A Moogle  🏳️‍🌈 💻 🎮 🎶 🇸🇪

That's kind of what I'm saying - the jailbars are only visible due to interference from the surrounding circuitry, but the interference only happens at a low enough frequency that it should be able to be notched out without affecting sharp chroma/luma transitions.

i.e. the jailbars are only "vertical" because they're temporally consistent across each line, just like anything would be that involves PPU data accesses on any given line. But they're still only at roughly 1/8 PCLK.

The only downside with the notch strategy, is that if something is intentionally displayed on-screen with literal jailbars, it will be attenuated harshly in the output signal, right?

Yes ! That's what I meant ! Am I being stupid or would that happen ?

This is pretty much where a comb filter fills in. Like the one in your television, its job is to separate (and in this case, discard) a low-frequency signal component (in your TV, the Y/C signal; in this scenario, 1/68 of a line +-10mV) from one or more high-frequency components.

The specialization here is that the attenuation it performs is selective to both frequency and amplitude of the waveform, which isn't sinus (it's more of a round square wave) but is periodic enough to select against.

But I still don't understand how the comb filter wouldn't remove any of the clean Y signal. I thought that comb filters only worked to separate Y and C because those two have "interleaved" harmonics.

Hmm. I'll rethink, it's entirely possible that maybe I'm misunderstanding something. The idea is to essentially chop down the jailbars in the scanline only if they're inphase with the pseudo-waveform that jailbars produce, as any intentional reproduction by the NES would not fit