If the rules are "change any sequence of characters that isn't a-zA-Z into a single underscore", you might like tr// better than a regex:

$ perl
my $s="UK Mobile - Vodafone [GBRVF] [MSRN]";
$s=~tr/a-zA-Z/_/cs;
print "$s\n";
^D
UK_Mobile_Vodafone_GBRVF_MSRN_
[download]

Still needs the second pass to get rid of MSRN_, but clean and straightforward otherwise.

Edit: I see someone already updated another comment node to suggest this, apologies for the dupe.