Distribution of the non-colocated Time-Reversal MUSIC null spectrum under realistic array, medium, SNR, and data models

Time-Reversal MUSIC (TR-MUSIC) localizes point-like scatterers from the noise subspace of the multistatic data matrix measured between separate transmit and receive arrays. Ciuonzo and Salvo Rossi (IEEE Signal Processing Letters, 2017) derived the asymptotic (high-SNR) distribution of the non-colocated TR-MUSIC null spectrum: via first-order perturbation of the singular value decomposition (SVD), the null-spectrum values at the true scatterer locations are squared norms of a complex Gaussian vector with a computable covariance, obey chi-square-type laws, and yield closed-form null-spectrum-deviation (NSD) stability measures for the transmit-mode, receive-mode, and generalized imaging variants. That characterization is explicitly restricted to a homogeneous background, negligible mutual coupling, ideal antenna patterns and polarization, single-frequency (narrowband) data, a single snapshot, and first-order perturbation terms. The open problem is to extend this distribution-level statistical theory to the realistic conditions the 2017 analysis excluded: mutual coupling and realistic antenna patterns/polarization (named as future work in the source), inhomogeneous or random propagation media (named as future work in the source), and — as faithful derived extensions removing stated scope restrictions — non-asymptotic validity at moderate or low SNR and broadband/multi-frequency or multi-snapshot data. No published work extends these distribution laws to date; the thread is open.

Background

Time reversal (TR) is a focusing technique in which a recorded wave field is re-emitted time-reversed so that the medium refocuses the energy onto its origin; in computational TR the refocusing is done numerically by back-propagating measured multistatic data with the background Green's function. A multistatic data matrix (MDM) collects the responses of all transmit–receive antenna pairs at a single frequency. Its singular value decomposition (SVD) underlies DORT (Décomposition de l'Opérateur de Retournement Temporel, the decomposition of the time-reversal operator), which detects and selectively focuses on scatterers through the dominant singular vectors (Prada, Manneville, Spoliansky, and Fink, 1996). TR-MUSIC (Devaney, 2005) combines this with the MUSIC (MUltiple SIgnal Classification) principle from array signal processing: for point-like scatterers, the background Green's function vectors at true target locations are orthogonal to the noise subspace (the singular-vector subspace associated with the smallest, noise-dominated singular values), so a null-spectrum function — the squared norm of the projection of the Green's function vector onto the noise subspace — vanishes sharply at the scatterers, while its reciprocal serves as a pseudo-spectrum whose local maxima mark the scatterers for localization. In the non-colocated (multistatic) setting the transmit and receive arrays are distinct, giving left and right noise subspaces of the MDM and hence three TR-MUSIC variants: transmit (Tx) mode, receive (Rx) mode, and a generalized variant combining both subspaces. Scattering is modeled by the linear Born approximation (BA) or the nonlinear Foldy–Lax (FL) multiple-scattering model.

A classical literature studies the statistical performance of MUSIC for direction-of-arrival (DOA) estimation, but those results assume stochastic sources, many independent snapshots, and an asymptotic regime in the number of snapshots, so they cannot be transferred to TR-MUSIC, where scatterers are deterministic and only a single snapshot of the MDM is available. For TR-MUSIC, Ciuonzo, Romano, and Solimene (IEEE Transactions on Signal Processing, 2015) derived the high-SNR localization mean-squared error from first-order SVD perturbation; Ciuonzo and Salvo Rossi (Digital Signal Processing, 2017) derived the asymptotic distribution of the TR-MUSIC null spectrum for the colocated (monostatic) case; and Ciuonzo and Salvo Rossi (IEEE Signal Processing Letters, 2017) — the source of this problem — settled the non-colocated case. In the 2017 letter the key object is a perturbation vector $\boldsymbol{\xi}_k$ collecting the projections that determine the Tx- and Rx-mode null-spectrum values at a true scatterer location $\boldsymbol{x}_k$: since measurement noise enters the MDM through an additive Gaussian matrix $\mathbf{W}$, a first-order (high-SNR) SVD perturbation expansion renders $\boldsymbol{\xi}_k$ linear in $\mathbf{W}$, hence circularly symmetric complex Gaussian, $\boldsymbol{\xi}_k\sim\mathcal{CN}(\mathbf{0},\boldsymbol{\Xi}_k)$, with a covariance matrix $\boldsymbol{\Xi}_k$ computable in closed form. The null-spectrum values at the scatterer locations are quadratic forms in $\boldsymbol{\xi}_k$ and follow (complex) chi-square-type laws, so their stability can be summarized by the NSD, the normalized standard deviation $\mathrm{NSD}_k=\sqrt{\mathrm{var}\{P(\boldsymbol{x}_k)\}/\mathbb{E}\{P(\boldsymbol{x}_k)\}}$: Tx and Rx modes give constants $1/\sqrt{N_{T\mathrm{dof}}}$ and $1/\sqrt{N_{R\mathrm{dof}}}$ (the noise-subspace dimensions), while the generalized variant's NSD depends on the scatterer and measurement geometry, which is the letter's main structural finding. Monte Carlo validation showed the asymptotic expressions are accurate above roughly 10 dB SNR.

The derivation is explicitly confined to an idealized envelope: a homogeneous propagation background, no mutual coupling between array elements, ideal isotropic antenna patterns without polarization effects, narrowband single-frequency data, a single snapshot, and only first-order perturbation terms. The letter's conclusions state verbatim: "Future works will analyze mutual coupling, antenna pattern and polarization effects, and propagation in inhomogeneous (random) media." The single-frequency, single-snapshot, and high-SNR restrictions are stated modeling assumptions of the 2017 analysis.

Problem Statement

Extend the asymptotic statistical characterization of the non-colocated Time-Reversal MUSIC null spectrum given by Ciuonzo and Salvo Rossi (2017) beyond its idealized measurement envelope. The shared scientific objective is a distribution-level theory of the same object — the laws of the null-spectrum values at the true scatterer locations for the Tx-mode, Rx-mode, and generalized TR-MUSIC variants, their covariance structure (the analogue of $\boldsymbol{\Xi}_k$), and closed-form stability measures such as the NSD — under realistic conditions, with each extension required to reduce to the known high-SNR, homogeneous, ideal-antenna, single-frequency, single-snapshot laws in the corresponding limit. Concretely: (a) incorporate mutual coupling between array elements and realistic (non-isotropic) antenna patterns with polarization into the multistatic data model and derive the resulting modified perturbation-vector distribution, null-spectrum laws, and NSD expressions (explicit future work in the source); (b) quantify how inhomogeneous or random propagation media — deterministic inhomogeneities or statistical medium fluctuations — modify the covariance $\boldsymbol{\Xi}_k$, the chi-square-type laws, and the NSD expressions (explicit future work in the source); and, as derived extensions removing the analysis's stated scope restrictions, (c) establish the accuracy of the first-order approximation and the necessary corrections at moderate or low SNR; and (d) generalize the single-frequency, single-snapshot results to broadband or multi-frequency measurements and to multiple independent snapshots. Each requested result is a distributional statement about the same null-spectrum functional under an enlarged model, jointly serving one objective: replacing the 2017 idealized-envelope theory with a statistical characterization valid under the measurement conditions actually encountered in TR-MUSIC imaging systems. Each of (a)–(d) is an independently valid extension of this shared objective: a solution that fully establishes the distribution-level theory under any one of the enlarged models (a)–(d) resolves the task, and the directions left unaddressed remain open extensions of the same objective.

References

1. D. Ciuonzo, P. Salvo Rossi, "Noncolocated Time-Reversal MUSIC: High-SNR Distribution of Null Spectrum", IEEE Signal Processing Letters 24(4):397–401, 2017. DOI: 10.1109/LSP.2017.2661246; arXiv:1701.07516, https://arxiv.org/abs/1701.07516
2. D. Ciuonzo, G. Romano, R. Solimene, "Performance Analysis of Time-Reversal MUSIC", IEEE Transactions on Signal Processing 63(10):2650–2662, 2015. DOI: 10.1109/TSP.2015.2417507, https://doi.org/10.1109/TSP.2015.2417507
3. D. Ciuonzo, P. Salvo Rossi, "On the asymptotic distribution of Time-Reversal MUSIC null spectrum", Digital Signal Processing 69:50–58, 2017. DOI: 10.1016/j.dsp.2017.06.011, https://doi.org/10.1016/j.dsp.2017.06.011
4. D. Ciuonzo, "On Time-Reversal Imaging by Statistical Testing", IEEE Signal Processing Letters 24(7):1024–1028, 2017. DOI: 10.1109/LSP.2017.2704612; arXiv:1705.02705, https://arxiv.org/abs/1705.02705
5. A. J. Devaney, "Time reversal imaging of obscured targets from multistatic data", IEEE Transactions on Antennas and Propagation 53(5):1600–1610, 2005. DOI: 10.1109/TAP.2005.846723, https://doi.org/10.1109/TAP.2005.846723
6. C. Prada, S. Manneville, D. Spoliansky, M. Fink, "Decomposition of the time reversal operator: Detection and selective focusing on two scatterers", The Journal of the Acoustical Society of America 99(4):2067–2076, 1996. DOI: 10.1121/1.415393, https://doi.org/10.1121/1.415393
7. B. Hu, X. Cao, L. Zhang, Z. Song, "Weighted Space-Frequency Time-Reversal Imaging for Multiple Targets", IEEE Signal Processing Letters 26(6):858–862, 2019. DOI: 10.1109/LSP.2019.2910929, https://doi.org/10.1109/LSP.2019.2910929
8. Z.-H. Cheng, M.-L. Ma, F. Liang, D. Zhao, B.-Z. Wang, "Low Complexity Time Reversal Imaging Methods Based on Truncated Time Reversal Operator", IEEE Transactions on Geoscience and Remote Sensing 62:1–14, 2024. DOI: 10.1109/TGRS.2023.3347624, https://doi.org/10.1109/TGRS.2023.3347624
9. Y. J. Song, H. S. Park, B. K. Ahn, S. K. Hong, "Effective Buried Nonlinear Target Detection Under Faulty Antenna Conditions Using Extended DORT Approach", IEEE Transactions on Instrumentation and Measurement 74:1–13, 2025. DOI: 10.1109/TIM.2025.3556916, https://doi.org/10.1109/TIM.2025.3556916
10. F. Foroozan, P. Sadeghi, "Super-resolution ultrawideband ultrasound imaging using focused frequency time reversal MUSIC", IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2015, pp. 887–891. DOI: 10.1109/ICASSP.2015.7178097, https://doi.org/10.1109/ICASSP.2015.7178097
11. C. Fan, L. Yang, Y. Zhao, "Ultrasonic multi-frequency time-reversal-based imaging of extended targets", NDT & E International 113:102276, 2020. DOI: 10.1016/j.ndteint.2020.102276, https://doi.org/10.1016/j.ndteint.2020.102276