#!/usr/bin/perl -w
use strict;

sub fixn($) {
    my $str = "@_";
    $str =~ s/[^a-zA-Z0-9^_\-\\`\.\[\]\{\}~]//g;
    return $str;
}

sub fixn2 {          # no prototype needed
    my $str = shift; # since we only ever want 1 argument
                     # which will be converted to a string
                     # if needed anyway.
    $str =~ s/[^a-zA-Z0-9^_\-\\`\.\[\]\{\}~]//g;
    return $str;
}

my $input = 'a b \\ c d';
print "input='$input'\n";
my $myvar = fixn($input);
print "myvar='$myvar'\n";
my $better = fixn2($input);
print "better='$better'\n";
[download]

input='a b \ c d'
myvar='ab\cd'
better='ab\cd'
[download]

In other words, your regex is fine, but your code is sloppy, and your problem is probably somewhere in do_whatever_I_want_with()

update:

I have the feeling, that backslashes are lost in my $str = "@_"; <= that quotes turn them into a escape sequences.

That doesn't happen: backslashes are only interpreted in literal strings, they are not "recursively" interpreted in interpolated variables. The only exception is when you embed a variable in a regex, like so:

 $var = "bla[abc]\\t";  # $var contains bla[abc]\t
 /$var/;      # is equivalent to /bla[abc]\t/
 /\Q$var\E/;  # is equivalent to /bla\[abc\]\\t/
[download]

"Please do not use prototypes if you don't know why they are problematic."

Also ;-), please do not use prototypes if you know why they are problematic. It is generally a bad idea to use Perl prototype.

I solved the problem :)
Instead  my $str = "@_"; I used  my $str = $_[0]; which is the same as shift.
Interesting ideas, I'll meditate on them :)

You might be looking for Gory details of parsing quote-like constructs, in perlop. There's enough information there to last a lifetime. Backslash interpretation occurs in a pass before variable interpolation, BTW.