Detection and Suppression of Intentional EMI Attacks to Smart Speakers

Abstract

A new digital technique to detect and suppress intentional electromagnetic interference (IEMI) attacks on smart speakers (SSs) without disrupting their nominal operation during the attack is proposed in this article. The effectiveness of the proposed approach is verified by experiments on a proof-of-concept SS prototype equipped with an artificial intelligence speech recognition algorithm. Thanks to the proposed suppression technique, the power of malicious signal in the audio band is suppressed by 19 dB, thus effectively preventing the execution of malicious commands in more than 99.6% of the cases under IEMI attacks of up to 5 dBm injected RF power in 10–100 MHz bandwidth. Under the same test conditions, the proposed detection technique, operating independently from the suppression method, makes it possible to successfully identify IEMI attacks in more than 99.8% of the cases. Experimental results also show the proper recognition of nominal commands while IEMI attacks are performed.