In　this　paper,　we　investigate　sub-Nyquist　sampling　method,　which　can　work　at　the　sampling　rate　below　the　Nyquist　rate.　Nowadays,　the　sampling　rate　of　the　classical　sub-Nyquist　sampling　is　higher　than　discrete　Compressed　Sensing　(CS),　since　the　sparseness　of　the　original　signal　is　always　destroyed　by　basis　mismatch.　We　propose　a　new　multi-channel　co-prime　sampling　system　along　with　Sparse　Fourier　Transform.　The　analog　signal　of　each　channel　is　firstly　mixed　by　a　series　of　exponential　functions.　Then,　it　is　fed　into　integrators　and　commercial　analog-to-digital　converters　(ADCs).　It　is　noted　that　the　frequency　parameters　of　each　subsystem　are　co-prime.　Chinese　remainder　theorem　(CRT)　is　utilized　for　the　joint　processing　of　all　sets　of　samples　and　the　reconstruction　of　original　signals.　Finally,　the　effects　of　the　signal　sparsity,　the　sampling　rate　and　the　signal　to　noise　ratio　are　simulated　in　this　system.　In　addition,　the　experimental　results　show　that　the　proposed　system　can　work　at　the　rate　lower　than　Random　Demodulation　(RD),　Modulated　wideband　converter　(MWC)　and　their　improved　version.