Lobby > Amp and DAC Measurements

FFT Process Gain in Spectrum (Distortion and FR) Measurements

(1/4) > >>

jkeny:
When looking at SNR from FFTs shouldn't the FFT gain be taken into account? In other words the noise floor seen on the FFT is the noise floor of the FFT bins NOT the noise floor of the device & the FFT process gain needs to be taken into account to calculate the noise floor of the device & thereby it's correct SNR.

Have a look at http://www.designnews.com/author.asp?doc_id=236273&dfpPParams=ht_13,industry_aero,industry_gov,industry_machinery,aid_236273&dfpLayout=blog where FFT gain is calculated as FFT gain = 10*log(M/2) where M equals the number of points processed in the FFT plot.

The 6384 pts used in these plots gives a FFT gain of about 39dB - thus a 120dB "noise floor" seen on an FFT plot is actually 81dB real noise floor

firev1:
For noise we(or I as a mere hobbyist) would mostly look at the noise in time and as a number which disregards FFT length. If I wanted a graphical representation I could use a PSD spectrum.

Over here we are not so much looking for noise but rather we are looking for low level distortion and spuraie so the high FFT lengths are to give us a clearer picture of these low level products.

jkeny:
Sure, I understand the use of longer FFT runs (more points) for raising the spurie from above the FFT "grass" but it's worth mentioning about the noise floor as it seems to be a common mistake made with FFTs - misinterpreting the "grass" as the device's noisefloor.

Solderdude:
Hi John,

Welcome to the forum...
It is customary in this forum to introduce yourself here:
http://www.changstar.com/index.php/topic,1784.0.html
Most will know you though... I reckon... as some use your products.

The noise floor in Merv's measurements is determined by the used soundcard and usage of a 1:10 probe.

You are correct though that the real noise floor is well above the 'seen' noise floor in the plots.
All the THD spikes from a complete audio signal spans along the whole frequency range lifting the actual noise floor to levels as determined by the highest harmonics (and jitter) and depending on the 'density' of the spikes the 'energy content' of those spikes will add like random noise adds.
In other words the actual noise floor could actually be raised about 20dB above that of the highest THD spike as found when -10dBFS to -20dBFS.

Still, I believe (based on observations of noise floors of actual recordings) that the noise floor in recordings usually still is well above the noise added by a decent DAC.

Based on atomic bobs measurements I estimate the real life noisefloor of the Yggy to be between -100dB and -110dB, still well below any recording I know.

jkeny:
Thanks for that, Ian - yes, didn't want to make too big a deal about the noise floor thing, just point it out.
Considering that the current model of hearing has the concept of filter banks (frequency bins) at it's core, it may well be that the pscychoacoustics of hearing actually are somewhat hinted at by the bins of an FFT than a standard noise floor measurement. Of course it's too simplistic an analogy but it does hint at some of the sort of dynamic variabilities that might be at play in the perception of hearing?

I will proceed to intro section & write something short

Navigation

[0] Message Index

[#] Next page