I'm getting barely audible static noise from my amplifier, but I don't know how I can find the source. I don't have equipment to measure levels this low, so I don't know how loud the noise actually is. Here's a quick circuit:
I don't know much about noise sources or how to calculate them, but I would like to be able to calculate the inherent noise created by the amplifier itself (while the input pot is turned all the way down). The only noise sources I know of is the inherent noise of the op-amp itself and the thermal noise caused by the external resistors. Here's some figures from the datasheet of the op-amp:
I found a formula for thermal noise on wikipedia: v^2=sqrt(4kB*TR). For room temperature and 100kOhm (input resistor) value gives a result of 40nV/sqrt(hz).
The datasheet gives an inherent noise density figure of 35nV/sqrt(hz) @ 10kHz. Adding these two figures together gives sqrt(40^2+35^2)=53nV/sqrt(hz). Since this is an audio amplifier my bandwidth is 20kHz so total noise is 53*sqrt(20000)=7495nV. With a gain of 3 the resulting output noise should be 22uV or are my calculations way of? I'm just taking a shot in the dark here, could someone help me out?
Is 22uV of noise really audible?
An externally hosted image should be here but it was not working when we last tested it.
I don't know much about noise sources or how to calculate them, but I would like to be able to calculate the inherent noise created by the amplifier itself (while the input pot is turned all the way down). The only noise sources I know of is the inherent noise of the op-amp itself and the thermal noise caused by the external resistors. Here's some figures from the datasheet of the op-amp:
An externally hosted image should be here but it was not working when we last tested it.
I found a formula for thermal noise on wikipedia: v^2=sqrt(4kB*TR). For room temperature and 100kOhm (input resistor) value gives a result of 40nV/sqrt(hz).
The datasheet gives an inherent noise density figure of 35nV/sqrt(hz) @ 10kHz. Adding these two figures together gives sqrt(40^2+35^2)=53nV/sqrt(hz). Since this is an audio amplifier my bandwidth is 20kHz so total noise is 53*sqrt(20000)=7495nV. With a gain of 3 the resulting output noise should be 22uV or are my calculations way of? I'm just taking a shot in the dark here, could someone help me out?
Is 22uV of noise really audible?
That level of noise is not audible. Why just the 100k at the positive input - what's the source impedance?
Last edited:
Use diodes instead of resistors for biasing your output. If resistors are used it's more sensitive to power supply noise - since you have just a voltage divider any variation in supply voltage would simply show up at the base of your transistors, giving you in this case a whopping -26dB PSRR.
22uV is -93dB relative to 1V, can be heard as a whisper on ear/headphones. Inaudible on speakers usually unless you put your ears against the tweeter.
22uV is -93dB relative to 1V, can be heard as a whisper on ear/headphones. Inaudible on speakers usually unless you put your ears against the tweeter.
The source impedance will depends upon the source though, won't it? It has a 3.3uF input cap as well, but I've shorted the input pot to ground in this example so the 3.3uF didn't have much going for it.
Having diodes will seriously cripple the amplifier output voltage capabilities compared to having a voltage divider though, won't it? How did you come up with -26dB PSRR? I know that there's about 1Vpp of ripple on the supply, but humming is not audible. I did some testing with my most sensitive headphones and a gain of as much as 25, humming could barely be heard, but at this level the static hissing was completely dominant.
I would like to know if my calculations are correct and if the two noise sources that I have included are the only ones? The noise is really really low and I'm testing with headphones that has a sensitivity of 109dB SPL/V.
This picture gives a better... picture... than words:
Source
-26dB is based on open-loop. With feedback the amp will try to correct it and greatly reduce the output error, assuming the amp and feedback loop is fast enough. Usually for audio frequencies it is fast enough, but RF not always.
Anyway, consider the open-loop case, with your op-amp output at ground. We make an assumption here: Output voltage = Voltage at Base of transistor minus Vbe. If you're biasing using voltage divider, then voltage at Base = Power supply voltage multiply by (some value). Meaning any fluctuation in power supply causes the Base voltage and subsequently output to fluctuate at 1/(some value). That (some value) is -26dB based on those resistor values.
Now, one interesting thing about this noise, is that it goes away if the same noise is on both + and - supply rails. If the noise is only present on one rail (say the +), the higher Base voltage of one transistor cause more current on one transistor compared to the other - this causes output voltage to shift from ground. But if both transistors get higher Base voltage at the same time, current through both is increased, the voltage changes are symmetrical, so output voltage is unaffected.
With most audio power supplies a center-tapped transformer is used so the charging cycle for both + and - rails happen at the same time and this noise cancels out.
So I've mentioned why the 50Hz hum isn't audible, but random noise behaves differently and hence can still show up. One, if random noise is on both + and - rails, it will not cancel out like in the previous case, but would instead add together. Second, random noise is across all frequencies including RF - which can modulate some way some how and affect the audio frequencies, and may not be filtered away by your feedback and etc.
Well that's my hypothesis.
An externally hosted image should be here but it was not working when we last tested it.
Source
-26dB is based on open-loop. With feedback the amp will try to correct it and greatly reduce the output error, assuming the amp and feedback loop is fast enough. Usually for audio frequencies it is fast enough, but RF not always.
Anyway, consider the open-loop case, with your op-amp output at ground. We make an assumption here: Output voltage = Voltage at Base of transistor minus Vbe. If you're biasing using voltage divider, then voltage at Base = Power supply voltage multiply by (some value). Meaning any fluctuation in power supply causes the Base voltage and subsequently output to fluctuate at 1/(some value). That (some value) is -26dB based on those resistor values.
Now, one interesting thing about this noise, is that it goes away if the same noise is on both + and - supply rails. If the noise is only present on one rail (say the +), the higher Base voltage of one transistor cause more current on one transistor compared to the other - this causes output voltage to shift from ground. But if both transistors get higher Base voltage at the same time, current through both is increased, the voltage changes are symmetrical, so output voltage is unaffected.
With most audio power supplies a center-tapped transformer is used so the charging cycle for both + and - rails happen at the same time and this noise cancels out.
So I've mentioned why the 50Hz hum isn't audible, but random noise behaves differently and hence can still show up. One, if random noise is on both + and - rails, it will not cancel out like in the previous case, but would instead add together. Second, random noise is across all frequencies including RF - which can modulate some way some how and affect the audio frequencies, and may not be filtered away by your feedback and etc.
Well that's my hypothesis.
There are many more noise sources around the first stage opamp, than you have allowed for.
I recently stumbled up a Ti pdf that gave a very thorough, but understandable, analysis for predicting the noise output of their opamps. It was linked for us from a Thread here.
I'll see if I can find the copy on my HDD, but that won't have the link to the original.
However, if you have ONE dominant noise source and use that, the error in your prediction as a result of ignoring all the other non dominant noises, is very small.
With 100k tacked onto your opamp, then I would guess you have correctly identified the dominant noise.
But like the other Member I have to ask, why 100k, what's wrong with 6k666? (to match 10k||20k0)
Would that be the dominant noise now?
I recently stumbled up a Ti pdf that gave a very thorough, but understandable, analysis for predicting the noise output of their opamps. It was linked for us from a Thread here.
I'll see if I can find the copy on my HDD, but that won't have the link to the original.
However, if you have ONE dominant noise source and use that, the error in your prediction as a result of ignoring all the other non dominant noises, is very small.
With 100k tacked onto your opamp, then I would guess you have correctly identified the dominant noise.
But like the other Member I have to ask, why 100k, what's wrong with 6k666? (to match 10k||20k0)
Would that be the dominant noise now?
The input bias current of this op-amp is so small that it doesn't matter. I just smacked on the 100k to make sure the cutoff frequency was well below 20hz. I could reduce the 100k resistor just to see what happens to the noise. I'll take a look at that pdf also, thanks a lot AndrewT
Also, how do I determine what noise is caused my the amplifier itself and what comes from the outside? With input cables connected and volume turned all the way up the hissing is intense, but disconnecting the input cables removes the hiss, but I'm not sure what this means? The internal input wires run from the back of the box, past the mains transformer and to the front amplifier PCB, but how do I isolate the noise that this cable is picking up (if any)?
Last edited:
if the input cable is open ended then you are hearing/measuring the noise of the smallest resistor across the input. If that resistor is 1M (to discharge the leakage of the DC blocking cap) then you have a lot of noise.
If the vol pot is the smallest resistance then you have 100k (10dB quieter than the 1M).
If the input to the full open vol pot is shorted or connected to a Rs=100r, then that short becomes the smallest input resistor and the noise disappears below audibility.
One never has an open ended input with a "connected" system, provided the Source is powered ON.
Your 109dB/V headphones when fed with a noise of -93dB below 1Vac will reproduce ~16dB SPL at your ears.
That may not be audible. It depends on the background noise, how recently you listened to louder noises, how fast/hard your heart is pumping, etc.
If the vol pot is the smallest resistance then you have 100k (10dB quieter than the 1M).
If the input to the full open vol pot is shorted or connected to a Rs=100r, then that short becomes the smallest input resistor and the noise disappears below audibility.
Just pace a "short" across the far end.
One never has an open ended input with a "connected" system, provided the Source is powered ON.
Your 109dB/V headphones when fed with a noise of -93dB below 1Vac will reproduce ~16dB SPL at your ears.
That may not be audible. It depends on the background noise, how recently you listened to louder noises, how fast/hard your heart is pumping, etc.
Last edited:
Shorting the input had no effect on the noise, neither has adjusting the volume pot (which is only 10k Ohm so that makes sense). Using my cellphone instead of desktop as a source removed that intense high-volume hissing so I definitively have a noise sound card
The noise (when volume all the way down) is audible, not low enough to depend upon my heartbeat and such so It's definitively more than -93dBv. Either my calculations aren't correct (I would like someone to take a look at them if they have time) or there's some noise source that I haven't taken into account. The point made by wwenze about high frequency static noise making it's way into the output stage might be something to check out? I understand how the op-amp can have troubles compensating for high frequency noise (parasitic capacitance in the feedback path etc), but how does this lead to audible noise? Also, if I remake my circuit using diodes instead of voltage divider, would I be able to accurately set the bias current? I'm just thinking that it would depend on the voltage/current relationship between the diodes and base-emitter junction?
As jackinnj said, the OPA454 can drive up to 100mA. I haven't come across any thermal instability, the bias current is always in the order of 70-100mA (slightly rising as temperature rise of course) with these transistors and resistor values. The supply is +-21V by the way.
The noise (when volume all the way down) is audible, not low enough to depend upon my heartbeat and such so It's definitively more than -93dBv. Either my calculations aren't correct (I would like someone to take a look at them if they have time) or there's some noise source that I haven't taken into account. The point made by wwenze about high frequency static noise making it's way into the output stage might be something to check out? I understand how the op-amp can have troubles compensating for high frequency noise (parasitic capacitance in the feedback path etc), but how does this lead to audible noise? Also, if I remake my circuit using diodes instead of voltage divider, would I be able to accurately set the bias current? I'm just thinking that it would depend on the voltage/current relationship between the diodes and base-emitter junction?
As jackinnj said, the OPA454 can drive up to 100mA. I haven't come across any thermal instability, the bias current is always in the order of 70-100mA (slightly rising as temperature rise of course) with these transistors and resistor values. The supply is +-21V by the way.
Then you are doing something wrong.That is correct. Using my cellphone as a source makes the noise constant regardless of volume position. The intense hissing was caused by my crappy motherboard probably
a.) First check noise at the last component in the chain. Short the input and measure the noise at the output.
b.) Add on your interconnects. Short the input. Measure the noise at the output.
c.) Add on the volume control/pre-amp. Short the input. Measure the noise at the output.
d.) Add on the Source. Reduce vol to minimum. Measure the noise at the output.
e.) Mute or something similar to the Source. Increase volume. Measure the noise at the output.
Now to your claim. If the Source is connected and the volume is turned to maximum then shorting the input to the volume control/pre-amp will result in shorting the source signal. This takes the system back 3 steps to c It is clear that if the system passed step a, b and c then adding the short takes you back to c, then it must also pass.
Do what you are actually claiming to have done.
If that's the amp from his other thread (and it seems it is), it's got signal ground connected to PE. Connect a PC, and you've got a first-rate ground loop, with who knows how much power supply crap being made audible.
And as mentioned, yes, the biasing leaves much to be desired. I'm actually surprised it's not thermal runaway personified, though I guess it doesn't get that bad in a single-stage buffer yet. Maximum output sims at about 7 W into 8 ohms or so, the opamp still is well short of its current limit at that point.
Is the gain really only 3x? I could understand if there were some noise at 21x (20k/1k), as the OPA454 is a pretty noisy bugger. Even that would still be at decent integrated amp level though.
All in all, I bet results would have been better with an LM4766/LM4780 or somesuch.
And as mentioned, yes, the biasing leaves much to be desired. I'm actually surprised it's not thermal runaway personified, though I guess it doesn't get that bad in a single-stage buffer yet. Maximum output sims at about 7 W into 8 ohms or so, the opamp still is well short of its current limit at that point.
Is the gain really only 3x? I could understand if there were some noise at 21x (20k/1k), as the OPA454 is a pretty noisy bugger. Even that would still be at decent integrated amp level though.
All in all, I bet results would have been better with an LM4766/LM4780 or somesuch.
+1 for firing up TINA-TI. It's a free download from Analog, Embedded Processing, Semiconductor Company, Texas Instruments - TI.com.
~Tom
wwenze: I have a question about this way of biasing a class AB stage:
If I go with global feedback on this configuration the amp won't have any DC gain so the op-amp can't change the DC output, right? It will just ram the output to either +V or -V unable to compensate for the DC offset, right? How is this solved? I find settling for local feedback unsettling
An externally hosted image should be here but it was not working when we last tested it.
If I go with global feedback on this configuration the amp won't have any DC gain so the op-amp can't change the DC output, right? It will just ram the output to either +V or -V unable to compensate for the DC offset, right? How is this solved? I find settling for local feedback unsettling
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.