If you try again to coregister both images again and the interferogram is not sufficient:
- Consider a higher orbit interpolation degree (differences are shown here: ALOS SLC L1.1 interferometry processing parameters and here Problem with Using ALOS Data)
- There was a nice study which explained the linear fringes by rotation of the sea ice during both acquisitions: https://elib.dlr.de/72754/1/0003594.pdf Could this partly explain the gradient you demonstrated here? A wierd gradient over my result