Sparse　component　analysis　is　a　signal　processing　method　based　on　sparse　representation.　It　is　widely　used　in　underdetermined　blind　signal　separation.　When　estimating　the　mixing　matrix,　the　aliasing　level　of　the　mixed-signal　scatter　at　the　origin　center　is　too　high,　and　it　will　affect　the　accuracy　of　estimating　the　mixing　matrix.　In　this　paper,　in　order　to　overcome　the　shortcoming,　a　new　＂Vanish　Circle　K-means＂　clustering　algorithm　that　can　estimate　the　mixing　matrix　more　effectively　is　proposed,　combined　with　time-frequency　analysis　to　achieve　the　instantaneous　linear　aliasing　blind　separation　signals　and　obtain　a　good　separation　effect.