Cramér-Rao Bound Analysis and Near-Optimal Performance of the Synchronous Nyquist-Folding Generalized Eigenvalue Method (SNGEM) for Sub-Nyquist Multi-Tone Parameter Estimation

The synchronous Nyquist folding generalized eigenvalue method (SNGEM) realizes full frequency/amplitude/phase estimation of multitone signals at extreme sub-Nyquist rates by jointly processing the original signals and their time derivatives. In this paper, accurate Cramer-Rao bounds for amplitude ratio parameter R=A/B=1/(2\pif) are derived for two channels with equal SNR. Monte-Carlo simulations confirm that SNGEM achieves machine accuracy in noise-free conditions and closely approaches the derived CRB at all SNR levels, even at 10- 20x compression, whereas classical compressive sensing OMP exhibits irreducible error flattening due to DFT grid bias and aliasing noise. These results establish SNGEM as a statistically nearly optimal deterministic sub-Nyquist parameter spectrum analysis

💡 Research Summary

The paper presents a rigorous statistical analysis of the Synchronous Nyquist‑Folding Generalized Eigenvalue Method (SNGEM), a deterministic sub‑Nyquist technique that jointly samples a wide‑band multi‑tone signal and its discrete time‑derivative. By exploiting the exact relationship B = 2πf A between the amplitudes of the original and differential channels, the authors derive closed‑form Cramér‑Rao bounds (CRBs) for the amplitude‑ratio parameter R = A/B = 1/(2πf) under the realistic assumption of equal‑SNR additive white Gaussian noise in both channels. Starting from the classical Fisher information for each amplitude estimate (Var(Â) ≥ 2σ²/N, Var( B̂ ) ≥ 2σ²/N), they propagate uncertainty through the ratio to obtain Var(R̂)/R² ≥ 8/(N·SNR), which translates to a frequency CRB Var(f̂)/f² ≥ 2/(N·SNR). This result shows that frequency estimation via the amplitude ratio incurs only a 3 dB penalty relative to the optimal single‑channel deterministic bound, a surprisingly small loss given the aggressive compression.

Monte‑Carlo experiments with 5–15 random tones, random phases, and compression ratios ranging from 8× to 20× confirm the theoretical predictions. In the noise‑free regime SNGEM reaches machine‑precision accuracy (RMSE < 10⁻¹⁴) for frequency, amplitude, and phase, demonstrating that the method is numerically stable even at extreme undersampling. In noisy conditions, across SNRs from –10 dB to 50 dB, the empirical RMSE follows the derived dual‑channel CRB within a factor of 1.1–1.3, indicating near‑optimal statistical efficiency. By contrast, a conventional compressive‑sensing approach using Orthogonal Matching Pursuit (OMP) on the same measurements exhibits an error floor (≈10⁻⁵–10⁻⁴) that does not improve with SNR, reflecting inherent grid‑bias and aliasing‑noise effects.

The authors emphasize that SNGEM requires no random measurement matrix design; only synchronized sampling of the signal and its derivative is needed, which can be realized with modest hardware modifications (e.g., an analog differentiator or digital finite‑difference). This deterministic nature eliminates calibration complexities and makes the method attractive for real‑time wide‑band applications such as radar, next‑generation 6G communications, and high‑resolution instrumentation where ADC bandwidth is a bottleneck. The 3 dB theoretical penalty is offset by the substantial reduction in sampling rate (10–20× compression) while preserving full parameter fidelity.

In conclusion, the paper establishes that SNGEM achieves statistically near‑optimal performance for deterministic sub‑Nyquist parameter estimation, outperforming classical compressive‑sensing techniques both in accuracy and robustness. The derived CRBs provide a solid benchmark for future algorithmic developments, and the demonstrated practical feasibility suggests a clear path toward hardware implementation and extension to multi‑channel, non‑linear, or real‑time scenarios.