My goal is to implement a mathematically correct multivariate normal distribution using Sobol QRNG sequence as a source of randomness. The implementation should NOT produce the whole set of a given size at once, but instead it should reveal next values one by one, also depending on getting new values out of the Sobol sequence. So the algorithm should be streaming the data with minimum memory requirements.
This guy claims he implemented MVN out of Sobol, which is officially endorsed on the MATLAB website. However, he has another call to rand() in the middle of his code, so it feels he does not fully understand what he is doing. That's why I hope for your help with a strictly mathematically and statistically valid approach here on Math.