Dynamic Holography

Signal Seperation: Optical Ears

Overview

The cocktail party problem highlights the human ability to listen to a single voice in a crowded room despite the interference of many other voices and various sources of noise. This problem has apparent generalizations to communications domains other than speech and is in fact indicative of a broad range of signal processing scenarios. One practical abstraction of the cocktail party problem is referred to as blind signal separation, the objective of which is to descramble unknown signals that have been mixed in some unknown way.

We have built an analog optoelectronic system incorporating dynamic holography that performs blind signal separation on signals that have been linearly mixed. The dynamics of the core portion of the system (See Figure) depends on the statistical character of its input signals. To lowest order in the dynamical variables signals can be classified as super-Gaussian, Gaussian, or sub-Gaussian, depending on whether the forth-order statistical moment is greater than, equal to, or less than that for a Gaussian distributed random signal. If at least one signal is sub-Gaussian the system dynamics is winner-takes-all, meaning one of the original signals is extracted from the mixture; the remainder of the signals can be processed by replicas of the core system. A variant of our system can perform winner-takes-all on super-Gaussian signals, although the communication signals of interest to us are usually sub-Gaussian distributed. We have also measured the performance of our system on a variety of statistical signals. The system reveals mixed signal separation performance of 40 dBV or better.

The functionality of our analog system is similar to that of computer algorithms that perform blind signal separation by independent component analysis (ICA) [1]. ICA algorithms are computationally intensive and are hard-pressed to perform blind signal separation in real-time even on signals of speech bandwidth using very fast specialized digital signal processing hardware. Our current demonstration processes signals having 100 kHz bandwidth; we expect it can be extended to signal bandwidths approaching 1 GHz and carrier frequencies to several tens of gigahertz. (The former is limited by a circuit delay time and the latter is limited by one's ability to modulate the laser beam.)

Vapor Detection: Optical Nose

This is a multidisciplinary project drawing heavily on Physics, Electrical & Mechanical Engineering, Chemistry, and Biology while at its core is the application of Dynamic Holography for signal processing in a practical real world application. We work to develop a paradigm in adaptable olfactory detection of any chemical, without limit to the chemical transducer, utilizing an Adaptive Holographic Interferometer (AHI) as a signal processor. Or, quite simply, we use light to probe chemical sensors as a means of smelling.

As chemical solvents or any airborne chemical that is able to chemically interact with a material comes into contact with it some reaction occurs. Whether it bonds to the surface, diffuses deeper to bond, or merely acts to break apart bonds the structure is somehow changed. This change manifests itself as an index and or dimensional change. Both of these affect the optical path length and can therefore be used to create an amplitude-modulated signal on the output of an interferometer.

The interferometric system uses a Diffusion dominated Photorefractive crystal to create what is referred to as an adaptive holographic interferometer. This essentially means That where a standard Michelson or Mach-Zehnder would require precision beam-forming optics to achieve a uniform phase front to have single fringe modulation. The photorefractive creates an index grating that couples light from one beam to another effectively compensating for any less than stellar phase-front negating any need for complicated optical alignment, spatial filtering, and columniation. As one beam experiences a phase shift relative to the other, if the shift is faster than the dynamic hologram's ability to respond and rewrite the grating, energy gets coupled back to the darker beam yielding a phase-to-amplitude demodulation. Our system uses lasers in the 532nm or green wavelength to pickup a phase shifts and the entire interferometer ranges in optical path length from half a meter in some systems, to a few cm's in others.

When detecting airborne particulates, our system is essentially limited by the transducer response and so we often characterize the system itself by its fundamental sensitivities and then adjust the chemical concentration sensitivity depending on the transducer. Our system has proven sensitive to a phase and displacement response currently limited to 1.9microrad/sqrt(Hz) and 1pm/sqrt(Hz) respectively. Using polymers, the old mainstay for solvent olfactory detection, we have detected down to 90ppb/sqrt(Hz) of an airborne solvent. A strong advantage of our system is we are able to use any sensor that is optically transparent to the probing wavelength, which nearly anything is if its thin enough. This means if someone can make a sensor to detect something, we can most likely use it allowing for great adaptability and future improvements in detection as the chemistry and biology improve. Finally, we have demonstrated the ability to taste or detect aqueous contaminants and are currently developing means of refining this detection.