Agenda

Signal Processing Seminar

3D direction of arrival estimation of multiple audio sources with spherical microphone arrays

Despoina Pavlidi
University of Crete, Department of Computer Science, Heraklion, Crete, Greece

Abstract: Direction of arrival estimation plays a central role in numerous signal-processing applications, such as smart home automation, surveillance systems, etc. Until recently the research community was mainly interested in single-dimensional direction of arrival (DOA) estimation by deploying linear or planar microphone arrays. Nowadays the focus has turned also towards spherical microphone arrays, which enable the more accurate capturing of the acoustic wavefield, hence enabling two-dimensional DOA estimation, i.e., the azimuth and elevation of an active audio source. In this talk we will present our proposed methodologies for DOA estimation in the 3D space. Our first proposed method relies on energetic analysis. We estimate the sound intensity vector on selected time-frequency elements of the spectrum and post-process the estimates utilizing 2D histogram representations. We enhance our approach by applying beamforming around local intensity vector directions. We call our hybrid approach spatially constrained beamforming (SCB). Our second proposed method improves the performance of two grid-based approaches, namely the steered response power (SRP) and the multiple signal classification (MUSIC) algorithm, both formulated in the spherical harmonic domain. We propose to derive local DOA estimates from the power map for SRP and the pseudospectrum for MUSIC. From these local DOA estimates we form a 2D histogram that we process to derive the final multiple sources directions.

Overview of Signal Processing Seminar

Agenda

MSc SS Thesis Presentation

Guillermo Ortiz Jiménez

Multidomain Graph Signal Processing: Learning and Sampling

Sparse sampling for tensors and graphs

MSc SS Thesis Presentation

Haidong Hao

Vessel Layer Separation of X-ray Angiographic Images using Deep Learning Methods

Fast solutions based on a fully convolutional network (FCN) trained by conventional loss or adversarial loss.

MSc TC Thesis Presentation

Feng Ma

Respiration monitoring based on information fusion from Impedance pneumography and Electrocardiography

With known respiration information, corresponding parameters in both time domain (respiratory rate) and frequency domain (respiratory power) can be extracted to indicate the health condition.

Signal Processing Seminar

Pim van der Meulen

Low-cost sparse sensing designed for specific tasks

MSc TC Thesis Presentation

Lichen Yao

Bluetooth Direction Finding

This thesis project focuses on the algorithm developement and practical considerations for Indoor Direction Finding feature that will be incorporated in the next generation Bluetooth standard.

Signal Processing Seminar

Tuomas Aittomäki

Tutorial on: Sum-of-squares Representation in Optimization and Applications in Signal Processing

Signal Processing Seminar

Krishnaprasad Nambur Ramamohan

Signal processing algorithms for acoustic vector sensors

Signal Processing Seminar

Farnaz Chamanzadeh

Accurate timing and positioning through an optical-wireless distributed time and frequency reference

Signal Processing Seminar

Aydin Rajabzadeh

manufacturing defect detection