AVR Freaks

Perhaps take a look of a FM sound decoder.

One problem for you is that you don't have a sine wave. You will be sensitive to the logic trigger level it will give the signal a DC level (2.harmonic noise).

So perhaps only look at high going edge, at this low speed all can be done with counters.

But main thing make a PLL that run at carrier (slow changes so you can't see data). And then compare it with the incoming signal.   

22.5 deg at 200KHz should give 5 clk on a 16MHz cpu, so 10 clk diff for 0 and 1  for 1000 (500 or 2000 depending of modulation). So it should not be to bad.

Since the baud rate is so low, (many hole periods for each bit about 1000 depending of modulations form), then perhaps clk a timer with the 200KHz and then a interrupt for something like each 200 clk , and then use the time difference between each interrupt.

if you fall into a shift there should be 10 clk (@16MHz) difference. And some with about 0 and then one with 10 clk the other way. (if you don't keep track of where the shift could be you will need look over two bloks)

normally:

0 0 10 0 0 0 0 -10 0 0 0 0 0 0 0 0 0 0 10 .......

but it could be

0 0 6 4 0 0 0  -5 -5 0 0 0 0 .........   

Yes, that would work, but needs a crystal at each end so jitter does not swamp the small step looked for, and you also limit the time-precision to 'somewhere inside the interrupt'. ~1ms at 200 clk checks. 

Another approach entirely, is to focus on your phase comparator, and use a HC4046/7046/9046/LV4046 type phase lock loop.

That is set up to lock to 200KHz with high gain, and the analog control voltage will have an impulse step on every modulation edge.

the phase of the carrier advancing or retarding by 22.5° will give a easy to see analog pulse, or because this is mostly locked (impulse every 40ms), you could use the digital PCP_OUT via a filter to interrupt the MCU

Getting the phase would be possible with a quick sample of PC2 out, tho that is tri-state pin, or maybe the PC3 on the LV4046 can demodulate directly ?

With such an external part the MCU loading drops significantly.

Is both transmitter and receiver clk very accurate?

And why do you need to know exactly when the shift happen? (does something need to be in sync?)

if you only need to decode the data (that is slow) then it doesn't matter if it's delayed 1/10 of a bit (or does it).

I would say that a tiny 85 running with a internal clk will spend less that 10% of the time decoding the signal. 

Perhaps look of how DGPS works that is about the same freq and data speed.

An other way would be to use the ADC and under sample, if you sample with lets say 12.5KHz and have your 200KHz PLL control the time so the sample is close to the zero crossing (high delta V), it will give close to same value all the time (because that is actually what control the PLL), but there should be two values depending of the phase.     

A book like Digital Signal Processing in Communication Systems is good to get ideas.

Add: 

In this case the 200KHz PLL also only need to be 12.5KHz

Add:

Remember that there are more information in a 12.5KHz sinewave than in a 200KHz binary signal.

So what is max legal jitter on the 1PPS ? (we know that there are no long term error).

Does this mean that the transmitter always send data? 

or stay at one angle most of the time and first shift is the point where all get's in sync?

Can you tell what it is ? (perhaps give a link)

But does it send data all the time?

or start sending each sec?

It sounds like it has about the same function as a radio controlled clock, to put everything in sync.

I will say that 1uS is a small value when your base clk is 200KHz (5us), if you only look on on off signals, perhaps have a 200KHz timer interrupt that "fire" something like 45 deg before the shift (with manchester coding there for sure is a shift at center of the bit), and then make 1-2 MHz ADC samples the next 90 deg. so you can see exactly when the shift was (remember that the HW filters give a delay), and use that 0.9999.. sec later (basically update the 200000 Hz timer)   

if it is measuring a pulse length- then ATTiny13A + 20Mhz external osc, INT0 and PCINT0 can do it.