Comparison of Efficient FPGA-Oriented Direction-of-Arrival Algorithms on an RFSoC 4x2 Board

Abstract

Reliable UAV navigation in GNSS-challenged environments often requires complementary sensing methods that can operate when satellite positioning is degraded or unavailable. Radio-frequency signals of opportunity can support this type of navigation, but their practical use depends on intermediate processing stages such as direction-of-arrival estimation. This work compares three FPGA-oriented DoA estimation approaches, MUSIC, ESPRIT, and an LU-decomposition-based estimator, within a common RFSoC 4x2 and Vitis HLS implementation framework. Instead of proposing a new estimator, the study evaluates existing hardware-oriented methods under shared assumptions about the receiver, array, fixed-point, signal model, and measurements. All methods use the same four-element ULA configuration, broadside angle convention, 256-snapshot frame structure, and captured RF data. The evaluation compares angular accuracy, HLS kernel latency, update interval, and FPGA resource usage. MUSIC provides the strongest full-range angular robustness, while the LU-based estimator gives the lowest latency and resource cost but produces larger edge-angle error tails. ESPRIT provides intermediate accuracy while retaining timing and resource characteristics close to those of MUSIC, since both methods rely on EVD-based subspace processing.