This LM336-5.0 gets its current through a resistor, right? If this resistor will limit any current into Vref to less than 10mA if Vcc is smaller than Vref or equal to ground potential, then nothing bad can happen. Another idea is to connect a diode from Vref to Vcc, so that overvoltages at Vref are limited to one forward voltage drop.



The methode where all the different grounds are connected to ONE single point, like shown here (forget, that here only one supply voltage is used):



Maybe you have misunderstood me? With forming a ground loop along the transformer secondary windings I meant the following:

Pick two separate and arbitrary points on your digital plane. Now assume, that a ground return current flows between these two points. And now think about, whether there's any possibility, that a certain portion of this current is ALSO flowing on analog ground plane. That's the important question here. Now analyze your layout and this above and you will see, that with your design a digital ground return current can flow from the one point to +5V power supply, traverse the barrier of secondary windings via interwinding capacitance, run back to analog plane and through the connection of analog and digital ground plane (under your ADC) back to the other point on digital ground plane! That's what I meant with 'ground loop' here.



By filtering I did not mean to cut everything away! Of course if signal contains some higher frequency content and this is characteristic for this signal, then it must not be filtered away. In such a case, symmetrical signal routing is needed, which CAN easily be filtered, without degrading signal integritiy, namely by common mode filtering.
But I think your zero cross detector will not show higher frequency contents than some hundreds of kilohertz? In such a case you can filter out everything higher than 1MHz and would still benefit from removed very high frequency noise. By the way, imedance of a soft ferrite choke becomes relevant at frequencies higher than about 10MHz. A 1kOhm resistor shows this high impedance for all frequencies. So, when you can insert a resistor without affecting your signal conditioning, then do it.

How could you insert such a resistor?
This can be provided if you do not connect everything directly to micro, but via a single transistor inverter, like it's done with level conversion. Means you connect the zero crossing signal not directly to micro, but to the base of such a transistor stage. Then you can run the signal over the barrier gap by introducing some impedance.

Kai