YT YouTube2Text

YouTube2Text → Transcripts → IoT Viva Theory from Scratch | Complete Internet of Things in 42 Min | SPPU TE AI&DS 2024 Pattern

IoT Viva Theory from Scratch | Complete Internet of Things in 42 Min | SPPU TE AI&DS 2024 Pattern — Transcript

by Pranav Malshikhare · 5,640 words · 971 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01Hello Pranav [music] and welcome to your
0:03complete internet of things viva
0:04revision. This video follows [music]
0:06your SPU syllabus for third year AI and
0:10data science 2024 [music]
0:11pattern course MDM331.
0:14We will cover every [music] theory
0:16topic, every sensor, all 12 lab
0:18assignments and the questions your
0:20[music] examiner is most likely to ask
0:22with crisp answers. Watch it once end to
0:25end with full attention and [music] you
0:27will walk into your orals confident.
0:29Let's begin.
0:32[music]
0:32Chapter one, the fundamentals of the
0:34internet of things.
0:37So what [music] exactly is IoT? Remember
0:40this one line definition for your Viva.
0:43The internet of things is a network
0:45[music] of physical objects fitted with
0:47sensors, software, and connectivity. So
0:49they can collect and [music] exchange
0:51data. These things sense their
0:54environment, communicate over a network,
0:56take decisions, and even [music] act
0:58with minimal human intervention. Each
1:01device has a unique identity [music]
1:03such as an IP address or a device ID, so
1:07it can be addressed on the network.
1:09Think of a smart electricity meter,
1:11[music] a fitness band, or a smart
1:13irrigation system that waters the field
1:16only when the soil is dry.
1:19Every IoT system is built from six
1:22[music] components. Learn them in order
1:24from the physical world to the user.
1:27First, sensors. [music]
1:28They measure physical quantities like
1:30temperature, light or gas and convert
1:32them into electrical signals. They are
1:34[music] the input side. Second,
1:37actuators. They do the opposite. They
1:39convert an electrical [music] signal
1:41into physical action like an LED
1:44glowing, a motor rotating or a [music]
1:46relay switching. Third, the processing
1:48unit such as Arduino Uno, Raspberry
1:51[music] Pi or ESP32 which reads sensors
1:55and runs our logic. [music] Fourth,
1:57connectivity, Wi-Fi, Bluetooth, low
2:00energy, Laura [music] or cellular
2:02networks carry the data. Fifth, the
2:04cloud or server where data [music] is
2:06stored and analyzed. And sixth, the
2:09application, a dashboard or mobile app
2:12where the [music] user finally sees the
2:14information and controls devices.
2:18A [music] very common Viva question.
2:20What is the difference between physical
2:22design and logical design of IoT?
2:26Physical [music] design talks about the
2:28actual things and the protocols they
2:30use. Logical design is an abstract view
2:33of functions [music] without going into
2:35hardware details. Physical design
2:37includes devices, [music]
2:39sensors, interfaces and protocols.
2:42Logical design includes [music]
2:43functional blocks, communication models
2:46and communication a PIS [music]
2:48like rest and websocket. For example, a
2:51Raspberry Pi with a DHT11 over Wi-Fi is
2:55a physical design. [music] The chain of
2:57device communication services management
3:00security [music] and application is the
3:02logical design. In short, physical
3:05design answers what is used. Logical
3:08design answers how [music] does it work?
3:11Why do we need the cloud in IoT? Because
3:14small devices have very [music] limited
3:17memory and processing power. The cloud
3:20gives almost unlimited storage for
3:22sensor data [music] that keeps coming 24
3:24by7. It provides scalable computing so
3:28we can run analytics [music] and machine
3:30learning on huge data sets. It gives
3:33remote access so a user can monitor the
3:36home or the factory from anywhere
3:38[music] in the world. It also handles
3:40device management overthe-air [music]
3:42firmware updates and rules that trigger
3:45alerts. Popular platforms are a [music]
3:48WSOT core, Microsoft Azure IoT hub and
3:52things speak for students.
3:55Always [music] be ready with a few
3:57applications and connect them to a
3:59sensor. Smart homes [music] with
4:01automatic lights, smart locks and energy
4:04monitoring. Smart agriculture where soil
4:07moisture sensors switch irrigation pumps
4:10automatically.
4:12Healthcare [music] with wearables and
4:13remote patient monitoring of heart rate
4:16and oxygen level. [music] Industrial IoT
4:20where vibration and temperature sensors
4:22predict [music] machine failure before
4:23it happens. This is called predictive
4:26maintenance.
4:27Smart [music] cities with smart parking,
4:29adaptive traffic signals and smart
4:31street lights and environmental
4:34monitoring [music] like air quality
4:36indexes and flood warning systems.
4:40[music]
4:41Chapter 2, IoT architecture.
4:45Let's start [music] with M2M which means
4:47machine to machine communication.
4:50M2M [music]
4:51is direct communication between two
4:53machines without any human involvement.
4:57It is usually pointto-point [music]
4:59often over a cellular or wired link. M2M
5:03solutions are generally proprietary and
5:05closed built for one specific purpose.
5:08Classic examples are in a [music] TM
5:11talking to the bank server or a vending
5:13machine reporting low stock to the
5:15supplier.
5:18Now [music] compare M2M with IoT. This
5:21table is a favorite of examiners.
5:23[music] Communication. M2M is
5:26pointtooint. IoT is many [music] to many
5:28over the internet using IP protocols.
5:32M2M [music] often uses proprietary nonIP
5:35protocols. IoT uses open IP based
5:39protocols like MQTT, co-op and [music]
5:41HTTP.
5:43Focus M2M focuses on hardware. [music]
5:46IoT focuses on software data and cloud
5:49analytics.
5:51Scale [music] M2M is limited. IoT is
5:54massive and scalable. Billions of
5:56devices [music] and data. In M2M, it
5:59stays local. In IoT, it goes to [music]
6:02the cloud. One line to remember, M2M is
6:05a subset of IoT.
6:08The simplified IoT [music] architecture
6:11is explained in layers. Three layer and
6:13five layer models are [music] both
6:15popular. Let's do the five layer one,
6:17which includes the three layer one. At
6:20the bottom [music] is the perception
6:22layer. the sensors and actuators that
6:24touch the physical world. [music] Above
6:26it, the network layer which transmits
6:28data using Wi-Fi, Laura, cellular
6:31networks and gateways. Then the
6:34processing or middleware layer where
6:36data [music] is stored and processed in
6:38the cloud at the edge and in databases.
6:41Then the application layer which
6:43delivers services to [music] the user
6:45like a smartarthome dashboard. And at
6:48the top the business layer which manages
6:50the whole system business models and
6:52decisions. For the three layer model
6:55simply say perception, [music] network
6:57and application.
6:59The IoT protocol stack maps protocols to
7:03[music] layers. Remember it from top to
7:05bottom. Application layer MQTT co-op
7:09HTTP [music]
7:10a MQP XMPPP and websocket.
7:14Transport [music] layer TCP and UDP.
7:17Security sits here too. TLS [music] runs
7:20over TCP and DTLS runs over UDP. [music]
7:25Network layer IP version 4 IP version 6
7:29low WPNEN and [music] RPL the routing
7:32protocol for low power networks. Link
7:35and physical layer [music] Wi-Fi
7:37Ethernet Bluetooth low energy ziggb on
7:40802.15.4 [music]
7:42Four, Lara and NB IoT.
7:47[music] Chapter 3, how IoT devices
7:49communicate.
7:51There are four communication models in
7:54[music] IoT. This is a guaranteed
7:56question. One, request response. The
7:59client sends a request. The server sends
8:01a response. [music] It is stateless.
8:04HTTP is the best example. Two, [music]
8:07publish subscribe. Publishers send data
8:09to a topic on a broker and the broker
8:11[music] forwards it to all subscribers
8:13of that topic. Publishers and
8:15subscribers do not know each other.
8:17[music] MQTT is the best example. Three,
8:21pushpull data producers push data into a
8:24queue and consumers pull data from
8:26[music] it. The queue acts as a buffer
8:28so producers and consumers can work at
8:30different speeds. Four, exclusive
8:32[music] pair. A full duplex birectional
8:35connection that stays open between
8:37client and server. [music] Web socket is
8:39the example.
8:42Next, the communication patterns [music]
8:44named by who talks to whom. Device to
8:48device, two devices communicate [music]
8:50directly, usually over short range
8:52protocols like Bluetooth or Ziggb, a
8:55smart bulb and a smart [music] switch,
8:57for example. Device to cloud. The device
9:00connects directly to a cloud service
9:02over Wi-Fi [music]
9:03or Ethernet like a smart thermostat
9:06sending readings to its cloud. [music]
9:08Device to gateway. Devices connect to a
9:11local gateway like a smartphone or a
9:13Raspberry [music] Pi hub which
9:15translates protocols and forwards data
9:17to the cloud. Useful when devices cannot
9:20[music] speak IP. Back-end data sharing.
9:23Data from one cloud is [music] shared
9:25with other cloud services or third
9:27parties through a PIS. So it can be
9:29combined and analyzed. [music]
9:33Chapter four, networks and connectivity
9:36protocols.
9:38The TCP IP model [music] has four
9:41layers. Application layer which gives
9:44services to the user like HTTP, MQTT and
9:48DNS.
9:49Transport [music] layer which gives end
9:51toend delivery between processes using
9:54port numbers [music] TCP and UDP live.
9:57here internet layer which handles
10:00logical addressing and routing [music]
10:02of packets that is IP and the network
10:05access layer which physically sends bits
10:08over Ethernet or Wi-Fi.
10:11TCP versus UDP. [music]
10:14TCP is connection oriented. It first
10:17sets up a connection with a three-way
10:19[music] handshake. SIN sin arc UDP is
10:23connectionless.
10:24TCP is reliable. It uses sequence
10:27numbers, acknowledgements [music] and
10:29retransmission. UDP gives no guarantee
10:32of delivery. Because of this, TCP is
10:35slower with more overhead while UDP is
10:38fast and lightweight. [music]
10:40In IoT, MQTT and HTTP run on TCP while
10:45co-op runs on UDP.
10:49Why does IoT need [music] IP version 6?
10:52Look at this table. IP version 4 uses
10:5532bit [music]
10:56addresses about 4.3 billion. That is not
10:59enough for billions of IoT [music]
11:01devices. IP version 6 uses 128 bit
11:05addresses [music] practically unlimited.
11:08IP version 4 is written in dotted
11:11decimal. IP version 6 is written in
11:13hexadimal separated by [music] colons.
11:16IP version 6 supports stateless address
11:19auto configuration so devices configure
11:22[music] themselves.
11:24IP6 security is optional in version 4
11:27and was [music] designed into version 6
11:30and version 6 removes the need for n so
11:33every device can have its own global
11:35address that is exactly what IoT needs.
11:40Six low W pan is an important topic. It
11:44stands [music] for IP version 6 over low
11:47power wireless personal area networks.
11:49[music]
11:50It allows IP version 6 packets to travel
11:53over it E82.15.4
11:57radios [music]
11:57whose frame size is only 127 bytes while
12:01[music] an IP version 6 packet can be
12:041280 bytes. So six low W pen adds an
12:08adaptation layer that compresses [music]
12:10the large IP version 6 header and
12:13fragments packets into smaller pieces.
12:16This gives [music] even tiny battery
12:18powered sensors a real IP address. It is
12:21used in the thread smarthome protocol.
12:25Communication protocols [music] are
12:27divided by range. RFID and NFC work at a
12:32few centimeters [music]
12:33used for identification and payments.
12:36Bluetooth low energy works up to [music]
12:39about 10 m for wearables. ZigGB works
12:42from 10 to 100 m in a mesh used for home
12:46automation.
12:47Wi-Fi [music] covers 50 to 100 meters
12:50with high data rate and long range low
12:54power wide area networks [music] called
12:56LPWN like Laura WAN and NBOT cover
13:00several kilometers with very [music] low
13:02power. Remember the trade-off longer
13:04range and lower power means lower data
13:06rate.
13:09RFID [music]
13:10radio frequency identification
13:12identifies objects using radio [music]
13:14waves without contact or line of sight.
13:18An RFID system [music] has three parts.
13:21The tag which is a chip with an antenna,
13:23the reader and [music] the host system
13:25with a database. Passive tags have no
13:28battery. They get energy from the
13:30reader's electromagnetic [music]
13:31field by induction and reply with their
13:34ID. Active tags have their own battery
13:37and can transmit over much longer
13:39distances [music] like vehicle tracking.
13:42Frequency bands are low frequency 125
13:45kohertz [music]
13:46high frequency 13.56 MHz which is what
13:50[music] our MFRC522
13:52uses and ultra high frequency [music]
13:55860 to 960 MGHertz.
13:59Laura [music] and Laura WAN. Students
14:01often confuse these two so be clear.
14:04Laura meaning long range is [music] the
14:06physical layer. It uses chirp spread
14:08spectrum modulation which makes signals
14:11robust over long [music] distances.
14:13Laura one is the protocol on top of it.
14:16It defines the [music] M a C layer and
14:18the network architecture. The topology
14:22is a star of stars. [music] Nodes talk
14:24the gateways and gateways forward to a
14:26network server over the internet. It
14:29reaches 2 [music] to 5 km in cities and
14:32over 15 in open areas and a sensor can
14:34run for years on a battery. [music] The
14:36trade-off is a very low data rate.
14:39Loroan devices have three classes.
14:41[music] Class A has the lowest power and
14:44listens only briefly after it transmits.
14:46Class B adds scheduled receive windows.
14:48[music] Class C listens almost all the
14:51time and uses the most power. In India,
14:54Laura uses the unlicensed [music]
14:56865 to 867 megahertz band.
15:02Now, cellular [music] IoT, which uses
15:04licensed mobile networks, NB IoT,
15:08narrowband IoT, [music] uses very little
15:10bandwidth, penetrates deep indoors, and
15:13uses very low power. Perfect for smart
15:16water and gas meters. LTEM [music]
15:19also called CATM M1 supports mobility
15:22and even voice. So it suits [music]
15:24vehicle trackers and wearables and 5G
15:27brings massive machine type
15:29communication supporting [music] up to a
15:31million devices per square kilometer.
15:34Wi-Fi defined by Ile E2.11 [music]
15:38has variants B, G, and N work at 2.4
15:42GHz. A C works at 5 GHz. [music]
15:45A X is Wi-Fi 6 which handles many
15:48devices efficiently. An 802.11H [music]
15:52called Wi-Fi hollow works below 1 GHz,
15:55[music] reaches about a kilometer and
15:57was designed specially for IoT.
16:01Your syllabus [music] also mentions a I
16:04enabled networking. It means using
16:06machine learning to [music] manage the
16:08network itself. A I can predict traffic
16:11patterns and allocate bandwidth
16:13dynamically. [music]
16:14It enables self-healing networks that
16:17detect faults and reroute automatically.
16:20It detects [music] intrusions and
16:21anomalies in network traffic, improving
16:24security. And it does energy [music]
16:26aware routing so battery powered nodes
16:29last longer.
16:32Chapter [music] 5, the hardware and how
16:35data is acquired.
16:37Let's look [music] at the two boards in
16:39your lab. First, the Arduino Uno.
16:42Arduino [music]
16:43Uno is a microcontroller board based on
16:46the AT mega 328P running at 16 MGHertz.
16:50[music] It has 14 digital input output
16:53pins, six of which support PWM and six
16:57analog input [music] pins, a 0 to a 5.
17:00Its analog to digital converter is 10
17:03bit. [music] So analog readings range
17:05from 0 to 1,023.
17:07It works at 5 Vs. It has 32 [music]
17:11kilobytes of flash and only 2 kilobytes
17:13of SRM and it runs no operating system.
17:16[music] It runs one program forever. It
17:19is programmed in C and C++ using the
17:22Arduino IDE. [music] Every sketch has
17:25two functions. Setup runs once and loop
17:28runs again and again.
17:31Next, the Raspberry [music] Pi.
17:33Raspberry Pi is a full single board
17:36computer with [music] a quad core RM
17:38processor running Raspberry Pi OS which
17:41is Linux. It has a [music] 40pin GPIO
17:44header. Very important. Its pins work at
17:473.3 [music] volts and are not 5V
17:50tolerant. Connecting 5 volts directly
17:53can damage the Pi. It has built-in
17:55[music] Wi-Fi, Bluetooth, Ethernet, USB
17:58and HDMI. So, it [music] is perfect as
18:01an IoT gateway or edge device. A classic
18:05[music] Viva trap. The Raspberry Pi has
18:07no built-in analog to digital converter.
18:10To read an [music] analog sensor like
18:11MQ2, we use an external ADC chip such as
18:15the MCP 3008 connected [music] over SPI.
18:20We program it in Python using libraries
18:22like [music] RPI.GPIO
18:25or GPIO.
18:28Arduino versus Raspberry Pi. [music]
18:31Expect this one. Arduino is a
18:34microcontroller. Raspberry Pi is a
18:36microprocessor based [music] single
18:38board computer. Arduino has no operating
18:42system. [music] Raspberry Pi runs Linux.
18:45Arduino works at 5V [music] logic.
18:47Raspberry Pi at 3.3 volts. Arduino has
18:52analog [music] inputs with a 10 bit ADC.
18:54Raspberry Pi has none. So Arduino is
18:58best for simple realtime control of
19:00sensors and actuators [music] while
19:02Raspberry Pi is best for edge computing,
19:04machine learning, cameras and
19:06networking. In many projects both are
19:08used together.
19:11Now the hardware interfaces in your
19:13syllabus [music]
19:14start with GPIO a DC and DSC. [music]
19:18GPIO generalpurpose input output is a
19:21pin you can program as input [music] to
19:23read a sensor or as output to drive an
19:26LED. It has only two states high and low
19:30[music] and a DC analog to digital
19:32converter converts a continuously
19:35varying voltage [music] into a digital
19:37number that the processor can
19:38understand. The resolution decides the
19:42number of levels and n bit [music] a DC
19:44gives 2 to the power n levels. Arduino's
19:4710 bit a DC [music] gives 1,024 levels.
19:51So each step is 5 volts divided by 1,024
19:56about 4.9 m volts. If [music] the
19:58reading is 512 the voltage is about 2.5
20:02volt. A DSC digital to [music] analog
20:05converter does the reverse. Arduino Uno
20:08has no true DSC. So it uses [music] PWM
20:11pulse width modulation to imitate an
20:14analog output like dimming and LED.
20:19The three serial protocols UR, SPI and I
20:23squared C [music] are very important.
20:25Learn this table well. Wires URes
20:28[music] two wires transmit and receive.
20:30SPI uses four M OSI MISO [music]
20:34the clock seck and slave select. I
20:37squared C uses only two. The data line
20:40SD [music] and the clock line Sccl
20:43clock UR is a synchronous. There is no
20:46clock wire. So both sides must agree on
20:48a B rate like 9,600. [music] SPI and I
20:52squared C are synchronous with a shared
20:54clock devices. U RT is pointtooint.
20:57[music]
20:58SPI has one master and many slaves each
21:01needing its own slave select line.
21:03[music] I squared C connects many
21:05devices on the same two wires using
21:08seven bit addresses. Speed [music] SPI
21:11is the fastest in megahertz. I squared C
21:14is medium 100 or 400 kilohertz. [music]
21:17UR is typically slow. Examples. The
21:21serial monitor uses UR. [music] Our RFID
21:24reader MFRC522
21:27uses SPI. OLED displays often use I
21:31squared C.
21:32>> [music]
21:33>> Data acquisition does not end at reading
21:36a sensor. Realtime [music] streaming
21:39means sensor readings flow continuously
21:42each with a time stamp. A buffer or Q
21:45[music] temporarily stores data when the
21:47network is slow or down. So readings are
21:50not lost. That is exactly the pushpull
21:52idea. Sampling rate [music] matters.
21:55According to the Nyquis theorem, we must
21:57sample at least twice the highest
21:59frequency in the signal. The [music]
22:01challenges in IoT data acquisition are
22:04noise, missing values, sensor drift
22:07[music] over time, and the need for
22:08calibration,
22:10plus limited battery power, [music]
22:12limited bandwidth, latency, and
22:14heterogeneous data coming in different
22:16formats from different sensors.
22:20[music]
22:20Chapter 6, a gallery of every sensor and
22:23actuator you used in the lab.
22:27The [music] DHT11
22:28measures temperature and humidity.
22:31Inside it has a capacitive humidity
22:34sensor [music] and an NTC thermister
22:36whose resistance falls as temperature
22:38rises. An internal chip converts both
22:41[music] readings into a digital signal
22:43sent on a single data wire. Each reading
22:46is 40 bits. 16 bits of humidity, 16
22:50[music] bits of temperature and an 8 bit
22:52checksum to detect errors. It measures 0
22:55to 50° C, [music] accurate to plus -2°
22:59and 20 to 90% relative humidity. It
23:03should be read at most once every
23:05second. [music] Its big brother, the
23:07DHT22,
23:08is more accurate and has a wider range.
23:12The IR obstacle sensor detects [music]
23:15whether an object is present in front of
23:17it. It has an infrared LED [music] that
23:21transmits and a photo diode that
23:23receives. When an object is present,
23:25[music] infrared light reflects off it
23:28and reaches the photo diode. [music] An
23:30LM393
23:32comparator compares the received signal
23:34with a reference [music] set by the
23:36potentiometer. Turning the potentiometer
23:38changes the detection distance. The
23:41output is digital and usually active
23:43low. [music] LOW means obstacle
23:46detected. High means no obstacle. Always
23:48mention that it depends on the module.
23:51Its range is about 2 to [music] 30 cm.
23:54It can be fooled by sunlight and black
23:56surfaces absorb [music] infrared so they
23:59are harder to detect.
24:02The [music] MQ2 gas sensor is used for
24:04gas leakage detection. It detects LPG,
24:08methane, propane, smoke, and hydrogen.
24:12It contains a [music] thin dioxide
24:14semiconductor layer heated by an
24:16internal heater. [music] In clean air,
24:18its conductivity is low. When
24:20combustible gas is present, conductivity
24:23increases. So, resistance falls and the
24:26output [music] voltage rises. It has two
24:29outputs. A O the analog [music] output
24:31gives the gas concentration. Read by
24:34analog read from 0 to 1,023. [music] Do
24:38the digital output switches when the
24:40onboard threshold is crossed. It must be
24:42[music] preheated before readings are
24:44reliable and calibrated in clean air to
24:47[music] avoid false alarms.
24:50A few more sensors you may be asked
24:52[music] about. An LDR light dependent
24:56resistor has a resistance that falls as
24:58light increases. High resistance in
25:01darkness, [music]
25:01low resistance in bright light. We put
25:04it in a voltage divider with a 10 kilo
25:07ohm resistor and read the middle point
25:09[music] on pin as zero. LM35 is an
25:13analog temperature sensor. It gives 10
25:16molts per degree [music] C. So 300 m
25:19volts means 30°.
25:21The ultrasonic sensor [music] HCSR04
25:25sends a 40 kohertz sound pulse [music]
25:27and measures the time until the echo
25:30returns. Distance equals speed of sound
25:33multiplied by time divided by 2 because
25:36the sound travels to the object and
25:38back. In code it is time in microsconds
25:41[music] time 0.0343
25:44divided by 2 giving cm.
25:48Now [music] actuators an actuator
25:50converts an electrical control signal
25:52into [music] a physical action. An LE
25:56needs a resistor in series to limit
25:58current with a 5V supply. an LED drop of
26:01about 2 volts and 20 milliamps of
26:03current. [music] R is 3 volt divided by
26:06002 which is 150 ohms. So we use the
26:10next [music] safe value 220 ohms. A
26:14buzzer gives an audible alert and an LE
26:16[music]
26:17gives a visual alert. Using both is
26:19redundancy. So the alert is noticed
26:21either [music] way. A relay lets a small
26:245V pin safely switch a high voltage load
26:27[music] like a fan, a pump or a gas
26:30solenoid valve and servos, [music]
26:32DC motors and solenoid valves produce
26:35movement.
26:38[music] Chapter 7, cloud platforms and
26:41edge computing.
26:43Let's look [music] at the big IoT cloud
26:46platforms. A WSOT core from Amazon
26:50provides a managed MQTT [music] broker,
26:53device shadows that store the last known
26:55state of a device, and a rules engine
26:58that routes data to other AWS services.
27:01Microsoft [music] Azure IoT hub provides
27:04device twins, device to cloud and cloud
27:07to device messaging and the device
27:09provisioning service. Google Cloud IoT
27:12[music] core was retired in August 2023.
27:16That is a great point to impress your
27:18examiner. Google now relies on PubSub
27:20and partner platforms for [music] IoT.
27:23For student projects, things speak,
27:26Blink and Adafruit IO are simple and
27:28free to start. Common cloud [music]
27:30services include storage, analytics,
27:33machine learning, visualization, and
27:35alert [music] rules.
27:38Now, the compute stack from devices to
27:40the cloud. At the bottom, devices and
27:44sensors generate [music] data. Edge
27:46computing processes data right on the
27:49device or the gateway [music]
27:50where it is generated, giving responses
27:53in milliseconds. Our Raspberry Pi acts
27:55as the edge device. Vog computing is the
27:59layer in between. It uses local nodes,
28:02routers, and [music] servers in the
28:03local network to process data from many
28:06edge devices. The term was coined by
28:08Cisco. The cloud sits at the top with
28:11[music] massive storage and computing
28:13used for training models and long-term
28:15analysis. One line, edge is on the
28:17device, fog is near the device, cloud
28:20[music] is far away.
28:23Edge versus cloud, five points.
28:26Location, [music] edge is near the
28:28device. Cloud is in a remote data
28:30center. Latency edge is very low in
28:34[music] milliseconds because there is no
28:35round trip. Cloud is higher. Bandwidth
28:39edge sends only useful or summarized
28:41data. [music] Cloud receives all raw
28:43data. This is bandwidth optimization.
28:46Offline edge [music] keeps working even
28:49when the internet is down. Cloud needs
28:51connectivity.
28:52Privacy. With edge, raw data stays
28:55local, which enhances privacy. [music]
28:57These four benefits: latency, bandwidth,
29:00reliability, and privacy are exactly
29:02what your syllabus lists.
29:06>> [music]
29:06>> Chapter 8. Intelligent IoT or IoT.
29:11A IoT [music]
29:13means combining artificial intelligence
29:15with the internet of things. IoT [music]
29:18gives the senses. It collects data. I
29:21gives the brain. It learns from data and
29:23takes decisions. Why do we need
29:25intelligent [music] analytics? Because
29:27IoT generates data that is too large and
29:30too fast for humans to analyze manually.
29:34A I is used for [music] three main
29:35tasks. Prediction like future
29:38temperature. Classification like
29:40recognizing an activity [music] and
29:41anomaly detection like spotting a faulty
29:44machine. Examples include predictive
29:47[music] maintenance, smart traffic
29:49control and health alerts from
29:50wearables. And tiny [music] ML is the
29:54trend of running small machine learning
29:56models directly on microcontrollers.
29:58[music] The ultimate edgeri.
30:02The syllabus specifically lists three
30:04challenges of IoT [music] data. Noisy,
30:06incomplete and heterogeneous.
30:09Noisy data has random fluctuations
30:11[music] caused by electrical
30:13interference or cheap sensors. We handle
30:16it with smoothing [music] like a moving
30:17average filter. Incomplete data has
30:20missing values. When a sensor [music] or
30:22network drops out, we handle it by
30:24imputation like mean, median, [music]
30:27forward fill or interpolation.
30:30Heterogeneous data comes from different
30:32sensors in different formats, units
30:34[music] and sampling rates. We handle it
30:36by standardizing formats, converting
30:39units and aligning timestamps. [music]
30:41We also deal with outliers, duplicate
30:44records and misaligned timestamps.
30:48Predictive [music] analytics using
30:50machine learning follows a clear
30:52pipeline. Collect sensor data with
30:55timestamps. [music]
30:56Clean it by handling missing values,
30:59duplicates, and outliers.
31:01Engineer [music] features like previous
31:03values called lags, rolling averages,
31:06and time of day. Train a [music] model
31:08like linear regression, random forest,
31:11or an LSTM for time series. Evaluate it
31:15with metrics like [music] ME, RMS, and R
31:18squar, and deploy it [music] to make
31:21predictions and trigger alerts in real
31:23time.
31:26Chapter nine, [music]
31:27security in the internet of things.
31:31Security is a major part of part B.
31:33First three [music] terms, a threat is a
31:36possible attack. A vulnerability is a
31:39weakness [music] that the attack can
31:40exploit. And risk combines them. Risk
31:44equals [music] threat time vulnerability
31:46time impact. If any one of these is
31:48zero, risk is low. The most common
31:51vulnerability is [music] weak or default
31:53passwords. In 2016, the Marai botnet
31:57infected [music] hundreds of thousands
31:58of cameras and routers using default
32:01passwords and used them to take down
32:03major websites. Unencrypted [music]
32:05traffic allows sniffing and
32:07maninthe-middle attacks where the
32:09attacker secretly [music] sits between
32:11device and server. Other threats include
32:14denial of service, device spoofing,
32:17firmware [music] tampering, and physical
32:18attacks on devices and vulnerabilities
32:22like missing updates, [music]
32:23insecure a PIS and badly configured
32:26cloud storage.
32:29Security goals [music] are summarized by
32:31the CIA triad. Confidentiality only
32:35authorized parties can read the data
32:37achieved by encryption. Integrity data
32:41is not altered in transit. achieved by
32:43hashes and [music] message
32:44authentication codes. Availability. The
32:48system works when needed, [music]
32:49protected against denial of service.
32:52Now, a classic question. [music]
32:54Authentication answers, who are you? It
32:56verifies identity using a password,
32:59token or [music] certificate.
33:01Authorization answers, what are you
33:03allowed to do? [music] It happens after
33:05authentication using access control
33:08lists. For example, a sensor may publish
33:10to its own [music] topic but not to
33:12others. Authentication first, then
33:14authorization.
33:17Encryption converts readable [music]
33:19plain text into unreadable cipher text.
33:22There are four categories to know.
33:24Symmetric encryption uses [music] the
33:26same secret key to encrypt and decrypt.
33:29A ES is the standard. [music] It is
33:31fast, so it encrypts the actual data.
33:34Asymmetric encryption uses a key pair.
33:37Data encrypted with the public key can
33:39only be decrypted with the [music]
33:40private key. RS and elliptic curve
33:43cryptography are examples. ECC is
33:46preferred in IoT [music] because it
33:48gives the same security with much
33:50smaller keys. Hashing is one way. It
33:53produces [music] a fixed size digest
33:55like SH at 256. You cannot reverse it.
33:59[music] It is used for integrity checks
34:00and storing passwords. and lightweight
34:03cryptography like ASCAN [music]
34:05chosen by NIST in 2023 is designed for
34:09very constrained [music]
34:10devices.
34:13How does TLS transport layer security
34:16actually secure connection [music]
34:18through the handshake? The client sends
34:20a client hello with the cipher [music]
34:22suites it supports and a random number.
34:25The server replies with a server hello
34:28and its [music] digital certificate
34:29signed by a certificate authority. The
34:32client verifies it. This is server
34:34authentication. [music] Then both sides
34:37perform a key exchange, usually elliptic
34:39curve defy helman to agree on a shared
34:42secret without ever sending it on
34:44[music] the network. Both derive session
34:47keys and send finished messages. From
34:50now on, all data is encrypted with a
34:52[music] fast symmetric cipher like a ES.
34:55TLS thus gives confidentiality, [music]
34:58integrity, and authentication. In mutual
35:00TLS, the client also presents a
35:03certificate.
35:05What about UDP? TLS [music] needs the
35:08reliability of TCP. So for UDP, we use
35:12DTLS.
35:14DTLS [music]
35:15datagramgram transport layer security
35:18gives the same security as TLS but over
35:21UDP.
35:22It [music] adds its own sequence numbers
35:24and retransmission of handshake messages
35:27because UDP can lose or reorder [music]
35:30packets. It is used to secure co-ap.
35:33Secure co-ap runs on port [music] 5684.
35:37Memorize the ports. MQTT is 1883. Secure
35:41MQTT is 8883. [music]
35:44HTTP is 80. HTTPS is 443. And remember
35:49SSL [music] is the old insecure
35:52predecessor of TLS. Today we use TLS
35:55version 1.3.
35:58[music]
35:59Chapter 10. The application layer
36:02protocols.
36:04HTTP [music]
36:05is the protocol of the web and also used
36:08in IoT. It follows [music] the request
36:11response model over TCP. Its headers are
36:14text based which makes it heavy for tiny
36:17devices.
36:18>> [music]
36:19>> HTTPS is simply HTTP over TLS on port
36:24443.
36:25REST representational state transfer
36:27[music] is an architectural style. Every
36:30resource like a sensor has a URI and
36:33each request is stateless [music]
36:35carrying all the information needed.
36:37Rest uses HTTP methods. Get to read,
36:41post to create, put [music] to update,
36:43delete to delete. For example, get /
36:45sensors/ temperature and status codes.
36:49200 is okay, [music] 201 is created, 404
36:53is not found, 500 is server error.
36:58MQTT is the most [music] important
37:01protocol for your orals. It stands for
37:03message queuing telemetry transport
37:05[music] created by IBM in 1999.
37:09It follows publish subscribe [music]
37:11through a broker like MSquido, Hive MQ
37:14or EMQX.
37:16It runs on TCP [music] port 1883 or 8883
37:20with TLS. Its fixed header is just two
37:23bytes which is why it is so lightweight.
37:25[music]
37:26Topics are hierarchical strings
37:28separated by slashes like home/ room
37:311/mp. A [music] subscriber can use wild
37:34cards. Plus matches exactly one level
37:36and hash [music] matches all remaining
37:38levels. Quality of service has three
37:41levels. CO0 at most once, fire [music]
37:44and forget. Go S1 at least once
37:46acknowledge, but duplicates are
37:48possible. Go S2 exactly once using a
37:51four-step handshake. [music] The safest
37:53but slowest. A retained message is
37:56stored by the broker and given
37:57immediately to new [music] subscribers.
37:59The last will and testament is a message
38:02the broker publishes if a client
38:04disconnects unexpectedly and keep alive
38:06pings tell the broker the client is
38:08still alive. [music]
38:10Finally, a persistent session lets the
38:12broker remember subscriptions [music]
38:14and cued messages for a client that
38:16reconnects.
38:19Three [music] more protocols.
38:21Co-op the constrained application
38:23protocol is like rest for tiny devices.
38:26It runs on UDP port [music] 5683 with a
38:30four byte header and supports get, post,
38:32put, and delete. Co-op messages [music]
38:35can be confirmable, non-confirmable,
38:38acknowledgement, or reset. Its observe
38:40option lets [music] a client subscribe
38:42to changes of a resource. A MQP,
38:46advanced message queuing protocol, is
38:48for [music] enterprise messaging.
38:50producers send to exchanges which route
38:52messages into cues based on rules
38:54[music] port 5672.
38:57It is very reliable with routing and
38:59transactions. Rabbit MQ is the famous
39:02broker. XMPP [music]
39:05the extensible messaging and presence
39:07protocol is XMLbased. It was built for
39:10realtime chat and presence like whether
39:12a user is online. [music] It is heavier
39:14due to XML on port 5222.
39:19Let's compare all five in one table. A
39:22[music] screenshot of this in your mind
39:24is worth many marks. Transport only
39:27co-op uses UDP. All others use TCP.
39:31[music] Model MQTT is publish subscribe.
39:34Co-op and HTTP are request response.
39:38Co-op with observe. A MQP uses Q's.
39:40[music] XMPPP supports both publish
39:43subscribe and pointtooint.
39:46Wait. [music] MQTT and co-op are very
39:48lightweight. HTTP, A MQP and XMPPP are
39:52heavier. Security [music] co-op uses
39:55DTLS. The others use TLS with SSL for
40:00authentication in a MQP and [music]
40:02XMPP.
40:04and best use. MQTT for telemetry, co-op
40:08for highly constrained nodes, HTTP
40:10[music] for web, a PIS, a MQP for
40:13enterprise, XMPP for chat and presence.
40:18[music] Chapter 11. Privacy in IoT
40:21systems.
40:24Privacy [music] is different from
40:25security. Security protects data from
40:27attackers. Privacy protects people from
40:30misuse of their data even [music] by
40:32authorized parties.
40:34IoT devices collect identity, location,
40:37health data, behavior patterns, and even
40:40audio and video. The risks [music] are
40:43profiling, tracking, inference attacks,
40:45surveillance, and data leaks. For
40:47example, a smart meter's power pattern
40:49[music] can reveal when you are at home
40:51and even when you are sleeping. The
40:54challenges are that sensing is always on
40:56and many devices have no screen [music]
40:59to ask for consent. Laws protect us. The
41:02GDPR in Europe and in India. The digital
41:06personal data [music] protection act
41:072023.
41:10How do we preserve privacy? [music]
41:12Seven techniques. Data minimization.
41:15Collect only the data truly needed for
41:18the purpose. This is your assignment.
41:20Five. Anonymization removes identity
41:23completely. Pseudonymization replaces
41:26identity with a code [music] which can
41:28be reversed only with a separate key.
41:31Encryption both at rest in storage and
41:33in transit over the network. [music]
41:36Edge processing so raw data never leaves
41:38the device. Only results are sent.
41:41Differential [music] privacy adds
41:43carefully calibrated noise to data or
41:46results. So no individual can be
41:48identified [music]
41:49while overall patterns remain accurate.
41:52Federated learning trains a shared
41:53[music] model by sending model updates
41:56not raw data to the server. And finally,
41:59[music] consent, strict access control,
42:02and limited data retention. All of these
42:04together form privacy by design.

About this transcript

This page contains the full transcript of IoT Viva Theory from Scratch | Complete Internet of Things in 42 Min | SPPU TE AI&DS 2024 Pattern by Pranav Malshikhare, generated from the public captions YouTube serves with the video. The transcript has 5,640 words across 971 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.