The Seer

Our proposed system will be focused on streamlining the current method of NR data transmission by using a machine learning approach. We would like to create a neural network that can use the signal received from different antennas in an array, to determine the direction-of-arrival of a low-band 5G signal. The system will be focused on Sub-6 5G NR within the frequencies of 600 MHz (n5) and 850 MHz (n71) bands. The radiation patterns across an antenna vary greatly with position, and analyzing the amplitude and phase of these radiation patterns can provide useful data for the creation of a deep learning model. Use of a neural network will allow for a more efficient method of pinpointing the location of an incoming 5G signal, allowing companies that utilize these systems to save energy and money, which in turn lowers the carbon footprint of these baseastations as well as the cost of implementation of such a system.

We will use an array of antennas, processing the received signal through RTL-SDRs and running the collected data through a neural network implemented in Keras for the training phase. Using a HackRF One to transmit the signal, we will take measurements with the Tx in varying locations relative to the Rx array. Once we have trained the neural network we will test its efficiency on a new measurement and determine if the accuracy and loss are acceptable. GNU Radio will be used to program and control the Tx and Rx, allowing for a high level of hardware control. This will be imperative since our data collection methods must be accurate in order to allow our neural network to create a representative model. We hope that our system will greatly improve the current methods for determining the direction-of-arrival of a low-band 5G signal, bolstering the communication between a base station and transmitter, while lowering the economic impact of these large scale 5G systems.