Noise Floor seems too high

[Matthew J Poes]

I recall John mentioning that the FFT method allowed a reduction in the noise floor from the room's actual noise floor or that of the mic. The new beta version of REW gives the noise floor, which I'm attaching. This was taken with a Dayton USB Measurement mic, which is known to have a noise floor right around 35-40db. My room is a soundproof space which measures below the noise floor of every SPL meter I've tried. A quieter mic I own has measured a noise floor at its limit of about 25db. Is the noise floor indicated in the distortion plots the rooms actual noise floor and not the processed signals noise floor (I think I may have misunderstood John on this)? Would this be the lower limit to use for CSD plots (any information below this would just be noise)? Since the mic has a known high noise floor, I assume that a quieter mic would improve this, correct? The measurement was about 73db's on average.

Anything concerning or worth interpreting in this data that may imply a noise contamination problem in the signal path? Like would the spike at 120hz just be electrical noise? The 90hz? Seems to perfectly lined up with 90hz. What about the rise at 2-3khz?

 

[John Mulcahy]

The noise floor in the distortion plots is obtained from the spectrum of the signal captured before the sweep is output, it combines any acoustic and electronic noise and includes the process gain of the FFT length used. Better not using a normalised view when looking at noise floor since the level of the fundamental will be contributing boosts wherever the fundamental dips, there is some discussion of that in the help. USB mics tend to be limited by the noise floor of the mic's front end electronics. In a region with 60 Hz mains it would be common to see spikes at multiples of 60 Hz, and with USB mics there can also be spikes at multiples of the 1 kHz USB frame rate, depending on how clean the USB power is and the mic's supply filtering and PSRR.

 

[Matthew J Poes]

Thanks John, that was done with a 1M length measurement.

I just grabbed an image when I did the normalised distortion plot, and never bothered to change it. What you say makes perfect sense.

I can try this again with an MOTU preamp and Earthworks M23 to see if I get a lower number.

I have a Class 2 mic and NIOSH meter app on my iPhone which they actually found to meet class 1 specs. I have a handful of poor quality calibrated SPL meters that all can't measure below about 30db or so. All of these are at the noise floor. I took a measurement once with that iPhone mic through REW using 1/3 octave bins in the RTA app and the final plot was hovering around 20-25db other than some spikes in the low frequencies and a ride toward 20khz. That measurement was calibrated with a 94db calibrator.

I was a little surprised to see this noise floor measurement be so much higher unless it was including the mic/system noise.

Is the RTA approach the same or better for measuring noise floor as you would for NC (noise criteria) style?

 

[jtalden]

Reading this, I Just looked at one of my latest measurements out of curiosity. I posted it below for you as a point of comparison in case it is helpful.

This is a 256k sweep with 'normalize' unchecked as John suggested. The measurement was done with a Beyerdynamic MM1 mic and a Focusrite 2i2 interface and with the HVAC system off. When the HVAC is on the bass range noise is significantly higher; maybe 10dB. The basement AV area here is not sound treated.

I expect your results with the better mic and 'normalize' off will be significantly better than this at least in the difficult bass range.

 

[Matthew J Poes]

Hey John I went back to create non-normalized versions and noticed something a little odd. The Measurement I took without Dirac on has a noise floor that seems to be about 10db's lower in the midrange. Any thoughts on what would cause that?

Without Dirac


With Dirac


Noise Floor with quieter mic/RTA method, This was with no door knobs and HVAC and Whole House Humidifier on, that seems to be the LF noise.

 

[Matthew J Poes]

Reading this, I Just looked at one of my latest measurements out of curiosity. I posted it below for you as a point of comparison in case it is helpful.

This is a 256k sweep with 'normalize' unchecked as John suggested. The measurement was done with a Beyerdynamic MM1 mic and a Focusrite 2i2 interface and with the HVAC system off. When the HVAC is on the bass range noise is significantly higher; maybe 10dB. The basement AV area here is not sound treated.

I expect your results with the better mic and 'normalize' off will be significantly better than this at least in the difficult bass range.
View attachment 4943

Thank you very much this is very helpful. I wonder how much of that is your room noise floor and how much is electronic noise floor. I'm fairly certain the rise in high frequency noise in my measurements is purely electronic noise.

I'm looking to pick up a used EW M30, which should help.

 

[jtalden]

I would love to own an M30 and Moto Interface. Let us know if it helps your measured the noise floor significantly.

 

[Matthew J Poes]

I would love to own an M30 and Moto Interface. Let us know if it helps your measured the noise floor significantly.

A friend of mine has the MOTU and an M23. For the same reason that he owns that, I need to get a proper mic myself. The M30's are available used at a good discount compared to the M23. Plus I can measure to 30khz which might have use? I'd rather not have to buy a MOTU, but to do burst measurements you need a very low latency USB sound card. We don't know why, but we do know that at least some inexpensive interfaces have issues with this. We assume its latency, but if nothing else, we know its not a problem with MOTU. If I could get away with Behringer or focusrite I would.

I'll report back as soon as I get some new measurements.

 

[John Mulcahy]

Hey John I went back to create non-normalized versions and noticed something a little odd. The Measurement I took without Dirac on has a noise floor that seems to be about 10db's lower in the midrange. Any thoughts on what would cause that?

Looks more like a 4 dB shift than 10 dB. I imagine Dirac may need some headroom for its processing and so may reduce gain to allow that, which would correspondingly reduce dynamic range.

 

[Matthew J Poes]

How about 6.5? That's the DB headroom set in the software. I finally found that feature. That explained it. When I changed the setting it changed the noise floor.