Embedded Systems and IoT Architecture

Introduction

The Internet of Things (IoT) connects physical objects to digital systems so that they can sense conditions, process information, communicate, and sometimes act autonomously. An IoT system may monitor soil moisture, control a motor, track a vehicle, measure electricity consumption, or detect faults in industrial equipment. Although cloud platforms and mobile applications are visible parts of IoT, the foundation is usually an embedded system located close to the physical process.

An embedded system is a computing system designed as part of a larger product or process. Unlike a general-purpose computer, it is normally built for a defined range of functions and must operate within constraints such as cost, power, memory, size, timing, reliability, and safety. IoT extends embedded systems by adding communication, remote management, data analytics, and integration with other systems.

This chapter establishes the vocabulary and architectural models needed to study embedded IoT systems. It distinguishes major computing platforms, explains the path from sensing to cloud services, introduces edge computing, and presents a practical development lifecycle.

Learning Objectives

After studying this chapter, you should be able to:

1. Define embedded-system terminology and identify common constraints.
2. Distinguish microcontrollers, microprocessors, systems-on-chip, and IoT nodes.
3. Explain sensing-to-cloud architectures and layered IoT models.
4. Describe the role of gateways, edge computing, cloud platforms, and actuators.
5. Select an appropriate architecture for a practical IoT application.
6. Explain an embedded-IoT development lifecycle and its engineering risks.
7. Identify important security, privacy, safety, and governance considerations.

1. Embedded Systems: Meaning and Characteristics

An embedded system combines hardware and software to perform a dedicated function within a larger product. A washing-machine controller, digital thermometer, vehicle airbag controller, smart electricity meter, and wireless environmental sensor are all examples.

The word embedded indicates that the computer is integrated into another system rather than being the primary product itself. The user may not see the processor, but the processor controls or monitors important operations.

Important terminology

• Hardware: The physical electronic components, including processors, memory, sensors, communication interfaces, power supplies, and actuators.
• Firmware: Software stored in non-volatile memory and closely associated with the hardware. Firmware commonly initializes devices, reads sensors, controls outputs, and manages communication.
• Application software: Code that implements the product-specific function.
• Real-time system: A system in which correctness depends on both the result and the time at which the result is produced.
• Hard real-time system: Missing a deadline may cause unacceptable or dangerous failure, as in an airbag controller.
• Soft real-time system: Occasional deadline violations reduce quality but do not necessarily cause catastrophic failure, as in video streaming.
• Bare-metal system: Software running directly on the hardware without a general-purpose operating system.
• RTOS: A real-time operating system that provides scheduling, timing, communication, synchronization, and resource-management services.
• Peripheral: A hardware module that supports the processor, such as a timer, analogue-to-digital converter, serial interface, or digital input/output port.
• Bootloader: Initial software that starts the system and may support firmware updates.
• Device driver: Software that provides a controlled interface to a hardware device.
• Interrupt: A signal that temporarily causes the processor to execute an event-handling routine.
• Watchdog timer: A timer that resets or otherwise responds to software that fails to operate correctly within an expected interval.

An embedded product may be simple, such as a temperature sensor with one input and one output, or complex, such as an automotive control unit containing several processors and multiple communication networks.

2. Constraints in Embedded and IoT Design

Embedded engineering is a process of balancing competing requirements. A faster processor may improve performance but increase energy consumption, cost, heat, or software complexity. A larger battery may improve operating life but increase size and weight.

Processing and memory

Processors have limited clock speed, computation capacity, and sometimes specialized instruction sets. Memory is commonly divided into:

• Flash or ROM for program storage;
• RAM for temporary variables, stacks, buffers, and operating-system data; and
• Non-volatile data memory for configuration, calibration, logs, or device identity.

The design must fit executable code and data within available memory. Communication buffers and sensor-processing algorithms can create unexpected memory pressure.

Timing and determinism

A temperature controller may need to sample a sensor every 100 ms, while a motor-control loop may require much shorter and more predictable intervals. The designer must consider worst-case execution time, interrupt latency, scheduling delays, and communication delays—not merely average performance.

Energy and power

Battery-powered nodes often spend most of their time asleep and wake periodically to sense or communicate. Energy is consumed by sensing, computation, radio transmission, reception, memory access, and leakage. In many wireless systems, transmitting frequently is more expensive than performing a small local calculation.

A simple battery-life estimate can be made under stated assumptions. If a battery has usable charge Q in ampere-hours and the average current is I_avg in amperes, then:

This is an idealized estimate. It assumes that the rated battery capacity is usable, current is approximately constant, temperature effects are negligible, and the battery is not discharged beyond its safe limit. For a 2,000 mAh battery supplying an average current of 10 mA, the ideal estimate is 200 hours, or approximately 8.3 days. In practice, duty cycle, battery chemistry, radio peaks, temperature, and voltage regulation reduce the achievable value.

Size, cost, and environmental conditions

IoT devices may be installed inside machinery, outdoors, underground, or in locations that are difficult to access. Enclosure design may need resistance to dust, water, vibration, corrosion, and temperature extremes. Cost includes not only the circuit board but also assembly, certification, installation, communication service, maintenance, and end-of-life disposal.

Reliability, safety, and maintainability

A device that cannot be physically accessed must tolerate faults and support secure remote updates. Reliability features include watchdog timers, brownout detection, input validation, redundancy, error logging, and safe default states. Safety-critical products require hazard analysis, controlled change procedures, verification, and applicable industry standards.

3. Computing Platforms for Embedded IoT

Microcontroller

A microcontroller unit (MCU) integrates a processor core, memory, timers, GPIO, and communication peripherals on one chip. Many MCUs also include ADCs, pulse-width-modulation outputs, cryptographic accelerators, and low-power modes. They are well suited to sensor nodes, appliances, wearables, and control systems.

MCUs generally provide predictable operation, low power consumption, low cost, and direct hardware control. Their limitations include constrained memory and lower performance for complex operating systems or graphical applications.

Microprocessor

A microprocessor unit (MPU) is primarily a processing core and normally depends on external memory and peripherals. It usually provides greater processing capability and can run operating systems such as Linux. MPUs are appropriate for gateways, cameras, industrial computers, and devices requiring networking, user interfaces, or substantial data processing.

They generally consume more power and require more supporting hardware than MCUs. Boot time, operating-system management, and security maintenance may also be more demanding.

System-on-chip

A system-on-chip (SoC) integrates several major computing functions on one piece of silicon. Depending on the design, it may combine CPU cores, memory controllers, radio interfaces, graphics, security modules, digital signal processors, and peripheral controllers. Many wireless IoT chips are SoCs because they combine an MCU with Wi-Fi, Bluetooth, cellular, or another radio subsystem.

The terms MCU and SoC overlap in practice. An MCU can be regarded as a specialized, highly integrated SoC, but industry usage normally emphasizes MCUs for small control-oriented devices and SoCs for broader integration or heterogeneous processing.

IoT node

An IoT node is a deployed endpoint that participates in an IoT system. It may contain a sensor, processor, power source, communication interface, and optionally an actuator. A node is defined by its system role rather than by one particular chip. For example, a node may be a low-power soil sensor, an industrial vibration monitor, or a smart lock.

Gateway and cloud platform

A gateway connects local devices to another network or service. It may translate protocols, aggregate data, provide local storage, enforce security policies, and perform edge analytics. A cloud platform provides scalable storage, device management, dashboards, event processing, digital services, and integration through APIs. Cloud processing is useful but should not be assumed to be the correct location for every decision.

4. Sensing-to-Cloud Architecture

A complete IoT path usually contains the following stages:

1. Physical process: The environment or machine being observed or controlled.
2. Sensing and actuation: Sensors convert physical phenomena into signals; actuators convert commands into physical action.
3. Embedded processing: The node filters, calibrates, timestamps, compresses, or interprets data.
4. Connectivity: Wired or wireless communication transfers data and commands.
5. Gateway or edge layer: Local systems aggregate data, translate protocols, and make time-sensitive decisions.
6. Cloud or enterprise layer: Data is stored, analyzed, visualized, and integrated with business systems.
7. User and application layer: Operators, mobile applications, automation services, and decision-makers consume information.

Figure: Bidirectional sensing-to-cloud architecture, showing data moving upward and commands or updates moving downward.

The architecture is bidirectional. Measurements travel toward applications, while control commands, configuration, security credentials, and firmware updates travel back toward devices. Each boundary can introduce delay, data-format conversion, authentication requirements, and failure modes.

5. Layered Models of IoT

Layered models help engineers divide a complex system into responsibilities. A common five-layer interpretation is:

1. Perception layer: Sensors and actuators interact with the physical world.
2. Network layer: Connectivity and routing move information between components.
3. Processing or middleware layer: Devices, gateways, platforms, databases, and analytics process data.
4. Application layer: Services such as healthcare monitoring, smart agriculture, or industrial maintenance are delivered.
5. Business or management layer: Policies, business processes, governance, value assessment, and compliance are addressed.

Another practical model uses device, edge, platform, and application layers. These models are not competing physical standards; they are ways of assigning responsibilities. Security, identity, observability, and device management should be treated as cross-cutting functions across all layers rather than confined to one layer.

Communication choices

Communication selection depends on range, data rate, power, topology, cost, spectrum, and environmental conditions. Examples include:

• short-range protocols for personal-area devices;
• local wireless networks for buildings and homes;
• low-power wide-area networks for small messages over long distances;
• cellular networks for wide-area mobility and managed coverage; and
• wired industrial or Ethernet networks where reliability and power availability are important.

No protocol is universally best. A sensor transmitting a few bytes each hour has different requirements from a camera streaming high-resolution video.

6. Edge Computing and the Distribution of Intelligence

Edge computing places computation near the data source or user rather than sending every item to a distant cloud service. It may occur inside a sensor node, gateway, local server, vehicle, factory controller, or access network.

Typical edge operations include threshold detection, filtering, feature extraction, protocol translation, anomaly detection, local control, and temporary buffering during network loss. Edge processing can reduce latency, bandwidth use, cloud cost, and exposure of sensitive raw data.

Cloud computing remains valuable for large-scale storage, model training, fleet-wide comparison, dashboards, long-term trends, and centralized management. A strong architecture usually distributes functions rather than choosing edge or cloud exclusively.

For example, a vibration-monitoring node may calculate a small set of frequency features locally and send those features to the cloud. If an emergency threshold is exceeded, the gateway may stop a machine immediately without waiting for a cloud response. The cloud can still retain historical data and improve the detection model.

Edge computing introduces its own challenges: limited resources, difficult physical access, distributed software versions, local cybersecurity, and more complicated debugging. An edge device must continue safely when connectivity fails and must eventually reconcile locally generated data with central records.

7. Development Lifecycle for Embedded IoT Systems

A disciplined lifecycle reduces technical and operational risk.

1. Requirements and feasibility

Define what must be measured or controlled, acceptable accuracy, response time, operating environment, battery life, communication coverage, expected service life, cost, safety requirements, and regulatory obligations. Requirements should be testable. For example, 'the node shall report temperature every five minutes with an accuracy of ±0.5 °C over the specified operating range' is more useful than 'the node shall measure temperature accurately.'

2. Architecture and technology selection

Assign functions to the node, gateway, edge, cloud, and user application. Select sensors, processor, memory, communication technology, power source, enclosure, operating environment, and data model. Consider lifecycle availability because an IoT product may remain deployed for many years.

3. Hardware and firmware design

Create schematics, printed-circuit-board layouts, power designs, interface drivers, boot software, application tasks, and fault-handling mechanisms. Hardware and firmware must be developed together because sampling accuracy, timing, power behavior, and update capability depend on both.

4. Connectivity and platform integration

Define message formats, device identities, topics or endpoints, authentication, encryption, retries, time synchronization, buffering, and cloud interfaces. Avoid assuming that networks are always available or that messages are delivered exactly once.

5. Verification and validation

Verification asks whether the system was built according to its specifications. Validation asks whether it solves the real user problem. Testing should include unit, integration, hardware-in-the-loop, environmental, performance, security, usability, recovery, and field tests.

6. Deployment and operations

Provision devices securely, record identity and configuration, monitor health, manage certificates, collect logs, and control firmware updates. Production operation is part of the system design, not an activity added after development.

7. Maintenance and retirement

Plan for security patches, component shortages, calibration, battery replacement, data migration, customer support, decommissioning, and secure deletion. A device that remains connected after its support period becomes a security and governance liability.

8. Illustrative Case: Cold-Chain Monitoring

Background

A food distributor transports vaccines and temperature-sensitive products in refrigerated vehicles. Existing practice relies on manual temperature checks, which can miss short excursions and provide incomplete records.

Problem

The company needs continuous monitoring, rapid alerts, evidence for audits, and reliable operation in areas with intermittent cellular coverage. Battery life must cover a delivery cycle, and unauthorized personnel must not be able to alter records.

Analysis

A sensor node inside the storage compartment measures temperature at regular intervals. The node requires calibration, a timestamp, local storage, and a temperature range appropriate to the product. A cellular gateway can transmit data when coverage exists, but the system must buffer measurements during outages. Immediate local alarms are necessary because waiting for the cloud could allow spoilage.

Approach and solution

The node uses a low-power MCU, a calibrated digital temperature sensor, non-volatile memory, a battery, and a communication interface. Firmware samples the sensor, checks for implausible values, timestamps records, and stores them locally. The gateway forwards encrypted records to a cloud service. The cloud provides dashboards, alerts, audit records, device status, and reports. A local buzzer or vehicle display warns the driver when a limit is exceeded.

The architecture separates safety-related local action from long-term cloud analytics. Data is signed or authenticated, and firmware updates require authorization. Calibration records are associated with the device identity. When communication resumes, buffered records are uploaded with their original timestamps rather than being treated as new measurements.

Outcome

The system provides more complete evidence than manual checks, reduces response time, and supports analysis of recurring refrigeration faults. It also creates new responsibilities: batteries must be maintained, sensor calibration must be verified, alert rules must be reviewed, and access to temperature records must be controlled.

Learning points

1. The cloud is not a substitute for local control.
2. Intermittent connectivity must be an explicit design assumption.
3. Measurement quality depends on calibration, placement, sampling, and validation—not only on sensor resolution.
4. Security and auditability are requirements from the beginning.
5. Operational procedures determine whether a technically successful device creates real value.

9. Security, Privacy, Ethics, and Governance

An IoT device can be attacked through physical access, debug ports, firmware, radio communication, cloud APIs, mobile applications, or supply-chain components. Core practices include unique device identities, secure boot, signed firmware, encrypted communication, least-privilege access, protected credentials, disabled unused interfaces, input validation, logging, vulnerability management, and tested update procedures.

Privacy is relevant whenever devices collect information about people, homes, workplaces, health, movement, or behavior. Designers should collect only necessary data, explain its purpose, restrict access, define retention periods, and protect data in transit and at rest. Aggregation or anonymization may reduce risk, but it should not be assumed to guarantee anonymity.

IoT systems may affect physical safety. A faulty smart lock, medical monitor, industrial controller, or vehicle system can cause harm. Safety analysis must identify hazards, define fail-safe or fail-operational behavior where appropriate, and ensure that an unavailable network does not create an unsafe condition.

Ethical design also includes accessibility, non-discrimination in automated decisions, responsible disclosure of vulnerabilities, environmental impact, repairability, and responsible disposal of batteries and electronic components. Compliance requirements vary by country and application, so professional teams must identify applicable radio, electrical safety, cybersecurity, data-protection, and sector-specific rules.

10. Contemporary Developments and Best Practices

Modern embedded IoT systems increasingly use secure hardware elements, over-the-air updates, lightweight machine learning, digital twins, time-sensitive networking, and energy harvesting. Tiny machine-learning models can classify patterns locally, but model accuracy, bias, computational cost, update control, and explainability remain important engineering concerns.

Digital twins represent a physical asset, its state, and relevant historical information in software. They are useful for maintenance and simulation but depend on trustworthy sensing and well-defined data models. Energy harvesting can extend battery life in suitable environments, but available energy is variable and must not be treated as an unlimited power source.

Best practices include:

• begin with measurable requirements and explicit assumptions;
• design for failure, offline operation, and recovery;
• separate safety-critical decisions from non-critical analytics;
• minimize data collection and protect identities;
• maintain traceability between requirements, implementation, and tests;
• use reproducible builds, code review, static analysis, and automated testing;
• plan device provisioning, updates, monitoring, and retirement before deployment; and
• evaluate total lifecycle cost rather than only purchase price.

Chapter Summary

Embedded systems are dedicated computing systems integrated into larger products or processes. Their design is governed by constraints involving timing, energy, memory, cost, size, reliability, safety, security, and environmental conditions. Microcontrollers favor low-power control, microprocessors support more demanding operating systems and applications, and SoCs integrate multiple computing and communication functions. An IoT node combines physical interaction, embedded processing, and connectivity.

A sensing-to-cloud architecture moves information from a physical process through sensors, embedded computation, networks, edge systems, and cloud services. Commands, configuration, and updates move in the opposite direction. Layered models clarify responsibilities, while edge computing places selected processing near the source to reduce latency, bandwidth, and dependence on connectivity. Effective IoT development covers requirements, architecture, hardware and firmware, connectivity, testing, deployment, maintenance, and retirement. Security, privacy, safety, ethics, and governance are cross-cutting responsibilities throughout the lifecycle.

Key Terms

Embedded system: A computing system integrated into a larger product or process.

Firmware: Hardware-oriented software stored in non-volatile memory.

Real-time system: A system whose correctness includes meeting timing requirements.

Microcontroller: An integrated processor, memory, and peripheral controller intended for embedded control.

Microprocessor: A processing unit generally requiring external memory and supporting components.

System-on-chip: An integrated chip containing multiple major computing and peripheral functions.

IoT node: A deployed endpoint that senses, processes, communicates, or actuates.

Sensor: A device that converts a physical quantity into a usable signal.

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

Gateway: A system that connects local devices to other networks or services.

Edge computing: Processing performed near the data source or user.

Cloud platform: Centralized or distributed services for storage, analytics, applications, and management.

RTOS: An operating system designed to provide predictable timing and task services.

Interrupt: A hardware or software event that requests processor attention.

Watchdog timer: A mechanism that detects stalled software and initiates recovery.

Bootloader: Initial software that starts a device and may support updates.

Telemetry: Measurement data transmitted from a device.

Over-the-air update: Remote delivery and installation of device software.

Secure boot: A startup process that verifies the authenticity or integrity of software.

Digital twin: A software representation of a physical asset or process.

Review Questions

Short Answer Questions

1. What is an embedded system, and how does it differ from a general-purpose computer?
2. Define microcontroller, microprocessor, and system-on-chip.
3. Why is a watchdog timer useful in an IoT node?
4. What is the purpose of an IoT gateway?
5. State two advantages and two challenges of edge computing.

Long Answer Questions

1. Explain the major constraints that influence embedded-system design.
2. Describe a sensing-to-cloud architecture, including the reverse path for commands and updates.
3. Compare MCUs, MPUs, SoCs, and IoT nodes using suitable applications.
4. Explain the stages of an embedded-IoT development lifecycle.
5. Discuss security, privacy, safety, and governance requirements in IoT systems.

Analytical / Application Questions

1. Design an architecture for a battery-powered agricultural soil-monitoring system. Justify the processor, communication method, edge functions, cloud functions, and power strategy.
2. A factory machine must stop within a short and predictable time when excessive vibration is detected. Decide which functions should occur locally and which may occur in the cloud. Explain your reasoning.
3. A remote monitoring device has a 2,000 mAh battery and an estimated average current of 10 mA. Using the ideal relationship T_hours ≈ Q / I_avg, estimate operating time in hours and explain why the actual value may differ.

Multiple Choice Questions with Answers

Question 1

Which characteristic most clearly distinguishes an embedded system from a general-purpose computer?

A. It is designed for a defined function within a larger system.

B. It always uses a graphical user interface.

C. It must use cloud storage.

D. It cannot contain a processor.

Correct Answer: A. It is designed for a defined function within a larger system.

Question 2

Which component is most directly responsible for converting a physical temperature into an electrical or digital representation?

A. Actuator

B. Sensor

C. Bootloader

D. Gateway

Correct Answer: B. Sensor

Question 3

Which platform is generally most suitable for a low-power sensor node with modest computation and direct peripheral control?

A. Microcontroller

B. Desktop workstation

C. Cloud database

D. High-end server

Correct Answer: A. Microcontroller

Question 4

What is a primary purpose of an IoT gateway?

A. To replace every sensor in the system

B. To connect local devices with other networks or services

C. To eliminate the need for device identity

D. To guarantee that batteries never discharge

Correct Answer: B. To connect local devices with other networks or services

Question 5

Which operation is most appropriately performed at the edge when an immediate response is required?

A. Long-term fleet-wide trend analysis

B. Archiving years of historical data

C. Stopping a machine after a dangerous local condition is detected

D. Training a large cloud model using all company data

Correct Answer: C. Stopping a machine after a dangerous local condition is detected

Question 6

In a hard real-time system, missing a deadline can be:

A. Unimportant in all circumstances

B. Acceptable only when the cloud is available

C. Unacceptable or dangerous

D. Required for correct operation

Correct Answer: C. Unacceptable or dangerous

Question 7

An ideal battery-life estimate uses Q = 2,000 mAh and I_avg = 10 mA. What is the estimated operating time?

A. 20 hours

B. 200 hours

C. 2,000 hours

D. 20,000 hours

Correct Answer: B. 200 hours

Question 8

Which practice best supports secure firmware updating?

A. Accepting any firmware image received over the network

B. Using signed updates and authorized installation

C. Disabling all device logging

D. Sharing one permanent password across all devices

Correct Answer: B. Using signed updates and authorized installation

Question 9

Which statement about layered IoT models is correct?

A. Security can be ignored because it belongs only to the cloud layer.

B. Layers help assign responsibilities across a complex system.

C. Every IoT product must use exactly one universal layer model.

D. The application layer must contain every sensor driver.

Correct Answer: B. Layers help assign responsibilities across a complex system.

Question 10

During the development lifecycle, validation primarily asks whether:

A. The system solves the intended real-world problem

B. The source code contains no comments

C. The processor has the highest possible clock speed

D. The device uses the largest available memory

Correct Answer: A. The system solves the intended real-world problem