASWN'06), Berlin, Germany, 2006, pp. 47-58
This paper presents a concept of data storage and routing protocol for a high mobile ad hoc network using Distributed Hash Tables (DHT). The paper optimized the standard DHT approach by incorporating mechanisms to deal with high volumes of disconnection. The focus for application is to place wireless sensor on cattle that will track behaviors and interaction with other cattle. This data is used to determine the proximity of a cow in relation to one that is infected with a disease such as mad cow.
The sensors are mounted on the animal in the form of a collar and collect data based on the animal’s health, location and interaction. The sensors have a unique 128bit id that is used to map to a specific ear tag number. The systems is designed such that information about the about a specific cow is broadcast and store on node that already contain data from the broadcasting node, or have an id similar to the broadcasting node. The system is setup such that when a user issues a query about a specific animal (denoted by the ear tag) the query is disseminated through an ad hoc network via a local gateway. If the query is found in an existing table the information is forwarded to the gateway, otherwise the query is disseminated through the network until it reaches the node under investigation. Once it reaches the specific node that node will broadcast its information back to the gateway. All nodes along the way that have similar id’s or already contain information about the node will update it’s table. Additionally the protocol disseminates information freely if there has been no network traffic for some period of time. This is done in order to reduce the conflict between data dissemination and query gathering.
The authors evaluated the proposed system to determine overhead of using such a system and the success ratio of a give query. They evaluated the network overhead and success ratio of the system while varying the velocity of a node, number of nodes in range, density of the network, and rate of queries. They evaluated the system using no data caching, passive caching, and proactive & passive caching. They found that the proactive caching method had a high success ratio up to about 5m/s of mobility while maintaining a relative low network overhead. During the evaluation of communication density, the authors found that passive and proactive & passive caching perform identical in terms of success ratio, but passive caching resulted in a less network overhead. The latter two evaluations produced similar results as the network density with a closely related success ratio with the difference becoming present in the network overhead.
From the result presented in the paper it seems that a system of this type would lean more towards the use of just a passive caching system. This conclusion is because passive and proactive & passive caching perform very similar in evaluation, but passive caching doesn’t require the computational overhead that proactive caching does. One important factor that seemed to be left out of the evaluation is the effect of power management to support this type of data storage and routing. Because node are placed on animals it would not be beneficial to the end user to have to continuously maintain power supplies for a herd of cattle.
Monitoring Volcanic Eruptions with a Wireless Sensor Network
Geoff Werner-Allen, Jeff Johnson, Mario Ruiz, Jonathan Lees, and Matt Welsh, Monitoring Volcanic Eruptions with a Wireless Sensor Network, EWSN'05.
The paper presented a sample sensor network placed on Volcan Tungurahua in central Ecuador. The purpose of the sensor deployment was to prove the ability of a sensor network to monitor volcanic activity. The sample system was constructed using Mica2 motes equipped with infrasonic sensors to measure the low-frequency sound pressure waves generated from volcanic activity. Traditional monitoring systems used consist of wired sensor stations that consume high amounts of energy. Additionally data collected from the sensor is stored locally on the stations and must be retrieved manually.
The sensor network proposed will collect data from infrasonic sensor attached to wireless linked motes. The motes are placed such that they are only one hop away from the base station. This method is used in order to reduce the bandwidth of the network by having to multi-hop information to the base station. Once at the base station the data is aggregated and sent via a long-range wireless link to an observatory 9km away. The system was deployed and was able to collect 54 hours worth of data. The data collected for the deployed sensor network compared closely with data collected from the wired sensor stations.
The paper presented an excellent example for the application of a sensor network, but the deployment used didn’t correctly exemplify a sensor network. Sensor nets should be designed such that information is passed through the network efficiently in order to reduce network bandwidth and extend network life. This application did neither! The nodes were designed to simply take a sensor reading and forward the information to the aggregation node. This repetitive sensing and sending causes nodes in the network to die off faster. If more care were taken to reduce the power usage of the network by alternating sampling throughout nodes in the network, this example system would have been a better judge of the use for sensor networks.
April 2, 2007
The Semantic Web
The paper introduces an extension of the World Wide Web called the Semantic Web. The paper introduces the semantic web as a form of data structure that is added on to a website so computer are able to autonomously interpret information. The proposed structured data comes in the form of XML and RDF. XML allows for tags or identification to be mapped to a description of data. The problem with XML markup is that it does not identify how one set of data corresponds to another set. Because of the issue the author introduced RDF. RDF is a form of relationship identification that is similar in form to an English language sentence, subject-verb-object. An example used in the paper was “field 5 in database A – is a field of type – zip code”. In the event that data is requested from separate databases and these database have different identifiers for the same data the program must have a means identifying common meaning between the two databases. Ontologies, the third concept of Semantic Web, is a set of rules that define the relationship of different sets of data to each other, and inferences are made based on these rules. The overall idea for this paper is to present a framework that with help to relate everything to everything else!!
The Semantic Web Revisited
This paper takes a second look at Semantic Web presented in the above paper. It identifies where some of the concepts are being used, and gives a few reasons where the adoption of the concept would be applied. Additionally the paper spent a get deal of time explaining the formation of ontologies. Additionally, the paper mentioned the need for the formulation of a standard for development, and attributed this to a reason of way Semantic web are still not in predominate existence. This paper provides a higher level look at Semantic Webs than the above paper, while introducing advances made in Semantic Web approaches since the publication of Tim Berners-Lee’s introduction to the concept.
OpenGIS Sensor Model Language (SensorML) Implementation Specification
This paper introduces a specification for the sensor modeling language (SensorML). In short, SensorML is a framework that models sensors and transducers as process. These processes can be connected and participate equally within a process chain, and can utlize the same process model chain as other processes. SensorML is a way to describe sensors in a general way based on the phenomena being sensed, the sensors provided performance characteristics, and the process by which the observation was obtained. This form of generalization is carried out using XML. The paper gives the basic definition of a sensor, application where SensorML could be used, and some processes that sensors are involved in. The XML decription of a sensor will contain information such as quantity, count, and type. The use of XML to provided a model for which sensor are defined seems to have a great deal of promise. The applied to integrate more information about a sensor will allow for a more descriptive result.
Plug-and-Play Sensors in Wireless Networks
This paper introduces a solution to WSN using IEEE 1451. CrossNet, presented in this paper, is a WSN that comprises a smart sensor, node, and a data sink. CrossNet uses the Bluetooth wireless communication between node and data sink. Data sink can be or any form from a handheld to a PC equipped with a Bluetooth radio. The CrossNet system captures data from a connected sensor digitizing it in preparation for transmission. The digitized data is transmitted to a data sink along with the TEDS for the particular sensor. Once at the sink, the information can be identified using the TEDS information. Since its publication in 2001, there have been many advances in the form of wireless communication options. While the concepts allied by the CrossNet system are very beneficial when it comes to power management the choice of wireless communication is a major draw back to the design.
IEEE 1451: A Standard in Support of Smart Transducer Networking
This paper gives an outline of the IEEE 1451 standard for smart sensors. The outline consists of a general overview for the four models encompassed in IEEE 1451. IEEE 1451.1 provides the common object model of the transducer and the software interface specification. The object encapsulates the details of the hardware implementation. IEEE 1451.2 defines the Transducer Electronic Data Sheet (TEDS). The TEDS are stored in NVM and contain information about the type, operation, attributes, and calibration of the transducer. IEEE 1451.3 provides a specification for a multidrop distributed system where multi sensors need to communication and collect data in a synchronized manner. IEEE 1451.4 defines the interface of a mixed mode devices were the device will monitor an analog signal and transmit a digital signal. The concept introducing a standardized interface for transducer will allow for a shorten time to market as well as a reduction in complexity to the standardized format of the sensor.
April 9, 2007
Minimum Power Configuration in Wireless Sensor Networks
G. Xing, C. Lu, Y. Zhang, Q. Huang, and R. Pless, Minimum Power Configuration in Wireless Sensor Networks, ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc), May 2005.
This paper presents a merger of three different forms of minimum power configurations for sensor networks. The first is topology control, where maintains the lifetime of the network through reduction in transmission power. The second is power aware routing. This is a process of routing information through nodes in the network that can afford the power demand of transmitting a message. The final approach is sleep management. Depending on the workload of the sensor network, the majority of nodes contained in the network will be in an idle state. Sleep management places nodes that are in an idle state and are not being used for message routing to be placed in a sleeping state. Sleeping node transition from the sleeping state and take over network routing in order to increase the overall network lifetime. While there has been much research into each individual approach this is the first to combine all three approach. The paper states that the proposed problem in NP-complete to obtain the optimal solution. Because of this the authors choice to use heuristic approach. The paper investigated four approaches for determining the appropriate path to take through the network, matching based algorithm, shortest path heuristic, incremental shortest path heuristic, and constant-ratio approximation algorithm. The paper discusses the development of a distributed protocol design (MPCP) that uses the incremental shortest path heuristic. The protocol requires that each node in the network maintain a routing table that contains associated routing costs to the data sink. The paper presented a very important concept, but the simulation conducted didn’t seem sufficient. The simulation was based on a ‘many-to-one’ communication model where the network only had one data sink. Before the protocol can be considered an asset to sensor network I feel the simulation should be ran to evaluate the protocol operation in a ‘many-to-many’ relationship.
Flexible power scheduling for sensor networks
Hohlt, B. Doherty, L. Brewer, E., Flexible power scheduling for sensor networks. Information Processing in Sensor Networks, 2004. IPSN 2004. Third International Symposium on, 26-27 April 2004, pp. 205-214
This paper introduces flexible power scheduling (FPS). FPS is a scheduling protocol that’s main goal is, “to reduce power consumption while supporting fluctuating demand in the network for data collection.” The protocol schedules transmission and receive slots for each node and additionally places the node in a sleep state when idle. The process of FPS is as follows. A node can be in any one of six different states, Transmit, Receive, Advertisement, Transmit Pending, Receive Pending, and Idle. To begin communication a parent node sends a child an advertisement for data at which time the parent is placed in a receive pending state. When the node receives this advertisement it immediately response with a confirmation and is placed in a transmit pending. At a specific time slot (specified by the advertisement message) the node begins transmission to the parent. At this time the node and parent are placed in the transmit and receive states respectively. Because of the use of time slots the node and parent need to be time synchronized. Because of the coarse grain used for time synchronization, there is some latency that is introduced. An analysis is presented that shows the reduction in power consumption, but it also shows that as the duty cycle of the system is reduced (more time asleep) the latency increases.
Low-energy sensor network time synchronization as an emergent property
Bush, S.F., Low-energy sensor network time synchronization as an emergent property. Computer Communications and Networks, 2005. ICCCN 2005. Proceedings. 14th International Conference on. 17-19 Oct. 2005, pp. 93-98
From the above paper we can see how synchronization can play a huge roll in the power consumption of a sensor network. This paper introduces a low power time synchronization approach that is a variant of pulse coupled oscillation (PCO). The approach taken in this paper uses a 5ms window for synchronization. Within the 5ms window there are 0.2ms pulses. The 5ms was chosen to accommodate for the start-up of the node. The paper states that the oscillator of the node requires 450 micro seconds to stabilize. This settling time causes the node to not be able to turn on/off rapidly. The 0.2ms pulses are used to compute synchronization. The difference between the first pulse and the current pulse is computed and normalized by some interval. This normalized valued converges on a number between 0 and 1, 1 being of desired synchronization. The paper evaluated the proposed method for time synchronization. The analysis shows promise for the mentioned method. In addition the evaluation shows that the synchronization method works best for mobile node networks.
TSync : A Lightweight Bidirectional Time Synchronization Service for Wireless Sensor Networks
H. Dai, R. Han, "“TSync : A Lightweight Bidirectional Time Synchronization Service for Wireless Sensor Networks”, ACM SIGMOBILE Mobile Computing and Communications Review, Special Issue on Wireless PAN & Sensor Networks, vol. 8, no. 1, January 2004, pp. 125-139.
This paper presents a TSync a bidirection time synchronization protocol. TSync achieves bidirectional time synchronization through uses to methods for time synchronization, a pull method and a push method. In the push method a host controller, one that contains the clocked to be synced to, initiates synchronization by broadcasting a message to its neighbors specifying a particular neighbor to participate in the synchronization. All neighbors, at the time of receiving the broadcast, set a time value t2 that corresponds to the time at which the broadcast was received at the node, and sets a second time t1 that corresponds to the time that the host node set the message. The neighbor node that was specified by the host stores the transmission time from host (t1), the time the message was received (t2), and sends the host a response containing the t2 and the time the message was sent (t3). The host node gathers the information sent from the requested neighbor and computes the difference. This difference is sent to all neighbors and a synchronization is done based on time t2 the difference computed from the selected node, and the reception time of the sync packet. The pull method is similar to the push method except a node wishing to sync its clock will send a query weighted with a depth to its parents. The parent will in turn send the query up the tree until the requested weight is expended. When the query reaches a reference node the synchronization information is forwarded to the requesting node. Analysis of the protocols was conducted using each of the proposed protocols separately. The analysis shows that the push protocol perform with a mean error of 21.23 for a single hop and 29.47 for a three hop clock estimation. While this protocol seems to perform well with little overhead, it seems that the amount of messages required to sync and the amount of time that the nodes will need to be active would greatly affect node lifetime.
Calibration of Sensor/Actuator Networks in Unknown Environments
Kamin Whitehouse and David Culler, “Calibration of Sensor/Actuator Networks in Unknown Environments”, 2002 ACM International Workshop on Wireless Sensor Networks and Applications, Sept.2002.
This paper presents a calibration approach to sensor/actuator networks that is based on parameter estimation. The paper discusses three methods for calibration, iterative calibration, mean calibration, and joint calibration. In the paper the focus of calibration was based on methods to determine locations, or distances between communication nodes. In the iterative calibration approach a single transmitter and receiver is chosen as reference a point. These reference points are used to calibrate receivers and transmitters respectively. This approach leaves only one uncalibrated device in the network. If a measurement from the pair is in error the error is attributed to the uncalibrated device. In the second calibration approach, mean calibration, each receiver’s parameters are collected from the mean of data gathered from all transmitters. This approach reduces the error for the receivers, but falls to reduce the error of the system as a whole. The final approach, joint calibration, is a calibration method proposed by the author that is used in an ad-hoc localization system, called Calamari. This method performs calibration on the system as a whole. This macro-calibration method computes calibration parameters in three steps: (1) parameterize each individual device and model system response using these parameters, (2) collect data from the system as a whole, and (3) choose parameters for individual devices such that behavior of the entire system is optimized.
The paper conducted an experiment to evaluate the performance of the mentioned calibration systems using MICA motes. The experiment revealed that an uncalibrated system presented an average error of 74.6%. The iterative calibration produced an average error of 19.7%, mean calibration produced a 16.0% average error, while joint calibration produced an average error of 10.1%.
This paper presents a solution to a problem that is existent in all sensor networks. The one portion of sensor networks that was not addressed to the proposed solution was the effect on network lifetime. I feel that this would be a great place to start for reevaluation of the approach.
April 16, 2007
Wikipedia: Ultra-wideband
Ultra-wideband (UWB) is a technology used for transmitting information spread over a large bandwidth. A large bandwidth in this article is considered as a bandwidth greater than 500 MHz or 25% of the center frequency. UWB transmitters operate in the unlicensed band from 3.1-10.6 GHz as ordered by the FCC. UWB transmitters can be found in use in the IEEE 802.15.4 wireless communication technology. The article states that the use of UWB technology has relative immunity to multipath fading, but not to intersymbol interference. Additionally UWB provides a high resolution for “time of flight”. Applications for UWB vary from PAN wireless connectivity and long range, low data rate applications. Currently there are more than four dozen UWB devised certified to date, the vast majority of which are radar and imaging systems.
Hierarchical Temporal Memory: Concepts, Theory, and Terminology
Hierarchical Temporal Memory (HTM) is a method of adapting learning and cognitive behavioral functions into computers. Currently computer have a difficult time performing task that human find to be simple. Tasks such as pattern recognition, speech recognition, and navigation, which are easy for humans, yet despite years of research there has been no algorithm or adaptation of technology, developed to perform these tasks efficiently. The HTM concept is derived from biology. HTM’s are modeled as is the cortex of the brain performing four primary tasks: discover causes, infer causes, make prediction, and direct behavior. As the HTM performs the ‘causes’ it builds a hierarchical structure that is based on the decisions and conclusions derived from various sensory inputs. As do humans HTM learns specific details about sensory input by observing changes based on time. The adaptation of HTM to WSN with open many doors for applications, but due to the withholding both wireless communication reliability and adaptive learning techniques it may be some time before we see the integration of HTM in WSN.
Presentation
February 5, 2007
Understanding Packet Delivery Performance In Dense Wireless Sensor Networks
J. Zhao and R. Govindan, Understanding Packet Delivery Performance In Dense Wireless Sensor Networks, ACM Conference on Embedded Networked Sensor Systems (SenSys'03), November 2003.
Powerpoint presentation : Slides
February 19, 2007
Adaptive Data Placement for Wide-Area Sensing Services
S. Nath, P.B. Gibbons, and S. Seshan, Adaptive Data Placement for Wide-area Sensing Services, USENIX Conference on File and Storage Technologies (FAST), 2005.
Powerpoint presentation : Slides
March 19, 2007
TinyDB: An Acquisitional Query Processing System for Sensor Networks
S. Madden, M. Franklin, J. Hellerstein, and W. Hong, TinyDB: An Acquisitional Query Processing System for Sensor Networks, ACM Transactions on Database Systems, 2005.
Powerpoint presentation : Slides
Sensor Net Simulations
April 23, 2007
Adaptive Traffic Flow Management Using Sensor Networks
Power Point Presentation: Project.ppt
Simulation Files: TFMS.zip
