Development of a low cost EMG sensor, 2019

Fetal arrhythmia detection, 2019

Neurodegenerative disease classification using gait signal, 2019

Weather condition independent cell phone location detection system, 2018

[Bachelor Thesis] Blood pressure (BP) measurement using radial artery pressure waveform (RAPW), 2017

Undergraduate Projects

Design of a 3-bit Odd Parity Generator using Cadence, 2017

Real-time object tracker, 2016

Two-wheel self-balancing robot, 2016

Line tracking and maze solving robot, 2016

ASL alphabet deciphering device, 2015

Design and implementation of 12V 160mA AC to DC converter, 2015

Digital calendar and clock with the facility of sending an alarm notification to the user's phone, 2015

Design of a 100 kVA, 3 phase, 50 Hz distribution transformer, 2014

Design and implementation of a low-cost FSK demodulator, 2014

Digital weight measuring machine, 2014

Publications

Conference paper

Islam MR'*, Pavel SR, Tunaz SA, "Neurodegenerative Disease Classification Using Gait Signal Features and Random Forest Classifier." International Conference on Electrical Information and Communication Technology (EICT 2019). (DOI: 10.1109/EICT48899.2019.9068822)
Pavel SR', Islam MR'*, Siddiqee AM, "Fetal arrhythmia detection using fetal ECG signal." International Conference on Telecommunications and Photonics (ICTP 2019). (DOI:10.1109/ICTP48844.2019.9041789 )
Bhattacharjee A'*, Islam MR, Ullah MA, “Weather Independent Positioning of MS Using Power Slope Patterning Algorithm via One BS”. International Conference on Electrical, Computer and Communication Engineering (ECCE 2019). (DOI: 10.1109/ECACE.2019.8678919 ) [Download]

'First author

*Corresponding author

Research Experience

Development of a low cost EMG sensor, 2019

Motivation: For patients without health insurance, a prosthetic arm typically costs less than $5,000 for a purely cosmetic arm, up to $10,000 for a functional prosthetic arm that ends in a split hook, and up to $20,000-$100,000 or more for an advanced myoelectric arm, controlled by muscle movements, with a functioning artificial hand [1] .

Proposal: A low-cost EMG sensor can pave the way to create an affordable prosthetic arm.

Method: First, I took the EMG signal using 3 electrodes and then amplified the difference using INA 118 instrumentation amplifier. Then used an active bandpass filter to obtain the EMG signal of interest. Finally, I used a precision rectifier and a low pass filter to get the muscle movement signal.

Fig. 1: EMG Sensor Circuit

Result: The developed sensor costs $8.4 and can sense as low as 50mV sEMG signal and it's gain is adjustable.

Contribution: Though the research, I developed a cost-efficient gain adjustable EMG sensor.

Fetal arrhythmia detection, 2019

Motivation: 1% of fetuses are diagnosed with arrhythmia, and about 10% of these are considered potential sources of morbidity. Thus, up to 1 fetus in 100 needs their arrhythmia to be closely monitored and, if indicated, treated in-utero using antiarrhythmic therapy [1].

Proposal: Arrhythmia can be predicted by carefully analyzing the ECG signal. Thus, a classifier can be trained with significant features of ECG signal (both regular and arrhythmia recording) to predict fetus arrhythmia.

Method: 8-time series and 5 statistical features of 12 arrhythmia ECG recording and 14 normal ECG recordings were collected from the Physionet database. Then using Q, R, S, and T peaks of the ECG signal were extracted from each signal using a peak detection algorithm that I developed. Then a total of eight features were used to train support vector machine (SVM) classifier with RBF kernel. As the number of fetal ECG signal was pretty low, we used the leave-one-sample-out cross-validation method for performance analysis.

Fig. 2: Fetal ECG (Normal and Arrhythmia) analysis

Result: The proposed method had 83.4% fetal arrhythmia classification accuracy.

Contribution: I developed a unique ECG signal feature detecting algorithm to classify fetal arrhythmia more efficiently.

Neurodegenerative disease classification using gait signal, 2019

Motivation: Gait disturbances, as a consequence of neurodegenerative diseases (NDD), has become a severe health issue nowadays [2]. However, modern computer-aided systems in classifying NDDs (Huntington’s, Amyotrophic Lateral Sclerosis, and Parkinson’s) did not exceed a 90% accuracy level [2].

Proposal: Gait dynamics of an NDD affected patient become abnormal due to the decay of motor neurons. Thus, gait signal analysis can help in early diagnosis of NDDs. NDD classification accuracy can be improved by training a classifier with unique and significant statistical and time-series features of the gait signal.

Method: I have collected 60 patients (15 Parkinson’s disease affected, 17 Huntington’s disease affected, 13 Amyotrophic Lateral Sclerosis disease affected, and 15 healthy patients) gait signals, which was collected ultra-thin pressure-sensitive switches, and differentiated that signal. Then, I extracted the maxima and minima to calculate the stride, stance, and swing intervals. I used these linear features, approximate entropy, and some other statistical data to train multiple classifiers and compared their performances. The random forest classifier showed the best performance in classifying NDDs. I also used the kernel principal component analysis with the Gaussian kernel to prevent overfitting and improve accuracy.

Fig. 3: Gait signal processing for the proposed system

Result: The classifier showed 96% accuracy in classifying Amyotrophic Lateral Sclerosis disease, 95% accuracy in classifying Huntington's disease, and 93% accuracy in classifying Parkinson’s disease.

Contribution: The proposed method outperformed all of its peers.

Weather condition independent cell phone location detection system, 2018

Motivation: Motivation: The methods that use a cell phone's control channel (CC) signal strength for location detection, frequently get affected by the weather condition because signal strength deviates with the weather [3].

Proposal: A mobile station (MS) always keeps communicating with its nearest base station (BS) through CC for the information required for initializing a successful call at any instant of time. The power level of CC is always constant, even if a call is in progress. The control channel's level of power only shifts when the MS changes position. We can use the rate of change of the slope angle of the power data of the receiving signal with respect to distance to calculate the cell phone's location.

Method: To detect location, first, the CC signal strength is measured. Then, we calculate the distance of the phone from BS using the Friis transmission equation for five consecutive times. Next, we calculate the slope angle of the power data of the receiving signal with respect to distance, and we compare the data with the previously stored data. The pattern with the least percentage error under a BS resembles the road in which the phone is currently on. Provided that the correct road is successfully determined, the exact location of the phone is determined by correlating its current slope data with pre-stored database data.

Fig. 4: Transmitted signal power in dry and humid weather

Result: The system showed 93.4% accuracy and a standard deviation of only 1.38 meters.

Contribution: I delivered a weather condition independent, unique cell phone's location detection system. This can be used as an alternative to GPS.

[Bachelor Thesis] Blood pressure (BP) measurement using radial artery pressure waveform (RAPW), 2017

Motivation: Popular BP measuring methods like Auscultatory, Oscillometric, PTT method has certain drawbacks like white coat effect, artery damage due to cuff pressure, and the dependency on the mechanical properties of artery wall [4, 5, 6].

Proposal: RAPW is detectable in diseases like sclerosis and calcification of artery, obesity has less impact on wrist and radius’s flattened bony structure gives a firm base for the sensor. Thus, I proposed a cuffless BP measuring method using RAPW so that I could lessen standard measurement methods drawbacks stated above.

Method: First, I developed a RAPW recording device using a piezoelectric sensor and recorded RAPW with the sensor and actual blood pressure using ALP K2 from 133 volunteers. Then, to cope with the non-linearity of RAPW features and actual BP, I used six classifiers and fused their predictions using a weighted majority decision fusion algorithm and completed my BP measuring device. Then following the Association for the Advancement of Medical Instrumentation (AAMI) standard, I measured another 40 volunteers BP using the developed device, ALP K2, and Omron M400 and compared the results to analyze the performance of the proposed method.

Fig. 5: Extracted and filtered RAPW using proposed BP measuring device

Result: The device had 86% BP measurement accuracy while compared with standard BP measuring devices (ALP K2 and Omron M400).

Contribution: The study provides a method that measures BP without inserting pressure on the artery and also has little to no dependency on the artery wall’s mechanical property. Also, the proposed device can be integrated into equipment like a smartwatch and contribute to modern technologies like the wireless healthcare system.

Undergraduate Projects

Design of a 3-bit Odd Parity Generator using Cadence, 2017

I designed a circuit to generate parity bits for a 3-bit input number. I designed the circuit at the sub-section level and then combined the sections to attain the desired output. I started with a gate-level schematic design. Next, I completed top-level layout placement and routing and then transistor-level design and cell layout. Finally, I finished the chip layout by placing the ESD protection circuit and I/O pads. Also, I checked LVS, performed DRC, and prepared a GDS file of the design. [Report]

Real-time object tracker, 2016

I developed a program in Microsoft Visual Studio using the OpenCV library to track an object. In the program first, I converted the image into HSV format. Next, I adjusted the hue, saturation, and value to select the object and then eroded and dilated the object for more accurately tracking. Finally, I drew a contour and used moments method to get the coordinates for tracking.

Two-wheel self-balancing robot, 2016

I created a two-wheel balancing robot that can stay stable and move according to the user's command. I used a gyro sensor (MPU 6050) to detect angular displacement and then developed an adaptive PID algorithm that used that displacement reading to adjust the motor speed and rotation direction to balance the robot. I created a joystick to control the robot and used two Bluetooth devices (HC-05), one on the joystick (master), and the other on the robot (slave) to establish communication between the joystick and the robot.

Line tracking and maze solving robot, 2016

The robot that I developed was able to track a black line in white surface and vice-versa and uses an adaptive PID algorithm to control the motion. It was also able to solve a maze using both the right-hand and the left-hand rule. The user had the option to choose whether to use the right-hand or left-hand rule.

ASL alphabet deciphering device, 2015

I developed an American Sign Language (ASL) alphabet deciphering device using a pair of gloves with ten flex sensors attached on each finger, a gyro sensor (MPU 6050), an Arduino (NANO), and an LCD. ASL has a specific finger and hand position for each alphabet, and I used the flex sensors to read the finger position and the gyro sensor to detect hand elevation. Finally, I combined those data to identify the exact hand and finger position and mapped them to recognize the alphabet and displayed them in an LCD.

Design and implementation of 12V 160mA AC to DC converter, 2015

I designed an AC to DC converter using two capacitors ((2.2uF, 400V, X rated; 1000uF, 50V), two resistors (470k ohm, 1W; 100 ohms), a bridge rectifier and a 12V 1W zener diode. The X rated capacitor dropped the 230V AC voltage to 24V 160 mA. Then the bridge rectifier rectified the AC to DC. The 1000uF capacitor then filtered out the ripples, and finally, I used the zener diode to get 12V DC output.

Digital calendar and clock with the facility of sending an alarm notification to the user's phone, 2015

I used a PIC16F877A microcontroller and a 20X4 LCD to develop a digital clock and calendar. Also, it allowed the user to set an alarm that could be sent as a notification to the user's phone via a HC-05 Bluetooth module.

Design of a 100 kVA, 3 phase, 50 Hz distribution transformer, 2014

I designed a 100 kVA, 3 phase, 50 Hz, 11 KV/415 V, delta/star, oil-cooled distribution transformer with tapping at Tapping ±2.5%, ±5% on high voltage side. The magnetic loading was 1.7 Wb/m2 with 4.5 voltage/tern and CRGO steel lamination of 0.35mm as core material. The project included winding design, core design, and radiator and tank design. I also calculated magnetizing VA, No-load loss, Cu loss, stray loss, and load loss at 750C. I used AutoCAD to design the final layout.

Design and implementation of a low-cost FSK demodulator, 2014

I developed an FSK demodulator using a resistor-capacitor low pass filter, a NOR gate, and an operational amplifier. The low pass filter filtered the carrier wave but flipped the actual signal along the horizontal axis, the NOR gate flipped the output from low pass filter and also quantized the signal. Finally, the operational amplifier amplified the signal to get a better output. The low pass filter could be calibrated to cut out the carrier wave.

Digital weight measuring machine, 2014

I build a digital weight machine using a linear variable resistor, two steel slaps, spring, a pic microcontroller (PIC16F877A), and an LCD. I installed spring between two steel slaps and attached the nob of a variable resistor with the upper slop. Then, I recorded the digital output value for various weights. Finally, I mapped those values to the corresponding weights and displayed them in a 16X2 LCD to finish the weight measuring device.