Hi,
Because the F-range is factor 500, is a simple integration not practicable...
I can think only on some peakdetecting/sample & hold, what is per (a free running oscillator)1KHz to zeroing?
Or if signal amplitude is enough high, you can count the pulses and after i.e. 10 reset the peak detector/S&H
K.
Added after 8 minutes:
Maybe a wideband (active) rectifier + SW-median filtering is the good solution?