This is a data collection of Covid19 Exposure Notification Protocol data, collected in Berlin subway between mid November 2020 to early January 2021.
Information about the collection process is documented in our blog (currently only in German). This article also contains some analytics and information about data quality. The source code of both sensor and server can be found in the Covid32Counter repo.
The dataset contains an array of frames. Each frame represents a mesurement of Exposure Notification Beacons. Mesurements were taken for 2 seconds every minute. There are exceptions in this interval, when we detected, that the train was parked in a tunnel or depot. We also collected BSSIDs of wifi networks to calculate locations of the mesurements. For privacy reasons, we don't publish the BSSIDs but the calculated geolocations. For the same reason we don't publish the received beacon data but the result of the key matching, including the associated metadata of a positive keymatch. The frame also contains the total number of beacons in this mesurement. We added an ID to the positive frames metadata, so they can associated with each other when they derivate from the same day key.
Example frame:
{
"id": 195605, // uniqe id of the frame
"client_id": 46, // id of the sensor
"timestamp": "2020-12-28T11:44:55Z", // timestamp in UTC
"beacon_count": 3, // number of total Exposure Notification Beacons received in this mesurement
"lat": 52.5149664, // geocoordinates in WGS84, where available
"lng": 13.4183284,
"positives": 1, // number of beacons later identified as covid19 positive in the key matching process
"positive_key_infos": [ // metadata from the positive beacons
{
"key_published": "2020-12-30", // Date when this key was published on the RKI keyserver
"rssi": -92, // signal strength of the BLE beacon
"risk_level": 7,
"days_since_onset_of_symptoms": 1,
"keyid": 40 // ID of the day key which matched the beacon
}
]
}
As decripted in the article the data has some major flaws:
- It's not enough data / sensors
- The app install base is likely to be unevenly distributed across the population
- December was not an ideal time to collect data, because of holidays, change of lockdown rules, low PCR-Test coverage during a high covid incidence period
- Sensors were not evenly distributed over the subway routes of Berlin
- Opening of a new subway line extension during this period
- Sensors were placed in different locations in the trains, also because of different types of trains