My Headphone Amp Design!!!

[tonal]

I needed a headphone amp to drive my pair of headphones, not like they needed amplification in terms of SPL but in terms of SQ (Sennheiser HD555 with foam pad removed and damping added on the surrounding walls around the drivers). 
 
I wanted to design my own amp that would upgrade the sound quality and driving capability. I also wanted it to be capable to drive better headphones than the ones I'm using right now.
And finally I wanted an external PSU for various (good) reasons that I will explain later.
 
Soooo, I first made a choice of the topology. That would be two opamp non inverting gain stages and a buffer inside the feedback loop of the second stage.
 
Because the options I had were many and I wanted to choose the one that suited me best, I made a modular design (in fact I made two so I could compare between opamps or buffers).
The opamp swapping is easy - just an 8DIP mounting base and you're done.
 
But the buffer swapping was somewhat tricky. That's why I ended up with a pin header with a fixed arrangement (power supply rails, ground, input, output)
Then all I had to do was to layout the buffer boards in order to match that specific pin arrangement.
 
I tested BUF634, LME49600, LME49610, OPA633, LT1010 (8DIP and TO220) and discrete buffers.
I made some more bare pcbs (a class-A SE with current sources, a class-AB push-pull and some low power buffers,
all discrete designs) but I had already made up my mind after listening so I didn't proceed to their evaluation.
 
I checked a few dual opamps as well to find the one that suited me best. Among them were 
TI :           OPA2132P, OPA2134PA, OPA2107AP, OPA2111KP, and OPA1642 (SMD)
National :  LME4562NA, LME49720NA (8PDIP), LME49720HA (metal can TO99), LME49860NA,
ADI :        AD8620A, AD8066A (both SMDs)
Linear :    LT1057
and a few generic parts like LF353 and NE5532.
The SMDs were mounted on DIP adapters.
Of course I can swap the opamps any time but I really wanted to end up with a limited number of options that sound well and suit my application and my taste.
 
 
The buffer modules, the two modular headphone amps and some power supplies I used:
 

 

 
 
The amp with my favourite buffer modules mounted and the dual regulated power supply:
 

 
 
 
Another take with the opamps I've used:
 
 

 
I first gave a good listen to all the buffers and I ended up choosing the discrete buffer. The differences among (same topology) buffers were not that great
but even with my headphones they were noticeable.
The LT1010s were almost disappointing apart from the fact that they are rated to only 100mA which is marginal for driving headphones appropriately IMO.
I mean, c'mon there are even opamps that can handle 100mA loads. And since I wanted a headamp that can handle all possible kind of headphone loads
(I'm not sure about planars but spec-wise it's possible) the LT1010s had no chance.
 
If I had to rate the buffers according to my preference, first comes the discrete diamond buffer, then the LME49600/LME49610s, after them the BUF634, then the OPA633 and last comes the LT1010s.
 
Things that I like with the discrete buffers is that I can choose topology, the amount of bias and the way that it is applied (variations of current sources, resistors etc), I can also choose all the parts,
the power handling is greater than the IC buffers, there is no thermal feedback because the output transistors are not in the same package with the rest of the circuitry just to name a few.
Of course none of these would matter if it eventually sounded worse than an IC. I'm happy it didn't.
 
In the beginning I used the same potentiometers (miniature alpha) in the two amps in order to keep things equal when comparing buffers and opamps but after making up my mind I used an ALPS RK27 10K log potentiometer
with one of the two amps in order to squeeze more "juice" out!
 
I guess that will be all for tonight (to be continued) 

 

[tonal]

All that prototyping frenzy took place about a year ago.
 
Now about the external PSU, the reasons that I made that choice were practical but also technical.
First of all if I had to place the transformer inside the headphone amp box then I had to provide some serious real estate for a 100VA toroidal transformer, smoothing caps (2x6800uF) and the first stage of regulation.
But then my amp would have to be comparable to the size regular hi-fi device 30 to 40 cm wide and 25 to 30 cm deep and that was not what I wanted on my desk.
 
The technical reasons is interference from the transformer and mostly from the current spikes that charge the filter caps after rectification. The current that charges the smoothing caps flows only for short periods
of time, 100 or 120 times (twice the mains frequency) per second and is not a sine wave but spikes that may range up to quite a few Amperes even if the circuit draws a few hundred milliamperes. That means that
the spectrum of the current flowing into the caps is not a single frequency of 100/120 HZ but many harmonics that extend up to one KHz or even higher. On top of that 50/60 Hz from the mains must be added to the noise sources. 
 
I guess these are good reasons for moving the transformer, the rectifiers and the smoothing caps away from the main signal processing circuitry. But that will not suffice.
If there is no control over the current flowing from the PS filter then some part of the charging current spikes will flow into any caps virtually paralleled to the large smoothing caps.
So in order to overcome this problem the external PSU must be regulated.
 
So voltage regulation is applied in the external PSU and the headphone amplifier. Theoretically that means that total attenuation of supply fluctuations is doubled.
For an 80dB ripple rejection of a single regulator you get (theoretically) twice as much - that is a total of 160dB rejection. Of course even if this stands true the 
noise of the regulator will be higher than that so there are other mechanisms which will limit the performance.
 
So after a few months of listening and testing I ended up with the final circuit. So I laid out the pcbs of the amplifier and the psu.
 
Here is a pic of the amplifier pcb (I have no pics from the bare psu board) :
 

 
It is a double sided board with copper wire bridges between top and bottom layer where needed.
 
 
And here are some shots from the finished prototype of my headphone amplifier:
 

 

 
 

 
 

 
 
And here are some close ups of the headphone amplifier (made a new chassis from 2mm aluminium sheet)
 
 

 

 

 

 

[tonal]

 
 

 
I liked it so much that I got me some professionally made double sided through hole plated boards!
 
I heard the prototype in comparison to a respectable headphone amp with better headphones  than mine and really hi-end sources and I liked it a lot over the other amp.
The folks that heard it liked it too. The differences were not just subtle but clearly heard. The bass was more controlled, the dynamics were excellent and the whole sound was more relaxed
but not laid-back or lifeless. The sound improved in every aspect.
I knew that it was something really good so I thought I should get the best out of it. 
 
I've slightly modified my PCB's layout in order to get rid off some extra vias that were necessary for my prototype since it didn't have plated through holes.
 
The trafo is 100VA, the rectifiers are Schottky, the audio connectors are all gold plated Neutriks, the electrolytics are all 105C Panasonic (TSHA and FC) - a total of 20,000uF capacity -,
the plastic caps are Panasonics stacked metal film, polypropylene and WIMAs FKP, the signal relays are high quality Panasonics, the discrete semis are all ON and Vishay and
the resistors are KOA Speer, the trimmers are BI technologies cermet multiturn, the potentiometer is an ALPS RK27. All the parts are high quality, sourced from trusted distributors
like Mouser (most of the parts), RS, Farnell and some online european retailers like Banzaimusic.and Vintage hi-fi.
 
I expect in the next couple of weeks to have my amp ready to roll. 

 
And a screenshot from RMAA. The loop is the measurement of the sound card alone with output fed back to input and the other column is with the amplifier in the loop while driving a 50 Ohm load with the volume set so that the total gain of the amp equals unity in order to keep the settings of the sound card fixed:
 
 

 
If I'm right with my calculations the dynamic range and noise is about -120 dB wrt the input and the distortion is practically way less than 0.0035% (it is probably way less than -100dB).
 
The amplifier itself is very quiet. Even if I turn the potentiometer all the way up the noise is hardly heard with the K555s. With a pair of AKG K330 in-ear headphones the noise is audible but at that level if any kind of signal is applied anyone would get permanent hearing damage before the headphones set on fire. With a pair of headphones with a sensitivity of around 100dB@1Vrms I'm sure that even when the volume is set to maximum nothing will be heard.
 
With a total gain of 12dB, a source with 2Vrms maximum output and a volume setting around 12 to 1 o'clock will be more than enough to adequately drive headphones like the HD650s.
 
The amp is designed to be driven by a 2Vrms source and will not clip even if the volume is set at maximum.
The input voltage compliance will be around 4.5Vrms. Because the amplifier is a two stage design,  the input voltage must be kept lower than a maximum limit
in order to prevent clipping or excessive distortion of the first amplification stage.
 
The pre-out is muted when a pair of headphones is connected, and there are two inputs selected from the miniature toggle switch in the front that controls the relays.
 
The input impedance is 20K, the headphone output impedance is 1 Ohm, and the pre out has an output impedance of 120Ohms.
 
A toggle switch at the back is used to connect signal ground directly to chassis  or through a small value resistor if any ground loop problem shows up.
The chassis is connected to the safety ground.
 
The output transistors are biased into class-A. That means that a peak current around 160mA can be delivered before leaving class-A, which is a lot!
 
reason for editing: some typos fixed

 

[nassq8]

 
wow impressive technicality. Good job in driving you phone without the need to any one, who sometime considers the consumers as a stupid ATM.  

 

[Telstar]

What a beautiful work
If you would like to share the schematics, I'm curious to see it.
 
About noise, I'm not sure that you reached 120dB, but sure looks good.

 

[tonal]

Thanks for the comments.
 
The new amplifier is going to look even better.
 
As far as noise concerned, unfortunately I have no decent ADC/DAC combination to measure it. An E-MU 0404 USB might do the job but for now the only thing I can do is just make an estimate.
With SpectraLAB I get a noise floor of less than -130dB across the BW with the potentiometer fully on and no input connected, so I might be really really close to what I've estimated.
91.1dB with the amp hooked on and 91.5dB for the loop, a mere difference of 0.4dB when the noise of the source and the input refered noise of the amplifier are added, maybe 3dB less than that
assuming noise of the amplifier and the on-board soundcard are uncorrelated. Sorry if I'm being too technical.
 
The schematic is two non inverting stages, with a diamond buffer inside the feedback loop of the second stage. Power supply is two stages of LM317/337 regs.
What really counts is the PCB layout which is meticulously designed, and the hefty (for a headphone amp) PSU as far as capacitance and transformer power rating is concerned.

 

[Telstar]

Quote:

The schematic is two non inverting stages, with a diamond buffer inside the feedback loop of the second stage. Power supply is two stages of LM317/337 regs. What really counts is the PCB layout which is meticulously designed, and the hefty (for a headphone amp) PSU as far as capacitance and transformer power rating is concerned.

The volume pot is at the input?
If you dont say otherwise, I assume the first stage is input with some gain, à la Ayre. The diamond buffer IS the second stage, or there is something before it (like some drivers?)
 
I agree about PSU, but have you thought about better regulators, such as something from Linear tech (esp. for the negative side, if the amp is DC coupled, which i dont remember if you said that)? I'm using AMB's sigma22 (a discrete series-reg psu) for my HP amps and I find it not lacking in anything.
How hefty? I go between 30 and 80VA depending on the amp design and its PSRR -bigger transformers have better self-regulation, but also creates more EM, which require better shielding.

 

[tonal]

Quote:

The volume pot is at the input?
If you dont say otherwise, I assume the first stage is input with some gain, à la Ayre. The diamond buffer IS the second stage, or there is something before it (like some drivers?)
 
I agree about PSU, but have you thought about better regulators, such as something from Linear tech (esp. for the negative side, if the amp is DC coupled, which i dont remember if you said that)? I'm using AMB's sigma22 (a discrete series-reg psu) for my HP amps and I find it not lacking in anything.
How hefty? I go between 30 and 80VA depending on the amp design and its PSRR -bigger transformers have better self-regulation, but also creates more EM, which require better shielding.

The volume pot is between the two opamp stages. Each stage has a gain of +2. The second stage has the diamond buffer inside its feedback loop (see the LME49600 datasheet or anything like that to see what I mean).
 
Yes the amplifier is DC coupled. The maximum DC offset is below 1mV (usually around 500uV) depending solely on the opamp used. So any decent DC performing opamp will exhibit such a low offset.
 
Regulators from LT would be an option only if LT1033 was not obsolete (paired with LT1085).  LM2940/2990 were not an option because a small AC overvoltage would cause the input voltage of the first regulation stage to be higher than (the absolute in value) limit.
I also dislike same polarity regs used for negative and positive supply for technical reasons.
 
Hefty like 100VA and around a total of 20.000uF rated at 63V plus it has external supply so the interference is limited even if the amp and the PSU are stacked.

 

[Telstar]

Quote:

Regulators from LT would be an option only if LT1033 was not obsolete (paired with LT1085).
 
Hefty like 100VA and around a total of 20.000uF rated at 63V plus it has external supply so the interference is limited even if the amp and the PSU are stacked.

What about the LT1083? I think the newer ones are for low current, but I didnt check all, they are too many
 
Yeah, that's hefty But now you made me curious in trying bigger transformers/reservoir banks!
 
Note that your wood frontpanel does allow some EMI/RFI to escape the psu box (unless you put some copper or aluminium foil on the inside). AFAIK the transformer should be fully metal shielded to achieve this goal.

 

[tonal]

Quote:

What about the LT1083? I think the newer ones are for low current, but I didnt check all, they are too many
 
Yeah, that's hefty But now you made me curious in trying bigger transformers/reservoir banks!
 
Note that your wood frontpanel does allow some EMI/RFI to escape the psu box (unless you put some copper or aluminium foil on the inside). AFAIK the transformer should be fully metal shielded to achieve this goal.

The LT1083 is a 7.5A positive reg and it is in a TO3P package.
 
The LT1033 is supposed to be the negative regulator equivalent of the LT1085 3A reg and is obsolete.
 
These (1085, 1033) are pin compatible with the 317/337 regs but the LT1033 is hard to source, is not a viable option and costs around 10€ per piece from Farnell.
The LT1085 costs another 10€ so for two pairs of positive/negative regs the cost rises to 40€!!! That is a lot and I really don't think it's worth it.
 
The classic 317's/337's do their job really well. With double regulation and generous carefull decoupling they are perfectly fine. No audible noise, no hum or whatever.
 
Finally there is no aperture in the enclosure. The wooden faceplate serves only the looks and is backed by aluminium sheet.

 

[jcx]

looks like you already have dual secondary, dual bridges - why not use the same polarity reg for both supplies - usually one is much better performing than the other of manufacturer's recommended pairing of pos and neg regulator chips
 
LM317 better than 337
 
same polarity reg, common parts for both supply polarities easier to source too

 

[tonal]

Quote:

looks like you already have dual secondary, dual bridges - why not use the same polarity reg for both supplies - usually one is much better performing than the other of manufacturer's recommended pairing of pos and neg regulator chips
 
LM317 better than 337
 
same polarity reg, common parts for both supply polarities easier to source too

Yes I could do it. For the matter of fact usually the positive voltage regs perform better than their negative voltage equivalents.
 
But putting a positive voltage regulator for the negative power supply induces asymmetry.
What I mean is that any common mode interference in the power supply will show up in both negative and positive rails and both grounds if negative and positive regulators are used
so the relative differences between the rails and the ground AND between the grounds will remain he same (it would be ideal if the line regulation wrt frequency was the same for negative and positive regulators).
 
If same polarity regulators are used, then after the regulators the ground of the negative rail where the positive regulator is used will exhibit less of the common mode interference so
there will be a difference in potential between the grounds of positive and negative supply.
 
That's why I keep my ground clear of any active circuit and avoid using same polarity regs to generate both negative and positive supply rails.
 
As far as availability there is no issue considering regulators such as the LM317/337. All the other parts are the same for negative and positive rails..

 

[jcx]

not sure I follow your argument - the output Z of the loaded reg + output bypass C will be low for any polarity reg, "the direction" that AC pri-sec C coupled line noise "flows" through this very low Z is meaningless - its AC
 
and the loaded Reg + output C combination Z should be 10^3-10^6 lower than the pri-sec parasitic C at all frequencies, 10^6 where the reg provides low Z, at least 10^3 at high frequency where the output C is working alone
 
the pri-sec C coupled noise has low Z paths to all 3 wires of the dual PS output regardless of reg polarity
 
then you have the question of the symmetry/equality of each of the transformer pri-sec parasitic C - which will be way different for a layered winding for any type reg
 
if you were really concerned about line noise coupling you would be using shielded transformer and/or split bobbin EI or R-core

 

[tonal]

Quote:

not sure I follow your argument - the output Z of the loaded reg + output bypass C will be low for any polarity reg, "the direction" that AC pri-sec C coupled line noise "flows" through this very low Z is meaningless - its AC
 
and the loaded Reg + output C combination Z should be 10^3-10^6 lower than the pri-sec parasitic C at all frequencies, 10^6 where the reg provides low Z, at least 10^3 at high frequency where the output C is working alone
 
the pri-sec C coupled noise has low Z paths to all 3 wires of the dual PS output regardless of reg polarity
 
then you have the questrion of the symmetry/equality of each of the transformer pri-sec parasitic C - which will be way different for a layered winding for any type reg
 
if you were really concerned about line nosie coupling you would be using shielded transformer and/or split bobbin EI or R-core

The supply is external so even if I had used a low radiation shielded transformer the cable that connects the external supply to the amp would remain.
 
The wires connecting the PSU and the amp are five (V+,GND+,V-,GND-, safety ground connected to the chassis). If they were three even if I wanted I couldn't use same polarity regs.
 
I could run a few simulations and see what's going on but the whole thing is mostly theoretical apart the fact that same polarity regulators would indeed offer exactly the same performance considering line/load regulation and ripple rejection.
 
Finally the specs of interest between LM317 and LM337 are close enough for me.
 
Regulation is double and total ripple rejection (RR) is high (theoretically ripple rejection doubles that way because any AC ripple is attenuated twice by the same amount).
So does it really makes a difference if RR is 140dB for positive supply rail and 120dB for negative supply rail without even considering PSSR of the opamps used. I really do not think so 
 
After all I think the whole thing is just a matter of preference (and a little bit of OCD 

 regarding myself about using same polarity regulators for negative and positive rails).

 

[jcx]

looks like you did use a "low radiation" transformer - a toroid radiates the least mag field - but they have the highest pri-sec parasitic winding C == capacitively couples line noise through the insulation
 
split bobbin can have 5-10x lower pri-sec C giving them lower line noise coupling to the secondary - but they are the ones that radiate more magnetic field, most benefit from being boxed separately