Skip to main content

Topic: RMAA fails to pick-up audible hiss (Read 10681 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • 2Bdecided
  • [*][*][*][*][*]
  • Developer
RMAA fails to pick-up audible hiss
Reply #25
Your files show identical behaviour in both conditions.

I've tried the same thing with my clip+, stock firmware (V01.02.15A), and I get a clear difference (using my own silent files). I tried 16-bit digital silence, 16-bit dithered silence, and 16-bit noise-shaped dithered silence, all generated in Cool Edit Pro from 32-bit silence  .

I don't have a Y splitter, so could only listen via earphones or connect directly to the PC - hence the direct connection to the PC is not a proper test of the Clip+'s performance when feeding earphones.

Through earphones at full volume (an impossibly loud volume setting for listening to full-scale waveforms without destroying your ears) I can hear that the noise level from a silent (all zeros, no dither) file drops dramatically when I press pause. With pause pressed, the output is essentially silent, whereas when playing silence it is not. Putting a fixed DC offset in the file vs all zeros, or a very slow ramp - no difference - same level of hiss playing all three. I could not hear any difference between dithered and undithered. I can hear the attenuator in the Clip+ acting rapidly when pause is pressed, especially when pausing the DC offset - it's not a hard click - it's a buzz caused by the attenuator stepping rapidly through a few discrete settings (about 12), which gives a stepped wave on the output.

Recording into my PC, the 0dB FS signal replayed at full volume on the Clip+ registered -10.3dB via my audiophile 2496 hence I am not using its full resolution.

Recording at 24-bits, you can see (and hear, if you crank the volume or digitally boost the signal) the change in noise level when pause is pressed. The recording of the paused state is not quite so silent as through earphones, but it is equally as silent as recording with the device switched off and/or disconnected, so I think this is the limit of my 24-bit sound card. You cannot see or hear any difference between dithered and undithered source but a third "silent" file with just noise-shaped dithering has the noise-shaped high frequency part easily visible on the recording (though within the audible range it looks and sounds the same, so all benefit of noise shaping is lost).

Recording at 16-bits (actually at 24-bits and converting, so I could control the conversion), the paused parts are still audibly and measurably less noisy than the playing-silence parts, but the difference is less pronounced than with the 24-bit recording. With dithered conversion, the difference is detectable but small.

Occasionally, especially when activated near the end of a file, the pause function is not silent at all, but is just as noisy as when "paying" - this seems to be a little (usually inaudible) "bug" which is worth watching out for.

It seems my sound card, even with more than 1-bit of headroom lost due to too low a level, is more than sufficient to capture the noise performance of the Sansa clip+ when its playing, but 16-bits is insufficient to capture quite how silent it is when paused. 24-bits on this sound card may be sufficient, but it's impossible to make that call without doing the test properly (Y splitter, correct levels somehow?).

Hope this helps.


P.S. a test for another day would be a -90dB and then -100dB sinewave with noise shaped dither.
  • Last Edit: 03 May, 2013, 10:10:45 AM by 2Bdecided

  • stephan_g
  • [*][*][*][*]
RMAA fails to pick-up audible hiss
Reply #26
The amount of dynamic range required in measurements can be surprisingly high. As mentioned, there's the issue of differing reference levels, and then there still is the difference of a device's instantaneous dynamic range vs. its total dynamic range. The DAC inside a Clip+ may only give about 90 dB of instantaneous DR, but it's followed by a 40-odd dB PGA. Ultimately minimum noise level at low volume settings (or muted) will be the PGA's at minimum gain, plus whatever comes in via the virtual ground buffer (see AS3543 datasheet).

Even assuming pretty horrid voltage noise performance from these CMOS amps, minimum noise floor should be around or under 10 ┬ÁVrms (20 kHz BW), so with the typical maximum output of 800 mVrms, total DR should be around 100 dB or thereabouts. With low-noise amps and higher supply voltages, total DR can be 130+ dB, 140+ dB if you're really good. (Actually these are the very same factors limiting SNR in converters these days, and consumer DACs and ADCs usually are CMOS parts operating from +5V only. So even if you do get noise down, you won't be able to match a super low noise bipolar amp on +/- 15 V. You would have to run 8 channels in parallel - with externally combined output - at the very least, as in fact some parts can do.)

Having total DR > instantaneous DR is important since it allows shifting DR up and down as required. You won't need more than about 70 dB of instantaneous DR for audio playback, but headphones alone may show a variation in sensitivity of 40 dB and more (see e.g. 5 Pro vs. K240M). Total DR requirements are often misjudged. You'd be surprised by the amount potentially required even in a nominally "lo-fi" application such as shortwave reception. (There's nothing quite as annoying as broadband hiss when your actual signal stops at like 3 kHz...)
  • Last Edit: 05 May, 2013, 07:00:20 PM by stephan_g
My little "blogalike":

  • stv014
  • [*]
RMAA fails to pick-up audible hiss
Reply #27
Tests in this other thread suggest that the noise voltage of the Fuze+ only decreases by a few dB at low volume vs. full volume. This explains why there is audible hiss when using very sensitive IEMs. Although the EMU 0204 does add some noise to the recording, even at a low volume setting of the Fuze+ it only seems to make the overall noise level worse by 1-2 dB (considering also that the noise measurement of the EMU also included its DAC).