Safety-related Studies on Non-Invasive Biomedical Signals and

Its Aptness Usage in Design of Fault Tolerant Multimodal Human

Health Monitoring System

L. V. R. Prasada Raju

, Madhubabu Anumukonda

and Shubhajit Roy Chowdhury

Center of VLSI and Embedded Systems Technologies, International Institute of Information Technology, Hyderabad, India

School of Computing and Electrical Engineering, Indian Institute of Technology (IIT), Mandi, India

1 RESEARCH PROBLEM

Biomedical Signals monitoring and diagnosis need

to be properly checked to avoid improper

medication. The parameters measured by the human

health monitoring systems(HHMS), specifically like

cardiac health monitoring systems (CHMS) that

measures, heart rate (HR), blood pressure (BP) and

other related vital parameters need to be effectively

diagnosed for illness and disease symptoms, to do

proper nursing.

The patient’s pre- and post-operation health

monitoring & diagnosis are particularly important

for effective nursing. These diagnostic systems

should present uninterruptedly the authentic health

monitoring vital, non-vital and desired parameters

even if any disturbances occur during critical

monitoring periods for a patient. In real time, the

systems should uninterruptedly present, authentic

vital health monitoring measurements. This involves

certain challenges like systemic failures and random

failures. These critical systems are known to

continuously replenish, but the mechanisms

involved are not adequate and largely unknown to

mitigate these failures.

Our main research problem is to address these

challenges and to improve in fault identification and

analysis mechanisms such that, the system should

uninterruptedly present, authentic vital health

monitoring measurements towards improvement in

accurate predictability of the illness with

pathological completeness. For this we preferred to

investigate these human health monitoring systems

(HHMS) by using safety related design architectures

and generation next technologies to counteract such

failures. This uses in design and development of

with or with-out redundancy approaches, run time

monitoring and implementing fault diagnostic

algorithms in designing fault tolerant systems.

The present available medical systems

specifically like non-invasive HHMS and related

point of care (POC) medical systems, using for

diagnosis, may not be having embedded with

approaches like redundancy, fault tolerant

mechanisms. This is due to various reasons like cost,

technological factors. However, if the mentioned

failures are effectively mitigated by applying these

mentioned approaches into this system, it improves

in authentic health monitoring and diagnosis of the

symptoms. Further, applying correlation techniques

on diagnostic data will provide relevant new insights

towards pathological completeness on identification

of the illness.

This design implementations and its safety

related correlation of data study is expected to

significantly improve our understanding on fault

tolerant diagnosis of the illness and may open a new

avenue in sensor fusion technologies advancements.

2 OUTLINE OF OBJECTIVES

This doctoral research project applies and evaluates

the safety-related design architectures usage in

specific cardiac health monitoring systems (CHMS),

or in a general safety critical patient monitoring

systems (PMS) and a low cost POC medical devices.

The main objective of this project is to

investigate the implementation of safety-related

design architectures like 2oo3 (two-out-of-three),

2oo2 (two-out-of-two), 1oo2 (One-out-of-two)

…etc, and its related safe computational approaches

based on the selection of monitoring vital parameter

of a human. These collected data evaluations and its

computational approaches results in applicability to

new areas in further advancing medical systems by

using-in methods & technologies like smart fusion

sensors, machine learning, data analytics, artificial

intelligence and internet of things (IOT).

We intend to collect data from multimodal

sensor fusion elements in a non-invasive mode by

sensing & analysing the biomedical signals from a

Prasada Raju, L., Anumukonda, M. and Roy Chowdhury, S.

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health Monitoring System.

In Doctoral Consortium (BIOSTEC 2019), pages 3-14

specific organ like Heart and to do the diagnostic

correlative computations inline to this safety related

approaches. Thus, this research studies helps in

improvements in mechanisms in addressing the

challenges and to achieve the following objectives:

 The designed HHMS ensures in providing

uninterruptable authentic vital data.

 The design and development of fault tolerable

system, ensures in reduction of undesired

spurious alarms for any single point failures.

 The designed fault tolerant multimodal human

health monitoring system enhances the capability

to provide inferences in predicting an illness, by

processing the multi-signal input data and

performing data analytics on the data collected

from these diverse sensors for pathological

completeness.

 Implemented safety-related health monitoring

systems, ensures the ability to reduce fault

alarms, for effective nursing by reducing alarm

fatigue.

 Enhances the health monitoring systems

capability in performing self-system diagnosis

for effective detection of system faults, and if

any fault detects, the system ensures safe-

degradation or fault-tolerable i.e. without loss of

system functionality.

 The design approach ensures, a non-invasive

patient monitoring systems (PMS) or wearable

diagnostic systems (WDS) can be made portable

with low cost having safe functionality.

In present systems, the availability of diagnostic

systems with functions like fault tolerant,

multimodal sensor fusion interfaces with portable or

wearable devices is limited.

We proposed to design and develop a portable &

wearable monitoring medical system prototype,

using safety -related architecture to collect the data

from primarily three diverse sensors, by detecting

electric potential, light and sound signals via non-

invasive. These diverse sensors used to detect bio-

medical signals to obtain data of electrocardiogram

(ECG), Photo-plethysmogram (PPG) and

Phonocardiogram (PCG).

As a strategy, we designed three independent &

diverse computing channels using ECG, PCG &

PPG sensors and a validated algorithm is used in

each channel to measure the heart rate. The

correlative diagnostic algorithms and related

procedures are developed to evaluate the

implemented safety-related architectures.

To predict the illness effectively, we decoded the

multi-modal data acquired by processing the various

biomedical signals and data analytics is performed

with correlative algorithms to unravel the illness. In

order to collect the data efficiently, a smart wearable

smart sensor-fusion suite is developed for

investigating the identified illness parameters like

cardiac related ailments.

The ultimate objective is to provide a mechanism

for human health monitoring systems using safety

related architectures for improvements in predicting

the illness effectively. These safety-related multi-

modal fault-tolerant studies will be combined with

the pathological studies, helps in development of

various gadgets from point-of-care devices to safety

critical patient monitoring systems, with aptness

usage on various organs health monitoring.

Additionally, this type of devices supports further

research in development of fusion-sensors, and on

usage of machine learning, deep learning algorithms.

These medical devices embedded with

histopathological studies further helps in predicting

the illness still more effectively.

3 STATE OF THE ART

Recent studies (Edworthy et al., 2018), (Bach et al.,

2018) (Hravnak et al., 2018) indicated that most of

the vital patient monitoring medical systems follows

IEC 60601 and related specific standards and

various design implementation techniques (Sheng et

al., 2013) (Alemzadeh et al., 2012) (Börcsök et

al.,2004) for better safety. It has been further

reported (Vescio et al., 2018) (Hsiao et al., 2017)

(Miao et al., 2017) (Selvaraj et al., 2008) that there

is a possibility to measure the same vital parameter

by monitoring diverse biomedical signals using

varied sensors. However, all the mentioned studies

have common limitation, that is, most of medical

related systems have been designed in 1oo1 (One-

out-of-one) i.e., one sensor measurers one or more

vital parameters but stops functionality and

generates alarm if any disturbance occurs as a safety.

In this present digital age, with advancements in

technology, medical systems are evolving from

analog to digital and simple to complex systems.

Over the last decade or so years, due to rapid growth

in technology and innovations, the systems are

shrinking in size and a lot of demand arise for new

methods to mitigate the challenges, for making the

medical systems resilient & reliable along with

safety improvements. In critical pre- & post

monitoring of the patients at ICU scenarios, if any

abnormal condition occurs, a system should provide

timely notification to operators, and performance

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies

should degrade gracefully rather than abruptly,

especially in life-sustaining medical systems. This

level of capabilities for a reliable safety system

requires mainly Availability & Reliability of

measurements in any given scenarios. Additionally,

it requires, reliable sensing with or without

redundancy structures, system modes of operation &

real-time response, self-monitoring built in test

functions, well-organized fault identification &

isolation of faults with safe degraded function, a

well-defined system negation mechanism for all

types of errors, to generate the related alarms, and

overall in implementation, the system should meet

the guidelines set by standard IEC 61508 and IEC

60601(Medical devices) for the best practice of

functional safety. Thus, having this type of system

requirements will reduce the false or fault or

insignificant alarms and enhance the safety feature

of the device for consistent vital sign measurements.

This proposed approach of fault-tolerant safety

related design strategy, such as 1oo2- (one-out-of-

two), 2oo2- (two-out-of-two), 2oo3- (two-out-of-

three) or MooN (M-out-of-N) logic for reliable

sensing and computation of cardiac vital parameter

like heart-rate measurement using field

programmable gate array (FPGA). This design uses

a hybrid sensor (fused sensors in combination of

electrocardiogram (ECG) or phonocardiogram

(PCG) sensors or photo-plethysmogram (PPG)

sensors) for reliable sensing. These sensors are

connected to each channel independently and a safe

voting function uses Pearson’s correlation

coefficient method for computing the correlation of

coefficient ‘r’ between the heartrate values measured

from any two independent diverse channels. The

resultant coefficient ‘r’ is used by a fault tolerant de-

gradable safe function & built-in test(BIT) function

for isolation of fault and enables a reliable heartrate

values for display at no fault condition. This safe

function uses negation error codes to generate the

related alarm for each significant detected fault and

log the results. The accuracy of the heartrate

measurements and the coefficient of correlative

results of vital sign measurements between two

channels are analyzed using Bland–Altman and

correlative plots (Bland et al., 1986). The recorded,

failure detected signal & heartrate measurements at

each channel output along with results of safe

function output are analyzed for effective

functioning of fault isolation and reduction of single

point of failures (SPOF).

This doctoral research project focuses primarily,

in reduction of SPOF & false or fault alarms by

using safety related architectures without using

redundancy along with a fault-tolerant safe degraded

algorithmic function for continuity and consistent

vital sign measurements. This design approach

contributes high availability of the system with

consistence in vital sign measurements. Availability

of the system, explains that, both channels should

detect a failure for the system to fail, or else, at least

one channel functioning is enough to operate the

system. It further explains how a same single cardiac

vital sign parameter is measured in parallel channels

by using diverse sensors and the results are

correlated along with safe voting function for

consistent vital sign measurements. Thus, evaluating

these results by additionally applying the

pathological data, helps in obtaining a state of the art

medical device for further research studies on

monitoring various organs for early detection of

various abnormalities.

4 METHODOLOGY

The experimental activities of this research project

are divided into mainly 3 main parts:

 First part dedicated to the evaluation of the

aptness to use the safety related architectures in

the medical systems to measure basic vital

parameters, and identification of suitable sensors

for sensing biomedical signals like sensing thru

electric potential, sound and light.

 Second part dedicated to the design and

realization of a three-independent channels with

sensor and FPGA based research prototype for

experimental studies.

 Final part dedicated to the calibration of the

research platform (sensor, configurable safety

architectures and FPGA circuits), by performing

lab & field trails in evaluating the vital parameter

like heart-rate, in addressing the mentioned

challenges as reduction of fault alarms, algorithm

limitations and uninterruptable functionality with

safe degradation.

4.1 Investigating of Existing

Safety-related Architectures

The standards IEC 61508 and IEC 60601(Medical

devices) have been referred for the best practice of

functional safety and now recognized all over the

world. In the IEC 61508 standard, several

architectures such as 1oo2-, 2oo2-, 1oo3-, and 2oo3-

are introduced for safety related systems. However,

the selection of the architecture depends on

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health

Monitoring System

application requirements such as safety, reliability

and availability levels. Here, some of the common

architectures are investigated and evaluates the

proposed architectures for suitability for medical

monitoring systems to address the mentioned

challenges.

1oo1(1-out-of-1): The system based on a single

channel architecture, shown in Figure 1, Figure 2

and is typically designed for low level safety

applications. Most of the point-to-care health

monitoring devices and low-cost patient monitoring

systems are based on this principle. In this system, if

any single failure occurs in sensor or at device, the

output represents a single switch shows wrong

results or loss of the safety function or a process will

dangerous shut down by raising the alarm.

Figure 1: Basic system 1oo1(1-out-of-1) architecture.

Figure 2: 1oo1-with self-checking pair safety architecture.

1oo2(1-out-of-2): The architecture 1oo2- has two

outputs (based on two 1oo1 channels) connected in

series as shown in Figure 3. Thus, the system

improves the performance of safety integrity of

safety system, since, any single contact is required to

dangerously shutdown the process by raising the

alarm. The disadvantage is it increases twice the

potential for nuisance failures. Thus, neither 1oo1-

nor 1oo2- architectures has any capability to reduce

the potential failures or alarms. However, a self-

diagnosis and switch over mechanism to 1oo1

possibilities can be explored.

Figure 3: 1oo2 (1-out-of-2) basic safety architecture.

2oo2(2-out-of-2): The 2oo2- has two outputs (based

on two 1oo1- channels) connected in parallel, as

shown in Figure 4 and Figure 5. The system

advantage is, if the system to go for dangerous

shutdown both channels should fail to raise the

alarm, else if, any single channel functionally

operational, the system operates normally. Thus,

implementing additional safety measures like

diverse sensors & diverse algorithms for

computations in each channel, this system has the

capability to reduce the potential failures or alarms.

Figure 4: 2oo2 (2-out-of-2) safety basic architecture.

Input

circuit

Logic

Computation

Input

circuit

Logic

Computation

Output

circuit

Device‐B

Device‐A

sensor

Figure 5: 2oo2(2-out-of-2) with dual self-checking pair

safety architecture.

2oo3(2-out-of-3) or Triple Module Redundancy

(TMR): The 2oo3- based system has three channels

with three outputs connected to a complex output

voting circuit as shown in Figure 6. The system

advantage is, if the system to go for dangerous

shutdown, any two channels should fail to raise the

alarm i.e., The system continuously operates even

when any single channel dangerously failed. These

2oo3 systems (TMR) or similarly 2oo4- quadruple

modular redundancy(QMR) is usually used in fault

tolerant applications, where the system must

continue functioning despite a failure—most often

in, life-support medical devices. Thus, implementing

additional safety measures like diverse sensors &

diverse algorithms for computations in each channel,

these systems has the capability to reduce the

potential failures or alarms.

Figure 6: 2oo3 (2-out-of-3) basic safety architecture.

Referring to (Börcsök et al.,2004) and based on

the above investigations, the 1oo2-, 2oo2- and 2oo3-

architectures are suitable due to safe & high

availability with diverse redundancy. The possibility

of fault-tolerant and continuity of system functional

operation can be achieved by using safe system

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies

degradation mechanism with proper log data

analytics. These architectures can be configured

based on the chosen health parameter monitoring in

medical systems.

4.2 Conceptual Design with 1oo2, 2oo2

& 2oo3- Architectures

The proposed approach for safe computation and

fault detection is based on composite fail safety

principle with 1oo2-, 2oo2- or 2oo3 architecture.

The design architecture consists of three micro-

processing devices (Device A, Device B, Device C

and Device O), in which safety critical functions are

partitioned and performed through three devices (A,

B & C) with diversity, while a fourth device (O) is

responsible for performing safety related functions.

Due to modularity in this system design approach,

the system provides an advantage of architecture

configurability based on feasibility & criticality of

the desired parameter to be measured. Thus, an

advantage of concurrent multi-signal, multi-

parameter, independent processing and possibilities

of effective correlative analysis of the values can be

explored by using this system. Based on the system

configuration such as 2oo3-, the three devices (A, B

& C) do the cross checks between themselves using

interconnection links and along with the fourth

device (O) they perform built-in-tests & negation

operations as per the defined safe state for each fault

detection.

In this system design, functional faults have been

attempted to be detected. At each fault detection, a

defined safe state is to generate an alarm and a

negation operation is performed i.e., a code or a

known defined value need to be generated and

recorded for the defined fault/error. This

fault/negation value indicates the type of

fault/failure.

Figure 7: 2oo2(2-out-of-2) Architecture design for HHMS.

The system uses an independent device units A,

B, C & O with stand-alone diverse computations,

and the implemented logic is programmed into each

Field Programmable Gate Array (FPGAs) devices

respectively, thus an agreement is required between

the three devices A, B and C to ensure composite

fail-safety. The feedback loop of failure detection

and appropriate execution of negation operation,

ensures the fail-safe operations of the system to the

incidence of data errors and their effectiveness in

reducing the false alarms, and ensures the systems

toward resiliency.

The proposed conceptual design architectures

2oo2- and 2oo3- are shown in Figure 7 and Figure 8.

Figure 8: 2oo3(2-out-of-3) Architecture design for HHMS.

4.3 System Hardware Description

4.3.1 System Overview

The system is designed in a modular approach and

each device unit consists of a Field Programmable

Gate Array(FPGA) and an analog front-end(AFE)

device. The flexibility in this approach is used to

configure the device in any one of the proposed

safety architecture for safe functional computations

in diverse method.

A system block diagram shown in Figure 9, has

three independent diverse channels of Device-A, -B

& -C and voting logic output in Device -O. The

entire system operates at 100 MHz and each channel

specifically processes, single data type of multi-

input signals like ECG signal is based on electrical

sensing data type, PPG signal is based on Optical

sensing data type and PCG signal is based on Sound

wave sensing data type. Each device -A, -B &-C is

having a FPGA XC6SLX45 & an analog front-

end(AFE) and the device – O uses an FPGA having

a controller with 16MB flash memory & 64MB

dynamic RAM, that processes the voting logic and

interacts with graphical user interface (GUI). The

CHMS GUI is a front-end software tool developed

in MATLAB, which controls & interacts with the

system via high speed serial interface. The CHMS

GUI is a multipurpose tool operated from a PC and

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health

Monitoring System

can be utilized for real time analysis of the received

sensor sample data from three different channels.

Figure 9: HHMS -System Block Diagram.

4.3.2 Device A– ECG Signal Computation

Channel Design

Overview: The Device -A, is an independent

channel, specifically to process the ECG signals.

This device consists of a Sensor, AFE & FPGA

module.

EPIC Sensor: The present system uses an electric

potential integrated circuit (EPIC) sensor chip,

which is a non-contact or dry contact sensor, to

sense the electro potentials on the surface of the skin

using a capacitive sensing technique. This advanced

sensor chip acts like a near perfect electro voltmeter

and eliminates the subject or patient shaving the hair

on the skin, usage of gels & other contact-enhancing

substances. This sensor is a perfect suitable sensor

for this type of portable/wearable CHMS

applications measuring the ECG, since, there is no

need for potentially dangerous low impedance

circuits across the heart. The availability of sensor

resolution, is as good as or better than conventional

wet electrodes.

Figure 10: (a) EPIC Sensor, Courtesy of photo from

Plessey Semiconductor sensor datasheet and (b) Internal

sensor schematic.

A single EPIC sensor, when placed on or near

the patient, an ECG signal can be recovered and is

capable of monitoring continuous ECG as well as

making more exacting clinical diagnostic

measurements. Using an array of EPIC sensors,

which are placed on the chest in a traditional 12-lead

configuration positions can recreate the signal

resolution as good as or better than the achieved

using traditional electrodes. These EPIC sensors can

be used to diagnose various heart diseases, which

can be measured and interpreted through ECG

recordings.

Two EPIC sensors are used, and the outputs are

connected to the pins of ADS1298R chip 1NN &

1NP in differential i.e., connected to ADC channel-

1, whereas the chip is configured in differential

mode of operation.

Status Ch‐1 Ch‐2 Ch‐12

24bit 24bit

24bit

DRDYn

CSn

SCLK

DOUT

(a)

(b)

Figure 11: (a) Circuit Schematic ADS1298R of Device-A

and (b) SPI Protocol for data sample reading.

Analog Front-End(AFE): The device uses an

analog front-end AFE1298R chip from Texas

Instruments and it can be used to acquire bio

potential such as 12-lead ECG signals or EEG

signals. The chip has low power eight-channel, 24-

bit-delta-sigma (ΔΣ) analog-to-digital converters

(ADCs) with built-in programmable gain low noise

amplifiers(PGAs) with simultaneous sampling

functionality.

The device is configured to obtain 1000 Samples

per second (SPS) and is interfaced to two electric

potential integrated circuit (EPIC) sensors in

differential mode. The AFE device is configured by

FPGA via SPI interface in Read Data Continuous

(RDATAC) mode and read data continuously in 8bit

burst cycle, such that all 216 bits per device [24

status bits + 24 bits per channel) X 8 channels] and

additional second AFE device interface in daisy

chain configuration for additional 4-channels are

read by an FPGA via SPI interface. A highlighted

schematic and SPI data timings shown in Figure 11.

ECG Data Pre-Processing using FPGA: An FPGA

is used to perform the desired computation of

measuring the heart rate by detection of R-peaks in

real time from the received ECG data collected from

a subject or a patient.

On power-up, the FPGA is configured by the

GUI for the desired functions to perform. The

designed FPGA consists of an ECG controller, SPI

controller and ECG function specific IP cores as

shown in Figure 12. Once configured by the GUI,

ECG controller coordinates by configuring the AFE

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies

ADS1298R and starts reading the ECG data via SPI.

Based on the desired function set by GUI, the ECG

controller uses specific function IP core, which is

enabled (where a specific algorithm is executed) in

the ECG processing Engine (EPE) and sends the

measured values to the Device –O for safe

correlation logic.

Figure 12: ECG-FPGA for function computation.

In this proposed system, a R- peak detection

function computation is performed on every 1000

samples per second (SPS). The sampled ECG data

continuously received for a defined period, which is

configured via GUI. The detected R-peak count

values are processed to measure the heartrate and

this parametric data is correlated using safe function

in Device –O.

R- Peaks Detection: To detect the R-peaks,

detection of QRS complex is most important. The

algorithm in (Chowdhury et al., 2012) has

successfully detected QRS with an accuracy 99.5%

using single channel ECG with entropy criteria. This

algorithm is implemented in the present work for

heartrate measurement.

4.3.3 Device B– PPG Signal Computation

Channel Design

Overview: The Device -B, provides an independent

channel, specifically to process the photo-

plethysmogram (PPG) signals. This device consists

of a Sensor (LED & opto-detector), AFE & FPGA

module.

PPG Sensor and Detector: PPG is an optically

obtained volumetric measurement of an organ. In

principle, the measurement of PPG is by

illuminating the skin and subcutaneous tissue with

radiation of a specific wavelength. This radiation

will come from a light emitting diode (LED). This

light when illuminated at the measuring point on a

patient or subject, is either absorbed, passed through,

or reflected-back from the capillaries below the skin.

A photodiode measures the light that is either

transmitted or reflected, depending on where it is

placed relative to the LED. The photodiode then

converts the measured light into an electrical signal.

In this system we used two LED sources of

specific wavelengths of light-- red, which is 660 nm,

and infrared, which is 940 nm. For photodetector we

used OP101 IC for detecting the transmitted light.

Analog Front-End (AFE): An AFE4490 chip is

used in device -B. This chip is a fully-integrated

analog front-end (AFE), which is ideally suited for

pulse-oximeter type of applications. This device is

suited for measuring heartrate and other blood

parameters. The device consists of a low-noise

receiver channel with a 22-bit analog-to-digital

converter (ADC), an LED transmit section, and

diagnostics for sensor and LED fault detection.

The device is configured, to process 200 Samples

per second (SPS) and is interfaced to FPGA. The

AFE device is configured by FPGA via SPI interface

and read data continuously for further processing in

FPGA. A highlighted schematic and sample data

timings shown in Figure 13.

PlethysmographSignal

PhotodiodeCurr entor

I‐VOutputPul ses

AmbientLevel

(DarkLevel)

IR‐LED‐‐ONSignal

RED‐LED‐‐ONSign al

(a)

(b)

Figure 13: (a) Circuit Schematic of AFE4490 of Device-B

and (b) Sample data read timings.

PPG Data Pre-Processing using FPGA: The PPG-

FPGA sub system is used to perform the desired

computation like measuring the heart rate by

detection of P-peaks in real time from the received

PPG data collected from a subject or a patient via a

AFE SPI interface.

On power-up, the FPGA is configured by the

GUI for the desired functions to perform. The

designed FPGA consists of a PPG controller, SPI

controller and PPG function specific IP cores as

shown in Figure 14. Once configured by the GUI,

PPG controller coordinates by configuring the

AFE4490 and starts reading the PPG data via SPI.

Based on the desired function set like heartrate by

GUI, the PPG controller enables the specific IP core

(where a specific algorithm is executed) in the PPG

processing Engine (PPE) and sends the measured

values to the Device –O for safe correlation logic.

A P- peak detection function computation is

performed on every 200 SPS. The sampled PPG data

continuously received for a defined period, which is

configured via GUI. The detected R-peak count

values are processed to measure the heartrate and

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health

Monitoring System

this parametric data is correlated using safe function

in Device –O. The functional block diagram of PPG-

FPGA implemented logic is shown in Figure 14.

PPG

IPCore‐1

PPG

Controller

SPIMaster

con troller

PPG‐FPG

PPG

IPCore‐n

PPG

IPCore‐2

PPGProcessingEngine(PPE)

AFE

4490

BusInterface

Accelerometer

Figure 14: PPG-FPGA for function computation.

P- Peaks Detection: Each cardiac cycle sends a

pressure wave through the cardiovascular system.

This pressure wave causes the blood vessels to

expand and contract, which gives the PPG a

characteristic waveform. Since the period of the

PPG waveform repeats with each cardiac cycle, it

too can be used to calculate a patient's heart rate.

The algorithm in (Paradkar et al., 2015) has

successfully detected the pulse rate with an accuracy

99.39% using PPG with entropy measures. This

algorithm is implemented in the present work for

heartrate measurement.

4.3.4 Device C– PCG Signal Computation

Channel Design

Overview: The Device -C, is an independent

channel, specifically to process the

Phonocardiogram (PCG) signals. This device

consists of a Sensor (Digital MEMS microphone)

and an FPGA module.

PCG Sensor: The device uses four individual

Digital MEMS microphone MPDT01 sensors and

each sensor to cover the four heart valves (Aortic,

Tricuspid, Mitral, and Pulmonary). The criteria of

selecting this sensor is due to low noise,

miniaturised device, low cost and simple interface to

process the binary sound signal. The digital MEMS

microphone has digital output type of Pulse Density

Modulation (PDM) format, with a high sensitivity of

-26 dBFS, signal to noise ratio (SNR) of 62.6 dB

and has a flat frequency response of 20 Hz to 15

KHz. The set of sensors is placed on human chest

for capturing the heart sounds at four heart valves

simultaneously with good quality due to its high

sensitivity and flat frequency response.

The digital MEMS chip has inbuilt signal

preconditioning, filtering and signal enhancement

module that provide digital PDM output for further

processing in an FPGA. The sensors are interfaced

serially to FPGA and captures the data at 1 MHz

inline to the data read timings specified in the

MPDT01 datasheet. The highlighted interface circuit

schematic and its data read timings is shown in

Figure 15.

FPGA

Digital

MEMS

3.3V

GND

DATA

CLK

C1‐

100nF

C2‐

10uF

Out

HiGH‐Z

Cloc k

PDM‐R

PDM‐L

(

)

(

)

Figure 15: (a) Device-C Circuit and (b) Data read timings.

PCG Data Pre-Processing using FPGA: The PCG-

FPGA, is used to perform the desired computation

like measuring the heart rate by detection of S-peaks

in real time from the received PCG data.

On power-up, the FPGA is configured by the

GUI for the desired functions to perform. The

designed FPGA consists of a PCG controller, which

enables the PCG processing engine (PCE) to capture

& process the received digital data from the MEMS

as shown in Figure 16. Based on the desired function

set like heartrate by GUI, the PCG controller enables

the specific IP core (where a specific algorithm is

executed) in the PCG processing Engine (PPE) and

sends the measured values to the Device –O for safe

correlation logic.

A S- peak detection function computation is

performed on a 2000 SPS. The converted PDM to

PCM sampled PCG data is continuously received for

a defined period, which is configured via GUI. The

detected S-peak count values are processed to

measure the heartrate and this parametric data is

correlated using safe function in Device –O.

PCG

IPCore‐1

PCG

Controller

PCG‐FPGA

PCG

IPCore‐n

PCG

IPCore‐2

PCGProcessingEngine(PPE)

PCGSensor‐2

BusInterface

Accelerometer

PCGSensor‐3

PCGSensor‐4

PCGSensor‐1

Figure 16: PCG-FPGA for function computation.

S- Peaks Detection: At each cardiac cycle major

heart sounds S1 and S2 are produced due to closing

and opening of the heart valves. The heart rate is

directly proportional to the number of S1 peaks per

minute, so the S-peaks detection is considered for

the heart rate monitoring.

The algorithm in (Anumukonda et al., 2015) has

successfully detected the S1 peaks using spectral

analysis of PCG data and measure the number of S1

peaks per minute. This algorithm is implemented in

the present work for heartrate measurement.

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies

4.3.5 Device O -Safe-Selection Logic

Implementation using Karl Pearson’s

Correlation Coefficient Method

To measure the magnitude of the relationship

between two variables, (Vescio et al., 2018) (Hsiao

et al., 2017) (Miao et al., 2017) (Selvaraj et al.,

2008) we used Karl Pearson coefficient method for

calculating the correlation coefficient ‘r’.

r = (∑XY)/√(∑X^2 × ∑Y^2 ) (1)

X = x - x

(2)

Y= y - y ̅ (3)

where x

are "mean"

The ‘r’ values always lies in between -1< r < 1,

and the interpretation of ‘r’ is as below:

When r = 1, there is perfect +ve relationship

between the variables, when r = -1, there is a perfect

-ve relationship between the variables and when r =

0, there is no relationship between the variables.

And, if the correlation is +1 or –1, it signifies

that there is a high degree of correlation. (+ve or –

ve) between the two variables. So, if r is near to zero

i.e., 0.1, -0.1, (or) 0.2 there is less correlation.

As a rule-of-thumb for interpreting the

coefficient of correlation value ‘r’, the below Table

1, shows the standard interpretation of relationship

between two variables.

The measured two variables A, B from two

diverse signals are correlated and calculated

correlation coefficient ‘r

’ in a set of pre-

determined continuous samples.

Table 1: Rule of Thumb for Interpretation of the

correlation coefficient 'r' values.

Coefficient of Correlation ‘r’ Interpretation

0.90 to 1.00 (-0.90 to -1.00) Very high +ve or -ve

0.70 to 0.90 (-0.70 to -0.90) High +ve or -ve

0.50 to 0.70 (-0.50 to -0.70) Moderate +ve or -ve

0.30 to 0.50 (-0.30 to -0.50) Low +ve or -ve

0.00 to 0.30 (-0.00 to -0.30) Negligible

4.4 Experimental Set-up and

Evaluation Framework

4.4.1 Experimental Set-up

The experiment was carried out in steps.

 Configuration of the CHMS using GUI

 Preparation of the subject for Experiment

 Sensors placement on a wearable-suite

 Simulation Experiment and Evaluation

 Clinical Experiment

A MATLAB tool based Graphical user Interface

(GUI) is developed for configuring the CHMS as

shown in Figure 17. The tool provides 1. CHMS

control panel – which provides options of real-time

algorithm model development & evaluation, setting

simulation mode and options for analysis, 2. Basic

patient information, 3. Device configuration panel,

4. Sensor selection for correlation, 5. Parameter

selection, 6. Safety architecture selection, 7.

Diagnosis artefact selection, 8. Signal capture

controls, 9. Raw signal to Processed signal capture,

10. Duration record, 11. Report generation.

5 6

Figure 17: MATLAB based GUI tool for system

configuration and Data sample collection.

Multimodal sensors are placed on a wearable

suite and shall be wore by the subject as shown in

Figure 18, which will be connected to the portable

monitoring device prototype.

(a)

(b)

Figure 18: (a) Experimental set-up and (b) CHMS - PPG,

PCG & ECG Sensors placement.

4.4.2 Application Protocol

The system has been preliminary validated with the

data collected of 5- subjects with various age groups

conforming to the declaration of Helsinki. The

subjects aged between 15 years to 55 years, were

available for measurement and testing after taking

informed consent of which 5 normal healthy patient

data have been used.

Recommendations before taking measurements:

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health

Monitoring System

The advisories are roughly the same as those

used for blood pressure(BP) measurements. Most of

the physical or psychological factors can influence

the assessment of heart rate. We mainly followed,

1. Heart rate can be influenced by exercise,

alcohol, nicotine, and coffee. These should be

avoided in the hours preceding measurement. 2.

Subject should be preferably seated in a chair with

legs uncrossed, with comfortable room temperature,

and no noise, before taking readings. 3. The subject

should refrain from talking during taking readings,

and at least 5 minutes should elapse before the first

reading is taken. 4. Hemodynamic variable need to

be assessed in patients under medical investigation

for hypertension or cardiac disease before the heart

rate is measured. 5. Avoid taking reading from the

subjects, if they are receiving pharmacological

therapy, and the doctor should be aware that many

cardiovascular drugs can either decrease or increase

the heart rate.

The CHMS is configured via GUI, by enabling

the ECG & PCG sensors in 2oo2 mode, and the

measured heartrate readings are recorded for a

period of 1-hour duration for each subject. The

recorded is evaluated in real time. As per (Wikipedia

Contributors, 2019), calculated trigger levels for

max

(MaxHR), HR

min

(MinHR), Targeted HR

(THR)-Upper limit & Lower limit levels and

Adjustable (ADJ) Upper limit (UL) & lower limits

(LL) at 5% & 10% are set for the defined criticality,

to generate the alarms. The safe logic

implementation does data correlations and generates

alarm for variances at ‘r

<0.5’ and evaluates the

faults in each channel independently. The authentic

data with no fault at each channel is enabled for

output. With 2oo2-logic, if both channels faults, then

the alarm is enabled with no data display, else if

fault noticed only at single channel, then the system

degrades to safe working 1oo1 channel and continue

to operate un-interruptedly by logging the record.

Thus, the alarm and data computations values are

generated and recorded. These resultant values are

displayed from each channel, ECG-1oo1 & PCG-

1oo1 along with 2oo2 resultant mode of operation

for investigations.

4.4.3 Results and Discussion

The processed data and fault alarm signals are

captured from channels ECG-1oo1, PCG-1oo1,

configured 2oo2- outputs are presented in Figure 19,

Figure 20, and Figure 21 for a single subject. As it is

analysed that, the reading from all the 5 subjects are

almost similar, and in Figure 22, presents the 2oo2-

results, which shows the reduction in alarms.

In Figure 19, presented the data captured from

ECG 1oo1 channel with HR measured data and

inverted alarm signal for a duration of 1 hour.

Analysed and captured alarm is with respect to the

set ADJ-UL & LL are shown. Similarly, for PCG

1oo1 channel shown in Figure 20.

Figure 19: 1oo1 -ECG HR measurement Vs Alarm Signal.

Figure 20: 1oo1 - PCG HR measurement Vs Alarm Signal.

The correlation coefficient ‘r

’ values and the

ECG & PCG HR measurements are captured in

Figure 21 and analysed for deviances at each

channel.

Figure 21: ECG & PCG HR measurements, r

Correlation Coefficient values Vs r

Signal.

The comparative and safe logic degradation logic

implementation analysis using alarm signals are

captured in Figure 22. Which shows a significant

reduction of alarms with consistency HR data record

in 2oo2 design architecture.

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies

Figure 22: r

Signal and Alarm outputs (ECG -1oo1,

PCG -1oo1 and 2oo2 configured Alarm signal).

5 EXPECTED OUTCOME

The proposed aim of this research project is to

provide aptness usage of safety related architectures

in design of Non-invasive human health monitoring

systems (HHMS), Point-of-Care (PoC) Medical

systems with fault-tolerable and portable to

multimodal sensors. This is done in mitigating the

teething challenges and achieving the outline

objectives mentioned in section-2 and to provide a

platform for further research scalability with scope

to improvement in predicting the diseases at early

stages.

We expect to deliver the design evaluation

results on a selected vital parameter ‘heart-rate

(HR)’, along with the research platform designed

with safety-related architectures, compatibility with

multimodal sensor interfaces and fault-tolerable

function.

However, to achieve the outline objectives,

primarily, we designed a research platform for

experimentation and ensured to perform evaluations

at hospital environments. Secondly, we choose to

evaluate the 2oo2 (two out of two) architecture and

presented the results in section 4.4. Thus, a safe

degradation logic and fault tolerable functions are

evaluated, and the results shows reduction in alarms

and additionally shown an improvements in

presenting uninterruptable authentic HR data record.

Further research studies need to be performed in

evaluating safety related architectures such as 2oo3

(two-out-of-three), 2oo2 (two-out-of-two), 1oo2

(One-out-of-two), along with various combination of

sensors like ECG verses PPG, ECG verses PCG,

PPG verses PCG and combination three sensors.

Applying correlation techniques on diagnostic data a

planned systemic research need to be performed in

achieving the objectives and providing relevant new

insights towards pathological completeness on early

identification of the illness or diseases.

Few of the criteria considered in building this

HHMS research platform, is to applicability of these

design concepts to wide range of medical systems

from safety-critical Intensive Care Unit (ICU) based

medical monitoring systems to complex-automated

patient operated robotic surgery medical systems, to,

as low as, simple real time monitored point-of-care

(PoC) medical gadgets or wearable medical

monitoring devices.

We used the validated algorithms for measuring

heart-rate and selected the available sensors like

EPIC (Plessey Semiconductors Inc), Digital MEMS

microphone (ST Microelectronics), Light Emitting

Diodes (LEDs) & Opto-detectors. We designed and

developed the safety mechanisms without

redundancy. As part of the project, we designed a

MATLAB based GUI for data analytics for results

evaluation during experiments. The preliminary

results during lab & field trails are encouraging. We

expect to, following the success of this task, propose

a low-cost wearable medical device by meeting the

criteria for commercial usage. We will then bring

this configurable prototype platform to real factory

conditions due to collaboration with our partner

companies interested in this concept.

6 STAGE OF THE RESEARCH

The uninterruptable monitoring, reduction of false &

fault alarms, and authentic display of measured vital

parameters is the need of the hour for any critical

patient monitoring systems during system under

operation even if any fault occurs. At critical times

of ICU patient monitoring or at general patient

check-ups, the predictive detection of illness is more

important for critical nursing.

The current research project is related to

evaluation of safety design architectures with and

with-out redundancies, and its aptness usage in

design of fault-tolerant non-invasive medical

system. A detailed literature has been reviewed in

the fields of medical systems like patient monitoring

systems, point-of-care medical devices and related

available functional system specifications for

processing of non-invasive biomedical signal

processing preceded the current research in-order to

approach this challenges in the most appropriate

way. At this point, we believe that it is possible to

0.5

0 500 1000 1500 2000 2500 3000 3500 4000

2oo2

AlarmSignal

Time

2oo2AlarmOutput

0.5

0 500 1000 1500 2000 2500 3000 3500 4000

PCG‐1oo1

AlarmSignal

Time

PCG‐1oo1Alarm

0.5

0 500 1000 1500 2000 2500 3000 3500 4000

ECG‐1oo1

AlarmSignal

Time

ECG‐1oo1Alarm

0.5

0 500 1000 1500 2000 2500 3000 3500 4000

AB

Signal(>0.5 )

Time

AB

Signal (>0.5)

Safety-related Studies on Non-Invasive Biomedical Signals and Its Aptness Usage in Design of Fault Tolerant Multimodal Human Health

Monitoring System

improve the medical systems resiliency by applying

the safety- related design approaches and we build

prototype of a multimodal wearable sensor suite

along with FPGA based medical diagnostic system.

The preliminary experiments have been

conducted, using this prototype with 2oo2 safety

architecture configuration with ECG & PCG sensors

and measured the HR to understand the cardiac

physiology. The initial recordings and its analysis

results are quite encouraging with uninterruptable

and authentic HR data with reduced alarms.

Further, with this experimental set-up, we extend

to configure the CHMS system with 1oo2-, 2oo2 and

2oo3 settings and perform research experiments to

collect the data. This serves us to study further on

the correlative analysis of non-invasive biomedical

signals on measured parameters in combination of

pathological data to infer predictable & early

detection of unknown illness along with mitigating

the desired challenges.

ACKNOWLEDGEMENTS

The author would like to thanks the Faculty of

Center for VLSI and Embedded Systems

Technologies (CVEST), International Institute of

Information Technology (IIIT) - Hyderabad and

School of Computing and Electrical Engineering

(SCEE), Indian Institute of Technology (IIT)-Mandi

for their valuable support during this research.

REFERENCES

IEC, I., 1998. 61508 functional safety of electrical

/electronic/programmable electronic safety-related

systems. International Electrotechnical Commission.

Börcsök, J., 2004. Electronic safety systems: hardware

concepts, models, and calculations. Hüthig.

Edworthy, J.R., McNeer, R.R., Bennett, C.L., Dudaryk,

R., McDougall, S.J., Schlesinger, J.J., Bolton, M.L.,

Edworthy, J.D.R., Özcan, E., Boyd, A.D. and Reid,

S.K., 2018. Getting Better Hospital Alarm Sounds Into

a Global Standard. Ergonomics in Design, 26(4), pp.4-

13.

Bach, T.A., Berglund, L.M. and Turk, E., 2018. Managing

alarm systems for quality and safety in the hospital

setting. BMJ Open Qual, 7(3), p.e000202.

Hravnak, M., Pellathy, T., Chen, L., Dubrawski, A.,

Wertz, A., Clermont, G. and Pinsky, M.R., 2018. A

call to alarms: Current state and future directions in

the battle against alarm fatigue. Journal of

Electrocardiology.

Sheng, H., Schwarz, M. and Borcsok, J., 2013, October.

Modeling a safety-related system for continuous non-

invasive blood pressure monitoring. In Biomedical

Circuits and Systems Conference (BioCAS), 2013

IEEE (pp. 254-257). IEEE.

Alemzadeh, H., Di Martino, C., Jin, Z., Kalbarczyk, Z.T.

and Iyer, R.K., 2012, June. Towards resiliency in

embedded medical monitoring devices. In Dependable

Systems and Networks Workshops (DSN-W), 2012

IEEE/IFIP 42nd International Conference on (pp. 1-

6). IEEE.

Chowdhury, S.R., 2012. Field programmable gate array

based fuzzy neural signal processing system for

differential diagnosis of QRS complex tachycardia and

tachyarrhythmia in noisy ECG signals. Journal of

medical systems, 36(2), pp.765-775.

Paradkar, N.S. and Chowdhury, S.R., 2015. Fuzzy entropy

based detection of tachycardia and estimation of pulse

rate through fingertip photoplethysmography. Journal

of Medical and Bioengineering Vol, 4(1).

Anumukonda, M., Ramasahayam, S., Raju, L.P. and

Chowdhury, S.R., 2015, December. Detection of

cardio auscultation using MEMS microphone. In

Sensing Technology (ICST), 2015 9th International

Conference on (pp. 173-177). IEEE.

Bland, J.M. and Altman, D., 1986. Statistical methods for

assessing agreement between two methods of clinical

measurement. The lancet, 327(8476), pp.307-310.

Wikipedia Contributors (2019) Heart rate, Wikipedia.

Wikimedia Foundation. Available at:

https://en.wikipedia.org/wiki/Heart_rate.

Vescio, B., Salsone, M., Gambardella, A. and Quattrone,

A., 2018. Comparison between Electrocardiographic

and Earlobe Pulse Photoplethysmographic Detection

for Evaluating Heart Rate Variability in Healthy

Subjects in Short-and Long-Term Recordings.

Sensors, 18(3), p.844.

Miao, F., Fu, N., Zhang, Y.T., Ding, X.R., Hong, X., He,

Q. and Li, Y., 2017. A novel continuous blood

pressure estimation approach based on data mining

techniques. IEEE J. Biomed. Health Inform, 21,

pp.1730-1740.

Selvaraj, N., Jaryal, A., Santhosh, J., Deepak, K.K. and

Anand, S., 2008. Assessment of heart rate variability

derived from finger-tip photoplethysmography as

compared to electrocardiography. Journal of Medical

Engineering & Technology, 32(6), pp.479-484.

Hsiao, C.C., Horng, J., Lee, R.G. and Lin, R., 2017,

October. Design and implementation of auscultation

blood pressure measurement using vascular transit

time and physiological parameters. In Systems, Man,

and Cybernetics (SMC), 2017 IEEE International

Conference on (pp. 2996-3001). IEEE.

DCBIOSTEC 2019 - Doctoral Consortium on Biomedical Engineering Systems and Technologies