Filings peaked at 49 in 2017 and have not returned to that level since, with 2022 sitting at 42 as a rough midpoint. The partial 2026 figure of 8 looks like a drop but is understated because publication typically lags filing by around 18 months, so the true 2025-2026 filing rate will only become clear once those applications publish. Overall the pattern reads as a mature, flat-to-declining field rather than one with fresh growth momentum.

G01R (electric and magnetic measurement) and G06F (digital data processing) are the two largest subclasses in this corpus, at 440 and 412 records respectively, and between them account for most windowing and spectral-leakage claim activity. H03H (impedance networks and filters) is a clear third at 193 records, covering sidelobe-suppression and filter-adjacent claims. Smaller but still meaningful activity sits in G10L (speech and audio), H04L and H04B (transmission), G01S (radar/sonar/positioning) and A61B (diagnosis and surgery), which mark where spectral-analysis methods get applied to a specific downstream signal or device.

The recent-year momentum data shows several historically active filers — including Fraunhofer, Advantest, Samsung Electronics, Ericsson, Westinghouse Air Brake and Hughes Aircraft — with zero filings in the latest year, which suggests either a pause in this exact claim space or continuations still working through the publication pipeline. The most-cited individual records instead trace back to defense, biomedical-instrumentation and audio-coding assignees, including the U.S. Navy's adaptive-window FFT patent and a cluster of biopotential-analysis and lapped-transform-coding patents. Chinese state-grid entities appear repeatedly in the co-assignee pairs, pointing to grid-monitoring applications of spectral analysis as an active but separate cluster from the classic instrumentation lineage.

The clearest gaps sit in the smaller IPC subclasses relative to the core: G01S (radar, sonar and positioning, 61 records) and A61B (diagnosis and surgery, 43 records) both use spectral-analysis and windowing methods but carry far less filing density than G01R or G06F, and no single assignee dominates either. A workable strategy is to claim a specific window-function or sidelobe-suppression method tied to a concrete application step in one of those domains, rather than claiming the transform or windowing math on its own, since that abstract layer is where the oldest and most-cited prior art already sits.

Largely yes for the foundational methods. The most-cited records in this corpus — spectral-analysis systems for biopotential signals, an adaptive-window FFT processor, and a lapped-transform coding patent — date from the 1980s and 1990s and are cited heavily because they were foundational, not because they are still commercially dominant. High citation counts inside a searched corpus favour older records almost by construction, so they should be read as markers of influence on later claim drafting rather than as evidence of what is currently being commercialised or actively litigated.