Simple question with most likely a complex answer. I'm posting it here rather than in the headphone forum as the guys over there rarely get further than worrying over which op amp to use...
On one hand, it is easy to determine class A bias current: decide on the maximum power output, calculate the maximum current for the minimum load, and set the bias current to be slightly in excess of the peak value (1.4x rms).
For a headphone amp, (arbitrarily) specifying 20 mW, 16 ohms, rms is 35 mA, peak 50 mA.
Thing is, for a headphone amp its fairly easy to built it out for higher bias. The more bias current, the more linear the circuit gets for the same output current. At the same time, the noise from the output devices also increases with current, forcing the designer to choose: low noise or low distortion?
Is the device noise issue real or not significant in typical applications (e.i. does the input noise upstream dominate)?
Is there an ideal compromise point or is it better to just up the bias to the practical maximum and forget about it?
Once you've decided on the bias current, how to you go about choosing the best transistor to achieve lowest noise?
LTSpice sim attached, though its a crutch. It would be better to calculate the device noise rather than rely in model simulation.
On one hand, it is easy to determine class A bias current: decide on the maximum power output, calculate the maximum current for the minimum load, and set the bias current to be slightly in excess of the peak value (1.4x rms).
For a headphone amp, (arbitrarily) specifying 20 mW, 16 ohms, rms is 35 mA, peak 50 mA.
Thing is, for a headphone amp its fairly easy to built it out for higher bias. The more bias current, the more linear the circuit gets for the same output current. At the same time, the noise from the output devices also increases with current, forcing the designer to choose: low noise or low distortion?
Is the device noise issue real or not significant in typical applications (e.i. does the input noise upstream dominate)?
Is there an ideal compromise point or is it better to just up the bias to the practical maximum and forget about it?
Once you've decided on the bias current, how to you go about choosing the best transistor to achieve lowest noise?
LTSpice sim attached, though its a crutch. It would be better to calculate the device noise rather than rely in model simulation.
Attachments
Last edited:
That would have been my intuitive conclusion. Good. Thanks.
My listening evaluation suggested a slightly higher noise floor, and I was wondering if the increased bias current was the cause. Your answer says no, and actually the measurements I've just made show its not a noise floor increase per se. but interference of some kind (hash rather than white noise). I'll have to look more carefully at the grounding and bypassing layout on the new boards.
My listening evaluation suggested a slightly higher noise floor, and I was wondering if the increased bias current was the cause. Your answer says no, and actually the measurements I've just made show its not a noise floor increase per se. but interference of some kind (hash rather than white noise). I'll have to look more carefully at the grounding and bypassing layout on the new boards.
try adjusting the Re value to suit the peak output current value that your load may demand.
A loudspeaker demanding typically 100mApk to 1Apk tends to have Re ~ 0r1 to 0r47
A headphone amp demanding 10mA to 100mA would tend to use Re ~1r0 to 4r7
A line driving amp might have to deliver 1mA to 10mA and would suit Re 10r to 47r.
Adjust the Vre to optimally bias the output stage and you end up with bias currents that roughly scale with the load current demand.
A loudspeaker demanding typically 100mApk to 1Apk tends to have Re ~ 0r1 to 0r47
A headphone amp demanding 10mA to 100mA would tend to use Re ~1r0 to 4r7
A line driving amp might have to deliver 1mA to 10mA and would suit Re 10r to 47r.
Adjust the Vre to optimally bias the output stage and you end up with bias currents that roughly scale with the load current demand.
does the load need 100mA pk or 100mAac?
Let's assume 100mAac.
the peak value will be ~140mApk.
A push pull amplifier biased to 70mA can output a theoretical 140mA, without half switching off.
But that last bit gets half the transistor into the region where they are starting to turn off.
So set the bias to 80mA.
Assuming a 1pr output stage optimally biased into ClassAB.
We know that Vre should be around 26mV.
Re = 0.026 / 0.08 = 0r325 use 0r33
Set the bias to 26mVre and the amplifier operates in optimal ClassAB.
The Maximum ClassA current is ~140mA.
The amplifier stays in ClassA when driving a 100ohm headphone to 10vac (14Vpk)
That is far too loud when sensitivity is usually between 90 and 100dB/mW
A 32ohm headphone could be driven to ~3.2Vac and stay in ClassA. Still very loud.
Let's assume 100mAac.
the peak value will be ~140mApk.
A push pull amplifier biased to 70mA can output a theoretical 140mA, without half switching off.
But that last bit gets half the transistor into the region where they are starting to turn off.
So set the bias to 80mA.
Assuming a 1pr output stage optimally biased into ClassAB.
We know that Vre should be around 26mV.
Re = 0.026 / 0.08 = 0r325 use 0r33
Set the bias to 26mVre and the amplifier operates in optimal ClassAB.
The Maximum ClassA current is ~140mA.
The amplifier stays in ClassA when driving a 100ohm headphone to 10vac (14Vpk)
That is far too loud when sensitivity is usually between 90 and 100dB/mW
A 32ohm headphone could be driven to ~3.2Vac and stay in ClassA. Still very loud.
Indeed loud. The idea, somehow, is that larger bias currents are desirable if the thermal budget will allow, even if the maximum class A power delivery far exceeds the real world demand. LTSpice insists that it leads to lower distortion. I'm not entirely convinced.
Up to 200 mA is practical for low voltage headphone amplifiers. Now ... does that really sound better? At the moment my thinking is that 20-30 mA should work just as well. I plan to try out different settings and see if I can hear any difference.
The optimal Vre is 26 mV? That seems entirely reasonable and all, but I don't follow the reasoning exactly. Since it allows the derivation of Re for the selected bias current, though, I'm all good.
Up to 200 mA is practical for low voltage headphone amplifiers. Now ... does that really sound better? At the moment my thinking is that 20-30 mA should work just as well. I plan to try out different settings and see if I can hear any difference.
The optimal Vre is 26 mV? That seems entirely reasonable and all, but I don't follow the reasoning exactly. Since it allows the derivation of Re for the selected bias current, though, I'm all good.
There is much discussion on Vre for setting optimal ClassAB bias.
If that is used for a push pull ClassA amplifier and for some reason the load demands extra current that exceeds the ClassA value then the amplifier transitions into ClassAB with the least crossover distortion while that extra current is being sourced.
If the load never demands that extra current then the amplifier remains in ClassA and there is no crossover distortion.
For the demands of headphones this seems an ideal way to drive them.
The only thing that needs some extra thought is choosing a topology and devices that that suit the low noise required.
If that is used for a push pull ClassA amplifier and for some reason the load demands extra current that exceeds the ClassA value then the amplifier transitions into ClassAB with the least crossover distortion while that extra current is being sourced.
If the load never demands that extra current then the amplifier remains in ClassA and there is no crossover distortion.
For the demands of headphones this seems an ideal way to drive them.
The only thing that needs some extra thought is choosing a topology and devices that that suit the low noise required.
For almost all headphones, peak power demand is not more than about 1-2 mW. Since there is very little thermal constraint, is it better to design the output stage for 2 mW, 20 mW or 200 mW of class A output? That's one thing.
The other thing is absolutely this question of noise. I chose to use an audio op amp like the OPA134 for the voltage gain stage, on the idea that the noise would be as low as any typical audio preamplifier. I've assumed that the input stage determines the noise - see posts above - and the op amp has pretty decent voltage and current noise values. So is there anything else I can do to improve things?
The other thing is absolutely this question of noise. I chose to use an audio op amp like the OPA134 for the voltage gain stage, on the idea that the noise would be as low as any typical audio preamplifier. I've assumed that the input stage determines the noise - see posts above - and the op amp has pretty decent voltage and current noise values. So is there anything else I can do to improve things?
The problem, if there is one, is that headphones come in a wide variety of impedance.
A universal headphone driver needs to be able to drive an 8ohms as well as it drives a 600ohms.
5Vpk and 140mApk are possibly enough to cover that range. Many builders seem to set their targets even higher.
The problem, if there is one, is that headphones come in a wide variety of impedance.
A universal headphone driver needs to be able to drive an 8ohms as well as it drives a 600ohms.
5Vpk and 140mApk are possibly enough to cover that range. Many builders seem to set their targets even higher.
5Vpk into 600ohms is ~20mW
140mApk into 8ohms is ~40mW, if the allowance for transient current is set to 1.4times.
A 90dB/mW headphone will reach ~103dB on 20mW. Many headphone listeners will declare this is too quiet.
A 100dB/mW headphone will reach ~ 116dB on 40mW. Most will agree this is adequately loud for transient peaks.
I don't experience more than 100 dB peak when listening to headphones, and I tend to listen loud. And that's once-in-a-blue-moon, theoretical maximum based on the maximum input signal off the CD and the net gain of the complete signal chain. It's most likely 6-10 dB below that since I would not listen to such a "hot" CD at my typical volume position.
Granted, the lower the headphone impedance, the greater the current and the lower the voltage required to drive it. And low sensitivity models need more power. What I'm curious about though is for me, for my 300 ohm, 97 dB/mW headphones, only needing 2 mW max, will it sound better with more and more bias current?
I've tried about five different settings, from 8 to 70 mA. I find there is a sweet spot around 30 mA. I can't think to explain why. It works out to 120 mW class A into 300 ohms, a pretty ridiculous figure given my listening requirements.
Granted, the lower the headphone impedance, the greater the current and the lower the voltage required to drive it. And low sensitivity models need more power. What I'm curious about though is for me, for my 300 ohm, 97 dB/mW headphones, only needing 2 mW max, will it sound better with more and more bias current?
I've tried about five different settings, from 8 to 70 mA. I find there is a sweet spot around 30 mA. I can't think to explain why. It works out to 120 mW class A into 300 ohms, a pretty ridiculous figure given my listening requirements.
Usually more idle current results in lower distortion; it may become slightly complex if there are multiple stages with partial cancellation though. But as you've found, current noise also increases and may eventually become a problem in an open-loop buffer like you're using. Another reason why people are going through all this trouble with feedback. In practical AB loudspeaker amps idle current is also limited by considerations such as thermal stability, operating temperature and secondary breakdown.
Recently I got some really sensitive in-ears again (Soundmagic E10, estimated sensitivity about 10 dB over spec or almost 130 dB/V, which is really high for dynamic drivers and close to 20 dB more sensitive than any of my "full-grown" cans) and was surprised at the levels of noise they made audible. Apparently the FiiO E11 barely emits 6-7 µV(A) of noise, yet is distinctly hissy with these. To think that I've battled TDA2822 noise (~300 µV at stock gain) by using my least sensitive 600 ohm cans in the past...
I don't think a step-down xfmr is necessary though. An O2 should do fine (~2.5 µV worst-case, usually less). Too bad it's relatively bulky when compared to typical sources and loads.
A 300 ohm, 97 dB/mW load (HD580/600?) is quite undemanding really. Your 2 mW translate to 775 mV (0 dBu), or 3.6 mApk. A Class A push-pull output would drive this conveniently at about 18 mA up when using a factor-of-10 rule of thumb. (And the march of technology means that a measly Clip+ is not particularly hard-pressed to get there either.)
BTW, the initial example with 20 mW into 16 ohms at 16 mA of idle current was not particularly well-chosen - it already is outside of class A operation, as you can also tell by the dominant odd-order distortion. 25 mA would be needed at the very least.
Considerations for emitter resistors in pure class A operation are not as complex as for AB. In the latter, you want to ensure that gm does not vary too much between the "A range" (where both tranaistors operate) and the "B range" (where only one does). With that out of the way, it becomes mainly a question of thermal stability, permitted output impedance and performance degradation into low-impedance loads (when Re/2 approaches or even exceeds Rload). If possible, I would always design a headphone amp to run in class A most/all of the time, even if it means that performance will be worse than ideal in those rare cases where AB operation is demanded.
Recently I got some really sensitive in-ears again (Soundmagic E10, estimated sensitivity about 10 dB over spec or almost 130 dB/V, which is really high for dynamic drivers and close to 20 dB more sensitive than any of my "full-grown" cans) and was surprised at the levels of noise they made audible. Apparently the FiiO E11 barely emits 6-7 µV(A) of noise, yet is distinctly hissy with these. To think that I've battled TDA2822 noise (~300 µV at stock gain) by using my least sensitive 600 ohm cans in the past...
I don't think a step-down xfmr is necessary though. An O2 should do fine (~2.5 µV worst-case, usually less). Too bad it's relatively bulky when compared to typical sources and loads.
A 300 ohm, 97 dB/mW load (HD580/600?) is quite undemanding really. Your 2 mW translate to 775 mV (0 dBu), or 3.6 mApk. A Class A push-pull output would drive this conveniently at about 18 mA up when using a factor-of-10 rule of thumb. (And the march of technology means that a measly Clip+ is not particularly hard-pressed to get there either.)
BTW, the initial example with 20 mW into 16 ohms at 16 mA of idle current was not particularly well-chosen - it already is outside of class A operation, as you can also tell by the dominant odd-order distortion. 25 mA would be needed at the very least.
Considerations for emitter resistors in pure class A operation are not as complex as for AB. In the latter, you want to ensure that gm does not vary too much between the "A range" (where both tranaistors operate) and the "B range" (where only one does). With that out of the way, it becomes mainly a question of thermal stability, permitted output impedance and performance degradation into low-impedance loads (when Re/2 approaches or even exceeds Rload). If possible, I would always design a headphone amp to run in class A most/all of the time, even if it means that performance will be worse than ideal in those rare cases where AB operation is demanded.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.