Real-Time Anomaly Detection and Signal Aggregation Framework for Adaptive RF Awareness in Wideband Naval Systems

DTJ Management LLC proposes to develop a real-time anomaly detection and signal aggregation system designed to enhance the Navyís capability to detect, classify, and respond to known and unknown radio frequency (RF) emissions across wideband operational environments. The proposed innovation addresses a critical gap in current signal intelligence (SIGINT) tools, which rely heavily on pre-trained classifiers that cannot adapt to unfamiliar or evolving signalsóa vulnerability that adversaries increasingly exploit through waveform modification, protocol evasion, and low-probability-of-intercept (LPI) techniques.ÝOur solution introduces a machine learning-based cue generator that operates within a high-dimensional feature space built from time-frequency transformations (STFT and CWT), statistical descriptors (entropy, variance, kurtosis), and modulation-specific features (symbol rate, pulse width). These feature vectors will be processed using unsupervised clustering algorithms such as k-means, DBSCAN, and Gaussian Mixture Models (GMMs) to establish known signal regions and identify statistical outliersóemissions that deviate significantly from any learned cluster. Anomaly scoring will be performed using adaptive distance metrics, including Mahalanobis and cosine similarity, with dynamic thresholding informed by ROC curve optimization to balance detection sensitivity and false alarm rates.The system will be implemented on commercial off-the-shelf (COTS) processors and integrated into a lightweight, operator-facing cue generator interface capable of supporting real-time visualization, classification confidence scoring, and downstream signal recording. Synthetic datasets encompassing known and intentionally varied signals, contaminated by noise and interference, will be used to train and validate the algorithm under realistic RF conditions. Performance will be evaluated in terms of detection accuracy, latency, and clustering fidelity during hardware-in-the-loop demonstrations using USRP-based signal emulation.ÝThis Phase I effort will demonstrate the technical feasibility of the proposed system and produce a deployable prototype ready for Phase II integration into Navy DSP architectures and wideband receiver systems. The technology is directly aligned with the Navyís goals of transitioning adaptive, low-latency AI/ML tools into acquisition pipelines and fleet operational use. Broader applicability includes integration into U.S. Air Force and Army SIGINT platforms, as well as commercial infrastructure sectors where spectrum awareness and wireless threat detection are paramount.ÝBy combining robust unsupervised learning with an explainable, operator-in-the-loop interface, DTJís innovation will equip the Navy with a cognitive RF sensing capability capable of uncovering adversarial behavior that traditional systems cannot perceiveóturning uncertainty in the spectrum into actionable awareness.