CARLab seminars are usually held during semester (teaching weeks) on Tuesdays at 1pm in room 830, Building J03. This is always subject to change, however, so please check this page close to the day of the seminar of interest or check by email here. Directions to CARLab can be found here.
FPGA-based signal processing for spherical microphone arrays
1pm, Tuesday 21st August, 2012
Current interfacing between a spherical microphone array (SMA) and a computer requires a large system of amplifiers and multiplexers which is power inefficient and subject to high latency. The commercially available systems are expensive and consist of proprietary interfaces and protocols which limit their flexibility. This research focusses on the design of low-cost real-time FPGA-based embedded systems for processing of SMA signals. The aim is to provide an efficient acquisition/transfer interface for delivering SMA signals to computers in a way that minimises the computational load for the CPU/GPU. The challenge is due to the limited resources of the FPGA and the real-time requirement associated with audio processing. This presentation will focus on a 64-channel (48kHz, 24-bits) SMA and its signal processing on a Virtex-6 FPGA. Some of the related low level audio processing techniques and design issues will be discussed.
Evaluation of different stimulation and measurement patterns based on internal electrode: application in cardiac impedance tomography
1pm, Tuesday 29th October, 2012
Joubin Nasehi Tehrani
Time Series Prediction as a Cloud Service
1pm, Tuesday 14th August, 2012
Time series prediction technology is crucial in a wide range of disciplines including finance, energy, networks, traffic prediction, resource management and logistics. Unfortunately, it has remained expensive to deploy due to a requirement of specialized knowledge in database systems, machine learning and visualization. We present a cloud-based architecture for a scientific collaboration web portal, allowing researchers and non-expert users to collaborate by exchanging data, building tools for time series prediction and comparing results through a web based portal. A unique feature and a key advantage of this system is its common framework, which allows researchers to do reproducible predictions on different data sets, creating a collaborative and competitive environment for research and improvement. Also different customers looking for a good prediction algorithm can try different algorithms from a web based interface on their dataset and compare results. This would assist different groups in finding each other and in making best use of the information they have at their disposal. This work will also address issues related to Intellectual Property (IP) protection, leading to the possibility of a prediction market, where different contributors can sell their prediction services without revealing their IP. Effectively a matchmaking service is formalized between users and developers in a manner similar to App stores, with security, access rights and pricing among the implementation concerns.
Initial work on Relating Head Related Transfer Functions to Head and Ear Morphology
1pm, Tuesday 5th June, 2012
For personalized 3D audio the relation between head and ear shape to hearing is both an important and interesting question for science and engineering. Our head, torso and particularly external ear shape influence the way sound is reached to our ear drums. This influence is in a directionally dependent way. The filtering of the sound by our head is called Head Related Impulse Response (HRIR) and in the frequency domain it is called Head Related Transfer Function (HRTF). Traditional ways of calculating the HRTF are both expensive and time consuming. This talk aims to present initial work on a proposed method for finding the HRTF via the head and ear shape.
An FPGA-based Floating Point Unit for Rounding Error Analysis
1pm, Tuesday 22nd October, 2012
Detection of floating point rounding errors normally requires run-time analysis in order to be effective and software-based tools are seldom used due to the extremely high computational demands. In this presentation I will discuss a field programmable gate array (FPGA) based floating point co-processor which supports standard IEEE-754 arithmetic, user selectable precision and Monte Carlo Arithmetic (MCA). When coupled with a hard or soft processor, this co-processor enables the detection of catastrophic cancellation, minimizes the floating point precision in reconfigurable computing applications, and and can also be used to determine whether single precision floating point is sufficient for CPU and GPGPU applications. Moreover, programs compiled for the processor can be made self-validating without any changes to the source code.
Delay-based Pattern Recognition and FPGA Implementation
1pm, Tuesday 15th October, 2012
Chong Hau Ang
A typical pattern recognition circuit is usually implemented using conventional design practices and standard electronic hardware components. The work to be presented looks into the use of time delays for pattern recognition and the exploitation of FPGA hardware resources as programmable delays for implementing a pattern recognition circuit.
Electrical Impedance Tomography for Pulmonary Perfusion Imaging
1pm, Tuesday 8th October, 2012
Doan Trang Nguyen
Electrical Impedance Tomography (EIT) has many advantages as a medical imaging modality. It is non- invasive, radiation-free and portable. This technique could be used for the diagnosis of pulmonary embolism, which up to now does no have a satisfactory solution. However, the current resolution of EIT is very low, rendering its sensitivity for per- fusion defects not yet suitable for clinical application. In this talk, I present a simplified computer simulated model using 3D EIT to image the perfusion defects. A FEM mesh of 122 880 elements is used to solve the forward voltages using algorithms built-in to EIDORS, an open source software for EIT. The EIT system simulated has 4 rings of 16 electrodes, excited in adja- cent patterns. The resulting voltage measurements with added white noise at 24 dB are reconstructed on a coarser mesh of 7680 elements. The results show that 3D EIT can image perfu- sion defects as small as 5% of a lung volume in the upper lung regions. Larger defects are observable from the reconstructed images in all regions of the lung. Although the study is simple and therefore suffers from several drawbacks, the results are encouraging for future application of this technique in perfu- sion defect imaging.
Blind Source Separation Algorithms for Spherical Microphone Arrays Using Compressed Sensing
1pm, Tuesday 1st May, 2012
Ping-Kun Tony Wu