An　analysis　of　previously　derived　Ambisonics　decoders　shows　that　the　mid-high　frequency　performance　of　these　decoders　has　poor　agreement　with　subjective　perception,　since　the　optimized　decoder　coefficients　could　not　well　meet　the　objectives　developed　by　the　energy　vector,　and　the　polar　pattern　of　speaker　feeds　is　suboptimal.　In　this　paper,　new　objectives　are　added　to　fitness　function　for　the　optimal　speaker　response,　and　the　sum　of　weighted　rations　genetic　algorithm　is　applied　to　search　a　good　set　of　decoder　coefficients　to　meet　all　the　objectives.　The　decoder　of　an　example　application　of　ITU-4.0　speaker　array　exhibits　a　significant　improvement　in　localization　performance　at　the　rear　of　the　listeners.　In　addition,　the　binaural　evaluation　demonstrates　that　the　performance　of　the　proposed　decoder　better　conform　to　the　subjective　perception.　©　2014　IEEE.