IoT, geolocation and SX1280 ranging : the continuation of the experiment conducted by WI6LABS
The WI6LABS teams have been experimenting for several months a new geolocation technology. We tell you everything about sx1280 ranging. You can find the first part of the experimentation here.
2 years ago, WI6LABS was experimenting with the localization technology offered by Semtech’s SX1280 ranging.
WI6LABS continued the work and developed a complete localization system. We will tell you the rest of the story!
The benefit for the customers
Thanks to our localization solution, it is now possible for an industrial company to know the position of its assets within its environment (warehouse, factory, tertiary building, parking,…) thanks to anchors and 1 tag. The infrastructure allows to manage several buildings around the world. A lost tool, a car not found… Everything is visible on the web application interface.
The solution is deployed in sectors such as aeronautics, agriculture and of course industry.
Thanks to this geolocation technology, we can use asset location data for tracking, geofencing, storage time measurement etc…
But the tag can be useful for purposes other than geolocation: the use of 2.4GHz allows us to consider high data rates to transfer large volumes of data and be reachable by the network at any time. This is an advantage for medium distance networks (500m to 1km) compared to longer range technologies. The network’s 6LoWPAN architecture allows us to cover a large number of applications.
Some elements of comparison
Compared to other 2.4GHz technologies, the SX1280 ranging offers impressive capabilities. It allows to have a speed at least equivalent to its competitors, while ensuring a much better range.
In addition, we note that the sx1280 offers a very good compromise between range and accuracy. It is also suitable for IoT due to its low consumption and low deployment cost.
And on the user side, what does it look like?
The management of this platform is ensured by WI6LABS and its partner Alkante: Maplink is a sensor-based information exploitation solution that collects, analyzes and shares geolocated data.
On the user interface (backend), we can place anchors and reference the sensor fleet. We provide the plans of industrial buildings, the names of the sensors, their number of ranges per day, the name of the border router that manages the anchors, the tag field etc….
All this then applies to the web application interface for the end user
Anchors and tags: We explain everything to you!
The sensors to be geolocated are tags. They are located in areas covered by anchors and a border router.
The tag, whose position is unknown, measures the distances between it and the different anchors and transmits this information to the border router. The latter pushes this data to the geolocation server. This entity is in charge of applying “trilateration” algorithms to determine the positioning of the tag. The processed data is then provided to the application server.
The bonus brought by the application server
In addition to performing trilateration, the application server integrates several optional Machine Learning (ML) models. The addition of learning allows for a significant improvement in measurement accuracy.
First, the model has to learn. In order to collect new data in quantity, it is therefore necessary to install training tags: a training tag has the same function as a tag, but its position is known. A tag and a training tag perform the same ranging position. See below how the ML is integrated into the project :
To cover an area of 10,000m² of an industrial building, a minimum of 5 anchors is required to avoid any loss of measurement. If the building is complex, it will be preferable to densify the network. This depends essentially on the structure of the building.
Accuracy can be up to 2 meters: the more points recorded for the learning machine, the more accurate the measurement will be. It strongly depends on the environment in which you are operating.
It is also possible to keep reference tags in position that will allow the ML algorithm to adjust according to changes in the environment.
A few months ago, the tests carried out at WI6LABS gave us an accuracy of about 2m with respect to the real point. In 2020, new tests have been carried out and now give an accuracy of 1m. These tests could be carried out thanks to the ML model which allows us to find correction factors much more precise than before. To obtain a ML model, it is necessary to have a large database, retrieved thanks to the training tags. The use of Artificial Intelligence (AI) makes it possible to provide all this input data.
Here are the different results observed: we can observe an improvement of the correction tables thanks to the Machine Learning model :
In more complex buildings, we are subject to disturbances such as reflections and we have to take into account cases of multi-paths. We have therefore worked on setting up algorithms to compensate for these difficulties related to the terrain, while ensuring that the number of anchors required for geolocation is kept to a minimum. The algorithm automatically adapts to its environment.
Today, in a very constrained industrial environment, our measurement accuracy is on average 5m.
The construction of a Machine Learning model is a solution for multi-path cases. Its use improves the accuracy of the system. Thanks to it, we gain stability in the data and the geolocation of the tags is improved.
Unlike our competitors who need a lot of anchors, our promise is to install the minimum number of anchors, i.e. 5, so that the cost of the infrastructure is as low as possible. Our primary goal is to limit the complexity of the infrastructure.
The expertise of our development team
The team worked hard on the development and development of the solution (tag, anchor, border router, localization solver, web application).
Software integration, with a 6LoWPAN stack for example, makes it an easy solution to implement and integrate into an existing system.
Developments on localization floors allow us to further increase accuracy by adapting to working environments.
Also, very significant efforts have been made in the design and integration of the antennas. It is a critical design element that guarantees the accuracy of the system