My apologies, I’ve just noticed I’d misread what you’d written – I’d understood that you had single-shell data acquired at b=2000s/mm², which is clearly not the case! If you have multi-shell data, you should definitely just be using the multi-shell multi-tissue CSD approach, as recommended in the documentation.
Regarding the requirement for high b-value data, my previous statement still applies, with the additional caveat that the requirement for high b-values actually becomes much muddier in this case, and basically no longer applies. This does impact on interpretability, but not on the validity of the analysis itself – see discussion of the topic in e.g. this post.