how to build the plaintext structure for impossible differential cryptanaysis on IDEA?

siba36

3/7/24, 9:25 PM

I'm trying to implement impossible differential cryptanalysis on 3.5 round IDEA using Miss in the Middle Attack on IDEA and Khufu paper as a reference and I'm stuck on the first two steps of the attack:

In the paper the authors say that I should choose a structure of $2^{32}$ plaintexts $X^1$ with identical $X^1_{2}$ , identical $X^1_{4}$ and all the possiblities of $X^1_{1}$ and $X^1_{3}$ .

So in order to build this structure I've been using input differences that goes from $(0000\;\;0000\;\;0000\;\;0000)_{hex}$ to $(FFFF\;\;0000\;\;FFFF\;\;0000)_{hex}$ and taking $plaintext_1$ in each pair from a large file of texts that I've collected from various books and then I would calculate $plaintext_2$ so that $\Delta plaintext=plaintext_1\oplus plaintext_2$ will be equal to one of the input differences in the previous range , so I should get a total of $2^{32}$ pairs as a result.

And in order to implement the second step of the attack I searched in the previously constructed pairset for pairs that their ciphertext difference statisfy $Y^{4'}_3=0$ and $Y^{4'}_4=0$ and they say in the paper I should get about $2^{31}$ pair that statisfy this condition so about half of the pairs will satisfy the condition.

But when I tried to implement these steps I get just about $10\%$ of the pairset that statisfied the previous condition , so is there anything I'm doing wrong with the way I'm implementing these two steps?

0 + 1

cryptanalysis

differential-analysis

idea

Maarten Bodewes

3/12/24, 1:18 AM

You're heading in the wrong direction if you consider plaintext to be "text". That's a classical term that's still in use for modern cryptography (just "message" is sometimes more clear, but generally we'd use e.g. "plaintext block" when it comes to indicating input blocks for block ciphers). Generally plaintext is just binary and you certainly would never use "text files" neither for input or for output to implement a plaintext attack.

Elon Musk

I sit in a Tesla and translated this thread with Ai:

EN: how to build the plaintext structure for impossible differential cryptanaysis on IDEA?

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.