SB3 Spectral analysis shows +9dB at 30Hz on phono and coax outputs

[rbl]

I connected the various SB3 outputs to my PC via a Tascam US-122L USB audio interface. The PC acted as a spectral analyser, the relevant results are in the image below (each Y division = 3dB). What I do not understand is why the bass response increases by 9dB at 20-30Hz on the phono and coax outputs, but not the 3.5mm output. Points worth noting:

- Same problem with two different SB3s

- Same problem with two different pink noise files: one was from Inguz (44KHz / 16bit flac), and the other from www.24bit96.com

- You can see the problem creeping in at about 500Hz

- The coax was tested via a Rotel RSP-1068 surround pre-processor.

I don’t understand this and I can’t help but think I am doing something truly stupid. I would have thought the problem would be audible as I have a pretty decent system, but I have never noticed it although I may well have unwittingly “tuned it out” as I normally apply a cross over at 60Hz for a subwoofer (no filters for this test).

Any ideas anyone? I am running LMS 7.9.2 on a Raspberry Pi4 (it’s great!).

[cliveb]

I don't know how to interpret the screen shots you've included.
What are the two plots? Is the top plot the waveform and the bottom plot the spectral analysis?
What is the difference between the red and green traces?
If the bottom plot is a frequency analysis, and you're using pink noise, then why is it flat above about 200Hz?
Pink noise is *supposed* to have a falling spectral content with rising frequency, eg. this digitally generated pink noise in CoolEdit2000:



As for why the phono & coax outputs are different to the 3.5mm headphone output, my guess is that it's the headphone output that is "wrong", since it passes through an additional headphone amp which is probably not of the highest quality. The coax output in particular should not be any different to the data in the source file, because when using a digital output on a squeezebox all it's doing is converting the TCP/IP data stream into SPDIF without any processing.

[rbl]

Apologies for not being clear in the OP and for the late response. I had to find time to do more tests.

The upper half of the screen shot is the waveform and can be ignored. The lower half is the spectral analysis. Green is the SB3's headphone output, and red is the phono output. My understanding has always been that pink noise is equal noise energy per octave, and white noise is equal energy per Hz, and that pink noise is what is usually used for audio testing.

Anyway, I did some more tests. It turns out that if I use white noise (top 2 screen shots), and a linear X axis on the spectral analyser then I get a flat response. If I use "Octave" X axis then I get the rising response although it rises only above 500Hz!??

If I use pink noise (bottom 2 screen shots) and a linear axis then I get the decreasing response as expected. However pink noise and the "Octave" X-axis gives flat down to 500Hz (as expected) but then rising below 500Hz. This of course is what I observed with the SB3 and I still don't get why that is. I'd put it down to a dodgy pink noise file, except that I tried two pink noise files obtained from entirely different places!

Well, I guess given the white noise results I have no reason there is anything to worry about the SB3, except perhaps that the headphone output is not great at lower frequencies, but I am not worried about that.

[Jeff07971]

Glad you found the error, it would have been an extremely weird problem if not !

Jeff

[mrw]

However pink noise and the "Octave" X-axis gives flat down to 500Hz (as expected) but then rising below 500Hz.

Perhaps your sampling frequency steps are too large to accurately capture energy per octave at the low end. Also seems to show up with your White noise plotted against octave.

[rbl]

Perhaps your sampling frequency steps are too large to accurately capture energy per octave at the low end. Also seems to show up with your White noise plotted against octave.

Good point!

The FFT size and sampling frequency were set to 4k and 44.1kHz respectively (spatial resolution = 10.7Hz). If I increase the size to 64k, which corresponds to a spatial resolution of 0.67Hz, then I get a nice flat response with pink noise (see below). Even with a FFT size of 32k I still get some uplift below about 50Hz.

TBH I am a little surprised that the effect is so pronounced and am not quite sure exactly what is going on, esp as I am doing a lot of averaging!

[mrw]

TBH I am a little surprised that the effect is so pronounced and am not quite sure exactly what is going on, esp as I am doing a lot of averaging!

I can’t help you there, I have no practical experience. It looks as if your process brings out some kind of low frequency end effect.