I am looking for a perfect OTP design, so let's see if this design is good.
There are 2 issues when it comes to a good OTP system, the key and the plaintext, we will use XOR as cypher:
If the plaintext is like a message, then it should be long enough, otherwise it's pointless. What is the point in encrypting "Hello world", or something like that. So it should be a long message, so that it can't be guessed. If the plaintext is a private key itself, then it should be truly random and long as well, otherwise this scheme is pointless. This has nothing to do with the OTP I will talk about, it's just a caveat.
The key is the important part. It has to be truly random, the same size as the message or longer, and used only once, hence an OTP (onetime pad). So if we XOR the plaintext with the key, we get an unbreakable OTP.
The weakest link is probably the randomness of the key. However there is a solution:
What if we XOR it multiple times with different keys, like this:
$$Cyphertext = ( ( ( ( ( (Plaintext \oplus K1 ) \oplus K2) \oplus K3) \oplus K4) \oplus K5 ) .... \oplus K_n$$
So we would just Xor the plaintext
n times with
n different one time keys. Even if the key is biased or contains low entropy, the Xor function is known for it's de-biasing features, so Xor-ing multiple keys togeter would decrease the bias, and even for a very low quality random key, I think 5-6 rounds are enough, the bias will definitely shrink after 6 rounds to insignificance. But it also adds entropy into the system. By having 1 weak key with unknown entropy, we can have more keys mixed into the system thus increasing the overall entropy of the keys.
Would this be a perfect OTP design?