University of Twente Student Theses


Signal Recovery using CλaSH

Wentink, D.J.M. (2017) Signal Recovery using CλaSH.

[img] PDF
Abstract:Ion-mobility Spectrometry (IMS) devices are used for identifying small amounts of substances in an air sample. A problem in increasing the level of detection is the effects of noise and finite resolution. Signal recovery algorithms are used to recover the input signal of a system with a known response, and can be used to improve the level of detection of IMS devices. A signal recovery algorithm based on the Maximum Likelihood (ML) principle approaches the limit on information that can be recovered from noisy data, but is computationally very complex. The computer systems that are currently used to solve ML problems are not suited for portable devices. A field-programmable gate array (FPGA) is a reconfigurable chip which is potentially more energy efficient than traditional computer systems. C¸aSH is a high-level programming language for creating synthesize FPGA architectures which is well-suited for mathematical algorithms. In order to downsize IMS systems, an energy efficient FPGA implementation of a ML-based signal recovery is made using C¸aSH. In the background study the ML signal recovery problem is analysed mathematically, and it is shown to be an optimization problem. The conjugate gradient algorithm can be used to solve this problem in a fixed number of iterations. The computational complexity comes froma large amount of matrix-vector multiplications. A processor architecture has been designed to solve the ML signal recovery problem in a fast and energy efficient manner. The architecture uses parallel computations, multiply accumulators and a custom memory architecture to be fast atmatrix-vector multiplications. To determine energy efficiency, the FPGA implementation is compared to an implementation using a modern graphics processing unit (GPU). The FPGA is at least a factor 8 more energy efficient, at the cost of processing time. In the worst case, the FPGA takes 1.7 seconds of computation time whereas the GPU needs 0.7.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology, 54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page