IoT projects and AI at the edge

Paul Tofan
Paul Tofan
21 Feb 2020

With IoT becoming a global phenomenon, the interest in 5G and low-power wide-area networks known (aka LPWAN) is growing exponentially. They both cover long distances and are expected to drive IoT in various applications from smart factories to smart cities. While the rollout of the 5G infrastructure is still going on, there are solutions that do not need the ultra-high-speed and can benefit from solutions built on top of LPWAN.

So we went to #TheThingsConference on 30-31 Jan in Amsterdam to find out more.
Read along to find out more about: Lora as the connectivity leader in LPWAN, it's the application in IoT projects and where you could start from, how tinyML and AI at the edge can make a difference for you.

Read along to find out more about:

  • Lora as the connectivity leader in LPWAN
  • it's application in IoT projects
  • where and how to start your IoT prototype
  • how tinyML and AI at the edge can make a difference for you

The Things Conference, Amsterdam 2020

The Things Network (aka TTN) started in NL and is focused on enabling low power Devices to use long-range Gateways to connect to an open-source, decentralized network to exchange data with Applications. The network is available across the globe and it uses LoRaWAN as the secure messaging protocol used by the network. LoRaWAN is the most used protocol in the area of LPWAN technology for IoT projects. We'll dive into details later and see how this fits into the spectrum with the other connectivity options that IoT projects have.

From the beginning, this felt like an event that had the capacity of bringing together different tribes, from hardware manufacturers to embedded programmers to software studios and business consultants. The common worldview that was shared across these tribes was that the moment is now to leverage the ability to collect data from everything and extract information to drive better decision making.

The opening keynote, "Unlocking the economies of scale for LoRaWAN deployments", did a great job to create an authentic passionate vibe that lasted throughout the entire 2 days duration of the conference.

Dark side of the moon

When it comes to IoT projects, the consensus among the consultants involved in the event was that about 60% of the deployments fail within a year. It seems that success in IoT is for those that have seen the graveyard of failed proof of concepts.

There were 2 main points brought up as differentiators for successful projects:

  • The strength of the business case
  • The easy access to the solution

When it comes to projects, usually the technological aspect is strong. Sometimes a R&D project becomes so compelling and potentially innovative that immediately is adopted as a full-blown productive solution and scaled at maximum speed. What can happen is that it does not scale well, in the sense where you would want to scale mostly the profit and not the cost.

Data first, Things after

This particularly can be the case in IoT projects where data collection starts to grow, storage cost starts to grow, and everything exponentially (imagine 1 petabyte per day) - if you're not managing and expiring your data this is a sure method to fail.

It is recommended thus to have a data-first, things after approach. To break down this actionable data-driven approach:

  • Start with the answers you are looking for
  • Move to the data that can be processed to give you those answers
  • Deploy the devices that can collect that data
  • Manage the data after collection, have an expiration policy

Because things and their data is expected to grow, it will become unmanageable without it.

Easy access

Because of the complexity of the IoT chain, it is no longer sufficient to just have good technology in place.

The various players in the ecosystem need to learn to partner and leverage each other's strengths to create value on top of the technology.

Typically you can make a breakdown of an IOT LPWAN based solution to four layers:

  • Sensors/hardware - generically named things
  • Connectivity - many protocols, many networks
  • Data storage and analysis
  • Visualization and application, or the actionable aspect of the data

Divergent connectivity, the main concern of IoT projects

Typically the main benefits targeted by an IoT project are cost savings and production process improvements, consolidated by the future possibility to use collected data for new revenue generation. Therefore it comes as no surprise when among the challenges and concerns regarding the development of IoT projects, connectivity seems to be the most difficult decision.

According to the research done by Libelium, it seems that Wifi, Ethernet and Cellular protocols are preferred for IoT projects followed closely by LoRaWAN, the most used LPWAN technology. You can read more details in the research from http://www.libelium.com/iot-survey-key-concerns-and-barriers-to-develop-successful-projects/.

Some practical observations on the distinctions between LPWAN protocols and how to take them into account when making a business case "LPWAN Is Not a Red Ocean Market - Paul Pinault (Disk91)"

Why LoRaWAN ?

The reasons why LoRaWAN is so much in focus when it comes to IoT solutions is that it comes with certain advantages that the other connectivity options don't have, like:

  • extended range (15km outside, 10km inside city - around one radio gateway range)
  • Free to use (unlicensed frequency band -no use permit is required)
  • Scalability (up to 20.000 end sensors per gateway)
  • Low cost of end devices/sensors and gateways
  • Low power consumptions (sensors can run on battery for about 4-5years)
  • Most of the solutions built on this technology in areas of industry, agriculture, connected cities, and water management.

Getting started with it

One of the best ways to learn about technology remains hands-on. It is, therefore, a great help to have a well supported open community like the one around TTN with very good coverage throughout Europe.

All you need to get started is a Lora compatible device that acts as the data collector and a Lora gateway that extends the service of TTN to your device so you can push your data into the network server. After you register an application to the network server and register the devices to it. you are ready to go prototyping and validate your potential use cases. I wholeheartedly recommend the documentation on https://www.thethingsnetwork.org/docs/ and the webinars from "How To: The Things Network Console Introduction"

Open-source and enterprise

In case you want to set up your own customized infrastructure, TTN provides also an open-source stack (The Things Stack V3) which is a fully-featured LoRaWAN network server with management for gateways, applications, devices, and users. You can integrate the stack in solutions deployed in the cloud or on-premises through the open APIs.

Another popular open-source solution for LoRaWAN Network Server is https://www.chirpstack.io which provides open-source components for LoRaWAN networks and can be used in commercial projects.

When it comes to enterprise versions, The Things Enterprise Stack (TTES) was just announced at the conference.

Another popular enterprise solution is https://www.loriot.io.

Connectivity from space

I mentioned that connectivity is a top issue in IoT, so the discussion about supporting connectivity through the use of satellites was something really interesting. Even if this is only going to become operational end of 2020, it will be available as a free service initially and taking into account that dev kits will be available from July - I can imagine a lot of interesting solutions for remote global areas and use cases where you don't have the advantage of an existent physical network (or you don't want to deploy one).

AI at the edge

Artificial intelligence promises particular benefits for industrial processes, imagine the impact of machines that can react to contextual information or predict needed maintenance before the failure of the system occurs. In some cases, this intelligence is spread across the network and the device at the edge is merely the collection or the interaction point. But in many cases, pre-processing of data can be done on the device saving the costs of transmitting (especially in LoRaWAN where bandwidth is limited) and potentially offering a light-touch privacy model.

Enter TinyML, a collaborative effort that broadly encapsulates the field of machine learning technologies capable of performing on-device analytics of sensor data at extremely low power. One of the developer platforms provides a TinyML pipeline where developers can contribute to and extend both algorithms and target device support.
Find out what a sheep has to do with it in an amazing presentation from their CTO: "Adding Intelligence to Your LoRaWAN Devices - Jan Jongboom (Edge Impulse)"

And a bonus, if you need to do radio planning and deploy your gateway properly for an optimal link budget - drop me a note (and thank professor Remko Welling who taught the class in 60 perfect minutes!).

This was a great conference in terms of well-prepared content, innovative ideas and creating just the right space to get inspired and develop practical skills.

Some personal ratings for #TheThingsConference :

  • Interesting: 5/5
  • Usefulness or applicability: 4/5
  • Fun factor: 5/5
  • Schmooz factor: 4/5
More fxBITS articles