So... I am recently working independently on the problem of reproducing NES sound on the SNES, and I seem to find that the SNES has a very strange highpass filter effect on its hardware noise?
On the left is a spectrum plot of Famitracker playing high NES noise frequencies. More or less flat response, as expected. On the right is from Mesen playing SNES hardware noise at its highest frequency (FLG low bits = $1F), and... it's very dramatically not flat? This also matched a recording of the same ROM on my SNES, so I believe the emulation is accurate:
This is not explained by it outputting the whole LFSR instead of just 1 bit. The effect of that
should just be the mild lowpass that was mentioned earlier in this thread. (I also tried some empirical testing with an ideal waveform just to make sure there wasn't some aspect of this we hadn't thought of.)
So... has anyone else seen this? If you make the SNES output plain hardware noise at $1F, is this what you get? Does it have this very strong highpass coloration? Or... am I somehow doing something horribly wrong? Unfortunately I don't have a ROM I can share at the moment, but an independent test would also help to rule out something strange in my implementation.
(EVOL=0,0 EON=0 FLG echo disable bit is set, so I don't think it can be echo... PMON=0... am I missing something that can do this?)
Emulators do seem to be getting this correct vs. the same ROM tested on hardware, but I have no idea where this effect comes from. Couldn't spot anything relevant in the source code. Noise does not seem to be subject to the BRR filter at all...