Every implementation of dual n-back that I've ever seen uses the set of letters 'C', 'H', 'K', 'L', 'Q', 'S', 'R' and 'T'.
Why is that?
The set is not always the one you suggest. There are many implementations that don't even involve letters, but use random shapes, numbers, spatial locations, or other stimuli.
In Jaeggi et al., 2007, they used spatial locations for the visual task, and
In Takeuchi et al. 2010, they used numbers and locations: