Minimal Headphones Amp

[Spacecowboy]

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

Would you mind sharing the values for a +/-15V supply? I’d be interested in trying this out.

 

[thor.zmt]

Would you mind sharing the values for a +/-15V supply? I’d be interested in trying this out.

C15 - start with 470pF, fine-tune on the PCB.

R52 - 1k
R43 - 2k2

You are shooting for around 5mA (exact value uncritical) through the two resistors.

R41 - 2k2
R-42 - start with 2k2...

...adjust (temporary adjustable resistor) for the desired quiescent current by measuring the voltage across R39 with the load disconnected.

Install a fixed resistor once the correct value of determined.

Watch dissipation in the Transistors when selecting the quiescent current.

Options:

26mA is theoretical optimum class AB, 36mA RMS out in class A (1.15V @ 32R). Dissipation = 0.39W, can run without heatsink.

52mA is "warm" class AB, 72mA RMS out in class A (2.3V @ 32R). Dissipation = 0.78W, small heatsink required.

200mA is pretty much full Class A, 280mA RMS out in class A (9V @ 32R). Dissipation = 3W, each channel needs a 2K/W heatsink, using the casework heatsink usually can deliver this.

It may be preferred to use 2SA1837/2SC4793 as outputs for class A, these are isolated TO-220 "Fullpack" and with some thermal grease need a 5K/W heatsink per channel.

Generally speaking output Transistors should have a beta of >> 100 @ 100mA and an Ft of > 100 MHz @ 30...100mA

As long as this observed, use whatever you like.

With good PCB Layout the output stage is usually stable, if it is not, add base stopper resistors of 10R.

Thor

 

[ruffrecords]

For SMT to DIP adapters, ebay/AliExpress.

If making PCB's a few not excessively small SMT parts can be hand soldered, especially SOIC8 Op-Amps. I always put a SOIC8 pattern under any DIP8 footprint. More choice.

Thor

Unfortunately now I am in my 70s my arthritic hands are no longer capable of soldering much SMT by hand.

Cheers

Ian

 

[thor.zmt]

To realize a deep null, frequency and phase response of the two paths must be close to identical.

Yup. Often enough we are dealing with 1% tolerance resistors and 5% tolerance capacitors. Not easy getting a deep null between two nominally identical PCB's fresh off the line.

Then comparing different devices in these PCBs becomes a fools errand.

Design engineering involves maximizing performance while minimizing cost (for most of us).

I would adjust that to optimising the performance vs. cost/size/weight tradeoffs.

But even that I do not see in practice, I often see expensive and single vendor parts that simplify and speed up development, but offer no material performance benefits.

If cost is no object we can always improve something on the test bench, but a null test provides us with some perspective.

In my experience to get a pair of nominally identical PCB's fresh off the line to allow a meaningful null test is next to impossible.

To evaluate null products I listen to the apparent sound quality of the null.

An FFT can also be instructive.

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.

Getting sufficiently deep nulls to hear anything but products that are caused by phase, frequency response and amplitude errors is non-trivial.

What I find null testing useful for is to "clone" the sonics of a given product.

Everyone agrees Amplifier A sounds great. And Amplifier B measures better, but is considered to sound worse. We can "tweak" Amplifier B (likely) to replicate the sound of Amplifier A. In theory.

In practice, less so. But we are drifting OT again.

Thor

 

[Newmarket]

Unfortunately now I am in my 70s my arthritic hands are no longer capable of soldering much SMT by hand.

Cheers

Ian

Understood. I used to be able to solder SOIC with no spectacles or magnification. Not now

 

[ricardo]

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.

Agree with many of your comments but perhaps not your conclusions.

We did ABC tests rather than ABX cos you get 'statistical significance' a lot faster. 5 pages of pontificating on low sample testing deleted
Also you can test your instrument (your DBLT panel) by making 2 of the 'presentations' the same.
It's a huge topic. I could write a book on this but this is very unlikely as I'm now a professional beach bum.
We tested loadsa stuff including the appendix of AES E-Library » Intermodulation Distortion Listening Tests The paper isn't only about Intermod
and refined our methods over nearly 2 decades.
Early on, eg the AES paper tests, it was obvious we needed loadsa OPAs which were transparent. Much work on cobbling different discrete i/p stages to existing OPA. When the Mullard TDA1034 came out (we were Beta testers), we sent all that work to the Don't Recycle Bin. It was the first OPA we found to be undetectable. Some of our early efforts could be 'undetectable' but only under very specific conditions.

And BTW, for those who have never tried to do DBLTs properly, it is VERY $$$$

 

[JohnRoberts]

Hi Ricardo long time no talkie.....

hows the beach? It should start getting warm again soon.

JR

 

[ricardo]

Hi Ricardo long time no talkie.....

hows the beach? It should start getting warm again soon.

Hi John,

It should be getting warmer but the wind isn't dying down so no diving & fishing

I've had a couple of HD crashes but somehow, it doesn't seem important anymore.

 

[john12ax7]

Sorry if this was already discussed, but why not just use some discrete opamps? Plenty of drive current, some buildout resistors and you're done. Though not necessarily the cheapest solution.

One thing I've wondered, is AC coupling necessary to protect headphones? Have seen it done with and without.

 

[NOON]

Just dropping in to show what can go wrong with poorly designed headphone amps. This is from an SMPro 6 channel unit where the bass player managed to plug his lead into the headphone box instead of the DI. The engineer noticed when smoke started coming out of the headphone amp. Still stinks.

 

[thor.zmt]

Sorry if this was already discussed, but why not just use some discrete opamps?

Few discrete Op-Amps I know have sufficient drive current and low distortion into typical headphone loads.

Which one would you suggest?

Of course, most "Speaker Amplifiers" if not using IC's are "discrete Op-Amps".

And one could roll ones own. The Neve BA440 could be a starting point.

Perhaps a differential N-channel J-Fet + PNP input, current mirror into a N-channel MOSFET and a N-MOS instead of NPN as first stage of the output circuit, just to avoid to be boring.

One thing I've wondered, is AC coupling necessary to protect headphones? Have seen it done with and without.

I tend to add a dual mosfet switch using an 8205 Spec Mosfet per active line with an AC bootstrap circuit (makes a very linear ~ 40mOhm on resistance switch) and use the MCU I usually have in there anyway as DC detector to turn off the output until the output DC is within range.

DC fault protection is a good idea, overall it is best to view the headphone amp as a small speaker amplifier.

BTW, the same MCU tends to also get get a signal from a overcurrent detector tied into the output stage emitter resistors (meaning it triggers at ~ 0.6A for a single output pair in my designs) to stop short-circuits and other abuse from blowing out the output transistors.

Modern technology makes protection systems so much more reliable while being essentially transparent to signals and the ability to ignore short repetitive overload states that do not endanger function.

Thor

 

[JohnRoberts]

overall it is best to view the headphone amp as a small speaker amplifier.



Thor

+1 mainly a lower power audio amplifier...

My last headphone amp could drive loudspeakers (but I power limited it to protect the 1A wall wart's thermal fuse).

JR

 

[Tubetec]

Glad this excellent thread recovered itself ,
Pardon my having to get the gloves off and give someone a good verbal pounding ,
but its the only tried and tested method Ive found to stop gobshites in their tracks ,
I see the majority of what that individual said was redacted but the empty spaces speak volumes .

I would also like to add , despite Thor's sometimes bull in a china shop like approach ,
Cheers , and I'm sure I say that for the vast majority of people here .



I did plan on building a TDA 2030 based low powered speaker and headphone amp ,
sensitivity and noise werent favourable , I didnt need the all that gain ,

I was wondering about Johns statements relating to 'tricking the gain' of devices ,
Anyway an op amp could be wrapped around the TDA 2030 to give balanced input , lower noise and distortion , in a simple compact format and easy to diy ?

Looks like that part is going the way of the dodo .

 

[thor.zmt]

I did plan on building a TDA 2030 based low powered speaker and headphone amp ,
sensitivity and noise werent favourable , I didnt need the all that gain ,

TDA2030 is kinda old. I was younger than my GF is now when I first worked with it.

Better options exist. LM3886/LM4780 for example. Realistically, in 2023 Class D often is used.

I was wondering about Johns statements relating to 'tricking the gain' of devices

At issue is usually the minimum stable gain, more precisely the minimum.stable noise gain.

This can be "faked" with an RC circuit between +/- input. I have, for example, used this to turn a LM3886 into a unity gain "current dumper" combined with a low power tube amplifier as driver.

As usual in the current dumper configuration the driving amplifier supplies part of the actual current in the load.

Result? A 300B SE amplifier with 50W+ / 4 Ohm output.

Anyway an op amp could be wrapped around the TDA 2030 to give balanced input , lower noise and distortion , in a simple compact format and easy to diy ?

Not that trivial.

Can I recommend LM3886 instead?

Neurochrome has a ready diy kit:

Modulus-86 Stereo Kit

Note - I have no links with Tom, get no money from nor do I personally get particulary well on with him.

Or better, while they still exist, get some lateral FETs from exicon and use a bootstrapped Op-Amp (to overcome PSU Voltage rating limitations) as driver.

Slight modification of this:



This can make a very simple and stable/reliable Amp with great results.

Looks like that part is going the way of the dodo.

Most analogue power amp IC's are. They have a huge silicon area, need obsolete or obsolescent Fab processes and rarely perform materially superior to modern Class D designs.

We now have digital input class D Amplifiers with multi-MHz switching (TI), with multi level switching (Infinon Merus) and build in AD based digital domain feedback.

Combine these Technologies (so far still WIP) and include GAN based on chip power switching, on chip GAN based Buck PSU Switching (synchronized) et voila, a "fully digital amp" that can take on the best analogue ones, has high power and needs a very, very small PCB area.

High switching frequencies with Multi-Level outputs reduces output filtering issues and reduces the distortion rise towards higher frequencies.

Fully digital post filter feedback loops lineraise the whole system, all the way to "remote sensing" directly at the speaker voice coil.

Having an on board high power buck converter means at low volume we can run the amplifier with low PSU rails that in turn mean we get a greater dynamic range at lower power levels.

I find these all digital systems, if of sufficient quality, quite interesting.

Thor

 

[JohnRoberts]

Glad this excellent thread recovered itself ,
Pardon my having to get the gloves off and give someone a good verbal pounding ,
but its the only tried and tested method Ive found to stop gobshites in their tracks ,
I see the majority of what that individual said was redacted but the empty spaces speak volumes .

don't break your arm patting yourself on the back...

I would also like to add , despite Thor's sometimes bull in a china shop like approach ,
Cheers , and I'm sure I say that for the vast majority of people here .

I can't speak for the majority here, perhaps this is a skill similar to mind reading?

I did plan on building a TDA 2030 based low powered speaker and headphone amp ,
sensitivity and noise werent favourable , I didnt need the all that gain ,

It is common for such off the shelf ICs to have high nominal closed loop gain for stability considerations. A decompensated amp can deliver higher open loop gain and higher slew rates.

I was wondering about Johns statements relating to 'tricking the gain' of devices ,

A very old and I ASSume well known work around for decompensated op amps. Back in bad old days lots of popular op amps were not unity gain stable. The NE5534 is only stable for closed loop gains greater than 10dB (3x) . One well known "trick" is to add a RC to ground from the - input (or between + and - pins). This makes the amplifier think it is operating at a higher closed loop gain than it actually is by attenuating the NF at HF. The RC is sized to not have much affect down at lower audio frequencies so the apparent higher closed loop gain does not boost audio frequency noise.

Looks like that part is going the way of the dodo ?

for the same reason...

JR

 

[thor.zmt]

The NE5534 is only stable for closed loop gains greater than 10dB (3x) .

Incorrect. The NE5534 has external compensation pins and work just dandy and super stable at unity gain, AS LONG as the correct value capacitor is connected between pins 5 & 8.

Incidentally, if you connect a resistor from pin 8 to out and an equal resistor from pin 1 to ground, you can lower open loop gain by applying local looped feedback, creating a greater open bandwidth at the cost of lower open loop gain, which provides the requirements for stable operation.

Using an external input pair we can massively increase slew rate by running more current and by upping the current in the second differtial pair (and lower noise) and Pin5 allows us to run more current in the "VAS" stage (again, skew rate up, open loop gain and lineraity too) and add an external output buffer.

The 5534 is really quite interesting due to all the extra pins that give so much flexibility that more modern parts lack.

Then again, the correct more modern parts have such a higher performance baseline, that these extra tricks are not needed.

Thor

 

[JohnRoberts]

Incorrect. The NE5534 has external compensation pins and work just dandy and super stable at unity gain, AS LONG as the correct value capacitor is connected between pins 5 & 8.

wow you got me there

The 5534 is NOT unity gain stable and requires added external compensation to be stable below closed loop gains of 3x.

Back when, there were sundry popular op amps that likewise were not unity gain stable, trying to milk out the last drops of open loop gain and slew rate. I don't recall all the old part numbers but I think there were even some quad opamps that lacked access to internal compensation nodes.

Some very popular early op amps (like LM301s) described alternate compensation tricks (like feed forward in inverting mode). Back in the day we did not have the luxury of inexpensive, fast, unity gain op amps.

[TMI warning] Another trick to stabilize under compensated IC amplifiers. Back in the 70s Is used a then new phono preamp IC LM387. This IC was only stable for closed loop gains >10x. Not generally a problem for most phono preamps but even back then I was painting outside the lines to realize more improved RIAA playback response. The pro forma LM387 circuit as shown is a typical non-inverting topology but with an added C to ground from the - input to insure a closed loop gain of >10x at high frequency to prevent oscillation. Of course this is no bueno for HF performance. Instead of the cap from - input to ground, I added a resistor in series with the NF network between the op amp output and the typical NF network capacitors. My added resistor made the op amp think it was running at 10x but I took my audio feed inside the unit from the top of the NF network so I was getting roughly accurate RIAA response (inside my kit I was able to control the loading on this node to minimize any errors). Getting all tweaky even this improved approach diverged from the true RIAA 75uSec roll off at very HF because a non-inverting topology cannot go below unity gain. I corrected that in several later phono preamp designs even though the RIAA EQ technically does not care about what happens above 20 Khz (actually above 15kHz back when).

/TMI sorry about yet another veer...

JR

Incidentally, if you connect a resistor from pin 8 to out and an equal resistor from pin 1 to ground, you can lower open loop gain by applying local looped feedback, creating a greater open bandwidth at the cost of lower open loop gain, which provides the requirements for stable operation.

Using an external input pair we can massively increase slew rate by running more current and by upping the current in the second differtial pair (and lower noise) and Pin5 allows us to run more current in the "VAS" stage (again, skew rate up, open loop gain and lineraity too) and add an external output buffer.

The 5534 is really quite interesting due to all the extra pins that give so much flexibility that more modern parts lack.

Then again, the correct more modern parts have such a higher performance baseline, that these extra tricks are not needed.

Thor

 

[john12ax7]

Few discrete Op-Amps I know have sufficient drive current and low distortion into typical headphone loads.

Which one would you suggest?

The 990 can do 200+ mA into 75 ohm.

 

[thor.zmt]

Back when, there were sundry popular op amps that likewise were not unity gain stable, trying to milk out the last drops of open loop gain and slew rate.

Yup, I loved using the OPA637 (decompensated OPA627).

I especially liked it in I/U converters for multi bit DAC's (and phono stages, but that's less an issue) , these of course have a capacitor in the feedback loop so theoretically they always get to unity gain.

Add a cap four times (or whatever delivers correct noise gain) the value of the feedback cap from -in to gnd and magic happens. It actually helped no end dealing with extremely high edge rates from the DAC.

I don't recall all the old part numbers but I think there were even some quad opamps that lacked access to internal compensation nodes.

OP27 was unity gain stable, OP37 was decompensated. I remember an RC prefix part (who used RC, my brain is dead) that was a dual OP37 and absolutely smashing as upgrade in cases where a 5532 was operated with enough gain.

Some very popular early op amps (like LM301s) described alternate compensation tricks (like feed forward in inverting mode). Back in the day we did not have the luxury of inexpensive, fast, unity gain op amps.

Yup, today we get crazy stuff, compared to the 072's and 5532 of yesteryear. Passive parts also have now got options that are insane.

Thor

 

[Newmarket]

Incorrect. The NE5534 has external compensation pins and work just dandy and super stable at unity gain, AS LONG as the correct value capacitor is connected between pins 5 & 8.

Not incorrect though is it ?
It seems clear that the meaning is that the device is not unity gain stable in itself. It's commonly known that it can be made so by a single (22p IIRC) capacitor. As well as more involved ways as described.