Can we generate two secret keys from same source?
The idea is we have time varying signal. We capture a window of say few milliseconds,$t$. I can chop values of signals at a rate $N_S$ and get $tN_S/1000$ samples. I put a threshold and assign 0 and 1 to values below and above it. Since this signal is shared I get a secret key of length $tN_S/1000$ bits.
Now the length of key depends on $N_S$ for a given window. We do agree that it gets harder to track original data if I decrease this $N_S$.
My problem is can I generate two different secret keys using this time varying signal. Approach can be to to use different functions say: digitization and derivatives. I can encode derivative to some bits.
These two functions can give two separate sequences. Now issue is I cannot say I have got additional level of security because: Signal can be compromised and that will compromise both functions results. This is what happens in general cyrpoto key management. Another point is I can use first method(thresholding) to generate back signal and guess output of second sequence hence its useless to generate that sequence!
My point is viability of back generation of signal depends on sampling rate.If its high the second sequence can be broken. But if sampling is low enough then back generation will be erratic hence second sequence cannot be guessed correctly.
I feel sampling can be optimized and keys generated through one process is not able to capture the full prowess of source and I can generate more keys from the signal which is still secure be it of a small length.
Is there any method already studied in literature of generating two secrets from a signal?
Hash is one practice that is effective in key management but the security level is decided by parent key no matter how many hashed keys we generate ahead. But if we generate two keys from a physical signal(assuming this signal is shared and random enough and cannot be guessed by eavesdropper) can the security level of system be increased?
If not are there analogous to hash functions on such time series signals? If not one way functions how can we compare feasibility of choosing two such functions? Is cross correlation a good choice?