It doesn't matter where the sequence originated, if you have to do a protein phylogeny then do it with amino acid sequences. If, when you have done that, there isn't enough variability to resolve the proteins in the way you want (ie they are too similar) then you could repeat the process with nucleotide sequences as you will get much more variation.
In truth you might find that almost all the amino acid sequences you are dealing with are "conceptual translations" of DNA sequences. Very few sequences originate from protein sequencing, and lots come from DNA sequences that have been translated. This doesn't matter for your purposes.
This article (linhttp://nar.oxfordjournals.org/content/31/13/3537.fulltio k text) may answer your doubt. Without doubt protein sequences have less signal to noise ratio as compared to DNA sequences and furthermore protein alignments also benefits from substitution matrices and moreover Phylogenetic links disappear more rapidly between DNA sequences (discrepancy) than between protein sequences.
Details about the best approach to build a phylogenetic tree and the advantages and disadvantages of using DNA and protein sequences can be read from, " Bioinformatics and functional genomics" Jonathan Pevsner.
Wish u luck :)