An investigation into the fidelity of incoherent optical matrix-vector multipliers with some practical solutions to the non-ideality of the optical sources

Many optical signal processors employing the basic Stanford matrix-vector multiplier have been described in the literature, but few discuss the practical problems encountered when attempting to make such structures perform with reasonable numerical accuracy. During the course of the authors´ investigations into such a device they have identified various significant sources of error. The principal of these is the non-ideal behaviour of the LED sources employed to input the electrical data in respect of their linearity and temperature sensitivity. In order to overcome these limitations of currently available LED sources they have devised and demonstrated a number of modified drive schemes. These will be described, together with an experimental assessment of their efficacy