What is Reference Network in time-series InSAR?

I have a question about the process of time-series InSAR.

What is the construction of a reference network?

Based on my understanding of the time-series InSAR process (e.g., PSI):

  1. N SLC images are selected, and co-registration and interferogram generation are performed.
  2. PS candidates are selected based on ADI (Average Deformation Index).
  3. Various criteria are applied to filter out PS candidates.
  4. Flat Earth and DEM are used to correct for topographic effects.

I’m curious about how the reference network is utilized in this process.
It seems like one of the criteria used to filter out PS candidates, but I’m not sure about the details.

Could you explain what the reference network is, why it is used, and how it is utilized in the process?

The picture below shows different types of reference networks from link.