Talbot Andrew wrote:

[snip]
Anyway, direct zero IF conversion will still need extensive input
filtering for anti-alliassing and in this case will be even more
stringent unless an artificially high sampling rate is employed.  For
example, assuming 80dB spurious levels, the input signal has to be at
-80dB at half the sampling rate separation from the edge of the
passband.   If a high speed A/D is considered - lets sample at 100kHz
for now - then we need -80dB at 50kHz away from the centre of the band.
(In fact, within this allias bandwidth are some very strong signals, so
the 80dB requirement may have to become 100dB anyway)  so we end up with
filtering more stringent than the double conversion approach.  And at
least to get rid of 1MHz signals, more filtering can be thrown at the
finished unit without degrading in band performance.
[snip]

Andy and the group,
               perhaps I didn't make myself clear. What I was suggesting was not
to digitize directly at 136 kHz.
My idea is along the following line :
Using a DDS with quadrature outputs (AD9854, if memory serves), feed the
two signals to two mixers, each of them gets also the wanted RF (in phase).
The DDS operates at the direct frequency you want to receive.
The two analytic signals generated, I and Q, are in the 0 - 3 kHz range,
while any off band signal will generated far removed frequencies.
So at this point you need just a couple of low pass filters to keep only what
you want. We still are in the analogue domain at this point, so no aliasing will
take place. After the low pass filters, you digitize with a low sampling rate, 
let's
say 8 kHz, the two I and Q components, then you enter into a dedicated DSP
chip for the processing you need (demodulation, filtering, FFT, AGC, etc.).
Using those chips I mentioned in a my previous post, this can be done with
minimal hardware, without the stability and temperature dependance problems
that would affect a purely analogue design (please don't shot me !).

73   Alberto   I2PHD