This dissertation investigates the potential transmission capacity improvement in optical communication links due to the use of single photon avalanche diode (SPAD) optical receivers. Currently, for non-optically amplified optical communication links that use intensity modulation with direct detection, the highest optical sensitivity is achieved using avalanche photodiodes. However, the excess noise generated within the avalanche photodiode (APD) limits the receiver sensitivity. Operating the APD above its breakdown voltage, as a SPAD, can eliminate this excess noise.

Photon detection efficiency (PDE) is a key parameter of SPAD arrays. At present, PDE is normally experimentally measured. Current PDE models cannot accurately predict the experimentally observed oscillatory behavior of the persistent avalanche current or its fast collapse under strong feedback conditions. For the first time, a stochastically self-regulating SPAD model, which includes the oscillatory behavior of the persistent avalanche current and the load feedback, together with a Monte Carlo simulation method, is used to evaluate the effective PDE of SPAD arrays. The model enables the first detailed study of the minimum multiplication required for each photon to be detected. The model indicates that a strong feedback tends to stabilize this threshold multiplication. In addition, an improved perspective of the avalanche triggering probability is introduced which proves that some photons can be detected without generating a self-sustained avalanche. The accuracy of the proposed Monte Carlo method increases with allocated computing power.

Multilevel modulation schemes are usually employed to improve the spectral efficiency of optical communication links. Currently, no suitable SPAD-array signal model for multilevel modulation has been reported. Therefore, the first SPAD-array contention signal and noise model, suitable for multilevel modulation schemes with signal processing, is proposed. The model is verified by comparison with published experimental results. For the first time, using this model, the signal variation after equalization or high speed distortion can accurately be described. This model provides a foundation for simulating high speed SPAD-based optical communications. A method for calculating the bit error rate (BER) of a digital SPAD receiver for higher order pulse amplitude modulation (PAM) modulation with equalization is proposed. The method is verified by comparison with published experimental results. It is found that the Poisson model, previously used in the literature, fails to characterize the signal variation after frequency dependent distortion or equalization.

In addition, the first numerical investigation for SPAD-based, high data rate, free space, visible light communication, using higher order PAM with matched filter, linear and second order non-linear Volterra post-equalizations, is presented. Via simulation, it is demonstrated that a 3 Mb/s, 40 cm, PAM-8, SPAD link can be achieved using Volterra and feed forward equaliser (FFE) + decision feedback equaliser (DFE). This is a significant improvement compared to the previously reported unequalized, 100 kbit/s, 15 cm transmission.

Plastic optical fibre (POF) has great potential for use in short reach communication, especially for home and automotive networks. Therefore, currently, the comparison of high speed, low-cost, POF links with either a large area APD, or a large array of SPADs is of particular interest. For the first time, the use of SPAD receivers in POF links is analyzed in comparison with APD receivers via simulation studies. The simulations show that a SPAD-based receiver is more sensitive in a POF link for a 4 Gb/s scenario given the condition that symbol period, integration period and dead time are equal. It is concluded that if the summation of PDE, fill factor and SPAD array saturation penalties are smaller than the SPAD additional power budget compared with an APD, then the SPAD array will have a higher sensitivity. To achieve this, the number of SPADs in an array should be large enough to prolong the linear region, leading to a reduction of saturation penalty. In addition, the dead time should scale with the symbol period. It is theoretically verified that SPAD arrays have the potential to support gigabit per second data rates with a higher sensitivity than conventional APD receivers in typical optical wireless or guided wave communication links that require large area detectors.