Adaptive compensation has been popular in a variety of biomedical signal processing applications. One of the major limitations of this method is the requirement for a primary signal + noise channel and a secondary noise channel, with the noises in the two channels being strongly correlated. In the majority of real-life biomedical applications this requirement can rarely be fulfilled since the secondary channel inevitably contains certain representation of the signal from the primary channel thus jeopardizing the adaptive compensation process. In the present work we describe a method for tuning an adaptive compensator containing a non-linear real-time multiplier in the reference channel in such non-ideal noise environment. The method is based on iterative adjustments of an exponential non-linear function which modifies in real-time the signal-to-noise balance in the reference channel in favor of the noise, preserving the correlation of the latter with the noise in the primary (signal + noise) channel but diminishing the signal content in the reference channel. Sets of model signals and noises are used to illustrate this methodology with a particular emphasis on biomedical signal applications. A quantitative description of the iterative tuning procedure is provided and the capabilities and limitations of this novel method for adaptive compensation are outlined.