We know that for the last step of QRNG: we need to separate quantum and classical noises from each other so we use extractors, after extractor we need privacy amplification step. At this point: if extractor is a strong extractor we can use it as a privacy amplification too. However I could not understand differences between extractor and privacy amplification. If I use the strong extractor for the second time, will I be able to complete my privacy amplification step? What will change in privacy amplification? If what happens, privacy amplification will be completed? Thanks
Privacy amplification (PA) cannot be done at all with only a QRNG (which I take to be a quantum random number generator). Privacy amplification resides further along the application stack, and requires two co-operating parties (Bob and Alice) and an opponent (Eve) listening in on their quantum channel within their quantum key distribution network (QKDN) :-
Essentially privacy amplification (pink bits above) in QKDNs aims to allow two parties, Alice and Bob, to distil a secure final key from a partially secure bit string. The degree of security (bits) is governed by the Leftover Hash Lemma (LHL), which also can be applied to simple random number generation for regulating the output bias.
The purple rectangle is where true random numbers are generated within a QKDN. Clearly within a QRNG that would be the end of the story. A randomness extractor may or may not be incorporated into the output stage of the QRNG as they're not always needed. If you did need a randomness extractor, they're used to alter the probability mass function of the quantized signal into an almost perfectly uniform distribution.
That's a lot of stuff, so an example is probably in order. See High-speed Privacy Amplification Scheme Using GMP in Quantum Key Distribution, specifically §2.2.