The optimal options for TopHat in mapping 35bp SOLiD reads

[holywoool]

I've been mapping 35bp ColorSpace reads to hg18 using TopHat(v-1.3.2).As the read length is too short I use the following command:

tophat -C -g 1 -p 4 -a 8 -Q --integer-quals --segment-length 17 --segment-mismatches 2 --initial-read-mismatches 2 /opt/bowtie-0.12.7/index/hg18 brainf_35.csfasta brainf_35.qual

As TopHat recommends I set segment-length to half the read length.Allowing 2 mismatches(color mismatch) in each segments,which means one base mismatch is allowed in each segment's alignment.
The initial mapping was succesful,~30% reads were uniquely mapped to the reference.However,on the second mapping process,TopHat failed in mapping becasue of non-uniquely mapped alignments,and no junctions were found.
It is without any doubt that the short length is the reason.
Can I just use 35bp(not split into segments) in finding junctions?Or the possible resulting can be?
Thanks