data compression in wireless sensor networks
DESCRIPTION
various data compression techniques(algorithms) used in wireless sensor networks.TRANSCRIPT
DATA COMPRESSION IN WIRELESS SENSOR
NETWORKS
PRESNTATION BY
PRAKASH.K.M4GM08CS029
OVERVIEW
• Power consumption-critical problem in WSN’s
• Proposed techniques-The DC scheme reduces transmitted data over wireless channels.
• DC reduces no.of intermediate nodes
SKETCH OUT
• Introduction• DC in WSN’s for real world requirements• Comparison criteria for DC algorithms• DC algorithm classification• Distributed DC approaches• Local DC approaches• Performance of DC approaches in WSN’s• Conclusion
INTRODUCTION
• Ubiquitous computing era.– machines that fit the human environment
instead of forcing humans to enter theirs.– Computers disappearing into surrounding
objects – Supporting tech-Wireless Sensor Networks.
WSN’s•Self Organized group of sensors •Monitors the activities of target
WSN’s design challenges • Information management architecture -
manage information & resolve conflict• Encryption and Authentication• Power management in WSNs
Cont…
What is Data Compression?• DC is the representation of an information
source (e.g. a data file, a speech signal, an image, or a video signal) as accurately as possible using the fewest number of bits retaining its meaning.
DATA COMPRESSIONCont…
DC ALGORITHM IN WSN’S FOR REAL WORLD REQUIREMENTS Common requirements in real-world WSN applications
• power saving• Commercial sensor nodes requirements
– Processing constraints in sensor nodes
Uses micro controllers- Atmel Atmega128L and Texas Instruments MSP430 –instruction mem128kb & 48 kb respectively.
Specific requirements in real-world WSN applications
• 2 Types- Tracking And Monitoring
Patient/citizen centered healthcare based on wireless biosensors
Animal Tracking
Enemy Tracking In Military Operations
Study Environmental Behavior
Cont….
COMPARISON CRITERIA FOR DC ALGORITHMS
• Following criteria will be used to review, analyze and compare the performance.– Compression performance.– Power saved from reducing in transmission(x).– Power used when performing DC algorithms(y).– Net of power saving=(x)-(y).– Algorithm code size: with available node memory
the code size is analyzed.– Suitability of using data compression classes (lossless
or lossy)
DC ALGORITHM CLASSIFICATION
Compression in WSN’s
Distributed Approach
Distributed sourc
e modeling-DSM
Distributed Transform
Coding
(DTC)
Distributed sourc
e Coding(DSC
)
Compressio
n sensi
ng
Local Approach
Lossless Lossy
Dense sensor networks Sparse sensor networks
DISTRIBUTED DATA COMPRESSION APPROACHES
• Applied in dense sensor networks• Classified into four main techniques:
– Distributed source modeling (DSM).– Distributed transform coding (DTC).– Distributed source coding (DSC) and – Compressed sensing (CS) techniques.
DISTRIBUTED SOURCE MODELING (DSM)
Ex:parametric modelingd1,d2…dn-input dataq1,q2….qn-quantized values
Equation for data transmission
dn-random variable(data)
θ-is an unknown expectation value
search for a best fit function/model for set of inputs of a specific group of sensor nodes using parametric modeling and non-parametric modeling. Parametric modeling: treats sensor data as random process and processes it knowing mean and variance
dn=Θ+sn
• Optimizing quantization based on -estimation targetThis algorithm aims to minimize the transmitted power by optimizing the level of adaptive quantization while meeting estimation error.
• Optimizing quantization-based power target (OQPT) OQET’s objective is to optimize adaptive quantization by
estimating the performance and total power transmission target.
Cont…
Data representation-based hidden Markov model (DRHMM)
• Models a set of data by exploiting its temporal and spatial correlations.
• Manipulates observed values to obtain a binary value. • A binary value can result in either a target event occurring or a
target event not occurring. • Filters all inputs (binary data)- sent to the sink node &identifies
event.• Hence communication b/w nodes and sink is reduced & power is
optimized.
Cont…..
Non-parametric modeling• when data are sparse or prior knowledge is unclear.
Distributed kernel linear regression model (DKLR)• Models a local correlation of data using the kernel linear regression
function while using less communication.• there are two important correlations: temporal and spatial. • By exploiting these correlations, the model used to represent local
data is the regression
Distributed non-parametric kernel-based scheme (DNKB)
• Fits sensor data with non-linear regression functions.
Cont…..
DISTRIBUTED TRANSFORM CODING (DTC)
• It decomposes a source output using on transform theories into components/coefficients that are coded according to their characteristics.
• Approaches -Karhunen–Loeve transform, cosine transform and wavelets transformation
• used in wireless multimedia sensor networks.
Karhunen–Loeve transform (KLT) • Each data vector is compressed or encoded by means of KLT.• And encoding matrices are sent to the fusion center, it reconstructs
the entire data vector from these compressed sensor observations with minimal MSE.
• Reduce the data to be transmitted over the WSNs, they did not evaluate their power consumption performance
wavelets transform approaches• It allows a faster implementation of the wavelet transform
along with a full in-place calculation of the coefficients.
• The scheme consists of 3 steps: split, prediction and update. Split step, the signals s(n) are split into even signals and odd signals.• Both types of signals are then processed in prediction (P)and update (U) steps.• Finally, the detail coefficient d(n) and smooth coefficients y(n) are obtained
Cont….
DISTRIBUTED SOURCE CODING (DSC)
• Is a popular approach for data compression in WSNs. • This approach follows the Slepian and Wolf theorem -
separate encoding is as efficient as joint encoding for lossless compression.
COMPRESSED SENSING (CS)• Is based on a sampling theory that uses compressibility without
relying on any specific prior knowledge or assumption on signals. • considers a n-sample vector of signal (x) which is sparse. A vector x
is said to be p-sparse if it has at most p non-zero entries, with p<n.Projection vector A =[0.2,0.1,0.3,0.4] ,
• projection value y=A=0.2x1*0.1x2*0.3x3*0.4x4 is obtained by nodes without sending their sensor readings to the sink node.
• This can be achieved by the sink node passing a message along the route S–1–2–4–3–S using source
LOCAL DC APPROACHES
• Compression locally on each sensor node & no distributed Collaboration,
• suitable for sparse n/w with low spatial correlation• Lossless and Lossy approaches
LOSSLESS COMPRESSION
• Dictionary approaches-used to compress all kinds of data.• The second type is the lossless compression based on
predictive coding approaches.
• Two-modal transmission (TMT) scheme:
predictive coding configuration as follows:1. The error term were fitted with a Laplacian distribution rather than Gaussian distribution.2. A linear predictor was also employed.3. Arithmetic coding was also chosen as the coding scheme.
• Error term= original value ~ predicted value, It degrades the coding efficiency . • 1st modal transmission-compressed mode, error terms falling inside the interval [-R, R]. • 2nd modal transmission, called non-compressed mode, original raw data of error terms falling outside the interval [-R, R]
LOSSY COMPRESSION• Compression rate, which is higher than the lossless
compression algorithm.
Lightweight temporal compression (LTC) scheme • Designed for climate monitoring applications• Error is either uniform or unknown. Ex:
Differential pulse code modulation-based optimization
• 3 key points. 1st, the consecutive observations collected by WSNs .
• 2nd, important metrics e.g., compression ratio, information loss and power consumption. Finally, existing noise . which typically exploits the strong correlation that usually exists between neighboring samples.
• Additionally, de-noise techniques were applied on these data sets where loss of information is noise.
TMT LTC DPCM
~40% ~45%-75% ~85%-94%
12 instructions for saved bit
~49% instructions for a saved bit
6 instructions for a saved bit
~36% Yes ~66%
Lossless Lossy lossy
~40% ~45%-75% ~85%-94%
Do not exploit spatial correlations in WSN’s
DSM DTC
QQET/ QQPR
DRHMM DNKB KLT DWT-lifting
yes yes yes yes yes
Low cost Low cost/O(N)
O(N) Low cost Low cost
yes yes yes yes yes
lossy lossy lossy Loss less Loss less
optimal n/a n/a n/a Depends on SNR
limited to some specific applications
Limited to specific topologies & need inter sensor communication
DSC CS
n/a Yes
Low cost Low cost
50.7% yes
lossless Lossy
n/a Depends on data sparsity
Need to know an exact correlation among nodes
Need a suitable transformation in order to make good sparsity for real
world data
Comparison topics
Communication saving
Computational cost
Net power saving
Lossless/ lossy
Compression rate
Limitations/issues
PERFORMANCE & COMPARISION OF DC APPROACHES
ADVANTAGES
• Increase in performance of WSN’s by optimizing the power.
• Decreases the transmission time , cost and space.
• Provides good quality of service compared to aggregation
CONCLUSION & SCOPE
• We analyzed dc approaches based on real world requirements.
• Distributed approaches-need specific assumptions, local-less assumptions .
• Local compression-on Multiple data types is better than that of distributed.
• The effectiveness of data compression algorithms for a particular application is still an open issue requiring further investigation in WSNs.
• This issue is also open for further research study.
TO KNOW MORE….
• “Reducing power consumption in wireless sensor networks using a novel approach to data aggregation”- Croce S, Marcelloni F, Vecchio M.
• “Energy conservation in wireless sensor networks: a survey”-Anastasi G, Conti M, DiFrancesco M, Passarella A.
• “A distributed and adaptive signal processing approach to reducing energy consumption in sensor networks”. -Chou J, Petrovic D.
Thank yu….
Queries ???