Improving Covert Channel Bandwidth by Variable-Length Exponent Re-Coding in Floating-Point Representations

In previous work, covert channels in network payload data were introduced that recode floating-point representations, yet introduce errors by modifying mantissa bits. The present research refines those covert channels for the case of fully reversible covert channels, i.e., where the original data is reconstructable by the covert receiver. By using knowledge about the distribution of the data over the number range, a variable length encoding of the exponent is derived that allows to improve covert channel bandwidth to levels comparable with previous work but without introducing errors to original data. The approach is evaluated numerically for two example distributions with respect to bandwidth and detectability.