TOPIC:

My NanoAVR analogue outputs are producing more white noise than I would have expected. The tweeter's noise spectrum is here:



That is measured with a UMIK in REW (w/ calibration file, 1/12 oct smoothing, spectrum). The bump around 10khz is likely due to nearfield effects from the proximity of the mic to the tweeter which was necessary to elevate the signal above the UMIK's own noise floor.

This noise is produced at all times by the nanoAVR when powered on, and disappears from the speaker only when the relevant RCA cable is disconnected (from either end of the cable) between the amp and nanoAVR, or for brief periods (~1 second) when switching the HDMI input (is the DAC is turned off during the switch?). I've also run the nanoAVR on battery power only, with a single RCA output (no HDMI inputs), and that channel did indeed have the same noise level (suggesting it is not a ground loop/power supply noise).

My current system is aimed at reaching 105db, so I was hoping the noise floor would be a little lower so i wouldn't be able to hear noise at the main seat during quiet music/scenes. Currently, the noise is high enough to notice at the seat, which is significantly higher than the amplifier's noise (Emotiva basx A-500, noise=100db(A)@1 watt).

I guess my question is; is this a normal noise level for this device? I can't see any hard specs for the nanoAVR's noise levels to actually compare against. I'm hoping someone might see something i've overlooked so I can fix this. The only alternative I can see is to find a 5 channel pre-amp with gain control, or a separate 5 channel HDMI DAC (neither of which are preferred due to cost reasons).

Hello,
the nanoAVR HDA uses the miniDAC8, equipped with the AK4440. It's SNR is specified in the datasheet to be 98dBA min, 105dBA typ.
So, with 2Vrms output voltage, the remainig noise output voltage should be not lower than 11µVrms approximately.
Regards, Timo

Okay, if that's the case, i should be getting a ballpark of <7dB noise at the seating position (105-98). I'm getting much more than this though, being able to hear it over the road traffic noise that makes it into the room. I'll see what I can do to make some semi-accurate measurements of the noise levels coming directly from the nanoAVR within a day or so.

My estimate (by ear) would be 20-30 dB at the seating position, which would translate into a SNR of 75-85 dB. But take my estimate as a grain of salt - it's only a semi-educated guess.

So I tried a few attempts at measuring the noise levels, and the only method that seemed to give me a confident result was the following:
Leaving the speaker connected to the emotiva amplifier, i connected my handheld digital oscilloscope (DSO Nano - very cheap device, prob 90% accuracy with voltages, but it'll have to do) to the speaker terminals in parallel with the speakers. I then measured the following noise levels (all in RMS):
Oscilloscope short circuit (noise floor): 1.15 mV
Amplifier powered off: 1.15mV
Amplifier powered on, nanoAVR disconnected: 3.8mV
Amplifier powered on, nanoAVR connected: 45mV
Noise due to nanoAVR: ~41mV

As the Emotiva amp has 29dB gain (from the spec sheet), my calculations put the noise floor at -63dB for my nanoAVR [20*log(2/0.041)+29]=63. This would translate into a noise at the main seating position of approximately 105-63=42dB, i.e. audible in an average room.

I acknowledge my equipment isn't the highest accuracy, but I believe it is sufficient to get a reasonable approximation of the real noise level.

I've also tested routing of power and signal cables, but had no noticeable improvement.

Also note that measurements were with an oscilloscope, not a spectrum analyser, so it might be possible my data includes noise outside of the audible spectrum.

Is my testing methodology acceptable, and do my equations match the method? It's been a while since I've worked with converting decibel equations, so it's quite possible I've made a mistake.

If my method+math is correct, is there anything I've overlooked, or is my device likely to be somewhat faulty?

Hello,
the noise level seems to be too high, if one relates it to the nano AVR output.
Maybe, it contains (higher) mains harmonics, so a spectrum would be helpful. Otherwise you could try to trigger on peaks inside the noise, visible on the oscilloscope by varying the trigger level slowly to search for dominant frequencies.
Here I described my version of cabling:
www.minidsp.com/forum/minidsp-for-newbie...lanced-outputs#36971
Differential signal transmission and careful cable making are really helpful and not necessarily expensive. Unfortunately the AK4440 and therefore the nanoAVR HDA do not support this.
Look into the suggested papers!
Regards, Timo

Before I made the first post, I'd spent quite a while looking for a hint of the noise source using my analogue oscilloscope, looking at pretty much every variation of x1, x10 probes, time/division setting, and trigger setting/setpoint. I haven't been able to see any patterns emerge that would show a dominant frequency. I don't have a spectrum analyser (other than using the UMIK with REW, but that mostly has too much noise to make accurate measurements), and i'm not really game enough to put a power amp's output directly into my laptop's mic input to measure the noise that way in case small voltage spikes ruin the soundcard.

The best I can offer is the REW measurement in the first post, which shows a broadband noise over the tweeter's crossover range. I can also now verify that the strange shape is indeed due to the mic's placement, as a generated louder white noise signal shows the exact same shape, only increased in amplitude.

Hi,
it's a common practice for measuring amplifiers to connect their output to the soundcard's input.
For protection you may insert a 10:1 (or ~28:1 matching the 29dB gain ) voltage divider.
But didn't you want to measure the nanoAVR's output? So connect this to the soundcard, possibly in parallel to the amps input, because this may affect the "cabling" noise. Try to connect what you think should be "signal ground" temporarily, so you will see immediate effects at the spectrum.
Regards

I've tried a few more things today;

1) Moving all equipment to a testing bench, but same house power - no difference
2) Replacing the emotiva amp with my older yamaha amp, then adjusting the gain to match the emotiva - This gave me some interesting data, and i'm not entirely sure what to make of it. These are my measurements:



As we can see, the amplifiers both have the same gain at the test signal (pink noise 4kHz-20kHz - the curve is due to UMIK placement). However, they are outputting a different noise level at the speaker by over 5dB. However, when disconnecting the AVR, both amps product no noticeable noise at the speaker. What's more, is that when i used my phone to play a silent track through the emotiva, the noise was also absent. I'm struggling to figure out where in the signal chain the noise could actually be coming from.

3) I also directly connected the nanoAVR to my soundcard to take measurements.



This shows that in this configuration the nanoAVR has a noise floor around -85dB (the difference between the highest signal and -inf dB signal. The AVR was also powered down to measure the sound card's noise floor to verify that the results are meaningful (they are in high frequencies).

Looking at this data, we have a -85dB noise floor which is higher than i'd like, but is much better than the -63dB i calculated with my digital oscilloscope. But I'm not sure why there is a difference between these numbers.

---

So after today I'm totally at a loss. I don't know which device (if any) is faulty, or where the noise is being generated. The emotiva isn't noisy with other inputs, and the nanoAVR noise level is probably acceptable if it really is -85dB (my room isn't terribly quiet). So I can't understand why pairing the two devices ends up with so much (~20dB) more noise.
I wish I had another nanoAVR to test because that's pretty much the only thing in my system that I haven't been able to change.

EDIT: fixed graphs

Hi,
hmm, the graphs do not look so well. I have difficulties to get the right facts from them. Please don't put to much information into one graph.
May I ask you for a second try? An older example, how could it look like or similar is this:
ibtk.de/project/class_d/measurements/20060318_UcD_4Ohm.htm
or this:
ibtk.de/project/class_d/measurements/200...d_fg_distortion.html
Try the electrical measures first. Double-check all the levels into and out of the soundcard. Do a loopback measurement first.
Regards

Sorry about my lack of explanation of the graphs. To give some context, my intention with the graphs was to show the relative differences between each of the lines rather than a straight SPL/voltage measurement. By observing where the lines start getting messy and overlapping, we can see the noise floors of my equipment (soundcard, UMIK microphone and nanoAVR).

The 1st graph is to show that different amplifiers (which are calibrated to have the same gain) are showing different noise levels from the nanoAVR. The -60dB white noise signal was fed to each amplifier (and then measured with a microphone at the speaker) in order to calibrate their gain. However, upon removing the signal (-inf dB) but leaving the nanoAVR connected and powered, we can see the two amplifiers are outputting a 5dB difference in noise levels coming from the nanoAVR. I chose to overlay that data on a single graph to demonstrate the 5dB difference visually.

For example, the 2nd graph shows that noise levels make the data <4kHz unusable at low levels, but the relative scale shows that the soundcard's noise floor is still good enough to give an indication of the nanoAVR's noise floor. By subtracting the lowest signal (-inf dB) from the highest signal (-3dB) signal, we can see that the nanoAVR's noise floor is at about -85dB (at high frequencies where measurements are valid). I should note that the absolute scale of the 2nd graph means nothing as it is a voltage being measured rather than an actual SPL, but the voltage is calibrated. But that's why we all love the dB's log scale isn't it

If you want me to separate out the graphs, can you tell me specifically what graphs you're looking for? That way I can try to make it a bit easier to read without having to dump dozens of graphs.

As for doing a loopback test, what would we be looking for? If I'm looking at the noise floor of the soundcard would that be as accurate as the "nanoAVR no power" signal shown at the bottom of the 2nd graph? I can do the test, but I'd like to know what i'm looking for too, as i'm trying to learn the context of everything i'm doing too.

EDIT:
I just re-read your previous post:

So connect this to the soundcard, possibly in parallel to the amps input, because this may affect the "cabling" noise. Try to connect what you think should be "signal ground" temporarily, so you will see immediate effects at the spectrum.

I don't understand what you mean by connecting the "signal ground". Do you mean the RCA's outside connection should be earthed to my PC when I make the measurement? As for connecting the nanoAVR output the the PC's soundcard and amplifier in parallel, I wasn't able to find any plugs around the house to do this, so it was a direction connection between the nanoAVR and soundcard. Next time I go to the shops I'll get one and post the new results.

Hello,
1. To support any help, it would be kind to simplify your measurements and the results. E.g. 1 measurement -> 1 graph as a first step.
2. The problem(?) is an electrical, not an acoustical one. So do electrical measurements:
a) soundcard out -> cable -> soundcard in (loopback) to get the right levels (for maximum SNR/minimum THD) and an impression of the soundcard's capabilities
b) soundcard out -> cable -> amplifier in; amp out -> resistor divider (protection) -> soundcard in
c) nanoAVR HDA (flat response, all gains = 0dB) out -> cable -> soundcard in
Do not use white noise except you want to get a quick overview of the spectrum. Use log sine sweep (Farina sweep), as provided by many measurement programs as ARTA.
Read the program's manual or appropriate guides.
The peaks at 50Hz, 150Hz... are hum and show, that your "grounding" or cabling scheme seems to be wrong.
If you are satisfied with the electrical results you may start the acoustical measurements (equalising etc.).

Sorry for the delay - I had to order some parts to connect my soundcard inline with the nanoAVR/amp/speakers etc.

I've also spent a lot of time trying to calibrate the system to get accurate levels, as well as learning how to use spectographs to identify noise issues. I wont bore anyone reading this with the details, but suffice to say that the nanoAVR is indeed at fault in this case. It's actually the combination of a few issues.

Basically, the noise noise floor of the AVR does seem to follow the DAC's spec sheet of 95-103dB (i'll use 95dB to account for the real implemention as opposed to theoretical standalone chip performance). This was when measuring with a clean power supply. The nanoAVR feeds directly into the amplifier, giving 29dB of gain, with a max SPL of 105dB. This would give a noise floor of 10dB.
However I have also lowered the each speaker's gain in the nanoAVR by about 9dB to balance their levels and give DSP headroom. This makes the noise floor 19dB.
In addition, I have 5 speakers, which causes the noise energy to increase 5x, or about 7dB. Now the noise floor is 26dB
I live out in a rural area, and at night there is usually no traffic noise or wildlife/insect noises, meaning that my listening room's noise floor is very quiet (though I don't have sensitive enough equipment to measure the actual value). So I can hear a significant difference between no speaker noise and 26dB of white noise.

On top of all this, there is an issue with non-isolated HDMI and USB port earthing, causing ground loops when connected to certain devices. The level of noise is dependant on the attached devices, but in my case it gives a slight electrical noise about the same level as the white noise. So another +3dB of noise takes us to 29dB of noise.

Regarding the HDMI/USB port ground loops, the worst offender is my laptop's power supply causing electrical noise to jump to about 40-50dB. This occurs for example when connected via USB to perform DSP modifications, or HDMI to provide audio testing signals. However, this isn't a problem in my normal use, and doesn't bother me as much as the perpetual noise.

---

I'm not sure if the designers of the nanoAVR will read this, but i'd like to suggest some improvements for this product to alleviate the noise issue for most people. Note that I accept that I do not design electronics as a profession, so I don't understand cost implications, or the more technical reasons why some of these issues exist - My intention is to provide feedback on this product's flaws, and how they have affected me.

1. prevent the HDMI and USB ports from sharing a common ground with the analogue audio system, causing electrical noise on the RCA ports
2. Master volume control could be performed by an analogue pre-amp stage (after the DAC) rather than in the digital domain. This may be better done by a separate analogue-only unit, however I've been unable to find an 8-channel pre-amp with a master gain control. Perhaps minidsp could fill this niche?
   Obviously, this would lower the noise level when listening at low volumes, though would still be an issue in quiet scenes at high volumes
3. Give analogue audio-out specifications on the nanoAVR-HDA product page. If I saw 95dB SNR/dynamic range, I may have avoided this problem altogether.

I may end up returning this item, but i'll have to see what other options I have before I do so. I'm a little disappointed with the analogue stage of the nanoAVR, but the digital (DSP) part works brilliantly, so I may just use it for the DSP, and find another HDMI DAC.

Thank you for helping me Tiki - I really needed that push to check everything as accurately as I could. Much appreciated.

Hello zmaile
hello devteam,

some measurements may show the different noise floor of the naoAVR HDA's AK4440 and other DACs.



The RME Fireface UC shows the same noise level for open and shorted (Mic-) inputs, respectively.
The nanoAVR HDA shows approximately 10dB higher noise floor plus lots of (1kHz-?) 2kHz-harmonics.
They do not disappear at higher levels, but will be masked.



And of course this noise level can be heard unfortunately. With the levels set properly, this noise is annoying
through Hypex nCore 252MP and stacked speakers Seeburg GL16. The amplifier itself is really quiet.
The noise seems to be completely independent of any DSP-setting including all mutings.
There's some room for improvements...

Cheers!

Hello devteam,

it would be nice to get an explanation for the unexpected noise and harmonics of the nanoAVR HDA to be seen in these measurements.
Are these existent in the digital domain already (as a part of the I2S-Signal)? Or is it a distortion inside the analog domain, possibly caused by an external part?

A month ago I wrote an support request. (Case #54383)
Part of your automated response: "...Note that our typical response time is within 48-72h from Mon-Fri 9am-5pm (HK timezone)..."
April, 17th I sent a private message through this forum to you regarding this issue.
Again no response...

Thank you!
Cheers, Timo

Tiki,
When I was doing my noise analysis on the HDA, the noise seemed to be in the analogue domain. I read the I2S signal and it showed 0 when there was no sound, and only noise.

At the time I suspected it was the DAC itself, as it isn't a particularly low-noise DAC. However, I replaced the HDA under warranty (due to another fault), and the one I got back had a much lower noise floor (~5-10dB by ear, though I haven't measured it). As I haven't got the two to compare, I can't say where the noise is coming from, whether it be noisy voltage regulator, the DAC chip, a bad solder joint, or some other component. I'm fairly sure the circuit boards are all of the same version in both HDAs.

If you have an oscilloscope, you can check the I2S signal yourself. The I2S headers are the same as the rest of the miniDSP products. Even an analogue multimeter or oscilloscope will show 0 on the data lines when there is supposed to be no signal. Obviously reading the individual bits will require more specialised equipment though.