Cryptonomicon

He gets to the point, conceptually, where his grandfather was when he commenced breaking the Arethusa messages. That is, he has a theory in mind now of how Arethusa worked. If he doesn't know the exact algorithm, he knows what family of algorithms it belongs to, and that gives him a search space with many fewer dimensions than he had before. Certainly few enough for a modern computer to explore. He goes on a forty-eight-hour hacking binge. The nerve damage in his wrists has mounted to the point where he practically has sparks shooting out of his fingertips. His doctor told him never again to work on these nonergonomic keyboards. His eyes start to go out on him too, and he has to invert the screen colors and work with white letters on a black background, gradually increasing the size of the letters as he loses the ability to focus. But at last he gets something that he thinks is going to work, and he fires it up and sets it to running on the Arethusa intercepts, which live inside the computer's memory but have never yet been displayed upon its screen. He falls asleep. When he wakes up, the computer is informing him that he's got a probable break into one of the messages. Actually, three of them, all intercepted on 4 April 1945 and hence all encrypted using the same keystream.

Unlike human codebreakers, computers can't read English. They can't even recognize it. They can crank out possible decrypts of a message at tremendous speed but given two character strings like

SEND HELP IMMEDIATELY

and

XUEBP TOAFF NMQPT

they have no inherent ability to recognize the first as a successful decryption of a message and the second as a failure. But they can do a frequency count on the letters. If the computer finds that E is the most common, followed by T, and so on and so forth, then it's a pretty strong indication that the text is some natural human language and not just random gibberish. By using this and other slightly more sophisticated tests, Randy's come up with a routine that should be pretty good at recognizing success. And it's telling him this morning that 4 April 1945 is broken. Randy dare not display the decrypted messages onscreen for fear that they contain the information that Wing's looking for, and so he cannot actually read these messages, as desperately as he'd like to. But by using a command called grep,which searches through text files without opening them, he can at least verify that the word MANILA occurs in two places.

Based on this break, with several more days' work Randy solves Arethusa entirely. He comes up, in other words, with A(x)= K,such that for any given date x he can figure out what K,the keystream for that day would be; and just to prove it, he has the computer crank out Kfor every day in 1944 and 1945 and then use them to decrypt the Arethusa intercepts that came in on those days (without displaying them) and does the frequency count on them and verifies that it worked in each case.

So now he has decrypted all of the messages. But he cannot actually read them without transmitting their contents to Wing. And so now, the subliminal channel comes into play.

In cryptospeak, a subliminal channel is a trick whereby secret information is subtly embedded in a stream of other stuff. Usually it means something like manipulating the least significant bits of an image file to convey a text message. Randy's drawn inspiration from the concept in his labors here in jail. Yes, he has been working on decrypting Arethusa, and that has involved screwing around with a tremendous number of files and writing a lot of code. The number of separate files he's read, created, and edited in the last few weeks is probably in the thousands. None of them have had title bars on their windows, and so the Van Eck phreakers surveilling him have presumably had a terrible time keeping track of which is which. Randy can open a file by typing its title in a window and hitting the return key, all of which happens so fast that the surveillance people probably don't have time to read or understand what he has typed before it disappears. This, he thinks, may have given him just a bit of leeway. He has kept a subliminal channel going in the background: working on a few other bits of code that have nothing to do with breaking Arethusa.