Intrinsic Physical-Layer Authentication of Integrated Circuits

An Optimization Framework for Generalized Relevance Learning Vector Quantization with Application to Z-Wave Device Fingerprinting

“…The resultant collected Signal-to-Noise Ratio (SNR) was at SNR C = 24.0 dB. Independent, like-filtered Additive White Gaussian Noise (AWGN) was added to collected signals to achieve desired operating conditions of SNR  [0 24.0] dB [21], [22].…”

“…RF-DNA fingerprints were generated, as in [21], [22], by 1) dividing each response into N R contiguous equal length bins, 2) calculating N s features within each bin and across the entire response (N R + 1 total bins), and 3) computing regional fingerprint vectors as,…”

“…Classification is considered for "one vs. many" scenarios as in [21], [22]. Two performance measures are considered: 1) SNR (dB) "Gain" (G SNR ) defined as the reduction in SNR for two methods to achieve a given average percent correct classification (%C) [23], and 2) Relative Average Percentage (RAP).…”

“…Computed for ID verification are True Verification Rate (TVR) and False Verification Rate (FVR) for the y-axis and x-axis, respectively, of authorized device Receiver Operating Characteristic (ROC) curves [46]. Two measures are considered herein to quantify verification performance: percent authorized [21], [22], and mean AUC (AUC M ).…”

“…Consistent with [21], [22], ID verification performance is commonly evaluated by a percentage correctly authorized from a binary grant/deny network access decisions based on a verification criteria, e.g. TVR ≥ 90% at FVR ≤ 10%.…”