Moving beyond connectivity

6 mins read

When it comes to designing applications for the Internet of Things (IoT) engineers have tended to focus on the issue of connectivity and the reliable sending and receiving of data via a connected network.

Whether it’s for a factory, a smart building or a consumer device, all of them need to be connected to other devices using either a central or network of servers.

Devices use the IoT to share data and are constantly sending information about their status, actions, and the like and that data is then used automatically to make decisions and to monitor devices and systems.

The IoT has dramatically over the past few years and devices are now benefitting from advances in Artificial Intelligence (AI) and Machine Learning, as well as from developments in 5G.

So, when it comes to developing an IoT application what does an engineer need to take into account so as to develop a suitable device that meets the needs of its users?

From the cost of new hardware or building the necessary hardware to addressing security issues or simply obtaining the right hardware that’s capable of meeting future requirements the list is certainly a long one.

According to Ross Sabolcik, VP, GM Industrial and Commercial IoT Products, at Silicon Labs, “Today it’s never been easier to connect devices capable of pumping data to the cloud – whether that’s door sensors or smartphones – but now, perhaps more so than in the recent past, there needs to be a solid and valid business case when developing for the IoT.

“Take commercial lighting. While a customer may be looking to retrofit a property to save money through energy saving, the IoT can also deliver much improved and more accurate lighting control which will help to ramp up the return on investment.”

Where vendors may be pushing better energy control by incorporating location services it will also be possible to track people and ‘things’ as they move through and around buildings.

“You may have a primary ROI but that can help drive other services that the IoT then makes possible,” suggested Sabolcik.

“Another good example of how an IoT solution can deliver more than was originally intended is with electronic shelf labels. Initially deployed to do low value work – automating price changes, for example – they can also be used to deliver a ‘click-to-cart’ service and connected to a shopper application can be used to trigger flashing labels to help customers identify products and offers - driving more value in the process.”

According to Sabolcik, “If you roll the clock back, just five years, the challenges were not about adding additional value but were rather all about connectivity. There was a lack of widely available open networks, and the networks that did exist tended to be both proprietary and limited, in many cases they were one-off solutions.

“Today, we have open standards and connecting devices has been made far easier. What that means is that developers are freed up to focus on where they can really add value by deploying IoT devices.”

That shift has meant, according to Sabolcik, that rather than the focus being on the technology it is now more about how best to leverage an IoT network to get the most value.

“Standards have evolved rapidly and the latest – The Matter Standard – is a huge leap forward. It simplifies devices and their use and delivers on what customers want – devices that simply work, without having to worry about specific protocols.”

Making life easier

Design starts with the use case, according to Sabolcik. “Whether a device is connected to the cloud, or not, is no longer important. The question that needs to be asked, and answered, is ‘how does this make life simpler or more secure’. That has to be at the heart of any new project.”

Once that question has been answered then the designer can start to look at the device and how they will design it – and those decisions should be based on what the device is looking to achieve.

 “Everything will be driven by that,” according to Sabolcik. “Take battery life. Is the solution being developed going to be designed to optimise power and provide an extended lifetime when it comes to the battery? What levels of power will be required, and will the device be upgradeable?”

Another issue is security. IoT devices have tended to run older operating systems that cannot be patched, or that use weak or default passwords, and in many cases are not monitored for security breaches. Consequently, IoT devices are often used as an entry point into a network to gain access to data that they can then exploit through malware and ransomware.

It has never been more important for IoT device developers to prioritise cybersecurity and give customers enterprise grade security and, in the process, peace of mind.

“Data is easy so why shouldn’t security be easy too,” said Sabolcik. “Again, it’s all about what the design is intended to do. There are certainly a range of security concerns when it comes to the IoT, and they will be influenced by the end application. If you take smart meters, for example, there is a worry about security and techniques are being adopted, such as validating firmware and locking down debug ports, that are now ‘must haves’ when it comes to meter solutions.

“When you turn to the consumer space, security can be viewed more as a continuum, from securing the device to securing the data in the cloud. But however it’s viewed, semiconductor suppliers are now building security into their devices.”

While connectivity may no longer be the issue it was a few years ago design engineers still have to consider whether to use Wi-Fi, Bluetooth or cellular platforms.

“There’s certainly plenty of choice,” Sabolcik conceded. “So, you need to consider what the ecosystem is that you want to play in. If you’re looking at the smart home market then you’ll be looking at the Matter ecosystem, and if you’re looking at smart meters then you’ll have to look at a cellular platform.”

“Power and data rates determine and ultimately drive the wireless technology that you will look to use,” added Sabolcik.

With the rise of the edge the IoT is now seeing more applications that look to use artificial intelligence and machine learning.

“Wake words, for example, are all done at the edge and more work is being conducted into carrying out the processing in the end device.

“So, we’re looking at audio processing, occupancy sensors and motion sensors and low-resolution video applications that can be used to count people, along with predictive maintenance.

“Customers are now taking tools and looking to use AI and ML in interesting ways.”

AI and ML

According to Brian Rodrigues, a Senior Field Applications Engineer at Silicon Labs more and more artificial intelligence is now migrating to data acquisition applications - especially constrained applications running on functionally restricted microcontrollers (MCUs) as Sabolcik described above. Consequently, it’s important that building and deploying edge and embedded AI systems is made simpler, reducing risk, time to market and accelerating adoption.

“This is especially true in the rapidly growing IoT market,” Rodrigues explained.

To address this, Silicon Labs has developed a wireless MCU that supports IoT applications with hardware-accelerated AI, based upon an embedded matrix-vector processor (MVP) – the EFR32xG24.

“Systems use ML to improve responses and make predictions through repetitive training based upon a ‘model’,” explained Rodrigues. “The model is a decision-making algorithm that requires training prior to use – and this training can be ongoing. Training is completed using existing datasets, by collecting data or by a combination of both.”

This once required powerful computers, but edge computing has made it possible for inference to now be run on more constrained devices.

In terms of the IoT edge computing, which is often based on devices such as ARM’s Cortex-M MCUs, can be found in basic sensors or actuators including lighting, thermostats, door sensors, meters and many other similar applications.

According to Rodrigues there are benefits associated with edge computing, most importantly it is not reliant on any other entity and so allows for localised decision-making.

“From a practical perspective this gives designers benefits such as a faster response times – data is not transferred to the cloud, decisions appear in real-time on the device and there’s reduced Internet bandwidth – sensors generate huge amounts of data that consume bandwidth, even if there is no valuable information in the data. Edge computing saves bandwidth and cost and local data analysis requires less power than transmitting data, which is a significant benefit in IoT systems that are often battery-powered,” said Rodrigues.

The EFR32Xg24 is also a hardware accelerator for ML models inference – or an MVP. This allows ML inferences to be run efficiently, often with a six-fold reduction in power and a doubling or even quadrupling of speed when compared to the commonly used ARM Cortex-M (without hardware acceleration).

“This a relatively simple MCU, so it only addresses a subset of the use cases AI/ML might require,” added Rodrigues. “Principally, it addresses four categories that include real-life applications: sensor signal processing; audio pattern matching, voice commands and low-Resolution vision.”

There are two primary steps to be taken when creating an ML-enabled application. The first involves the creation of a wireless application, usually based upon a protocol such as Zigbee, BLE, Matter, or any proprietary 2.4 GHz protocol-based application – even one which is non-connected. The next requires the building of an ML model that is then integrated with the application.

Silicon Labs has generated a number of sample applications based upon the TensorFlow AI/ML framework, which is Google’s end-to-end open-source platform for ML and which offers a comprehensive ecosystem of tools, libraries, and community resources.

“At the end of the day ML engineers need to focus on the data sets that they’re looking to train against,” said Sabolcik. “Have you been collecting data and do you have the necessary data sets? If you’re looking to use AI/ML it starts with the data, every time.

“Finally, if you want to use AI/ML when it comes to the IoT you need to ask yourself: “What type of AI developer am I? Do you simply need the tools to advance your project, or do you need help in understanding how AI/ML will benefit your device?”

Either way there are plenty of solutions and answers out there when it comes to developing IoT enabled products.