I think you have to make the assumption that all the bitstrings are equally likely. With that assumption, the hamming weight has a Binomial(n, p) distribution with p = 1/2. From this you can get the mean and standard deviation and apply Chebyshev's inequality.