TSD-DOA-Benchmark-v1: A 1.5M-Frame Public Benchmark for RF Direction-of-Arrival Estimation

TSD-DOA-Benchmark-v1 is a public RF direction-of-arrival benchmark datasetcontaining 1,518,000 labeled IQ frames generated using the Stoica-Nehorainarrowband ULA signal model. The dataset spans 11 array sizes (N=4 to N=64),46 SNR points (-15 to +30 dB in 1 dB steps), and 1-3 simultaneous emitters,with 1,000 Monte Carlo trials per configuration. Three algorithms are benchmarked head-to-head on identical data: the TSDengine, MUSIC (Schmidt 1986), and ESPRIT (Roy and Kailath 1989). At theprimary operating point (N=16, SNR=+10 dB, K=100, 1 source), TSD achieves39% lower RMSE than MUSIC and 44% lower RMSE than ESPRIT while running atESPRIT-class throughput (0.21 ms per frame). The TSD advantage scales witharray size, reaching +90% over MUSIC and +69% over ESPRIT at N=64. All results are bit-for-bit reproducible from master seed 20260424. Theincluded README documents the data format, results CSV schema, statisticalmethodology, and step-by-step verification procedure for independentreimplementation of the baselines. License: CC-BY-4.0