Please have a look at this topic, the same question was discussed quite recently: Abnormal Values in DEM from Sentinal-1 Data
Making a DEM from Sentinel-1 data is very likely to end in miserable results and there are many reasons for this. Unfortunately, only few of them can be tackled by the selection of suitable image pairs. As long as there is a temporal baseline of at least 6 days, most parts of the world will not allow the generation of a proper DEM from repeat-pass C-band data.