[morsel]

I created these graphs a couple of years ago but never posted them.

blue = linear 100K
red = log 100K
green = linear 100K shunted with 5K

(It was much worse with a 15K shunt.)

Now look at it graphed on a log scale.

Shunted pots present a variable impedance to the source in addition to not conforming very closely to a true logarithmic response.

 

[JohnFerrier]

Hi Morsel,

I had did the same thing in MS-Excel and reached the same conclusion...that shunted linear pots aren't close enough to work well. It really is not much different than a linear pot without the shunt. As you've shown, it's the log scale that best reveals this...and, of course, our ears work on a log scale...


JF

 

[morsel]

I should have used only 3 decades of logarithmic range instead of 5. 3 decades = 1000x = 60db, a good useful range for a volume control. However, the principle is still valid. I think I used 5 decades because Mathematica wanted to show 5 and I didn't get into it enough to change that, but hey, it was 2 years ago.

 

[00940]

The project on ESP 's website mentions than shunted linear pots offered a better log curve than most commercial log pots. This because the commercial log pots often are only log in name and are done with two linear sections put one after the other. I think a shunt resistor of 22K was adviced instead of 15K.

The varying input impedance problem is certainly a problem but not that much of a problem. Typically we only use half of the course of a pot, sometimes even less.

Anyhow, I've been using this arrangement recently, with great results, sonically speaking :

It has been discussed a long time ago in this thread, it was Tomo who came with it : http://www4.head-fi.org/forums/showt...t=shunt+linear

 

[morsel]

I made new graphs, this time with a 3 decade range. The Y axis now represents the change in resistance instead of a specific value, so it applies to any pot.

blue = linear
red = logarithmic
green = linear shunted with 10% of the pot value
cyan = linear shunted with 20% of the pot value
magenta = 2 segment log pot
black = 3 segment log pot

Segmented log pots are superior to shunted pots:

 

[sanaka]

Thank you Morsel, this is extemely interesting. The ESP site suggests the law fake resistor should be 1/6.67 (~15%) of the pot value, which would presumably put it's curve right in between your 10% and 20% curves, so I'm not sure where the ESP suggestion comes from. Also, only a 25dB range is graphed there, improving the appearance of the curve a bit.

To his credit, Mr. Elliot doesn't propound the law fake as the only best way, just as better than a 'standard' log pot, by which maybe he refers to worse pots than the Alps blue or even the Panasonic EVJ. Obviously, actual mathematical log is the 'most log' of any of the options. But how do say an Alps blue or EVJ actually measure compared to your ideal blue line?

Perhaps other benefits, like better stereo tracking, the wider availablitiy of good linear pots, are more pertinent to the why of using law fake. As long as impedance requirements are met. Still seems worthy of some listening comparisons to me.

00940: Your method doesn't change the pot's law and hence uses a log pot, yes? This is the method discussed at great length at DIY Audio on this thread. It's main benefit to me seems to be that the pot is pretty much removed from the signal path. Have you tried/found that using different types (carbon comp. vs. film vs. metal film etc.) of resistor across the input/output changes the sound much?

Peace,
Sanaka

 

[morsel]

The red line is the log line, the blue line is the linear line.

Pot shunting uses linear pots. If the pots were log to begin with they would not need to be shunted.

If the pot was removed from the signal path it could not attenuate the signal.

You are correct on the point of stereo tracking - it is important that the channels match closely to keep the channels in balance. It is also important to present a constant impedance to the source, which a shunted pot does not do.

As I write this Tangent is plotting the Alps Blue to see what the curve looks like.

 

[tangent]

Quote:

Tangent is plotting the Alps Blue to see what the curve looks like.

Here's the new data: Zipped Excel spreadsheet. This will render adequately in OpenOffice.org's Calc module, if you don't have Excel. OOo is free and runs on most all platforms, so there's no excuse for not being able to look at the spreadsheet!

The spreadsheet has curves for an ALPS RK27, the fake ALPS stepped attenuator, the Panasonic EVJ, the Radio Shack dual 10K audio pot by ALPS, and the DACT CT-2. This replaces my previous effort, pot-ch-matching.zip.

The test method: I put 10.00V across the pot with my bench supply, with the input and output lugs of each pot section tied together. I used 10V because I really wanted 1V, but 10V gave me better resolution with my meter; I scaled the data down by 10x when recording the data points. Then I measured volts at each wiper with a Fluke 189, using the millivolts setting up to the point where it dropped from 5 to 4 significant digits, at which point I switched to the volts setting. I made a scale in a vector drawing program with rays 15 degrees apart. I printed that out, glued it to a square of acrylic, and drilled a mounting hole in the center. With the pot attached to the scale and with a well-marked knob on the pot, I was able to get reasonably close to 15 degrees per adjustment.

Once I had the data, I made a manual best-effort at fitting a curve to the data of the form 10^(2x)/y, where x is a value from 0 to 1 proportional to the degree of rotation, and y is a scaling factor so that the high end of the resulting exponential (log) curve meets the actual data.

I didn't try to match the lower end of the curve because audio attenuators aren't made with a simple base-10 log curve. All five units I tested taper off at a higher rate than a straight log curve as you approach the point of greatest attenuation. This would be more obvious in the graphs in the spreadsheet if I had used a log Y scale, but that distorts the data so it's less intuitive to look at. This is also the reason why I chose a 40dB base-10 log curve: it matched the upper range of the units I tested better than anything else I tried. Since an amp should be designed so that the pot stays in its upper range most of the time, this seemed the best tradeoff.

Regarding the channel matching data, beware that only one sample of each type was tested. The average unit will have much better channel matching than its specs indicate, but it's possible for the actual performance to be much worse than the exemplars I used while not straying outside the specifications. Use this information to gauge relative performance of the units, not as an absolute guide to expected performance.

 

[morsel]

I made new graphs, this time with a 2 decade range (40dB) to match Tangent's research.

blue = linear
red = logarithmic
green = linear shunted with 10% of the pot value
cyan = linear shunted with 20% of the pot value
magenta = 2 segment log pot
black = 3 segment log pot

Segmented log pots are closer to true log than shunted pots even for a 2 decade range.

 

[sanaka]

Morsel and Tangent: You Rock. This seems to be some vital basic research.

Quote:

Pot shunting uses linear pots. If the pots were log to begin with they would not need to be shunted.

There are two methods of 'shunt' setup I've referred to above (and in that other thread).

One is the "Law Fake," as referenced (ad nauseum) on the ESP site, whose purpose is to supposedly approximate log taper with a linear pot, by paralleling the pot's output to ground through a fixed resistor. As you, Morsel, have shown, the graph of this is arguably not even an approximation. FWIW, Rod Elliot here asserts (scroll down for 'Changing the Law...') a max deviation from real Log of ~3dB, with a 100k pot and 15k resistor, over a 25dB range. I'm not sure how to interpret dB to check that against your resistance/rotation graphs. He also implies in several places that a good log taper conductive plastic pot is a better solution.

The other, more interesting, case is what I'd call a "shunt pot," in which the pot itself forms the paralleled path to ground (pot and resistor are in reverse locations from Law Fake setup). This is exactly what 00940's schematic on this thread looks like to me, and is the topic of the DIY Audio thread I've linked to. As you point out: Quote:

If the pot was removed from the signal path it could not attenuate the signal.

So I apologise for that misstatement. I should have said that in this case the pot is not in series with the signal path. My understanding is that this method does not change the taper of the pot being used, so if you want log taper you must use a log pot. What I find intriguing about this idea is that the series connection from input to output of the attenuator stage is a single fixed resistor.

As you've pointed out, that both of these methods present a varying impedance to the source should not be ignored. While I don't personally have an opinion on it because it's at the misty fringes of my small electronics comprehension, some seem to think that's not always necessarily a problem, however.

Peace,
Sanaka

 

[morgan jones]

I've tried fudging linear pots to give logarithmic attenuation too. The best approximation I could achieve was with a shunt resistor 1/6 of the value of the pot, but the results are pretty poor (see graph).

 

[tangent]

I've polished my previous post and the spreadsheet, and have added it to my site:

http://tangentsoft.net/audio/atten.html

Lots of new info and comments in the article.

 

[JohnFerrier]

Nice. Thanks.


JF

 

[sanaka]

Cool stuff as always Tangent, thank you.

I'd love to know what you and Morsel make of the other shunt configuration, the pot-as-shunt-to-ground-not-in-series-with-signal-path one.

Shooting from the hip and basically regurgitating stuff I've picked up here but don't understand to deeply: It seems, assuming someone didn't mind the deviation from log law and wanted whatever other benefits they could get, that the varying impedance problems of either a law fake or shunt pot could be fixed with some kind of buffer arrangement.

But then that must be weighed against whatever is perceived as 'bad' about just running a log pot in series...

I'm close to banishing further consideration of the idea, at least until I get more of a grasp on impedance issues in general.

I'm sure you all wanted to share in Sanaka's muddled neural activity like this...

Peace,
Sanaka

 

[sanaka]

I am close to banishing further thought about this, but not there yet. This thread has been the most informative and satisfying stuff I've read about the (in)famous law fake, but I have yet to read a similarly satisfying, dissenting opinion on the other shunt setup. So close though!! Morsel or Tangent, have you any thoughts about or impulse to analyze it? It would be so cool. Mahalo.

Peace,
Sanaka