In　response　to　the　problem　of　data　redundancy　and　resolve　complex　of　the　indoor　location　based　on　high　fine-grained　channel　state　information　(CSI),　when　using　CSI　as　fingerprint,　a　CSI　indoor　location　method　based　on　stacked　sparse　auto-encoder　and　support　vector　machine　(SVM)　was　proposed.　First,　the　amplitude　and　phase　data　of　physical　layer　channel　information　were　combined,　and　the　stacked　sparse　auto-encoder　was　used　to　extract　the　deep　location　features　in　the　nonlinear　fingerprint　feature　space.　Then,　sparse　feature　fingerprint　was　generated　and　target　location　was　determined　by　support　vector　classifier.　The　application　of　sparse　feature　fingerprint　reduces　the　size　of　CSI　fingerprint　database　by　about　92.6%.　Meanwhile,　experimental　results　show　that　the　proposed　method　can　achieve　an　average　positioning　error　of　1.205　m　in　a　complex　indoor　environment　with　mixed　line-of-sight　and　non-line-of-sight　propagation　paths,　and　the　positioning　accuracy　is　significantly　improved　compared　with　other　methods.　©　2021,　Editorial　Department　of　Journal　of　Beijing　University　of　Technology.　All　right　reserved.