It's nothing more than the sum of the power noise of two random and uncorrelated signals, divided by the amount of signals = average.
That's is not smoothing, that is most definitely averaging the noise out, since the noise is totally random to begin with.
Smoothing would mean that any correlated signals will also be averaged out, which they actually don't.
In fact, if they are strongly correlated they sometimes even get more obvious after averaging.
But again, read your books.
They get averaged too is all. So if there is any variation in their amplitude from each FFT pass what we see after averaging is the average value of the correlated signal. IOW, we are are low pass filtering any "noise" in the correlated signal amplitude.
However, there is an average value of noise Magnitude in the noise floor too, and because most spectral analysis is void of phase information there will be no cancelation. For that you need phase of noise too, not just amplitude.
Use of the word smoothing was in its common language meaning.
Exactly. If we want to reduce RMS noise (and keep all signal-related stuff), we need to apply time-domain averaging first (which then requires sample-synced recording).
Noise floor only really has a meaning in the context of effective bandwidth. Higher resolution with a larger FFT chops the bins into smaller pieces with less noise in each but they still add up to the same number for the same bandwidth. The higher resolution enables "seeing" deeper into the noise. Coherent averaging, however, will enable removing noise pretty dramatically revealing stationary signals.
I don't know a lot about the Mac audio subsystem but I can say that Windows will enable sample rate conversion and different bit depth if you don't check or bypass every opportunity for Windows to mess with your audio. if you can use ASIO it's usually better. However even that can run into issues. This is one virtue of the dedicated systems with software (AP, Quantech etc.); those system problems are largely locked out.
I don't know a lot about the Mac audio subsystem but I can say that Windows will enable sample rate conversion and different bit depth if you don't check or bypass every opportunity for Windows to mess with your audio. if you can use ASIO it's usually better. However even that can run into issues. This is one virtue of the dedicated systems with software (AP, Quantech etc.); those system problems are largely locked out.
The Java based drivers supplied with REW seem to be really good. The macOS one was already good, but a very smart fellow wrote custom drivers for Windows and Linux use in REW. These allow for exclusive access to the devices, too. I haven't found a single case where the Java solution isn't superior to ASIO with REW. At least with recent versions of REW. But, that's just me.
Anyway, I tried running my own Cosmos ADCiso with a MacBook Air M1 just like SylphAudio's and had no problem with the noise floor. (Posted above.) I tried using the same settings as he (she?) did in REW, too. It took me longer to find all the right cables than it did to get it all up and running for the first time.
I suspect that SylphAudio has a bad connection, a subtle setting wrong, or something like that. That's easy to do, and I'm not being critical here. I certainly have experienced all of those.
Anyway, I tried running my own Cosmos ADCiso with a MacBook Air M1 just like SylphAudio's and had no problem with the noise floor. (Posted above.) I tried using the same settings as he (she?) did in REW, too. It took me longer to find all the right cables than it did to get it all up and running for the first time.
I suspect that SylphAudio has a bad connection, a subtle setting wrong, or something like that. That's easy to do, and I'm not being critical here. I certainly have experienced all of those.
What do you mean by sample-synched recording? In synch with the digital source? Or preseving the phase? What good would either do here?
Is your Mac version Sonoma 14.4.1? Looks like REW can't get exclusive control over the ADCiso since the system treats the ADCiso as a microphone input (and it's applying noise cancellation). I already updated to the latest non-beta REW (5.31.1).
REW only displays the RTA graph from ADCiso when there's an input signal. If there's no input signal, then the RTA display is totally blank. (where it should display the ADCiso's noise floor). Like some sort of a noise cancellation.
On windows, it can display the noise floor. This is not possible on my macbook.
Already unselected the ADCiso(record interface) as the system sound's microphone input, but still REW doesn't display the graph when there's no signal input feeding the ADCiso.
If I remove REW on microphone security and privacy settings, then REW cannot totally access the ADCiso, it can be selected as the input on REWs preferences, but the RTA graph won't display anything. (Even if there's a signal feeding the ADCiso) Obviously this access is required.
Last edited:
Yes, I also noted something strange when using my Cosmos ADC on my M3 MBP. There is sometime fishy going on with the mic setting and I got interruptions every second or so in the recording.... I didn't use REW but recorded sound to file.
So it isn't only REW.. someting has been changed and it must be MacOS.
//
So it isn't only REW.. someting has been changed and it must be MacOS.
//
I was also surprise to see the Mic thing in the top bar... perhaps time for a trouble report to Apple...
https://developer.apple.com/bug-reporting/
Did you check the MIDI Audio setting?
//
https://developer.apple.com/bug-reporting/
Did you check the MIDI Audio setting?
//
SylphAudio, I have no mac os PC and can't help you but so far nobody complained about Coasmos ADC+Mac performance being limited by 16bits or so.
PS: try to ask that in our discord channel, perhaps Mac users may help.
PS: try to ask that in our discord channel, perhaps Mac users may help.
The former, the recording must be in sync with the digital source, if there is one, that is. At an integer multiple or fraction of the sample rate.
Use cases there are many, basically every time you want to look way down into the noisefloor, especially in the time-domain.
Two practical examples:
https://www.audiosciencereview.com/...ve-sample-values-are-always-offset-by-1.6865/
https://www.audiosciencereview.com/...-hump-revisited-khadas-tone-board-v1-3.30136/