What is the most difi cult part of producing ssb for voice signals by using the phasing methods?