One of the best demos I’ve seen at the Qualcomm 5G Summit is the 5G Positioning Evolution which uses a combination of 5G signal and Artificial Intelligence to yield a device’s positioning with a 1-feet precision.
As you may know, consumer-level indoor positioning has been in the works for over a decade. However, you still don’t have access to anything remotely practical such as in-mall positioning services.
There are several reasons for that. Most indoor positioning systems require specific infrastructure (which no one wants) and use wireless protocols that were not yet standard. A direct line of sight between the emitter and receiver can be lost, resulting in poor position accuracy. Finally, the signal can pass through objects like trees that interfere with the signal.
Qualcomm’s approach to 5G positioning is two-fold. First, 5G positioning uses the wireless communications infrastructure. It’s already there and not installed solely for location services.
It works by having your phone (or some other 5G device) send a very simple 5G signal to nearby 5G access points with a time stamp. The amount of data transmitted is tiny and uses virtually no bandwidth.
Each access point receives the signal at a slightly different time, and from there, it’s possible to triangulate the emitter’s position. It’s the same principle as the GPS system, but indoors. In itself, this is an ingenious and practical solution.
However, that’s not quite enough. Unfortunately, the real world is full of situations where walls, corners, trees etc. can obfuscate the signal. All these things can make the precision as loose as 8 yards which is too inaccurate for many use cases.
That’s where the AI comes to the rescue. Qualcomm’s engineers have demonstrated that it’s possible to train a neural network to correct the environment’s interferences with the 5G signal.
The training consists in placing QR codes that would precisely mark 100% accurate positions and using this “ground truth” to train the AI. Engineers then use the approximate 5G positioning as inputs to the neural network and compare it with the ground truth data to back-propagate incorrect results until the AI can no longer improve the accuracy of the 5G positioning.
It works! There are instances where the positioning precision went from 8-yards down to 1-feet, which is a remarkable result.
Such accuracy is fantastic for consumer-level indoor positioning, but it’s also usable for industrial robotics. As more and more enterprises want to deploy private 5G networks in their facilities, this is a great “bonus feature” that comes with a 5G infrastructure.
Although this is a very convincing proof of concept, we don’t know when you and I could use such a feature in our everyday lives. The good news is that our 5G phones already have the required hardware. As 5G deployments become denser, the opportunity for indoor location services will increase. We’ve never been so close to having a straightforward and accurate indoor positioning system. For nerdy details, watch the video below.