Using Trajectory Sensor Data Stream Cleaning to …...Using Trajectory Sensor Data Stream Cleaning to Ensure the Survivability of Mobile Wireless Sensor Networks in Cyberspace (new

Using Trajectory Sensor Data Stream Cleaning to Ensure the Survivability of Mobile Wireless

Sensor Networks in Cyberspace (new project)

Dr. Niki Pissinou, PI

Dr. S.S. Iyengar, Collaborator, FIU Dr. Charles Kamhoua , Collaborator, AFOSR

Dr. Sitthapon Pumpichet, Dr. Ali Bakthiar ,Dr. Xinyu Jin

Ph.D. Students: Concepcion Zulema, Sanchez de Rodriguez, Samia Tasnim, MingMing, Laurent Lavoisier Yamen

NSF-DOD REU SITE, NSF RET Site participants

Dr. Frederica Darema, AFOSR, DDDAS Program Review, December 1, 2014

Mr. E. Lee, AFOSR 12/1/14

AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting 1

Outline

• Research Objectives

• Research Context

– Ad-hoc Networks and Wireless Sensors

– Design Issues

• Related Work

• Preliminary Research – Virtual sensor based method

– Belief based method

– Sketch based method

• Future Directions

2 12/1/14 AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting

Research Objectives (10/2014-10/2017)

1. Design and develop a context-based trajectory data model that illustrates the physical, spatial-temporal, symbolic, absolute, relative and uncertain context of trajectories in mWSNs.

2. Design and develop a game-theoretic trajectory secrecy safeguarding mechanism in mWSNs.

3. Design and develop a lightweight energy-efficient framework that integrates trajectory secrecy with security protection techniques.


Research Context

12/1/14 AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting 4

• Mobile, Trajectory, Wireless Sensor Network

– Sensor Nodes talking to each other

– Nodes talking to “some” node in another network

– Dynamic and Data Driven Application Systems (DDDAS)

– Limited resources

• E.g., limited in power, computation, memory prone to failures, sensor nodes may not have global id

5

Access PointAccess Point

Basic Scenario

12/1/14 AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting

• Node mobility

– A node participating as source/sink (or destination) or a relay node might move around

– Deliberately, self-propelled or by external force; targeted or at random

– Happens in both WSN and MANET

• Sink mobility

– In WSN, a sink that is not part of the WSN might move

– Mobile requester

• Event mobility

– In WSN, event that is to be observed moves around (or extends, shrinks)

– Different WSN nodes become “responsible” for surveillance of such an event 6

Different Sources of Mobility


• Do not only deliver bits from one end to the other

• Provide information, not necessarily original bits

– E.g., manipulate or process the data in the network

– E.g., aggregation

• Apply composable aggregation functions to a converge cast tree in a network

– Typical functions: minimum, maximum, average, sum,

7

In-network Processing

12/1/14 AFOSR DDDAS Program Review Dec 1-3, 2014 PI

Meeting

• Exploit location information

– Required anyways for many applications; can considerably increase performance

• Exploit activity patterns and context

– E.g., stop and go

• Exploit heterogeneity

– By construction: nodes of different types in the network

– By evolution: some nodes had to perform more tasks and have less energy left; some nodes received more solar energy than others

• Exploit DDDA paradigm

8

Project Design Principles


• How to select a proper spatio-temporal granularity to annotate meaningful semantics to each trajectory type in a heterogeneous trajectory?

• How to classify the trajectory relations and design the annotation algorithms to be computationally efficient ?

• With the uncertainty and reduction of location reports, how can we characterize the relationship between two trajectories in online fashion?

• If there is a sensor that its sensor readings are highly correlated to those of a given sensor, what is the optimum approach to select that sensor instead of others?

• How can we balance the rate of data reduction with the data cleaning performance?

• How is the most relevant and helpful annotation data defined and chosen for cleaning sensor data?

• …..

Characterizing Sensor Data


Meeting 9

Problem

• Data received at the base station is dirty --

o missing, incomplete, noisy, unordered, duplicated, etc.

• Poor data integrity: sensor nodes base station

• 40% of sensor data collected in Intel Lab is missing[Int04]

• Node mobility in MSN makes data dirtier o Transient connectivity, Node isolation, Data Collision, etc.

• Most of current data cleaning methods are either for cleaning data streams in stationary WSNs or for offline static databases.

• Objective: To clean sensor data in mWSNs in an online fashion.


Related Work


Related Work

• Data cleaning in static databases

• Cleaning data streams in static WSNs

o Sequential based methods

o Non-sequential based methods

• Cleaning data streams in mobile WSNs


Data Cleaning in Static Databases

Key works in data cleaning in static databases

Outlier detection [SPP+06]

Data consistency maintenance[CFG+07]

Uncertainty reduction[CCX08]

These methods need a complete set of static data.

Not practical in sensor applications – too high volume of data running into the base station.


Meeting 13

Cleaning Data streams in Static WSNs

• Sequential based methods

o Extensible Sensor stream Processing (ESP) framework [JAF+06]

o Quality estimation of the cleaned data streams [SJFW06]

o Smoothing for Unreliable RFID data (SMURF) [JGF06]

• Non-sequential based methods

o Bayesian based cleaning for noisy sensor data [EN03]

o Neuro-fuzzy regression approach [PS07]

o Belief-based data cleaning framework [BPM09]


Cleaning Data Dtreams in Mobile WSNs

• Few works dedicated research work attempting to clean sensor data in mobile environments

• Trajectory and mobile object related literatures have been reviewed. o Location prediction [MPTG09]

o Trajectory classification [LHLG08]

o Nearest neighbor query in uncertain trajectory [TTD+09]

o Outlier detection in trajectory data [BCFL09]


Preliminary Research


Virtual sensor for mobile sensor data cleaning

S. Pumpichet and N. Pissinou, “Virtual Sensor for Mobile Sensor Data Cleaning,” in Proceedings of IEEE International Conference on Global

Communications (GLOBECOM 2010), Miami, USA, December 6-10, 2010, pp. 1-5.


Contributions

One of the first data cleaning methods designed for mWSNs.

Developed a use of Virtual Sensor (VS) concept and combined with adaptive filter in data prediction.

AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting 18 12/1/14

Methodology and Assumptions

• Idea analogy: A host guides visitors about local info

Host Virtual sensor

Visitor Target sensor

• A Virtual Sensor is actually a memory space at the base station

• This memory space is linked to 2 data: location and area

Host location VS location

Host’s home size VS area

• System administrator will assign the VS location and VS area

19 12/1/14 AFOSR DDDAS Program Review Dec

1-3, 2014 PI Meeting

VS Data Update

• The memory space will store measurement samples that represent measurements of VS

• When missing/dirty data occurs from a (target) sensor, the missing data will be replaced/cleaned by samples from the nearby VS.

• How VS get data update?


VS Data Prediction

• If no sensor passes by, VS data cannot be fetched.

• VS data is predicted by using Adaptive filter-based prediction model.

o No prediction longer than N consecutive missing data.

o Not start predicting unless N consecutive data updates from passing-by sensors


Evaluation



Results: Varied Node Density


Results: Varied Speed

24

Avg. speed = 9m/s


Results: varied speed


Meeting 25

Avg. speed = 18 m/s Avg. speed = 9 m/s

Results: Varied Speed

26

Avg. speed = 18 m/s Avg. speed = 27 m/s


Meeting

Results: varied missing data rate


Meeting 27

missing data = 20% missing data = 40%

Results: Varied Missing Data Rate

28

missing data = 60% missing data = 40%

12/1/14 AFOSR DDDAS Program Review Dec


Summary

We developed a centralized data cleaning method for mWSN using the virtual sensor concept.

With classic adaptive filter, the method can clean data and outperform the self-cleaning temporal method in a dense network.

The method does not need to implement any additional in-field hardware.


Belief-based cleaning in trajectory sensor streams

Sitthapon Pumpichet, Niki Pissinou, Xinyu Jin, Deng Pan: Belief-based cleaning in trajectory sensor streams. ICC 2012: 208-

212 12/1/14

AFOSR DDDAS Program Review Dec

1-3, 2014 PI Meeting 30

Contributions

Developed a belief-based sensor selection method to

o identify the group of sensors that is helpful in cleaning data based

• on their current trajectories

• the quality of their data streams

Evaluated the cleaning performance with various

o mobility models,

o speeds

o node densities.


Methodology and Assumptions

• Assumptions

o We assume that there are multiple sub-areas forming up the area of interest.

o The boundaries among sub-areas are also assumed to be known.

• The cleaning process computes the replacement of dirty data by utilizing the readings from a group of sensors that are believed that they offers enough reliable readings from a specific sub-area.

• Two-step process:

1. sensor selection

2. cleaning process



Belief-based Sensor Selection

• Area-based calculation of belief degree

• We use a belief table that contains belief degree of each sensor in a sub-area. (1 belief table/ 1 sub-area)

• The belief degree of each sensor represents how trustworthy a sensor could help cleaning the dirty readings measured within the sub-area at a specific time.

• It is based on two parameters: 1. alibi degree

2. detection rate of dirty data


Alibi Degree

• The alibi degree is computed from residence vector and the frequency of existence in the sub-area.



Detection Rate of Dirty Data

• the detection rate of dirty data shows the quality of streaming data received from a sensor residing in the corresponding sub-area.

• As an online method, we simply calculate the detection rate of dirty data as a ratio of the cumulative number of detected dirty samples to the number of all samples that the sensor measured within a sub-area.


Belief Degree and Sensor Selection

• The belief degree would be proportional to the alibi degree but inversely proportional to the detection rate of dirty data.

where

: belief degree

α : belief coefficient

AN : normalized alibi degree

D : detection rate of dirty data

• The sensors with the β value higher than a belief threshold (βth) would then be selected to collaborate in the cleaning process.



Cleaning Process

• The calculation of cleansed data considers

1. the time difference between the time that each available data of the selected sensors are sampled and the time when the target sensor senses the dirty sample.

2. the distance between the selected sensors and the target sensor when the target sensor senses that dirty sample.


Cleaning Process

38

𝑑𝑐 =

𝑑𝑖 ∙1

∆𝑡 𝑑𝑑 , 𝑑𝑖 ∙

1∆𝐿 𝑑𝑑 , 𝑑𝑖

𝑘𝑖=1

1

∆𝑡 𝑑𝑑 , 𝑑𝑖 ∙

1∆𝐿 𝑑𝑑 ,𝑑𝑖

𝑘𝑖=1

k : The number of data samples of selected sensors residing in the sub-area

dc : Cleansed data of the target sensor

di : The eligible data from selected sensors

∆t(dd ,di) : Difference in sampling time of the dirty sample dd and the eligible data di

∆L(dd,di) : Location distance of target sensor and selected sensors when the target sensor senses the

dirty sample dd 12/1/14 AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting

Evaluation

Compared with VS method, and Moving average based method [ZCWL07]

BonnMotion mobility generator-

Mobility models: Random waypoint,

Nomadic, and Random street.

Metric: Cleaning rate

Cleaning rate= Successfully cleaned

All detected dirty data




Random waypoint


41

Nomadic


Results: varied node density


Random street

Results: Varied Missing Data

12/1/14 AFOSR DDDAS Program Review

Dec 1-3, 2014 PI Meeting 43

Random waypoint



Nomadic



Random street

Results: Varied Mode Speeds


Random waypoint

Results: Varied Node Speeds


Nomadic

Results: Varied Node Speeds



Random street

Summary

• We introduced a belief-based sensor selection to help clean sensor data streams in sparse mWSNs.

• Cleaning process consider asynchronized sampling time of mobile sensors.

• Area boundaries need to be known; however, in many scenarios they might be unknown and dynamic.


Sketch based cleaning in sensor data streams

S. Pumpichet, N. Pissinou, X. Jin and D. Pan, “Belief-based Cleaning in Trajectory Sensor Streams,” in Proceedings of IEEE International

Conference on Communications (ICC 2012), Ottawa, Canada, June 10-15, 2012, pp. 208-212.


Contributions

• We developed a sketch-based data cleaning method for data streams in mWSN environments, where moving sensors do not measure a shared phenomenon and/or are deployed in a sparse network. – internally sensing

• We modified the use of a sketch technique [CM05] that attempts to summarize the counting of item frequency of the events of interest.

• We applied the unique features of a super-increasing set to help formulate the sketch and cleaning process.



Methodology

• The hybrid cleaning scheme: the base station needs information only from a sensor to clean data for the corresponding sensor.

• There are 2 main processes: Sketch and Cleaning process. o Sketching at each sensor node

o Cleaning at the base station

• Sketch and Clean process are periodically perform every N sensor samples.


Before the Sketch Process: Initialization

• Output: A sketch packet containing a. a sketch array

b. counter array

c. One sketch packet is for cleaning N sensor samples

• Before sketch process, a sensor initializes a dxw sketch array and a dxw counter array. All array members are set to zero.

• We assume that the upper and lower bound of valid values of a measurement are pre-defined to each sensor. o The measurement with values out of this valid range (R) will be simply

judged as an outlier. Sketch resolution (r), r = R/wd.


Before Sketch process: Super-increasing Set

We will use a super-increasing set (S) to update sensor readings into the sketch packet.

Definition 1. The super-increasing set (S) is the set that the value of members is a positive integer and value of the ith member is greater than the sum of all the 1st to (i-1)th members.

S = {𝑠𝑖 | [𝑠𝑖> 𝑠𝑘]} 𝑖−1𝑘=1

• A unique characteristic of the super-increasing set is that if there is a number, which is equal to a sum of members in a super-increasing set when no members are added more than one time, the set of members which produce the summation is unique [MH78].


Before Sketch process: Super-increasing set

Lemma 1. The minimum value of the ith member in the super-increasing set (S) is equal to 2i-1.

Smin = {𝑠𝑖𝜖 S | si = 2i-1}

• Proof: The lemma 1 can be proven by using the induction proof. (1) The minimum value of the first member of set S is the minimum positive integer, s1,min =1 = 21-1. (2) The minimum value of the 2nd member of set S is s2,min =1+1 = 22-1. (3) For i ≥ 3, si > si-1+…+s2+s1. Because 2𝑚 = 2𝑀 − 1,𝑀−1

𝑚=0 the minimum value of the ith member of set S equal to si,,min =2i-1.



Sketch process (cont’d)

For i = 1 to N,

1) For a sensor reading (ni), find a suitable sub-range value(rsub) rsub = ni/r

2) Find the value (vi) to add into the sketch array from sequence i,

i 𝜖 [1,..,N] by vi = s(i mod m)

3) Find the set of array members that need to be altered (U),

U = {a(p,q) | q = 𝑟𝑠𝑢𝑏 / 𝑤𝑝−1 𝑚𝑜𝑑 𝑤}

4) Update the sketch array by a(p,q) a(p,q) + vi

5) Update the counter array by c(p,q) c(p,q) + 1


Cleaning Process

• For available sensor samples,

Update the sketch array by a(p,q) a(p,q) - vi

Update the counter array by c(p,q) c(p,q) – 1

• The remainder values in sketch array and counter array are from dirty/missing samples.

• With the unique feature of the super-increasing set, we can extract the value of missing samples. (Given that the sequence numbers of missing samples are known.)

• The recovered values could be incorrect if there are any two missing samples share the same update value (vi)



Evaluation and Analysis

1) Analysis of Cleaning Performance o Prob. of the event that the missing samples do not have same vi

o Cleaning performance without missing sketch packet

o Cleaning performance with missing sketch packet

2) Analysis of the Communication Cost

3) Cleaning performance on Synthetic Data



Analysis of Cleaning Performance


The cleaning performance of the

sketch-based method and the simple

retransmission.

The cleaning performance in the

function of the missing rate with

varying N/m ratios.

Analysis of Communication Cost

We analyzed the energy spent on a link between tx-rx sensors

We adopted the energy model [SAM03] as follows.

𝐸𝑛 = 𝑃𝑡𝑒 + 𝑃𝑜𝛼 + 𝑙 + 𝜏

𝑅+ 𝑃𝑡𝑠𝑡𝑇𝑡𝑠𝑡 + 𝑃𝑟𝑒

𝛼 + 𝑙 + 𝜏

𝑅+ 𝑃𝑟𝑠𝑡𝑇𝑟𝑠𝑡 + 𝐸𝑑𝑒𝑐

We used the values of each parameters according to the experiment

results in [SAM03].



Analysis of Communication Cost



Comparison of energy consumption

in cleaning N sensor samples

among the sketch-based method

with R=102, R=103 and the

retransmission

Cleaning performance of the sketch-

based method compared with that of

the retransmission method with

adjusted energy consumption

Cleaning Performance on Synthetic Data: Varied Speeds


10 meters/min

Cleaning Performance on Synthetic Data



20 meters/min

Cleaning performance on Synthetic Data


30 meters/min

Summary

• We have developed a novel sketch-based data cleaning method to recover the values of missing samples in sensor data streams.

• The sketch-based cleaning method relies only on a sketch packet, which plays a role in a summary of N sensor samples.

• It does not rely on data from other nearby sensors with any types of contextual relationships so that it is can clean sensor data streams when sensors operate in sparse networks and do not measure a shared environment.

• It requires a small portion of additional power consumption to transmit a sketch packet.


Current Work


1-3, 2014 PI Meeting 66

Design and development of a semantic trajectory relationship model

• Build a trajectory-based relationship model that illustrates how the physical, spatial-temporal, symbolic, absolute, relative trajectory contexts affect correlations among sensor data in mWSNs.

•

• A semantic trajectory relationship model could be designed to support the development of a sensor selection process by providing the definition and format of dependencies among trajectory relationships.

12/1/14

AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting 67

Development of a context-aware annotation method for trajectory sensor streams.

In scenarios, context of sensor operations can be used to find out correlation of sensor streams better than sensors’ trajectory relationships.

– E.g., the readings of two soldiers’ stress level are not necessarily reflected by the proximity of the soldiers. Instead, the readings might be reflected by other “context semantics,” such as the number of hours a solider sleeps per night and number of hours on duty.


Goal (3): Design and development of a comprehensive data cleaning method that tolerates the uncertainty of trajectory readings

The main approach of this direction is to develop a comprehensive data cleaning method for mWSNs that integrates context awareness and semantic trajectory relationships.

In fact, sensors’ trajectory data are uncertain: localization uncertainty, indoor data loss, etc.

One should focus on developing an efficient online trajectory knowledge extraction technique for estimating imprecise sensor locations that tolerates trajectory uncertainty.


Conclusion

• Described our current research to clean the real-time sensor data streams in different system restrictions in mWSNs.

o Dense, Sparse and Internally DDDA sensing environment

o Concept of virtual sensor to clean sensor data streams in mWSNs

o a belief-based sensor selection to improve sensor data cleaning in sparse networks.

o a protocol using the sketch technique to clean data streams for sparse and internally sensing device.

• Preliminary results on

o context aware trajectory data cleaning that tolerates uncertainty of trajectories

o Design and develop a game-theoretic trajectory secrecy safeguarding mechanism in mWSNs.


References

[BCFL09] Bu, Y., Chen, L., Fu, W., & Lui, D. (2009). Efficient anomaly monitoring over moving object trajectory streams. Proceeding of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.159-168.

[BPM09] Bakhtiar, Q. A., Pissinou, N., & Makki, K. (2009). Belief based data cleaning for wireless sensor network. Wireless Communications and Mobile Computing. doi: 10.1002/wcm.970

[CCX08] Cheng, R., Chen, J., & Xie, X. (2008).Cleaning uncertain data with quality guarantees. Proceedings of ACM Very Large Data Bases Endowment. 1(1). 722-735.

[CFG+07] Cong, G., Fan, W., Geerts, F., Jia, X., & Ma, S. (2007). Improving data quality: consistency and accuracy. Proceedings of the 33rd International Conference on Very Large Data Bases Endowment. 315-326.

[CM05] Cormode, G. and Muthukrishnan, S. (2005). An improved data stream summary: The count-min sketch and its applications. Journal of Algorithms. 58-75.

12/1/14

AFOSR DDDAS Program Review Dec 1-3, 2014 PI Meeting

71

References [EN03] Elnahrawy, E., & Nath, B. (2003). Cleaning and querying noisy

sensors. Proceedings of the 2nd ACM International Workshop in Wireless Sensor Network. 78-87.

[Int04] Intel Lab Data. (2004). http://db.csail.mit.edu/labdata/labdata.html

[JAF+06] Jeffery, S., Alonso, G., Franklin, M., Hong, W., & Widom, J. (2006). A pipelined framework for online cleaning of sensor data streams. Proceedings of the 22nd International Conference on Data Engineering.

[JGF06] Jeffery, S. R., Garofalakis, M., and Franklin, M. J. (2006). Adaptive Cleaning for RFID Data Streams. Proceedings of the 32nd International Conference on Very Large Data Bases. 163-174.

[LHLG08] Lee, J., Han, J. Li, X., & Gonzalez, H. (2008). TraClass: trajectory classification using hierarchical region-based and trajectory-based clustering. Proceedings of the 34th International Conference on Very Large Data Bases Endowment, 1(1). 1081-1094.



References [MH78] Merkle, R. C, and Hellman, M. E. (1978). Hiding information and

signatures in trapdoor knapsacks. In IEEE Transactions on Information Theory. Vol. IT-24, No. 5.

[MPTG09] Monreale, A., Pinelli, F., Trasarti, R., & Giannotti, F. (2009). WhereNext: a location predictor on trajectory pattern mining. Proceeding of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 637-646.

[PS07] Petrosino, A., & Staiano, A. (2007). A neuro-fuzzy approach for sensor network data cleaning. Proceedings of the 11th International Conference on Knowledge-based and Intelligent Information and Engineering Systems. 4697. 140-147.

[SAM03] Sankarasubramaniam, Y., Akyildiz, I. F., and McLaughlin S. W. (2003). Energy efficiency based packet size optimization in wireless sensor networks. In Proceeding of the 1st IEEE International Workshop on Sensor Network Protocols and Applications. 1-8.

[SJFW06] Sarma, A. D., Jeffery, S. R., Franklin, M. J., and Widom, J. (2006). Estimating Data Stream Quality for Object-Detection Applications. Proceedings of the 3rd International ACM SIGMOD Workshop on Information Quality in Information System.


References

[SPP+06] Subramaniam, S., Palpanas, T., Papadopoulos, D., Kalogeraki, V. & Gunopulos, D. (2006). Online outlier detection in sensor data using non-parametric models. Proceedings of the 32nd International Conference on Very Large Data Bases.187-198.

[TTD+09] Trajcevski, G., Tamassia, R., Ding, H., Scheuermann, P., Cruz, I. (2009). Continuous probabilistic nearest-neighbor queries for uncertain trajectories. EDBT. 874-885.

[ZCWL07] Zhuang, Y., Chen, L., Wang, X., & Lian, J. (2007). A weighted moving average-based approach for cleaning sensor data. Proceeding of the 27th International Conference on Distributed Computing Systems. doi: 10.1109/ICDCS.2007.83


Publications • S. Pumpichet, X. Jin, and N. Pissinou, “Sketch-based Data Recovery in Sensor Data

Streams,” in Proceedings of IEEE International Conference on Networks (ICON 2013), Singapore, December 11-13, 2013. (to appear)

• X. Jin, N. Pissinou, S. Pumpichet, C. Kamhoua, and K. Kwiat, “Modeling Cooperative Selfish and Malicious Behaviors for Trajectory Privacy Preservation using Bayesian Game Theory,” in Proceedings of IEEE International Conference on Local Computer Networks (LCN 2013), Sydney, Australia, October 21-24, 2013. (to appear)

• S. Pumpichet, N. Pissinou, X. Jin and D. Pan, “Belief-based Cleaning in Trajectory Sensor Streams,” in Proceedings of IEEE International Conference on Communications (ICC 2012), Ottawa, Canada, June 10-15, 2012, pp. 208-212.

• X. Jin, N. Pissinou, C. Chesneau, S. Pumpichet, and D. Pan, “Hiding Trajectory on the fly,” in Proceedings of IEEE International Conference on Communications (ICC 2012), Ottawa, Canada, June 10-15, 2012, pp. 403-407.

• S. Pumpichet and N. Pissinou, “Virtual Sensor for Mobile Sensor Data Cleaning,” in Proceedings of IEEE International Conference on Global Communications (GLOBECOM 2010), Miami, USA, December 6-10, 2010, pp. 1-5.



Thank you very much


1-3, 2014 PI Meeting 76

Documents

Using Trajectory Sensor Data Stream Cleaning to …...Using Trajectory Sensor Data Stream Cleaning to Ensure the Survivability of Mobile Wireless Sensor Networks in Cyberspace (new