The algorithm consists starting with the block of 128 bits, adding some bits of the key to it (mod 2), and then doing 14 rounds to ``mix'' it more and more. In each round the bits are changed in a nonlinear way and some ``round key'' is added in. The last round has a slight modification.
The round key is derived by mixing up the 256-bit key using some of the same methods (to be described) that occur in the rounds themselves. Each key bit is used in several rounds to make the rounds more interdependent.