Toeplitz Inverse Eigenvalue Problem (ToIEP)

• The Toeplitz Inverse Eigenvalue Problem (ToIEP) is a matrix analysis challenge that determines the existence of a real or complex Toeplitz matrix given a set of eigenvalues and prescribed lag moduli.
• It leverages sign assignment in the real-symmetric case and Newton's method for the Hermitian case to efficiently solve for phase and modulus constraints in array calibration.
• Applications span high-precision direction-of-arrival estimation and adaptive beamforming, with recent methods achieving robust phase RMSE improvements under statistical convergence conditions.

The Toeplitz Inverse Eigenvalue Problem (ToIEP) is a foundational inverse problem in matrix analysis, signal processing, and array calibration. ToIEP asks: for a given set of eigenvalues and (optionally) additional structural constraints (such as prescribed moduli of elements), can one find a real or complex-valued symmetric Toeplitz matrix realizing those eigenvalues while satisfying the constraints? This problem is central in the context of uniform linear array (ULA) processing, where calibration and structure-exploiting reconstructions of covariance matrices underpin high-precision direction-of-arrival estimation and adaptive beamforming. The ToIEP captures the unique algebraic structure of Toeplitz matrices—constant along diagonals—and its solution space is tightly governed by group-theoretic and spectral properties. Recent advances include complete resolution for the real-symmetric case, algorithmic frameworks for the Hermitian (complex) case, and concrete application to ULA phase calibration (Abramovich et al., 2023).

1. Formal Statement of the Toeplitz Inverse Eigenvalue Problem

The classical real-symmetric form of ToIEP is defined as: Given a set of $N$ distinct real eigenvalues $\{\lambda_1,\dots,\lambda_N\}$ and the absolute values $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$ of the first row of a symmetric Toeplitz matrix $T$, find the vector $v = [t_0, t_1, ..., t_{N-1}]^T$ such that the $(i,j)$th entry of $T$ satisfies $T_{ij} = t_{|i-j|}$, $T$ is symmetric Toeplitz, and $\text{spec}(T) = \{\lambda_k\}_{k=1}^N$ with $\{\lambda_1,\dots,\lambda_N\}$0 as prescribed.

For the Hermitian (complex) case, ToIEP generalizes to seeking a Hermitian Toeplitz $\{\lambda_1,\dots,\lambda_N\}$1 with given eigenvalues and $\{\lambda_1,\dots,\lambda_N\}$2 matching observed statistics, relevant to phase calibration in array processing under unknown per-element phases.

2. Uniqueness, Symmetry, and Solution Structure

In the real-symmetric case, the set of feasible solutions is sharply characterized: for a prescribed set of eigenvalues and lag moduli, there exist exactly two isomorphic solutions for the sign pattern $\{\lambda_1,\dots,\lambda_N\}$3. These correspond to the original Toeplitz matrix $\{\lambda_1,\dots,\lambda_N\}$4 and its "flipped" version $\{\lambda_1,\dots,\lambda_N\}$5, where $\{\lambda_1,\dots,\lambda_N\}$6 is the backward-identity. Only these two patterns preserve symmetry and lead to Toeplitz matrices with the desired spectrum (Abramovich et al., 2023).

The physical relevance of each solution is determined by examining the associated spectral symmetry. In array calibration, the matrix whose spatial spectrum is centered about broadside is usually selected, while the other exhibits non-physical, shifted spectral properties.

For the complex Hermitian case, the solution set becomes substantially richer. The moduli and spectrum alone do not confine possible solutions to the set $\{\lambda_1,\dots,\lambda_N\}$7 (with $\{\lambda_1,\dots,\lambda_N\}$8 diagonal), leading to non-uniqueness and the need for additional constraints such as enforcing that the element-wise ratio $\{\lambda_1,\dots,\lambda_N\}$9 is rank-one with constant-modulus eigenvector.

3. Computational Algorithms

3.1 Real-Valued Case via Sign Assignments

Given lag magnitudes $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$0, assign signs $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$1 and minimize the spectral error $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$2. A greedy coordinate-descent (sign-flip) algorithm efficiently searches the sign space, converging in $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$3 steps. Only two global minima exist, which can be disambiguated by spectral symmetry tests. This renders the real-symmetric ToIEP a computationally tractable problem for practical array dimensions.

3.2 Hermitian Case via Newton's Method and Rank-One Alignment

Unknown phases $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$4 parameterize the ambiguity in $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$5. Define the modulus-matrix $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$6 from observed data and solve for $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$7 to ensure $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$8 is rank-one. Newton's method iteratively updates $\{|t_0|,|t_1|,\ldots,|t_{N-1}|\}$9 using the Jacobian of the eigenvalue-based objective $T$0 (for $T$1), with careful step-size control. Alternating projections enforce the modulus and eigenvalue constraints to high numerical precision, guaranteeing convergence to a physically plausible solution family.

3.3 Statistical Convergence and Sample Complexity

Solution accuracy is conditioned on the accuracy of the sample covariance $T$2, not merely its leading eigenstructure but its element-wise fidelity. Under mild spectral regularity conditions, the ML Toeplitz estimator converges as $T$3. Empirical results indicate that achieving phase RMSE within $T$4 requires $T$5 to $T$6 times the array size ($T$7) in snapshot data.

4. Applications: ULA and Minimum Redundancy Array Calibration

The ToIEP is directly applied to calibration of uniform linear arrays and minimum redundancy arrays (MRAs), exploiting array covariance structure. For a ULA, starting from a sample covariance estimate, one extracts lag moduli, solves the ToIEP (real or complex as appropriate), selects the physical solution, and derives per-element phase errors by comparing observed and reconstructed entries. In MRAs, uncalibrated MRA data and the MRA-to-ULA selection matrix enable reconstruction of the full ULA covariance, provided the missing Toeplitz eigenvalues are set to noise-floor values and the real-valued ToIEP is solved.

The procedure provides a purely data-driven calibration route, without explicit knowledge of source signals or injection of calibration tones, requiring only sufficient sample support.

5. Key Equations and Theoretical Underpinnings

Key Operation	Mathematical Formulation	Context
Toeplitz construction	$T$8	Defining the matrix
Eigenvalue constraint	$T$9, $v = [t_0, t_1, ..., t_{N-1}]^T$0	Imposing spectrum
Real sign-assignment	$v = [t_0, t_1, ..., t_{N-1}]^T$1, $v = [t_0, t_1, ..., t_{N-1}]^T$2	Real ToIEP solution
Hermitian rank-one ratio	$v = [t_0, t_1, ..., t_{N-1}]^T$3, enforce $v = [t_0, t_1, ..., t_{N-1}]^T$4	Removing ambiguities
Newton update (complex case)	$v = [t_0, t_1, ..., t_{N-1}]^T$5	Solving for phases
MRA extension	$v = [t_0, t_1, ..., t_{N-1}]^T$6	Array selection mapping

The group-theoretic result underlying the real case—the stability group of Toeplitz matrices—restricts the number of sign patterns producing feasible solutions. For the Hermitian case, spectral and modulus constraints admit a rank-one ratio condition to extract a physical calibration.

6. Practical Usage: Step-by-Step ULA Calibration Protocol

A standard ULA calibration via ToIEP proceeds as follows:

1. Form a Toeplitz modulus estimate from the observed sample covariance.
2. Solve the real ToIEP (section 3.1) if the spatial spectrum is real-symmetric, or the complex case (3.2) otherwise.
3. Recover the calibrated Toeplitz matrix and select the solution yielding a broadside-symmetric maximum-entropy spectrum.
4. Estimate array phase errors as the argument differences between observed and reconstructed off-diagonal entries.
5. Optionally, remove global phase bias based on known reference sources.

Achieving robust calibration with phase RMS below $v = [t_0, t_1, ..., t_{N-1}]^T$7 typically requires $v = [t_0, t_1, ..., t_{N-1}]^T$8–$v = [t_0, t_1, ..., t_{N-1}]^T$9 independent samples, achievable under standard system constraints (Abramovich et al., 2023).

7. Extensions and Open Directions

Recent works have extended ToIEP methodology to MRAs and other non-standard array geometries, highlighting ToIEP's flexibility in structure-exploiting signal processing. Open research includes characterizing the uniqueness set in higher-dimensional or non-Toeplitz structured matrices, advancing algorithms for large-scale instances, and exploiting ToIEP in robust and adaptive beamforming.

The ToIEP thus provides a principled, algebraic backbone for calibration and inverse spectral problems in array processing, with algorithmic and theoretical frameworks now firmly established for both real and complex-valued cases, and practical protocols with theoretically justified statistical guarantees for convergence and accuracy (Abramovich et al., 2023).