Here is a little something that I've been playing around with for a few months. Mostly for my own pleasure and learning (it is DIY after all!), but I thought I'd share.
Orpheus is an op-amp driven, DC Servo equipped class-AB headphone amp with a real-world THD of 0.0044% (THD+N = 0.0074%). Clipping starts around 4.2v into a 32 ohm load.
The circuit presented comprises:
The non-inverting amp stage has been set at about 2.1x gain, which I found to be plenty. A potentiometer (R13) allows for an additional 1x gain as well as channel level matching. Gain can also be increased by changing the 1k feedback resistor (R5). At 6x gain, THD increases to 0.0044% and THD+n to 0.0074%. The amp stage has a feedback capacitor which lowers the gain of higher frequencies for stability.
The output topology is simple and gets the job done with 4 resistors, 2 transistors and 4 diodes per channel, having slightly less than unity gain. One characteristic of this topology is the possibility of thermal runaway, so the diodes should be mounted on the heatsinks. The output stage in this amp is quite lightly loaded, so this shouldn’t be a problem. I used BD139/140 pairs for the output, but others such as the TTA004B/TTC004B should also be quite suitable.
The DC Servo (courtesy of Rod Elliot’s wonderful website) using the venerable TL072 cancels out the DC down to between -1.5mV and +4.5mV, which is perfectly acceptable. The servo eliminates the need for input and output capacitors, and gives the amp a polite power on/off behaviour.
The PSU is a simple +/-15v CRC filtered supply that I designed around the 10W Recom RAC10-15DK/277 SMPS module. It provides plenty of regulated power with good smoothing and rejection. It has provisions for a ground loop breaker if required (independent of the Protective Earth of course).
A more expensive second PSU was also designed using the Triad Magnetics FS36-170 transformer and LM317/337 regulation. The transformer is dampened with snubbing (Cs, Rs, Cx) and again, GLB provisions and independent Protective Earth are there too. The voltage can be varied a fair bit, which makes this PSU suitable for other applications also.
The amp PCB is 99x100mm and fits into the Hammond 1455 series enclosures. I opted for individual ground traces with a separate ground plane.
Op-amp selection:
For the DC servo, the TL072 with it's JFet-input is the best choice. The NE5532 will work, but not ideal.
For the amp, the NJM4556 is an excellent choice, but the circuit works well with others such as the NE5532. The circuit should be stable for other substitutions.
But how does is measure?
The frequency response measured acceptably flat and the amp has descending harmonics, dominantly H2.
* There is a 4KHz and 8KHz artefact from the Focusrite, possibly USB related.
But how does it sound?
I’m no connoisseur, but I’ll try to describe it the best I can.
I found the sound to be very detailed and the amp responsive to complex pieces. I heard details in songs that I haven’t previously detected and the reproduction of voices felt accurate and realistic. I found the amp pleasant and not fatiguing to listen to over extended periods.
Orpheus is powerful enough to drive all of my headphones, from my 150 ohm Yamaha HP-50 Orthodynamics, the 38 ohm ATH-MX50, 32 ohm ESS 422H Hybrid AMTs and my HE-X4. All of them paired nicely, but the best detail and bass extension came from the ATH-MX50 and the easiest, most relaxed listening was from the HE-X4. The ESS were a lot softer in detail, but I find that is their nature and not entirely to my taste.
Noise wise, the amp is dead silent at full volume with no input connected.
The finished amp:
Thanks for reading.
Orpheus is an op-amp driven, DC Servo equipped class-AB headphone amp with a real-world THD of 0.0044% (THD+N = 0.0074%). Clipping starts around 4.2v into a 32 ohm load.
The circuit presented comprises:
- A basic input filter;
- A simple voltage follower buffer;
- A non-inverting op-amp stage with an adjustable gain from around 2x up to 3x;
- A simple diode-biased, Class-AB push pull output stage;
- An op-amp driven DC Servo.
The non-inverting amp stage has been set at about 2.1x gain, which I found to be plenty. A potentiometer (R13) allows for an additional 1x gain as well as channel level matching. Gain can also be increased by changing the 1k feedback resistor (R5). At 6x gain, THD increases to 0.0044% and THD+n to 0.0074%. The amp stage has a feedback capacitor which lowers the gain of higher frequencies for stability.
The output topology is simple and gets the job done with 4 resistors, 2 transistors and 4 diodes per channel, having slightly less than unity gain. One characteristic of this topology is the possibility of thermal runaway, so the diodes should be mounted on the heatsinks. The output stage in this amp is quite lightly loaded, so this shouldn’t be a problem. I used BD139/140 pairs for the output, but others such as the TTA004B/TTC004B should also be quite suitable.
The DC Servo (courtesy of Rod Elliot’s wonderful website) using the venerable TL072 cancels out the DC down to between -1.5mV and +4.5mV, which is perfectly acceptable. The servo eliminates the need for input and output capacitors, and gives the amp a polite power on/off behaviour.
The PSU is a simple +/-15v CRC filtered supply that I designed around the 10W Recom RAC10-15DK/277 SMPS module. It provides plenty of regulated power with good smoothing and rejection. It has provisions for a ground loop breaker if required (independent of the Protective Earth of course).
A more expensive second PSU was also designed using the Triad Magnetics FS36-170 transformer and LM317/337 regulation. The transformer is dampened with snubbing (Cs, Rs, Cx) and again, GLB provisions and independent Protective Earth are there too. The voltage can be varied a fair bit, which makes this PSU suitable for other applications also.
The amp PCB is 99x100mm and fits into the Hammond 1455 series enclosures. I opted for individual ground traces with a separate ground plane.
Op-amp selection:
For the DC servo, the TL072 with it's JFet-input is the best choice. The NE5532 will work, but not ideal.
For the amp, the NJM4556 is an excellent choice, but the circuit works well with others such as the NE5532. The circuit should be stable for other substitutions.
But how does is measure?
The frequency response measured acceptably flat and the amp has descending harmonics, dominantly H2.
* There is a 4KHz and 8KHz artefact from the Focusrite, possibly USB related.
But how does it sound?
I’m no connoisseur, but I’ll try to describe it the best I can.
I found the sound to be very detailed and the amp responsive to complex pieces. I heard details in songs that I haven’t previously detected and the reproduction of voices felt accurate and realistic. I found the amp pleasant and not fatiguing to listen to over extended periods.
Orpheus is powerful enough to drive all of my headphones, from my 150 ohm Yamaha HP-50 Orthodynamics, the 38 ohm ATH-MX50, 32 ohm ESS 422H Hybrid AMTs and my HE-X4. All of them paired nicely, but the best detail and bass extension came from the ATH-MX50 and the easiest, most relaxed listening was from the HE-X4. The ESS were a lot softer in detail, but I find that is their nature and not entirely to my taste.
Noise wise, the amp is dead silent at full volume with no input connected.
The finished amp:
Thanks for reading.
Last edited:
Very pretty PCB layouts, congratulations!
I am a little nervous about using 0.2 ampere diodes in the ground loop breaker subcircuits. Their job is to protect the human user in case the mains live wire accidentally shorts to the metal chassis. But those dinky little 1N4148s cannot sink 20 amperes for 2 seconds, which they need to do in order to trip the house wiring's circuit breaker.
I am a little nervous about using 0.2 ampere diodes in the ground loop breaker subcircuits. Their job is to protect the human user in case the mains live wire accidentally shorts to the metal chassis. But those dinky little 1N4148s cannot sink 20 amperes for 2 seconds, which they need to do in order to trip the house wiring's circuit breaker.
Thanks Mark!
The Protective Earth goes directly from the IEC earth terminal to the chassis via a 4mm bolt & star washer independently of the loop breaker:
Physically the diodes fitted are 1n4007, but the chassis is always connected directly to PE no matter whether or not there is a loop breaker in place.
The Protective Earth goes directly from the IEC earth terminal to the chassis via a 4mm bolt & star washer independently of the loop breaker:
Physically the diodes fitted are 1n4007, but the chassis is always connected directly to PE no matter whether or not there is a loop breaker in place.
Very nice project! Some years ago I was experimenting with very similar idea.
Later decided to use IC output stage instead of discrete transistors. Mainly because of temperature matching issues you mentioned and IC are additionally overcurrent and overtemperature protected which is really handy in headphone amplifier. Shorting output with jack plug isn't that unusual and adding protection circuits to discrete output stage is a hassle.
If I recall correctly adding about 100 ohms resistor between input and output and two capacitors across diodes improved THD significantly. I think it was described somewhere in Art of Electronics.
Later decided to use IC output stage instead of discrete transistors. Mainly because of temperature matching issues you mentioned and IC are additionally overcurrent and overtemperature protected which is really handy in headphone amplifier. Shorting output with jack plug isn't that unusual and adding protection circuits to discrete output stage is a hassle.
If I recall correctly adding about 100 ohms resistor between input and output and two capacitors across diodes improved THD significantly. I think it was described somewhere in Art of Electronics.
That's good info, thanks. It would be an easy mod to make, so I think I'll try that out on one of the spare boards and see how it compares. I have the book on my shelf at home, so I'll have a dig through and see if I can find the reference.
That's a very good question.. I couldn't really get the DC servo to behave properly in the simulation when I tried to just control the final output. Perhaps I was putting the connections in the wrong spot, but the offset was too high.I'm intrigued as to why you used the servo to correct the 5532 output rather than the final output?
Very nice though
The servo behaves nicely in the sim and in real life connected as is, but maybe there is a more correct way? Where should the ServoIn and ServoFB lines be to wrap around the final output?
It won't hurt to switch over to the 1N5405 as they are only $5.90 for 10 at Jaycar. I am always nervous designing anything with AC, so I made sure of the creepage distances, used the recommended fuse for the SMPS and used a board mounted IEC rather than relying on loose wire connections.
The voltage on R8 on your sim (so ground) sets the 'reference voltage' for the servo to work to.
The take off point for the servo should be the point you want to maintain at the reference voltage value (which is ground or zero volts in this case). The opamp then does what is needed to make those two points equal in voltage. If you set the reference at 2 volts the servo would then bring the take off point to 2 volts and so on, make the reference -4.2 and the take off point would assume that voltage.
With resistor values as high as 2.2meg anything other than a FET opamp will probably give significant DC offsets due to opamp input bias currents.
At 25mA idle current that's a class A amp for 99.9999% of all headphones on Earth... I really need to understand the need for dc coupling headphones...I listened Meze Empyrean on a highly reveered 10K$ Chord dac and when measuring the Chord it showed 3...6db less base under 300hz... You need less base with headphones not more so linearity at low frequency is actually the wrong game for headphones.I'm curious what is a dc design bringing audibly more than an old Studer amp...The usual FT trz in headphones is 100...300Mhz...even pure class B headphones amps sound clear as hell. I don't see often headphones capable of 0.004% thd reproduction. I feel there's too much fuss about surgical headphones amplification and a waste of time to put such work into it unless you're going really exotic like tubes or germanium...Even my phone and my laptop can drive pretty loud and clear any of my headphones...That pretty box would have made a good tube amp .
.Attachments
Last edited:
This was mainly to see if I could do it and make it work, but also down to price. The components to run the DC servo came in at $3.89AUD whereas the cheapest Silmic II 2200uF is $6.64AUD and all have long lead times.
Interesting your comments regarding needing less bass for headphones. I found some of the amps that I have built in the past sounded quite anaemic and I had to boost the bass with DSP. This one sounds more balanced and realistic when I listen to something like Japanese Taiko drums - the flutter of the drum skin is much more evident. Can't beat seeing a live Taiko performance and feeling your organs vibrate and dance to the music though
I can always CNC another pretty box as a starting point for a tube design... Not that I need another project
Never used Silmics...but compared 2000uF nichicon muse bipolars to 9000uF much cheaper(0.5 bucks for 4700uf/35v ) unipolar JB and the higher capacitance and double(20mA vs 10mA) idle current won against the Muse on 64 ohms..leakage...almost identical.I'm pretty sure the idle current was the one guilty for the better base not the JB csps.Too much fuss about audiophile caps...You just need your values right.Having the whole circuit supplied by 10V dc and the output strapped to mid tap of 2 series 35 v dc caps will keep the leakage curent very low even for a very cheap unipolar cap.
Be brave and buy cheap oversized caps !
Be brave and buy cheap oversized caps !
Attachments
Interestingly, the simulation shows an improvement in distortion by a factor of ten when adding the resistor and 2 capacitors. Adding the capacitors plus a diode in reverse across the 4 diodes as shown on the schematics dreamth posted yields a slightly worse but similar result.
The sim seems to suggest that values as low as 10 ohms would be stable (and result in even lower thd), but I'd have to test out some values to see what actually works outside of Sim Land. It will be good to see how this modification changes the sound of the amp, because the sim did show a change in the FFT harmonics.
The sim seems to suggest that values as low as 10 ohms would be stable (and result in even lower thd), but I'd have to test out some values to see what actually works outside of Sim Land. It will be good to see how this modification changes the sound of the amp, because the sim did show a change in the FFT harmonics.
Put all of them in the schematic, put all of them in the PCB layout, then build a few prototype boards implementing each of the optional variations. Measure them, listen to them, read tea leaves, deal tarot cards, burn incense, ask your spouse to listen ... use whatever procedure you prefer, to choose one. Then mark all the others "DNS" (do not stuff , do not solder) on the schematic and the PCB. Victory!
I have modified the board with 22u and 100r as per the diagram in post #4.
The amp still sounds great, but unfortunately I am locked in my bedroom with Covid, so I can't do some measurements just yet.
I have been playing with the modified circuit in LTspice and this is what I have found:
From Art of Electronics 3rd Edition:
Adding the capacitors across the diodes changes their square wave behaviour from this:
To this:
Am I right in saying that the additional distortion was from diode switching noise and that the capacitor + resistor is dampening the diodes? The square wave behaviour looks like an RC time constant output. I'm trying to wrap my Covid-fogged brain around what the mechanism is here.
Also, does anybody know why 100 ohm for the resistor? The sim is happy all the way down to 1 ohm, but I don't know if it is a good / bad idea to reduce it that far.
The amp still sounds great, but unfortunately I am locked in my bedroom with Covid, so I can't do some measurements just yet.
I have been playing with the modified circuit in LTspice and this is what I have found:
- When using 100r, the lowest distortion occurs with 10uF capacitors - simulated 0.000045%.
- Reducing the value of the resistor also reduced distortion - 10uF and 10r = simulated 0.000025%.
- Connecting without the biasing resistors as per the AoE diagram (below) results in 0.000099% distortion but dissipates 7.5W in Q1 and Q2 as opposed to 430mW.
- 2 diodes in series result in significantly less crossover distortion than the single diode when the biasing resistors are used - 0.00047% versus 0.013%
From Art of Electronics 3rd Edition:
Adding the capacitors across the diodes changes their square wave behaviour from this:
To this:
Am I right in saying that the additional distortion was from diode switching noise and that the capacitor + resistor is dampening the diodes? The square wave behaviour looks like an RC time constant output. I'm trying to wrap my Covid-fogged brain around what the mechanism is here.
Also, does anybody know why 100 ohm for the resistor? The sim is happy all the way down to 1 ohm, but I don't know if it is a good / bad idea to reduce it that far.
Managed to take some readings of the amp with the modifications made:
Distortion comes in at half of what I had before - 1KHz THD 0.0022%, noise 0.0034%; down from 0.0044% and 0.0060%.
From my loopback measurements, it seems that I have hit the limit of my measuring device. So the amp may or may not be lower than what I have measured, but it is at a point beyond what I can economically measure, and certainly well beyond the threshold of perception.
As for the characteristics, there are a lot less noise spikes on the plots. The H2 has reduced from around -92dB to about -98dB. The other harmonics appear unaltered.
Again, there is a lot of trash around the 4Khz and 8Khz marks, but these are present in the loopback and the source of which I have been trying to pinpoint for some time.
Out of interest, I also took some square wave measurements, but just with a resistive load.
They are also only 1 and 2 KHz as I do not expect the Focusrite able to generate solid square waves at higher orders:
Finally I took a THD Sweep to see the distortion behaviour with respect to frequency:
When I get out of quarantine, I'm going to try and put another board together but with TTA-004/TTC-004 transistor pairs and see how they go.
Distortion comes in at half of what I had before - 1KHz THD 0.0022%, noise 0.0034%; down from 0.0044% and 0.0060%.
From my loopback measurements, it seems that I have hit the limit of my measuring device. So the amp may or may not be lower than what I have measured, but it is at a point beyond what I can economically measure, and certainly well beyond the threshold of perception.
As for the characteristics, there are a lot less noise spikes on the plots. The H2 has reduced from around -92dB to about -98dB. The other harmonics appear unaltered.
Again, there is a lot of trash around the 4Khz and 8Khz marks, but these are present in the loopback and the source of which I have been trying to pinpoint for some time.
Out of interest, I also took some square wave measurements, but just with a resistive load.
They are also only 1 and 2 KHz as I do not expect the Focusrite able to generate solid square waves at higher orders:
Finally I took a THD Sweep to see the distortion behaviour with respect to frequency:
When I get out of quarantine, I'm going to try and put another board together but with TTA-004/TTC-004 transistor pairs and see how they go.
Attachments
It's been a while since I made an update.
A while back, I made a V3 of this amp taking on board some of the suggestions made.
For example, I moved the servo FB loop to the output side of the outputs as suggested by Mooly and changed the footprint to accept 1N5404 diodes on the power supply loop breaker option as suggested by Mark Johnson,
The final scheme I went with is as follows (just LH channel and servo shown, inputs etc remain the same):
The desire was always to create a 'cube' type enclosure for this amp, so I went ahead and stacked the layers and made a 3d-printed case for it to go into:
Bottom layer is the PSU, unchanged from the original design.
Middle layer is the amp. Here it is fitted with the Burson V6 Vivids that were generously donated by Burson some time ago (not related to this project).
And the top layer is a carrier for a ES9023/PCM2706 USB DAC.
There is a 3.5mm input jack which when disconnected, bypasses to the DAC.
Small input pot knob on the rear.
The final fitout looked like this:
Measurements were as follows:
0.00272% THD with the RC4558P.
0.00188% THD with the Burson V6.
Burson THD:
The amp has no audible noise with inputs shorted at max volume and the servo is rock solid.
I've been using this amp for some time now as a travel companion and I've really enjoyed its performance and sound.
A while back, I made a V3 of this amp taking on board some of the suggestions made.
For example, I moved the servo FB loop to the output side of the outputs as suggested by Mooly and changed the footprint to accept 1N5404 diodes on the power supply loop breaker option as suggested by Mark Johnson,
The final scheme I went with is as follows (just LH channel and servo shown, inputs etc remain the same):
The desire was always to create a 'cube' type enclosure for this amp, so I went ahead and stacked the layers and made a 3d-printed case for it to go into:
Bottom layer is the PSU, unchanged from the original design.
Middle layer is the amp. Here it is fitted with the Burson V6 Vivids that were generously donated by Burson some time ago (not related to this project).
And the top layer is a carrier for a ES9023/PCM2706 USB DAC.
There is a 3.5mm input jack which when disconnected, bypasses to the DAC.
Small input pot knob on the rear.
The final fitout looked like this:
Measurements were as follows:
0.00272% THD with the RC4558P.
0.00188% THD with the Burson V6.
Burson THD:
The amp has no audible noise with inputs shorted at max volume and the servo is rock solid.
I've been using this amp for some time now as a travel companion and I've really enjoyed its performance and sound.
I see you have the output of the servo feeding into the output of the driver opamp via a 22k
Typo? The servo still needs to act on the opamp as you had it originally. Only the take off point changes.