Mesa 7i96S step and direction signals go high on power loss

Hi, Troubleshooting a series of mishaps in design.Background.

Problem 1. Operators using power switch as an e-stop.
Problem 2, Killing control power on this machine doesn't kill power to the drives and motors - Welp adding another relay now.
Problem 3, Mesa step and direction signals go high on power loss until the caps die out. (Which is what I'm actually curious about in this thread)
Problem 4, Teknic eclipse drives read a high step as a series of pulses. (Following up with them on that).

7i96s is wired with Step connecting to Step input on eclipse drive and Dir connected to Direction on eclipse. One motor has a RC delay timer on the enable, as the brake takes a moment to engage or the Z-Axis falls under its own weight. While waiting for the brake it jerks very violently.

Is there a way to make the Mesa card step and direction go low on power loss? 

Scope - Green is step, purple is enable delay, 100ms/div
 

I suspect that the drives see marginal output levels + drive generated noise as a series of steps.
Marginal output levels are possible with a slow ramp of the power as it drops to 0

Estop should really drop primary power to the drives but not control power. 

Suggestions:

1. Use one 7I96S isolated output to enable all drives. This will go false (off)  on power loss or watchdog bite
so the step/dir states during power-up/power down will not matter.

2. Use the STEP- and DIR- pins (and invert them in hal)

#1 is a more reliable choice

Even with the noise - it over accelerated the motors passed the over torque limit which stopped them.

Adding the relay was my first step and that works pretty well. Has 8ms of reaction time which motors on a test bench show about 30degrees.

I'll try the pin inversion right now in addition.

Thank you!

Quick confirming I leave the step + disconnected?

Looks like the swap kills the problem completely. Although I still prefer the relay solution.

There isn't a lot of noise on the step line, +/-50mv which shouldn't be enough to cause steps I would think?

The problem is, if the drives are enabled and powered when the step/dir signals decay 
there is a point where even 10mv of switching noise (at the drive PWM frequency so maybe 20 KHz)
will be interpreted at step signals, especially if there in no hysteresis past the optocoupler in the drive.

For actual servo drives, I would use a 7I96S isolated output to enable the drives.