Constant impedance relay-resistor logarithmic attenuator

[peranders]

Originally posted by x-pro


Yes - DPDT relays (like Omron G6K I've used) will do stereo w/o balance. Channel crosstalk was below -100 dB up to 20 kHz with these relays and a careful layout.

I made long time ago a T-link attenuator with 600 ohms impedance and it worked high up in frequency, don't remember the number but I think it was towards 1 MHz for low attenuation. I never came the control section.... but the pcb's are nice. I have the somewhere in my junkbox.

[x-pro]

Originally posted by janneman


No, thank YOU!

Alex, how do you proceed to calculate this? You select a load impedance first, then work backwards towards the input? I would be intrested in this with different values etc.

Jan Didden

Calculations are simple enough, however you can just change values from the diagram proportionally for any other impedance than 10 K

Originally posted by peranders
X-pro, you could also use T-link or Pi-link to get 1 dB, 2 dB, 4 dB etc or whatever step you choose.

Have you tested T-link?
Did your attenuator "click"(I mean electrically ) when it was operated? Did you have to trim the turn-off times for the relays(using special coil arrangements)?

Did you have a microcontroller or did you use a simple counter with a demultiplexor for the control?

How about the frequency response when you used 10 kohms impedance?

x-pro, do you have a nice formula for calculating a link?

Fred, you don't have worry. I'm only curious.

Check this out!
T-link attenuator with constant impedance


As I understand T-link is good for little attenuation and and PI-link is better for much attenuation? Is that right?

T and P - links require twice as many contact groups and more precision resistors . I did mentioned these in my article, thought.

There are some ways to illiminate "clicks" without doing anything special to the relays . I had the production version controlled by PIC which did NOT produce any audible clicks. Couple of early prototypes used just a counter and demultiplexor and worked fine, thought with some clicks.

Frequency response for 10K version was well over 100 kHz and depended mostly on the load capacitance. In a worst case the ouput impedance of the attenuator is about 2.5K and it can take quite a lot of capacitive load.

There is a formula to calculate links, however I'll leave it to you guys to work it out . That is DIY after all - it shouldn't be too easy...

Cheers

Alex (x-pro)

[peranders]

Originally posted by x-pro
There are some ways to illiminate "clicks" without doing anything special to the relays . I had the production version controlled by PIC which did NOT produce any audible clicks. Couple of early prototypes used just a counter and demultiplexor and worked fine, thought with some clicks.

Did you "tune" in any way the turn-on and turn-off times for the relays.

Did you use zero-crossing detection?

[Jens]

This little spreadsheet my help to refigure the resistors for different input impedance and/or attenuation blocks (like 1.25, 2.5, 5…).
BTW: Have a look here http://www.diyaudio.com/forums/showt...844#post120844

[x-pro]

Originally posted by peranders


Did you "tune" in any way the turn-on and turn-off times for the relays.

Did you use zero-crossing detection?

Zero-crossing is quite useless with relays - they're too slow for that. However it is possible to make the switching in a particular order which reduces the clicks down to inaudible level. Hint - "timing is everything" .

Alex (x-pro)

[Upupa Epops]

Most elegant and most expensive attenuator based on T or Pi you can do with " two ways " of signal trace : you have two identical attenuators in each cannel and in output of both you have relay, which is switched on way, which is not at this moment switched, while switched is other " way ". If there is switching stoped, relay switch on this way and switching goes on second etc.

[x-pro]

Originally posted by Upupa Epops
Most elegant and most expensive attenuator based on T or Pi you can do with " two ways " of signal trace : you have two identical attenuators in each cannel and in output of both you have relay, which is switched on way, which is not at this moment switched, while switched is other " way ". If there is switching stoped, relay switch on this way and switching goes on second etc.

It is not really nesessary - with the right relay switching sequence the clicks are inaudible. Even if you supply relays drivers directly from a CMOS counter you will have only an occasional click and I personally can live with this. A microprocessor control could be a bigger problem, unless you kill the clock completely when it is not required.

Alex (x-pro)

[Kay]

"this circuit is copyright of Creek Audio Limited, however it is already in a public domain for more than a year and half. "


Really?

I saw the circuit first at ELEKTOR 1991/7-8, page 23.

I post this circuit at July on

http://www.audiodiskussion.de/foren/....php?idx=37117

[Brian Guralnick]

Originally posted by x-pro


Here it is - as a scanned extract of my article in RadioHobby with a (hopefully) clear enough diagram:

Alex Nikitin (x-pro)

Thanks a million.

Originally posted by x-pro


1) It provides completely constant input impedance independant of the attenuation position.

Alex Nikitin (x-pro)

Really important.

Originally posted by x-pro


2) It gives absolutely precise logarithmic curve - the only limiting factor here is the tolerance of resistors. For 0.1% resistors the maximum error would be less that 0.02 dB over most of the attenuation range.

Alex Nikitin (x-pro)

Since I finally learned the math of db, I now see the importance of an audio pot being log. An example of my 256 linear step module going from 0/255 (-infinity db) to 1/255, it's already at -48db. That's if I did the math right.

------------------------------------------------------------------
a volume of 0 = ' 20*log( 0/255) ' = -(infinity) db.
a volume of 1 = ' 20*log( 1/255) ' = -48 db.
... ... ... ... ... ... ... ... ... ... ... ... ... ...
a volume of 254 = ' 20*log(254/255) ' = -0.03 db.
a volume of 255 = ' 20*log(255/255) ' = -0 db.
------------------------------------------------------------------

Originally posted by x-pro


3) It uses only one switchover relay contact for each of the binary bits, so a full stereo single-ended (or mono balanced) attenuator with 64 dB range could be build with only 6 relays (you may need one more for a complete mute) . You can have a 0.5 dB steps with one more relay and so on. Multichannel attenuators could be easily made this way.

Alex Nikitin (x-pro)

-When you say 6 relays, you mean, 64db range, inside 64 x 1db steps?
-When you say adding 1 relay, it will now give us still a 64db range, but, with 128 x 0.5db steps? What if we want a 128db range with 128 x 1db steps?

Originally posted by x-pro


4) It has reasonably low output impedance (not more than a usual potentiometer type volume control)

Alex Nikitin (x-pro)

In the magazine schematic, your input impedance is listed with 10k input impedance. Does this mean I should drive the attenuator with a 0 ohm drive & expect the load of the attenuator to be 10kohm?

Originally posted by x-pro


Each part of the attenuator (when engaged) always presents the same load to the previous segments and also needs to be loaded onto a same load - on the diagram it is 10 kOhm, however it is easy to change this to any reasonable value. The only requirement is that the eventual load, i.e. amplifier input, should present the same load as well. This way the attenuation of each segment does not depend on the position of any of them.

Alex Nikitin (x-pro)

What's the level of affect on this attenuator as the output load changes? For example, if I make 2 driving 2 amps, and 1 amp has a load of 10.1k, and the other 9.9k, how messed up will the balance be?
I guess if the output load is unknown, this circuit should have an added output buffer. Sort of kills the ability to use this circuit externally as a passive device.

Originally posted by x-pro


Something similar is used in RF attenuators for constant input-ouput impedance attenuation. However the usual "RF" approach requires at least 3 resistors and two switchover contacts per segment. My version is much cheaper and simpler.

Alex Nikitin (x-pro)

RF guys done this to match signal lengths & number of resistive elements as the relays are switched.

This is the 1 main beef I have about this circuit.

In my attenuator, the signal being fed through always appears to have 8 switches closed & another 8 open, it also appears that there are always 8 resistors in the signal path.

My final schematic makes this fairly easy to see:
http://pages.infinit.net/helloftp/attschemnew.png
Except now, I need to work out a log version.

[JoeBob]

Thew diagram does show 128 steps of 1db, I do believe (I've been known to be wrong ). If you wanted 128 steps of 0.5dB you would remove the 64dB relay and resistors and add a relay on the side of the 1dB relay and set the resistor values to attenuate 0.5dB.