There are a number of feedback patterns on a chain of dividers used for random numbers, CRCs, scramblers and so on which work well. Most potential patterns don't work at all, or have poor behaviour. That might be bias in a PRBS sequence, or a lack of self-synchronisation for a data scrambler.There are a number of more complex patterns, like the Galois approaches, where logic is inserted between the divider stages. Some of those work well, but most do not. Nobody has a really solid way of deriving from first principals which camp any feedback pattern will fall into. Even for patterns which give the general characteristic you want, like maximal length for a PRBS generator, many will produce clear patterns in some parts of the sequence, so they are weak as PRBS generators. If you have a sequence that behaves well for your needs, be happy.