RMAA fails to pick-up audible hiss, reports nearly perfect noise level Options

[skamp]

We've been scratching our heads about this on Anything but iPod and in #rockbox on Freenode. With sensitive IEMs (like my Shure SE425s), there's a faint, but audible, hiss with the Rockboxed Fuze+, when no music is playing. The hiss is a lot more pronounced when playing a digitally silent file. I don't hear hiss with the iPod Classic. At least 2 other people have made the same observation.

The problem is, RMAA can't seem to pick it up. The Fuze+ gets exactly the same noise and dynamic range scores as the Classic, unloaded as well as loaded with the Shures and the Denons: 95dB, a nearly perfect score for a 16 bit device. More to the point, while the Fuze+ and the Classic are audibly different (with regards to hiss), why do they get identical scores in all situations?

You could slap me with the TOS #8 hammer, but I don't know how we can prove what we hear. The difference is easily heard when turning the Fuze+ on and off. And since we can't determine the nature of the noise, or its cause, the Rockbox developer (pamaury) doesn't really know what to do about it, or if it's even fixable at all.

I'm no electrical engineer. Is there something special about what we perceive as "hiss", that would be audible, but that wouldn't show up in RMAA-type tests / recordings? There's no way I could hear noise if it actually was at -95dB.

I have no idea if this is relevant, but FWIW, the Fuze+ has an output impedance of 0.9Ω, and the iPod Classic, 5.5Ω. The Shures are 22Ω / 109 dB/mW, and the Denons are 25Ω / 100-106 dB/mW (depending on whom you ask).

Edit: I forgot that I already mentionned that problem, but it's bugging me, and I guess it doesn't hurt to start a dedicated thread.

This post has been edited by skamp: May 1 2013, 10:55

[bennetng]

RMAA always add dither in its test signal so you are not really testing digital silence.
Expand the [+] Noise level tab and you can see "Dithering signal" is always checked and can't be unchecked.

I've modified the 16-bit 44k test signal so that it really contain digital silence (generated by RMAA version 6.2.3 free)
http://s000.tinyupload.com/index.php?file_...470431937870794

This post has been edited by bennetng: May 1 2013, 11:15

[skamp]

You must have done something wrong, your sample lasts 26s while the original lasts 71s. Also, don't compress with 7zip, just use FLAC.

This post has been edited by skamp: May 1 2013, 11:26

[Brazil2]

Maybe give Visual Analyser a try:
http://www.sillanumsoft.org/intoduction.htm
http://www.sillanumsoft.org/prod01.htm

[bennetng]

You must have done something wrong. Your sample is too short and doesn't look like the original signal at all.

Audacity screenshot of the original: http://outpost.fr/stuff/20130501122056.png
Audacity screenshot of your sample: http://outpost.fr/stuff/20130501122019.png

My test signal did not contain some test so it is shorter, including:
Frequency response (swept sine)
THD (set of tones)
Impulse/Phase response

Other than these there is nothing wrong, except I modified the "Noise level" test to remove the dither. You can safely use my signal and generate a nice looking html report with it.

[skamp]

I don't think that's working properly. RMAA reports a noise level of -18.6dB with both my Fuze+ and my O2/ODAC (which is not possible, it's dead silent).

[bennetng]

Try to uncheck the following in RMAA before testing. Those tests won't show in the html report anyway.

Frequency response (swept sine)
THD (set of tones)
Impulse/Phase response

If it still doesn't work I will upload a full length test file again.

[skamp]

I did that. RMAA wouldn't accept the file if I didn't.

[bennetng]

Full test file
http://s000.tinyupload.com/index.php?file_...990245098663944

[skamp]

Thanks. Dynamic Range is still the same (95dB), while the noise level is even lower: -98.6dB.

[bennetng]

Do you mean you can still hear the hiss but the measurement is -98.6dB? How did you perform the load test? Is it similar to this?

http://www.xbitlabs.com/articles/multimedi...-kamabay_5.html

[skamp]

Do you mean you can still hear the hiss but the measurement is -98.6dB? How did you perform the load test?

As always, with the Shures attached to a stereo splitter, which is plugged into the headphone out of the Fuze+ and connected to the line-in of my ADC.

[skamp]

Also note that RMAA does pick up noise, when it's there. I used a fairly bad ADC when I first started doing RMAA tests, and the noise level was significantly high.

[2Bdecided]

Could the maximum signal output vs the background hiss be a greater range than 16-bits (96dB)? Is your ADC capable of capturing this range?

FWIW there's a feint but audible hiss on the Sansa Clip+ with factory software using sensitive earphones. I couldn't cope with listening to a loud recording at full volume through the same equipment. Never measured the dynamic range though - don't think I have the equipment to do it.

Cheers,
David.

[skamp]

Could the maximum signal output vs the background hiss be a greater range than 16-bits (96dB)? Is your ADC capable of capturing this range?

I'm not sure what you mean. Surely if the difference was greater than 96dB, I wouldn't be able to hear it. FYI, my ADC (EMU 0204 USB) only records in 24 bit, and the maximum DR that I ever measured with it, is its own: 112.7dB.

[2Bdecided]

Surely if the difference was greater than 96dB

That depends on how loud "full scale" at "full volume" is with your IEMs.

Cheers,
David.

[bennetng]

In RMAA's test signal, 2.5s to 6.5s is the noise level test. Save the recorded file and listen to it with your 0202. If you can hear nothing then it simply means your test method cannot reveal the hiss.

I have no confidence to test headphone amps with RMAA as well, I only trust line level tests.

[saratoga]

My guess is that you don't have enough dynamic range on your ADC to detect it. The full scale voltage on your ADC could easily correspond to an extremely high level on your headphones. If the noise is independent of the volume level, it may be below the noise floor of the ADC.

For example if the noise is -100 dbfs and your headphones are 125 db per volt, and your ADC had 100 db SNR, you will not detect it with the ADC but will easily hear it with a silence file. This is what happened with some of the clip players for example.

[skamp]

My guess is that you don't have enough dynamic range on your ADC to detect it. […] if your ADC had 100 db SNR […]

112.7dB.

This post has been edited by skamp: May 1 2013, 18:05

[saratoga]

My guess is that you don't have enough dynamic range on your ADC to detect it. […] if your ADC had 100 db SNR […]

112.7dB.

Is that enough to record the noise?

[skamp]

Is that enough to record the noise?

Yes, compared to a 24 bit silent file, with the volume cranked up ridiculously high.

[2Bdecided]

So if you just record using your equipment the noise is there in the resulting wav file, but RMAA doesn't see it?

[skamp]

I find this all very confusing, so instead of drawing the wrong conclusions, I uploaded two recordings of my Fuze+: one with nothing playing (just the Fuze+ powered on), and one with digital silence playing.

Note that previously, I had generated my 16 bit silent file with Audacity. As I suspected, it does something more than just generate silence (it adds some very audible noise). This time I generated my 16 bit silent file with shngen from shntool. With the new silent file, I don't hear a difference between playing it, and playing nothing.

EDIT: I forgot to mention that both recordings were made loaded with my Shure IEMs.

This post has been edited by skamp: May 3 2013, 11:30

[bennetng]

Note that previously, I had generated my 16 bit silent file with Audacity. As I suspected, it does something more than just generate silence (it adds some very audible noise).

The noise is probably dither. You should be able to disable that in Audacity's preferences. 16-bit dither without noise shaping is around -96dB. It should be inaudible unless you turn up the volume to an uncomfortable level. Also, digital silence means nothing, 8-bit digital silence is identical to 16-bit and 24-bit digital silence.

[Arnold B. Kruege...]

My guess is that you don't have enough dynamic range on your ADC to detect it. […] if your ADC had 100 db SNR […]

112.7dB.

Is that enough to record the noise?

One relevant parameter is what 0 dB means in terms of real world voltages.

An audio interface designed for pro audio generally puts out 4-6 volts with digital FS.

Digital Fs on a portable digital player would probably be more like 1 volt.

There is 12 dB difference between 1 volt and 4 volts.

For the purpose of measuring portable digital players, the Emu 0204 may have only about 100 dB dynamic range, and that puts it well within 10 dB of the UUT, which is starting to be too close for good accuracy.