11 Jan
2014
11 Jan
'14
11:20 a.m.
On Fri, Jan 10, 2014 at 11:20 PM, Fons Adriaensen <fons(a)linuxaudio.org> wrote:
On Fri, Jan 10, 2014 at 10:48:15PM +0000, James Stone
wrote:
It means something... but what it means is another question.
Assume
* the noise is more or less white (if not: real trouble)
* the spectrum analyser uses an 1024 point FFT (typical),
* and a raised cosine window (also typical),
Then the full bandwidth (1/2 Fs) is covered by 512 filters,
and each of these has a bandwidth that is 1.5 times that of
a rectangular filter.
So each filter sees 1.5 / 512 of the total power.
10 * log10 (1.5 / 512) is -25.3 dB
So the noise level would be around -93 + 25 = -68 dB.
You could check this with jnoisemeter.
It's reported as -81dB on line-in with jnoisemeter.
The mic stage when turned right up (for my cheap SM57 copy) has
background noise of -65 dB (with mic still on, and quite a bit of
background noise, so not a great test), when the mic is switched off
(but still connected) it is -95dB. In practice, it doesn't seem that
noisy to my uneducated ears, but I guess others will have different
opinions.
The one thing I would say is that I am not that impressed with
realtime performance overall. It is really only stable down to about
128/2. If I use Mixbus, I have to use 256/2 on jack. According to
jack_iodelay this is around 27ms round trip. However, despite this, I
haven't really noticed any audible delay that makes recording
difficult. 128/3 seems better - around 15ms round trip according to
jack_iodelay.
With a realtime kernel - 3.6.4-bbq-rt-rt10, I can push jack right down
to 64/2 - (approx 10ms round trip according to jack_iodelay) but this
leads to xruns with minor stresses on the CPU.
James
Do you
mean the spectrum analyser showed a trace at -90 dB ?
Yep - the spectrum analyser showed a trace at around -93dB across all
bands. Is this meaningful?