Direction-of-Arrival Estimation Using Signal Processing on Graphs

Abstract

Direction-of-arrival (DOA) estimation, the ability to find the direction from where a sound source originates, is a long-studied problem in array signal processing. Single-snapshot DOA estimation, in particular, brings a unique set of challenges and opportunities for research, particularly in scenarios where environments vary rapidly or limited computational resources are available. The development of single-snapshot DOA estimations systems has long been directed, however, by conventional signal processing. One limitation with conventional signal processing is that data is assumed to lie on a fixed domain of regularly spaced points, resulting in fixed estimation performance. Current methods for single-snapshot DOA estimation can overcome this limitation, and hence gain more flexibility, if the problem is studied from the point of view of signal processing on graphs (SPG). In SPG, data no longer lies on a fixed uniform grid, but instead on vertices of graph. Our conjecture is that we can improve DOA estimation performance by utilizing the framework of SPG. To support this conjecture, SPG principles and tools were employed to produce a new single-snapshot DOA estimation system, based on an existing estimation method from conventional signal processing, that incorporates graphs into the data processing chain. The system was then evaluated using real and simulated data inputs, and explanations were developed to understand the system's output behavior. By leveraging the flexibility offered in SPG, this work shows that our approach to single-snapshot DOA estimation can not only produce outputs equivalent to those in conventional signal processing, but also, depending on the choice of graph structure, outputs with more improved DOA estimation accuracy.