>appropriate range of attenuation. Messy, perhaps, but you know what
>these hi-fi types are like ;-)

Yep.


>I've puzzled over appropriate ways of doing this using the minimum
>number of relays & resistors. The 'logDAC' type semiconnductor devices
>seem to use a large R-2R ladder (like, 17 element pairs), giving 2^17
>possible settings, and then internally choosing a much smaller set to
>give the appropriate law. This seems a bit excessive in terms of relays
>even for me! (I'm looking for around 30 discrete attenuation steps)

How about something other than a R-2R network?  I've also had plans for a
stereo attenuator, but I wanted to have a binary weighted (in dB) network
of relays and resistors.  You could use 6 DPDT relays for each channel and
have 64 steps, so that would cover 64 dB in 1 dB steps.  (I was thinking
more like 128 dB in 0.5dB steps, but that's only 'cause I've got lots o'
surplus relays ;-)  I haven't built it yet--draw your own conclusions.)

I got the idea for the networks from none other than a Tektronix 'scope
book.  Each stage of the attenuator network (L pad, I think you'd call it)
must have the same input impedance, so you can simply string them together,
using the relays to completely bypass the unwanted stages.

Of course to be really correct, you'd also want to ac compensate the
attenuator stages, much like the scope attenuators.  Might not be a problem
with audio, at least not on the low and moderate attenuation stages.

Each stage of the attenuator would require one DPDT relay and at least two
resistors.  Unlike the R-2R networks, these resistor values are not
standard, and there won't be any in common among the stages.  Some values
might have to be series / parallel combos of 1% values if you wanted to be
really precise about it.  (Remember, you not only have to get the ratio
correct, but the series impedance needs to be fixed as well.  Makes
trimming tricky.)

I can't recall for sure, but I may have written a little program to
calculate the required resistor values for each stage, and it may have
calculated the max error at each setting for different resistor tolerances.


newell