Topic: PCM-M10 noise floor

[wipeman]

I posted this in the PCM-M10 thread but it has been swamped by the "clock issue" discussion so I thought I'd try again in a new thread. In short, I'm wondering whether it is ever worth recording in 24-bit resolution on the PCM-M10. Even at zero and very low gain values I never achieve a noise floor better than -93dBFS when recording from external mic. I am using a battery mic with the power switched off to give a "silent" input to test the noise of the PCM-M10 itself.

With 16-bit quantisation introducing (IIRC) -96dB level of noise to the recording, what would 24-bit recording ever achieve if I never record better than -93dB?

NB I noticed this value when I was trying to measure the EIN of the device. I can assume therefore that the EIN for the PCM-M10 at zero gain is -93dB, though I had expected better.

Have I missed something?

[wipeman]

Thanks for your answer mshilarious.

So I should therefore get some small benefit from 24-bit as the usual dynamic range. I hadn't taken account of dithering (didn't know this happened) or any frequency-dependent noise performance. This is what I hadn't considered.

I had wondered whether the 24-bit was merely due to "feature wars" with other recorders and had little practical use.
Thanks again.

[pnoman]

With 16-bit quantisation introducing (IIRC) -96dB level of noise to the recording, what would 24-bit recording ever achieve if I never record better than -93dB?

First, I have not had my hands on the PCM-M10, but I have done a test of my Roland R-05, which I think is very similar (if not identical) to the Sony. The test is here: http://www.b4net.dk/?page_id=111 if you want to have a look. Might give you also some ideas of how to measure and correlate the various metrics.

Anyway, I also found a converter noise floor around -93 dBFS in 16-bit mode. If you go to 24-bit mode, the converter noise floor drops to around 98 dBFS, so around 5-6 dB improvement is what you get. Note that a signal that would register at the same level would of course be very noisy, but you would still be able to hear it... and you get "free dithering" at this level ;-)

I use always 24 bit recording, unless I would need to absolutely reduce the size of the files. It is maybe still just small improvement, but for a signal at, say -60 dBFS, a 24 bit recording will still give you more effective resolution (around 1 bit extra when compared to the noise) and maybe slightly less distortion also. But of course it is better to record at slightly higher levels.

NB I noticed this value when I was trying to measure the EIN of the device. I can assume therefore that the EIN for the PCM-M10 at zero gain is -93dB, though I had expected better.

No, you need to take into account the actual amplification in the preamp i.e. know the absolute input sensitivity. If, say, your nominal 0 dBFS level is at -7 dBu at the default settings, and you have additional attenuation in the preamp of, say, 10 dB for your measurement, then your EIN would be -93dB - 7dB - 10dB, i.e. -110 dBu. You cannot assume the specified sensitivity, but have to measure that also yourself.

In most cases, when you measure -93dBFS noise level (16 bit), it is the converter noise you see.

The best EIN will usually be achieved at max preamp gain, and you may rather measure -e.g. a 73dBFS noise floor when using the microphone input at +20 dB max attenuation on top of the "default" 0 dBFS level of -27 dBu, which means the EIN is around -120dBu, or some such.


-- Per.

[wipeman]

Just to confirm, I'm seeing noise floor in the absence of any input signal of -93 dB when recording at 24/96 resolution. I'm not recording at 16-bit.
My point was that I kind of expected better than this noise floor at 24-bit resolution.
Will take a look at your site. Thanks for the link.

[pnoman]

Just to confirm, I'm seeing noise floor in the absence of any input signal of -93 dB when recording at 24/96 resolution. I'm not recording at 16-bit.

Sorry if I misunderstood you then.

The noise floor you see in the resulting recording is a function of the noise in the "preamp" stages and the "converter" stages (just to simplify a bit) -- often somewhat like a geometric sum of the individual components.

When you detect the noise floor at -93 dBFS, what are your gain settings then? If the Sony is a modern "balanced" design, as I believe it is, then at minimal gain, you should see the converter noise being dominant, and at max gain, the preamp noise should be dominant.

On the Roland R-05, there are 80 gain settings from 1 to 80 (and an "off" setting at 0). When I apply a signal to any of the inputs and set the gain to 1, I measure around -93dBFS(A) in 16-bit mode and -98dB(A) in 24-bit mode, if I remember right. At gain settings 80, the noise floor grows to -88 dBFS(A) on the line-in and around -82dBFS(A) and -73dBFS(A) on the Mic Lo and Hi input(s). I would expect the Sony to be similar, but am very interested to hear what you find out -- but remember the actual attenuation has a big impact on the noise floor. There is a hidden setting that can increse the gain on the R-05 to the same level as I think the Sony can go to. At least the EIN for the Sony and the Roland look almost identical, as far as I can tell.

Try to adjust your gain setting to min (not off, though) and apply a "non-signal" input like you did, and measure the noise level of your recording. Try this on both the line-in and the mic inputs, and in both 16-bit and 24-bit modes. Try also with the gain set to max -- you should be able to detect a difference between the line-in and mic in at various sensitivity settings.

Look forward to hear your findings,


-- Per.