RJM Audio Sapphire Desktop Headphone Amplifier

I thought the whole point was to run the buffer closed loop, you know, keep op amp's good characteristics with increased power capabilities. Of course, I understand what you wanted to do in your project, but still...

Have you made any closed-loop measurements with considerably smaller quiescent current (<5 mA)?
 

rjm

Member
Joined 2004
Paid Member
That's the conventional wisdom, yet don't you find it just a little inconsistent? In audio lore, global negative feedback is generally considered a bad thing except, you know, when its a good thing, then suddenly everything is better for being in the feedback loop, and no amount of feedback is too much... wait, what? :)

Numerically, noise level would decrease if the buffer was closed loop. Distortion also, enough that the bias current could be decreased and the diamond buffer run in class AB. The argument for negative feedback is contingent, however, on those improved numbers correlating with a better sounding component. I think we've built up enough empirical evidence over the years to state as fact that this correlation may not always be present...

Negative feedback is a time (phase) delayed correction of the input by the output. The more feedback, and the more complex the nature of the correction, the more likely something is going to go wrong i.e. instability, error. For an op amp driving a gentle, resistive load, all the feedback does is correct for the inherent non-linearities of the op amp itself. Driving headphones, not only is there a lot more current (and thermal fluctuations) but the feedback has to work against the inductance and capacitance of the cable and voice coils, etc. Putting a buffer in there cuts the current output of the op amp itself - it only now has to drive the buffer input - but the all the corrections are still programmed into the feedback loop, and now, in addition, those corrections include any non-linearities in the buffer circuit, too. The feedback circuit is working a lot harder than it would if it was just wrapped around the opamp alone.

Now, maybe that's fine, there are no stability issues, and it sounds great. On the other hand, maybe keeping the feedback loop isolated from the load sounds better. At present, believe what you like and act accordingly. I have no proof to support my argument, except that the Sapphire sure is one damn fine sounding amp...

The boards have a convenient jumper to switch between open and closed loop. I haven't tried closed loop - running on faith, as usual - while Kevan ran into stability trouble when he tried it (go figure). So it remains out of the realm of practical verification for the moment.
 
Last edited:
But what if it cannot be measured?
Do you honestly believe that? It sound plain wrong.
Or, as is often the case, what is measured cannot be matched to what is heard?
That's where personal preference comes into play.

The thing with Sapphire is that if you run it open loop you are essentially making it a two-stage design, and if you close the loop you're back in one stage. If I were building my own interpretation of the Sapphire, I would keep the two-stage design, but I would add an op amp for the buffer stage. I should be able to reduce quiescent current considerably and still retain the same audio quality that you experience. Also, If I understand the concept correctly, phase-shift distortion should all but disappear, op amp having to deal only with the output buffer. Would I hear the difference? Probably not. But it would make me sleep better at night. :)
 
There's no compensation for personal preference.
Don't remember where I saw this, but must be one of those religious threads here.

I was always intrigued by the local feedback concept. While wrapping everything inside global feedback and pampering it till stable tends to result in lowest distortion, strangely when we make music, we deliberately introduce all sorts of distortion.

I've been contemplating at different output stage topologies that distorts most pleasantly, so far I am liking single ended emitter follower in moderately high class a bias, but it's always nice to have something running cooler and cuts my electricity bills.

Just wondering, how is the DC stability of this output buffer? I might rack one up if I can find some spare parts lying around.
 

rjm

Member
Joined 2004
Paid Member
@hwfanatic

Yes I do, in the sense that the correct metric is unknown. Concrete example, most people here would swear that different coupling caps sound different. OK, we could measure something, say, dielectric loss, that might be perceptibly different for different capacitors, but we can't say for sure that this is what we should measure to properly evaluate which capacitors sound better. I could measure, at work, the dielectric sprectral response accurately from subHz to GHz, and surely each capacitor would look a little different. But again, I have no way of connecting that information, however detailed, with any differences I might hear.

I rely on measurement and testing for troubleshooting, confirming the basic operating points, and a kind of pass/fail minimum performance benchmark. Once it's clear the circuit is running to specification and is basically solid, the rest is done by listening evaluation.

@nereis

Because the Sapphire's diamond buffer is doubly-symmetrical, using the same transistors fro the driver and output, the DC offset, and DC stability is incredibly good. The total output offset is mostly from the op amp stage, the buffer itself contributes very little. Likewise the transients on turn on turn off are no worse than the op amp alone.

I don't think we need to play the personal preference card or invoke any quasi-religious philosohy. THD is a very crude metric of the level of distortion, it gives equal weight to all the distortion peaks, regardless of whether the harmonic is even or odd, high order or low. It's a time invariant measurement using a simple sine wave input signal.

By analogy, it's like judging how fast a car can go by measuring the diameter of the wheels. There are other, better measurements, but eventually each and every one runs into the same fundamental problem: you end up chasing the measurement rather than using it intelligently towards the end goal, which, since we are in the business of making great sounding audio equipment, should be clear to everyone present: better sound.
 
AndAudio.com • ???? - Diamond Buffer ??

I personally understand little about diamond buffers but this Chinese thread may be of some interest. It covers the performance of a few diamond buffer topologies through simulation comparing the pros and cons, and proposes innovative use of current mirror (as per figure 21), note that D3/D8/D4 diode string is meant to be replaced by a 1.5V red LED in practice.

If you look at simulation (fig. 70) of circuit shown in figure 69 from the above article, the 4 transistor diamond buffer seem to be dominated by 3rd followed by 2nd & 5th harmonic. My ears tend to prefer lower order harmonic distortion dominated by 2nd followed by 3rd and less higher order harmonics, so my guess is that a diamond buffer utilizing CCS bias (fig. 77, 78) might better suit my taste.

Figure 145 shows that adding C2 (0.1uF) between the base of output transistors improved slew rate and step response by a good margin, so maybe this could be a worth trying trick?
 

rjm

Member
Joined 2004
Paid Member
Any push-pull circuit, even run in class A, will have a higher proportion of odd than even harmonic distortion unless it is deliberately unbalanced.

Comparing equivalent class A pp and class A single-ended, the difference is class A pp has lower even and odd harmonics, but even harmonics are reduced by a much greater extent.

ps. that link has a lot of circuits and figures to look over. When I tried it, pspice was a very poor guide for distortion, since all devices with the same part number are simulated identically (parameter spread not taken into account) so simulated distortion is vanishingly low unless the output signal is brought within a whisker of the supply rails - where it it hypersensitive and anyway an unrealistic approximation of the makeup of the distortion at lower, "normal" output levels.

Apart from the efficiency (which is truly awful) I am not sure there is any deficiency with the 2x2 transitor LH0002 circuit that needs to be addressed.
 

rjm

Member
Joined 2004
Paid Member
Alright, let's do some measurements!

Measurements are only as good as the measurement tools. Also, measurements are a total PITA.

I'm using my Onkyo SE-200PCI soundcard at 24/96. In-out loopback using a nice set of oyaide interconnects generates the following FT spectrum, for a -1 dB 1kHz sine wave.

2nd harmonic -100 dB. 3rd harmonic -110 dB. Noise floor below -120 dB worst case, typically -130 dB. THD+N is 0.003%.

So far so good. Problems starts happening if the R and L grounds are connected anywhere in the test loop, which of course happens in the case of a headphone amplifier since the headphone jack has only three pins. Then, a lot of switching noise garbage gets injected into the output for some reason... most likely a problem (possibly unavoidable) in the circuit layout of the soundcard.

The investigation, as they say, is continuing.

The images below show the soundcard alone, they are the baseline before I test the amp.
 

Attachments

  • thd.png
    thd.png
    3.9 KB · Views: 1,773
  • noise.png
    noise.png
    3.9 KB · Views: 1,726
Last edited:

rjm

Member
Joined 2004
Paid Member
Sapphire 1. Computer 0.

Only significant event tonight was blowing the line input of my motherboard. Not my Onkyo card, thankfully, I was trying the on board chip to see if it made any difference... 1 kHz sine test input through the sapphire and into into the line input at (accidentally) full volume ... and pop goes the audio chip.

I'm kinda impressed (and relieved) that the Sapphire amp was utterly unaffected by the incident.

Testing suspended. Time is better spent listening than testing.
 
Last edited:
After my HD650 cable upgrade with navships (ebay) cable I found the Sapphire a bit bright but it has now run in and sounds incredible. I can't believe how much more detail I'm hearing. I used to prefer listening to my electrostatics before but not any more.

Llistened to Nightfly (Fagan) last night, something I have listened to for yonks but I was amazed. Even sibilance is so well controlled and LP static and pops arent emphasized at all something that I couldnt handle with headphones before.

Thanks for the kit RJM!!!! Got it when I got back from holiday on Monday.
Do you think the change to Multicaps was worth it?

I'm now looking for a closed back headphone after I flog off some old gear and an AKG headphone amp.

kffern
 
Last edited:

rjm

Member
Joined 2004
Paid Member
Glad the kit arrived safely, and I look forward to hearing what you think of the new boards.

I'm personally rather surprised at the difference I heard between the two. Rev 10 was fascinating, hyper detailed and clean, while rev 14 is comparatively "normal", with fuller, more relaxed character. I'm still looking into why this might be so.

On Multicaps. I use them because I can afford them and because they allow me to forget about the coupling caps once installed. Cheaper caps are probably just as good, but there is always the lingering doubt.
 

rjm

Member
Joined 2004
Paid Member
33R. Exact value not critical. 20-40 ohms.

Under normal operating conditions it is impossible to short the output, or rather its impossible to damage anything by doing so since the DC offset is normally extremely small and the circuit is AC coupled preventing input DC signal from causing any problems.

I've also checked that transients in the output offset during power up and power down are too small to worry about.

This extra resistance is not to protect the buffer against a short, though it has that function, but rather to prevent damage to 16 ohm headphones in the event one of the output transistors fails.

The biggest risk is you build one with one or more transistors installed incorrectly (or a defective transistor) and neglect to test the output offset voltage before plugging in the headphones "just to try". If the headphone impedance is 300 ohms there's no risk anyway, but 16 ohms will draw enough current to melt the voice coils (and your ears!!). Once the circuit is working correctly the chance of a transistor failing suddenly for no reason is vanishingly small.

By the way this an issue for most DC coupled discrete buffer designs, not just the Sapphire.
 
Last edited:
so if I take the precaution to check the dc offset I shouldn't need this resistor mod?

I have just finish wiring up mine and I figure I will do a preflight check by letting it run for a smoke test without the headphones.. let it "warm up" a bit and then check the offset voltage.. if it passes then I will let it run a bit more.. check again before I plug in my headphone..

is that a good enough check?

-joe
 
An externally hosted image should be here but it was not working when we last tested it.


Just finished the wiring of the amp.

I will be making the outboard power supply this week..

An externally hosted image should be here but it was not working when we last tested it.


Going to mod this one I built before.. outputs dual 15vac with two tordoial with parallel secondaries and the vandal switch circuits

-joe
 
Last edited:

rjm

Member
Joined 2004
Paid Member
so if I take the precaution to check the dc offset I shouldn't need this resistor mod?

That's my opinion, yes.

I figure I will do a preflight check by letting it run for a smoke test without the headphones.. let it "warm up" a bit and then check the offset voltage.. if it passes then I will let it run a bit more.. check again before I plug in my headphone..

is that a good enough check?

I would say that is definitely good enough. Unless you have wired the transistors incorrectly it will most likely work fine from the very beginning, and stay that way.

In addition to the output voltage, also check the voltage at the op amp power pins for DC level and noise. You should see about 10-11V and ~zero ripple.

Btw that's a sweet looking build. If you used that case for the power supply, you could get two larger transformers in there if you removed the relay and extra circuitry. Though I suppose you'd still need room for the rectifiers.
 
After some dry fitting, etc.. I decided to go with the 1455T (same size as the amp side) for the power supply. It will allow me to use a pair of 25va transformers and have room for the wiring and rectifiers.

rjm, you mentioned I should check the power pins at the op amp.. which pins specifically? how do I check for ripples on my dmm?

-joe