📘 Resource Overview & Documentation Details

You are viewing the technical directory listing for MATLAB IoT Development for Embedded Applications. This engineering resource is curated under our IoT Sensors division to support automation specialists, control systems engineers, and manufacturing technicians looking for verified reference materials.

🔍 Contextual Abstract & Technical Scope

Share Summary:LinkedInEmail

MATLAB IoT development can span more than cloud analytics and connected sensors. For industrial and consumer device teams, it can also provide a way to move algorithms from desktop prototyping to embedded hardware while managing constraints such as battery life, intermittent connectivity and limited bandwidth. The experience of engineering company Skyrad illustrates how these requirements can shape an IoT architecture over time.

Rather than relying on one standard cloud-to-edge pattern, Skyrad used different approaches for applications including leak detection, heating diagnostics and in-home thermal tracking. The common thread was developing and refining algorithms in MATLAB and Simulink, then choosing where those algorithms should run based on the needs of each device.

Why IoT architecture must fit the device

A common IoT architecture assumes that an edge device runs an operating system, has a continuous internet connection and sends data to cloud services for analysis. Cloud platforms can provide useful infrastructure, but that pattern is not automatically suitable for every application.

A battery-powered device may need to spend most of its time offline to conserve energy. A sensor may have only a narrow communications window, or the product may need to continue working when a network connection is unavailable. In those cases, developers must decide what to process locally, what to transmit and how to update the algorithm without putting too much at risk.

Skyrad’s projects show why those decisions are application-specific. Connectivity can support remote data access and algorithm updates, but local processing can reduce dependence on a live connection. The balance may also change as a product moves from proof of concept to production.

MATLAB IoT development in the LeakBot project

LeakBot estimates water flow by using the temperature difference between incoming water and the surrounding environment. Sustained flow over time can indicate a leak, although other events, such as a dripping tap or hose, may produce flow as well. The project’s stated detection performance was 98% of leaks above 5 ml/min.

The first version was a connected proof of concept. Connectivity made it possible to develop algorithms remotely and gather data. After the concept was established, the product requirements shifted: the device needed to operate on batteries, and a continuous connection was not considered essential.

Starting with standalone operation

The first production version worked as a standalone leak detector. It had no connectivity, was designed for a five-year battery life and beeped when it detected a leak. The MATLAB algorithm developed for the proof of concept was hand-coded in C for embedded hardware.

The device also logged its data. Plumbers could download those logs and return them to the development team, providing a way to examine classification errors and inform future code changes without requiring the device to maintain a network connection.

Adding connectivity without sending everything

In a later version, customer demand for data led to the addition of Sigfox connectivity. With a reported limit of 12 bytes per day, the link could not carry the full time series. Instead, the device pre-processed the data and sent a compressed representation to the cloud.

This architecture supported remote tuning despite the limited bandwidth. Initial cloud rules ran in C, while more advanced MATLAB algorithms ran on MPS. Empirical and machine-learning approaches could be compared in the cloud. The example demonstrates a useful design principle: constrained communications may require local data reduction, with selected analysis or tuning performed remotely.

A subsequent Wi-Fi version faced a different trade-off. Wi-Fi offered higher bandwidth but used more power, so it could be switched on for only a few seconds per day. Fast leaks, such as burst pipes, were detected locally. Other data was buffered and sent to cloud MPS every 24 hours. Because MPS ran an already proven and tested algorithm, enhancements could be checked against device logs before being moved into production with minimal revalidation.

Choosing where algorithms run

Skyrad’s HeatDoctor application monitors data from smart thermostats, such as Nest or Tado, to identify central-heating faults. Its initial architecture used PHP on an Apache web server to call a MATLAB algorithm running on MPS. When scaling the project, AWS Lambda was selected because its benefits outweighed the time required to recode the algorithm for deployment.

This choice highlights that a consistent development tool does not mean every production algorithm must run in the same environment. Teams can use MATLAB for prototyping and algorithm development, then assess deployment options against scaling needs and implementation effort.

Designing for offline operation and faster response

Project Pippa, an in-home target-tracking application using a thermal imaging sensor, brought together several lessons from LeakBot. The device still needed to run on batteries and use connectivity, but it also had to respond more quickly and remain useful when disconnected. The team wanted to refine algorithms offline and deploy them with limited retesting.

An initial proof of concept used a Raspberry Pi sensor connected over Wi-Fi to an AWS database. A Simulink algorithm running in desktop MATLAB analyzed the data and wrote results to an AWS table; the Pi read those results and provided feedback to the user. The setup gathered data in around 10 homes over 12 months.

For the final architecture, the algorithm was refined using logged data and moved onto the embedded device. Outputs were sent by SMS, using cellular connectivity while avoiding a requirement for a continuously active Wi-Fi connection. Embedded Coder generated C from the Simulink model. That code could be compiled for the Raspberry Pi during user testing, the intended low-power hardware and the associated Android and iOS app.

Compiling for the Raspberry Pi allowed users to experience the device while final hardware was still being developed. The same algorithm could then be compiled for the target device, with memory use optimized for that hardware. Reusing the model and generated code was intended to reduce the need to revalidate the algorithm across these deployment targets.

Practical implications for automation teams

These projects offer several considerations for engineers and system integrators planning connected products:

  • Set requirements before selecting an architecture. Battery life, response time, connectivity and offline behavior affect where processing should take place.
  • Use local processing where the application requires it. A device can detect urgent events locally while buffering less time-sensitive data for later transmission.
  • Design around communication limits. Pre-processing and compression can help when a device has a low-bandwidth link.
  • Plan the path from prototype to deployment. Logged data, embedded code generation and testing against target platforms can connect algorithm development with production requirements.
  • Choose cloud services pragmatically. Cloud infrastructure can support analysis or scaling, but the deployment environment should match the project’s operational needs.

MATLAB and Simulink provided Skyrad with a workflow for developing algorithms and deploying them across desktop, cloud, mobile and embedded environments. The practical value was not a single IoT architecture, but the ability to adapt deployment to each device while maintaining a common algorithm development process. For teams building connected equipment, that makes power, connectivity and validation requirements central design inputs—not details to address after the prototype is complete.

Access Official Technical PDF