diff --git a/lib_enc/fd_cng_enc_fx.c b/lib_enc/fd_cng_enc_fx.c
index 9c3a236d75d8ec84214f7397eeeb1526df695a3a..e51e2b68261873f63114710a610b5293c7d9ab84 100644
--- a/lib_enc/fd_cng_enc_fx.c
+++ b/lib_enc/fd_cng_enc_fx.c
@@ -2839,16 +2839,19 @@ void FdCngEncodeMDCTStereoSID_fx(
         {
             IF( lr_in_ptr_fx[ch][p] )
             {
-                t1 = BASOP_Util_Log2( lr_in_ptr_fx[ch][p] );     // Q25
-                t2 = L_add( t1, L_shl( lr_in_ptr_e[ch], Q25 ) ); // Q25
+                t1 = BASOP_Util_Log2( lr_in_ptr_fx[ch][p] );                 // Q25
+                t2 = L_add( t1, L_shl( lr_in_ptr_e[ch], Q25 ) );             // Q25
+                ms_ptr_fx[ch][p] = Mpy_32_32( t2, TEN_MULT_LOG10_2_IN_Q29 ); // Q23
+                move32();
             }
             ELSE
             {
-                t2 = 0;
+                // 10.f * log10f( EPSILON ) --> -150.0f
+                // Subsequent additions / subtractions happen on these numbers, so to avoid saturations
+                // this value is set to -128.0f in Q23
+                ms_ptr_fx[ch][p] = -ONE_IN_Q30; // Q23
                 move32();
             }
-            ms_ptr_fx[ch][p] = Mpy_32_32( t2, TEN_MULT_LOG10_2_IN_Q29 ); // Q23
-            move32();
             E_fx[ch] = L_add( E_fx[ch], L_shr( ms_ptr_fx[ch][p], 4 ) ); // Q19
             move32();
         }