How to do logarithmic downsampling

How to downsample the data in a vector by only keeping n samples per each log-cycle? For example I would like to downsample the data by keeping only 30 samples between 1e-4 and 1e-3 and similarly 30 samples between 1e-3 and 1e-2, and 30 samples between 1e-2 and 1e-1, and so on. If the number of original samples in each log cycle is already less than 30, then keep all of those.

[Jonas]

you could calculate floor(log10(data)) which gives you the decade in which the numbers were originally. then you could check the number of occurences per decade and decide which decade appears more than 30 times and remove those entries.