On-line signature learning and checking: experimental evaluation

An experimental evaluation of a signature monitoring technique called online signature learning and checking (OSLC) is presented. The experiments are based on physical fault injection on the system bus of a prototype. A set of 12000 faults was inserted in the system. The duration of the faults, the line affected, and the point in the program where the faults were injected were chosen at random. Analysis of the results have shown that OSLC can detect up to 99.4% of the faults affecting the control flow or the instruction/operand fetch and 94.5% of all faults, with a short latency. The same set of faults injected for OSLC evaluation was also injected with the prototype configured for emulation of the basic signature monitoring technique. In this way, the improvements due to OSLC are easily highlighted, and the comparison with other signature monitoring techniques is more meaningful