It is better to have a reference channel to compensate for small buffer differences, drift and non-specific binding to the matrix.
The best reference is a related molecule that will not bind the analyte.
When that is not available, use a non related molecule of the same size.
When creating a refrence channel try to match the loading of the ligand with that of the protein of interrest.
For the SA-chip you can think of blocking the streptavidin with biotin to create a 'neutral' surface.