Evaluating Wideband Beamforming Methods for Low-Frequency Spatial Awareness Applications : An Analysis of Computational Complexity and Performance

Sammanfattning

This thesis investigates the design and implementation of a low-frequency beamforming system using FPGA, with a focus on optimizing computational complexity. The methodology followed a structured approach, starting with simulations, followed by the development of a fixed-point model and FPGA implementation. Several beamforming methods, including frequency-domain subband, tapped delay lines, and time-domain subband, were compared for their computational efficiency, alongside the consideration of adaptive beamforming techniques such as MVDR. Based on this analysis, the frequency-domain subband method was selected for its low computational cost, particularly in scenarios where adaptive beamforming is not required. Two FPGA architectures were evaluated for implementing the selected FFT-based beamforming system. Architecture 1, which utilizes a single delay path FFT, was initially considered but found to be inefficient in terms of hardware utilization, particularly since parallelism was not necessary at the low sample rates involved. In contrast, Architecture 2, which employs serial processing, was implemented due to its ability to fully utilize the hardware while maintaining a relatively low clock frequency. The FPGA implementation, designed with 25×18 DSP multipliers and distributed memory, successfully demonstrated spatial perception, enabling the accurate detection of signal direction and power, even in the presence of noise. While the system effectively utilized available resources, accuracy of detecting angles could be enhanced by increasing the number of FFT points and expanding the sine/cosine tables.