See How ML Aids in Detecting Uncertified Drones
FREMONT, CA: Commercial drone applications are becoming increasingly customary, with recent forecasts indicating that drones represent increased market opportunities. Drones are extensively used, for newsgathering and inspection and mapping in industrial applications. In many cases, drones are required to transmit video feeds to their flight controllers imposing heavy uplink traffic load on the networks. Mobile networks are well suited to support drone communication, but this massive imposition demands mobile networks to identify if a User Equipment (UE) is a drone or a regular ground UE. But there may be an occurrence of uncertified use of mobile devices or subscriptions which may cause increased interference to the network. Interference mitigation techniques must be initiated, as soon as an uncertified drone is detected. Machine Learning (ML) is the key to Uncertified Aerial UE Detection. Read on!
The radio link features and mobility patterns are different for aerial and ground UE. The device in the sky is expected to have a close to the line of sight propagation that leads to the low discrepancy of Reference Signals Received Powers (RSRP) of the active cells. For indoor UEs, the penetration loss will be more in comparison to an aerial UE at a similar height, enabling the differentiate the two UE types. As ML algorithms need data to be prepared the measurement data from known legitimate drone UEs flying in the network, are collected in addition to the data from ground UEs. The gathered measurement data are then divided, into a training data set and test data set. The former is used for training a drone detection machine learning model and latter for assessing the performance of the model. The detection accuracy increases with drone height and more measurement reports.
Deploying machine learning algorithms in mobile networks for detecting uncertified drones is another challenge. Two options possible are a central entity storing all the data and training a single machine learning model or machine learning model being trained per cell or group of cells. Careful design is needed to strike the right balance between the amount of signaling between the nodes and uncertified drone detection performance.