Classical receiver architectures demodulate a high frequency bandpass signal to baseband before sampling the in-phase and quadrature components. With the advent of high-speed analog-to-digital converters (ADCs) and wide bandwidth sample-and-hold (S/H) circuits, it has become practicable to sample a bandpass signal directly without any demodulation operation and then process it with robust DSP technology. Direct sampling methods do present their own challenges. When a single channel is used to digitize the signal, not all frequencies above the Nyquist rate are allowed and only signals in certain frequency bands can be sampled at this minimum rate. As first shown by Kohlenberg, the restriction on spectral location can be removed with a two-channel time-interleaved ADC (TIADC) where two ADCs separated by a timing skew independently sample the signal.

In this dissertation, we propose a general and flexible technique for sampling the complex envelope of a bandpass signal using a nonuniformly interleaved $M$-channel TIADC. The bandpass signal is sampled directly by the sub-ADCs, and the overall TIADC sampling frequency is at or slightly above the Nyquist rate. Reconstruction of the complex envelope entails inverting a matrix of filters resulting in some TIADC timing skews being forbidden. The proposed sampling scheme requires the implementation of $M$ digital FIR filters and can be used to digitize bandpass signals with any carrier frequency in software defined radio applications. Reconstruction analysis is provided for the case of two, three, and four channels. Multi-tone and MSK signals are used in simulations to validate the proposed method and assess its performance. Quadrature sampling, a special case of the two-channel TIADC that assumes certain parameter relations, is investigated as an approximation technique.

It is well known that gain and timing skew mismatches can severely degrade TIADC performance. To mitigate the effect of these mismatches on complex envelope reconstruction, we present a novel blind calibration method which assumes that there exists a frequency band where the complex envelope signal has no power, due for example to oversampling. Mismatches give rise to errors in this band, which are extracted and used to estimate adaptively the gain and timing skew mismatches. Simulations with multi-tone, MSK, and bandlimited white noise signals demonstrate calibration can significantly improve reconstruction performance measured in the mean-square error (MSE) sense. Suggestions for further research are provided.