Lecture 1

Introduction to IoT
CS3237
BOYD ANDERSON
SCHOOL OF COMPUTING
Module Details

2
 Boyd Anderson
My research interests are in embedded
systems, wearable sensors, and gait
analysis (analysing the way you walk or
run). I also work on wearable sensors for
sports (such as fencing and sprinting).

Contact me at: [email protected]

My office: COM3-02-46 Drop by any time

3
 Lab Tutors
Our lab tutors will run your weekly labs, as well as mark your lab
work. They are former CS3237 students just like you!

Shreya Pande Wu Dan Liu Zhuohao Sirui Chen

4
 Our Labs are in new The Makers Lab @ SoC

Join the telegram here: https://t.me/MakersSOC 5

Schedule (Tentative)
Week Lectures Labs Project
1 Introduction to IoT
2 Embedded Systems Introduction
3 Comms and Serial Interfaces Getting started with Arduino
4 Statistical Models Communication Lab
5 Neural Networks Statistical Techniques Team Forming
6 Deep Learning Neural Networks
7 Backend Programming Deep Learning Proposal Submission
8 Midterm Backend / Transfer Learning
9 Power Management Power Management Lab Check in 1
10 Energy Sources
11 Communication Check in 2
12 Security
13 The Future of IoT / Summary Final Demo/Report

6
Assessments and Breakdown (Tentative)
This module is 100% CA. It is a project module.

In the second half of the module most of your time will be spent
working on your project

Important Assessments: Lecture Quizzes 5%

Labs (5% each) 35%
Midterm: Week 8 (during Lecture slot)
Midterm 10%
Final Demo: Week 12 or 13 (TBC) Project 50%
Final Report: Week 13 (TBC) Total 100%

7
The Project
It is a group project. You will be working in groups of 4 to 5 students.

The objective of the project is to build a complete IoT system

consisting of the IoT devices, an optional gateway (smartphone), and
the server (laptop or in the cloud).
The focus should be on a system that enables
real-time processing of critical data that requires Proposal 10%
instant action in the edge device (for example, Final Report 10%
fall detection, or motion sensors-based security) Final Demo 15%
along with long-term analytics in the cloud (for Check ins 10%
example, the patient’s activity trends). Peer review 5%

8
We are a diverse bunch! :D

Some are more confident in Deep Learning or Embedded Systems. Or

neither.

I expect all of you to challenge yourselves!

This is a module that touches on a lot of areas, and only in moderate
depth.
9
Introduction

10
Internet of Things (IoT)
Physical objects or
“Things” embedded with
computational
intelligence and
connected to the network.

Cooperation between
cyber-domain and
physical world

Image Credit: ilex

11
What is the Internet of Things?

Computational Connection
Thing
Intelligence to Internet
12
Smart Refrigerator

Image Credit: LIEBHERR

Leanna Garfield/Tech Insider
13
Internet of Things: Not Just Things
 IoT is much more than just the “things”:
• The “things” produce enormous amounts of data (more on
this later).
• Can leverage on this data to do very useful things:
• Predict the weather.
• Understand when and where crowding will occur.
• Plot the likely trajectory of an illness.
• Find key vulnerabilities in a physical system.
• …
 Thus, CS3237 covers both:
• Embedded systems – Knowledge and skills to design and build
the “things”.
• Deep Learning – Knowledge and skills to leverage on data to
perform intelligent actions.

14
IoT Architecture: Components
 Uniquely identifiable embedded systems
(Things) connected with the internet
infrastructure
 Node devices (Things) collect data
through sensors
 The data is sent to the internet through the
gateway
 Things and Gateway are generally
considered as the edge (of the Internet)
 Data analytics takes place in the cloud or
data center
 Following analytics, actuator commands
might be sent 15
Example: Smart Home

16
Smart Home IoT Architecture

17
Endless Opportunities

Source: McKinsey & Company 18

What You Will Learn (Quick Summary):
 The “senses and muscles” of IoT.
• The hardware aspects of IoT:
• Communicating with sensors and actuators.
• Power management.
• Signal processing.
• Hardware security.
 The “brain” of IoT.
• Machine learning:
• Statistical Methods.
• Neural Networks.
• Deep Learning Networks.
• Communications security.
19
 IoT Device Considerations OTA Updates?
Device Security? Edge Processing?
Power Management? Storage?
Computational Power? Cost?
Wireless Connectivity? and many more..
Actuator and Sensor Requirements?
Constraints of Embedded Systems 20 20
 Living within a Budget

Power Budgets
Heat Concerns
Computational Power
Space/Memory Limitations

Constraints of Embedded Systems 21 21

 Power Budgets / Heat Concerns
Unlike your desktop or server, embedded systems must live within a
limited power budget.
 Wired Power Delivery?
• Energy is practically unlimited
• Installation is difficult and costly
• Limited mobility of the sensors
 Battery Power?
• Offers greater mobility of the sensors
• Power budget limits the functionality of the sensors
• Depending on the location and scale, replacing
batteries might be challenging

Constraints of Embedded Systems 22 22

 Power Budgets / Heat Concerns
Embedded Systems can get hot of course! In Desktops, we will use
active cooling more than passive cooling.

Increase the amount of time the device is idling.

Turn off or sleep components when they are not needed.
Ties in with reducing battery consumption!

Memory / Space Issues

Memory comes at a premium in embedded hardware. Both program
storage and working memory are limited!
Constraints of Embedded Systems 23 23
Our Hardware Platform: WeMos D1

Microcontroller ESP-8266EX

Operating
3.3V
Voltage

Digital I/O Pins 11

Analog Input Pins 1

Clock Speed 80MHz/160MHz

Flash 4M bytes

Image: Proto Supplies

24
Microcontroller: ESP-8266
 Low-cost Wi-Fi microchip:
• Full TCP/IP stack and microcontroller
capability
• Produced by
Espressif Systems, China

Compatible with the Arduino IDE and

supported by Platform.IO

Can’t connect to NUS network due to

issues with PEAP 

25
Source: Wikipedia & Expressif Systems
25
Sensors / Actuators
1 x Infrared sensor receiver module
1 x Temperature and humidity sensor module
1 x Infrared emission sensor module
1 x Detect heartbeat module
1 x Microphone module
1 x Flame sensor module
…

26
Sensors
 Sensors connect the analog world to the
digital
 Each IoT sensor collects a stream of time-
correlated data that must be transmitted
securely, possibly analyzed, and possibly
stored
 Sensor data may not always be reliable, and
sensors can fail in the field
 Value of IoT is in the data in aggregate
from multiple sensors

27
 Understanding IoT Data
 IoT data presents a huge challenge:
• Too big:
• 40 houses with smart plugs will produce 4 billion pieces of data a month.
• In the UK, if every house had an IoT connected meter, they are expected to produce 2.64
quadrillion pieces of data a month.
• Too fast:
• Data streams from HD cameras, Lidars and
other sources of information can produce over
1 GB of data per second.
• Often must be processed in real-time
• Too diverse:
• Image data from cameras, heart rate data from
ECQ, gait data from motion sensors, etc.
• In 2014 70% of data scientist said that
diversity of data is a huge challenge that they
are struggling to cope with. 28
Understanding IoT Data
 But there’s lots of $$$ in IoT:
• In 2013 McKinsey noted a 300% growth in IoT in the previous 5 years.
• They predicted that IoT will make an impact of US$2.7 trillion to US$6.2
trillion by 2025.
• In 2015 these figures grew to US$4 trillion to US$11 trillion.
 Healthcare is expected to take up to 41% of the market.
• This is followed by industry taking up 33%, energy market at 7%, and
transportation, agriculture, urban infrastructure and retail taking up the
remaining 19%.

29
Understanding IoT Data
 Ways to understand IoT Data (Analytics)
• Descriptive:
• Answers the question “what happened”.
• Takes the form of describing, summarizing or presenting raw IoT data that has been
gathered.
• Data is decoded, interpreted in context and fused and presented in an understandable way.
• Diagnostic:
• Goes one step deeper than descriptive analytics by trying to discover and present the root
cause and explanation for the data.
• Discovery:
• Discovers events and information that we don’t know about and tries to explain those
events.
• Different from previous two analytics by detecting new and novel trends and information,
rather than just presenting or explaining existing data.
 These three techniques work on hindsight – Data that is already
recorded.

30
Understanding IoT Data
 Ways to understand IoT Data (Analytics)
• Predictive:
• Uses past (hindsight) data to answer the question:
“What is likely to happen?”
• Predicts future outcomes AND the ability to
assess the quality of these predictions.
• Prescriptive:
• Looks at past data and predicted outcomes to
suggest what to do.
• Enables decision makers to not only look into the
future for opportunities, but also presents the
best course of action to act on foresight in a timely
manner.
• Also provides ability to evaluate “what if”
scenarios.
 Unlike previous three techniques, these
two techniques are “foresight” techniques.

31
Previous Projects

32
Some Student Projects For Inspiration
Here are some projects that were completed by students in the
CS3237 and the SWS3009 summer workshop

SWS3009:
• 5 days of lectures + 5 days of project work.
• Gives you some ideas of what you can do.

33
Real-time mask detection and tracking

34
Real-time Distracted Driver Detection

35
Plant Incubator

36
Smart Rubbish Bin

37
Blind Navigation System

38
Baby Monitor

39