Processing Encrypted Floating Point Signals

An important problem in the context of signal processing in the encrypted domain (SPEED) is to perform secure computations on real-valued signals. This paper presents a first implementation of the IEEE 754 floating point standard and thus provides a comfortable solution to the aforementioned problem. We describe secure and efficient protocols, which allow to perform all arithmetic operations on encrypted floating point values. We further show how to enhance these protocols to allow for a basic exception handling and discuss strategies how to deal with exceptions in cryptographic protocols. Even for a 32-bit floating point architecture, our protocols require evaluation of garbled circuits with only a few thousand gates and a. marginal number of public key operations and are therefore ready-to-use in real world applications.