Congratulations to Konstantin Kapinchev, who has recently passed his PhD viva, with minor corrections. His PhD thesis was titled ‘Scalable Parallel Optimization of Digital Signal Processing in the Fourier Domain’.
Konstantin said: ‘As well as my supervisor and examiners, I would also like to extend thanks to Professor Adrian Podoleanu in the School of Physical Sciences for his expertise and willingness to help during my work.’
The aim of the research presented in this thesis is to study different approaches to the parallel optimization of digital signal processing algorithms and optical coherence tomography methods. The parallel approaches are based on multithreading for multi-core and many-core architectures. The thesis follows the process of designing and implementing the parallel algorithms and programs and their integration into optical coherence tomography systems.
Evaluations of the performance and scalability of the proposed parallel solutions are presented.
The digital signal processing considered in this thesis is divided into two groups. The first one includes generally employed algorithms operating with digital signals in the Fourier domain. Those include forward and inverse Fourier transform, cross-correlation, convolution and others.
The second group involves optical coherence tomography methods, which incorporate the aforementioned algorithms. These methods are used to generate cross-sectional, en-face and confocal OCT images. Identifying the optimal parallel approaches to these methods allows improvements in the OCT imagery in terms of performance and content. The proposed parallel accelerations lead to real-time generated comprehensive imagery. This improves the utilization of the optical coherence tomography systems, especially in areas such as ophthalmology, where detailed visual information provided in real-time is crucial for accurate diagnosis.
A number of reports on performance, in terms of latency and speed-up, provide information for a balanced choice between performance and OCT method and between performance and size of digital signal, which in many cases is linked to image quality. The proposed parallel approaches, along with providing solutions to a range of signal processing problems, illustrate the computational characteristics, both sequential and parallel, of the utilized microarchitectures, language specifications, and libraries. To reach an improved performance, computer programs need to reflect these characteristics.