Broadband analog opto-electronic blind source separation

This thesis addresses the problem of high bandwidth blind source separation and demonstrates an analog opto-electronic implementation of a processor capable of solving this problem for two linearly mixed signals. Using only the assumption of signal independence, real-time separation exceeding 20 dB is demonstrated. The opto-electronic circuit is capable of processing signals with 100 kHz bandwidth. The theory presented here show that the dynamics of the core portion of the system depend on statistical characteristics of the signals. If at least one of the signals is sub-Gaussian, the dynamical properties of the system are such that one of the original signals obtains all the system gain and is extracted from the mixture.; A dynamic feedback loop containing an electro-optic phase modulator operating in the nonlinear regime provides processing functions analogous to higher-order statistical correlation of the signals. The correlation portion of the processor is based on photorefractive two-beam coupling in BaTiO3 at an optical wavelength of 532 nm. The behavior of the feedback loop is described by an open-loop gain analysis and a steady-state stability analysis. The system demonstrates "winner-takes-all" competition between the two input signals.; The bandwidth of the processor presented here is limited by electronic amplifiers and can exceed 100 MHz with the higher bandwidth off-the-shelf components. The high bandwidth makes this processor applicable to communication blind source separation scenarios that are difficult to solve using traditional digital signal processing. In addition, this processor can be used directly at the microwave carrier frequency of tens of gigahertz.