AVR Freaks

The sine is a half rectified 0-200Hz signal, 0 to 80V. I believe it's too slow to trigger edge detect.
 

For the CMP I was also intending to use a resistor divider and a DAC with 0.55v REF rather than a single very high value resistor to limit current through the pins protection diode. For the CMP this offers much higher hysteretic change which a single resistor does not. However, the DAC doesn't run in power-down either.
 

I have the RTC PTC going at 2Hz to clear WDT, true I could speed that up and check the port status in code but the higher wakeups might be counterproductive to a deep sleep.
 

It comes down to, if the pin level is zero and I enable the pin LEVEL_LOW interrupt will it fire or will it wait until the level first goes high then returns to zero? If it does this then it'll be the best solution.

I'm probably over-optimising, but why not be the best. 

The power supply is a large super-capacitor (so it has quite a lot of leakage about 0.2mA) feeding into a boost converter which powers the AVR.

In idle mode I'm measuring the aforementioned half wave cycle using a CMPn and TCBn in frequency measurement mode. I'm planning to run the AVR at 250KHz if it's fast enough to get all the code ran within the RTC PIT's 2Hz cycle rate.

Once the half-wave cycle hasn't been detected for a while (4 minutes) most peripherals shut-down. At this point it would be ideal to enter power-down mode.

I don't think trigger edge will work, there's no hysteresis with a single resistor. Using the CMP permits using a voltage divider, the DAC and setting hysteresis. It's why I went with this for normal use to deal with noise.

For wake-up I don't care about noise of course, but changing to a single divider then forces this disadvantage during normal mode.

There's also one more issue with the single resistor approach, there's no discharge route and it may float above 0v a bit, I've seen this in a previous life with PIC's.

Here's where I'm at.

I'll use standby = idle all the time. This keeps the CMP and DAC running. 

In "Quasi standby" I'll switch CLK_PER to 32KHz, which given nothing else is using OSC20M directly will save the 125uA.

The RTC PTC fires at 2Hz. It checks the CMP INT flag (Int isn't enabled, I assume flag still gets set). If detected it switches CLK_PER to OSC20M. Now TCBx will start calculating the correct frequency. I have another software timeout shown in the code below that forces the quasi standby again.

I've made an assumption the CMP INT flag is being set when INT isn't enabled. This is how I've seen it on PIC16's before.

 static void quasiPowerDown (void) {

    HW_CMP_INT_RESET(MOTION_CMP);
     _PROTECTED_WRITE(CLKCTRL.MCLKCTRLA,CLKCTRL.MCLKCTRLA | CLKCTRL_CLKSEL_OSCULP32K_gc);
}
 
 
 while (1) {
        quasiPowerDown();
        ticks = 0;

        // the only wakeup will be from the rtc timer int, no need to check flag
        while(1) {
            sleep_cpu();
            wdt_reset();
            // any motion detected?
            if (HW_CMP_INT_STATUS(MOTION_CMP)) {
                // fast cpu
                 _PROTECTED_WRITE(CLKCTRL.MCLKCTRLA,CLKCTRL.MCLKCTRLA & ~(CLKCTRL_CLKSEL_OSCULP32K_gc));
                MOTION_Task();
                if (motion.moving) {
                    break;
                } else if (motion.hz) {
                    ticks = 0;
                } else {
                    ticks++;
                    if (ticks > MOTION_DETECTED_TICKS) { // must be > not = otherwise conflict with motion.moving
                        ticks = 0;
                        quasiPowerDown();
                    }
                }
            }
        }
        
    // normal bootup here, held in another while loop until quasi powerdown is required again
        
 }