Abstract:
Path-loss (PL) Prediction and Line-of-Sight (LOS) Identification play an impor tant role in modeling UAV-based mm-wave networks. Communication at mm Wave provides larger bandwidth with low latency for the fifth generation (5G)
and beyond. In UAV-based mm-wave communication, the channel characteris tics i.e. (PL-Prediction and LOS-Identification) are highly unpredictable and are
difficult to be modeled by statistical or deterministic methods with high accuracy
and low computational complexity. Moreover, no generalized mm-wave wireless
channel model exists that can be adopted in all kinds of communication environ ments. This paper presents two important aspects of the UAV-based mm-Wave
channel model through different Machine learning (ML) methods which are: 1)
PL-Predictions; 2) LOS-Identification. The ML-Models are built through the con sideration of multiple input variables related to the environment and channel char acteristics. For training the ML-Models the data set is generated through exten sive simulation performed in a self-developed Ray-tracer (RT) assuming UAV as
a flying base station over different heights. Specifically, we have exploited SVM,
RF, and K-NN for PL-Prediction and LOS-Identification. The ML framework in cludes a data Pre-processing step, hyperparameter tuning, and a feature selection
scheme to improve the accuracy and performance of the ML model.
