I'm not sure why you need banded population groups. Why don't you just choose a population distribution that first the age groups and use that to estimate death rates instead ?