How do I conduct a PCA on compositional data in R

Question

Compositional data are multinomial vectors that add up to 1 or 100 (for composition expressed in percent). Aitchison (1983) recommended an inverse logit transform (or link function) for these data (ln(xi-/xi)) where xi is the ith composition and xi- is 1-xi. As can be seen, this is very close to a logit transform, and probably a logit link function would work. I found a package logisticPCA in R, but the data that this package analyzes are vectors of binomial data, not multinomial composition data. My problem is that I have data where some of the composition components are 0, so it is not possible to calculate a logit transform. Adding an arbitrarily small value to these 0's (as is commonly done) is known to distort the analysis. Hence, I do not want to apply a logit transform directly. It seems to me that the logit needs to be incorporated into the MLE to work on my data. Any suggestions for how to implement Aitchison's approach or alternatives to Aitchison?