Indeed... it depends on the design goals... however, the devil is in the details... Sorry for the long post below, but for the more costly, more precision, here is an example... In most DIY cases, the above PDF is just grand and perfect especially if it's design goals (including cost) are desired, which they probably are...
Nick Franklin said:
http://proharmonic.com/articles/AT78_OTB.pdf
Enjoy
This is a nice design, and the binning of the pots and finding that a bunch are within 0.5dB rocks; perfect for DIY, hopefully not too many need to be purchased to get the magical part, and if both decks of the pot are from the same batch it is possible to have tracking capability...
However, if balanced gear is in a noisy environment with common mode noise coupling on the cables, then a better matched resistor on each leg for CMRR may be desirable, although relying on that over proper grounding and system design... The good news is that this circuit in the PDF is verging on a T-network style of attenuator... Thusly keeping the pot somewhat "out of the signal path".... 1% tolerance gives only about -40dB of CMRR... 0.1% -60dB... 0.01% -80dB (rule of thumb 20*Log(%match), e.g., 20*Log(0.01) for 1% matching)... The higher impedance of most passive stepped attenuators does render the need for shorter cables typically...
I guess it depends on desired design goals... 0.01 ~ 0.02 ~ 0.05 dB match +/- is what the big guys do (Dangerous Music, Manley etc...)... I have simply found that using 0.1% tolerance resistors which are still cheap for small projects can get the design to match the "specsmanship"...
Years ago, the 0.1% resistors may have been more expensive and not justifiable to commercial engineering designs (and taught that way in colleges and tech school as "verboten" - do not use them, even if your calculations determine that a 10.74635223754673858568934674 k Ohm or a 1.0001 k Ohm resistor is what is needed; been there, got the T-Shirt) unless absolutely precision is required at that circuit node, and without it, it will not meet spec... of course if you're manufacturing/selling millions of pieces of equipment, and you can save 1 penny on a resistor, while keeping the sale price the same, you are saving 1 million pennies of profit margin...
In this case, transparency is just the design circuit topology; many op amps or completely passive, or tube/transformer or whatnot... Precision and accuracy, especially when talking of the resistors' tolerances, are two different topics... Precision is how repeatable the settings are on the front panel, in this example, and accuracy is usually approached by calibration (of the manufacturer's measurement and binning of their resistor products) and absolute target value being reached in the design, perhaps with calibration capabilities if designed in... If all resistors thermally track and age the same in a certain circuit topology and the ratio of resistance (like in inverting gain equation of opamps Av=-Rf/Ri) is more important than the absolute value of the resistors (within reason of course) then you should get the same setting on the front panel every time.
Passive attenuators have their issues when switching in different pieces of gear with different impedances (especially if the passive attenuator was designed for one specific set of I/O impedances)... If it is a stepped design the attenuation steps may not be what is advertised on the front panel when switching in different impedanced gear... for a pot, the same position of rotation may not be the same attenuation for all combinations of gear switched in... For DIY purposes, this may be OK within reason of course, especially if you're expecting the outcome...
So some manufacturers put a buffer amp after the passive attenuator to get the impedances to play nice and keep the advertised gain/attenuation settings on the front panel...
My opinion, if'n yer gonna put an amp in there, why not do gain/attenuation as necessary... Especially if your going to listen at a calibrated monitor level that's repeatable...
Here is a single ended attenuator/gain block, which I was toying with, but never built, not necessarily for monitor gain/attenuation listening, more for input gain/atten capabilities; 0.5dB steps +/- 6dB (I think)... Using two of these for a balanced configuration is probably not desirable and a T-network solution with a balanced set of opamps would probably be more elegant as in the above PDF but with precision matched resistors on each leg as desired (emphasis on the match for CMRR not necessarily the absolute value, although that helps)... The T resistor need not be matched as it is a singleton resistor, but repeatable is always nice, as resistors age and mechanical stresses can change stability... could be a pot as shown in the PDF design above....
The relays/switches in the opamp loop can be a rotary switch and would be way cheaper than all the relays (about half price) !!!! (I just wanted recallable and automated reference settings with a micro controller)... You'll notice that there are 12 relays, similar to a 12 position Grayhill rotary (mere coincidence... or was it?).... The 200k and 100k are there for the momentary open/closing of contacts in the switch giving a temporary -6dB when switching... The calculations were tricky for this because of the 200k and 100k resistors paralleling and took a long time in the spreadsheet (and getting values to be close enough to render 0.01dB to 0.02dB of repeatability with standard 0.1% available parts)... The 100pF caps may be too much depending on your BW requirements...
Essentially this circuit is a variable stepped "pot" (with only one relay closed at a time in the opamp loop) where the wiper is in the virtual ground and the two legs of the pot form input resistor and feedback resistor of the inverter opamp... The Atten/Boost switch swaps the legs of the pot from input to output and vice versa rendering different input/feedback resistor values relative to the "wiper" to obtain gain or attenuation of the same value based on the front panel setting.. R151 and R152 add some trim as the swapping between boosting and attenuating is not perfect with the 200k/100k paralleling and the swap... I use inverting opamps to reduce purported crossover distortions... Notice 2 inverting opamps maintaining polarity...
The first opamp provides buffering from balanced and a possible +10dB gain.... The ALT_GAIN switch is there for -0.25dB for those inclined for the resolution... All resistors are 0.1% tolerance unless otherwise specified...
The opamps are specified to essentially be similar to a straight wire with gain/atten, hopefully, under most linear circumstances... The resistor values were kept low... I try to keep similar parts/values being reused throughout the whole design so as not to have too many different part numbers in stock and manufacturing likes that so that we are not binning a bunch of different parts or changing mindset (and hunting for different parts/locations) too many times while stuffing the board - stuffing the wrong part number in the wrong location etc.; not so evident here but it was attempted...
Those were some of the design goals... However, the caveat is converting from balanced to unbalanced and back again if so desired....
Since I had a hard drive crash, you may want to simulate the resistor values to make sure these are the proper values... simply the loop/input resistors in the 12 relay spots and the paralleling 200k/100k should suffice... I think this was the last version of the schematic before the crash...
Cheers,
-chris