A General Ziv–Zakai Bound for DoA Estimation in MIMO Radar Systems

Mohammadreza Bakhshizadeh Mohajer ∗ , Daniela Tuninetti † , and Luca Barletta ∗ · 2025

Read A General Ziv–Zakai Bound for DoA Estimation in MIMO Radar Systems by Mohammadreza Bakhshizadeh Mohajer ∗ , Daniela Tuninetti † , and Luca Barletta ∗ (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-08326.txt.

Introduction

Mohammadreza Bakhshizadeh Mohajer, Daniela Tuninetti, and Luca Barletta derive a general Ziv Zakai bound for direction of arrival estimation in co located MIMO radar. Such bounds limit the best possible mean squared error when locating targets, and older results cover simpler antenna setups with Gaussian inputs. The new bound handles multiple targets, a general transmit covariance, radar cross section statistics, and multiple snapshots in one closed form.

The formula shows plainly how error depends on transmit antenna count, target count, and signal to noise ratio. Simulations confirm the bound is tight where prior information dominates and tighter than the Cramer Rao bound at low signal levels. More antennas compress the threshold for entering the high accuracy regime.

Catalog record

CREATORMohammadreza Bakhshizadeh Mohajer ∗ , Daniela Tuninetti † , and Luca Barletta ∗
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-08326

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A General Ziv–Zakai Bound for DoA Estimation in MIMO Radar Systems Mohammadreza Bakhshizadeh Mohajer ∗ , Daniela Tuninetti † , and Luca Barletta ∗ Abstract This paper derives a Ziv–Zakai Bound (ZZB) on the Mean Squared Error (MSE) for Direction-of-Arrival (DoA) estimation in co-located Multiple-Input Multiple-Output (MIMO) radar systems and provides closed-form expressions that hold for multi-target scenarios. Unlike classical results that address single-input multiple-output systems with complex Gaussian input signals, the developed ZZB in this paper explicitly accounts for a general input covariance matrix, target radar cross-section statistics and multiple snapshot effects, and admits a compact expression that reveals the dependence of the MSE on the number of transmit antennas, number of targets, Signal-to-Noise Ratio (SNR) and the transmit covariance matrix. Numerical simulations validate the tightness of the ZZB in the a priori dominated region and show how the increase of the number of transmit antennas compresses the threshold SNR for the transition to the Cramér-Rao bound (CRB) while the variation of the number of targets shifts the bound’s behavior across SNR regimes. The analytical results and numerical simulations demonstrate that the ZZB is tighter than the CRB, particularly in the low SNR regime. I Introduction Multiple‑Input Multiple‑Output (MIMO) radar has attracted considerable interest for modern sensing applications because, by transmitting different waveforms from each antenna rather than scaled copies of the same signal, it provides increased degrees of freedom and greater operational flexibility [ 1 ] . These additional degrees of freedom directly benefit core estimation tasks such as Direction‑of‑Arrival (DoA) estimation, which is central to radar, sonar, and wireless sensing systems and determines spatial resolution and interference resilience [ 2 ] . At the same time, next‑generation wireless networks, such as 6G, aim to combine high‑fidelity connectivity with accurate, resilient sensing. Integrated Sensing and Communication (ISAC) meets this need by unifying sensing and communications on a common hardware platform and waveform, allowing systems to exploit MIMO’s waveform diversity to improve DoA estimation while reducing form factor and power consumption [ 3 ] . Mean Square Error (MSE) is the standard metric for assessing estimator performance, but for nonlinear parameter estimation problems such as DoA estimation there is no closed‑form expression for the minimum achievable MSE [ 2 ] . This absence motivates the use of lower bounds on the MSE to characterize estimator performance. The Cramér–Rao Bound (CRB), derived from the inverse Fisher information, is the most widely used lower bound; however, because it characterizes only small estimation errors, it is a local bound and can be ineffective for assessing estimator performance in non‑asymptotic regimes [ 4 ] . To capture the a priori performance region at low SNR, a variety of Bayesian lower bounds have been developed over the past decades, including rate‑distortion–based bounds [ 5 ] , bounds derived from functional inequalities such as log‑Sobolev inequalities [ 6 , 7 ] , and bounds built on variational representations of divergence measures [ 8 ] ; in this work we adopt the Ziv–Zakai Bound (ZZB) [ 9 , 10 ] as a global MSE lower bound. The scalar ZZB was introduced in [ 9 ] , and extensions to vector parameters were developed in [ 11 ] . The vector ZZB was applied to 2‑D DoA for the single target case in [ 12 ] , while [ 2 ] generalized the ZZB from 1‑D DoA single‑target to multi‑target scenarios by using order statistics and by ordering both estimated and true DoAs to resolve permutation ambiguity. These prior works, however, focus on Single-Input Multiple-Output (SIMO) architectures and assume zero‑mean complex white Gaussian transmit signals, which limits their applicability for analyzing the sensing–communication trade‑off in ISAC systems. In this paper we extend the ZZB to co‑located MIMO radar and derive a closed‑form expression for transmitted waveforms with a general covariance matrix. Contributions: The main contributions of this paper are: • Derivation of a closed‑form ZZB for multi‑target DoA estimation in co‑located MIMO radar systems. • Analytical treatment for a general transmit covariance matrix that captures the most general case, including ISAC waveforms. • Insights into the operating regimes where the CRB becomes invalid and the ZZB provides a tight performance prediction. Paper Organization: The remainder of the paper is organized as follows. Section II presents the system model, MSE, and CRB. Section III summarizes the preliminaries on the ZZB. Section IV provides a closed‑form ZZB expression for transmit waveforms characterized by a general input covariance matrix. Section V presents numerical results, and Section VI concludes the paper. Notation: Scalars are denoted by italic letters (e.g., \theta,\lambda ), while vectors and matrices are denoted by bold lowercase and uppercase letters (e.g., \mathbf{c},\mathbf{X} ), respectively. The transpose, elementwise complex conjugate, Hermitian transpose, trace, and determinant of a matrix \mathbf{A} are denoted by \mathbf{A}^{\mathsf{T}} , \mathbf{A}^{*} , \mathbf{A}^{\dagger} , {\text{tr}}\left(\mathbf{A}\right) , and |\mathbf{A}| , respectively. The Euclidean norm of a vector \mathbf{a} is \|\mathbf{a}\|_{2} , and the Frobenius norm of a matrix \mathbf{A} is \|\mathbf{A}\|_{F} . The Hadamard product is denoted by \odot . The expectation operator is \mathbb{E}[\cdot] . f_{\theta}(\phi) denotes the probability density function (pdf) of the random variable \theta evaluated at \phi . \mathbb{C} is the set of complex numbers. The distribution of a proper complex multivariate normal random vector with mean \bm{\mu} and covariance \bm{\Sigma} is denoted by \mathcal{CN}(\bm{\mu},\bm{\Sigma}) . The identity matrix is \mathbf{I} , while {\bf 1}_{K} denotes the all-ones column vector of length K . II System Model We consider a co-located MIMO radar with N transmit and M receive antennas observing K point targets with DoA \mathbold{\theta}=[\theta_{1},\ldots,\theta_{K}]^{\mathsf{T}} . The received signal is expressed as \displaystyle\mathbf{X} \displaystyle=\sum_{k=1}^{K}\mathbf{a}(\theta_{k})b_{k}\mathbf{v}^{\mathsf{T}}(\theta_{k}){\bf\Phi}+\mathbf{Z}, (1) \displaystyle=\mathbf{A}(\mathbold{\theta})\,\mathbf{B}\,\mathbf{V}^{\mathsf{T}}(\mathbold{\theta}){\bf\Phi}+\mathbf{Z}, (2) where \mathbf{X}\in\mathbb{C}^{M\times L} with L being the number of snapshots (or channel uses), \mathbf{A}(\mathbold{\theta})=[\mathbf{a}(\theta_{1}),\dots,\mathbf{a}(\theta_{K})] and \mathbf{V}(\mathbold{\theta})=[\mathbf{v}(\theta_{1}),\dots,\mathbf{v}(\theta_{K})] denote the receive and transmit steering matrices with \displaystyle\mathbf{a}(\theta_{k}) \displaystyle=\left[e^{-j\frac{2\pi}{\lambda}d_{1}\sin{\theta_{k}}},\dots,e^{-j\frac{2\pi}{\lambda}d_{M}\sin{\theta_{k}}}\right]^{\mathsf{T}}\in\mathbb{C}^{M\times 1}, (3) \displaystyle\mathbf{v}(\theta_{k}) \displaystyle=\left[e^{-j\frac{2\pi}{\lambda}d_{1}\sin{\theta_{k}}},\dots,e^{-j\frac{2\pi}{\lambda}d_{N}\sin{\theta_{k}}}\right]^{\mathsf{T}}\in\mathbb{C}^{N\times 1}, (4) where the d_{i} ’s are the antenna positions in the linear array, and \mathbf{B}=\text{diag}(\mathbf{b}) is a diagonal matrix of target reflection coefficients \mathbf{b}=[b_{1},\dots,b_{K}] that are assumed to be independent and identically distributed (i.i.d.) complex random variables with finite second moment, i.e., \mathbb{E}[|b_{k}|^{2}]<\infty . The rows of {\bf\Phi}\in\mathbb{C}^{N\times L} are the transmitted waveforms in time with identically distributed columns, and \mathbf{Z}\in\mathbb{C}^{M\times L} is Additive White Gaussian Noise (AWGN) with i.i.d. \sim\mathcal{C}\mathcal{N}(0,\sigma_{n}^{2}) entries. III CRB and ZZB Preliminaries Permutation ambiguity denotes the invariance of the channel likelihood f_{\mathbf{X}\mid\mathbold{\theta}} to any relabeling of the DoA vector \mathbold{\theta} : the measurements depend on the set of source directions, not on their ordering. Consequently, practical estimators impose a canonical labeling (e.g., sorting the estimates in ascending order), which maps the parameters to order statistics thus modifying the joint prior pdf f{\theta} . Therefore, the MSE should be computed on the ordered parameters as [ 2 ] \displaystyle\text{MSE}=\frac{1}{K}\sum^{K}_{k=1}\mathbb{E}\left[(\hat{\theta}_{(k)}-\theta_{(k)})^{2}\right], (5) where \hat{\theta}_{(k)} and \theta_{(k)} are the k -th smallest order statistics in \hat{\mathbold{\theta}} and \mathbold{\theta} , respectively. A widely used lower bound on the estimation error in DoA estimation is the CRB, which equals the inverse of the Fisher information matrix. For a random vector \mathbold{\theta} , a lower bound on the MSE is given by the expected CRB as \displaystyle\frac{1}{K}\sum^{K}_{k=1}\mathbb{E}\left[(\hat{\theta}_{k}-\theta_{k})^{2}\right] \displaystyle\geq\frac{1}{K}{\text{tr}}\left(\mathbb{E}_{\mathbold{\theta}}[\text{CRB}(\mathbold{\theta})]\right) (6) \displaystyle=\frac{1}{K}{\text{tr}}\left(\mathbb{E}_{\mathbold{\theta}}[\mathbf{J}^{-1}(\mathbold{\theta})]\right), (7) where the ij -th entry of the Fisher information matrix \mathbf{J}(\mathbold{\theta}) can be

Read the complete text