Some further thoughts about discrete op amp modules and output offset voltage.
The key spec given the Sapphire circuit is the input bias current. Looking at "typical" values for different op amps,
OPA134 5 pA (that's what JFET inputs buys you)
OPA27 15 nA (that's what internal bias cancellation circuitry buys you)
NE5534 500 nA (that's what microfabrication buys you)
John Hardy 990C+ 2.2 uA (that what competent design optimization buys you)
Sparkolabs SS3601 4.6 uA (that's probably still better than what you'd get if you made it yourself)
Burson SS V5 180 uA (I have no freaking idea what is going on here.)
The basic issue is the input bias current generates a potentially large voltage across the resistor R2 (47k x Ibias). R2 is large to keep the input impedance of the amplifier high, allow the use of a relatively small coupling capacitor C1.
Since the input offset is multiplied by the circuit gain at the output, it should really be kept lower than 10 mV. This limits the desired Ibias to 200 nA and less. So, for the stock configuration only JFET types like the OPA134 or input cancellation bipolar types like the OP27 can be recommended, though the NE5534 is in practice fine. (A few hundred mV output offset isn't going to damage even 16 ohms headphones.) None of the discrete bipolar modules are going to work.
We can modify the circuit to better cope with high bias currents however.
Option 1 is to reduce R2, decreasing the volume control to the value of R2 or smaller, and increasing the value of C1 in proportion. The practical limit would be 10k. No one wants a headphone amp with an input impedance lower than that.
The other option is to increase R3 and R4 such that 1/(1/R3+1/R4) = R2. So to say R3 = 56k and R4 = 220k. This will drastically increase the output noise value, unless you use a JFET opamp with low current noise ... but then you don't need to bother!
Finally, you can mess about with offset compensation, trimming the output by injecting a cancelling DC signal.
As I see it none of these are satisfactory, and that for this application, namely the input stage following a volume control, JFET op amps, or, worse case bipolar ICs with factory trimmed and compensated bias current are the best option.
However, using a 10k volume pot, reducing R2 to 10k, increasing C1 to maybe 4.7 uF, and changing R2 to 10k and R4 to (R3xgain), you have a fair to good chance of being able to use the 990C successfully.
The key spec given the Sapphire circuit is the input bias current. Looking at "typical" values for different op amps,
OPA134 5 pA (that's what JFET inputs buys you)
OPA27 15 nA (that's what internal bias cancellation circuitry buys you)
NE5534 500 nA (that's what microfabrication buys you)
John Hardy 990C+ 2.2 uA (that what competent design optimization buys you)
Sparkolabs SS3601 4.6 uA (that's probably still better than what you'd get if you made it yourself)
Burson SS V5 180 uA (I have no freaking idea what is going on here.)
The basic issue is the input bias current generates a potentially large voltage across the resistor R2 (47k x Ibias). R2 is large to keep the input impedance of the amplifier high, allow the use of a relatively small coupling capacitor C1.
Since the input offset is multiplied by the circuit gain at the output, it should really be kept lower than 10 mV. This limits the desired Ibias to 200 nA and less. So, for the stock configuration only JFET types like the OPA134 or input cancellation bipolar types like the OP27 can be recommended, though the NE5534 is in practice fine. (A few hundred mV output offset isn't going to damage even 16 ohms headphones.) None of the discrete bipolar modules are going to work.
We can modify the circuit to better cope with high bias currents however.
Option 1 is to reduce R2, decreasing the volume control to the value of R2 or smaller, and increasing the value of C1 in proportion. The practical limit would be 10k. No one wants a headphone amp with an input impedance lower than that.
The other option is to increase R3 and R4 such that 1/(1/R3+1/R4) = R2. So to say R3 = 56k and R4 = 220k. This will drastically increase the output noise value, unless you use a JFET opamp with low current noise ... but then you don't need to bother!
Finally, you can mess about with offset compensation, trimming the output by injecting a cancelling DC signal.
As I see it none of these are satisfactory, and that for this application, namely the input stage following a volume control, JFET op amps, or, worse case bipolar ICs with factory trimmed and compensated bias current are the best option.
However, using a 10k volume pot, reducing R2 to 10k, increasing C1 to maybe 4.7 uF, and changing R2 to 10k and R4 to (R3xgain), you have a fair to good chance of being able to use the 990C successfully.
Last edited:
Disfunctionalshadow tested Burson SS and say, that have very good DC offset. I little search about this opamp. Datasheet values are same for many years (too for older gens). Burson V4 have JFET input and output (older gens I dont know). Maybe it is reason for low DC offset. Secondary reason can be better matching or trimming in new generations. In one statement from Burson was written, that SS Opamp have very low DC offset.
I'm actually using a John Hardy JE-990c+ now. One channel is sitting at -20mv (drifts down to -18mv warmed up) and the other sits at -30mv. I've swapped them between channels and the offset stays the same, os it's the bias injection at the end of the trim pot limits that's holding it there.
I tried the NE5534 and the offsets went down to 8 mv worst case one channel. Sonically, it's pretty good, but not as involving as the DOAs and the JE-990c+ brings out the notes - more weight or fundamental to the bass notes.
As for getting the other op amps to work (other than Sparkos'). I've run some numbers over at Tangent's on his calcs and lowering the gain to 12 dB by raising R3 to 4.5K, assuming a 10K pot, leaving R4 at 20K, and R2 at 47.5K, should put the JE-990 (both the DIY-990 and 990C+) around 50mv of offset with no other changes. I haven't actually tried this without the DC blocking caps to ground off R3. It's something that needs to be tested.
Then injecting bias either with the simple bias arrangement or the more complex one in the M1 circuit or use one of San Groner's, or use something like Analog Device's MT-37 tutorial to build one from scratch and refine it like Deane, John, and Sam have done, should all work.
One other thing - I did up the secondary capacitance (caps after the Z-Reg) to 500+ uFd and bypassed it with a .1uF polypropylene. I think the DOAs need more than the 300uFd in the current design - just and opinion, not engineering or testing!
I tried the NE5534 and the offsets went down to 8 mv worst case one channel. Sonically, it's pretty good, but not as involving as the DOAs and the JE-990c+ brings out the notes - more weight or fundamental to the bass notes.
As for getting the other op amps to work (other than Sparkos'). I've run some numbers over at Tangent's on his calcs and lowering the gain to 12 dB by raising R3 to 4.5K, assuming a 10K pot, leaving R4 at 20K, and R2 at 47.5K, should put the JE-990 (both the DIY-990 and 990C+) around 50mv of offset with no other changes. I haven't actually tried this without the DC blocking caps to ground off R3. It's something that needs to be tested.
Then injecting bias either with the simple bias arrangement or the more complex one in the M1 circuit or use one of San Groner's, or use something like Analog Device's MT-37 tutorial to build one from scratch and refine it like Deane, John, and Sam have done, should all work.
One other thing - I did up the secondary capacitance (caps after the Z-Reg) to 500+ uFd and bypassed it with a .1uF polypropylene. I think the DOAs need more than the 300uFd in the current design - just and opinion, not engineering or testing!
I forgot that important point: a blocking cap in series with R3 will make a huge difference. (instead of being multiplied by the gain, the output DC offset is just the input value) It's a bit of an unsightly mod on the existing board though.
@Disfunctonalshadow:
Thanks for the research, as always. Given the larger current draw some more capacitance near the modules would not go amiss, I agree. I haven't seen Tangent's calcs, but back-of-the-envelope your quoted values are close to what I had in mind. You may still need the blocking cap on R3 to reach under 50 mV, but can do without any trim/offset injection correction circuit.
@Disfunctonalshadow:
Thanks for the research, as always. Given the larger current draw some more capacitance near the modules would not go amiss, I agree. I haven't seen Tangent's calcs, but back-of-the-envelope your quoted values are close to what I had in mind. You may still need the blocking cap on R3 to reach under 50 mV, but can do without any trim/offset injection correction circuit.
There is another, somewhat more costly, solution to the offset issue. And it would limit some swapping of op amps. That is change the architecture a bit.
Instead of - pot, cap, op amp, buffer. Pull a NwAvGuy and do a - cap, op amp, pot, buffer architecture. Though, I'd personally (and am looking at doing this) do a - transformer, op amp, pot (LDR), buffer.
To do so would require doing what John Hardy does (and Deane Jensen did) in the M1/M2 - use a bias compensation circuit and a servo on the op amp.
The other limitation is the op amp. With a bipolar op amp and transformer - a Jensen JT-11P-1HPC, R3 would be 2K, R4 would be 10K, and R2 would be 10K (load for the transformer). Calculations say - 12.87 uV, -98 dB, with an offset of 21 mv for a JE-990C+, and 43 mv offset for Sparkos' op amp. Gain would be 6 or 16 dB.
Offset could be controlled fairly well with the servo after the initial offset adjustment is made.
For a FET op amp, either a second resistor (assuming a transformer) might be needed, which the 47.5K could be used. Though I haven't run the calculations for that to find the offset, the noise voltage would rise to -93 dB, -23 uV as that resistor dominates the noise. I suppose the resistor could be left off, but I haven't thought about FET op amps much at this time, for several un-interesting reasons.
Without the transformer, a 4.7 uFD cap could be used to couple the op amp's input. The -3 dB corner would be 3.4 Hz, using the bipolar op amps feedback loop and 10K R2.
Given the above would take more space, regardless of the choice of transformer or cap (4.7 uFD pp is pretty big), it'd mean taking the filter caps and Z-Reg off the board. If one were willing to remote the transformers, bridge, and filter caps - say add an RC filter and snubbers for the transformer secondaries, then cable to the headphone amp.
The regulators could be formed such that they match the diyaudio Jung/Didden SuperRegs footprint, then 1 output cap of 120 uFd would reside on the regulator boards and 2 x 220 uFd, plus a bypass pp cap for the op amp, would be on the headphone amp boards. A Z-Reg, SuperReg, or anything else - shunt reg, could be plugged in, assuming the sense lines were kept and all regs had the first filter cap on board, like the SuperRegs . That would leave enough room on the headphone amp boards to have 5 mm lead spacing to support Silmic II's which are 12.5 mm in diameter. And 5 mm lead spacing for Wima or Vishay / Roederstein Polypropylene bypass cap or caps.
BTW - none of the above is my original thought. The transformer idea is from SY - Heretical preamp, and the others are listed in the above text. I like the transformer, despite the cost because of the galvanic isolation, and CMRR of 124 dB @ 60 Hz, which gets rid of (minimizes) one of the distortions.
The final part of this whole delusion would be to change the current mirrors on the buffer section - because why leave something alone that works fine, when more time and money can be thrown at it? I had thought about using the Figure 6 of the mirrors given that Richard had used Figure 4 of, which would remove the thermal issue (not that I've seen this), but would cost 3 diodes to be added. However, being influenced by Walt Jung from "Sources 101: Audio Current Regulators for High Performance", the floating source/sink using an LM317, npn, TLV431, and 2 resistors - 1.5K and a 150 ohm Rset would provide 9 ma would seem to be an improvement from a noise standpoint.
Instead of - pot, cap, op amp, buffer. Pull a NwAvGuy and do a - cap, op amp, pot, buffer architecture. Though, I'd personally (and am looking at doing this) do a - transformer, op amp, pot (LDR), buffer.
To do so would require doing what John Hardy does (and Deane Jensen did) in the M1/M2 - use a bias compensation circuit and a servo on the op amp.
The other limitation is the op amp. With a bipolar op amp and transformer - a Jensen JT-11P-1HPC, R3 would be 2K, R4 would be 10K, and R2 would be 10K (load for the transformer). Calculations say - 12.87 uV, -98 dB, with an offset of 21 mv for a JE-990C+, and 43 mv offset for Sparkos' op amp. Gain would be 6 or 16 dB.
Offset could be controlled fairly well with the servo after the initial offset adjustment is made.
For a FET op amp, either a second resistor (assuming a transformer) might be needed, which the 47.5K could be used. Though I haven't run the calculations for that to find the offset, the noise voltage would rise to -93 dB, -23 uV as that resistor dominates the noise. I suppose the resistor could be left off, but I haven't thought about FET op amps much at this time, for several un-interesting reasons.
Without the transformer, a 4.7 uFD cap could be used to couple the op amp's input. The -3 dB corner would be 3.4 Hz, using the bipolar op amps feedback loop and 10K R2.
Given the above would take more space, regardless of the choice of transformer or cap (4.7 uFD pp is pretty big), it'd mean taking the filter caps and Z-Reg off the board. If one were willing to remote the transformers, bridge, and filter caps - say add an RC filter and snubbers for the transformer secondaries, then cable to the headphone amp.
The regulators could be formed such that they match the diyaudio Jung/Didden SuperRegs footprint, then 1 output cap of 120 uFd would reside on the regulator boards and 2 x 220 uFd, plus a bypass pp cap for the op amp, would be on the headphone amp boards. A Z-Reg, SuperReg, or anything else - shunt reg, could be plugged in, assuming the sense lines were kept and all regs had the first filter cap on board, like the SuperRegs . That would leave enough room on the headphone amp boards to have 5 mm lead spacing to support Silmic II's which are 12.5 mm in diameter. And 5 mm lead spacing for Wima or Vishay / Roederstein Polypropylene bypass cap or caps.
BTW - none of the above is my original thought. The transformer idea is from SY - Heretical preamp, and the others are listed in the above text. I like the transformer, despite the cost because of the galvanic isolation, and CMRR of 124 dB @ 60 Hz, which gets rid of (minimizes) one of the distortions.
The final part of this whole delusion would be to change the current mirrors on the buffer section - because why leave something alone that works fine, when more time and money can be thrown at it? I had thought about using the Figure 6 of the mirrors given that Richard had used Figure 4 of, which would remove the thermal issue (not that I've seen this), but would cost 3 diodes to be added. However, being influenced by Walt Jung from "Sources 101: Audio Current Regulators for High Performance", the floating source/sink using an LM317, npn, TLV431, and 2 resistors - 1.5K and a 150 ohm Rset would provide 9 ma would seem to be an improvement from a noise standpoint.
It's an idea I floated in the early stages of the Sapphire development. On first look the topology (cap, voltage amp, volume control, buffer) seems more attractive than putting the volume control at the input. And, from a noise perspective, it certainly is.
The problem is the output of the volume control is a weak (high, variable impedance, low, variable level) source. It, rather than the input signal, is what you have to be mainly concerned about vs. following it with an easy to drive, high impedance front end.
The diamond buffer, like all high current BJT stages, distorts in proportion to the source impedance. It doesn't work especially well following a volume control, so you'd have to, uh, buffer the buffer with essentially the same JFET op amp as you were planning to use on the input. The topology then becomes (cap, voltage amp, volume control, impedance buffer, current buffer) and it all become a largely redundant exercise.
In the O2 the buffer is pair of IC op amps, there is no particular problem putting them after the volume control.
The problem is the output of the volume control is a weak (high, variable impedance, low, variable level) source. It, rather than the input signal, is what you have to be mainly concerned about vs. following it with an easy to drive, high impedance front end.
The diamond buffer, like all high current BJT stages, distorts in proportion to the source impedance. It doesn't work especially well following a volume control, so you'd have to, uh, buffer the buffer with essentially the same JFET op amp as you were planning to use on the input. The topology then becomes (cap, voltage amp, volume control, impedance buffer, current buffer) and it all become a largely redundant exercise.
In the O2 the buffer is pair of IC op amps, there is no particular problem putting them after the volume control.
Indeed, something like the B1 would do nicely.
(Note the Sapphire is the same circuit as the simplified LMH6321 figure in Dave's blog, plus extra Sziklai transistors in the output.)
The problem with diamond buffers identified in the blog I circumvent by just running the buffer in class A. For headphone amplifiers or line drivers this means no additional complications like the electrolytic caps shown further down are needed.
(Note the Sapphire is the same circuit as the simplified LMH6321 figure in Dave's blog, plus extra Sziklai transistors in the output.)
The problem with diamond buffers identified in the blog I circumvent by just running the buffer in class A. For headphone amplifiers or line drivers this means no additional complications like the electrolytic caps shown further down are needed.
My Sapphire is working!!! Huge thanks Richard for great design, PCBs and support.
Any details of my design:
- Toroids with rectifiers are in separate chassis
- Supplies are 18V with 18V BZX85 zeners. Voltage on opamp is 17,17V.
- Tranzistors are Philips 2N5401 and 2N5551 matched for hfe from 2x200 pieces
- Resistors are Vishay RN and CMF
- Supply capacitors Elna Silmic II
- Board capacitors Nichicon Muse
Amplifier is not yet fully finished, I wait for any parts and prepare latest mods:
- Mini XLR connectors for supply-amp interconnect
- Copper sheet as shield between channels in both chassis and maybe for RCA-pot wire
- Vishay Dale stepper attenuator
- Back switch for switching 2 inputs
- Secured bottom switch for input capacitor bypass
- Better Jack connector
- Finaly will be amplifier wired with pure silver
After first switch on DC offset with OPA134 was 5mV and 2mV. At listening is whole time under 10mV (about 5mV ordinary).
I can say sound is fantastic - extremely neutral. Lot of power, very detailed without any harsness. I never hear so huge stage on my headphones. I listen third day, switching genres - all is OK. From pop to metal, from bass to treble. Nothing miss, nothing above.
I tested too any opamps. OPA134 was good, but little warmer for me, OPA637 is better detailed and punchy. OPA128SM is winner - detailed almost as OPA637, but more stage and musical without too warm. I generaly prefered detailed sound over relaxed.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Any details of my design:
- Toroids with rectifiers are in separate chassis
- Supplies are 18V with 18V BZX85 zeners. Voltage on opamp is 17,17V.
- Tranzistors are Philips 2N5401 and 2N5551 matched for hfe from 2x200 pieces
- Resistors are Vishay RN and CMF
- Supply capacitors Elna Silmic II
- Board capacitors Nichicon Muse
Amplifier is not yet fully finished, I wait for any parts and prepare latest mods:
- Mini XLR connectors for supply-amp interconnect
- Copper sheet as shield between channels in both chassis and maybe for RCA-pot wire
- Vishay Dale stepper attenuator
- Back switch for switching 2 inputs
- Secured bottom switch for input capacitor bypass
- Better Jack connector
- Finaly will be amplifier wired with pure silver
After first switch on DC offset with OPA134 was 5mV and 2mV. At listening is whole time under 10mV (about 5mV ordinary).
I can say sound is fantastic - extremely neutral. Lot of power, very detailed without any harsness. I never hear so huge stage on my headphones. I listen third day, switching genres - all is OK. From pop to metal, from bass to treble. Nothing miss, nothing above.
I tested too any opamps. OPA134 was good, but little warmer for me, OPA637 is better detailed and punchy. OPA128SM is winner - detailed almost as OPA637, but more stage and musical without too warm. I generaly prefered detailed sound over relaxed.
very nice build. i like 
have you tried the AD744? its very quick and detailed at first i thought that the bass was missing till i realised that it was just starting and stopping so quick.
hmmm must try the OPA128SM
so far i have tried:
OPA134
OPA132
OPA551
OPA27
OPA177
OPA227
OPA228
AD711
AD744
AD797 (volts to high)
Signetics NE5534N (volts to high)
have you tried the AD744? its very quick and detailed at first i thought that the bass was missing till i realised that it was just starting and stopping so quick.
hmmm must try the OPA128SM
so far i have tried:
OPA134
OPA132
OPA551
OPA27
OPA177
OPA227
OPA228
AD711
AD744
AD797 (volts to high)
Signetics NE5534N (volts to high)
once i have had a word with Richard to order some V3 boards then i will mess with the v2 AD744 and pin 5. until then i'm liking what i hear
i built the V2 without any bypass caps or zobel then swapped opamps about until i got ones i liked. since then i have only put the bypass caps on the opamps. i was disappointed till the caps burned in which did not take long and now they are its brought the open sound back. funny thing is the bass is getting better each night or its just me getting used to the sound. i liked the OPA134 but the AD744 is just sooo much more open and 'quicker' which is more to my taste.
on DSOTM Time when the toms do their bit every single opamp i have tried has 'torn' the sound until the AD744, its still there but ever soooo slightly now that i dont wince.
i built the V2 without any bypass caps or zobel then swapped opamps about until i got ones i liked. since then i have only put the bypass caps on the opamps. i was disappointed till the caps burned in which did not take long and now they are its brought the open sound back. funny thing is the bass is getting better each night or its just me getting used to the sound. i liked the OPA134 but the AD744 is just sooo much more open and 'quicker' which is more to my taste.
on DSOTM Time when the toms do their bit every single opamp i have tried has 'torn' the sound until the AD744, its still there but ever soooo slightly now that i dont wince.
Notes for OPA128SM:
Voltage noise is 15-92 nV/sqrtHz. (NE5534A is closer to about 3.5 nV/sqrtHz.)
GBWP (unity gain bandwidth) is only about 1 MHz. (NE5534A is about 8 MHz.)
On the face of it it wouldn't be my first choice, but it should work fine as long as the gain isn't set too high.
AD744 13 Mhz bandwidth (fine), en = 16-45 nV/sqrtHz (a little on the high side).
Voltage noise is 15-92 nV/sqrtHz. (NE5534A is closer to about 3.5 nV/sqrtHz.)
GBWP (unity gain bandwidth) is only about 1 MHz. (NE5534A is about 8 MHz.)
On the face of it it wouldn't be my first choice, but it should work fine as long as the gain isn't set too high.
AD744 13 Mhz bandwidth (fine), en = 16-45 nV/sqrtHz (a little on the high side).
Q, can i omit the volume control and instead use the gain R4 (V2 boards) resistor to set the output?
let me explain a bit about this, i never move the volume control more than 1 click on the Stepped Ladder Attenuator and it sits at 1 o'clock so i thought of removing the volume which in turn would be less in the audio path.
or even more of a stupid idea would be to use a switch with say 4 poles as a sort of basic SLA and have it set up for turning the gain up/down via R4.
let me explain a bit about this, i never move the volume control more than 1 click on the Stepped Ladder Attenuator and it sits at 1 o'clock so i thought of removing the volume which in turn would be less in the audio path.
or even more of a stupid idea would be to use a switch with say 4 poles as a sort of basic SLA and have it set up for turning the gain up/down via R4.
interesting. lets say i put a Resistive Optocoupler in place of R4 and then 'trim' R3 so when the LED of the RO is at full off it would be 25db and then use a trim on the power that's feeding the LED on the RO to set the full on at 15db, then use a potentiometer for adjusting the power of the LED on the RO that would give a semi-adjustable gain. the RO is around 40k when full off.
to me having a volume control is a waste when i never use 90% of it.
to me having a volume control is a waste when i never use 90% of it.
You are paying for the feature - an adjustable voltage divider. I can't say I subscribe to the view that it's a waste if you don't use the full range available. It's the convenience of doing it that way vs. the cost and convenience of doing it another way.
I pretty much always use mine at about the same volume setting, within 2-3 clicks of 10 o'clock on the Goldpoint attenuator. Of the 24 steps, the 12 steps above the 12 o'clock are never used. I think it's largely all in the head - we like to imagine we have the option of going 'to eleven' even though we never do.
You could instead of a 50k pot use a 5k pot in series with a 45k resistor. The volume would max out at about -20 dB. Which if you never went higher would give you a wider control in the useful range. Oddly though people dont like that so much, as the amp feels 'weak', like there is no power on tap.
I pretty much always use mine at about the same volume setting, within 2-3 clicks of 10 o'clock on the Goldpoint attenuator. Of the 24 steps, the 12 steps above the 12 o'clock are never used. I think it's largely all in the head - we like to imagine we have the option of going 'to eleven' even though we never do.
You could instead of a 50k pot use a 5k pot in series with a 45k resistor. The volume would max out at about -20 dB. Which if you never went higher would give you a wider control in the useful range. Oddly though people dont like that so much, as the amp feels 'weak', like there is no power on tap.