3 Channel Headphone Amps & Virtual Grounds

[RocketScientist]

No problem. It's actually worth noting that almost all the advertised "3 channels" DIY headphones amplifiers works that way, at least those from PPL, Tangent or Amb.

I've updated the article.

I have one on my bench right now that does not. The headphone ground is the virtual ground used for the entire design. And, from what I've seen so far, it works better than the Mini3 does. But that's likely because it uses a much more robust virtual ground. I'm also still testing it so I may yet find problems.

 

[00940]

Iirc, jcx advertised such setups for situations in which a virtual split supply was necessary. I'll let him comment.

An old post that might be of interest to you: HeadWize: DIY Workshop > Another go on ground, current return and power caps

A problem with dual batteries setup is a pet peeve of mine though: inserting a fresh battery and a dead one in an amplifier is a very easy failure mode to achieve and can lead to fairly spectacular results... of the wrong kind. Some might say I'm paranoid, so be it.

 

[RocketScientist]

Iirc, jcx advertised such setups for situations in which a virtual split supply was necessary. I'll let him comment.

An old post that might be of interest to you: HeadWize: DIY Workshop > Another go on ground, current return and power caps

A problem with dual batteries setup is a pet peeve of mine though: inserting a fresh battery and a dead one in an amplifier is a very easy failure mode to achieve and can lead to fairly spectacular results... of the wrong kind. Some might say I'm paranoid, so be it.

Thanks for the link. The battery issue is real and I specifically address it in the article. I would not trust expensive headphones to the Mini3 or a dual battery design on a regular basis.

Even a single battery virtual ground amp can (and likely will) lose its DC operating point and put out DC when the battery drops too low. You can only hope the battery is so low by that point it doesn't destroy your headphones in the process. But I'm not willing to take that gamble over the long haul with a $500 pair of headphones. DC protection is the best answer.

Another answer would be a split rail DC-DC converter with a low voltage shutdown but that still won't save your headphones if the output amp dies.

 

[Earfanatic]

I think the source of these discussions was Your writing referring to an existing site and item directly, and not said like "there are circuits out there" or "some sites says so...". That would be more fair I think as there should be many more designs out there with the same errors and statements too.

As I enjoyed Your article i only suggest You to not point to someones work directly because they may find it offensive. I think this one sentence could sum this up pages before.

For me your article showed some aspects as I plan a headphone amp myself, but i'm not the guy with lots of cash and time, so I will only build one and no more. I search for design ideas and sometimes found pros and cons. For me your statements were Your opinion and I feel like it can be true, then follow it but there weren't any "must do - and - must do not" in that as someone said so.

 

[RocketScientist]

An old post that might be of interest to you: HeadWize: DIY Workshop > Another go on ground, current return and power caps

Reading through the thread linked above, it's the same thing. JCX is arguing solid engineering principals, and AMB doesn't want to hear it and concludes with:

I won't defend active ground topology any more because, as I said, I am not the first nor the only advocate of the idea, and it's all just a rehash of old discussions ad nauseum. Frankly, I am getting tired of the subject. It's up to the readers to believe, or not.

I'm not surprised he's getting "tired of the subject". He's way out on a limb. Thanks again for the link.

 

[RocketScientist]

I think the source of these discussions was Your writing referring to an existing site and item directly, and not said like "there are circuits out there" or "some sites says so...". That would be more fair I think as there should be many more designs out there with the same errors and statements too.

This is a good point but, in my experience, it's a "no win" situation. If I make only general references I get complaints about being "too general" and not citing specific examples. And If I cite specific examples I get slammed for "attacking" a particular design.

 

[jcx]

head-wize/fi, AMB's "3-channel" is different from "active gnd"

I have commented over the years
Started by jcx - Search
mostly discusses the issue

"active gnd" is a "supply splitting" method to avoid output coupling caps with a single supply amp (or the similar passive AC gnd + resistive divider)

above I posted a link to a "proper" connection of an "active gnd" with a "star gnd"


"3-channel" separates the input signal gnd from the output gnd - the output gnd is a "active gnd" usually using the same type of amplifier stage as the R/L channels

using the input reference (which may be a lower current "active gnd" like the TLE2426) to the "3rd-channel" as the signal input and feedback gnd and connecting the headphone gnd to the output of the 3rd-channel amplifier is a where I believe they've taken a wrong turn

the thread you (diditmyself) pointed to has other problems but I chose these 2 AMB posts which I think summarize the "3-channel" difference
Quote:

Originally Posted by amb
The ear is at the same time an extremely sophisticated instrument that it has an extraordinarily wide dynamic range and (coupled with processing in the brain) have uncanny capabilities such as picking out a single conversation in a very noisy room. However as a measurement device it is also crude and easily fooled by preconceived notions and psychoacoustics.

Your analogy of the stage play doesn't really work here, because the "performance" of the play is a subjective matter. If the goal of an amplifier is to accurately amplify what comes in (i.e., straight wire with gain), then subjectives don't really come into play. Either there is a difference in the reproduced signal or there isn't, and if there is, it should be measurable with proper instrumentation and methodology. Whether any such differences are audible is a whole 'nother matter.


so we now have a criteria that can be decided entirely on engineering terms - AMB votes for accuracy

unfortunately the last part of the following is easily shown to be wrong on accuracy terms when applied to "3-channel" topology:

Quote:

Originally Posted by amb
To relate back to the original subject of this thread -- with active ground and class A providing totally constant power supply currents, the capacitors on the rails are not called upon to do much other than keeping the impedance low.

I should also mention that constant PSU draw isn't the only benefit of active ground and balanced topologies. I've often posted about how not dumping load return current into ground eliminates signal ground pollution, but here is a fresh perspective of the synergy between active ground and class A operation that shouldn't be overlooked.



in the "3-channel" configuration the signal is amplified by the R/L amplifiers relative to the input signal gnd, the output current of the R/L channels passes through the headphone drivers and flows into the output of the 3rd channel - so the voltage appearing across the drivers is

input signal * gain - 3rd-channel output

3rd-channel output differs from "0" (= signal and feedback gnd) by the the noise, output Z*I_r+l of the 3rd channel amp and the distortion caused by 3rd channel amplifier nonlinearity while sinking that current

elsewhere "3-channel" advocates have argued that the errors of the 3rd channel output can cause distortion cancellation or other euphonic changes, tuning the "sound" - a frank admission of divergence from accuracy

my claim is that my illustration of "star gnd" shows that "gnd pollution" can be avoided by purely passive means - which doesn’t introduce a 3rd amplifier's noise, output Z or distortion and can be more accurate

The Pimeta’s "3-channel" topology is fundamentally flawed - most technical reasons given for it are simply not true or can be bettered by proper layout

"active supply splitter" can be useful when wanting to use single V supply - the single active gnd should be used for input, feedback and load gnd

but "3-channel" puts the gnd “buffer” amp in series with the load - with the input and feedback grounds separated from the load gnd - this means all of the error of the "output gnd" buffer is in series with the headphones - giving distortion and "output gnd" impedance crosstalk that is worse than with a good "passive" gnd of heavy conductor with signal input gnd, feedback gnd "star" connected with the load common ground pin of the TRS connector


A "active gnd" showing star at output - can better "3-channel" performance:

...

in amb's last post in the Headwize thread he finaly does the meausrement of "gnd contamination" crosstalk correctly and has to conceed that the TRS jack plug resistance does dominate
He tries to still claim superiority for his "3-channel" arrangement but without comparing aginst a well designed output star gnd

 

[RocketScientist]

Thanks for weighing in JCX! You do a better job at this stuff than I do.

I just wanted to put an article on my blog that (at least in the first half) tries to put it all in terms more people can hopefully understand. Promoting inferior designs isn't doing anyone any good, and at the least, those contemplating a 3 channel amp should at least understand all the trade offs involved.

 

[qusp]

Ok, perhaps not the best analogy, but I was trying to relate electrons to gravity. Both behave in ways that were largely well documented long ago. And, for the purposes of this discussion, the science is not open to debate. Linuxworks seems to think the basic rules of electricity are somehow open to discussion like it's some unproven "gray area" that nobody (least of all me) can be certain of. That's convenient for audiophile myths, but just not true in this case.

i take your meaning, Einstein's work may have predicted the existence of the Bose-Einstein condensate, but i'm not sure he would have predicted the manner we control them and exactly how deep the rabit hole goes

 

[DualTriode]

Hello,
A mechanical engineers’ perspective:
Yes water goes downhill and the water level in the manhole varies with the quantity of water flowing in. Sometimes the water backs up and sometimes you need a pump. The virtual ground is the pump.
The virtual ground does not reduce the impedance, it actively keeps the water level in the manhole constant as the pressure and quantity of the water flowing in fluctuates.
Laws of nature are documented not invented or broken.
Then there are lunar cycles
DT
All just for fun!

 

[regal]

Great work, now look at the the balanced headphone amp design nonsense.

 

[qusp]

exactly what about pulling a balanced output from a balanced source is 'nonsense'? he wont have any problems measuring performance difference there. i mean compared to what? an amp fed by a balanced dac that has its outputs folded down into a SE signal? you simply wont find a dac that performs at the top level that isnt balanced these days. some that arent of course sound very good, but measure them and se will = fail in all areas.

plus i dont know about you, but i loive in an area that has quite a bit of gsm activity and on public transport its unavoidable with se gear to be struck by gsm devices pinging the tower sending a locate signal, and with low impedance iems its not so great sounding, i do not get that since building a balanced transportable buffalo 2 with ackodac balanced iv stage with just a tiny bit of voltage gain (like 1.3x) cmrr to the rescue

 

[agdr]

crosstalk vs. crossfeed

Some headphone amps intentionally mix versions of one channel into the other, of course, to help eliminate the unnaturally extreme separation of headphone vs speakers:

Learning Center - HeadRoom Crossfeed | HeadRoom Audio

Learning Center - Headphone Imaging | HeadRoom Audio

Headphone amplifier - Wikipedia, the free encyclopedia

Maybe part of any crosstalk present with virtual grounds in headphone systems audibly acts as partial crossfeed. Thoughts? Seems like the various post-build posts from people who have owned and used virtual ground headphone amps are generally positive about what they are hearing.

 

[RocketScientist]

Maybe part of any crosstalk present with virtual grounds in headphone systems audibly acts as partial crossfeed. Thoughts? Seems like the various post-build posts from people who have owned and used virtual ground headphone amps are generally positive about what they are hearing.

This is an interesting theory, and one I've thought about. My first reaction is I wouldn't want any form of "crossfeed" or reduced separation that I cannot turn off. But that's what it is in this case.

And crossfeed is not just simple crosstalk. The kinds of crossfeed people tend to like are more complex. This would be a less desirable kind.

Just from a psychoacoustics perspective, it's possible some prefer the sound as they can indeed hear some of the crosstalk--levels around -40 or so dB are, by themselves, plainly audible. The question is if that amount of crosstalk is effectively masked in other ways--such as the acoustic leakage from the left earpiece to the right one. It's an interesting possibility.

In terms of sound quality my larger concern isn't the crosstalk, but the shared non-linearities. I showed the spectrum while driving each channel at a different frequency well below the clipping point. It's seriously, and audibly, ugly. Again, the psychoacoustics are likely complex. I don't know how the brain will process hearing distortion in your left ear that's related to the music in your right ear and will be at least partially non-harmonic.

It's possible the above "cross distortion" mechanism is perceived as a form of "euphonic" distortion to some people much like tube distortion. The right blind listening tests could at least shine some light on that issue.

If 3 channel audio was being "sold" much like tubes--i.e. sounding better at the expense of measured performance--that would be fine with me. I'd leave the whole issue alone. But objective claims of better performance are being made for 3 channel designs that range from highly suspect to genuinely impossible. That seems like something worthy of correction.

 

[jcx]

again; head-wize/fi, AMB's "3-channel" is different from "active gnd"

"active gnd" for supply splitting is a useful option for projects using a single battery or wall wart if you want to avoid output Caps

it is also claimed to be advantageous to balance discharge in a 2-battery circuit which may help cycle life with rechargeable batteries

active gnd wastes power - doubles current draw - some are willing to pay that price in small headphone amps to avoid output coupling C or to simplify power supply


but at some point the idea of "avoiding gnd contamination" by using 2 "gnds" separated by the active gnd buffer was invented - the "input gnd" being used for signal, feedback reference and the "output gnd" being used only for the driver return current

it should be immediately obvious that this puts any V that develops between the gnd buffer input and output in series with the headphones and will cause additional sound output - at best small enough to be inaudible

it is claimed that the "3-channel", 2 gnd scheme reduces "gnd contamination" by preventing the larger output current from flowing in the signal gnd

further claims include distortion cancellation or vague euphonic effects due to the matching of the gnd channel amp to the R,L output amps

even harmonic distortion cancellation in Bridged outputs is well known but doesn't happen in "3-channel" amps - any active gnd implementation requires the gnd amplifier to sink the sum of the R,L channels
in fact you get the new possibility of "cross channel IMD" - where single tones in each of R,L channels mix in the "gnd" amp output and give difference frequency IM products in both channels

a good "gnd" amp could keep the distortions and output impedance low enough to be inaudible - noise will always be added by the extra active device in series with the outputs


I claim that simple layout techniques manage "gnd contamination" with hiearhical clean/dirty gnd branches and star gnding principles - at higher efficiency, and accuracy, linearity using the standard dual polarity supply
if a active gnd supply splitter needs to be used you can still use just the "output gnd" for a the signal, feedback and driver return current with proper layout


the "3-channel", 2 "gnd" philosophy has been extended to desktop headphone amps with purpose built AC power supplies - AMB recommends "3-channel" Beta22 builds - using 3 Beta22 amplifier boards
the Beta22 may well be a good enough amp that the “3-channel” perversion isn’t audibly different from a good dual supply with proper gnd layout – but at a high cost – a dual supply is lots cheaper than a 3rd Beta22 build

at Head-fi this fashion even seems to overshadow "dual mono" (separate dual polarity supplies for R,L channels) - pretty much the ultimate in eliminating gnd contamination (if you use "Balanced" 4-pin headphone cable connector)



as mentioned there is little you can say about people's opinions formed in ignorance of or actively hostile to blinded, controlled subjective testing

but claims of objective, measurable performance superiority for the "3-channel", 2 gnd scheme are not supported by engineering analysis

 

[RocketScientist]

Thanks JCX. You repeated (and hence confirmed) a lot of my exact points but managed to be nicely succinct--particularly about the difference between "2 grounds" versus "1 ground".

but at some point the idea of "avoiding gnd contamination" by using 2 "gnds" separated by the active gnd buffer was invented - the "input gnd" being used for signal, feedback reference and the "output gnd" being used only for the driver return current

Based on what I've seen published, my best guess is this has its roots in high power bridged/balanced amps for speakers. Bridging only works with 4 wires, so someone decided something similar could "work" with 3 wire headphones. It's also evident there was some "trial and error" design where the 2 channel headphone amp used for the trial likely had poor grounding to begin with. That may have led to the erroneous conclusion ground isolation is a good thing when it's really just a band aid.

it should be immediately obvious that this puts any V that develops between the gnd buffer input and output in series with the headphones and will cause additional sound output - at best small enough to be inaudible

It's very obvious in my measurements and, with sine waves at least, it's also very audible.

in fact you get the new possibility of "cross channel IMD" - where single tones in each of R,L channels mix in the "gnd" amp output and give difference frequency IM products in both channels

The above is a huge red flag in the Mini3 measurements. Anyone who values accuracy should realize the shared 3rd channel makes things worse not better.

if a active gnd supply splitter needs to be used you can still use just the "output gnd" for a the signal, feedback and driver return current with proper layout

I have just such a design on my bench right now. It uses a robust rail splitter that serves as the ground for the entire device including the headphones. So far, it's measuring much better than the Mini3. It's far from an apples-to-apples design comparison but it still may help validate your point above.

Grounding is a difficult area to analyze because of all the parasitic values present. It's difficult to accurately model the entire grounding system of an amplifier so regular calculations and simulations only provide a partial answer. I think if there's any objective point of dispute here, it might be in this area--i.e. is there a problem worth trying to solve in a proper star ground 2 channel amp?

Guys like Douglas Self have chased all kinds of measurable flaws, including crosstalk, to vanishingly small amounts. And he's done so in high power amps with many amps of load currents using entirely conventional, but well designed, grounding. If you reduce the amount of load current by an order of magnitude, or two, as with headphone amps, I just don't see how it can even remotely still be a problem.

AMB recommends "3-channel" Beta22 builds - using 3 Beta22 amplifier boards
the Beta22 may well be a good enough amp that the “3-channel” perversion isn’t audibly different from a good dual supply with proper gnd layout – but at a high cost – a dual supply is lots cheaper than a 3rd Beta22 build

Some may think this is a cheap shot, but AMB does gets to sell more PC boards and parts for 3 channel B22s. And that thought has crossed my mind as to why he might be defending something so strongly that's rather hard to objectively defend.

but claims of objective, measurable performance superiority for the "3-channel", 2 gnd scheme are not supported by engineering analysis

My point exactly. Thanks again!

 

[wakibaki]

The battery issue is real and I specifically address it in the article. I would not trust expensive headphones to the Mini3 or a dual battery design on a regular basis.

Even a single battery virtual ground amp can (and likely will) lose its DC operating point and put out DC when the battery drops too low. You can only hope the battery is so low by that point it doesn't destroy your headphones in the process. But I'm not willing to take that gamble over the long haul with a $500 pair of headphones. DC protection is the best answer.

I think this issue is exaggerated. While of course it is not impossible for a battery to fail suddenly or for a user to connect a charged battery and a discharged battery, battery decline is not like throwing a switch, although again some types drop off comparatively quickly.

My point, however, is that when sufficient difference in battery output voltage develops to cause DC offset at the amplifier output, the offset will be small and the current drain on the batteries becomes differential and self compensating at that point, tending to draw down the battery with the higher voltage rather than that with the lower voltage. I haven't tested this (it's a bit awkward to arrange and prove conclusively, and anyway I don't care that much), but I suspect that the current will not reach levels likely to damage headphones.

Obviously I'm talking about dual-battery supplies.

w

 

[RocketScientist]

If you look at it from a statistical point of view, the longer the battery life, the less often the batteries will get low putting your headphones at risk. This puts the Mini3 at a big disadvantage with its typical 4-5 hour battery life vs say a headphone amp with 80 hours of battery life.

It's also worth noting commercial designs tend to do it right. Even the $20 FiiO E5 dies gracefully without so much as a loud pop. It just shuts off. That's probably a feature built into the Li-Ion battery management IC which also handles charging functions. The E7 gracefully manages the battery and even displays its level. And the TI chip amp is designed to power up and down gracefully including managing the internal charge pump that generates the negative supply rail.

There's no reason a DIY design can't use a similar power management chip that has low voltage shutdown built in. Or, for that matter, the same TI chip amp (although it might be tricky to solder).

 

[qusp]

there is a similar chip from linear tech that has lower quiescent current and imo better all round, i believe linear tech are the leaders in this area and it works well with their ldos, which are certainly the best and most prolific for audio due to low noise fast ldos like lt1963a, lt1764a. the supervisor chip is ltc2935 and i'm using it in my diy portable dac/amp.

its hard to solder by hand also due to the power pad, but if the pcb is designed with this in mind its possible to put a largish via under it in the middle of the pad to fill with solder and make your own custom land pattern to extend the pads out so you can solder the 'leads' which start on the side, but are under the package (its dfn style) similarly they have a bipolar ldo capable of +/-150ma i think from memory

 

[bpribadi]

I did some test similar to the topic of this thread. I built 2, 3, and 4 channels headphone amplifier, to hear the differences between them from the same circuit. I named this project: Pseudo Balanced Amplifier (PBA).

I found that 3 channels sounds better than 2 channels, and the 4 channels, the Pseudo Balanced Output sounds best, better than the 3 channels output.

I wrote a small article about PBA in HeadFi, more info about it here:
Pseudo Balanced Amplifier (PBA) - Head-Fi.org Community
(Earfonia is my nick in HF)

And a comment about PBA from my friend Mike (from Headfonia):
Earfonia’s Pseudo Balanced Amplifier (PBA) | Headfonia