Adaptive Subspace Signal Detection and Performance Analysis in Nonzero-Mean Clutter

Weijian Liu, Senior Member, IEEE, Zhenyu Xu, Jun Liu, Senior Member, IEEE, Hui Chen, and · 2026

Read Adaptive Subspace Signal Detection and Performance Analysis in Nonzero-Mean Clutter by Weijian Liu, Senior Member, IEEE, Zhenyu Xu, Jun Liu, Senior Member, IEEE, Hui Chen, and (2026) 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-2605-07434.txt.

Introduction

Weijian Liu and four co-authors design radar detectors for signals hiding in clutter with a nonzero average. Detecting subspace signals against such background is a practical radar problem. The authors build adaptive detectors from five classical test strategies, including likelihood ratio and Wald tests. Several turn out structurally identical to known zero-mean detectors.

Closed-form expressions for detection and false-alarm probabilities are derived for each detector. Two costs of nonzero-mean clutter emerge: one lost degree of freedom, and a loss in signal-to-clutter ratio. Simulations and measured data confirm the detectors work, supporting their value in real radar systems.

Catalog record

CREATORWeijian Liu, Senior Member, IEEE, Zhenyu Xu, Jun Liu, Senior Member, IEEE, Hui Chen, and
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE95 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.07434

Browse the full library · About the Tower of Babel

Full text

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

Adaptive Subspace Signal Detection and Performance Analysis in Nonzero-Mean Clutter Thanks: This work was supported by National Natural Science Foundation of China (Grant Nos. 62471485 and 62471450), Natural Science Foundation of Hubei Province (Grant No. 2025AFB873). Thanks: W. Liu, Z. Xu, and H. Chen are with Wuhan Electronic Information Institute, Wuhan 430019, China (e-mails: [email protected], [email protected], and [email protected]). Thanks: J. Liu is with Department of Electronic Engineering and Information Science, University of Science and Technology of China, Hefei, 230027, China (e-mail: [email protected]). Thanks: Y. Liu is with the College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China (e-mail: [email protected]). Thanks: Digital Object Identifier 10.1109/TSP.2026.3692130 Weijian Liu, Senior Member, IEEE, Zhenyu Xu, Jun Liu, Senior Member, IEEE, Hui Chen, and Abstract To solve the problem of detecting subspace signals in nonzero-mean clutter, we propose adaptive detectors, based on the strategies of generalized likelihood ratio test (GLRT), Rao test, Wald test, gradient test, and Durbin test. The results show that the detectors based on GLRT, Rao and Wald are structurally consistent with the subspace detectors in zero-means clutter. The analytic expressions for the probability of detection (PD) and probability of false alarm (PFA) of each detector are derived, and two major performance differences in the nonzero-mean clutter scenario are revealed. One is the loss of degree of freedom (DOF), which is reduced by 1 compared with the zero-mean clutter scenario. The second is the loss of signal-to-clutter (SCR) ratio. Simulation and measured data verify the effectiveness of the proposed detectors and demonstrate their practical value in real-world radar systems. I introduction Multichannel adaptive signal detection plays an irreplaceable role in radar, sonar, wireless communication and other fields [ 1 ] . Since Kelly proposed the first adaptive detector based on the generalized likelihood ratio test (GLRT) in 1986, denoted as Kelly’s GLRT (KGLRT) [ 2 ] , adaptive detection under zero-mean clutter has formed a relatively complete theoretical system [ 3 , 4 , 5 , 6 ] . In particular, based on the KGLRT, the adaptive matched filter (AMF) [ 7 ] , and De Maio’s Rao (DMRao) [ 8 ] , adaptive coherence estimator (ACE) [ 9 ] are constantly proposed. Nowadays, adaptive detection technology has been used in advanced radars such as frequency diverse array (FDA) radar [ 10 , 11 , 12 ] and multiple-input multiple-output (MIMO) radar [ 13 , 14 ] . The aforementioned detectors are primarily for rank-one signals, which are characterized by well-defined steering vectors. In practice, however, signals may occupy a given subspace rather than a single direction. Such signals are referred to as subspace signals. Notably, a rank-one signal represents a special case of a subspace signal when the subspace dimension equals one. The subspace signal model offers broader applicability. For instance, echoes from helicopter targets [ 15 ] and polarimetric radar returns [ 16 ] can both be effectively modeled as subspace signals. Subspace generalizations of the KGLRT, AMF, DMRao, and ACE detectors have been developed in [ 17 ] , [ 18 ] , [ 19 ] , and [ 20 ] , respectively. For brevity, the subspace-based variants of the first three detectors are termed the subspace GLRT (SGLRT), subspace AMF (SAMF), and subspace Rao (SRao). However, the aforementioned findings primarily focus on zero-mean clutter, while real-world scenarios may involve nonzero-mean clutter. For instance, in hyperspectral imaging, the mean spectral radiation of background targets is modulated by illumination conditions, atmospheric transmission and other factors, showing nonzero mean characteristics [ 21 ] . Moreover, in space-time adaptive processing (STAP) or array signal processing, the presence of outliers or interfering targets can cause the data to have a nonzero mean without the desired target being present. Outliers are typically caused by discrete clutter or interference [ 22 ] , while interfering targets originate from the sidelobe of radar receivers [ 23 ] . In these scenarios, ignoring the nonzero mean of clutter can lead to the loss of the constant false alarm rate (CFAR) feature of the detector, or a severe degradation of probability of detection (PD). The detection problem in nonzero-mean clutter modeled as elliptically symmetric distributions is investigated in [ 24 ] . Robust estimates of the clutter’s covariance matrix and mean vector are obtained using fixed point estimators (FPE), and founded on these estimates, an effective detector is developed utilizing the adaptive normalized matched filter (ANMF) structure. Furthermore, the same study derives the probability of false alarm (PFA) expression for this detector. For the signal detection problem in nonzero-mean Gaussian clutter, an adaptive detector with joint estimation of the mean and covariance matrix is constructed, and the analytical PFA is derived in [ 25 ] . In addressing the hyperspectral signal detection challenge in nonzero-mean clutter, a detector based on the ACE structure is designed in [ 21 ] given that both the nonzero-mean vector and covariance matrix of the clutter are known. A matrix-variate t-distribution is employed in [ 26 ] to model the nonzero-mean non-Gaussian clutter, where a GLRT is derived. The results indicate that this detector is an extension of the GLRT for nonzero-mean Gaussian clutter. The hyperspectral target detection problem in nonzero-mean matrix-variate t-distribution clutter is investigated in [ 27 ] . By utilizing two sets of training data with different means but the same covariance matrix, an effective GLRT-based detector is developed. For the hyperspectral signal detection problem, the statistical performance of the real-valued KGLRT, ACE, and AMF in nonzero-mean clutter is analyzed in [ 28 ] under two mismatch scenarios: an additive model and a replacement model. The former refers to the case where a mismatch is observed between the clutter means of the training and test data, while the latter involves an unknown amplitude mismatch in the clutter mean and an unknown mismatch in the covariance matrix between the training and test data. To address the detection challenge in more complex environments, the detection problem in nonzero-mean compound-Gaussian sea clutter is studied in [ 29 ] . Based on the GLRT, detectors are proposed for cases where the texture component follows inverse Gaussian, Gamma, and inverse Gamma distributions. The corresponding Rao and Wald detectors are provided in [ 30 ] . It should be noted that although numerous studies have addressed target detection in nonzero-mean clutter, none have yet investigated subspace signal detection in such environments. For signal detection in nonzero-mean clutter, due to the excessive number of unknowns, an optimal detector does not exist. Beyond the GLRT criterion, several other effective detector design strategies are available, including the Wald, Rao, Durbin, and Gradient tests. In some scenarios, these strategies can even outperforms the GLRT in detection performance [ 31 , 32 , 33 , 34 ] . Accordingly, this paper extends the rank-one signal model in [ 25 ] to a subspace signal model in the complex domain, designing detectors based on multiple strategies and evaluating their statistical performance. The most closely related prior studies are [ 25 ] and [ 28 ] , yet our work is distinguished from them in several key aspects. First, although effective detectors for nonzero-mean clutter have been proposed in [ 25 ] , their design is primarily based

Read the complete text