On the number of samples necessary to achieve observability