Power Allocation for Parallel Two State Gaussian Mixture

We consider power allocation on a set of independent parallel channels affected by additive impulse noise, which is modeled by a two-state Gaussian mixture interference model. We derive the optimal power allocation that maximizes the total capacity subject to a total power constraint.

Moreover, we show that the well known water filling solution, which is optimal for Gaussian noise, is not the optimal power allocation for impulse interference. We also consider channel selection, i.e. to allocate all of the available power to a single channel, and show that the capacity is not necessarily maximized by the channel with the highest SNR in general.