Authentication　is　the　first　line　of　defense　in　communication　security.　Physical　layer　authentication　(PLA)　based　on　Channel　Impulse　Response　(CIR)　offers　a　lightweight　approach.　Unlike　traditional　cryptographic-based　authentication,　physical　layer　authentication　identifies　data　packets　through　wireless　CIRs.　This　avoids　the　compatibility　requirements　of　upper-layer　protocols　and　presents　advantages　such　as　flexibility　in　authentication　methods　and　suitability　for　massive　heterogeneous　terminal　access.　However,　current　physical　layer　authentication　often　requires　illegal　CIR　samples,　which　are　hard　to　obtain.　Furthermore,　training　for　deep　learning-based　physical　layer　authentication　requires　numerous　samples,　and　the　low　latency　characteristics　of　Mobile　Edge　Computing　(MEC)　means　it　might　not　have　ample　time　to　gather　sufficient　signals,　reducing　authentication　performance.　This　paper　presents　a　physical　layer　authentication　scheme　based　on　deep　learning,　mapping　CIRs　to　device　locations,　and　further　mapping　to　their　authentication,　achieving　multi-user　physical　layer　authentication.　In　the　absence　of　illegal　device　CIR　samples,　it　discriminates　against　illegal　devices　and　classifies　legal　devices.　Moreover,　a　data　augmentation　technique　using　noise　injection　is　employed　to　enhance　the　performance　of　the　multi-user　physical　layer　authentication　scheme,　speeding　up　the　training　process.　Extensive　experiments　on　public　datasets　confirm　the　efficacy　of　our　approach.　©　2023　ACM.