Location - how it works

Transcript

So the Mist mobile SDK is responsible for receiving the BLERSSI information and combining it with sensor data from the mobile device and sending that back to the cloud every second while the mobile device is moving. Then this machine learning system continuously adapts to changing RF environments without the need for site surveys. Unlike single antenna BLE designs, Mist uses a single transmitter to drive eight unique directional antennas. Each beam contributes to the likely location of the device.

And this technology examines all the probability surfaces, one for each directional beam, and combines them to find the most likely point in the map where the device is located. So two terms of importance are ple and intercept. ple is the path los exponent. Mist uses this formula to determine the expected signal strength at various locations on a map based on a gain at that direction from the antenna taking into account orientation and ceiling height.

The intercept is another constant that must be derived. It is much like the intercept of a line and indicates the expected power at one meter from the antenna. Individual RF environments and even device types have different optimal ple and intercept values. This ensures the RF model adapts to the RF environment as it changes and accounts for the differences between mobile device types, providing a consistent user experience.

The essential element behind Mist machine learning is to seek the maximum agreement between the results of increasingly many location estimates, from that, construct individualized path loss formulas. As the machine learning algorithm begins, its using the default plf for a device type, meaning an iPad, an iPhone 6, an iPhone 6S. Machine learning is running continuously.

The only time you would see an AP at zero learning is when it was just added to the map or something that’s been changed and its configuration regarding ceiling height, orientation, et cetera. The thing to look for here is to make sure all of your APs are learning. To reach level one usually only takes a couple of minutes for a device. And that’s the most important level because 95% of the work is done reaching level one learning. After that, the algorithm is looking for new, unique data points.

We have created this visual to help you quickly understand the status of location machine learning. If an AP is not learned on a device it will be grey. When all the devices are learned they will show green. So what is going on here? Using this Show All button you can see the raw data. And note, you can also choose to expose just a single device. What the UI has done is to add up the ple and intercept.

Then, a median value is derived per device, and per AP, a variance from that median is shown. In a normal environment we would expect this variance not to be more than one or two. If it is three or four, the tile will show yellow and more than five it will be red. This is to alert you to take a look at why this AP may be so different from the others.

So in this case, this is what I would expect. If you saw a five or six, something that’s radically different from the others– immediately I would go look at the floor plan and see if that AP is in a strange spot that would justify it having a much different plf than the rest of the APs.

The Mist patented dynamic BLE array (included in all Mist AP’s)

Mist vBLE Array

The Mist AP has a vBLE array, which consists of a 16 Directional Antennae Bluetooth Array. Eight unique antennas are reflectors to help with directing the BLE energy outward away from the AP, and the other eight unique antennas are directional antennas that point in eight different directions, transmitting BLE energy in a flashlight-like fashion.
To better visualize this, imagine a flashlight shining narrow beams in eight different directions.

AP Directional Beam Angle

The best directionality comes from about a 45 degree angle, with the AP’s LED facing down towards the ground. Anything below 45 degrees acts more like a null radius, since you are directly under the AP at 0 degrees.
This is why the higher an AP is placed, the null radius under the AP increases and the RSSI from the client decreases.

AP Beams

Each AP is transmitting on eight unique directional beams.
The Mist SDK on the device hears these beacons from the beams, and sends the RSSI and device sensor information back to the Mist cloud. Through either Wifi or cellular.
The more beams your device hears the more accurate your location will be, as there is more data for the location engine to make a better location estimate.

Mist builds location from “Probability Surfaces” not triangulation. The more directional RSSI beam information your client hears, the more data for the location engine to work with creating the probability surfaces, which provides the best possible location accuracy with sub second latency.

Essentially your mobile device, with the Mist SDK, listens to the BLE coming from the beams from the Mist AP, once the Mist SDK receives the beams the SDK will then send the mobile information to the cloud where location engine will calculate and respond back with an x,y coordinate to your mobile device.

So the more beams your client sees from neighboring APs the more accurate your location will be as these probability surfaces build up. Machine Learning is constantly evaluating the Path Loss Formula (PLF) per device type and will modify the PLF over time as it learns the “optimal value/device/AP” for the actual environment. Below is an example of the probability surface where the center of the red blob is the peak of the surface and the highest probability of where you are and the blue is the least probability where you can be.

If properly deployed, the Machine Learning engine will get better and better over time has it sees more data over time so the network will be ‘constantly optimizing’ for your location experience.