Hello Marco:

Your fist post raises a number of points, so I'll try to cover them all one at the time.

Firstly, you said that "The reference Impedance value was determined at no signal present.", and if this is true, it will certainly explain why you're experiencing problems.

If you are intending on using the Impedance Monitoring feature as a method to determine a fault on a speaker line, you must first achieve a reliable and accurate impedance reading to use as Alarm Baseline Data. To do this, you must inject the amplifier with some signal.
I would recommend a sine wave tone of 19.5 KHz at around -24dBFS, as a quieter level may also result in inaccurate readings.
The load monitoring band should have its centre frequency set also at 19.5 KHz, and its bandwidth should be of 20Hz.

This way, you're concentrating on a narrow band of frequencies, and your measurements should not be disturbed by program material like paging or background music. Also, the frequency is too high to be audible on a 100V line system, and should therefore not bother anyone.

Of course, and as HST correctly pointed out, you should set your Tolerance, probably to something like 10% or so. This will be of course dependant on your individual circumstances, as factors like temperature will change the impedance measured on the line. Also, a 10% shift on impedance measured on a line with 20 speakers is very different to a 10% shift on a line with 5 speakers.

And this brings us neatly to the last point, which is your comment of "with a high tolerance it doesn`t recognize a missing or defective speaker".

I do not wish to enter a debate and I'm well aware this topic has been hotly debated on many occasions; but employing impedance monitoring as a method to discover whether a single speaker is defective or missing on a 100V line is a bit like counting grains of salt while blindfolded relying just in your sense of touch.

It can be done, but only under certain circumstances.

The reality is that if you have a number of speakers on a 100V line, the changes on the line impedance when a single speaker is disconnected may be quite small. Small enough, in fact, to be smaller than the changes in line impedance that will naturally occur on the line throughout the day.
Therefore, you could be caught in the impossible position of needing to set a tolerance of 5% or less in order to determine whether a speaker has been disconnected, but the temperature changes between day and night cause a shift on impedance readings of 10%, forcing you to set a higher tolerance to avoid false alarms.
If this is the case, your only solution is to reduce the amount of speakers on that line so that the removal of one of them causes a change in impedance that is noticeable enough to offer a reliable parameter to detect failure from.

I hope this helps,

Martin