Hi Sven , 
Thanks a lot for your reply. 
My question is if it possible to get the desired result by using the  seasonal differences  before compacting the high-frequancy data. This method , that is sdiff in the high frequency series, is compatible with the method of constructing a Midas dataset ? Or I must use a "special" function as  I did when using first deifferences , that is the hfdiff function istead of diff function? 

If I understand you correctly, then you should be OK with using sdiff in the hi-freq context. In a quarterly and monthly context, seasonal always means one year ago (or $pd periods, where $pd==4 for quarterly and $pd==12 for monthly), so that matches.

The same should hold for weekly data, I guess, but subject to testing and verification.

For daily data, however, things are likely to become trickier. There, the standard meaning of seasonal is (I believe) from week to week, and there you then have a mismatch with the seasonal definition for weekly data.

Also, all this assumes that there's just one layer of seasonality. For monthly data, in principle you could think of both a quarterly and an annual cycle. But so far gretl does not natively support several coexisting seasonalities, AFAIK.

cheers

sven