FFT Analysis Settings.

Hardware: custom made PC with onboard VIA AC’97 sound card
Software: Audacity ® 1.3.12-beta (Unicode), Spectrum Lab V2.75 b09.


I. Audacity settings I used before in the Round One testing:

 

 

 

 

Fig.1 No input signal on sound card (noise floor). Spectrogram by Audacity 

 

 

 

To insure that Audacity does a good job analyzing spectrum, I tried another program, Spectrum Lab V2.75 b09.

 

First run was done on the direct input signal, doing real time analysis. Spectrum, updated each second, had changed a little bit from time to time, specifically in the region below 100 Hz, but still was usable to make a decision on settings.    


II. Settings for Spectrum Lab:

 

1. Dependency on the FFT input size settings (taken with sample rate 11.025 kHz):

 

 

Fig. 2. FFT input size 8192, sample rate 11.025 kHz

 

 

 

Fig. 3. FFT input size 16387, sample rate 11.025 kHz

 

 

 

Fig. 4. FFT input size 65536, sample rate 11.025 kHz

 

 

 

2. Sound card sample rate dependency:

 

All above spectra are for the sample rate 11025 Hz: 

 

 

When changed to 22 kHz

 

 

Fig. 5. FFT input size 65536, sample rate 22.050 kHz

 

 

There is a well prononced tendency to show a higher weight of lower frequency components when using a higher sample rate.

 

3. Resolution (Bits per second) setting. All above results were obtained with 16 bits/sample. When I changed bits/sample to 24, spectrum just disappeared from the screen, so I've decided to keep 16 bits for the further tests.

 

4. FFT Spectrum settings:

 

 

 

 

To further test signal processing options I’ve compared results using the noise floor signal captured over longer time. I’ve recorded it using Spectrum Lab’s “save input in a file” option, and then used the option “Analyze audio file without DSP”.

 

Fig. 6. FFT input size 16384, sample rate 11.025 kHz, from recorded file

 

 

Fig. 7. FFT input size 65536, sample rate 11.025 kHz, from recorded file

 

 

From comparison of those two graphs, I would conclude that the algorithm used in the FFT analyzer is very sensitive to the FFT input size number specially in the region below 10 Hz, and that peculiar behavior of the spectrum produced by Audacity shown in the Fig. 1 in the area 0 – 6 Hz is just a consequence of the low FFT input size equal to 16384. (The same tendency shown on the Fig. 3 and Fig. 4)


III. Conclusion:

 

So, for the turntable tests Round Two I’ve decided to use Spectrum Lab V2.75 b09 program, with following settings:

 

Sound card (recording sample) parameters: Sample rate 11K, bit-depth 16
FFT processing: FFT input size 65536.
(Click on resulting spectrum Fig. 7 to see more details)