You are on page 1of 47

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/343452402

Tradeoffs Between Sensing Quality and Energy Efficiency for Context


Monitoring Applications

Presentation · January 2016

CITATIONS READS

0 15

3 authors, including:

Sujan Sarker Amit Kumar Nath


University of Dhaka Florida State University
28 PUBLICATIONS   199 CITATIONS    12 PUBLICATIONS   20 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Exploring RDMA Plugin for Apache Spark View project

Personal Image Retrieval View project

All content following this page was uploaded by Amit Kumar Nath on 05 August 2020.

The user has requested enhancement of the downloaded file.


Tradeoffs Between Sensing Quality and Energy
Efficiency for Context Monitoring Applications

2nd International Conference on Networking


Systems and Security (NSysS)
8 January, 2016

Authors
Sujan Sarker, Amit Kumar Nath, Md. Abdur Razzaque
Green Networking Research (GNR) Group
Faculty of Engineering and Technology
Dept. of Computer Science and Engineering
University of Dhaka
Bangladesh
Overview
2

Introduction
Related Works

System Model

Proposed Strategies for Energy Efficiency

Performance Evaluation

Conclusion

References
3

Introduction
Introduction
4
Introduction
5

 New generation smart


devices provide various
functionalities

 Smartphones contain
several embedded
sensors

 Sensing operations in
different domains are
enabled
Introduction
6
Introduction 7

 Context related information is


required in various tasks

 Context monitoring applications


provide such facilities

 Smartphone sensors can be used


for capturing contextual
information
Introduction
 For accurately capturing
8

contextual information, sensors


must operate continuously

 As a result huge energy


consumption occurs, decreasing
battery lifetime

 A tradeoff in between sensing


accuracy and energy efficiency is
required
Introduction 9

 Over sensing and under sensing


must be avoided to make the
tradeoff

 Optimal calibration of sensing


frequency is required!!

 Optimal calibration can be


achieved by dynamically adjusting
sampling frequency and duty cycle
10

Related Works
[4] Y. Wang, J. Lin, M. Annavaram, Q. A. Jacobson, J. Hong, B. Krishnamachari, and N.
11
Sadeh,“A framework of energy efficient mobile sensing for automatic user state
recognition”, in MobiSys ’09: Proceedings of the 7th international conference on
Mobile systems, applications, and services, New York, NY, USA, 2009.

Drawbacks:
• Sensors have fixed
duty cycles
whenever they are
active
• They are not
adjustable to
different user
behavior
12
[12] K. K. Rachuri, M. Musolesi, and C. Mascolo, “Energy-accuracy trade-offs in querying
sensor data for continuous sensing mobile systems”, in Proc. of Mobile Context Awareness
Workshop, vol. 10. Citeseer, 2010.

Drawbacks:
• Function based
approch doesn’t
always ensure high
performance
• Thresholds are set
manually
[5] O. Yurur, C. Liu, X. Liu, and W. Moreno, “Adaptive sampling and duty cycling for 13
smartphone accelerometer,” in Mobile Ad-Hoc and Sensor Systems (MASS), 2013 IEEE
10th International Conference on, Oct 2013.

Drawbacks:
• Adaptation method is
inefficient
• Both the parameters
(duty cycle and
sampling frequency)
cannot be adjusted
simultaneously
14

Contributions
Developed a framework for dynamic sensing and
monitoring applications

Developed an intelligent context monitoring


scheme based on K-means clustering algorithm

Used MIMD based dynamic controlling of sensing


frequency
15

System Model
16

The System Model


Extracts
Determines whether new and different contextual
further
Filters necessary information from raw
information
processing on the currently to determine a possible
sensed
sensor data
user state
data will be carried out ortransition
not

Keeps track of the operational Recognizes user state


frequency and the length of the duty transitions
cycle of the sensor
17

Proposed Strategies for Energy


Efficiency
Context Monitoring Mechanism 18

(1/4)
• Proposed an intelligent computational method to
recognize contextual information of the user
• Conventional methods process all raw sensor data
• Proposed mechanism notifies the application about the
user’s state only when a state transition occurs
• When a raw sensor data arrives, it is compared with
previously buffered data to decide whether it indicates
a possible change of user’s current state or not
Context Monitoring Mechanism 19

(2/4)
• Let, for each sensor there is a there is a buffer Bi of
size N which stores N previously sensed data
• Bi = {X1, X2,……, XN} where Xj denotes the jth data
tuple sensed by sensor Ai
• We applied K-means clustering
k is the number algorithm to partition Bi
into K clusters, of clusters
• Equation used for clustering: mean of k-th cluster
Context Monitoring Mechanism 20

(3/4)
• Let Xj is the current sample and Xj-1 is the previous sample stored
in the buffer
• Mean and standard deviation (k) of kth cluster, Ck are calculated:
Context Monitoring Mechanism 21

(4/4)
Sensor Operation Structure 22

(1/2)
Initialization time Termination
Active cycle time time
Sampling interval

Sampling period
Sensor Operation Structure 23

(2/2)
• Tc: Time required for an operational cycle
• Number of samples N, taken in Tc: N = f x d x Tc
• Ttotal: Total running time
• Sensor’s active running time for taking samples:
Trun = Ttotal - (Tinit + Tter)
• Number of active cycles run by a sensor between Tinit
and Tter, Ncycle = Trun/Tc
• Sampling interval within which time a sample is taken:
Ts = Tc/N
• Sampling period, ts = 1/f
Dynamic Frequency Calibration 24

(1/8)
• The actual number of required
i-th datareadings between two
(i-1)-th data
consecutive samples cantuple
be calculated
tuple as,
Max allowable
diff.
Dynamic Frequency Calibration 25

(2/8)
• Sampling frequency is limited to
• The maximum and minimum numbers of samples that
can be taken during Ts are –

• The actual number of required samples is given by –


Dynamic Frequency Calibration 26

(3/8)
• To reduce energy consumption, a pair of duty cycle
and sampling frequency is assigned to each sensor
• A tradeoff between energy consumption vs.
accuracy arises, which has to be balanced
efficiently
• Our proposed mechanism dynamically adjusts
sampling frequency and duty cycle to balance this
tradeoff
Dynamic Frequency Calibration 27

(4/8)
• We define base B as,

• We consider a number system of base B whose digits can be any


values between 0 and B - 1.
• For N-1 consecutive samples we generate two numbers Ncur and
Nmax taking Nreqi and Nmax as digits respectively
Dynamic Frequency Calibration 28

(5/8)
• Using Ncur and Nmax we can determine sampling
frequency for the next cycle:

• According to fnext, the duty cycle for the next cycle


dnext can be adjusted as-
Dynamic Frequency Calibration 29

(6/8)
• Since sensor’s frequency can’t be calibrated to any continuous
value, we must map the fnext and dnext pair to the available
discrete values
• Let F and D denote the sets of all available frequencies and duty
cycles.
• The required mapping function can be defined as,

• If then the mapping function can be


modified as:
Dynamic Frequency Calibration 30

(7/8)
• The sensor’s sampling frequency is calibrated to φ(fnext) and
φ(dnext) is assigned as the active duty cycle in the next cycle, as
summarized in Algorithm 2:
Dynamic Frequency Calibration 31

(8/8)
• Overall sensor operation scheme is presented in Algorithm 3:
32

Performance Evaluation
33

Environment Setup
• We study the performance of our proposed system with
comparison to the methods described in [5]
• For our convenience, we denote the methods described in
[5] as ASDC (Adaptive Sampling and Duty Cycling)
• Performance measurement is done in terms of accuracy
and power efficiency
• A smartphone application is implemented in order to
evaluate the performance of the proposed system
• The application collects contextual data using
accelerometer sensor
• Samsung Galaxy S-DUOS smartphone is used as the target
device
• Android Studio is used as software development tool
34

Performance Metrics (1/2)


• Accuracy
― Accuracy between ith and (i -1)th samples can be
calculated as:

• Average accuracy for N samples taken in Trun time:


35

Performance Metrics (2/2)


• Power Efficiency
– Power Efficiency is defined as the ratio of remaining energy and
initial energy budget. For calculation of power efficiency we adopt
power consumption analysis of [5]

• Simulation parameters:
Comparison Between ASDC and 36

Proposed Method (1/2)


Rate of increase in proposed method is
greater than that of ASDC
Comparison Between ASDC and 37

Proposed Method (2/2)

Rate of decreasing is slower


in our proposed method
38

Conclusion
39

Conclusion
• Introduced novel approaches to address the energy
accuracy tradeoff for smartphone sensing
• Developed Several algorithms to improve the energy
efficiency of context monitoring applications while
capturing accurate contextual information
• The use of MIMD in adaptively scaling the sensing frequency
has been explored and the results prove that it is more
effective than AIAD (additive increase additive decrease)
40

Future Works
• Formulation of an optimization function is being explored
• Optimization based approach might further improve
tradeoff levels
• Whenever the application environment parameters greatly
vary over time, optimization function would be able to
address and make necessary changes
41

References
References (1/3) 42
References (2/3) 43
References (3/3) 44
Any questions?

View publication stats

You might also like