RESEARCH ARTICLE

RESEARCH
ARTICLE

Edited by:
Azman Ahmad

First Edition
Copyright © KKTM Balik Pulau 2019

All rights reserved. No part of this book may be reproduced in any form or any
means without permission in writing from the authors.

Published by

KOLEJ KEMAHIRAN TINGGI MARA BALIK PULAU,
Jalan Genting, 11000 Balik Pulau,
Pulau Pinang, Malaysia

: +604 8665805 / 3670 : https://www.instagram.com/kktmbp/
: +604 866 5741 : https://www.facebook.com/kktmbp/
: http://balikpulau.kktm.edu.my/ : Lat,Long (5.34916,100.23938)

RESEARCH
ARTICLE

EDITORIAL BOARD

Ts. Hj. Azmin Ariffin
Director of Kolej Kemahiran Tinggi MARA Balik Pulau, Pulau Pinang

Ts. Dr. Azman Ahmad
Editor

Ts. Mohd Aidil Johari Abdul Hamid
Cover Design

Researchers: Azri Azman
Ts. Hasri Haris Mohd Azlim Mat Shariff
Dr Muhamad Zaki Yusup Mohd Hasruzairin Mohd Hashim
Ts. Dr Azman Ahmad Alimi Abdul Ghafar
Mohd Saiful Hazam Majid
Norhafiza Mohamed

Director’s Foreword

Thanks to Allah SWT for His permission, this Research Article book is
successfully published. This collection of articles/journals is
contributed by Kolej Kemahiran Tinggi MARA Balik Pulau’s staff serve
as a reference to the public to find the latest research information and
innovation ideas. The contributors for the articles/journals are those
who are expertise in their respective fields such as electronics
engineering, mechanical engineering and Technical and Vocational
Education and Training (TVET).

This book will also serve as encouragement to other staff to learn new
knowledge and to improve work quality and productivity. This is
important as it can enhance efficiency among the staff.

Lastly, I would like to express my deepest gratitude to the Editor and
the team members for publishing this Research Article book.
Hopefully this book can benefit to everyone. Thank you.

Ts. Hj. Azmin Ariffin
Director

Kolej Kemahiran Tinggi MARA
Balik Pulau, Pulau Pinang

Editorial

Alhamdulillah. Welcome to our First edition of Research Article, Kolej
Kemahiran Tinggi MARA Balik Pulau book, a collection of
articles/journals which involved of multi-discipline scopes such as
Artificial Intelligent, RF Microwave Engineering, Wireless System,
TVET – educational study, Lean Management, Tool and Die
technology; and which is for research sharing and knowledge by our
staff. Hopefully this book will be useful for anyone who interested in
educational and research field.

I’m gratefully acknowledge many individuals/contributors for their
generous contribution support to the Research Article book to be born
and made it happen. The list includes all current Editorial Board, the
researchers, and many others.

Azman Ahmad [PhD, PTech]
Kolej Kemahiran Tinggi MARA

Balik Pulau, Pulau Pinang
E-mail: [email protected]

RESEARCH ARTICLE page
1
contents

1. VISION-BASED PATH ESTIMATION FOR THE NAVIGATION
OF AUTONOMOUS ELECTRIC VEHICLE
Ts. Hasri Haris

2. NOISE FIGURE AND HIGH GAIN SINGLE STAGE DUAL 17
BAND LNA FOR WIRELESS APPLICATIONS
Ts. Dr Azman Ahmad

3. PERFORMANCE ASSESSMENT OF THE OPTIMUM 29
FEATURE EXTRACTION FOR UPPER-LIMB STROKE
REHABILITATION USING ANGULAR SEPARATION
METHOD
Mohd Saiful Hazam Majid

4. COMPARATIVE MEASUREMENT OF RF LINK BUDGET 47
BETWEEN SUI AND THEORETICAL MODELS IN A DENSELY

POPULATED AREA
Mohd Hasruzairin Mohd Hashim

5. RELATIONSHIP BETWEEN MANUFACTURING 59
SUSTAINABILITY AND LEAN PRODUCTION PRACTICES BY
MALAYSIAN MANUFACTURERS – A PRELIMINARY STUDY
Dr Muhamad Zaki Yusup

6. RESPONSE SURFACE METHODOLOGY (RSM) MODEL TO 72
EVALUATE SURFACE ROUGHNESS IN MACHINING OF
TITANIUM ALLOY (Ti-Al-4V) USING END MILLING
PROCESS
Mohd Azlim Mat Shariff

7. PRE-FORMING INSPECTION SYSTEM TO DETECT DEEP 91
DRAWING DEFECT DUE TO PUNCH-DIE MISALIGNMENT
USING IMAGE PROCESSING TECHNIQUE
Alimi Abdul Ghafar

8. PRACTICAL SKILLS ASSESSMENT IN MACHINING 106
COURSE BASED ON PSYCHOMOTOR DOMAIN

Azri Azman

RESEARCH ARTICLE page
119
contents

9. THE INFLUENCE OF ENVIRONMENTAL ACTIONS AND
CUSTOMER ACTIVITES IN GSCM ON OPERATIONAL
PERFROMANCE
Norhafiza Mohamed

10. IMPLEMENTATION OF LEAN AND CLEANER PRODUCTION 131
BY MALAYSIAN MANUFACTURERS – PRELIMINARY
SURVEY
Dr Muhamad Zaki Yusup

2018 Haris H.et al.

Vision-based Path Estimation for the Navigation of
Autonomous Electric Vehicle

Ts. Hasri Bin Haris*, Prof. Madya Ir. Ts. Dr Wan Khairunizam Wan Ahmad**,
Kamil S. Talha**

*Kolej Kemahiran Tinggi MARA Balik Pulau,
** School Of Mechatronic Engineering
Universiti Malaysia Perlis

[Email : [email protected]]

Abstract – Making an Autonomous Electric Vehicle (AEV) and able to
operate “unmanned” requires extensive theoretical as well as practical
knowledge. An AEV must be able to make decisions and respond to situations
completely on its own. Buggy car are used as Electric Vehicle (EV) and set up
with several equipment and sensor as an AEV. A camera is installed in front
of the AEV and is used to obtain image information of the road. On the other
hand, users or drivers do not have to directly contact with the main system
because it will autonomously control the devices by using fuzzy information
of the road conditions. This paper focuses on experimental used on vision
system of AEV. From the experimental results, the AEV has demonstrated a
robust performance for moving in straight line on line detection vision
navigation.

[Keywords: Autonomous Electric Vehicle (AEV), Vision System, Line Detection]

1. INTRODUCTION
Autonomous driving is increasingly attracting public interest due to

various research projects over the past years1-7. Usually, conventional cars are

1

converted with significant effort and many different sensors are placed on the
roof. The advance of electro-mobility provides the chance for completely
new vehicle concepts. By breaking away from classic approaches, it is
possible to consider and integrate autonomous driving into the vehicle
architecture with respect to IT and sensor systems, energy management and
design. These kinds of cars are the upgrade version of electric vehicle (EV).
Recently a lot of EVs and related vehicles such as a hybrid car have been
developed to solve environment and energy problems caused by the use of an
internal combustion engine vehicle. Developing such vehicles for solving the
environment and energy problems is a great idea. Currently, many researches
publish technical papers in journals, which are related to autonomous EV. In
their researches, steering wheel, brake and acceleration pedals are control by
using computers8-10. On the other hand, users and drivers do not have a direct
contact with them. A touch panel is installed in the EV and it serves a user
Graphical User Interface (GUI) for users and drivers interact with controlling
devices. Unfortunately, based on current outcomes more effort should be
done for making sure that autonomous EV could move with safety. Although
mechanism of mechanical could be used to solve safety and reliability issues
of autonomous EV, computational approach also very important. The
computational approach is for example the algorithm for controlling motor
device, the capacity of data transmission device, image processing technique
and etc.11-13. Autonomous vehicle with intelligent driving control are
developed to provide the driver assistance as well as unmanned driver for
road, logistics and flexible manufacturing system. It is an automatic guided
vehicle and able to move automatically along the road. This research will
focus on design and implementation of a sensor fusion system for navigation
and control of an autonomous electric vehicle. It also introduced the

2

intelligent vehicle trace, obstacle avoidance and speed control. In the first
part, a vision system for the electric vehicle will be develop by using image
processing techniques to recognize neighboring circumstances surrounding
the electric vehicle.

2. LITERATURE REVIEW

Vision system for lane lines detection is one important process based
driver assistance system and can be used for vehicle navigation, collision
prevention, lateral control, or lane lines departure warning system. Various
road conditions make this problem become very challenging including
different type of lane lines (straight or curvilinear), obstacles, lighting
changes (night time), occlusions caused by shadows, and so on. Therefore, in
the literature and recently, there have been many approaches proposed for
solving the above problems in lane lines detection. For example, Cheng et
al.14 used the colour feature to detect lane and utilized the size, shape and
motion for false lane region elimination. He et al.15 proposed a colour-based
vision system to detect lanes from urban traffic scenes. Yim and Oh16
combined three features including the starting position, intensity, and
direction for lane detection. In addition to the feature-based scheme, the
model-based scheme is more robust in lane detection when different lane
types with occlusions or shadows are handled. Kang and Jung17 proposed a
searching framework to group edges with similar directions as a road lane.
However, when complicated roads were handled, their method tended to
detect false candidates of lane. In Kluge and Lakshmanan18, a deformable
template model of lane structure is used to locate lane boundaries by using
intensity gradient information. Jung and Kelber19, used an approach that

3

combines an edge distribution function and the Hough transform with linear
parabolic model is developed for lane detection and lane tracking. Hechri et.
al.20, the Hough transform is very powerful because it can detect road
boundaries successfully even in an extremely snowy environment.

3. METHODOLOGY

A. Flow chart of the proposed works

RGB Image Acquisition

Converting to Gray Scale
Image

Converting to Binary Image

Edge Detection

Measure Distance (dL, dR)

Figure 1: Flow chart of proposed works

B. RGB Image Acquisition
For the line detection, first vision acquisition setup has been made.

Webcam with the resolution of 640 x 480 had been chosen. Image that
produces from webcam is in RGB color plane. By using the graphical
function in Lab VIEW, image from the RGB color space can be converted to
several color spaces. The best color plane in detection of the line will be
choosing to convert to gray scale image. For this research, we choose red
color plane as the line color. In the gray scale image, the threshold value for

4

the line will be determine for extracting line region in binary image.
The input data is a color image sequence taken from a moving

vehicle. A color camera is mounted in-front of the electric vehicle. It takes
the images of the environment in front of the vehicle. The on-board computer
with image capturing card will capture the images in real time (up to 30
frames/second), and save them in the computer memory. The lane detection
system reads the image sequence from the memory and starts processing. A
typical scene of the road ahead is depicted by Figure 2. In order to obtain
good estimates of lanes and improve the speed of the algorithm, the original
image size was reduced to 620x480 pixels by Gaussian pyramid.

Red line
marking on
the road

Figure 2: RGB Image

C. Converting to Gray Scale
To retain the color information and segment the road from the lane

boundaries using the color information this proved difficulties on edge
detection and also it will affect the processing time. In practice the road
surface can be made up of many different colors due to shadows, different
pavement style or age, which causes the color of the road surface and lane
markings to change from one image region to another. Therefore, color image
are converted into gray scale. However, the processing of gray scale images

5

becomes 84 minimal as compared to a color image. This function transforms
a 24-bit, three-channel, color image to an 8- bit, single-channel gray scale
image by forming a weighted sum of the Red component of the pixel value *
0.3 + Green component of the pixel value * 0.59 + Blue component for the
pixel value *0.11 the output is the gray scale value for the corresponding
pixel.

Red line
converted to
Gray Scale

Figure 3: Gray scale Image

D. Converting to Binary Image
A binary image is a digital image that has only two possible values for

each pixel. Typically the two colors used for a binary image are black and
white though any two colors can be used. The color used for the object(s) in
the image is the foreground color while the rest of the image is the
background color. In the document-scanning industry this is often referred to
as "bi-tonal".

A binary image can be stored in memory as a bitmap, a packed array
of bits. A 640×480 image requires 37.5 KiB of storage. Because of the small
size of the image files, fax machine and document management solutions
usually use this format. Most binary images also compress well with simple
run-length compression schemes.

6

Binary images can be interpreted as subsets of the two-dimensional
integer lattice Z2; the field of morphological image processing was largely
inspired by this view. Figure 4 shows the image after conversion to binary
Image.

Gray Scale
line converted
to Binary
Image

Figure 4: Binary Image

E. Edge Detection
Lane boundaries are defined by sharp contrast between the road

surface and painted lines or some type of non-pavement surface. These sharp
contrasts are edges in the image. Therefore edge detectors are very important
in determining the location of lane boundaries. It also reduces the amount of
learning data required by simplifying the image considerably, if the outline of
a road can be extracted from the image. The edge detector implemented for
this algorithm and the one that produced the best edge images from all the
edge detectors evaluated was the ‘canny’ edge detector14. To find the maxima
of the partial derivative of the image function I in the direction orthogonal to
the edge direction, and to smooth the signal along the edge direction.

The Sobel operator is used in image processing, particularly within
edge detection algorithms. Technically, it is a discrete differentiation
operator, computing an approximation of the gradient of the image intensity

7

function. At each point in the image, the result of the Sobel operator is either
the corresponding gradient vector or the norm of this vector. The Sobel
operator is based on convolving the image with a small, separable, and
integer valued filter in horizontal and vertical direction and is therefore
relatively inexpensive in terms of computations. On the other hand, the
gradient approximation that it produces is relatively crude, in particular for
high frequency variations in the image. The Kayyali operator for edge
detection is another operator generated from Sobel operator.

The operator uses two 3×3 kernels which are convolved with the
original image to calculate approximations of the derivatives - one for
horizontal changes, and one for vertical. If we define A as the source image,
and Gx and Gy are two images which at each point contain the horizontal and
vertical derivative approximations, the computations are as follows:

+1 0 −1 +1 +2 +1
= [+2 0 −2] ∗ = [ 0 0 0] ∗
+1 0 −1 −1 −2 −1

where here denotes the 2-dimensional convolution operation.

Since the Sobel kernels can be decomposed as the products of an averaging
and a differentiation kernel, they compute the gradient with smoothing. For
example, Gx can be written as

+1 0 −1 1
[+2 0 −2] = [2] [+1 0 −1]
+1 0 −1 1

The x-coordinate is defined here as increasing in the "right"-direction, and
the y-coordinate is defined as increasing in the "down"-direction. At each
point in the image, the resulting gradient approximations can be combined to

8

give the gradient magnitude, using:
= √ 2 + 2

Using this information, we can also calculate the gradient's direction:
= 2 ( , )

where, for example, is 0 for a vertical edge which is darker on the right
side.
F. Measure Distance (dL, dR)

The IMAQ Clamp Horizontal Max VI will measure a distance in the
horizontal direction, from the vertical sides of the search area towards the
center of the search area. This VI locates edges along a set of parallel search
lines, or rake. The edges are determined based on their contrast and slope.

A vertical hit-line to the object is calculated through the leftmost edge
detected. A second vertical hit-line to the object is calculated through the
rightmost edge. The distance between those two lines is returned. IMAQ
Clamp Horizontal Max can overlay on the image returned: the position of the
search area, the search lines, the edges found, and the result.

9

F = 640 pixels

Figure 5: Theoritical Frame for Distance dL and dR

Figure 5 show the theoretical frame to get the distance of dL and dR.
Where F is the size of camera frame to captured line on the road. dd is the
distance inside the line or width of the line. dL is the distance from left frame
to the line. dR is the distance from the right frame to the line. dy is the distance
from the left frame to the CoG. And CoG is the center of the gravity of the
line.

Based on CoG, dL is defined by the following distance equation;

= − (1)
2

and, = − (2)

Where F = 640 (size of the frame)

10

To filtering the noise from the graph, is defined by the following
filtering equation;

If > 35 (3)
And we assume that value of, = 30.2

4. EXPERIMENTS

A. Experimental Setup
To test and evaluate the proposed methodology on EV vision in

Chapter 3, the Buggy Car have been used and set up as stated on system
configuration. Figure 6 shows the buggy car used in the experiments. This
setup has been made at Putra Golf Club, Sungai Batu Pahat, Perlis. In the
experiment, the camera mounted in front of Buggy Car with noise reduction.

Figure 6: Buggy Car used in Experiments

11

This camera will detect the red line, which have been marking on the
road while Buggy Car moved. The image captured will be send to computer
(PC1). A Lab View 2012 was used as software to process the image as stated
in methodology (EV Vision).

Table 1: Experimental Parameters

Parameter Characteristic
Target Distance 5 meters
30 second
Time 10 times
Trial 6 cm
Line Width

B. Experimental Results

(a) Autonomous EV Controller Interface

12

Distance, dd(b) Screen shot of Tracking Image displayed at the touch panel
Figure 7: Autonomous EV Controller Interface displayed at the touch panel

Figure 7(a), show the screen shot of preprocessing the input image of
the line captured and converting to Gray scale and Threshold Gray scale
Image in the Lab View and Figure 7(b) is the screen shot of the tracking
image to get distance, dL and dR.

From the data of the experiments, we have plot the graph to show
AEV move.

Noises

Time (t)

Figure 8: Trajectories of line distance, dd with noise.

13

Distance, dd

Time (t)

Figure 9: Trajectories of line distance, dd after filtering.

Figure 8 shows trajectories of raw line distance dd with noise and
Figure 9 show trajectories of line distance dd after filtering. From the results,
to make the AEV move in straight line, it should be filtered.

Distance, dy Distance, dL

Figure 10: Trajectories of line distance for dy and dL

5. CONCLUSION
This paper presents the design and implementation of an autonomous

Electric Vehicle (EV) with intelligent driving control to provide the driver
assistance as well as unmanned driver. It is an automatic guided vehicle and

14

able to move automatically along the tracks in a given region. For the

purpose of prototyping, a buggy car has been used and several sensors are

installed. The camera has been installed to the EV as a vision system and

connects to the personal computer (PC) for the processing of image

information. Image processing algorithm will be employed for the detection

of line and the center of gravity (COG) of the road. For the future research,

another PC will be installed for controlling motors to operate acceleration

pedal, brake pedal and steering wheel. Information from several sensors was

fused to move the EV intelligently without control by the human.

REFERENCES

1. P.E Dumont, A Aitouche and M. Bayart, “Fault Tolerant Analysis for Multisensor Non
Linear Systems”, Application to An Autonomous Vehicle (IEEE Vehicle Power and
Propulsion), p198-203, 2004.

2. II Hammouri, M Kinnaert and E H El Yaagoubi, “Observer-Based Approach to Fault
Detection and Isolation for Nonlinear Systems”, (IEEE Transactions on Automatic
Control), vol 44, (No. l0) p 1879-1884, 1999.

3. M. A. Djeziri, A. Aitouche and B. Ould Bouamama, “Sensor Fault Detection Of
Energitic System Using Modified Parity Space Approach”, IEEE Conference on
Decision and Control, p 2578-2583, 2007.

4. K. Bouibed, A. Aitouche and M. Bayart, “Nonlinear Parity Space Applied to an
Autonomous Vehicle”, Journal of Energy and Power Engineering ISSN 1934-8975, Vol.
3 (No 12), (Serial No.25) p 10-18, 2009.

5. M. A. Djeziri, R. Merzouki and B. Ould Bouamama, “Robust Monitoring of Electric
Vehicle with Structured and Unstructured Uncertainties”, IEEE Transaction on
Vehicular Technology, ISSN 0018-9545, Vol. 58 (No. 9), p 4710-4719, 2009.

6. Han-Shue Tan, Rajesh Rajamani and Wei-Bin Zlhang, “Demonstration of an Automated
Highway Platoon System”, Proceedings of the American Control, 1998.

7. Systems Control Technology Inc. “Roadway Powered Vehicle Project rack Construction
and Testing Program Phase 3D”, California PATH Research Paper, 1994.

8. S. R. Cikanek and K. E. Bailey, “Regenerative Braking System for A Hybrid Electric
Vehicle”, Proceedings of the American Control Conference Anchorage, 2002.

9. H. X. Li and S. Guan, “Hybrid Intelligent Control Strategy: supervising a DCS-
controlled batch process”, IEEE Control Systems Magazine 21(3), 2001.

10.CHEN Quan-Shi, QIU Bin and XIE Qi-Cheng, “Fuel Cell Electric Vehicle”, Tsinghua

University Press, 2005.
11. Howlett P.G., Pudney P.J. and Xuan Vu, “Local Energy Minimization in Optimal Train

Control”, Journal of Automatica, Vol. 45 (No. 11), p 2692-2698, 2009.
12. General Definitions of Highspeed, International Union of Railways, 2009.

15

13. Masamichi Ogasa, “Energy Saving and Environmental Measures in Railway
Technologies: Example with Hybrid Electric Railway Vehicles”, IEEJ Trans. on
Electrical and Electronic Engineering, Vol. 3 (No.5), p 304-311, 2010.

14. H. Y. Cheng, et al., "Lane detection with moving vehicle in the traffic scenes". IEEE
Transactions on ITS, Vol. 7, pp. 571-582, 2006.

15. Y. He, H. Wang, and B. Zhang, "Color-Based Road Detection in Urban Traffic Scenes".
IEEE Transactions on ITS, vol. 5, pp. 309-318, 2004.

16. Y. U. Yim and S. Y. Oh, "Three-Feature Base Automatic Lane Detection Algorithm
(TFALDA) for Autonomous Driving". IEEE Transaction on ITS, vol. pp. 219-225,
2003.

17. D. J. Kang, M. H. Jung "Road lane segmentation using dynamic programming for active
safety vehicles". Pattern Recognition Letters, Vol. 24, Issue 16, pp. 3177-3185, 2003.

18. K. Kluge and S. Lakshmanan, "A deformable template approach to lane detection".
IEEE Intelligent Vehicle Symposium, pp.54-59, Sept. 1995.

19. C.R. Jung and C.R. Kelber, "A robust linear-parabolic model for lane following". The
XVII Brazilian Symposium on Computer Graphics and Image Processing, 2004.

20. Ahmed Hechri, Fayçal Hamdaoui, Anis Ladgham and Mtibaa Abdellatif, "Using fuzzy
logic path tracking for an autonomous robot". In International Review of Automatic
Control - Theory and Applications (IREACO), Praise Worthy Prize Publishers (Italy),
(ISSN: 1974 – 6059) Volume 4 (Issue 1).

21. Hasri Haris and Khairunizam Wan, “A Fusion of Sensors Information for Autonomous
Driving Control of an Electric Vehicle (EV)”, IOP Conference Series: Materials Science
and Engineering, Vol. 53, No.01 pp. 20-25, 2013.

22. M. Aly, “Real Time Detection of Lane Markers in Urban Streets”, Intelligent Vehicles
Symposium, p 7-12, 2008.

23. A. Hechri and A. Mtibaa, “Lanes and Road Signs Recognition for Driver Assistance
System”, International Journal of Computer Science, Vol. 8 (No.6), 2011.

[Publication from International Journal of Mechanical & Mechatronics Engineering IJMME-IJENS]

16

2018 Ahmad A.

Noise Figure and High Gain Single Stage Dual Band
LNA for Wireless Applications

Ts. Dr. Azman Bin Ahmad
Department of Industrial Mechatronics
Kolej Kemahiran Tinggi MARA Balik Pulau

[Email : [email protected]]

Abstract – A new technique to design a dual-band low noise amplifier (LNA)
using GaAs-pHEMT technology. The LNA utilized a single stage amplifier
with inductively source degeneration topology. A T-matching input network is
used to match the LNA load terminal, and a notch filter is used to separate
two concurrent frequency bands at 2.4 GHz and 5.75 GHz. The single stage
LNA was designed to improve and meet the standards of IEEE 802.11a/b/g.
The LNA simulation produced a noise figure (NF) of 0.39 dB and 0.76 dB at
center frequencies of 2.4 GHz and 5.75 GHz. The input reflection coefficient
|S11| of -13.2 dB and -24.25 dB, output reflection coefficient |S22| of -7.08 dB
and a -13.2 dB, power gain |S21| of 18.75 dB and 14.69 dB at a frequency of
2.4 GHz and 5.75 GHz. The supply voltage for the LNA is 2V.

[Keywords: Low noise amplifier, concurrent dual-band, input/output matching]

1. INTRODUCTION

The LNA key role is to provide sufficient gain and determine the
overall performance of the Radio Frequency (RF) receiver. The demand to
integrate the multi-band into a single transceiver is to save the die area and
power consumption compared to a single stand-alone LNA. In fact the multi-

17

band design also helps to minimize the packaging and testing costs1. Some
of the wireless gadget in the market such as Wi-Fi, WiMAX, LTE and mobile
phones offers multi-band coverage with dual-band, tri-band and even quad-
band which required multiplication of channel of the receiver. It is therefore
of high demand to design LNA’s for multi-band.

Figure 1 shows the concept behind the dual band LNA design. A new
technique need to improve the dual band LNA for wireless
telecommunication. Therefore, it is necessary to optimize the front-end
receiver performance, especially for the first device at front-end receiver
which is LNA. The front-end receiver should ensure that the signal is
received by the user has high gain, consumes low power and have an overall
low NF condition.

Ant1

Ant2

Ant1
Ant2

LNA

Figure 1: Dual band receiver design

Two front-end receivers with different frequency integrated into one
receiver and become a single LNA with a single signal path. The LNA also
has the same circuitry even it can reduce the size of the previous circuit. This
approach is implemented with new design of single stage dual band LNA.

18

2. LNA DESIGN CONSIDERATIONS

The basic LNA consists of three stages which are the input matching
network, the amplifier and the output matching network2. Figure 2 shows a
typical single stage LNA with input and output matching circuit. The LNA
design should meet the criteria required for stability, small signal gain and
bandwidth2. The initial development of the LNA design can be obtained by
deriving the formula and mathematical statements3.

ZS ZL

Input Transistor Output
Matching [S] GC Matching
Circuit Circuit GL

ΓS ΓL
ΓIN ΓOUT

DC Bias

Figure 2: Typical single-stage LNA

The designer must ensure that the input and output matching network
of the LNA design is to match the 50 Ω load terminal. The input and output
matching circuit is necessary to reduce the unwanted reflection of signal and
to improve the capability of transmission from source to load. The design
target specifications for the LNA are shown in Table 1.

Table 1: Targeted S-Parameters for LNA

S-parameter Single Stage LNA
Input Reflection S11 dB < -10
Return Loss S12 dB < -15
Forward Transfer S21 dB >+ 17
Output Reflection Loss S22 dB <-10
Noise Figure dB <3

19

Recently, most of the previous researcher used cascade topology with
more than 2 stage amplifier to design the dual band LNA. Furthermore, many
of the output gain also have been reported4-11 less than 15 dB. With a small
gain signal, the output from the front-end receiver signal also will be less or
weak. So, it is important to design the high gain and low NF to ensure the
stability of the circuit. Therefore, this new design proposes to improve these
criteria using dual-band concurrent LNA.

A. Power Gain
The power gain must be considered as the important parameter to

operate the LNA. Figure 3 shows the 2 port power gains with a power
network circuit impedance of the LNA. The gains are represented by
scattering coefficients classified into Operating Power Gain (GP), Power
Transducer Gain (GT) and Available Power Gain (GA)3.

ZS
V+ [S] V+

VS V- (ZO) ZL
ΓS ΓIN V-

ΓOUT ΓL

Figure 3: Input and output circuit of 2-port network12

The GP is the ratio of the power dissipated in the load ZL to the power
delivered from the source of the two-port network3. The GP can be expressed

in equation (1). ΓIN is a reflection coefficient of load at the input port of 2-

port network and ΓL is the load reflection coefficient.

= || || (1)

|| ||

20

While, the GT is the ratio of the power delivered to the load to the power

available from the source. The GT is given by an equation (2). ΓS is reflection

coefficient of power supplied to the input port.

= || | | | || (2)
|| |

The GA is the ratio of the power available from the network to the power

available from the source and can be obtained by the equation (3).

= || | | | | (3)
||

B. Noise Figure
In an RF communication system the receiver process very weak

signals, but the noise added by the system components tends to obscure those
very weak signals. Bit error ratio, sensitivity and NF are the parameters that
characterize the ability to process low level signals. Of these parameters, NF
is not only for characterizing the entire system, but also the system
components such as amplifier, mixer and IF amplifier that make up the
system. It provides the dominant effect on the overall system noise
performance3. Typically, noise figure of 2-port transistor has a minimum
value at the specified admittance given by14 as calculated in equation (5).

F  Fmin  RN | Ys  Yopt |2 (5)
GS

Usually the manufacturers provide Fmin, RN and Yopt data for low noise

transistors. The desired noise figure, N can be determined from equation (6).

N | s  opt |2  F  Fmin | 1  opt |2 (6)
 1 | s |2 4RN / Z0

When the stability of the active device is determined, input and output
matching circuits should be designed so that a reflection coefficient of each

21

port can be correlated with the conjugate complex number as shown in
equation (7) and (8).

IN  s*  S11  S12 S 21L (7)
1  S22L (8)

and

out *  S 22  S12 S 21s
 L 1  S11s

The source reflection coefficient should match with Γopt and the load
reflection coefficient should match with Γ*OUT with a complex conjugate
number to obtain a minimum noise figure using 2-port transistor as shows in
equation (9) and (10).

ΓS = Γopt (9)

and

L  o*ut   S22  S12 S 21s  (10)
 1  S11s

The GP is the ratio of the power dissipated in the load ZL to the power
delivered from the source of the two-port network3. The GP can be expressed
in equation (1). ΓIN is a reflection coefficient of load at the input port of 2-
port network and ΓL is the load reflection coefficient.

3. DESIGN OF SINGLE STAGE LNA
Dual band low noise amplifier design is based on the specifications

mentioned in the previous section. The transistor type of GaAs-pHEMT
(FHX76LP) used in this design. To arrive at a balance between noise figure,
gain and linearity, the device drain-source-current (IDS) was chosen to be
10mA with a 2V drain-to-source voltage (VDS); the gate-to-source voltage

22

(VGS) was -1V. The S-parameters for frequency 2.4 GHz and 5.75 GHz for
the chosen IDS and VDS from simulations of Advanced Design System 2008
(ADS) software is shown in Table 2.

Table 2: S-parameter from Transistor GaAs-pHEMT (FHX76LP).
[VDS=2V, VGS=-1V, IDS=10mA]

Freq. S11 S12 S21 S22
(2.4GHz)
0.949 4.874 0.065 0.455
Mag -84.075 93.425 34.450 -65.4
Ang

Freq. S11 S12 S21 S22
(5.75GHz)
0.949 5.411 0.033 0.555
Mag -35.480 142.56 65.200 -27.62
Ang

By using the ADS software, the overall performance of the dual band
LNA can be obtained especially for NF and GT. From the simulations, the ΓS
= 48.862 + j 22.125 and ΓL = 80.31 - j 54.68 at 2.4 GHz. Meanwhile at 5.75
GHz, the ΓS = 44.247 + j 0.519 and ΓL = 47.189 – j 25.108. The T- matching
network was used at the input and a notch filter in the output lumped reactive
elements and microstrip line impedance were used to design the element of
the circuit network. Figure 4 shows the circuit diagram of the proposed single
stage dual band LNA.

Vbias2

Vbias1 L5 C5 RF out
L1 L2 L6

C4

L4 C6

RF in C1

C2 L3
C3

Figure 4: The proposed single stage LNA amplifier

23

The design of single stage LNA has its own topology where notch
filter at the output impedance produce two notched frequency bands by
embedding a series L5C4 branch in parallel with an L6C5 (LC tank). Series
L5C4 gives minimum impedance at resonance, while LC tank gives maximum
impedance at its resonant frequency. The notch filter is placed after the gain
stage to avoid the decline in NF due to the loss of this filter16. The filter used
at the output matching network is to suppress the interfering frequency at 2.4
GHz and 5.75 GHz without affecting the noise figure (NF) of the LNA.

The single stage LNA was also constructed using an inductor
degenerated, L3. This technique reduces the interaction between output and
input stage; and can be separately optimized to improve the reverse
isolation17. Additionally, it gives the best performance to meet gain, NF,
stability and impedance matching goal with minimum power consumption

To obtain the maximum power transfer to the transistors, the input
and output matching networks must be considered. For this design we
proposed a T-matching network for input impedance which is C1, C2 and C3.
T-matching network gives the best performance to optimize the gain, NF
with minimum power consumption18.

4. SIMULATION AND RESULT
The result of simulations performed on the single stage dual band

LNA for the output gain shown in Figure 5 is 18.75 dB and 14.69 dB at
frequency band 2.4 GHz and 5.75 GHz respectively.

24

30dB(S(2,1))

18.75dB at 2.4GHz

20

10

0

-10 14.69dB at 5.75GHz

-20
12 34 567

freq, GHz

Figure 5: Output gain |S21| at 2.4 GHz and 5.75 GHz

The NF after load matching depicted in Figure 6. The simulation of NF is
0.39 dB at 2.4 GHz and 0.76 dB at 5.75 GHz.

2

0.39dB at 0.76dB at
2.4GHz 5.75GHz

nf(2) 1

dB(S(2,2)) 0
dB(S(1,1)) 1 234567

freq, GHz

Figure 6: NF at 2.4 GHz and 5.75 GHz

Figure 7 refers to the input return loss |S11| and output return loss |S22|, each
has value –13.2 dB and –7.08 dB at 2.4 GHz; – 24.25 dB and – 11.98 dB at

5.75 GHz respectively.

0

-5

-7.08dB

-10

-11.98dB

-15

-20 -13.2dB
-24.25dB

-25
1 2 34 5 6 7

freq, GHz

Figure 7: Input Return Loss |S11| and Output Return Loss |S22|

25

The comparison and lists of simulation/measured performance of the
dual band LNA are shown in Tables 3. It reveals that this work proves with
single stage amplifier also able to obtain the high gain and low noise figure
compared to a recent study by using two stage amplifiers.

Table 3: Comparison with recently dual band LNA

Author Stage Freq Gain NF (dB) S11 (dB) S22 (dB)
(GHz) (S21) dB
14.2/13.1 2.9/2.6 -10/-11.4 -12/-12
*[8] 2 2.4/5.2 21.8/12.5 2.5/2.4 -12.7/-17.4 NA
10.1/10.9 2.9/3.7 -10.1/-11
*[9] 2 2.4/5.2 -10.5/-17
18.8/14.7 0.39/0.76 -13.2/-24.3
[11] 2 2.4/5.2 -7.08/-11.9

*[This 1 2.4/5.75
work]

* simulation results

Figure 8 shows the layout of the proposed dual band gain with
single stage amplifier.

Figure 8: Layout of the concurrent dual band LNA with single stage
amplifier

5. CONCLUSION
This new dual band LNA is matched concurrently at the two

frequency bands and the input/output matching networks are designed with
T-matching and notch frequencies to shape the frequency response at 2.4

26

GHz and 5.75 GHz. The designed has been developed according to IEEE

standard of Wi-Fi and WiMAX applications. The dual band LNA has

multiple signal bands and provides a greater range and strength of the

overall signal. The advantage of dual band is that it is suitable for areas

that require more accessibility. For both bands of the LNA, the NF is less

than 1dB and it will provide overall low NFs for the front end receiver.

REFERENCES

1. S. Spiridon, C. Dan, and M. Bodea, “Overcoming the Challenges of Designing CMOS
Software Defined Radio Receivers Front-Ends Embedding Analog Signal
Conditioning,” 2012 13th Int. Conf. Optim. Electr. Electron. Equip., no. 1, pp. 1207–
1210, May 2012.

2. Yongguang Lu, Shu-hui Yang, Yinchao Chen “The Design of LNA Based on BJT
Working on 2.2-2.6GHz” 2010, 2nd International Conference on Signal Processing
Systems (ICSPS).

3. M. Pozar, David. Microwave and RF Wireless System. Third Avenue, N. Y. John Wiley
& Sons, in, 2001.

4. Perumana, B.G.; Zhan, J.-H.C.; Taylor, S.S.; Carlton, B.R.; Laskar, J., "Resistive-
Feedback CMOS Low-Noise Amplifiers for Multiband Applications," Microwave
Theory and Techniques, IEEE Transactions on , vol.56, no.5, pp.1218,1225, May 2008.

5. Kuan-Hsiu Chien; Hwann-Kaeo Chiou, "A 0.6–6.2 GHz wideband LNA using resistive
feedback and gate inductive peaking techniques for multiple standards
application," Microwave Conference Proceedings (APMC), 2013 Asia-Pacific , vol.,
no., pp.688,690, 5-8 Nov. 2013.

6. Hashemi, H.; Hajimiri, A., "Concurrent multiband low-noise amplifiers-theory, design,
and applications," Microwave Theory and Techniques, IEEE Transactions on , vol.50,
no.1, pp.288,301, Jan 2002.

7. Yu-Tso Lin; Shey-Shi Lu, "A 2.4/3.5/4.9/5.2/5.7-GHz concurrent multiband low noise
amplifier using InGaP/GaAs HBT technology," Microwave and Wireless Components
Letters, IEEE , vol.14, no.10, pp.463,465, Oct. 2004.

8. Dehqan, AR.; Mafinezhad, Khalil; Kargaran, E.; Nabovati, H., "Design of low voltage
low power dual-band LNA with forward body biasing technique," Electronics, Circuits
and Systems (ICECS), 2011 18th IEEE International Conference on , vol., no.,
pp.591,594, 11-14 Dec. 2011.

9. Kargaran, E.; Madadi, B., "Design of a Novel Dual-Band Concurrent CMOS LNA with
Current Reuse Topology," International Conference on Networking and Information
Technology, 2010 IEEE International Conference on , vol., no., pp.386-388, 2010.

10. S. Wang and B.-Z. Huang, "A high-gain CMOS LNA for 2.4/5.2-GHz WLAN
applications," Progress In Electromagnetics Research C, Vol. 21, 155-167, 2011.

11. Liang-Hung Lu; Hsieh-Hung Hsieh; Yu-Shun Wang, "A compact 2.4/5.2-GHz CMOS
dual-band low-noise amplifier," Microwave and Wireless Components Letters, IEEE ,
vol.15, no.10, pp.685,687, Oct. 2005.

12. Leon, Michael Angelo G. Lorenzo and Maria Theresa G.De. "Comparison of LNA

27

Topology for WiMAX Application in a Standard 90-nm CMOS Process." 12th
International Conference on Computer Modelling and Simulation. 2010. pp-642-647.
13. Ibrahim A.B,Husain M.N,Othman A.R,Johal M.S . “High Gain of Cascode LNA at
5.8GHz Using T-Matching Network for wireless Applications.” 2011, International
conference on Wireless and Optical Communication (ICWOC).
14. Othman A. R, Hamidon A. H, Abdul Wasli. C, Mustaffa M. F, Ting J. T. H, Ibrahim
A.B, “Low Noise, High Gain RF Front End Receiver at 5.8GHz for WiMAX
Application.”. Journal of Telecommunication and Computer Engineering (JTEC 2010).
15. Othman A.R, Hamidon A.H, Abdul Wasli. C, Mustaffa M. F, Ting J. T. H, “ High Gain
and Low Noise Cascaded Low Noise Amplifier Using T Matching Network,”. 4th
International Symposium on Broadband Communication (ISBC 2010).
16. Ch.Liang, P. Rao, T. Huang, and Sh.Chung, “Analysis and Design of Two Low-Power
Ultra-Wideband CMOS Low-Noise Amplifiers With Out-Band Rejection,” IEEE Trans.
Micro Wave Theory AND Tech,vol. 58, no. 2, pp.277-286, Feb. 2010.
17. Lian L.L., Noh N.M., Mustaffa M.T., Manaf A.B.A., Sidek, O.B., “A Dual-band LNA
with 0.18-μm CMOS Switches”, Micro and Nanoelectronics (RSM), 2011 IEEE
Regional Symposium on Micro and Nano Electronics , vol., no., pp.172-176, 28-30
Sept. 2011.
18. A.R.Othman, A.B.Ibrahim, M.N.Husain, A.H.Hamidon, and J.Hamidon, “Low Noise
Figure of Cascaded LNA at 5.8 GHz Using T-Matching Network for WiMAX
Applications,” 2012 International Journal of Innovation, Management and Technology
(IJMIT), vol. 3, no. 6, pp. 688–691, Dec. 2012.

28

2018 MSH Majid et al.

Performance Assessment of the Optimum Feature
Extraction for Upper-limb Stroke Rehabilitation using

Angular Separation Method

Mohd Saiful Hazam Majid*, Prof. Madya Ir. Ts. Dr Wan Khairunizam Wan
Ahmad**

*Department of Industrial Mechatronics
Kolej Kemahiran Tinggi MARA Balik Pulau,

**School of Mechatronic Engineering
Universiti Malaysia Perlis

[Email : [email protected]]

Abstract – Most of human everyday activities will require the use of their
upper-limb muscles. The pattern of upper-limb muscle movement can be used
to estimate upper-limb motions. Fundamental arm movement which is part of
upper-limb muscle rehabilitation activity have been studied in order to
investigate the time domain features, frequency domain, and time frequency
domain from the surface electromyogram (sEMG) signal of the upper-limb
muscle. The relationship of electromyogram (EMG) signal and the
rehabilitation exercise of related upper limb muscles movements are analyzed
in this study. Then the features from the three domains were compared using
Angular Separation Method to determine optimal feature. The result shows
that MinWT has the best value of similarity which is 0.98, followed by
MeanWT feature which resulted 0.91 of similarity. These results of EMG
signal feature extraction can be used later in the study of human upper-limb
muscle especially for analyzing EMG signal from patient undergone a
rehabilitation treatment.

[Keywords: Feature Extraction, Electromyogram (EMG), Upper-limb, rehabilitation,
Angular Separation Method]

29

1. INTRODUCTION

According to National Stroke Association of Malaysia (Nasam)1, a
large number of new stroke patient listed in Malaysia each year with an
estimation of 40,000 people. Stroke patient could experience one side of their
body to become paralyzed, a medical condition which is called hemiplegia,
or one side of their body to become weak, a condition termed as hemiparesis.
These conditions will effects the quality of life of the patients to perform
daily activities such as eating, drinking, brushing teeth, combing hair and
washing face. The main goal of stroke rehabilitation is to regain
independence and improve life’s quality2. Specifically upper-limb
rehabilitation will aim stroke patient to be able to control their limb and have
stronger and healthier muscle so that they will be able to perform their
everyday task without being dependent of others and thus will get their
quality of life back. Consequently optimum method needs to be selected to
provide the appropriate and the good result of stroke rehabilitation. In this
research, the surface electromyography (sEMG) signals of stroke patient
performing upper-limb rehabilitation will be analyzed and optimum muscle
and feature extraction will be selected.

The upper-limb rehabilitation will be done based on fundamental arm
movement and functional arm movement. Among the objective is for the
stroke patient to regain its muscle strength and coordination so that they can
function as normal as before the stroke attack and gain quality of life. The
research of using sEMG signal to study muscle performance during upper-
limb rehabilitation has yet to be explored especially in the aspect of
determining the best muscle to be monitored and to be evaluated.
Furthermore, in recent trend where rehabilitation is to be performed at home

30

or without hospitalization, the flexible and user friendly rehabilitation system
needs to be designed such as the selection of optimum muscle to monitor
patient performance and progress. In addition the selection of optimum
feature extraction could reduce the complexity of the system and calculation
time.

Monitoring of sEMG signal is the main feature of this upper-limb
rehabilitation research works. Electromyography is the study of muscle
function through analysis of the electrical signals emanated during muscular
contractions and activation of the muscle3. So studies of EMG are for the
purposes of analyzing the relationship of muscular function to movement of
the body segments and evaluate timing of muscle activity as well as muscle
activation pattern with regard to the movements of muscle3. There are two
types of electrodes used to record EMG signal from a person which are fine
wire electrodes and surface electrodes hence the name surface EMG coming
from the fact that surface electrodes techniques are used for signal recording.
Further information on how the sEMG signal being acquired and processed
will be discussed in the next section.

2. LITERATURE REVIEW

Feature extraction is the transformation of raw signals into a set of
parameters that represent the signal. This is necessary to minimize
complexity of implementation and to reduce the cost of information.
Generally, features extraction from sEMG signal can be obtained in time
domain, frequency domain and time-frequency domain4. The advantages of
using time domain in parameters extraction are easy and quick. Besides, it
has lower computational complexity5. Based on numerous studies, the

31

selection of parameters in time domain ought to be carefully considered. It
was found that a lot of redundant time domain parameters were used to
extract muscle fatigue signals but the results were not reliable because it has
lower accuracy4. However according to Fernando D Farfán in his paper, in
time domain there are two parameters that are commonly used in EMG
feature extraction which are the root-mean-squared (RMS) value and the
average rectified value. According to him also, both parameters are good
where EMG signal amplitude could be obtained from them6.

On the other hand, processing EMG signal in frequency domain
would be required if the signal’s frequency information is needed for
example in studying muscle fatigue. In this case feature like power spectral
density (PSD) in frequency domain could be used7-9. However there is also
drawback of analyzing EMG signal using frequency domain where time
information is excluded.

To move on further, there are also research on presenting feature
extraction in time domain specifically using Wavelet Transform analysis7, 10-
11. The time frequency domain analysis is said to be the solution which
cannot be solved by time domain analysis which is lack of frequency
information and to be the solution which cannot be settled by frequency
domain analysis which is lack of time information. In fact since the nature of
EMG signal is non-stationary signal, so the time domain analysis will be the
best tool to process this particular biomedical signal12. According to S.
Sharma et. al. in their paper13 however, a high-dimensional feature vector
will be produce if signal analysis are performed using Wavelet Transform.
Since among the objective of the designed technique is to reduce time
consuming, an increase in high-dimensional feature vector could mean an

32

increase in computational time thus it is not desired. As a result, they advised
in wavelet analysis, it is important to select an optimal dimensionality
reduction method to solve the problems.

So in conclusion, there are pros and cons of feature extraction in the
three domains. However the research in this area of determining best feature
will be much according to the objective of their research. For this study, 17
features from the three domains have been applied to find optimal feature that
will be used later in our research project which is designing an upper-limb
rehabilitation system for post stroke patient using sEMG signal as patient
improvement indicator. Thus the selection of optimal method, hence simpler
and time efficient systems is highly desired. So for this study, root mean
square, integral of EMG, simple square integral, mean absolute value and
modified mean absolute value of time domain are applied. Feature such as
auto regressive, total power and mean frequency of frequency domain are
also applied for comparison. Besides feature in the previous two domains,
feature in time frequency domain are applied to justify the optimal results.
Some researcher use method such as Euclidean Distance to select optimal
feature14 but in this study we analyze and compare the extracted features
using Angular Separation method.

3. METHODOLOGY

A. Flowchart of the Proposed Work
Figure 1 shows the flowchart of the proposed work for this research.

Subjects were asked to perform fundamental arm movement which is elbow
flexion four times to record sEMG signal from bicep brachii muscle. The
signals were then being extracted to obtain 17 features and consecutively the

33

features will be further analyzed using Angular Separation Method to find
optimum one.

Arm movement

EMG signal acquisition

Signal preprocessing

Wavelet decomposition

Wavelet coefficient subsets
(cDn cAn)

EMG feature extraction

Angular separation method

Figure 1: Flowchart of the Proposed Work

B. Experimental Protocol
1) Subject
Subject selected in this study were coming from healthy male subject,

five person ages between 21 to 27 years old. To further narrow the scope,
only right-handed subject were selected.

2) System design
This fundamental EMG bio-amplifier system has been used equipment for
acquiring sEMG signal. Figure 1 also can be regarded as the block diagram
of the system for this experiment.
These electrodes of type silver-silver-chloride (Ag/AgCl) self-adhesive
were used in the designs that were placed on subject’s skin to capture the
sEMG voltage signal. In the study, muscle belly of bicep brachii muscle were

34

selected as the placement of electrodes while reference electrode were placed
at the bony area of subject’s other hand.. The Ag/AgCl electrodes are chosen
for this experiment due to their half-cell potential is closer to zero. A total of
three Ag/AgCl electrodes were used in this experiment for each subject.
Figure 2 shows the electrodes placement of the subject muscle.

Figure 2: Electrodes placement on bicep muscle
The placement of surface electrodes is among the crucial issue in
EMG signal recording. According to Ahamed et. al. in their paper15, the best
result could be obtained if the electrodes were placed at muscle belly of the
selected muscle. Besides electrodes placement skin preparation prior to any
experiment recording is also important and has to be neatly prepared to avoid
unwanted noise in the recorded data. So in this study, electrodes placement
and skin preparation was done in according to procedure specified by
SENIAM (Surface Electromyography for the Noninvasive Assessment of
Muscles) and ISEK (International Society of Electrophysiology and
Kinesiology).
Subsequently, the signal were amplified and recorded by using
PHYWE COBRA3 bio-amplifier depicted in Figure 3 at sampling frequency
of 1 kHz. The signal were then transferred and processed by MATLAB
software in a personal computer whereas among the early process is

35

rectification and filtration. Further elaboration on the technique will be
discussed in the next section.

Figure 3: Bio amplifier PHYWE COBRA3
C. Signal Processing

In this experiment subjects were asked to perform fundamental arm
movement which is elbow flexion for four times. Each subject performs two
trial of the fundamental movement and data were recorded while subject
perform the movement in sitting position where their hand were rested on a
table initially. EMG data obtained from two electrodes placed on the muscle
belly of bicep brachii of each subjects.

In the rectification process, all the negative values of the signal were
translated to positive values for the next process. This is to ensure that that
data will not be zero if the averaging of mean were to take in the next step.
For the filtration process, the Butterworth low pass filter of the order of five
has been used in accordance to previous literature and the cut-off frequency
was set at 10 Hz to attenuate noise. Later, the linear envelope of the signal as
describe by R. Suhaimi et. al. in their paper16 was obtained.

36

Afterwards seven time domain variables are measured as a function
of time17. Because of their computational simplicity, time domain features or
linear techniques are the most popular in EMG pattern recognition.
Integrated EMG (IEMG), Root mean square (RMS), Waveform length (WL),
Simple square integral (SSI), Mean absolute value (MAV), Modified mean
absolute value 1 (MMAV1), Modified mean absolute value 2 (MMAV2) of
EMG are used to test the performance. All of them can be done in real-time
and electronically and it is simple for implementation. Features in this group
are normally used for onset detection, muscle contraction and muscle activity
detection. Moreover, features in frequency domain are used to represent the
detect muscle fatigue and neural abnormalities, and sometime are used in
EMG pattern recognition. Six features in frequency spectrum are performed
namely Autoregressive coefficients (AR), Total power (TTP), Mean
frequency (MNF), Median frequencies (MDF), Modified mean frequencies
(MMNF) and Frequency ratio (FR)17, 18. Table 1 shows the mathematical
definition of features in time domain and frequency domain that has been
applied in this study.

37

Table 1: Mathematical definitions of 13 sEMG feature extraction methods.

Let xn represents the nth sample of the sEMG signal (S). N denotes the length
of the sEMG signal. wn is the continuous weighting window function.

Threshold is used to avoid low-voltage fluctuations or background noises. Pj
is the sEMG power spectrum at frequency bin j. Aj is the sEMG amplitude
spectrum at frequency bin j. fj is the frequency of the sEMG power spectrum

at frequency bin j. M is the length of the frequency bin17, 18

Feature extraction Mathematical definition
IEMG = ∑ |x |
Integrated EMG
(IEMG) RMS = ∑

Root Mean Square
(RMS)

Waveform length WL = ∑ | − |
(WL) SSI = ∑ | |

Simple Square
Integral (SSI)

Mean Absolute Value MAV = ∑ | |
(MAV)
MMAV1 = ∑ | |;
Modified Mean
Absolute Value 1 = 1, 0.25 ≤ ≤ 0.75
(MMAV1) 0.5, ℎ

Modified Mean MMAV2 = ∑ | |;
Absolute Value 2
(MMAV2) 1, 0.25 ≤ ≤ 0.75
= 4 ⁄ , 0.25 >

4( − )⁄ , 0.75 <

Autoregressive =− +
Coefficients (AR)

Where is AR coefficients, error

sequence, and is the order of AR model

Total power (TTP) TTP = ∑

Mean Frequency MNF=∑ ⁄∑
(MNF)
= 1
Median Frequency =2
(MDF)

Modified Mean MMNF = ∑ ∑
Frequency (MMNF) FR = ∑ ∑

Frequency Ratio (FR)

38

Besides features in Table 1, features in time frequency domain were
used to do the non-stationary signal analysis19. This is due to the drawbacks
of time domain feature extraction where the result is not consistent with the
non-stationary nature of EMG signal as according to comment by F. A.
Mahdavi and his friends in their research paper12 while features extracted in
frequency domain has the problem of not having access of time domain. So
four extracted features in time frequency domain that were derived from
Wavelet Transform analysis are Minimum value, Maximum value, Mean,
and Standard Deviation.

Wavelet analysis based feature extraction for EMG pattern
classification usually requires using best Mother Wavelet to uplift the
functionality of EMG classification. Some other researchers had used db2
considering the good operation of Daubechies family in EMG feature
extraction12 but in this study db7 is used.

Four levels' decomposition (see Figure 4) of the signal was performed
using Mallat algorithm. cA is low-frequency components of the signals, often
called approximations. cD is the high-frequency components, called details.
After the wavelet transform, the low frequency and the high frequency are
separated from the signal (S) to obtain approximate values (cA) and detail
values (cD), respectively, as demonstrated in Figure 4. From the figure, one-
dimensional wavelet is used to decompose an original signal into four
levels18 that can be represented in equation (1).

39

S cD1
cA1
cA2 cD2
cA3 cD3
cA4 cD4

Figure 4: Four levels wavelet decomposition tree

S = cA5 + cD1 + cD2 + cD3 + cD4 (1)

From here, the feature extractions were accomplished by using
Minimum value (MinWT), Maximum value (MaxWT), Mean (MeanWT),
and Standard Deviation (SDWT).

In this study, prior to feature extraction step the preprocessed EMG
signal will first be segmented to 256ms adjacent windowing technique as
suggested in12. This is done to create extracted feature waveform from the
sEMG signal so that the waveform is easier to visualize by researcher for
further analysis. In this case the extracted feature waveforms will be
compared in the next method to determine the optimum feature. Furthermore
the windowing technique for EMG signal analysis could also be used to
avoid more information losses while performing feature extraction on the
whole signal data points.

The final step in the signal processing stage is the Angular Separation
Method. This method has been used to evaluate which feature gives the best

40

similarity measurement result. According to Kardi Teknomo in20 the formula
of Angular Separation can be expressed as equation (2).

= ∑ × (2)
×∑
∑

where is the element kth of signal i and is element kth of signal j.

In this research the two extracted EMG signal for each features
recorded from either same subject with different trial or from two different
subjects were analyzed using formula (2). Results of the experiment and the
comparison between features are discussed in the next section.

4. RESULTS AND DISCUSSION

Result of the filtered signal is the removal of the unwanted noise from
raw EMG signal such as movement artifact, DC offset as well as to attain
only the desired sEMG frequency where the pass band lies in 10 to 500 Hz.
Among the goal of research is to study muscle activation pattern that is the
pattern of the onset of the EMG amplitude when there is a contraction where
the muscle involves in this case is the bicep brachii muscle. So from graph
depicted in Figure 5, its can be seen that the amplitude arise to its maximum
value 4 times hence the subject flexed their elbow four consecutive times.
The timing for each contraction to occur was not measured specifically
because our goal is to do rehabilitation on real stroke patient where only
amplitude pattern will be monitored. In this figure there are four peak of
amplitude and the highest are 0.21 mV. The peak voltage differ between the
four peaks due to inconsistency of muscle contraction and force produce by

41

subject’s hand movement probably due unfamiliarity with the exercise and
the muscle fatigue21.

The main result for this paper were tabulated in Table 2 where 17
features extraction techniques were applied to gain useful information of the
recorded EMG signal. At a time two EMG signals were analyzed to obtain
two versions of each feature. Then they will be analyzed using angular
separation method to check similarities between the waveforms where the
most similar waveform comparison will get the result almost to 1.0. In the
table, for angular separation result between different subjects waveform,
feature MinWT gave the highest score which is 0.98 followed by 0.91 by
MeanWT. Similarly the result for waveform extracted from EMG signal of
same subject with two trials, MinWT gave the highest score which is 0.97
while MeanWT gave 0.95. The consistency of the result was obtained when
angular separation method was applied to compare extracted features with
more subjects and more trials.

So in future if less computational time is required and optimum result
is desired in EMG signal processing, MinWT will be the best selected feature
to extract. Next would be the MeanWT features.

42