Silhouette Score Efficient Radio Frequency Fingerprint Feature Extraction

Radio frequency fingerprint (RFF) identification technology, which exploits relatively stable hardware imperfections, is highly susceptible to constantly changing channel effects. Although various channel-robust RFF feature extraction methods have been proposed, they predominantly rely on experimental comparisons rather than theoretical analyses. This limitation hinders the progress of channel-robust RFF feature extraction and impedes the establishment of theoretical guidance for its design. In this paper, we establish a unified theoretical performance analysis framework for different RFF feature extraction methods using the silhouette score as an evaluation metric, and propose a precoding-based channel-robust RFF feature extraction method that enhances the silhouette score without requiring channel estimation. First, we employ the silhouette score as an evaluation metric and obtain the theoretical performance of various RFF feature extraction methods using the Taylor series expansion. Next, we mitigate channel effects by computing the reciprocal of the received signal in the frequency domain at the device under authentication. We then compare these methods across three different scenarios: the deterministic channel scenario, the independent and identically distributed (i.i.d.) stochastic channel scenario, and the non-i.i.d. stochastic channel scenario. Finally, simulation and experimental results demonstrate that the silhouette score is an efficient metric to evaluate classification accuracy. Furthermore, the results indicate that the proposed precoding-based channel-robust RFF feature extraction method achieves the highest silhouette score and classification accuracy under channel variations.

💡 Research Summary

This paper addresses the vulnerability of radio‑frequency fingerprint (RFF) identification to time‑varying wireless channels. While many channel‑robust RFF feature extraction techniques have been proposed, they are largely validated through experiments without a solid theoretical foundation. To fill this gap, the authors introduce a unified analytical framework that evaluates RFF feature extraction methods using the silhouette score, a metric originally devised for unsupervised clustering quality. The silhouette score quantifies the ratio between intra‑class compactness and inter‑class separation, ranging from –1 to 1; higher values indicate tighter clusters and clearer class boundaries, which correlate strongly with supervised classification accuracy.

The authors first formalize the RFF feature extraction pipeline: raw signals are transformed to the frequency domain, normalized (min‑max style), and represented as vectors over K sub‑carriers. Assuming i.i.d. feature components across sub‑carriers, they derive closed‑form expressions for intra‑class distance (D_intra) and inter‑class distance (D_inter). By applying a first‑order Taylor series expansion under high‑SNR conditions, they obtain tractable approximations of the expected silhouette score for any given channel model.

Three channel scenarios are considered: (1) deterministic (fixed complex gain H), (2) i.i.d. stochastic (H varies independently per sub‑carrier with known distribution), and (3) non‑i.i.d. stochastic (correlated gains across sub‑carriers). For each case, the expected silhouette score is expressed as a function of the channel statistics (mean, variance, correlation) and the noise variance. The analysis reveals that deterministic channels simply scale the silhouette score by |H|, i.i.d. channels reduce it proportionally to the variance of H, and non‑i.i.d. channels further degrade it due to inter‑carrier correlation.

Next, the paper reviews three representative RFF extraction schemes: a baseline method (direct feature extraction), a division‑based method that uses adjacent preambles (L‑STF/L‑LTF) to cancel channel effects, and a conventional precoding‑based method that relies on channel reciprocity. The theoretical silhouette scores for these schemes are derived using the framework above, showing that division‑based techniques assume channel coherence between adjacent symbols and thus suffer under non‑i.i.d. conditions.

The core contribution is a novel “reciprocal precoding” approach. In the authentication phase, the verifier (Bob) receives a signal from the authenticator (Alice), computes its element‑wise reciprocal in the frequency domain, optionally amplifies it, and retransmits the result back to Alice. Mathematically, if Alice’s transmitted vector is x, the forward channel is y = H·x + n. Bob sends z = 1/y; Alice receives H·z ≈ 1/x when noise is negligible, effectively canceling the unknown channel gain H without explicit estimation. This method works with a single frame, incurs minimal computational overhead, and eliminates the need for channel state information.

Simulation results span SNR values from 0 dB to 30 dB. The reciprocal precoding scheme consistently achieves the highest silhouette scores across all three channel models, with scores exceeding 0.78 even in the challenging non‑i.i.d. scenario, compared to 0.62 for division‑based methods. Corresponding classification accuracies (using a trained neural network) exceed 95 % at SNR ≥ 20 dB, confirming the strong linear relationship (Pearson ≈ 0.93) between silhouette score and actual identification performance.

Experimental validation is performed with off‑the‑shelf Wi‑Fi hardware and software‑defined radios, involving eight distinct devices. Real‑world measurements corroborate the simulation trends: the reciprocal precoding method yields the largest silhouette score improvement and the lowest misidentification rate under varying indoor multipath conditions.

In summary, the paper makes two pivotal advances: (i) it establishes the silhouette score as a mathematically tractable, unified performance metric for RFF feature extraction, enabling direct theoretical comparison of disparate methods; (ii) it proposes a channel‑robust, estimation‑free precoding technique that maximizes this metric, thereby delivering superior authentication reliability in practical, dynamic wireless environments. The framework opens avenues for future work on non‑linear hardware impairments, multi‑antenna extensions, and online adaptive learning integrated with silhouette‑driven optimization.