Adaptive Beamforming for Weather Observations using a Constrained Capon Method and the Advanced Technology Demonstrator at NSSL

Phased-array radars are being evaluated as potential replacements for the Weather Surveillance 1988 – Doppler (WSR-88D) radars, whose service life is expected to be sustained beyond 2035 to the 2040 timeframe by ongoing supporting engineering and maintenance. The Advanced Technology Demonstrator (ATD) has been developed and deployed as part of the research effort to show the ability for a phased-array radar to collect calibrated, dual-polarization weather data with rapid update rates. The ATD utilizes an overlapped-subarray architecture, which allows for digital beamforming on receive, including adaptive beamforming. This makes it possible to adjust the antenna pattern to improve the accuracy of estimated radar variables, especially in the presence of large reflectivity gradients. The Capon beamformer is an adaptive beamforming technique capable of minimizing contamination from discrete targets but is impossible to calibrate for distributed targets. In order to address this shortcoming, a “Constrained Capon” method is proposed, which adds more constraints to the Capon optimization problem. These constraints have been developed with the goal of shaping the main lobe of the adaptive beam in order to allow for calibration while providing flexibility to adjust the sidelobes to reduce contamination. To evaluate the proposed technique, a subarray in-phase/quadrature (IQ) data simulator has been developed to take archived WSR-88D data as inputs and produce subarray IQ data as though collected by the ATD. These simulated data are used in the adaptive beamforming technique to form optimized beam patterns, with the results demonstrating that the proposed technique leads to calibratable radar data with reduced biases from sidelobe contamination. However, due to the complex optimization problem associated with the proposed technique, real-time operations are not possible. In this study, we evaluate results of the Constrained Capon method, with a particular focus on the use of deep neural networks for eventual real-time usage. This will allow for the application of beamforming weights to all range gates in milliseconds. A proposed neural network architecture is discussed, and preliminary results are presented.