In the project-description I try to explain the non-discrete nature of magnetic media, and do-so quite poorly. Here's a simpler example: 

A 74595(?) shift-register can take in a serial stream of bits, and convert that serial stream into 8 parallel output bits. There are 8 discrete positions for a bit to be stored. 

Similar to any digital-memory, like SRAM, etc. Each bit is discretely positioned on the output of some 1-bit-wide logic circuit, e.g. a Flip-flop (SRAM), or a capacitor tied to a logic gate (DRAM). Again, that same circuit is repeated for each and every bit.

In the case of a shift-register, we essentially have an 8-bit SRAM whose input is serial and output is parallel.

So, say you load the serial bitstream 01011001 into the shift-register's input, then the 8 discrete outputs will each show one of those 8 bits.

But now, what when you change that input bitstream to have a tiny pulse of '1' between the two zeros? It's not like, suddenly, a new output circuit/pin will be inserted between bits 6 and 7 to show that '1'.

This may seem ridiculous to harp on, but the fact is that magnetic media, signals sent down wires, even RF and optical systems are more-often-than-not treated as discrete-time systems just like that serial bitstream fed into that shift-register.

That extra '1' bit, discussed earlier, is considered a glitch in most storage and communication systems. Similarly, any value-change on that data stream at any time *other* than those discrete bit positions... (TODO: timing diagram). These, too, are considered glitches in most current systems (aka jitter).

Frankly, I think that's a tremendous setback to our technological progress (maybe that's a good thing?).

Similar to that 'glitch', even more similarly to that jitter, is the PWM-system I've designed, here... Feeding that *directly* into a shift-register seems a bit ridiculous... It won't work *at all* with the original MFM or RLL system, its shift-registers nor timing constraints. But it could work with the same (slow) technology available at the time, and so much easier with that available today.

Continuing with its seeming ridiculosity: For it to work properly fed into a discrete-time system like a shift-register (similar to the MFM design of the era), the PWM-Nibble--5us long, with 16 values stored in the middle 1us--would require a shift-register at least 48 bits long (more like 72!). And that shift-register would have to be 16 times faster than those in the original MFM/RLL system. Further, Only 16 of those 48 shift-register outputs would contain data. And that data, when packed in binary, would easily be represented by 4 shift-register outputs.

Speed-wise, as described in the project description, this system *could* have been achieved with the slower technology of the era... Using similar techniques as a 100MS/s oscilloscope with 50MHz ADCs, alternately taking samples. Or, much easier, by using timer-counters to measure the width of the pulse... As I recall, even the original 74163(?) could run at 16MHz. But, either way, there would be quite a bit more logic involved, and that wasn't cheap in an era where a Floppy Drive cost $200.

So, absolutely, this method seems *completely* ridiculous as a means of storing or sending binary data.

Except that magnetic media, fiber-optics, RF, etc. are *NOT* discrete-time systems, despite regularly being treated that way.

So, now we exist in an era where FAST DSPs exist, an era where FAST shift-registers can be packed in by the thousands in a tiny cheap piece of silicon... And yet, we're still treating things like magnetic media, optical media, and coaxial media as though they're discrete-time and we need to pack on them as many binary bits, in discrete binary forms, each occupying some discrete amount of time or space.

I think we've missed something.

This PWM-nibble method I've come up with may be still discrete-time, cramming 16 values in 1us. But, it relies *significantly* on the fact that the magnetic media is non-discrete when compared to the normal MFM or even RLL 'bits' usually stored, individually, in *discrete* bit-wide spacing. And the PWM system ideally does so WHILE maintaining the discrete-value benefits of our so-reliable digital systems today. 

That's the idea. 

And there's no reason to limit that idea to all-but-extinct floppy drives.

-----

OK, as an aside, it *just* occurred to me how much this system is similar to my project AVR-LVDS-LCD (todo:link github?)... Wherein I use an AVR's Fast-PLL-based timer to create PWM signals that correspond to serial data-streams to drive an LVDS/FPD-Link LCD.

Not a big deal, I just thought it was an interesting thing that these projects are so similar, yet seem so different.

There, I've managed to squeeze 48 colors, pretty close to evenly-spaced on a color-map, out of PWM signals fed into a *serial* input expecting sequential binary data-bits. (Todo: timing diagrams!). As well as sending proper pulses for Hsync, Vsync, and Data Enable... And this from the PWM-output of an AVR running at 1/4 the speed.

PWM FTW!