Multi-modal distribution fitting to data that behaves like probability function

Ok, so I have a distribution of frequency dependent data (voltage versus 1/Time), which results from fitting an exponentially decaying signal with:

V(t) = sum( V0_i*exp(-t/Time_i) ),

and then plotting V0_i versus 1/Time_i in logspace to get the distribution.

I am trying to fit a multi-modal log normal distribution (i.e. a curve that is the result of several log normal distributions summed together) that fits the data using the above equation.

I've looked at gmdistribution.fit, but it is set up to deal with a random variable - in other words, the most common occurances in the data vector are considered the means. In my case the positions of the peaks along the MHz axis are the means, and the widths of each mode are the standard deviations.

Does anyone have any good ideas how to do this?

Thanks