Content area
Unpredictability is a key resource for many computational applications like cryptography, statistical simulations, probabilistic games (e.g., gambling), fair selection, or even quantum key distribution. Quantum random number generators promise guaranteed quality, secure, fully random output by utilizing the fundamentally indeterministic nature of quantum measurements. While these generators can be based on any quantum phenomena, advancements in quantum optics are making architectures based on various properties of photons more accessible, with continuously increasing capabilities. One such architecture is photonic time-of-arrival based generators, where the non-deterministic emission times of single photons are measured. One of the main advantages of this scheme is that it can be realized with a relatively simple measurement setup while offering substantial output entropy rates. The expected output of any random number generator is a uniformly distributed sequence of zeroes and ones. Actual physical measurement statistics, however, typically follow some other non-uniform distribution, mandating the need for various post-processing steps. Additionally, potential imperfections and non-idealities can also influence the measurement statistics and must be handled. Thus, the non-trivial task of post-processing is to transform physical measurement results to a high quality output bitstream while striving for optimal bit-generation efficiency with affordable computational costs. After briefly introducing the topic of quantum random number generation in the first chapter, this thesis explores and presents various challenges and possible solutions associated with post-processing the measurement results of time-of-arrival based quantum random number generators. The second chapter presents a scheme for dealing with unwanted correlations due to typical hardware imperfections. The third and fourth chapters present two different post-processing schemes to generate uniform output. The first is inspired by the continuous probability integral transform and can be used in practical scenarios where the measurement setup is well characterized, while the second one is based on universal hashing, offering increased error tolerances at the cost of elevated computational costs. The theoretical results and claims of the thesis are also verified experimentally.