IoT Architecture and Connected Devices

Introduction

The Internet of Things (IoT) is an organizational and technical system in which physical objects sense conditions, communicate data, and support automated or human-directed action. In business, IoT is not simply a collection of connected devices. It is an architecture that links physical processes with digital platforms, analytical services, operational systems, and managerial decisions.

An IoT architecture describes how information moves from the physical world to digital systems and how decisions move back to physical processes. A typical flow is:

Physical process → sensors → microcontroller or embedded device → network → gateway or edge device → cloud services and applications → control decision → actuator → physical process

The architecture is usually described as layered. Although organizations use different models, a practical business architecture includes:

1. Device or perception layer: sensors, actuators, microcontrollers, and embedded software.
2. Connectivity layer: wired and wireless networks, protocols, and gateways.
3. Edge and processing layer: local filtering, event detection, protocol translation, and rapid control.
4. Cloud and platform layer: storage, device management, analytics, artificial intelligence, and integration.
5. Application and business layer: dashboards, workflows, enterprise systems, and managerial decisions.

These layers are logical rather than strictly physical. A single industrial controller may perform sensing, processing, communication, and control simultaneously. The purpose of the model is therefore analytical: it helps managers and engineers allocate responsibilities, assess risks, and design scalable solutions.

Learning Objectives

By the end of this chapter, readers should be able to:

• explain layered IoT architecture and the relationships among its layers;
• distinguish sensors, actuators, microcontrollers, embedded software, gateways, edge devices, and cloud services;
• evaluate how IoT systems collect data and control physical processes;
• analyze trade-offs involving latency, reliability, cost, scalability, privacy, and security;
• apply architectural reasoning to a business case involving connected assets.

1. IoT Architecture as a Closed-Loop System

Many information systems are primarily observational: they collect data and present reports. IoT systems can be observational, but their distinctive capability is often closed-loop operation. A sensor measures a condition, software interprets the measurement, and an actuator changes the physical environment. The result is measured again, producing feedback.

For example, a cold-storage system may measure temperature, compare it with an acceptable range, and activate refrigeration when the temperature becomes too high. The system is closed-loop because its output affects the condition that it measures.

A simplified control relationship is:

Here, r(t) is the desired reference condition, such as 4 °C; y(t) is the measured condition; u(t) is the control action, such as compressor power; and f is the control rule. This expression is conceptual rather than a complete controller design. Its applicability assumes that the measurement is meaningful, the actuator can influence the process, and the control system has a defined response rule.

The architecture must preserve the integrity of this loop. If a sensor is poorly calibrated, if a network introduces unacceptable delay, or if an actuator is incorrectly configured, the system may make a harmful decision even when the cloud analytics are sophisticated. Consequently, IoT architecture is a socio-technical design problem involving hardware, software, people, processes, and governance.

Figure: A logical IoT architecture connecting sensing, processing, cloud services, applications, and physical control.

2. Sensors: The Foundation of IoT Architecture

A sensor is a device that detects a physical, chemical, or biological property and produces a corresponding signal. Examples include temperature, pressure, humidity, vibration, location, light, sound, proximity, electrical current, chemical concentration, and biometric sensors.

Sensors are the foundation of IoT because every later decision depends on their measurements. A sensor does not automatically provide business information; it provides a signal that must be interpreted in context. A temperature value, for example, becomes meaningful only when the system knows the asset, location, time, unit, calibration status, and acceptable range.

2.1 Sensor signal chain

A practical sensor path commonly contains four activities:

1. Transduction: conversion of a physical property into an electrical or digital signal.
2. Conditioning: amplification, filtering, isolation, or linearization of the signal.
3. Conversion: transformation of an analogue signal into digital values, often by an analogue-to-digital converter.
4. Interpretation: assigning units, timestamps, identity, quality status, and operational meaning.

A sensor may output an analogue voltage, a current such as 4–20 mA, a frequency, a resistance, or a digital message. The choice affects wiring, noise resistance, installation cost, interoperability, and maintenance.

2.2 Measurement quality

Managers should evaluate sensors using more than nominal range. Important characteristics include:

• Accuracy: closeness of a measurement to the true or accepted value.
• Precision: consistency among repeated measurements.
• Resolution: the smallest distinguishable change.
• Sensitivity: change in output per unit change in the measured quantity.
• Range: minimum and maximum measurable values.
• Response time: time required to respond to a change.
• Drift: gradual change in output when the actual condition is stable.
• Selectivity: ability to respond to the target variable without undue response to other variables.
• Reliability and durability: ability to operate under environmental and operational stress.

Accuracy and precision are not identical. A sensor can produce tightly clustered values that are consistently wrong, or widely scattered values whose average is close to the truth. Calibration, installation, environmental protection, and maintenance influence both.

For a simple measurement model:

where x_m is the measured value, x_t is the true or reference value, b is systematic bias, and ε is random error. The model assumes that bias and random error can be meaningfully separated. Calibration can estimate or reduce b, whereas repeated measurements may help characterize ε.

If a sensor reports 24.6 °C while a trusted reference reports 25.0 °C, the observed error is −0.4 °C. This single observation does not establish the sensor’s full accuracy; multiple tests across the operating range are required.

2.3 Sampling and data volume

A continuous physical signal is often sampled at discrete intervals. If a system samples every 10 seconds, its sampling frequency is:

where T_s is the sampling period in seconds and f_s is the sampling frequency in hertz. For T_s = 10 seconds, f_s = 0.1 Hz. The appropriate rate depends on how quickly the process changes and what decision must be made. Excessive sampling increases energy use, network traffic, storage, and processing costs; insufficient sampling can miss important events.

For periodic signals, a common engineering principle is that the sampling frequency should exceed twice the highest frequency of interest to avoid aliasing. This principle does not remove the need for anti-aliasing filters or careful engineering judgment, particularly where signals are noisy or nonstationary.

2.4 Sensor selection as a business decision

Sensor selection should begin with the decision the organization wants to improve. Questions include: What must be detected? What is the cost of a missed event? What response time is required? Is the device exposed to heat, moisture, vibration, chemicals, or tampering? How often can it be serviced? What evidence is needed for compliance or insurance?

A low-cost sensor may be appropriate for trend monitoring but unacceptable for safety-critical control. Conversely, high-grade instrumentation may be uneconomic when approximate data are sufficient. The correct choice is therefore based on total value and risk, not purchase price alone.

3. Actuators and Physical Control

An actuator converts a control signal into physical action. Examples include motors, valves, relays, pumps, heaters, locks, robotic arms, and variable-speed drives. Sensors report what is happening; actuators help change what is happening.

Actuators may be discrete, such as a relay that is on or off, or continuous, such as a valve whose opening can vary. They require consideration of power, response time, operating range, fail-safe behavior, mechanical wear, and safety interlocks.

A cloud application should not normally be allowed to issue unrestricted commands to dangerous equipment. Local control logic may impose limits, require authorization, or place the equipment in a safe state when communication fails. This principle illustrates a key architectural distinction: business optimization can be centralized, while immediate safety control often belongs close to the physical process.

4. Microcontrollers and Embedded Software

A microcontroller is a compact computing device containing a processor, memory, input/output interfaces, timers, and often analogue-to-digital conversion and communication peripherals. It is designed to control a specific device or embedded function, usually with low cost and low energy consumption.

A microcontroller may read a sensor, apply calibration, detect a threshold crossing, communicate a message, and drive an actuator. It differs from a general-purpose computer in its focused purpose, constrained resources, and close relationship with hardware.

Embedded software is software designed to operate within a device and interact directly with its hardware. It may include a bootloader, device drivers, a real-time operating system, communication stacks, control logic, diagnostics, and security functions. Firmware is often used to describe software stored in non-volatile device memory, although the distinction between firmware and embedded software is not always strict.

Embedded design involves trade-offs:

• Real-time behavior: whether a response must occur within a specified deadline.
• Energy consumption: especially for battery-powered devices.
• Memory and processing limits: which restrict algorithms and update mechanisms.
• Reliability: including recovery after faults or power loss.
• Security: including secure boot, authenticated updates, and protected credentials.
• Maintainability: including remote monitoring and over-the-air updates.

A device that controls a motor may need deterministic local response even if the wider IoT platform is unavailable. Embedded software should therefore distinguish essential local functions from optional cloud-dependent functions.

5. Connectivity, Gateways, and Edge Devices

The connectivity layer transports data and commands. Technologies may include Ethernet, Wi-Fi, Bluetooth Low Energy, cellular networks, low-power wide-area networks, industrial fieldbuses, and other specialized protocols. Selection depends on range, bandwidth, latency, energy, coverage, mobility, environmental conditions, and operating cost.

A gateway connects devices or networks that would otherwise be unable to communicate directly. It may translate protocols, aggregate messages, authenticate devices, buffer data, and provide a path to enterprise or cloud services. For example, a gateway may collect data from industrial equipment using an industrial protocol and forward normalized messages through an IP network.

An edge device performs computing near the source of data or near the physical process. A gateway can be an edge device, but not every edge device is merely a protocol gateway. Edge computing may filter data, run anomaly detection, compress messages, enforce local policies, or issue time-sensitive commands.

Edge processing offers several benefits:

• lower latency for operational decisions;
• continued operation during cloud or network outages;
• reduced bandwidth and cloud-storage cost;
• improved privacy by keeping sensitive data local;
• greater resilience in remote or intermittently connected sites.

It also creates management challenges. Organizations must deploy, patch, monitor, secure, and replace computing resources across many locations. Edge logic can become inconsistent if versions are not governed centrally.

A useful distinction is:

6. Cloud Services and IoT Platforms

Cloud services provide pooled computing, storage, networking, device management, event processing, analytics, identity services, and application interfaces. In IoT, a cloud platform commonly performs device registration, message ingestion, time-series storage, rules processing, visualization, alerting, digital-twin management, and integration with enterprise systems such as ERP, CRM, and maintenance platforms.

Cloud architecture is attractive because it supports elastic capacity, centralized governance, and access across locations. It is not a substitute for all local processing. A cloud-only design may be inappropriate where control requires millisecond response, connectivity is unreliable, or data cannot legally leave a site.

A sound architecture assigns each function to the location that best satisfies its requirements:

• Device: signal acquisition and immediate hardware protection.
• Edge: rapid decisions, filtering, local autonomy, and protocol conversion.
• Cloud: long-term analysis, fleet-wide comparison, model training, and cross-site coordination.
• Business application: workflow, accountability, financial impact, and managerial action.

Cloud services also introduce dependency on provider availability, pricing, data location, interface stability, and contractual exit provisions. Vendor lock-in can arise when devices, data formats, and applications depend heavily on proprietary services. Open interfaces, portable data models, documented APIs, and tested migration plans reduce this risk.

7. Data and Control Flow in Layered Architecture

IoT data is valuable only when it is trustworthy, interpretable, and connected to action. Data should normally include metadata such as device identity, timestamp, location, unit, calibration status, and quality indicator. Without metadata, a numerical value may be impossible to compare or audit.

A typical telemetry path is:

1. The sensor detects a physical condition.
2. The microcontroller samples and conditions the signal.
3. Embedded software validates, timestamps, and formats the reading.
4. A gateway authenticates, aggregates, or translates the message.
5. An edge device filters data or performs immediate analysis.
6. Cloud services store, analyze, and correlate the data.
7. An application presents a recommendation, alert, or workflow.
8. An authorized command travels back through controlled interfaces.
9. An actuator changes the physical process.
10. Sensors measure the result, closing the loop.

The architecture should specify what happens when a reading is missing, implausible, delayed, duplicated, or contradictory. Data-quality rules might flag a temperature outside the sensor’s physical range, reject impossible timestamps, or compare redundant sensors. These rules are operational controls, not merely technical conveniences.

8. Security, Safety, Ethics, and Governance

Connected devices expand the attack surface of an organization. Risks include weak default passwords, insecure update mechanisms, exposed interfaces, unencrypted communication, compromised gateways, malicious commands, and unauthorized collection of personal data.

Core controls include unique device identities, strong authentication, encryption in transit and at rest, least-privilege authorization, secure boot, signed software updates, network segmentation, vulnerability management, logging, backup, and incident response. Security must be designed across the lifecycle, from procurement and installation through decommissioning.

Safety differs from cybersecurity but interacts with it. A cyberattack that changes a valve setting can become a physical safety incident. Systems should define safe states, independent interlocks, manual overrides, alarm escalation, and recovery procedures. Safety-critical designs require appropriate engineering standards and formal assurance beyond general-purpose IoT guidance.

Ethical and governance issues include workplace monitoring, consent, data minimization, retention, algorithmic bias, transparency, and accountability. For example, employee-location sensors may improve workflow but also enable intrusive surveillance. A responsible organization identifies the legitimate purpose, limits collection, informs affected people, protects access, and evaluates proportionality.

Managers should establish ownership for device identity, data quality, software updates, incident response, and decision rights. An IoT system without governance may accumulate unreliable devices and unowned data even if its initial pilot succeeds.

9. Illustrative Case: Connected Cold-Chain Distribution

Background

A regional pharmaceutical distributor operates refrigerated vehicles and warehouses. Product quality depends on maintaining specified temperature conditions during storage and transport.

Problem

Manual temperature checks are infrequent. When a shipment is found to be out of range, managers cannot determine precisely when the deviation occurred or whether the equipment, door opening, sensor, or handling process caused it.

Analysis

The business requires reliable measurement, rapid notification, evidence for audit, and local protection when connectivity is absent. It does not require every raw reading to be transmitted continuously.

Approach and solution

Temperature sensors are installed at representative points in each vehicle and storage zone. A microcontroller samples the sensors, checks plausibility, timestamps readings, and records local data. An edge gateway aggregates readings and sends them through a cellular connection. If temperature crosses a defined limit, the gateway immediately activates a local alarm and stores the event even when the network is unavailable.

Cloud services maintain device identities, store time-series data, compare temperature across vehicles, and generate alerts. The business application links exceptions to shipment records and maintenance workflows. Authorized personnel can acknowledge an alert, inspect the evidence, and initiate corrective action. The refrigeration actuator remains governed by local control logic rather than unrestricted cloud commands.

Outcome

The architecture improves visibility and creates an auditable record. It also separates rapid local protection from longer-term cloud analysis. However, success depends on calibration, sensor placement, cellular coverage, battery management, alarm response procedures, and disciplined software maintenance.

Learning points

The case demonstrates that IoT value comes from the complete chain, not from sensors alone. A high-quality sensor without timestamps, identity, connectivity, response procedures, and governance produces limited business value. It also illustrates why edge autonomy is important when physical risk exists and communication may fail.

10. Contemporary Design Considerations and Best Practices

Modern IoT systems increasingly use digital twins, machine learning, containerized edge applications, and zero-trust security principles. A digital twin is a digital representation of a physical asset or process that is updated using operational data. It is useful when the representation has a defined purpose, reliable data, and a process for acting on insights; creating a visually impressive model without decision value is not sufficient.

Machine learning can detect unusual vibration, predict equipment failure, or classify events. Its performance depends on representative training data, sensor stability, concept drift, explainability requirements, and human oversight. Machine learning should not be treated as a replacement for instrumentation or domain knowledge.

Recommended practices include:

• define the business decision before selecting devices;
• design for failure, including power, network, sensor, and cloud outages;
• use consistent identity, units, timestamps, and metadata;
• place control functions according to latency and safety requirements;
• test calibration and sensor placement in real operating conditions;
• secure devices and update them throughout their lifecycle;
• use open interfaces and plan for portability;
• monitor device health as well as the physical process;
• measure value through reduced downtime, waste, risk, energy, or service cost;
• pilot under representative conditions before scaling.

Chapter Summary

IoT architecture connects physical processes with digital decision systems through layered functions. Sensors measure physical conditions, while actuators create physical effects. Microcontrollers and embedded software provide local acquisition, logic, and control. Gateways connect heterogeneous networks, and edge devices process data close to its source to reduce latency, bandwidth use, and dependence on cloud connectivity. Cloud services provide scalable storage, analytics, device management, and enterprise integration.

The most important architectural principle is appropriate allocation of responsibility. Immediate safety and control functions often belong locally; fleet-wide analysis and long-term coordination are well suited to cloud services. Measurement quality, metadata, calibration, cybersecurity, safety, privacy, governance, and lifecycle management determine whether an IoT system creates reliable business value. A successful design is therefore not merely connected; it is purposeful, resilient, secure, interpretable, and operationally governed.

Key Terms

Sensor: A device that detects a physical, chemical, or biological condition and produces a signal.

Actuator: A device that converts a control signal into physical action.

Microcontroller: A compact processor-based device designed for dedicated embedded control.

Embedded software: Software designed to operate within and directly interact with a device’s hardware.

Firmware: Software stored in non-volatile device memory and responsible for device operation.

Gateway: A device that connects networks, translates protocols, and may aggregate traffic.

Edge device: Computing equipment located near the data source or physical process.

Cloud service: Provider-hosted computing, storage, networking, analytics, or management capability.

Transducer: A component that converts a physical quantity into an electrical or digital signal.

Calibration: Comparison and adjustment of an instrument against a reference.

Accuracy: Closeness of a measurement to an accepted or true value.

Precision: Consistency of repeated measurements.

Resolution: The smallest change that a measurement system can distinguish.

Latency: Delay between an event, its processing, and the resulting response.

Telemetry: Data transmitted from a device for monitoring or analysis.

Closed-loop control: A system in which measured output informs subsequent control action.

Digital twin: A digital representation of a physical asset or process updated with operational data.

Interoperability: The ability of systems or devices to exchange and use information effectively.

Metadata: Information that provides context for data, such as time, unit, identity, and location.

Review Questions

Short Answer Questions

1. What is the difference between a sensor and an actuator?
2. Why is calibration important in an IoT system?
3. What are two functions commonly performed by a gateway?
4. How does edge processing differ from cloud processing?
5. Why should IoT measurements include metadata?

Long Answer Questions

1. Explain the layers of a practical IoT architecture and show how data and control commands move between them.
2. Discuss the roles of microcontrollers and embedded software in connected devices.
3. Evaluate the business and technical factors involved in selecting sensors for an industrial application.
4. Explain why cloud-only control may be inappropriate for safety-sensitive physical processes.
5. Discuss security, privacy, safety, and governance requirements across the IoT device lifecycle.

Analytical / Application Questions

1. A manufacturer wants to monitor vibration on remote pumps and predict failures. Design an architecture specifying the roles of sensors, microcontrollers, gateways, edge devices, cloud services, and enterprise applications. Justify where each function should be performed.
2. A warehouse sensor reports values that are precise but consistently 2 °C below a reference instrument. Diagnose the measurement problem and recommend an operational response.
3. A connected building loses internet access for six hours during extreme heat. Analyze which functions should continue locally, which can wait for cloud connectivity, and what safeguards should prevent unsafe actuator behavior.

Multiple Choice Questions with Answers

Question 1

Which component directly measures a physical condition such as temperature or pressure?

A. Sensor

B. Actuator

C. Cloud service

D. Gateway

Correct Answer: A. Sensor

Question 2

What is the primary purpose of an actuator in an IoT system?

A. To store historical telemetry

B. To convert a control signal into physical action

C. To authenticate a cloud user

D. To calibrate a sensor automatically

Correct Answer: B. To convert a control signal into physical action

Question 3

A gateway commonly performs which function?

A. Replacing every physical sensor

B. Connecting networks and translating protocols

C. Eliminating the need for embedded software

D. Guaranteeing that measurements are accurate

Correct Answer: B. Connecting networks and translating protocols

Question 4

Which statement best describes edge computing?

A. Processing data near its source or physical process

B. Storing all data only on removable media

C. Using cloud services exclusively for control

D. Measuring physical variables without electronics

Correct Answer: A. Processing data near its source or physical process

Question 5

A sensor samples once every 10 seconds. What is its sampling frequency?

A. 0.01 Hz

B. 0.1 Hz

C. 1 Hz

D. 10 Hz

Correct Answer: B. 0.1 Hz

Question 6

In the measurement model x_m = x_t + b + ε, what does b represent?

A. Random error

B. Sampling frequency

C. Systematic bias

D. Actuator output

Correct Answer: C. Systematic bias

Question 7

Why might a local edge controller be preferred for an immediate safety response?

A. It removes the need for sensors

B. It can respond despite network or cloud delay

C. It guarantees unlimited storage

D. It prevents all measurement error

Correct Answer: B. It can respond despite network or cloud delay

Question 8

Which characteristic refers to closeness of a measurement to an accepted value?

A. Precision

B. Resolution

C. Accuracy

D. Latency

Correct Answer: C. Accuracy

Question 9

Which information is useful metadata for an IoT measurement?

A. Device identity and timestamp

B. The brand colour of the device

C. An unrelated marketing slogan

D. A random unverified number

Correct Answer: A. Device identity and timestamp

Question 10

What is a central risk of allowing unrestricted cloud commands to control dangerous equipment?

A. The system may produce too much metadata

B. A communication or cyber failure may create unsafe physical action

C. The sensor will necessarily become more precise

D. The gateway will no longer translate protocols

Correct Answer: B. A communication or cyber failure may create unsafe physical action