• Disclaimer: This Vendor's Forum is a paid-for commercial area. Unlike the rest of diyAudio, the Vendor has complete control of what may or may not be posted in this forum. If you wish to discuss technical matters outside the bounds of what is permitted by the Vendor, please use the non-commercial areas of diyAudio to do so.

Aspen Headphone Amp

what's the psrr of the amplifier circuit? that will help determine if the regulation needs to be better.

mlloyd1
(who is also interested another nice headphone amplifier)

Knocked the regulator together and gave it the headphone test with 100 ohm load connected across the output. It is very quiet and no power supply hum is audible, but regulation is poor.

The output drops by 470mV when switching to 100 Ohm load and 700 mV when switching to 50 Ohm load. An error correction amplifier could improve matters but complicates the circuit considerably.

An LM317 may still be a better and simpler option.

Nico
 
Bigun, load regulation will be the same in any case unless we do one of two things, or both:

1: Put the 431 voltage divider at the output, introducing global NFB.
2: Use a CFP pass transistor configuration.

So what we're worrying about right now is line regulation.

I've got this huge post I'm about to post, so be ready...

- keantoken

I don't think I care about line regulation - but that might be ignorance:p

I may have to check the huge post in the morning, I'm in a later time zone than you I think... will look forward to it though.
 
Not really any technical reason, but that it adds some complexity where I'm not sure that it's needed. I see that it helps establish a current flow through a potential divider for the Z1 reference signal but I wasn't sure this was needed.

The other thing is, I'm still looking at this circuit as if it were an 'enhanced' cap multiplier in which a simple zener is replaced by the TL431. In this case a simple resistor works fine.

The reason I don't view this circuit as a voltage regulator is perhaps a bit over-simplified: If the TL431 maintains a constant voltage at the base of the pass transistor then the voltage at the emitter will depend on the current flow through the transistor. Besides, I can't see how it can keep a constant voltage because there is no feedback mechanism to tell the 'regulator' what is at it's output.

p.s. Sometimes I find that I learn more about a circuit under simulation by taking bits out and putting bits in.

The TL431 basically is a zener in its current configuration.

To update, I simulated the circuit with and without the CCS, and I didn't see a large difference in performance. This was kind of shocking. But not really surprising if you think about it. That thing only has .15 ohms ESR.

So this means that the main culprit is the pass transistor: Early affect (one of those things simulation will readily teach you). HF signals will also leak to the base of the transistor and pass through. This could be effectively dealt with by cascoding the pass transistor.

Here is my next version:

1: CCS was replaced with 2Q version. This one is soooo much better than the DQ versions. This change isn't really major.
2: CCS is made of 2N5087 low-noise transistor; same family as the MPSA18.
3: CCS is bootstrapped. This again helps isolate line noise.
4: all the above improvements make no major difference. :D
5: Pass transistor is cascoded through the LED (so we still geta power light). Line noise disappears COMPLETELY with this addition.
6: Load regulation is still not that great. This can be fixed easily by putting the 431's voltage divider at the emitter of the pass transistor. Then there would be nearly perfect line and load regulation. But this would introduce global NFB; is this something we want to risk?

So this is what I suggest.

1: Keep the CCS in just to be safe (it does improve line regulation insignificantly, not to mention varying current through the LED would cause wandering cascode voltage, possibly ruining the cascode advantage).
2: Cascode the output device. This destroys line noise. Also, we might be able to get away using even a cheaper, faster TO92 device for the pass transistor because it dissipates less than 1W of power.
3: If there is still not enough load regulation (it is actually quite poor by my standards), move the TL431 divider to the output, but put in a healthy amount of HF compensation to avoid HF peaking/BW-related distortion; this isn't a problem if our decoupling caps are good. I feel this is a good compromise for audio.

Good luck,
- keantoken
 

Attachments

  • Aspen_Headpone_PSU_K.png
    Aspen_Headpone_PSU_K.png
    22.3 KB · Views: 268
  • Aspen_hdph_4psu_K.txt
    7.6 KB · Views: 42
The TL431 basically is a zener in its current configuration.

To update, I simulated the circuit with and without the CCS, and I didn't see a large difference in performance. This was kind of shocking. But not really surprising if you think about it. That thing only has .15 ohms ESR.

So this means that the main culprit is the pass transistor: Early affect (one of those things simulation will readily teach you). HF signals will also leak to the base of the transistor and pass through. This could be effectively dealt with by cascoding the pass transistor.

Here is my next version:

1: CCS was replaced with 2Q version. This one is soooo much better than the DQ versions. This change isn't really major.
2: CCS is made of 2N5087 low-noise transistor; same family as the MPSA18.
3: CCS is bootstrapped. This again helps isolate line noise.
4: all the above improvements make no major difference. :D
5: Pass transistor is cascoded through the LED (so we still geta power light). Line noise disappears COMPLETELY with this addition.
6: Load regulation is still not that great. This can be fixed easily by putting the 431's voltage divider at the emitter of the pass transistor. Then there would be nearly perfect line and load regulation. But this would introduce global NFB; is this something we want to risk?

So this is what I suggest.

1: Keep the CCS in just to be safe (it does improve line regulation insignificantly, not to mention varying current through the LED would cause wandering cascode voltage, possibly ruining the cascode advantage).
2: Cascode the output device. This destroys line noise. Also, we might be able to get away using even a cheaper, faster TO92 device for the pass transistor because it dissipates less than 1W of power.
3: If there is still not enough load regulation (it is actually quite poor by my standards), move the TL431 divider to the output, but put in a healthy amount of HF compensation to avoid HF peaking/BW-related distortion; this isn't a problem if our decoupling caps are good. I feel this is a good compromise for audio.

Good luck,
- keantoken

I will have to study this further.

[You might also be surprised about how little benefit you see in the sims from the common mode choke too.]

We also have to look at Tranformer Options, at least I had a quick look at Digikey...

If we go with an outboard transformer in favour of a wallwart then there are some nice AC options (we want to keep the rectifiers and filtering on-board I assume)

For example, 24Vac with 500mA (Energy Star): $14

http://www.cui.com/pdffiles/EPA240050-P5R-SZ.pdf

If we stick wth a pcb-mounted transformer inside the box: IF-10-40-ND
this gives 20Vac at 500mA, $21
 
Okay, here is a more useful PSRR plot: watts into 32 ohms. This is exactly how much power goes into the headphones, so for 100db/mW, we really need some sort of regulation.

EDIT: Oh, and Bigun, I drew and use a different inductor symbol, because I absolutely abhor those pathetic quirly loopy-loop scribbles that take up way too much space :)faint:). So no problem, I'll just readjust the inductor positions. (just a style clash, that's all :))

- keantoken:
 

Attachments

  • Aspen_Heaphone_PSRR_Watts_32ohm.png
    Aspen_Heaphone_PSRR_Watts_32ohm.png
    10.8 KB · Views: 247
Last edited:
Good idea, we could still go all-out shunt. In the end it's Hugh's decision... But shunt is less efficient. Should we save heat and possibly money by using a more efficient PSU, therefore using a smaller transformer?

I have another big post coming, then I'll probably just wait for replies.

We could go Salas style, but that would be cheating... :)

- keantoken
 
Okay, to summarize:

1: Bigun pointed out that line rejection isn't major, which probably makes the output cascode not so important. (the pass cascode makes line rejection below -105db though!). Nico's tests seem to confirm this.
2: Load rejection really needs help (though we haven't quite decided whether load rejection is really important for this circuit or not).
3: I don't see a problem with using a CFP style output, but others seem to really not like the idea. So the most probable method for increasing load rejection would be to put the TL431 divider at the output, which would give better results than the CFP anyways.

Here are the specs for my version of the regulator:
Line rejection: below -105db across the audio band.
Load rejection: (I'll put this in terms of ESR since the AC simulation won't work right).
.37 ohms of ESR.

With GNFB, the load rejection goes to 1.7mohms of ESR, equivalent to about-55.5db. Attached is the schematic I used. GNFB strips out any chance of capacitance multiplier affect, unless we bypass R3 with a capacitor (which may add odd transients during start-up, but would also increase load rejection). C4 ensures slow start-up, but still adds a little more loading to the TL431 through the 220 ohm resistor (doesn't produce a large difference).

I apologize for not being present during day hours. This is largely because of school, but I also haven't been able to get to the computer so much.

- keantoken
 

Attachments

  • Aspen_Headpone_PSU_K_GNFB.png
    Aspen_Headpone_PSU_K_GNFB.png
    24.9 KB · Views: 100

iko

Ex-Moderator
Joined 2008
Realistically we're not talking about half an ampere here. Is an extra 125mA too much of a price to pay for outstanding line and load regulation, from a very simple circuit, and parts that are easy to find? Very wide bandwidth too, not that it would be necessary.

Of course, it comes down to personal choice. This would be mine.
 
Hi guys,

I think it's great that we have such a lively discussion and some outstanding ideas. I'm learning more and more every day from all this. It seems a pitty that there has to be a final design - I wonder if we will eventually have a chance to try out more than one of these ideas :)

The circuit Hugh proposed (which I posted a couple of days back) is the default plan, based on the common mode choke, single pass transistor backed up with the TL431. It's also the circuit which has been prototyped and verified as 'good' by Nico. But there's value in pulling it apart to understand it and value in looking at options for improving it.

The key for me, as others have pointed out, is understanding what we need; otherwise we will over-design this thing. The test results from Nico were key for me - the idea that you can't hear any noise (hum or otherwise) on the power rails seems like a very good test for psrr. Perhaps we should ask Nico to remind us of the test conditions just to be safe - e.g. was the load impedance low enough (23 Ohms) ?

Whilst the pcb could be laid out to incorporate the TL431 this doesn't mean everyone has to use it. Same story for the CCS. One can leave these parts out and by proper placement of resistors can instead end up with a simple cap multiplier - so it allows us to play if we want to.

So if ripple is a non-issue I also believe that channel-channel crosstalk will also be a non-issue because isolation improves over the audio band. That leaves only the question of how much line regulation is needed. I assume that the concern here is the amplifier pulling down the rails and causing distortion of the signal - this is different in my books than the psrr test. My gut feeling is that for any reasonable listening levels this won't be an issue - transients will be catered to by the energy stored in the rail capacitors (are they big enough ?).

does this sound logical ?
 
Last edited:
High Gareth,

There are two issue at hand that needs to be considered. The first is that of hum and noise on the power supply which is almost always mains related. The fundamental is 100Hz and includes many harmonics simply because your ripple is not a sine wave but sawtooth.

When there is little or no signal present (between tracks and soft passages) this is the most irritating sound that you would experience, since it is in your ear. Even though it is of minute magnitude, if there is nothing else you will hear it. As soon as the signal masks it, it becomes trivial.

Now the PSU as Hugh drew it is not at all bad in this regard because it is quiet for almost any load. It would be even if the TI431 is replaced with a common zener, and Kean is abolutely right in saying that the CCS driving makes no difference at all. It will perform as well using a resistor.

The second issue is load regulation or stability, now this is a completely different issue and not related to mains or ripple.

It is load related and this is where this power supply falls short. When current is drawn the output sags and when little current is drawn the output rises. In simple terms the supply voltage is modulated by the signal (as much as 1 Volt) and can be described as a pumping sound for lack of a better word.

Because headphones are almost void of any reflective and delayed signals which normally masks this phenomena by room boundaries in a speaker environment, it causes the signal to sound smeared, or muddy, in the headphone world (again my interperatation of the kind of sound).

Now if you look at Hugh's circuit you will find something common to all of his designs and that is of the diode that he inserts in series with the driver section - this is smart because this attempt to counteract precisely this phenomena by not allowing the front end to follow the power supply modulation, but the time constant could be an issue if too short.

How much the load regulation would affect the sound of this particular machine is hard to predict and impossible to simulate, because it would differ for music types, how loud one plays, the impedance of the headphones and their sensitivity but that it is audible with headphones is a given.

Therefor I think one needs to look at both these issues if you are going to become a serious headphoner.

A series/shunt regulator which in my experience sounds far superior to a plain series regulator. Although I am not building this amp, I would like those who are, walk away with the best sounding amp for their time and effort.

For those who are not very familiar with headphone listening, it is a total different and very revealing experience and if the equipment is not up to the task, you will hate it. Listening fatigue is a nice way of saying it sounds &%@#

This is probably the reason that few speaker enthusiasts actually enjoy headphone listening because under most circumstances headphones sound dreadfull - (with badly designed amplifiers).

I take heed of AndrewT's warning, this could turn out to take a little longer than anticipated. I would also like those silent DIYers to voice their feelings and expectations, become proactive. You do not have to be a technical wizkid to say what you like or dislike.

Kindest regards

Nico
 
Last edited:
Hi Keantoken,

Maybe this is up your alley - design a series shunt regulator for this project. Choose a series CCS at about 150% of the total current required by both amps and a shunt that stabilises the output at 24V. Plot the output impedance. Also sim it with a stepped load current of low medium and high. Finally look at generated harmonics of 100 Hz under full load.

Nico
 
Hi Nico,

I am definitely in the camp of people who aren't familiar with headphone listening !

And I think we are agreed, the ripple issue is behind us, it's the signal modulation of the supply rails we are afraid of.

I know some people build their headphone amps with no regulation at all (e.g. Pellerano's zen like hd amp - has anyone built this ?).

I would take the input of somebody who has listened to real head amps over the simulations.

If the output stage has adequate psrr (?) then as you suggest we can turn our attention to the front end, the LTP and the VAS. We can easily beef up the RC-diode filter here if we want.

We may not be able to simulate real listening, but we can look at the psrr of the front end as a function of output loading. We can look at worse case where we are driving 23 Ohm loads with full swing.
 

iko

Ex-Moderator
Joined 2008
Thanks Nico, I couldn't have said it better. I think a lot of people would be in for a very nice surprise then they experience for the first time a low output impedance psu. This translates directly into good load regulation. The output voltage (hence all biasing) does not sag/shift, in spite of the load "tugging" on the current. The most popular ways to get good load regulation is either a big bank of capacitors just before the amp/preamp, or a good shunt regulator. IMHO there are several disadvantages with the big bank of capacitors. I find it is easier to have control over the sound with a shunt regulator. But then, this would be when one is really shooting for outstanding results. My experience with headphones is similar to Nico's. They reveal everything about the chain upstream. You want the best psu you can get. I'm even weary of the tl431. That's just me though.
 
Hugh,
The psu needs to be clean and without regulators we need to ensure that channel-channel x-talk is kept low so that may mean dual supplies

Why? And with the crossfeed, yet. I think that trying to eliminate crosstalk altogether is an engineering exercise, not a sonic one. I have several amps that started with dual supplies and ended with one. I could hear no benefit to the dual supply, though I could hear crosstalk with one channel input shorted. Ever listen to records? Lot more crosstalk there than you will get through the supply. If they claim 30dB separation, they're stretching. Yet records can image just fine.

If there is some other reason relating to stability, then carry on.

Sheldon

I raise the point, because simplicity was one of the stated goals. One less supply is simpler.
 
Last edited:
I would also like those silent DIYers to voice their feelings and expectations, become proactive. You do not have to be a technical wizkid to say what you like or dislike.
Kindest regards
Nico

Hi, I'm one of the silent diy'ers who greatly appreciate the work that has gone into this so far. The technical expertise here among the principals is way beyond what I could contribute. But the chance to have a headphone amp w/ some of Hugh's protected charm installed with the CF is a project I very look forward to and can wait as long as it takes for a buffed finished piece.

I'll be right there when it comes time for a board order - Thanks for all your hard work and sharing!

Best - Stan
 
Hi Keantoken,

Maybe this is up your alley - design a series shunt regulator for this project. Choose a series CCS at about 150% of the total current required by both amps and a shunt that stabilises the output at 24V. Plot the output impedance. Also sim it with a stepped load current of low medium and high. Finally look at generated harmonics of 100 Hz under full load.

Nico

Here's one from Mihai (Roender): http://www.diyaudio.com/forums/showthread.php?p=1349112

Common parts, easily adapted for different voltage/current, single rail. Based on my experience with his amp, I bet it's a good design.

Layout for dual rail version: http://www.diyaudio.com/forums/showthread.php?p=1349246

BOM for dual rail version, @ 39V: http://www.diyaudio.com/forums/showthread.php?p=1354022
 
Last edited: