Greetings from your newest member. I've been reading the board for a few weeks now and decided to take the plunge into DIY headphone amplifiers.

I have a question or two about Fig. 9 of the "Designing an Opamp Headphone Amplifier" write-up. It shows a MOSFET gate (driven by an opamp) being biased by a voltage divider without a coupling capacitor.

1. Is the MOSFET within the feedback loop of the opamp? I am guessing that it is not, but would like to be sure.

2. If it is outside the loop, wouldn't that configuration negatively affect sound quality? If not, why not?

3. Would that configuration (if outside feedback loop) give better performance than the same circuit with a coupling capacitor between the opamp and the bias tie point?

Reason I am asking is because the final stages of the current design I am working on utilizes an opamp buffer (OPA134) driving a source follower output stage. I am trying to keep the source follower completely out of the feedback loop and not use a coupling capacitor if possible. Should I go with the configuration in Fig. 9 and quit worrying so much?

Thanks in advance for any response.

zagnar

Yes, and I believe this is wrong and should be re-thought, perhaps removed from the article.

The voltage divider has no effect: the opamp output will drag it around to whatever voltage it wants to be at.

Since it is both flawed and incomplete, I don't see much point in poking at it.

> same circuit with a coupling capacitor between the opamp and the bias tie point?

That would at least let the bias divider work without the opamp fighting it.

If you don't mind the MOSFET inside the loop, there is a FAR simpler way to put an opamp and MOSFET together, DC coupled, no capacitor, and no bias trim:

Lead-up and comments at:

http://www.head-...60&pagenumber=1

> I am trying to keep the source follower completely out of the feedback loop

Why?

MHz stability may be a good reason NOT to put the MOSFET inside the opamp's feedback loop. But doing so makes the bias far simpler, especially if you wish to avoid an output capacitor. With manual bias, any goof in trimming or any drift in the MOSFET will burn your cans. With the simple overall feedback, the opamp reduces output DC to zero (typically a few milliVolts).

The MOSFET may be clean enough to work without feedback (though it is easy to find >1% THD, especially with resistor-coupled). And it will be mostly even-harmonic. But if you are already committed to an opamp, you are committed to massive feedback in at least one stage. Why miss the good benefits of feedback (especially DC control) in the MOSFET stage?

-PRR

The only stupid question is the one you should ask, but don't. Nobody is born knowing this stuff... everybody had to learn some time. To learn, and then to laugh at some new learner, is stupid.

However, please read the thread on Head-Fi. (This plan starts after a dozen other posts.) Many of these "silly" points were covered already.

(I don't know why that link didn't work for you the first time... it is however going to be slow to load because I've referenced a huge thread with many pictures, and Head-Fi's servers are not as speedy as HeadWize's.)

> Are you sure that the simple feedback path will keep zero DC on output to safely avoid the capacitor?

If wired and powered correctly, I am sure the output DC will be about zero. A few milliVolts with typical chips.

OH! And the source (preamp, deck, etc) better have zero output DC! This simple amp will amplify any DC on the source. If unsure, an input capacitor is a very wise idea.

> It is a theoretical exercise or you have built it?

Long ago I built something like it, but with a bipolar transistor. Elroy has built this plan and posted nice words on Head-Fi. He is now working on a comparison between several amps, though I'm not sure if he will get to this one.

-PRR

[Edited by PRR on 05-07-2002 at 12:28 PM.]

Yes, there is a missing coupling cap in figure 9. I will have to correct it. If you look in the Szekeres article ( http://headwize..../szeke1_prj.htm ), you will see versions of the MOSFET amp with ac and dc coupling between the MOSFET and the opamp.

PRR, in your schematic, what determines the value of the gate voltage? How does the opamp know to output x volts without biasing resistors either at the MOSFET gate or at the input of the opamp?

Thanks again,
zagnar

The opamp (+) input is biased to ground by the 100K resistor (plus whatever is on the input jack, which better be exactly what you want, and you usually don't want to listen to DC, since it is boring).

The opamp (-) input is connected to a divider between ground and the output. Whatever is at the output, 1/6th of that appears at the opamp (-) input.

The opamp "wants" to have zero volts between (+) input and (-) input. It will swing the output wildly until this happens. (Actually, it will swing the output thousands of times more volts than the difference between (+) and (-) inputs.) Assuming the connection gives negative feedback, and it is not impossible for the opamp to find a voltage for its OUTput pin that produces the desired voltage at the main output and the (-) input, it will settle with (+) and (-) inputs at the same voltage.

Assume that we put in Zero volts. We expect/hope-for Zero volts out.

Assume that this FET, with this load resistor/CCS, needs the Gate to be 4 Volts more positive than the Source.

In this case, the opamp output pin has to rise to 4 Volts to make the FET Source go to zero. At that point, the opamp will be happy.

If we replace the FET with another that needs 3 Volts Gate-Source for the same conditions, the output would be +1V. The opamp (-) input would be at 1V/6 = 0.166V. The (+) input is still at Zero Volts. So the opamp has 0.166V across its inputs, the (-) input being more positive than the (+) input. That will force the opamp output to swing negative. When its output pin hits 3V, the FET Source will be at Zero volts, the feedback voltage drops to Zero, the opamp now has no reason to change, and it sits there.

In fact we could stick in a bipolar transistor like TIP42 and the opamp would settle to +0.65 volts, just right for the transistor. Or we could put in a vacuum tube, one that could feed the load resistor with a voltage that the opamp can deliver, and it would bias right up. The loads needed for headphones are too heavy for any tube to handle at these voltages, but if the load resistor were 10K and you put in a fat triode with low B+, so the grid voltage was within +-10V, it would work.

With a perfect opamp, the main output would sit at zero. Real opamps have non-infinite DC gain, only a million. Since the opamp has to swing its output to +4V, there is an error of 4V/1,000,000 or 4 microVolts. This is insignificant. There is also a typical 1mV to 2mV offset voltage error in real opamps, and this circuit multiplies that by 6, so we expect 5 or 10 mV of output DC. There is also input bias current in real opamps. I have made no attempt to balance this out. FET-input opamps will show less than 1mV added error with the 99K imbalance I show. Many modern bipolar input opamps also have vanishingly small input bias current, nanoAmps. Something like 5532, the hot old low-noise workhorse, with as much as a half-microAmp bias current worst-case, may have a real problem. 10K input resistor would help, at the cost of low input Z and loading on any input Volume pot. But all the same factors affect all opamp designs that do not use a DC blocking cap in the feedback leg to ground or a servo (which is just an elaborate DC feedback loop). If it works in a Cmoy, it will not have excess input error or output DC in this design. (And even some amps that give DC trouble in resistor-biased Cmoy may work here.)

Assuming the opamp could swing +-10V from the +-12V supplies, for any FET required drive voltage from +10V to -10V, the opamp will find a point where the FET Source sits at Zero volts. (At least it will sit idle for FET bias from +10V to -10V... if you want audio too you better use an FET that needs less gate voltage. Fortunately, most are between +2V and +6V at currents we would use in headphones.)

Operating point: assuming +-12V supplies, the FET Drain-Source voltage is 12V. Assuming a 100 ohm resistor from FET Source to V-, the voltage across the resistor is 12V. The current in the resistor and in the FET is 12V/100 ohms = 120mA. If we use a CCS instead, it also gets 12V and it (and the FET) passes the current programmed into the CCS.

See, we simply say "We want zero volts at this point", and pick supply voltages and load resistance/CCS to achieve the desired operating point when that happens.

Drawback: if the opamp (powered by +-12V) can swing up to +10V, and the MOSFET Gate-Source voltage is 4V (very typical at these currents), then the Source can not rise above 10V-4V = 6 Volts. That's the peak positive swing. It is kind of low for a +-12V amplifier. And because the full 12V and the idle current pass through the FET, dissipation is about twice what it could be with a "perfect" amp. And 6V swing is not a really-high level in some hi-Z phones. So this is not the all-round perfect amp. But it is neat and simple, and Elroy says it sounds good.

{Zowie this new server is fast.}

-PRR

[Edited by PRR on 05-07-2002 at 06:21 PM.]

If anyone builds this DC-coupled MOSFET amp, please post details and pics.

[Edited by cmoy on 05-07-2002 at 10:52 PM.]

To my small mind, neither this nor most of the half-dozen other opamp/MOSFET designs should be lumped under Szekeres' design. His has a simple elegance that is impossible to beat. While some of the others start from Szekeres and elaborate, mine has been kicking around in various forms for decades. (I really wish I could find my Hot-Box amp and post it. I bet the date-codes on the chips are older than some readers.)

> I will add this circuit to the opamp article

Best go steal some of the text from Head-Fi to go with it. I forgot to mention supply bypassing and Elroy had big problems. I also said some things that I later changed my mind about. Elroy's comments were good, I suppose I should ask him if they may be quoted.

If you are leery of stealing words from another site, I guess I should write something myself. {sigh} But I really need to finish this Chapel Choir CD and collect my fee, before the choir goes on tour. (Anybody here in the Pacific NorthWest? Great girl-group coming to a town near you!)

-PRR

[Edited by PRR on 05-07-2002 at 11:16 PM.]

"Well, spent last night listening to the PRR-Szekeres (P-S) with a wide variety of material and in two different configurations; one with no gate resistor and the other with a 100 ohm gate resistor. The drain resistor was an 8 ohm 20 watt non-inductive wirewound and I used a constant current source with a 5 ohm bias resistor.

I did not use any input/output capacitors and the power supply was a +/- 16.40 Vdc unit based on Kevin Gilmore's ultra-regulated design. The dc voltage at output throughout the evening was basically fixed at -.006 vdc on the left channel and .000 on the right channel!

All listening was done on a pair of Grado 325's and the signal was fed from;

1) MP3's from my PC played through a Soundblaster Live!
2) Plextor CD/digital out-->Soundblaster Live!
3) Analogue out from a Theta/AA DTI 2.0/DDE 3.0 digital frontend

With the 100 ohm resistor in place, the P-S had a really, really smooth presentation with non-fatiguing highs and very good, tight bass. There were many times where I'd just stop what I was doing and say to myself, "wow, that sounds really great! I never liked that song before...".

I don't want to give the impression that the P-S didn't sound detailed, but the midrange on this baby kicks so much #$#% that you don't really want more high end info. The presentation is balanced and extremely enjoyable to listen to. The bass was tight and very clean. The U2 Mango remixes made the Grados vibrate like nobody's business and I kept lifting the earcup off my head to make sure that driving beat wasn't my neighbors banging on my door.

Again, I don't want to say that the upper frequencies were "polite", but they just didn't stand out -- again, its the liquid midrange that's the real deal here--it's something special.

The P-S with no gate resistor is a totally different beast. The high end on this version will open up so much detail that your ears will be begging you for mercy on badly recorded/low bit rate encoded tracks. The highs are so high that I kind of suspect some degree of ringing past a certain freq. The bass and midrange are very clear and transparent--maybe even a bit more so than the 100 ohm version, but its the detailed highs that will grab your attention. On some tracks this P-S will literally have you hearing things for the first time on tracks that you thought you knew intimately. Background vocals, mumbled words, etc., become clear in a way here that's downright funny. You'll be like, "Ohh...so that's what that word was...I've been singing it wrong for 18 years..."

On the other hand, the high freqs. can get a bit tizzy on tracks that already have a large amount of treble/treble distortion and believe me, you will notice this excess energy. While listening to Yo La Tengo's studio and live versions of "Cherry Chapstick" I had to dive for the volume pot as the guitar chords began the distort. There is a LOT of treble information here! On the other hand, the ambience of the Hoboken venue where their live version was recorded will have you turning your head to see if some audience members voice isn't really someone standing behind you with a knife/baseball bat/frying pan. I kept doing that till I got too freaked out and turned down the volume...

The bass, again, is punchy, powerful and extremely clean--this version of the P-S is a real hot rod. On some tracks, it will blow you away with the level of detail and transparency. On other tracks, you will hear every single treble note and probably hear it for a while, as well.

The P-S with the 100 ohm resistor, on the other hand, has a very good level of high freq. detail with that killer midrange. Wow. Talk about smooth. smooth, smooth. And liquid. As in drip, drip, drip--you can hear the raindrops fade into silence, wrapped in their raindrop wetness...And the bass, through the Grados, will slap you upside your head and toss you around the room. This is easily my favorite Szekeres version (again, not that the other designs are 'not good', but that this version just hits all the right notes for me) and oddly enough, is probably the easiest one to build."

My best friend collapsed last night and is in hospital. So I may be off-line for a few days (or may hang here more to keep my mind off his trouble).

> I kind of suspect some degree of ringing past a certain freq.

Dang. The difference between no gate resistor and 100 ohms should be all above 250KHz. Therefore my analysis is wrong, it may indeed be on the edge of instability, and any specific copy may not work depending on part and layout variation. And I may be the only handy person with the tools and experience to scope out the issues. So I better build one.

> the ambience of the Hoboken venue where their live version was recorded will have you turning your head to see if some audience members voice isn't really someone standing behind you

That's how good audio should be. I was so happy when an un-biased critic praised my small monitor speaker rig. On stage, someone's necklace broke and beads hit the floor. My dog was dozing near the monitors, but woke right up at that. He even knew which side of the stage to look at. The beads sounded like dog food kibble hitting the floor, and were so realistic that he was fooled. He has good ears and has no reason to flatter my speakers. He isn't the least bit interested in wolf-howls on TV, because he knows it is just a cheap 4" speaker. But my monitors had him convinced that something bite-size was hitting the floor next to him. On my best recordings, someone walking across the stage can be tracked step-for-step, and a coff in the crowd makes you want to turn and shush.

-PRR

I'm sorry to hear about your friend...I hope things work out okay for him. Maybe working with these circuits and all this electrical design stuff will help you from stressing out too much:).

So to give you some food for thought, I'm pretty certain that there is some sort of oscillation/ringing happening here. I rebuilt my original Szekeres the other day with all new premium parts as part of my comparison reviewing and there's a noticeable difference in "listenability" at higher volume levels. Specifically, when the PRR-Szekeres reaches a certain level, it feels as if there is some pressure (?) build up that makes it uncomfortable to listen to for extended periods. I know that when I mis-align my speakers, for example, I get a similar sort of nodal peakiness that has a similar effect.

If this is some sort of oscillation, do you have any suggestions on where to go from here? I currently have the IRFZ 46N mosfets instead of the IRF510's thinking that since they have a lower volatage rating, they might ameliorate the oscillation, and they do sound different, but they have a higher input capacitance than the 510's so I'm not sure what's happening here...

The PRR-Szekeres is a funny animal, after more extended listening, I get the sense that on some tracks, the amp sounds terrific, and then on others, it seems to suffer a bit of harmonic disonance. I think its a very good idea for you to build one yourself and troubleshoot this issue (if indeed, there is one in the first place!)

Again, best wishes to your friend and thanks--

From DC to 50KHz, the MOSFET is under strict op-amp control. This gives low distortion and output impedance, and strict DC control. Above 100KHz, the op-amp just gives a nice steady drive voltage and the MOSFET works like a simple cathode follower.

Rgate is necessary (as you found) and now may be 47 to 1,000 ohms. Don't go outside this range. Inside this range, there will be some interaction with the op-amp's internal output impedance, so there may be slight effect with different chips. 100 or 220 ohms looks best to me for most situations. I'd be curious if you hear any change with other values. (I think you should not.)

C5 and R4 are new. These are the key to MHz stability. These values should be good for most situations. Changes here should have little to "no" effect in the audio band, but large effects at radio frequencies.

I have changed the values of the feedback resistors so C5 and R4 don't load the op-amp badly. Gain turns out the same.

The LF411 op-amp and IRF150 MOSFET were just what I have in my simulator. I don't think you can even buy an LF411 today, and the IRF150 is far bigger than this amp needs or wants. Use a pretty-good (not fantastic) op-amp with GBW of 10 to 30MHz and less than 10nA input bias current.

IDC is an "ideal" current source. A LM317 would be used in real life. 0.2A should be plenty in most headphones, and output Z or linearity is not an issue. And this current directly sets the total heat to be gotten rid of. However if you expect to clip the amp with LOUD levels in low-Z phones, be generous. If the amp runs out of current on negative peaks, the op-amp goes crazy trying to urge the now-cutoff FET into reverse current, and recovery is messy for a microsecond after each clip. A big 30 to 50 ohm resistor will also work, but with more heat for less output.

V11, V12 are ideal signal generators; Rsource is to make them un-ideal. In real life your input jack or pot goes here. Rsource is not really needed in most cases, values from zero (short) to 2K or so would be typical good practice.

I have shown 12V power supplies but this is not critical. Stay way over +-6V or you'll get gross clipping at small levels. +-20V is far more than a 32 ohm phone needs, just more heat. For Hi-Z-only duty, you might want +-15V to +-20V supplies (watch op-amp rating!) and a lower current in IDC (say 0.1A to 0.050A).

I show 1,000uFd power capacitors. If your power supply is very good and very close, these are not needed. If in doubt, use 1,000uFd or more very close to the amp. While it is Class-A, it has huge signal currents in the power supply and the supply needs to resist changes.

-PRR

I just finished PRR-Szekeres with above scheme.

Here is my PCB board scheme. (1 channel)
Look! extremely simple.

With my humble ears and English, I can't describe sonic characteristics of this amp.
I feel more details and transparency than basic Szekeres.
But less smooth and less warm.
I am very satisfied with PRR-Szekeres now.
This amp becomes one of my favorite headphone amps.
Thanks PRR !!!

Following is my PRR-Szekeres specification and measured data.

Spec.
=====
- MOSFET: IRF510 (Fairchild)
- Heat Sink: Half of 9V battery size. --> Yes. so small.
- OP amp: OPA2604
- IDC(Idle Current Source) not used, I used 100 ohms resistor(Rs).
- Gain = 2 (R2=10K)
- Test with Philips HP-890, Koss Porta-Pro
- Power Supply: Convert 24V/1.5A regulated DC to +-12V (BUF634 Virtual Ground)

*** I have been tested with this BUF634 virtual ground circuit up to 400mA in my DC-coupled Szekeres amp without any(including sonic quality) problem. My SDS(180mA) also use this power supply. Surface temperature of this BUF634 is just 30~40 deg C even without a heat sink.

- Almost other components are same as PRR's scheme.

Measured Data
=============
- Power supply: +12.07V, 0, -12.03V
- Output DC offset: +2.5mV(Left) and +4.1mV(Right)
- Heatsink temperature at Idle: about 70~75 deg C (too hot)
- Total idle current of this amp. = 0.26 A
- Vgs: 3.42V(Left), 3.47V(Right)
- Vgd: 8.59V(Left), 8.57V(Right)

[Edited by sijosae on 05-19-2002 at 10:49 PM.]

BTW, where is your 2nd opamp? Are you using a dual opamp to save space? I still can't believe you can hand solder something that small...

Congratulations!

[Edited by boyelroy on 05-16-2002 at 02:08 PM.]

Your success in world 1st PRR-Szekeres encouraged me to build this WONDERFUL AMP. Many informations from you also VERY useful. Thanks a lot.

[Edited by sijosae on 05-16-2002 at 11:33 PM.]

Thanks again for all the information. I have removed the relays from my schematic in lew of a switch. With the money saved by doing so, i can now put in multiple levels of crossfeed. I will post pics and writeup as soon as it's finished. It may be a while as building time is scarce.

BTW: sijosae, kudos on your beautiful amplifiers. Its amazing how small, yet elegant they are.

Thanks again,
zagnar

I like that name: PRR-Szekeres Amp. I hope CMoy uses this name in the Projects section.

To PRR:
I'm sorry about your friend's sudden hospitalization. Maybe you can do something for him. I suggest that you check out the link below. It's the site of The Life Extension Foundation (LEF.org). Once at that site, check out the Disease Protocols. These are excellent, up to date nutritional guidelines for all sorts of ailments and diseases. Take a look...

LEF.org...
www.lef.org

LEF.org's Disease Protocols...
www.lef.org/protocols/index.shtml

Believe me, this site has helped many of my friends. It's A+! Good luck with it!

[Edited by Gariver on 05-17-2002 at 11:02 AM.]

Thank you and Elroy so much for the reports and beautiful pictures! It feels great when something seen only in my mind comes together in someone else's hands, and gives them some pleasure.

> With my humble ears and English, I can't describe sonic characteristics of this amp.

English is my birth language, I use it a lot. My mother was a journalist, my grandmother was a school teacher in the middle of the USA.

I think English is a terrible language for describing the fine points of sound.

However it may be better than most other languages.

They say people in Siberia have 10 words for "snow" of different kinds, and in Hawaii they have many words for "lava". But in English there are very few words to describe the sense of hearing, other than a few basic words like "loud". I believe this is true of most other languages. I know some of the "Latin" and Italian words used by musicians to indicate how music should be played, and there are a lot of these words, but they are not as precise as we could wish them to be.

Much more than French or Spanish, in English we can take a word that means one thing, use it another way, and hope that listeners will understand. In English, "smooth" means the feel of velvet cloth or your girlfriend's skin. But was can say "smooth amplifier" and people do not think the case is hand-rubbed like a Rolls-Royce radiator grille. And because all our senses go to the same brain and are processed together, the English way of using a word for "feel" to describe "sound" is similar to what really happens inside our head. But it is never certain that we all mean the same thing when we say "smooth amplifier".

OK, in English:

> I feel more details and transparency than basic Szekeres. But less smooth and less warm.

The basic Szekeres has some distortion, probably mostly even-order and low-order. 2nd harmonic and a little 4th and even less 6th. We often say "sweet", and sometimes "smooth" or "warm". It may have more than 1% of such distortion, and can probably approach 5% before it starts to clip and go very harsh.

It also has a rise in distortion, again mostly even-low-order, above the midrange, because the nonlinear Gate capacitance loads the input. A tilt in the shape of the distortion versus frequency seems to have a big impact on how we hear music. (Low-order distortion is almost the same as musical timbre; and many instruments play "emphasis" not by playing with more power but by playing with more harmonics.)

The op-amp, if done right, makes most of the FET's distortion go away. This might be "details" and "transparent". If "smooth" means "a little sweet distortion sounds better than the original signal", then the op-amp is "less smooth".

At some point we must decide: "perfect" or "better"? In my work I tend to prefer "perfect" amplifiers, because I don't want my recordings to sound better to me than they will to my clients. For most home listening, I use a near-perfect amplifier, but I know the old Macintosh SE BJT preamp must have a veil of even-order harmonics. And when I used to party, I liked the large lush distortion of a 15 watt tube amp pumping out 30 watts.

Getting back to the plan.... right after I posted the latest schematic I realized that it has very little feedback in the top octaves. The sound of the FET will come through, and it is possible that response will droop (or rise) above 5KHz.

sjosae, do you have equipment to test frequency response? I'd like to know how the response changes from 1KHz to 10KHz to 100KHz, with enough detail to see "small" 0.2dB deviation 1KHz-20KHz and any bumps above 20KHz.

If it droops 1dB at maybe 10KHz to 50KHz and then falls off, that's good but the high-end droop is audible. Changing C5 to 1,000pF may make it flat to 20KHz, but I fear it will bump-up above the audio band, or even just oscillate.

-PRR