Availability as an IoT driver
This article explains what happens when data from electronics become part of a supply chain, and what you need to take into account when rolling out large wireless networks in an IoT context.
IoT is developing at a rapid pace, and many businesses are working hard to get their own systems in operation in Denmark. In recent years, we have seen a lot of pilot projects, but when it comes to putting these systems to use, there are new challenges we have to face, compared to those involved in creating a proof-of-concept. Normally, proof-of-concept projects take place on a smaller scale and in more controlled environments. When rolling out IoT systems on a large scale, the systems are challenged in the form of breakdowns and the resulting lack of data, due to the environment, network errors, differences in installation, variation in sensor data, and the system's overall integrity. In other words: the availability of a function.
In this article, we focus on what happens when data from electronics become part of a supply chain, and what you need to take into account when rolling out large wireless networks.
The supply chainMost electronics businesses still make their money on selling products. But as businesses continue to adopt IoT, more and more of them are considering service-based models, where instead of selling a product, they sell the functionality itself. An example of this would be lorry tracking. In this case, it is important to the transport company that data regarding the whereabouts of each lorry is delivered reliably. Transport companies don't usually have the resources required to administer and run a data network and perform device maintenance. For that reason, they often choose to work with a system supplier, which they will evaluate based on the service provided—that is, reliable delivery of data from the devices. This makes the electronics and the network part of the supply chain required to deliver the service. Therefore, the reliability of these components must be taken into account in managing the supply chain.
Wireless coverageIn connection with all this, Sigfox (a wireless network devoted to IoT) has introduced a certification program. Producers that supply electronics for the network are evaluated and tested based on various parameters relevant to ensuring that these devices can reliably deliver data to end users. One of these parameters is the antenna implementation. Devices are tested to find out how good their antennae are, since this directly influences the coverage level experienced by end users.
In figure 2, you can see the four classes into which devices are sorted based on the test. This can be directly linked to the actual level of coverage, revealing the risk probability that a tested device will lose its connection, with a deterioration of the service experienced by the end user.
In figure 3a & 3b, you can see Sigfox's coverage map for Denmark. The map to the left shows that Denmark has nearly 100% coverage, provided that producers' developers have ensured a good antenna performance and achieve class 1U. By contrast, if the antenna integration is not up to par, gaps in coverage will occur in areas of the country, as shown by the map to the right.
Statistical analysis requiredIt is essential to realise that in going from proof-of-concept (POC) to large-scale rollouts as an IoT developer, producers must evaluate their devices statistically. In POC projects, we often work with demonstrating that coverage exists and is sufficient for the system to fulfil its purpose. On the other hand, when we roll out thousands of devices throughout the country for long-term use, it becomes essential to take a statistical approach to considering coverage. Most often, a normal distribution can be used to evaluate the quality of signal strength nationwide - and by extension, the quality of the service delivered (see example in figure 4).
For an example of this, we can look at Svebølle's experience, where SEAS has created the first "smart village" in Scandinavia to test out IoT technology in the real world. They installed sensors for parking, pedestrian monitoring, street lighting, and smart meters for gas, electricity, water, and heating. Over a period of about nine months, FORCE Technology has been monitoring LoRA-based wireless communications as part of the Energy Bigger project.
To prevent FORCE Technology from having access to potentially sensitive data, encryption keys were not exchanged. In this manner, the data contained in the messages could not be read—only timestamps and received signal strengths were available. It is thus still possible to draw conclusions about the radio network itself based on the slightly less than 700,000 packets received during that time. Looking at just a single device (see figure 5a & 5b), it is possible to see that the mean received power is –125 dBm, with a standard deviation of 5.5 dB. This means that there is a 2.5% probability for the power to be less than –136 dBm, the sensitivity of a LoRA radio. In other words, there is a 2.5% probability that the data from the device will not be received in the relevant transmission.
It is essential to understand that, if the antennae are not implemented correctly, the mean value will be lower, thereby increasing the probability of packet loss. For example, if 5.5 dB is lost due to a bad antenna, the probability rises to 16% instead of 2.5%, though the device is installed in the same location. This is why FORCE Technology, working in the IoT & Wireless Club together with Anritsu and Bluetest, have constructed a testing facility to measure the performance of IoT devices in terms of antenna sensitivity.
Classic skillsIt is essential to understand that this approach is relevant not only to wireless coverage, but also other parameters that these devices are subjected to. The more devices in operation and the more time they will spend out in the field, the more extremes they will be exposed to. These might include temperature changes, environmental humidity, the magnitude of voltage spikes in the power supply, and vibrations in installation locations, etc.
For example, if a network of 10,000 devices is installed to be in operation for 10 years, they will go through a combined total of 100,000 years of operation — that is, about 850 million hours. Even though the probability of a device being exposed to 80C each hour in an industrial environment is only one in one million, statistically speaking, 850 devices will be exposed to these conditions throughout the lifetime of the network. If they must be replaced by a service technician for (e.g.) 650 DKK per hour plus mileage, this task could easily take three hours, costing 2000 DKK in repairs. Over the lifetime of the entire network, this adds up to 1.7 million DKK. All of this is based on a probability that is virtually impossible to predict in a proof-of-concept project.
Thus, when rolling out IoT systems on a large scale, it is essential to consider the probabilities that come into play when many devices are in operation over a long period of time, to ensure the availability of their data. This is directly related to the services that businesses can deliver when they sell both services and products.