Dynamic Programmable Clock

Frequency Using Machine Learning

Algorithms to Reduce Power
Consumption in Wearables

A. Ajin Roch, S. Karthik, and R. Arthi

Abstract In today’s world, wearable devices have become a key device in moni-
toring all internal activities of human being with the help of sensors. Humans are
provided with a wristband device fabricated with various sensors to detect overall
body condition in every day-to-day activity. Most of the wearable devices can be
used to monitor the beats per minute (BPM) by heart rate sensor (pressure sensor),
sleep and body movement activity by three-axis accelerometer sensor, the respi-
ratory rate by biosensors and electrocardiogram (ECG) by sensors (electrodes). It
is well known that soon in the near future, all sensors related to human activity
and health monitoring play a major role in human beings; especially, for old age
people, the wearable devices with well-packed sensor will soon be reaching out the
market. The most wearable devices are able to provide almost all sensors in a single
package, but fail to work if the power goes off shortly. The proposed model makes an
attempt to reduce the power consumption in the wearable device by introducing the
programmable clock frequency with machine learning concept. To reduce the power
consumption in wearable devices, a programmable clock frequency of 1 GHz has
been designed and monitored on a monthly basis. The proposed work uses different
frequencies based on different activities of wearable device with generated data set
of machine learning algorithms and is analysed using PYNQ boards. The proposed
work concludes based on the multilinear-based regression model to check the power
consumed for the given parameters observed that step count consumes 96% power
during the complete analysis.

Keywords Biosensors · PYNQ · Step count · Heart rate · BPM · Sleep activity

A. Ajin Roch (B) · R. Arthi

Electronics and Communication Engineering, SRM Institute of Science and Technology,
Ramapuram Campus, Chennai 600089, India
e-mail: [email protected]
R. Arthi
e-mail: [email protected]
S. Karthik
Electronics and Communication Engineering, SRM Institute of Science and Technology,
Vadapalani Campus, Chennai 600026, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2021 337

A. K. Bhoi et al. (eds.), Bio-inspired Neurocomputing, Studies in Computational
Intelligence 903, https://doi.org/10.1007/978-981-15-5495-7_19
338 A. Ajin Roch et al.

1 Introduction

Human has no more space for paper works, paper bills and prescriptions. As every-
thing is getting digital form, the E-health application is grabbing the topic in front.
E-health field is a multidisciplinary area to get inspired by distinct science and tech-
nology. The E-health application is a multipurpose, easily configurable and easy
to step up for various needs. Loneliness is a condition in old-age people that can
be the reason for dementia, traumatic and situationally induced negative health in
a desperate situation. Ubiquitous health care is turning out to be reality where the
sensors that meet customized human services with minimal effort and productive
patient observing are more popular are called as Next Generation E-health (NGeN)
which assist individuals with doing their everyday life exercises paying little heed
to any medical issue exclusively with no assistance. E-health services are normally
built using the components which have reusable services [1]. In most case, the home
environment has been considered for old-age people; obstructiveness, robustness
and cost are of much relevance. Smart objects need to be designed and deployed at
homes. Wi-Fi technology and practical sensors are age friendly that are effective for
practical application. Real-time monitoring systems were used to record, measure
and monitor the activities of the patient. Patient empowerment is most important,
thereby becoming active participants, and outcomes are better. A skilled practice
proctor has been advanced to envisage the fitness condition of patients centred on
everyday habitat measurement of palpitate, breathing level, in case of acute illness.
Wearable motion sensors that are suitable for practical and effective operation have
been used. The sensors are connected to cloud that has access flexibility and make
the personalization easy [2].
The expert system is a practical method that helps in developing healthy diary
supervising the system. The expert system plays an important role in decreasing the
time and cost of interpreting psychological facts generated by observing patients
under a regulated clinical setting. The Wi-Fi compliant sensors used for practical
operation are high power consuming and have low battery life. The major difficulty
in the expert system is the missing of data and execution of time-oriented processing
of the data. The other terms, classified as Board sensors and Fuzzy Pattern Tree
(FPT), do not require training but have an advantage of getting domain knowledge
from clinical partners. The other classifier sensors needed to provide a labelled set
of training data, the neural network and support vector machine.
Ubiquitous personalized e-health service architecture consists of sensors, security
mechanisms and IoT where the generated data are managed using cloud. In case of
real time, there occurs a delay due to transferring of data which becomes improper to
the cloud. Therefore, a smart gateway technique called as fog computing would be
preferred rather than cloud computing data centre. Fog computing solves the greater
latency problem, location awareness, reliability and relocating the data to the finest
location [3]. The smart house scheme uses sensors to monitor the activity of the
patient based on behavioural patterns and is reported in paper [4].
Dynamic Programmable Clock Frequency … 339

Mixture models are used to develop the probabilistic model of behavioural patterns
using the log back of events. Day-to-day activities can be differentiated by logging
to the data collected from an accelerometer, cardio tachometer and smart sensors.
The impact of collapses can be recorded and differentiated from regular day-to-day
events. Smart sensors borne on the body of an elderly people if, unintentional falls
can be noticed. Telehealth is the use of digital information through technologies in
the field of communication to access health care. It has provided medical education
over a long distance making use of advance telemedicine [5]. It has become possible
for physicians to treat patients whenever needed and wherever the patient is, by using
a SMART watch by sending mail. To enhance clinical decision support, new research
has been carried out to integrate cybercare along with analysis [6]. The sleep problem
has become common among older adults that results in various health issues. The
wearable activity sensor has been used to analyse the diurnal activity rhythm and
sleep patterns in their normal life [7].
During the long-lasting chronicle of the exercises of individuals that decline in
degrees of activity, relentlessness and force of the activity cadence, with uplifted
breaking down of musicality and daytime lack of involvement. The integrated activity
sensor is a wrist-worn device that collects the user’s activity data similar to traditional
actigraphy [8]. There are a lot of activities to get monitored using sensors, the data
from unremarkable, lightweight, and an able, well-organized wearable sensors are
used to record the physical activity. An incremental diagnosis method (IDM) gener-
ates data under supervised training to determine a medical illness with minimum
wearable sensors by using the dynamically adjusted sensor in their natural environ-
ment. The progress of less load functional sensors resolves to be comfortable for
monitoring the range of activities of inhabitants [9].
The wearable purchaser gadgets like inclination trackers, smart glasses and
SMART watches are considered not exclusively to be a glossy item, yet will be
in 100 s of million later on. Formal and informal survey predicts that the increase in
usage of wearable smart devices in the future will result in the price of these devices
as it has a wide application in society. E-health has been aimed towards home use
for elderly people by constantly monitoring them in the home environment without
disturbing them. At the same time, the system is sensitive in case of security and
privacy policy [10].
A loneliness monitoring system for old-age people has been made into experi-
mental values for a period of eight months [11] in which longitudinal linear mixed-
effects regression and out-of-sample cross-validation have been studied to prove
the accuracy of the system. Embedded-based health monitoring system for old-age
people to detect early changes in health has been made possible by continuous moni-
toring, for which 22 features have been analysed [12] from in-home sensor data with
k-nearest neighbour classifier along with fuzzy pattern tree.
In designing architectural model for monitoring patients’ health and to pass on
information about patient to the doctor, two-layer architecture model has been used in
which data and communication layer used with REST Web services [13] to develop
home-based telecommunication to effectively pass the information about the chronic
diseases of the patients. Data set is the key in machine learning that has been used
340 A. Ajin Roch et al.

for blood pressure (BP) with more than 16 million BP measurements; so in day-
to-day activity, supervised algorithms [14] shall be implemented over BP sensors
with real-time home-based monitoring system, through which cardiovascular risk
activities can be reduced. Particle swarm optimization technique for monitoring the
patient’s [15] state with the past and present measurement of various parameters has
been studied. Support vector machine for embedded-based applications [16] shall be
effectively utilized by the people with real-time world in the form of smart home.
The major problem faced by smart devices will be the power consumption, if
all the sensors are combined in a single package for health monitoring. To improve
the power consumption, if all the sensors are consolidated in a solitary bundle for
wellbeing observing, diverse clock frequencies for various sensors, with the inde-
pendent informational collection utilizing AI calculations, have been actualized in
the proposed work. In the proposed model, multilinear regression algorithm provides
better efficient output under supervised learning.

2 System Design

The proposed system architecture has been designed to perform different

programmable-based clockings as shown in Fig. 1. The chosen application based on
the wearable model with different clock frequencies has been applied with machine
learning algorithms in the ARM processor. The Python library has been used to
implement the coding with PYNQ-based field-programmable gate array kit through
which the power consumption related-problems for wearables and SMART watches
are being rectified.
In the proposed model, different parameters like BPM, step counts, calories
burned, temperature and sleep disorder has been considered with real-time values
using SMART watches and wearables. The chosen parameter has been monitored
using five different devices as shown in Fig. 2.
The nature of our proposed work depicts that the database was taken for a month
with three different wearables and two different SMART watches. Different wear-
ables and the SMART watch configuration are shown in Tables 1 and 2, respectively.
The tracking data set has been studied for a period of 30 days; though MI band uses

Fig. 1 Architectural view of the model

Dynamic Programmable Clock Frequency … 341

Fig. 2 Devices used for

analysis

different colours, it could be able to withstand more than 7 days with full sensors
being active for the entire period of testing when compared with other wearables.
It has been noticed here that the wearable technology uses different frequencies
ranging from 2400 to 2483 MHz, so a separate study was conducted with these three
devices for a period of 30 days and is shown in Table 2.
Two different SMART watches with frequency range of 1 GHz have been
compared for a period of 30 days so that the original real-time data shall be provided
for the proposed model. While comparing the data among the SMART watch of
different brands, power management is literally reduced since these devices have
baseband processor, DRAM, Wi-Fi, Bluetooth, RF components, NFC and MCU.

3 Results and Discussion

As proposed by the framework plan, various clocking with various sensors are
trained under Supervised Machine Learning Model. The comparison of average
power observed for a period of 30 days is as shown in Table 3.
The observed results clearly show that the major factor dealt with the wearable
and SMART watches consumes the less amount of power. From the data, it is clear
that watch product code series 3 stands for 1 day 13 h and 24 s. In the future, if all the
sensors are embedded in smart devices, the power may degrade within a day. In case
of old-age people, they will not charge regularly and need continuous monitoring of
their overall internal behaviours. So, utilizing the power effectively in this scenario
plays a major role. For the same reason, a novel table has been generated for different
 342

Table 1 Wearable configuration

Brand Code No. Battery life (days) Tracked data Bluetooth version Voltage (V) Max input current (mA) Battery (mAh)
Fitbit 85,913,025 6 Natural action (activity) 4 3.7 200 90
Kilo calorie (calories
burned), distance
(aloofness)
Sleep (slumber) activity
Pace (steps) taken, time,
BPM, workouts (physical
exertion)
MI XMSH07HM 14 Natural action (activity) 5 5.0 250 135
Kilo calories (calories
burned), distance
(aloofness)
Sleep (slumber) activity
Steps (pace) taken, time,
BPM, workouts (physical
exertion)
Fast track Reflex 2.0 10 Natural action (activity) 4 3.7 200 90
Kilo calorie (calories
burned), distance
(aloofness)
Sleep (slumber) activity
Steps (pace) taken, time,
BPM, workouts (physical
exertion)
A. Ajin Roch et al.
Table 2 SMART watch configuration
Brand Product Battery Tracked data Bluetooth Frequency Voltage Input Battery Processor
code life band (V) current (mAh)
(1 GHz) (mA)
Apple Series 3 1 day Natural action 338S00348 1 5 250 Li-Ion Apple
13 h (activity) 341 S3-Baseband-QUALCOMM
Kilo calorie (calories MDM9635M
burned), distance
(aloofness)
Sleep (slumber)
activity
Dynamic Programmable Clock Frequency …

Steps (pace) taken,

time, BPM, workouts
(physical exertion)
Samsung Gear 2 4 days Natural action 4.1 1 5 150 250 1 GHz dual-core Cortex A-7
(activity) processor
Kilo calorie (calories
burned), distance
(aloofness)
Sleep (slumber)
activity
Steps (pace) taken,
time,
BPM, workouts
(physical exertion)
343
344 A. Ajin Roch et al.

Table 3 Comparison of average power observed for 30 days

Brand Steps BPM Calories Distance Power Workout Sleep
taken (Average) burned (km) consumed (mins) activity
(Kcl) (for 100%) (Hrs)
Fitbit 7024 80 280 5.01 18 30 8
MI 6121 83 249 4.56 9 30 7
Apple 7028 79 276 5.0 72 30 8
Samsung 7027 79 274 5.01 47 30 8
Fast track 7011 78 276 5.02 15 30 7

Table 4 Variation of frequency for various parameters

Clock Normal Medium BPM BPM Walking Walking Aggressive
(MHz) sleep sleep (normal) (exercise) (leisure) (exercise) workout
200 Inactive Inactive Proactive Inactive Active Inactive Inactive
400 Active Inactive Inactive Inactive Inactive Inactive Inactive
600 Proactive Active Active Inactive Proactive Inactive Inactive
800 Inactive Proactive Inactive Proactive Inactive Active Proactive
1000 Inactive Inactive Inactive Active Inactive Proactive Active

parameters of the considered brand with different frequencies selected as given in

Table 4.
The tracked data are categorized under active, proactive and inactive for various
clock frequencies where the active functionally of device is shown in data set format.
The observed values shall be made in high-level performance using the machine
learning algorithms. In our approach, multilinear regression for performance moni-
toring and logistic regression for data-oriented monitoring have been applied. Since
both the concept come under supervised model, generated test data shall be used to
estimate with good accuracy.
Database gathered for as far back as one month gave an extensive continuous
real-time based yield of parameters as shown in Figs. 3 and 4. Utilizing multilinear
expectation model, the steps taken and BPM are in persistent checking, 96% of the
force has been used by steps taken that was followed with 88% by the BPM has been
analysed. The analysed data may change with respect to the climatic conditions; still,
it was observed that the overall data gathered provided best outcome with the available
wearable and SMART watches. As previously mentioned, multilinear modelling was
preferred for the data set to prove the overall power consumption from the gathered
information based on various parameters.
Figure 3 shows that percentage of power consumption was higher for steps taken
and BPM. The force parameter has been investigated by the shading variety that
noticed the rest tracker alongside step taken expends more level of intensity while
BPM with exercises devours less power. It is obvious that if it is considered to provide
Dynamic Programmable Clock Frequency … 345

Fig. 3 Percentage of power

consumption

Fig. 4 Order-wise power

monitoring of parameters

separate clock frequency for each parameter manually, it can be controlled and the
power consumption shall be reduced for the device.
Figure 4 shows the order-wise power monitoring of the parameters and their
influence from the generated data set. It has been clearly observed from the picture
that steps taken alone consume high power, followed by BPM, workouts and sleep
tracker.
346 A. Ajin Roch et al.

4 Conclusion and Future Work

The overall aim of the proposed work was to find the best solution for power-related
problems in smart and wearable devices. Even though, nowadays, normal battery
resolves the power issues year by year, for medical-related devices, portable batteries
could not be able to withstand the high precision-based arithmetic and floating-point-
based operations. So, it has become an important key factor for all the embedded
circuit designers to work for an efficient device that can balance all the health-
care applications in more effective power-based manner. In the proposed work, the
database collected for a period of 30 days with the available devices has verified
the applications that are responsible for high power consumption. The same can
be dealt by giving different clocks for different functions which will tremendously
improve the power consumption for which the data set was planned. The future work
can be based on providing the clock with different frequencies and to use logistic
regression so that the expected data set will be maintained to provide required power
consumption by Python-based programming, and it is also clear that PYNQ-based
FPGA boards shall be used for implementation.

References

1. Fengou, M.-A., Mantas, G., Komninos, N.: A new framework architecture for next generation
e-health services. IEEE J. Biomed. Health Inform. 17(1) (2013)
2. James, R., Stanley, M., Laiwah, A.: Monitor an expert system that validates and interprets
time-dependent partial data based on a cystic fibrosis home monitoring program. IEEE Trans.
Bio Med. Eng. 36(5) (1989)
3. Verma, P., Sood, S.K.: Fog assisted-IOT enabled patient health monitoring in smart homes.
IEEE Internet Things J. 5(3) (2018)
4. Barger, T.S., Brown, D.E., Alwan, M.: Health-status monitoring analysis of behavioural
patterns. IEEE Trans. Syst. MAN Cybern. Part A Syst. Humans 35(1) (2005)
5. Wang, J., Zhang, Z., Lee, S., Simon, R.: An Enhanced fall detection system for elderly person
monitoring using consumer home networks. IEEE (2014)
6. Branko, G., Ross, S.: Home telemonitoring of vital signs-technical challenges and future
directions. IEEE J. Biomed. Health Inform. 19(1) (2015)
7. Merilahti, J., Viramo, P., Korhonen, I.: Wearable monitoring of physical functioning and
disability changes, circadian rhythms and sleep patterns in nursing home residents. IEEE J.
Biomed. Health Inform. 20(3) (2016)
8. Chandra, S.: Wearable sensors for human activity monitoring: a review. IEEE Sens. J. 15(3)
(2015)
9. Junnila, S., Kailanto, H., Merilahti, J., Vainio, A.M., Vehkaoja, A., Zakrzewski, M. and
Hyttinen, J.: Wireless, multipurpose in home health monitoring platform: two case trials. IEEE
Trans. Inform. Technol. Biomed. 14(2) (2010)
10. Yan, H., Huo, H., Gidlund, M: Wireless sensor network based e-health system- implementation
and experiment results. IEEE Trans. Consum. Electron. 56(4) (2010)
11. Austin, J., Dodge, H.H., Riley, T.: A smart-home system to unobtrusively and continuously
assess loneliness in older adults. Point Care Technol. 4 (2016)
12. Skubic, M., Guevara, R.D., Rantz, M.: Automated health alerts using in-home sensor data for
embedded health assessment. Wearable Sens. Health Monitor. Syst. 4 (2015)
Dynamic Programmable Clock Frequency … 347

13. Lasierra, N., Alesanco, A., Garcia, J.: Designing an architecture for monitoring patients at
home: ontologies and web services for clinical and technical management integration. IEEE J.
Biomed. Health Inform. 8(3) (2014)
14. Quer, G., Nikzad, N., Chieh, Normand, A.A.: Home monitoring of blood pressure: short-term
changes during serial measurements for 56398 subjects. IEEE J. Biomed. Health Inform. 22(5)
(2018)
15. Arpaia, P., Manna, C., Montenero, G., Addio, G.D.: In-time prognosis based on swarm intel-
ligence for home-care monitoring: a case study on pulmonary disease. IEEE Sens. J. 12(3)
(2012)
16. Fleury, A., Vacher, M., Noury, N.: SVM-based multimodal classification activities of daily
living in health smart homes: sensors, algorithms, and first experimental results. IEEE Trans.
Inform. Technol. Biomed. 14(2) (2010)