ADuM1100 glitch filter window vs. very narrow pulses

Q 

Problem description:

I have found that narrow pulses of about 5ns cause the output to respond to
only one edge and produce a pulse of many hundreds of nanoseconds until the low
frequency path corrects it.

It happens on both polarities and it seems to happen over a very small range.

It appears that there is a very small range of pulse widths which defeat the
input filters.

This seems to be at odds with the spirit of this document

http://www.analog.com/static/imported-files/product_highlights/5406830898840Agil
ent_NVE_FMR_IsolatorsF.pdf

Find enclosed the circuit used to show the problem and Excel spreadsheet of the
results for both A and B grades.

 
Vdd1 and Vdd2 are 3.3V and the whole thing is on a copper-clad board
ground-plane with no isolation or split between the two grounds. (ie there is
only one ground)

The PWM generator circuit provided is what I used to generate the pulses as
shown in the Excel spreadsheet.

I included it so that others can replicate the problem.

The PWM generator circuit generates a 192kHz square wave, turns it into a
triangle wave and uses the SN65LVDS9637 as a comparator to turn that into
varying pulse widths.

Please can you advise what may be the cause of the problem?

 

A 

The min pulse width specification is exceeded by the 5ns pulse. The input
glitch filter at a certain small pulse width may still respond to
first edge of the narrow pulse width, but not the second, due to the timing of
the glitch filter window.
If you tried a wider pulse it may work ok. If you tried a narrower pulse it may
give no output as expected, since it would be smaller than the glitch filter
window.
The ADuM1100 has a refresh timer that will send a refresh pulse internally
every 1usec if no edge pulses are sent, and in this case a low input level will
cause a low refresh to make the output low 1 usec after the edge pulse to
correct the output.