On Thu, Dec 9, 2021 at 9:01 AM Fabien PELLET <fpellet.f4...@free.fr> wrote:
> Maybe it could be a track to follow also... I read it very quickly. > > I go down to 1e-9 and there is no influence on the attack behaviour. > > I try difference sampling rate (200e3 and 1e6 for output on a USRP) and it > influences decay and attack as expected. > > I come back to AGC instead of AGC2 and the rate value has an influence on > both attack and decay behaviour as expected. > > Hope this will help. > > If I would like to play on the code of AGC2, how can I recompile only this > module and not all gnuradio ? > You could create your own OOT module, copy much of the code, and try new algorithms there. There is no good way to recompile only one block. > Le 09/12/2021 à 14:52, Jeff Long a écrit : > > Maybe "tmp" should be compared to 0, not to gain. It does look suspicious. > Not sure what algorithm the author had in mind. > > On Thu, Dec 9, 2021 at 8:41 AM Fabien PELLET <fpellet.f4...@free.fr> > wrote: > >> Maybe I'm wrong but at line 141, it should be a division and not a >> difference ??? I'm telling that because in line 143, the result "tmp" is >> compared to the gain. There is a unit problem here I think : amplitude >> (result of the difference) can't be compared to a gain which do not have >> unit by definition.... >> >> I will try wider value for attack. >> Le 09/12/2021 à 13:23, Jeff Long a écrit : >> >> The change in gain is also proportional to the difference between the >> reference level and the signal level. So it probably is having some >> influence but it reacts more quickly. Try a smaller value for attack. Take >> a look at the source here >> >> >> https://github.com/gnuradio/gnuradio/blob/35a242f1cd4b724be0c4708d8690f82c804aafd8/gr-analog/include/gnuradio/analog/agc2.h#L64 >> >> for more info. It may just not be a great algorithm, or may not work for >> the rates and changes in signal you have. >> >> On Thu, Dec 9, 2021 at 7:07 AM Fabien PELLET <fpellet.f4...@free.fr> >> wrote: >> >>> Thanks for the explanation. >>> >>> Do you have any idea on why the attack as no influence ? >>> Le 09/12/2021 à 12:58, Jeff Long a écrit : >>> >>> Attack and decay are unitless per-sample multipliers, so 1.0e-2 gives >>> you a time "on the order of" 100 samples. It's really an exponential attack >>> and decay, but that gives you an idea. So, your observation is correct - a >>> smaller value gives you a longer attack/decay. >>> >>> On Thu, Dec 9, 2021 at 4:33 AM Fabien PELLET <fpellet.f4...@free.fr> >>> wrote: >>> >>>> Hello, >>>> >>>> I'm trying to work with the module AGC2 but I have some trouble using >>>> it. >>>> >>>> First problem is : what is the unit of the parameters attack and decay >>>> ? >>>> I thought initially that it was in second but after some test it is >>>> clearly not in that unit. Indeed, if I increase the value of the decay >>>> for example, the time of the increase of the gain at the end of a >>>> signal >>>> will go down where it should go up. >>>> >>>> Second problem (maybe in relation with the first) : the attack value >>>> have absolutly no influence.... From 10-6 to 10-2, I can't see any >>>> changes on the output signal. >>>> >>>> Does someone have any explanation ? >>>> >>>> Best regards, >>>> >>>> Fabien, F4CTZ. >>>> >>>> >>>>