In this abstract, we describe how to improve time domain full waveform inversion using source wavelet convolution, windowed back propagation and source side illumination. Instead of estimating the source wavelet from field data, a user defined source wavelet can be convolved to field data. This convolution makes waveform matching between modeled and field data easier. Increasing time window applied to residual enables top down velocity update and reduces the possibility of being stuck at a local minimum. The balance of gradient value can be improved by the illumination compensation using the square of source side wavefield. Well balanced gradient helps FWI restore the absolute value of velocity. We apply this method to estimate migration velocities using 2D and 3D synthetic and real data examples.