How much noise is acceptable?

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
This is a question about the art of the possible. I've built a stereo amp based on the LM1875 (datasheet x20 gain circuit), it has a microprocessor controlled set of relays controlling 3 sets of phono inputs and a DAC, these feed into a OPA2134 opamp unity gain buffer and then into a PGA2311 which in turn feeds the LM 1875s.

If I am not playing anything through the DAC and turn the PGA2311 up to maximum gain (+31.5db), I get a noise, just audible at 8 ft, through the speakers. This seems to be 50Hz related but doesn't include the fundamental hum. Once I turn the gain in the PGA down below about +20db this sound disappears (again from about 8ft) but obviously the noise component is still there in the signal.

Normal listening levels are settings on the PGA of -30db to -10db.

What are others accepting as a system noise level? Is it worth trying to chase this down or am I being too ambitious?


measure the noise level at the input to the amplifier and at the output from the amplifier.
A good amplifier should have less than 1mVac of noise at the output, when the input is shorted.
A very good amplifier should have less than 0.1mVac of noise at the output, with input shorted.

If the amplifier has a gain of 20times you would expect the input referred noise to be <5uVac to <50uVac. If you can measure anything above 0.1mVac at the output of your preamp then it is noisier than your amp. Get a new design of preamp.
I can't measure down to those levels.
My portable DMMs read 0.1mVac or 0.1mVdc at least significant bit. If that bit is off then it's likely that the voltages are below 50uVac or dc. The ac voltage scale have terrible bandwidth. Bench top DMMs can be much better, both in sensitivity and accuracy and rms reading bandwidth.

There are a few designs of +40,+60 and even +80dB audio band amplifiers to extend your meter or scope down to the uV sensitivity.
Last edited:
You can build a preamp for your DMM with a low noise chip like the AD797, LT1028 -- most DMM's however, are bandwidth limited and the frequency to which humans are most noise sensitive is in the range of 800Hz to 3.2kHz. You might do better measuring noise with a soundcard (and something like Pete Millet's sound card interface.)
If I am not playing anything through the DAC and turn the PGA2311 up to maximum gain (+31.5db), I get a noise, just audible at 8 ft ....

.... Normal listening levels are settings on the PGA of -30db to -10db.

You are being way too ambitious. Let's say you do listen normally at -20 dB. The noise is audible at +30 dB. That is a difference of 50 dB, or 100000x (power), which is a voltage difference of 316x. All systems have noise, and if you apply enough gain to any system you can bring the noise up to problematic levels. Why not just accept that the noise is completely inaudible at normal listening levels and leave it at that?

You may bring an improvement by optimizing the dynamic range of your system. You said you listen at between -30 dB and -10 dB. I'd assume that the sweet spot for the PGA is around 0 dB unity gain (neither attenuating nor amplifying). So try reducing the overall gain of your system by 20 dB so that your normal listening settings are between -10 dB and +10 dB. You could knock the LM1875 down from 20x to 10x for 6 dB of loss (can't go below 10x). You don't have gain anywhere else, so you could also put a fixed 1/5 attenuator between the PGA and the LM for the other 14 dB. Doing one or both of those should reduce the noise floor at a given output level. You'll need to try it to see if you like it. Changing gain of the LM in particular could change the character of the sound.
A good pre/power combo with Dac connected and no signal should be quiet enough at min volume setting to have only a whisper of white noise with your ear to the speaker. With vol full and no signal the hiss should be a little louder at the speaker but barely if at all audible at the listening position. In neither case should there be any hum or buzz.

Sent from my Android using Tapatalk

Thanks for the interesting and informative replies. I've played with a few things and discovered a few things as follows.

My scope does go down to 10mv per division so I could get quite accurate measurements.

The LM1875 amp with the input shorted produces a straight line on the scope connected across the outputs with no measurable DC offset so that looks good.

I did find that by moving the star ground connection on the amp from the power input to the signal ground I eliminated any 50Hz trace on the scope when using the PGA2311 to route the input. This idea came from the document.

If I remove the PGA2311 and bypass it by connecting "in" to "out" in the chip socket there is no noise from any signal source so the OPA2134 buffer is behaving well. But I still get noise (now more like white noise) when using the PGA in the +20 to +30db range.

So it looks like linuxworks is absolutely correct. These are noisy amplifiers when operated towards the top of their range.

My understanding is that internally the PGA bypasses its internal amp for 0 db and below. So the optimum solution appears to be to increase the gain on the power amp a bit so that even when I want it "loud" the PGA is attenuating.

Again, thanks all for the ideas and info

Best regards

I was developing code (arduino stuff) for the PGA chips and was dicking around for a few months with them. mostly I was doing headphone amp and DAC testing and my components were decent enough that I could tell when the PGA was in clean mode and when it wasn't. and below 0 was its good area ;)

I liked having some db above zero for strange cases like ac3->2ch downmix where you often can lose 10db in the process (don't ask me why but it does happen). having extra room up there helps in cases like that. but normally I make sure my DAC can drive the rest and the attenuator sits at -10 db for normal 'wow this is loud but still clear' listening. only very rarely should I even have to hit zero.

also, it seemed like giving the analog supply to the pga chip was important and you could always do with an even lower noise PSU system for it. I didn't take mine far enough (used 317 based regs) but an ultra low noise PSU might help make that chip really high-end capable. and keep the 5v digital supply line quite separate from the analog ones (I'm sure you know that already).
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.