TY: THES T1 - BSC - Bloom based stream cipher A1 - Neto, Nicolas dos Santos N2 - A Linear Feedback Shift Register (LFSR) is a building block that is frequently used to build fast, hardware-based stream ciphers. However, the fact that an LFSR is bit oriented makes it inefficient when implemented by microprocessors. On the other hand, LFSR?s have a very well-defined internal behavior, defined by a carefully chosen (primitive) feedback polynomial, which facilitates the evaluation of their quality using mathematical tools but also their cryptanalysis. This work consisted on creating a generalized LFSR where the information stored in each stage of the shift register is a 64-bit word, instead of a single bit. Furthermore, a variable feedback polynomial is used instead of a fixed one, for making cryptanalysis harder. The variability of the feedback polynomial is given by the state of a Bloom filter. A Bloom filter is a well defined construction used to detect a possible repetition of a value observed in the past, and was used in our stream generator to provide a hard-to-model, always changing state. The evolution of the Bloom filter state is cyclic, in the sense that during some iterations it accumulates ones (1?s), while in other iterations it accumulates zeros (0?s). The number of iterations in each case is not fixed, it is given by an accumulated number of collisions in the Bloom filter itself. UR - https://ria.ua.pt/handle/10773/28300 Y1 - 2018 PB - No publisher defined