The top is the raw bass signal, bottom is compressed using Live's Compressor F1 model (F2 does the same), both with a 1 ms lookahead. Switching to the FB model (which is not feed-forward) removes the artifact, as does increasing the look-ahead time to 10 ms.
I assume it's not an artifact of too-fast attack on bass (increasing the attack time here to even 40ms did not make a difference). There are other compressors (like the BT compressors from Nomad, etc.) which seem to do the same, so I assume they are feed-forward. But without an adjustment for the look-ahead, they can't be tamed. They have bass presets, but are useless for bass (or at least "electronic" bass - maybe are just fine for bass guitar or modeled basses that have higher freq. attack transients from slap/pick sounds, etc.)
I simply don't use those compressors for this type of sound, so I'm just curious. I assume the feed-forward models are modeled after the FET hardware compressors, etc., and are used to catch unwanted transients but still have a reasonable attack speed (so you get smooth compression but are able to catch the "crap"), and maybe were never intended to be used for smooth basses? They seem to work great on drum hits, etc.
(Except for the times I want to play with adding a compressor's "sound" - like smashing a parallel drum buss with Rough Rider, etc., I find myself almost always going back to Live's compressor - between the three models, peak/RMS, it just seems to cover every need without unwanted distortion, etc., and the FB model works just fine for this type of sound. I am however thinking about the DCAM dynamics - with the transient shaper, etc., it looks like a great addition. But I ramble...
)