EDFA NF from polynomial fit

[ojnas]

In the _calc_nf method in the EDFA class, for the advanced model, the average NF is evaluated from target gain and polynomial coefficients like this:

nf_avg = polyval(self.params.nf_fit_coeff, -dg)

where dg should be the difference between target gain and (average) design-flat gain but is calculated like this:

dg = max(self.params.gain_flatmax - gain_target, 0)

So for target gain > design-flat gain dg will be 0 and average NF will be constant. It this really a correct model? I think the polynomial fit should be valid for any target gain and dg should simply be calculated as:

dg = self.params.gain_flatmax - gain_target

[dutc]

Thank you for this feedback! @jeanluc-auge is putting together some comments and will provide feedback after the TIP Summit.

[jeanluc-auge]

Hi, thank you for your comments and taking the time to check the code. This is not a bug, and I stand by this implementation.

• If gain < GAIN_FLATMAX, the nf polynomial will indeed follow gain variations
• but if gain > GAIN_FLATMAX, you enter the so called extended gain range where the internal VOA is at its min value (so little NF improvement) and the amplifier starts to tilt. The range of this extended gain region is not fully limited in the code (partially in auto-design) and in real use, whereas the validity range of any polynomial fit is. Typically most polynomial fits may only be validated within the amplifier gain range and are not safe to use outside (strange rebounds may happen with high polynomial orders). As already mentionned: considering the very limited NF variation in this region, it was deemed a better choice to fix the NF to its minimum value.

[ojnas]

@jeanluc-auge, thinking a bit more about this, I have larger issues with the EDFA model. Your reasoning above applies to variable gain EDFAs (with internal VOA). I think an "advanced model" should be able to model any type of EDFA. For example, fixed gain EDFAs (without internal VOA) typically specifies a gain range as well as an optimal flat gain between gain_max and gain_min. So three gain parameters would be needed instead of the two (gain_min and gain_flatmax) that are currently allowed. For such an EDFA, the tilt is a function of the gain (e.g. represented by the dgt - dynamic gain tilt) and it is not possible to set a tilt target independently of the gain.

In the current code, distinction between fixed_gain, variable_gain and "advanced model" is only done when calculating NF, not when calculating gain profile. I think this is confusing since it does not represent what is normally meant with fixed vs. variable gain EDFAs.

[ojnas]

Two more questions related to this:

1. There is a hard-coded parameter TARGET_EXTENDED_GAIN in the select_edfa function. Shouldn't the extended gain range rather be specified per type_variety in eqpt_config.json?

2. The fact that the gain starts to tilt when entering the extended gain range does not seem to be implemented in the code, or do I miss something?

[jeanluc-auge]

1. The target_extended gain range is now parametrizable in the Span (eqpt_config.json) from the last Roadm equalization PR. But you are right it could make sense to have it specified in each amp type variety.
   2. I need to check too