<p>This master thesis has been performed at Sectra Communications AB in Linköping where they manufacture electronic encryption devices. In cryptology it is essential that an eavesdropper cannot find the plain text. To assure that a system is safe, all devices are thoroughly examined so that they don’t emit any secret information, so called Compromising Emanations (CE). This is done by comparing the secret signal with measured signals on the screen of an oscilloscope, a process that is very time consuming and furthermore, the risk of missing compromising emanations is high. The purpose of this thesis is to investigate the possibilities to let a computer do this job. Two methods for comparing the signals have been suggested and implemented; cross covariance and model estimation. In the cross covariance method, the cross covariance is calculated and its highest value determine how well the signals match. In the model estimation method, a number of ARX models are estimated, and if the best model is successful in explaining the output signal, it is assumed that there exists CE. </p><p>The work shows that both these methods are functional for solving the task, yet with a few limitations. Both methods announce CE for clock signals, which really aren’t compromising. The cross covariance method is limited only to find compromising emanations that look like the secret signal or certain frequently occurring non-linearities of it. The model estimation method is more general and detects CE for all linear systems, along with the known non-linearities. However, this method has a flaw: the time delay in the system must be known approximately. Further, the results show that the model estimation method is considerably more time consuming than the cross covariance method. The measurement is carried out with a computer communicating with oscilloscope and spectrum analyzer via the GPIB interface. The implementation has been done in LabView with calculation modules coded in C++.</p>

This master thesis has been performed at Sectra Communications AB in Linköping where they manufacture electronic encryption devices. In cryptology it is essential that an eavesdropper cannot find the plain text. To assure that a system is safe, all devices are thoroughly examined so that they don’t emit any secret information, so called Compromising Emanations (CE). This is done by comparing the secret signal with measured signals on the screen of an oscilloscope, a process that is very time consuming and furthermore, the risk of missing compromising emanations is high. The purpose of this thesis is to investigate the possibilities to let a computer do this job. Two methods for comparing the signals have been suggested and implemented; cross covariance and model estimation. In the cross covariance method, the cross covariance is calculated and its highest value determine how well the signals match. In the model estimation method, a number of ARX models are estimated, and if the best model is successful in explaining the output signal, it is assumed that there exists CE. The work shows that both these methods are functional for solving the task, yet with a few limitations. Both methods announce CE for clock signals, which really aren’t compromising. The cross covariance method is limited only to find compromising emanations that look like the secret signal or certain frequently occurring non-linearities of it. The model estimation method is more general and detects CE for all linear systems, along with the known non-linearities. However, this method has a flaw: the time delay in the system must be known approximately. Further, the results show that the model estimation method is considerably more time consuming than the cross covariance method. The measurement is carried out with a computer communicating with oscilloscope and spectrum analyzer via the GPIB interface. The implementation has been done in LabView with calculation modules coded in C++.