Michael Black wrote in
news:alpine.LNX.2.02.1410071314370.14980@darkstar. example.org:

If the input to the mixer isnt' well filtered, then I think you're going
to get all kinds of weird responses, since something can mix with a
harmonic of the oscillator and be converted down to the IF.

The best wheeze I came up with (never built it though), is a tracking filter:
a Schmitt trigger with a bit of positive feedback hysteresis to condition the
signal for a PLL and divide-by-100 based on two decade counters, to create a
stable clock signal for a 100X switched capacitor filter that takes a copy of
the original, direct signal. The output would be damn close to a sine wave no
matter what the input was like, but there is a clock signal that may need
removing, but possibly a simple one-pole filter will do that well enough.