Late last year, my thesis was approved by Virginia Tech. Fortunately, that meant I could graduate. Unfortunately, it probably means that my thesis will be socked in a (virtual) drawer somewhere to gather (virtual) dust. I had the option to produce a vanity publication through some shady German-owned publication house, but I opted out. I decided that I would publish my thesis to the Internet for anyone who was interested in such research. It can be found here:
http://scholar.lib.vt.edu/theses/available/etd-12082011-204951/
I looked at Hidden Markov Models (HMMs) as a method to enhance functionality in cognitive radios. HMMs have been used previously for pattern recognition, such as handwriting and speech analysis, but they have seen limited use in the wireless world for things like spectrum sensing and analysis.
CPUs (especially the really small ones found in modern wireless devices like radios and cell phones) have trouble keeping up with the demands of most HMM implementations. Therefore, I created both C and CUDA implementations of such HMM algorithms in order to compare their executions times. From what I found, graphics cards (GPUs) can surpass CPUs only when many states or many models are used at the same time. If you’re interested in the results, check out my thesis. If you’re interested in trying out the code or replicating my results, you can find my code hosted here: