Generating a Robust Dataset for Radar Detection in the CBRS Band

Abstract

Oftentimes wireless spectrums can become saturated when the demand for the spectrum is too high leading to lower quality communication. We choose to address this issue by employing spectrum sharing, which involves letting various users share the same frequency band in order to deliver certain data as bits through this spectrum, and leads to more optimized communication. There is interest in allowing cellular networks to communicate through federal bands that have so far been reserved for the government. However, one prevalent issue is that these communications need to be paused whenever the federal bands are being used to send a radar signal by the government in order to not interfere.

In order to recognize when the radar is being sent we propose developing a machine learning software to recognize radar signals within the spectrograms that are generated from the network and stop the data from being sent in the interfering frequencies without cutting off the connection that the users have. We deploy this on Open Radio Access Network (O-RAN), which is a virtualized, disaggregated solution for 5G/6G cellular networks, and is extremely beneficial in that any other cellular operator that uses O-RAN can use the same solution. For example, when there are different signal-to-noise ratios and when there isn’t, as well as when the radar is shifted overtime in the spectrogram. In fact, creating spectrograms of radar signals at different delays can be achieved through shifting existing data within a smaller dataset. After that, the graph is then overlaid with a LTE signal with varying noise levels in order to make sure that the radar can be recognized in different environments. Once we introduce multiple spectrograms with different delays as well as different noise levels, we will be making the dataset more robust leading to the machine learning model to detect radar signals more accurately.