In　this　paper,　we　propose　a　deep　learning-based　sparse　code　multiple　access　multi-user　multi-carrier　differential　chaos　shift　keying　(DL-SCMA-MU-MC-DCSK)　system,　for　the　sake　of　improving　the　spectrum　efficiency　(SE)　and　bit-to-error　(BER)　performance.　In　the　proposed　system,　the　transmitted　symbols　of　each　user　are　mapped　to　complex　codewords　which　are　randomly　generated　from　a　complex　normal　distribution,　and　the　codewords　overlap　on　sub-carriers　in　a　non-orthogonal　way　for　the　sparsity.　Subsequently,　the　real　and　imaginary　parts　of　the　resultant　complex　codewords　are　modulated　by　the　chaotic　signal　and　its　Hilbert-transform　version,　respectively.　At　the　receiver,　after　correlation　demodulation,　a　deep　learning-based　decorder　consisting　of　deep　neural　network　(DNN)　is　adopted　to　recover　the　transmitted　data.　We　also　compare　the　SE,　energy　efficiency　(EE),　and　complexity　with　benchmark　systems.　Simulation　results　demonstrate　the　superiority　of　the　proposed　system　in　terms　of　bit-error-rate　(BER)　performance.　Therefore,　the　proposed　DL-SCMA-MU-MC-DCSK　system　represents　a　remarkable　solution　for　low-power　and　cost-effective　short-range　wireless　communication.　IEEE