A device can reveal its encryption key without saying a word. Tiny electrical traces carry clues about the secret, but noise can bury them before anyone can read them. The traces used in side-channel analysis are essential to breaking the key of encryption, and signal quality greatly affects the correct rate of key guessing.

But collected signals are often mixed with strong internal, environmental, and other noise. The proposed framework is designed to extract useful features from traces with a very low signal-to-noise ratio, so preprocessing becomes an important step before trying to recover the key.

The core idea is like sorting a messy recording: keep the broad patterns that carry the message, remove patterns that look like background clutter, and soften the borderline ones instead of making an all-or-nothing decision. The framework combines hard threshold truncation with soft threshold operation, and includes several ways to decide which patterns to keep or soften.

The approach refines three ways to choose where the hard cutoff should occur in the decomposition of a trace. It also introduces two soft operations for shrinkage and damping within a hybrid denoising approach, rather than relying on only one kind of cutoff.

The evaluation then examines security measures and experimental results after preprocessing, asking whether the cleaned traces make the hidden information easier to use. That matters because the study evaluates preprocessing through security metrics and experimental results, not just through the preprocessing step itself.

The signal-to-noise ratio reflects how much key information can be obtained from the traces. After hybrid denoising, it increased for nearly all byte positions in one test set and for most positions in the other. The results show that the framework can significantly reduce the noise component in a trace and improve side-channel attack efficiency, performing better overall than wavelet denoising.

The correct key becomes clearly separated from wrong guesses at the strongest peaks, especially after cleaning the measurements. That sharper separation makes the key easier to identify reliably in both attack settings. With the same number of traces, the proposed preprocessing gives a higher attack success rate than the original traces, and the preprocessing operations are overall better than wavelet denoising.

The improvement is not identical everywhere: on the AES dataset, one attack method is only slightly better than the comparisons, while another is better than both the original and wavelet-denoised traces. The performance gain is especially prominent for correlation-based analysis, where the traces needed for the attack are reduced by almost half.

For mutual-information analysis, the reduction is only around ten percent, showing that the benefit depends on how the attack reads the signal. The experimental results show that the framework can better reveal side-channel information hidden in noise, improve the signal-to-noise ratio, and greatly reduce the number of traces required for the attack.

The results suggest that noisy measurements may no longer hide side-channel weaknesses as effectively: preprocessing reveals leakage, improves trace quality, reduces required traces, and raises attack efficiency and success. Future work will continue optimizing the hybrid framework, while developing a more adaptive method for selecting rank values than its current empirical threshold schemes.

Cleaning those traces with a flexible combination of cutting away weak patterns and gently shrinking uncertain ones made the hidden clues easier to use. In these tests, attacks generally needed fewer traces and worked more reliably.