Minimal Headphones Amp

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
Most will lack the required current.

NJM4556 are rated at 70mA output current into 150 Ohm, which is 10.5V peak. Few other Op-Amp's come even close. By comparison a 5532 only provides 38mA short circuit current (typical) less with load.

That's why I bought some NJM4556. There are two per channel so that's 140mA into 150 ohms. To drive the phones to 118dB SPL we only need 1V rms and 16mA rms current (16mW of power). Not a problem for the NJM4556
Downside, 3V/uS slew rate, 8MHz GBWP and 10nV|/Hz Noise.
Yes, but it is unity gain and pp output volts is less than 3V so at that level it can slew at a 1MHz.
Again, at unity gain 8MHz GBP is more than adequate.
10nV/rootHz noise at unity gain is 1.4uV at the output or -114dBu. I think I can live with that.

Next time I will certainly go with your BA640 approach. Looks like they took the BA440, stripped out the diff pair and VAS and replaced it with a 5534

Cheers

Ian
 
It's a headphone amp, not a PA power amp. 3V/us is more than enough.

Hmmm, this is always the argument we get, BUT, are we sure we do not have any signals passing into the circuit that are sufficiently fast? And is the slew rate actually indicative of design choices which even if the circuit does not enter slewing, compromise audio performance?

No problem for audio. They are used as unity-gain outputs drivers so 8MHz GBWP is also sufficient for audio.

Well, I disagree. The audio performance of the NJM4556 is well below what is needed for audible transparency in an HP Amplifier. It is sufficient only according to nwavguy who just picked out some random numbers out of this air and declared them to represent "ok for audio" and others like him.

Anyway, a modern audio grade Op-Amp that is unity gain stable and can output 100mA and sells for 3 USD in 1pc for a Dual from Mouser manages 24V/uS & 53MHz GBWP respectively and on top, UNLIKE the NJM4556 which has relatively high distortion AND strongly rising distortion with frequency above 1kHz manages -130dB THD @ 3V into 600 Ohm with minimum rise of HD at 20kHz to -120dB.

The NJM4556 on the other hand barely manages -110dB @ 3V and has a 20kHz HD of worse than -80dB. This is a direct consequence of the design which is in terms of technical parameters expressed as low slew rate and low GBP. It was this sort of behaviour I referred to by criticising the slew rate and GBWP as "relatively low", I should perhaps continue to declare it as relatively noisy and comparably high distortion with unacceptable large increase of HD from 1kHz to 20khz. And that of course creates big problems with CIFF IMD.

And yes, the limitations of 4556 are audible in the right context.

As I remarked, one possible solution (as the NJM4556 is 1/3rd the cost of the modern Op-Amp) is to use them as current dumper with a higher performance IC bottoming out the difference by acting as error correction Amplifier for the NJM4556.

Many other options exist that may offer superior results on less budget, or not.

Thor
 
Last edited:
That's why I bought some NJM4556. There are two per channel so that's 140mA into 150 ohms.

140mA into 75 Ohm actually, sorry to be persernicky.

To drive the phones to 118dB SPL we only need 1V rms and 16mA rms current (16mW of power). Not a problem for the NJM4556

That of course presumes the customer never changes his headphones.

Yes, but it is unity gain and pp output volts is less than 3V so at that level it can slew at a 1MHz.
Again, at unity gain 8MHz GBP is more than adequate.

It results in -80dB HD @ 20khz & 3V/100R Load. I personally find this at least borderline or past the borderline of audible transparency.

10nV/rootHz noise at unity gain is 1.4uV at the output or -114dBu. I think I can live with that.

Again, depends really on context and headphones, which are not fixed.

Next time I will certainly go with your BA640 approach.

Not exactly "mine" but Rupert Neve.

Who incidentally for his commercial headphone amplifier (the RNHP) elected to use the TI TPA6120 with what appears a 5532 as driver in a multi-loop (or "current dumper") style circuit.

Looks like they took the BA440, stripped out the diff pair and VAS and replaced it with a 5534

Yes, it mostly does. Personally I prefer the Biasing of the BA440 with a VBE Multiplier.

1694172440638.png

This circuit incidentally is quite interesting for using a dual feedback loop in 1974, way before they became fashionable. Slew rate incidentally is 7.3V/uS and is stated in the data, so this was already a concern in 1974, at least for RN.

TR3/4 form a 2-Stage Amplifier with a CCS load for TR4 (TR6) and an Output Buffer (TR5/7/8).

Input is to the base of TR3 and feedback to the Emitter (so yes, this is also "Current Feedback").

The feedback loop consisting of R7/R8 setting the whole output stage to a gain of 5.
TR1/2 form the "controlling amplifier", again 2-stage non-inverting with input to the base of TR1 and feedback to the emitter of TR1, so again "current feedback".

With a quick estimate the open loop gain (with the second amplifier stage loop closed) appears to be appx 105dB - all in all quite a respectable performance.

Incidentally, with a bit of squinting (and flipping the transistor polarities) you can see a fair bit of the NE5534 circuit structure here.

1694173669815.png

My own BA640 (with 440 output stage bias) derived take (some values omitted, but available on request) is this:

1694170437512.png

Output transistors are the TO-223 SMT equivalents of BD139/140. The LR buildout circuit makes sure Output impedance is low.

R41/R42 determine the quiescent current, the values depend somewhat on the current through R43/R52 which in turn depends on the actual PSU Voltages

Gain from unity all the way to 18dB. Noise can be lowered by reducing P1 from 100k to 10k.

Adjust C15 with the actual PCB for best small signal 10kHz square wave behaviour (minimal overswing at 0dB gain, good edges at 18dB gain).

Thor
 
Last edited:
Anyway, a modern audio grade Op-Amp that is unity gain stable and can output 100mA and sells for 3 USD in 1pc for a Dual from Mouser manages 24V/uS & 53MHz GBWP respectively and on top, UNLIKE the NJM4556 which has relatively high distortion AND strongly rising distortion with frequency above 1kHz manages -130dB THD @ 3V into 600 Ohm with minimum rise of HD at 20kHz to -120dB.
Which op-amp would you recommend?
 
Anyway, a modern audio grade Op-Amp that is unity gain stable and can output 100mA and sells for 3 USD in 1pc for a Dual from Mouser manages 24V/uS & 53MHz GBWP respectively and on top, UNLIKE the NJM4556 which has relatively high distortion AND strongly rising distortion with frequency above 1kHz manages -130dB THD @ 3V into 600 Ohm with minimum rise of HD at 20kHz to -120dB.
Sounds ideal. What is its part number and is it available in an 8 pin DIP package?

Cheers

Ian
 
Sounds ideal. What is its part number and is it available in an 8 pin DIP package?

Cheers

Ian
It might be instructive to see bench test results between the two parts in subject circuit topology. Better yet a null test when playing typical program (music).

Of course do what floats your boat. I don't need a headphone amp, so don't do any extra work for me.

JR
 
The NJM4556 on the other hand barely manages -110dB @ 3V and has a 20kHz HD of worse than -80dB. This is a direct consequence of the design which is in terms of technical parameters expressed as low slew rate and low GBP. It was this sort of behaviour I referred to by criticising the slew rate and GBWP as "relatively low", I should perhaps continue to declare it as relatively noisy and comparably high distortion with unacceptable large increase of HD from 1kHz to 20khz. And that of course creates big problems with CIFF IMD.

And yes, the limitations of 4556 are audible in the right context.
And what context would that be? I certainly wouldn't use it for high gain or where LN was important.

Genuine question as I've done a lot of DBLTs of OPAs in Jurassic Times.
 
Which op-amp would you recommend?

OPA1656/OPA2156 with a preference for the latter.

Here an illustration what the performance is like in an O2 style circuit:

HP-2

HP-2_BlockDiagram_600x600.png


It is a lot ahead of the NJM4556 at a - for DIY - pretty nominal price increase.

I do not recommend the "O2" style circuit, the "current dumper" I showed in this thread can be set up to offer much improved performance especially with lower load impedances.

Thor
 
Sounds ideal. What is its part number and is it available in an 8 pin DIP package?

OPA1656 - but it doesn't come in DIP8, it's SMT. Adapters are available.

It's my new "go to" general purpose audio op-amp, hard to beat in most applications.

And internal circuit structure is one that from my experience promotes "good sound" (folded cascode input, common source output).

Thor
 
And what context would that be? I certainly wouldn't use it for high gain or where LN was important.

It does incredibly poorly with lowish loads. My standard headphones are Fostex T50RP, magnetic planar headphones aimed at studio use and in continuous production since the 1980's.

These have ~ 50 Ohm resistive impedance and ~100dB/V sensitivity. And are much more transparent than any dynamic driver headphones I have experience with.

After small modifications (mainly adding damping material to the rest cavity and reinforcement of the cheap thin plastic of the whole acoustic system they come close to the transparency of STAX Electrostatic headphones.

By modern standards the T50RP is not "hard" to drive (that honour goes to DCA's Aeon) but it is a much more difficult load than most dynamic headphones, but they are much more of a workout for a headphone amplifier than common dynamic types.

Genuine question as I've done a lot of DBLTs of OPAs in Jurassic Times.

I am unsure what listening setup you used, what statistical analysis you used (ABX for example is worse than useless and cargo cult science) and what was the variable you were testing for, so I cannot comment.

In blind listening tests with fairly large numbers of subjects I was able to observe a strong preference for one specific OPA over others, despite THD & N of all Op-Amp's well below that of the source and in "non-stressful" conditions.

The same preference held when artificial stresses were introduced.

Over time at AMR/iFi we did many systemic listening tests (using preference in a somewhat complex questionaire) to find if among nominally similar items there was an observable and reliable preference for almost anything, from resistors (MELF Thin Film wins among all commodity options by.a huge margin, that surprised me) to active parts etc.

Some outcomes surprised me so much that I did re-runs to confirm that the results were reliable.

Note, my tests were always blind in the sense that listfners not only did not know identies of DUT's, but they did not know what actual variable was tested. The whole setup was tightly controlled for any possible biases, unlike the common "DB" tests widely promoted.

If the listener knows what is being tested and has any opinion or prejudice on the matter any DB test is impossible (placebo/nocebo effect).

Forced choice tests tend to create the equivalent of "examination Stress".

On the other, if I give you a permanent reference device that is arbitrarily assigned a score of 5 in each category (e.g. quality of bass, mid, treble, dimensions of virtual sound space, emotional engagement etc.) and I ask you to listen to multiple alternative devices (blind) and mark them for perceived quality AND at the end to give a preference marking overall (which one would you like more to take home) I changed the task and actually "gamified" the experience.

That leads neccessarily to radically different outcomes from common DB listening tests, which tend to be forced choice difference, which I find mostly useless anyway, as it gives me no actionable intelligence.

Add to that the incredibly poor statistics applied to the most popular ABX format (ABX) where a preponderance of a failure to reject the null hypothesis is designed in, naturally these tests return null results on all the strongest and easiest to detect audible differences.

I don't want this to get into another ABX is the gold standard vs. ABX is pseudoscience and intentionally biased towards null results debate which is OT in this thread.

If you want to debate DB testing, we should do so in a separate thread. You are welcome to open one.

Thor
 
Last edited:
It might be instructive to see bench test results between the two parts in subject circuit topology. Better yet a null test when playing typical program (music).

The "O2" which uses njm4556 is well documented with extensive measurements.

The Neurochrome HP2 uses de-facto the same topology (not exactly the same circuit) and uses OPA1656. It too is well documented with extensive measurements.

I have always found it difficult to set up a null tests using two different circuit boards, analogue, that gave a "null" deep enough to expose the predicted difference in HD.

I found it equally difficult to do it digitally in audiodiffmaker.

So what would that "null test" actually tell me?

Thor
 
It is a pity the OPA1656 and OPA2156 are not available from UK distribution. It is always a pian to have to go to Mouser/Digikey for small quantities.

Cheers

Ian
 
Yes. That's what I tend to do. Although in my current day job I don't do the layouts so applies to DIY only ATM and I haven't done a PCB for quite a long time now. SOIC spacing is pretty easy with good light/magnification and a decent soldering setup. Can do finer (forget the spacing but the finer pitched THAT Outsmarts device) with testing and solder wick to mop up.
 
The "O2" which uses njm4556 is well documented with extensive measurements.

The Neurochrome HP2 uses de-facto the same topology (not exactly the same circuit) and uses OPA1656. It too is well documented with extensive measurements.

I have always found it difficult to set up a null tests using two different circuit boards, analogue, that gave a "null" deep enough to expose the predicted difference in HD.

I found it equally difficult to do it digitally in audiodiffmaker.

So what would that "null test" actually tell me?

Thor
Perhaps it would tell us that there are diminishingly small improvements remaining.

A null test quantifies the objective difference between two audio paths. The better the two paths being compared are, the smaller this difference signal is. To realize a deep null, frequency and phase response of the two paths must be close to identical.

Design engineering involves maximizing performance while minimizing cost (for most of us). Pretty much any audio circuit can be improved, it's how some engineers amuse themselves.

If cost is no object we can always improve something on the test bench, but a null test provides us with some perspective. To evaluate null products I listen to the apparent sound quality of the null. Null products that result from phase response or frequency response errors are usually clean sounding audio. Null products caused by distortion will generally sound nasty, while some distortions are clearly worse sounding than others. It is common practice to add boost to this null signal so that should be factored into the evaluation.

In business there is pressure to improve performance specifications on paper to give marketing a story to sell...

JR
 

Latest posts

Back
Top