A Limit-Aware Sparse Frequency-Domain Decision Engine for EMI Risk Feedback in Resource-Constrained Systems.

Resource-constrained electromagnetic interference (EMI) management requires a frequency-domain feedback path, while FFT-based full-spectrum processing introduces redundant computation, storage, and data movement for decision tasks. This paper proposes a limit-aware sparse frequency-domain decision engine for internal EMI risk feedback. The engine redefines EMI analysis from spectrum reconstruction to selective exceedance verification and uses randomized spectral reordering, flat-window bucket aggregation, and folded sampling to compress the length-N spectral search into bucket-level observations. Then, by comparing bucket-level amplitude envelopes with local limit envelopes, the method excludes risk-negative buckets, and only uncertain buckets are further refined through phase localization and sequential verification. Degradation experiments involving continuous background uplift, main-harmonic sidebands, and parasitic resonance clusters clarify the applicability boundary of the proposed method, and measured GaN power-converter spectra acquired through an in situ EMI sensing chain remain inside the empirical usable region. RTL evaluation at 100 MHz shows that the proposed design achieves an average decision latency of 6.031 ms. Compared with two FFT baseline implementations, it reduces BRAM usage by 95.17% and 97.59%, dynamic power by 54.0% and 83.0%, and per-decision dynamic energy by 46.3× and 33.3×, respectively. The results show that the proposed decision engine reduces hardware overhead for frequency-domain EMI risk feedback in resource-constrained systems.