How do I wake an ESP8266 when a float sensor value changes?

Question

My goal is to reset an ESP8266 in deep sleep (pull RST low) when the value of my float sensor changes (which detects the water level in a tank), so that I can report the value change to the IoT platform. It's a low power design, so the MCU should sleep as much as possible.

Edit: At first it seemed like the sensor value was on/off, but on second thoughts, it could be analog. I'll do some testing and update.

I created a dual edge monostable trigger which wakes the MCU when the value changes. Is this the best approach for low power, or is there a better solution?

Can the monostable I came up with be improved? I designed a circuit which I can make out of junkbox parts, but it doesn't need to be, I'm happy to use an IC or MCU feature instead. I'm also happy to use a different MCU.

^{simulate this circuit – Schematic created using CircuitLab}

Note: Clock signal and Q10 simulates the sensor value changing; purely for simulation and not actually part of the real circuit.

Edit: I see from an answer on a similar question that PIC micros have something called "interrupt on change". I see that ESP8266 has GPIO interrupts, but it looks like only a specific pin can be used to wake from deep sleep. It looks like the ESP32 can wake from different GPIOs, but can it wake on either falling or rising edge? Also, from the same answer I see that I might have created an "edge to glitch converter"... is my circuit a good or a bad example of such a design?

Edit: Below image is the float sensor. Unfortunately, I'm not sure where the datasheet is, but I'll do some testing to see if it's digital or analog.

Edit: After testing, I see that the sensor produces a digital signal, as I had hoped.

Answer 1

Edit: I turned this into a GitHub project.

Revision 1

I refined the circuit after testing with the MCU. The pull up for the float sensor was consuming a little more power than I wanted, so I had to tweak the design to work with a 100k pull up.

Also, it turns out, Q1 alone was not enough to pull the MCU RST pin low. When in deep sleep (and D0 is connected to RST), the pin is pulled high with a fairly strong internal resistor, maybe 1k? So, the weak current on Q1 base was not enough to pull RST low. So, I added an amplifying NPN (Q11 below).

^{simulate this circuit – Schematic created using CircuitLab}

R25 and R23 are actually 500k in the real circuit. I can't seem to get the simulation to work with those values, but 1M and 5M seem to yield a nice graph.

Revision 2

I realised that I now have 3 BJTs, and a potentially unnecessary diode, so I rearranged the circuit to remove the diode. I had to change R28 to 1k (otherwise the falling edge trigger wouldn't be strong enough to pull RST low), but the graph still looks good and it apparently manages to pull the RST pin low. I haven't tried the 2nd revision as a real circuit yet though.

^{simulate this circuit}

After this 2nd revision, I noticed something. My circuit is starting to look very similar to Olin Lathrop's dual edge monostable.

Of course, an exclusive XOR gate and delay circuit would probably be the simpler approach. As Andy aka mentions in the comments...

you can buy single device XOR gates with a SOT-23 footprint