If you’ve ever used a rideshare app in central Boston, you may have experienced the frustration of seeing your driver going in circles, unable to locate you. As a historical city, parts of Boston have streets that resemble a clump of spaghetti — narrow and very close together. And this makes it very difficult for a smartphone’s navigation app to figure out which street you are on.
Your phone has a number of ways to figure out where you are: GPS, Wi-Fi, cellular, and more. All of these methods have inherent errors. Let’s take a closer look at each of these methods and their accuracy.
GPS - The Global Positioning System (GPS) uses 24 satellites that orbit the earth and computes your position by taking the distance between the GPS receiver and at least 4 satellites. Although the government-published GPS error radius is ≤7.8 meters (25.6 feet), the error range could be greater due to signal degradation from atmospheric and multipath effects. There are a number of ways to reduce the error, one of which is discussed in the last section of this blog.
Cellular - Cellular positioning triangulates a device’s position based on the precise and known locations of three nearby cell phone towers. This is combined with signal strength and distance (determined by round-trip signal time) to locate a device and, depending on the density and proximity of cell towers, is accurate to ~1,000 meters (3,280 feet).
Wi-Fi - Wi-Fi positioning works in an almost identical way as cell phone triangulation, but utilizes Wi-Fi routers and signal strength instead. Due to the almost ubiquitous use of Wi-Fi routers, triangulation is far more accurate than cell towers, and can locate devices to less than 80 meters (262 feet).
Beacons - Beacons are small, physical units, usually the size of a keychain-fob. Beacons contain a Bluetooth emitter that broadcasts a unique ID, which is captured by Bluetooth-enabled phones in the immediate vicinity. Beacons are often used in stores to determine the specific location of customers and are highly accurate - to within 1 to 4 meters. Usage is restricted to where units are deployed though, limiting their scale.
IP Geolocation - The granddaddy of location technologies, IP geolocation works by referencing IP addresses against a database of known locations. The upside is its ubiquity – every connected device has an IP address; the downside is accuracy. Typical accuracy is a couple of city-blocks.
Your phone probably uses a combination of methods to figure out where you are. So how accurate is all of this in aggregate?
Smart phone signal error can range from 1 meter to 200 meters. To put this into perspective, if you draw a circle around the footprint of an average Starbucks, ~10% of devices at the Starbucks have a lat/long that falls within that radius.
So how is this inherent error in signal overcome?
First, there needs to be an accurate list of places in the world. For example, if you are trying to determine if a device is in a Starbucks or the wine bar next door that’s out of business or closed for the night, you’ll need a list of places that’s highly accurate, up-to-date and accounts for the shuttering of businesses.
Next, devices need to be “place attached” to the right business or point of interest. As described above, relying on GPS, Wi-Fi and cellular signal is not enough.
Let’s use my weekday routine as an example. At 7 a.m., I’m at the gym, not the chicken restaurant across the street, even though the chicken restaurant is a more popular place when it’s open. And when I’ve been sitting at the intersection of the 405 and 10 freeways in Los Angeles for 30 minutes, I’m not actually at any place — I’m in transit.
A sophisticated model can learn over time by observing dense areas, understanding street topography and tracking open businesses and their hours to accurately determine where a device is located.
Figuring out where people are in the real world is hard and relying on a phone’s signal is not enough. You need a highly sophisticated model trained with high quality, real world data.
Factual’s place attachment model combines high quality business and point of interest (POI) data with a large number of additional factors to attach a device to a specific place.
For POI data, place attachment taps Factual’s own Global Places dataset, with detailed attributes on more than 100 million businesses and points of interest across 52 countries and built using a highly scalable data stack that aggregates, resolves, cleans and normalizes billions of inputs to ensure the highest quality data.
Some of the other factors considered in the place attachment model include:
Factual is constantly tuning and evolving the place attachment model to better understand real world situations and user behavior.