Custom Attenuator Curves

[luvdunhill]

Hey, I was wondering if anyone might know of other attenuation curves, other than the standard log? What I'm thinking of is something that uses multiple logarithmic slopes, but I cannot really expect this to be a novel idea on my part. I have an idea that I'm working out, but was wondering if there was anyone else that has gone down this path? My one requirement is constant input impedance

My second question has to do with SMD resistors. I'm looking at the Vishay Dale TNPW and Susumu RG/RGH (due to vanishingly low tempco values), but was wondering if there were other lines out there you guys might be aware of. I'm focusing on thin film resistors at the moment (unless there is a reason I shouldn't?). My requirements are <1% accuracy and a large number of values under 1K and the ability to buy in quantities <= 10 or so. The large number of < 1K values requirement is probably the most stringent. Anyways, any recommendations here? Any other resistor lines that I should consider?

[pabbi1]

Scarily, Arn Roatcap @ Goldpoint may be the most knowlegeable, as this is all he does... then there are guys that do calculators, like Stepped Attenuator Resistor String Calculator - Neville Roberts, who may have considered these possibilities. Maybe the TwistedPear guys, though most everyone is just copying ALPS, Noble and TKD, or, so it seems. You would think there are more options than linear and log A, and I know HP does a lot of work with attenuation, though mostly NOT in an audio spectrum.

[luvdunhill]

Here are the resistor lines I'm currently considering. Any opinions on which might be the lower noise and lower distortion resistors?

http://www.irctt.com/pdf_files/PFC-COM.pdf

http://www.farnell.com/datasheets/3698.pdf

http://www.farnell.com/datasheets/43145.pdf

By the way, this statement is more or less "Due to special technology used to produce tight tolerance, low TCR at high values the RP73 resistor is not individually part marked"

[cetoole]

The IRC PFC series is what I use when I want a SMD resistor a bit fancier than generic 1% thick films.

[pabbi1]

In looking at the Goldpoints, now 3+ years old, it looks like they are SPC Multicomp (your 3rd link), based on the external visual with the usual caveats (my poor vision, etc), but no way to affirm without pulling one, which ain't gonna happen.

[luvdunhill]

so... looks like I'm making my own spreadsheet..

So a question. Anyone know a function in Excel that is similar to VLOOKUP but doesn't have this caveat "not_exact_match determines if you are looking for an exact match based on value. Enter FALSE to find an exact match. Enter TRUE to find an approximate match, which means that if an exact match if not found, then the VLookup function will look for the next largest value that is less than value."

I'd like it to return the closest value period. None of this next largest nonsense.

[slwiser]

so... looks like I'm making my own spreadsheet..

So a question. Anyone know a function in Excel that is similar to VLOOKUP but doesn't have this caveat "not_exact_match determines if you are looking for an exact match based on value. Enter FALSE to find an exact match. Enter TRUE to find an approximate match, which means that if an exact match if not found, then the VLookup function will look for the next largest value that is less than value."

I'd like it to return the closest value period. None of this next largest nonsense.

There is a function on the web that I just used at work that will do this. It was call something like InterpolateVLookup. Check it out. Cut and pasted from the function below is the heading...

Function InterpolateVLOOKUP(x As Single, Table As Range, _

YCol As Integer)

'Returns the interpolated number from the first column of

'Table and the YCol column of table.

'x - the number between two numbers in the first column of Table

'Table - The range of values with at least YCol columns

'YCol - the column of Table used for the Y-side of the interpolation

[luvdunhill]

Cool! I'll try it out! The below results are using the VLOOKUP function.

So, this turns out to give some unexpected results. The basic schematic is a series attenuator like this:

Notice I'm using a three slope curve, not the traditional log curve. The issue is once you get to the higher attenuation steps, things go crazy. I'm still trying to figure out why. At first, I figured it was because there are course E96 values < 10 ohm in film resistor line I'm looking at. But, then I just set the values < 10 to the exact theoretical value, and Exhibit A (atten1.jpg) is the resulting graph. Notice the deviation. My other theory is somehow the error from the series resistors above somehow catches up and throws the balance off... anyways, any ideas?

So, next I plotted the error between the theoretical value and the E96 values on another graph, see Exhibit B (atten2.jpg) and the non linearity at both ends of the stepper becomes obvious.

Now, if you sit and play with the lower values, you can actually coerce the values a bit, but then you get this strange deal with the first 2 steps, where you have a NOOP of sort and the volume doesn't change. This is rather strange, but maybe acceptable. See atten3.jpg and atten4.jpg.

Still working on some other features, then I can post the spreadsheet.

A few questions in addition to comments on the above:

What's the best way to provide a "mute" stop (yeah, stupid question I know)?

Is there any reason I cannot use the current sensing thin film resistors for values < 10 ohm, or alternately, how hard is it to mount smaller SMD resistors on 0805 pads, if perhaps they are available < 10 ohm in smaller case sizes?

[luvdunhill]

meh, I figured all the above out, plus how to move the non-linearity to the last 2 loudest steps.. even then, my max deviation is 0.50% That VB function didn't work for me, but writing an appropriate one was rather simple it turns out.

Okay, so next order business is a Bode plot for frequency response. Of course, Excel seems to be a fail on that out of the box. So... rolling up the sleeves..

[morphsci]

Bode and Nyquist Plots in Excel

[luvdunhill]

ah, a subject that's deeply been suppressed it seems. Rather than messing with complex math, I just calculated the magnitude 1/(SQRT(1+(wRC)^2))... more to follow!