I know where you're coming in from, but whichever way I look at Win's graphs I'm seeing a straight line for that 1N4148 at 0-30mv levels. Yes, the scales are log-log but the constant of proportionality is dead straight linear. I.e
1mv RF in gives 1e-10 amps and 10mV in gives 1e-9 amps and pro-rata for all points in between (you did see the double decade increments?). This agrees with the 10Mohm value that's marked on the graph. It shouldn't matter if the RF signal is swept over a 1 to 30mV range or just a 5 to 5.01mV range, the DC out will be directly proportional to the 'DC' in and no distortion of the waveform can occur, hence no dc offsets or harmonics. The graph next door though , the 1n5819, looks like it could offer up a tad of rf dB-I rectification. Though to my eyes it still looks way more like a resistor than anything with a square or log law response. There'll be better devices out there that offer (say) quadratic like classical responses at these low levels but it's moving out of the 'common or garden' playground. regards john