Selecting Capacitors to Minimize Distortion in Audio Applications

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

bjosephs

Well-known member
Joined
May 10, 2021
Messages
104
Location
Massachusetts
Hi All,


I was googling around on capacitor selection criteria and found this little article from TI that looks to be only a year old. I didn't see it anywhere on this site so I'm posting a link. The author provides measurement detail on some different capacitor types/scenarios and includes mitigation techniques. The big takeaway is the correlation between dielectric permittivity (k) and distortion because high-k dielectrics, though capable of ever smaller package sizes, also vary in capacitance with applied voltage. Also included are mitigating design considerations that may I have been discussed before:
  • Increase circuit impedance so the voltage drop across the capacitor is made smaller
  • Reduce the -3db point as far below pass band as possible, also reducing voltage drop
  • In the case of polarized caps, apply a modest bias voltage to prevent rectification effects
Link: Selecting Capacitors to Minimize Distortion in Audio Applications - Zak Kaye
 
From that article " For high- performance applications, use C0G/NP0 capacitors when available and film capacitors when practical."

I didn't see any graphs of COG/NPO or quality film caps in that article.

JR
 
Agreed, that is an unfortunate exclusion. It was a great opportunity to back up a claim with measurements. Also they listed a few interesting samples at the bottom (Wima and orange drop) but don't show the results of any testing.

Regardless I found it very interesting because I've been looking over old Neve designs (BA283 specifically) and wondering about the impact of tantalum capacitors in one of the world's favorite preamp/eqs. I've always heard that tantalums are as bad as they get for distortion but clearly you can do worse with an XR5. Then by keeping the -3db point low (and adding a DC bias as the Neve's must in a single supply design) you can push it down low enough to potentially be negligible.
 
Tantalum were OK caps for 50 years ago,,, relatively small and low impedance . Their shortcomings are well known by now.
I guess I am not up on the world's favorite preamp/EQs. Are you suggesting it is because of tantalum capacitor non-ideal performance?

With a decent test bench you can measure for yourself.

JR
 
Tantalum were OK caps for 50 years ago,,, relatively small and low impedance . Their shortcomings are well known by now.
I guess I am not up on the world's favorite preamp/EQs. Are you suggesting it is because of tantalum capacitor non-ideal performance?

With a decent test bench you can measure for yourself.

JR
I’m suggesting that the shortcoming aren’t enough for anyone to have thrown their racks of 1073s in the trash or stop them from continuing to go up in price. Maybe the shortcomings are part of the sound (I’m not claiming to believe that, I really mean “maybe”) or perhaps the shortcomings were never actually impactful *because* of the implementation choices. I certainly wasn’t expecting to see a 15db drop in THD from just a few volts of bias for example. What others get from the article will vary.
 
The #1 one short coming of tantalum caps when used for PS decoupling is failing short circuit, causing spectacular events if not current limited. #2 was their know poor dielectric absorption. This was basically a concern in sample and hold circuits. Back in the 70s/80s audiophools used that to criticize tantalum. I used some back in the day for their relatively low ESR.

One of our respected forum members, Sam Groner did a deep dive into capacitor distortion years ago. hers's a link to his website publications page

Quadrature Bridge Measures Harmonic Distortion in Capacitors​

This article considers the measurement of harmonic distortion in capacitors using a bridge circuit. Each arm of the bridge consist of one resistor and one capacitor. The bridge is driven in quadrature, such that both the differential and common-mode output voltage are nominally zero. The article introduces a procedure to construct a capacitor with very low distortion contribution, which is used in the reference arm of the bridge. The resolution of the bridge is estimated to exceed −150 dB. Measurement results for various capacitor types are presented. These show that parts with ceramic C0G dielectric have more consistent, and generally speaking also lower, distortion than Polyester, Polypropylene or Polystyrene film capacitors.

Linear Audio, Volume 12, September 2016

This article is co-authored by Scott Wurcer.

As I recal Sam had to roll some of his own test equipment to get useful resolution floors.

JR
 
Make of this what you will wrt Tantalum capacitors as voltage rail decoupling devices. I've often specified and designed in AVX TAJ Series Tants into this function in scientific and industrial applications. Properly derated of course. Not aware of any failures except where installed with incorrect polarity.
And believe me people will shout when their factory or foundry is at a halt because of equipment failure.
 
Hi All,


I was googling around on capacitor selection criteria and found this little article from TI that looks to be only a year old...
Good info: but we'll trodden ground so that's probably why there's no attention on here. But yes - high quality dielectrics and low voltages across caps where practicable eg AC coupling as opposed to EQ filters.
 
Last edited:
I've always heard that tantalums are as bad as they get for distortion
That's not my experience. I've heard of (and experimented) dramatic failures with tants, but the consensus is that distortion is pretty good.
but clearly you can do worse with an XR5.
Using any of the low-cost ceramic caps for AC coupling (or filters) is bad practice.
Then by keeping the -3db point low (and adding a DC bias as the Neve's must in a single supply design) you can push it down low enough to potentially be negligible.
I use 100 uF polarized electrolytics for AC coupling circuits with >10k impedance (for a -3dB point at about 0.16Hz). That's exactly what is recommended in this article.
Then I establish the global LF response, with a film capacitor, located at a place where the noise build-up due to the increased impedance is negligible. Typically a 0.47uF with a 100k load results in -0.1dB @20Hz and a -3dB point at 3 Hz.
I've never been happy with adding DC bias.
 
Good info: but we'll trodden ground so that's probably no attention on here. But yes - high quality dielectrics and low voltages across caps where practicable eg AC coupling as opposed to EQ filters.
Indeed it's a recurrent subject, and what most people fail to realize is that these tests are made in conditions that no decent designer would put in practice, in order to make the measurements simpler and more explanatory. The usual conclusion is that only film caps are good enough, which is rather expreme.
 
+1 as I've shared ad nauseam, where capacitor non-ideal behavior can become a significant factor is in passive loudspeaker crossovers (because of current) and in series with mic gain pots (because of low resistance).

JR
 
I’m suggesting that the shortcoming aren’t enough for anyone to have thrown their racks of 1073s in the trash or stop them from continuing to go up in price. Maybe the shortcomings are part of the sound .
Tant distortion is indeed part of the overall distortion profile of the 1073, it's why we continue to use them in our 1073-alike products. If you substitute aluminium electros, you get less LF distortion, which is not the classic sound people expect.
I cover a lot of capacitor distortion characterstics in my book, including C0G.
 
Tant distortion is indeed part of the overall distortion profile of the 1073, it's why we continue to use them in our 1073-alike products.
I understand the decision to use tants when these products have been designed, but, along the years, was this choice questioned, as alu lytics became better and better?
Did Rupert keep on using them in his later endeavours?
If you substitute aluminium electros, you get less LF distortion, which is not the classic sound people expect.
Is it the only reason for keeping them tants? I know acquired taste is a strong motivation, but surely, some products that do not "feature" the "tant sound" have been largely successful...
 
Did Rupert keep on using them in his later endeavours?
Is it the only reason for keeping them tants? I know acquired taste is a strong motivation, but surely, some products that do not "feature" the "tant sound" have been largely successful...

No, Rupert Neve always designed for best performance, so as technology improved, he followed with it (solid aluminium was preferred for a while, as they last forever). But when someone buys a 1073-alike today, they expect it to sound the same as one the Beatles recorded on, so the tants remain.

The original reason for tants was low leakage and stability in unipolar circuits. As technology evolved from discrete to opamp, power supplies went from unipolar to bipolar, and there was no longer a great need for low-leakage coupling caps, so tants disappeared. It was the proliferation of ICs in the late 70s, rather than any sonic motivation, that pushed out the tant.
 
Last edited:
Tant distortion is indeed part of the overall distortion profile of the 1073, it's why we continue to use them in our 1073-alike products. If you substitute aluminium electros, you get less LF distortion, which is not the classic sound people expect.
I cover a lot of capacitor distortion characterstics in my book, including C0G.
Thank you Merlin!
I've been saying this about the tants in Neve's since forever. To little avail from some.

All they need do is measure the damn things. Besides, you'd probably hear a difference even if you were Mutt 'n' Jeff.
 
Last edited:
For what it's worth, I tried both aluminium electrolytic and tantalum capacitors in coupling positions in one particular Helios 0011 ("type 69") module clone I built recently. The original schematic specified tantalum, I assume for much the same reasons as Merlin sets out above. Both alu and tant measured within 1% capacitance and had the same voltage rating.

Objectively, the tantalum measured higher distortion than aluminium. I can't find my notes and THD sweeps now to back this up (thank you, hard disk failure), but on blind A/B testing I could consistently identify and preferred the tantalum as it sounded a little more, for want of a better word, "lively" in that circuit. It was subtle, but measurable and audible.

Of course, that's just a single datapoint (sans data) of a single implementation - and "better" is of course highly subjective and depends on your goals - but I'm no longer replacing tantalum coupling caps with aluminium on principle when refurbishing old gear now!
 
That's not my experience. I've heard of (and experimented) dramatic failures with tants, but the consensus is that distortion is pretty good.

Dramatic failures is a bit of an under-statement. Fire hazard liability risk is more accurate.

Really nobody is using regular tantalum capacitors for modern designs these days, except in very niche applications. If you use them in a commercial design, you'd better make damn well sure you understand their failure modes in the context of your application and you cover it thoroughly during product validation. Otherwise, make damn well sure you've got good lawyers on hand!
 

Latest posts

Back
Top