After-attack performance of parameter estimation systems

In this paper, attacks on systems estimating the value of an unknown deterministic parameter based on quantized noisy observations are studied. The attacks are categorized according to the information available to the attacking entity. The considered categories are attacks of full information, which are generally more devastating, and the information free attacks, which are generally less effective. Given an estimation algorithm and a quantization scheme, the presented work provides the actual error in estimation rather than a bound on it. More importantly, the error is provided for cases of misspecified observation models and sub-optimal estimation algorithms. A special property of the information free attack is presented. The error in estimation is studied under some possible attack mitigation schemes carried out at the fusion center.