Optical Signal Processors for - and Frequency - Adaptive Filtering

Optical signal processing concepts for the realization of time- and frequency-domain adaptive filtering algorithms are proposed. Specifically, four new optical signal processing architectures are presented, and variants on these designs are forwarded. Two of the optical signal processing systems proposed are time-domain implementations, namely (1) an implementation of the classic LMS algorithm for continuous inputs and weight functions, and (2) an implementation of the block LMS algorithm using a traditional acousto-optic time-integrating correlator and a novel acousto-optic tapped delay line filter. The detailed design, fabrication and testing of a 10-MHz bandwidth acousto-optic tapped delay line filter is described. This hardware realization uses a focused optical input to an acousto-optic delay line, and the resultant heterodyne output signal does not require an additional reference beam. The third and fourth optical signal processing systems implement frequency-domain algorithms. First, unconstrained frequency-domain adaptive filtering is achieved by taking blocks of time-domain signals and converting them optically into the frequency domain through Fourier or Hartley transforms. Once in the frequency domain, all processing is performed optically to generate the desired error signals and weight vector updates. The sliding discrete Fourier transform (DFT) algorithm takes a Fourier transform at each time step and is the last algorithm considered. An acousto-optic heterodyne spectrum analyzer is used with either electronic correlation cancellation loops or an optical system to perform the weight vector calculation. In addition to the presentation of the optical architectures, simulation of the algorithms is performed to assess the convergence of each approach. Computational complexity is also addressed, and it is shown that the frequency-domain techniques offer advantages in terms of both complexity and convergence speed, especially for large block sizes and correlated data inputs.