Yes it does. N+X aliases at N-X, if N-X is negative you can take its absolute value and repeat the process until it falls in your frequency range (negative frequencies are equivalent to positive ones). See: folding.
1 * 14915 hz does not alias
2 * 14915 aliases at 7778 hz ( 29830 = 7778 )
3 * 14915 aliases at 645 hz ( 44745 = 22695 = 645 )
etc.
The strength of these harmonics is actually dropping off quickly as they are attenuated by 1 divided by the harmonic number. I don't think harmonics after this will really show up on your graph (the resolution of your FFT and the window used will start to cover them in sidebands).
There are other spikes on your graph which are probably due to aliasing as well, but from a different part of the process. Any given peak could be from any number of frequencies above the nyquist frequency; there's a lot you can't really tell about this from an FFT. The current mixing algorithm jsr has implemented is not just calculating the value output from the chip at every sample, there are a few more steps in there; if you really want to know what's going on you need to analyze the algorithm itself.
Unless you were pointing out a flaw in the beta's rendering of that particular sound somehow, I don't see how it belongs in this thread at all.
Hardware recordings of things are welcome, but this is the beta thread. Put it in its own thread. Also, you completely neglected to tell us the source; didn't even tag the MP3 file. What use did you think it was here?
(I'm saying I am interested, but you've posted it in the wrong place and in the wrong manner.)
So, back to topic, I tried skimming the past few posts but since it was much a debate than anything else, TL;DR. The Nesdev wiki is also not clear about my question, so here I ask to those who observed closely the output of the chip.
So, if I understand correctly, the N163 doesn't sum up the channels, but instead output channels one at a time, much like what the YM2413 does? (illustrative example) So that's the switching noise that is much debated?
Seems to be the case. The YM2413 has the necessary equipment built-in to filter out this high frequency signal that is the result of clockig each channel one at a time, which Namco never bothered to add to the N163.
The YM2413 does it at a much higher frequency (~50kHz?); it's not strictly necessary to add a filter unless you have some need to remove the ultrasonic ripples from the signal. There's a LPF inherent in the NES' audio path, but I actually don't think there's an additional LPF on the VRC7 (from the recordings I made, it still produces the full range of audible frequencies).
The problem with the N163 isn't really the lack of a filter; it's the frequency. It's low enough to be audible.
Also, Jarhmander, I'm planning to give the N163 page on the NESDev wiki a good going over once I have some time to do a few more tests.
I have a horrible headache now because of the channel switching noise, so please add a switch/checkbox for turning it off, and add a warning that this mode is inaccurate to the real hardware. I don't want a headache every time I use the N163 (Seriously, the noise does give me headaches).
.gif){kind=avatar}
