The　observed　data　are　usually　contaminated　by　various　errors　in　practical　applications.　This　paper　deals　with　the　density　deconvolution　problems　with　super　smooth　errors　under　heteroscedastic　situation.　We　provide　a　new　wavelet　estimator　and　investigate　its　upper　bound　of　L-p　risk　over　Besov　ball.　Then　the　lower　bound　is　given　which　shows　that　the　defined　estimator　attains　the　optimal　convergence　rate.　It　turns　out　that　Theorem　3.1　extends　some　existing　theorems　of　Pensky　and　Vidakovic,　Fan　and　Koo,　as　well　as　Li　and　Liu　in　some　sense.　However,　since　the　convergence　is　just　logarithmic　rate　which　is　not　satisfactory　for　us.　Motivated　by　the　works　of　Pensky　and　Vidakovic,　Li　and　Liu,　we　study　the　wavelet　estimator　over　super　smooth　space　to　improve　the　convergence　rate.