| A typical scene encountered in an outdoor environment where the luminance varies over a wide range (e. g. exceeding 100dB) needs an ultra-wide dynamic range imager. However, for conventional voltage-based CMOS imagers, if the integration time is too long then the brighter pixels will saturate and if the integration time is too short then the darker pixels will be swamped in noise. This voltage-based sampling technique inherently has a low dynamic range.; In this dissertation, a two-degrees of freedom time-based sampling technique is proposed for ultra-wide dynamic range CMOS imagers. The time-based approach automatically optimizes the integration time for each pixel, i.e. the bright and dark pixels choose correspondingly faster and slower integration times. It has the advantages over standard integration-mode CMOS imagers in that complex analog readout circuitry and subsequent readout noise are no longer of concern since only digital signals need to be read out from each pixel. Also, an analog-to-digital converter is no longer required and image reconstruction is straightforward. To achieve ultra-wide dynamic range imaging in video applications, a varying voltage reference should be included to work together with the sampling time. Optimal sampling strategies are discussed in this dissertation based on maximizing the signal-to-noise ratio.; To increase the readout speed, bit plane compression readout schemes are proposed. Instead of reading out and storing the whole frame at each sampling time as conventional multiple sampling does, only the addresses of the fired pixels are read out through reconfigurable shift registers architecture.; Fixed pattern noise is evident in the captured images. Off-chip methods can calibrate somewhat the output data, but for more accurate results on-chip offset cancellation circuits are required. |
