Target Monitoring Terminals (TMTs)

• Target Monitoring Terminals (TMTs) are target-facing monitoring endpoints that collect data via distributed sensors and are integrated into larger control architectures.
• They separate sensing functions from high-level decision logic, optimizing deployment flexibility and system robustness in varied environments.
• Applications span wireless communication, robotics, power-grid supervision, and astronomical surveys, each adapting terminal designs to unique operational challenges.

Target Monitoring Terminals (TMTs) are terminal-based sensing or monitoring endpoints associated with a designated target, target region, or target-adjacent process. In current literature, the term is used explicitly in perceptive mobile networks and networked integrated sensing and communication, where TMTs are distributed sensing receivers; functionally analogous architectures also appear in robotic target search, electric-grid transformer supervision, in-detector target metrology, beamline instrumentation, and high-cadence astronomical monitoring. Taken together, these works suggest a common architectural pattern: target-facing sensing is separated from higher-level control, estimation, or decision logic to improve deployment flexibility, robustness, or operation in hostile environments (Xie et al., 2021, Xia et al., 6 Oct 2025).

1. Terminology, scope, and acronym ambiguity

The phrase “Target Monitoring Terminals” does not denote a single universally standardized device class. In wireless sensing papers, it refers directly to distributed sensing terminals connected to a larger network. In other domains, closely related systems are described without the same acronym. This is particularly important because the astronomy literature discussed here uses TMTS, not TMTs: TMTS stands for Tsinghua University–Ma Huateng Telescopes for Survey, a high-cadence survey system rather than a communications terminal (Lin et al., 2021, Guo et al., 2024).

This ambiguity is not merely lexical. In robotics, a “terminal” may be a candidate target location at which search can end; in cyber-physical monitoring, it may be an endpoint or sensing channel that can be activated according to a resilient schedule; in fixed-target experiments, it may be a permanently installed optical or photogrammetric monitor viewing the target inside a detector or beamline. A plausible implication is that “TMT” is best treated as a functional category—target-coupled monitoring endpoints—rather than as a single hardware archetype (Lyu et al., 19 Jan 2026, Sengupta et al., 2020, Cavoto et al., 2020).

Domain	Terminal interpretation	Representative sources
Perceptive mobile networks	Passive sensing receiver observing target echoes from BS waveforms	(Xie et al., 2021)
Networked ISAC	Single-antenna distributed sensing terminal forwarding reflected signals to a central controller	(Xia et al., 6 Oct 2025)
Power-grid monitoring	TMT-like monitor-capable endpoint or sensing channel activated under MTD	(Sengupta et al., 2020)
Robotics	Probabilistic terminal or monitored target region in planning abstractions	(Lyu et al., 19 Jan 2026, Dudash et al., 2024)
Astronomy	High-cadence survey platform functioning as a field-monitoring system, though acronym is TMTS	(Lin et al., 2021)
Detector and beamline metrology	Permanently installed optical target monitor in high-radiation or high-field environments	(Cavoto et al., 2020, Bhadra et al., 2012)

2. TMTs in integrated sensing and communication

In the explicit communications usage, TMTs are introduced to let a network sense targets without requiring full-duplex sensing at the base station. In “Perceptive Mobile Network with Distributed Target Monitoring Terminals: Leaking Communication Energy for Sensing” (Xie et al., 2021), TMTs are defined as passive sensing terminals deployed over the wireless network, typically as IoT-like monitoring devices, with “only perception functionalities, including radar, vision, and other sensing capabilities.” The base station transmits the downlink communication/sounding waveform, while the TMT receives the target and clutter echoes. In the paper’s simplified model there is one BS with $N_t$ antennas, one TMT with $N_r$ antennas, $K$ single-antenna UEs, and one sensing target. The TMT uses a hybrid combiner $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$ and a detection filter $\mathbf{w}_d$; sensing quality is measured by SCNR, communication quality by UE SINR, and the joint design is optimized by alternating optimization with a convergence proof. The paper’s central physical conclusion is that, instead of forming dedicated sensing signals, it is more efficient to redesign the communication signals for both communication and sensing purposes and “leak” communication energy for sensing; the amount of energy leakage from one UE to the ST depends on their relative locations (Xie et al., 2021).

A later networked ISAC formulation generalizes this architecture from a single TMT to multi-TMT localization. In “Coordinated Beamforming for Networked Integrated Communication and Multi-TMT Localization” (Xia et al., 6 Oct 2025), the system consists of $M$ BSs, $N$ TMTs, a central controller, and $K$ single-antenna communication users per BS. Each TMT has a single receive antenna, BSs and TMTs are connected to the central controller via fronthaul links, and clock synchronization can be established among BSs and TMTs via wired links. Because TMTs are low-cost terminals and may have only one antenna, the paper adopts ToA-based localization rather than AoA-based localization. The received signal at TMT $n$ is

$$r_n(t)=\sum_{m=1}^{M}\sum_{k=1}^{K} \xi_{m,n} \mathbf{a}^T(\theta_m)\mathbf{f}_{m,k}s_{m,k}(t-\tau_{m,n}) + n_n(t),$$

and the target localization metric is a closed-form CRLB

$N_r$0

This CRLB becomes either the sensing objective or the sensing constraint in coordinated beamforming. The paper develops a globally optimal SDR-based algorithm for a sensing-centric regime when each BS has more antennas than the total number of communication users, a globally optimal bisection-based algorithm for the single-BS communication-centric case, and a unified SCA-based algorithm for the general case. Its simulation result is unusually direct: deploying more TMTs is always preferable to deploying more BSs in the studied networked ISAC system (Xia et al., 6 Oct 2025).

3. Cyber-physical monitoring terminals and resilient configuration design

A power-system interpretation of TMTs emerges in “Moving Target Defense for Robust Monitoring of Electric Grid Transformers in Adversarial Environments” (Sengupta et al., 2020), although the paper does not use the term TMT. The monitored assets are high voltage transformers, and the sensing problem is cast on a bipartite graph

$N_r$1

where $N_r$2 is the set of transformers, $N_r$3 is the set of permissible sensor locations, and $N_r$4 connects a transformer to a sensor location if the transformer’s behavioral signal can reach that sensing location within an acceptable propagation radius. The paper assumes PMUs cannot be placed directly on HVTs, and signal propagation is limited to at most two hops because the signal-to-noise ratio deteriorates rapidly over larger distances. In a TMT-like reading, this implies that monitor terminals should not be treated as arbitrarily observable endpoints; their usefulness is topology- and propagation-constrained (Sengupta et al., 2020).

The baseline identifiability problem is the Minimum Discriminating Code Set (MDCS): for each transformer $N_r$5, the selected active sensors $N_r$6 must be unique. To make this robust to attack, the paper defines $N_r$7-$N_r$8MDCS, requiring $N_r$9 different MDCSs that are pairwise disjoint: $K$0 This is the paper’s concrete form of differential immunity. It proves that $K$1-$K$2MDCS is NP-Complete, formulates a QC-ILP for exact computation, and proposes a greedy iterative alternative that is faster but may return only $K$3. Once the configuration set is computed, the defender randomizes over those configurations via a Stackelberg security game. A TMT-like deployment inspired by this work would therefore emphasize multiple physically distinct sensing options per monitored transformer, configuration diversity across monitoring modes, the ability to switch active sensor subsets over time, and control logic that chooses these subsets strategically against an intelligent adversary. The design implication is explicit: if each MDCS has size $K$4 and there are $K$5 such configurations, the defender may need to place $K$6 sensors in the network while activating only $K$7 at any one time (Sengupta et al., 2020).

4. Robotic interpretations: probabilistic terminals, access monitoring, and mobile monitoring agents

In robotic target search, “terminal” becomes a planning abstraction. “RPT*: Global Planning with Probabilistic Terminals for Target Search in Complex Environments” (Lyu et al., 19 Jan 2026) defines the Hamiltonian Path Problem with Probabilistic Terminals (HPP-PT), where each graph vertex $K$8 has a probability $K$9 that the robot’s path terminates there because the target is found. The expected path cost of a Hamiltonian path $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$0 is

$(\mathbf{W}_{RF}, \mathbf{W}_{BB})$1

The challenge is history dependence: the cost of future movement depends on the order in which vertices were previously visited. The paper removes this by augmenting the search state to $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$2, where $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$3 is the probability the target has not yet been found and $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$4 is the visited set, yielding the Markovian transitions

$(\mathbf{W}_{RF}, \mathbf{W}_{BB})$5

RPT* is optimal; F-RPT* is $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$6-bounded suboptimal. The associated HATS system combines this planner with Bayesian filtering in known environments and frontier-based exploration in unknown environments. This suggests a TMT interpretation in which candidate monitoring locations are ranked not by geometric distance alone but by their probability of terminating the search (Lyu et al., 19 Jan 2026).

A related multi-target access-monitoring formulation appears in “Multi-Agent Team Access Monitoring: Environments that Benefit from Target Information Sharing” (Dudash et al., 2024). Here the “targets” are target regions $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$7 in a discretized 2D environment. Robots are placed so that every path from the border region $(\mathbf{W}_{RF}, \mathbf{W}_{BB})$8 to any target region intersects a monitored location. The paper replaces a prior minimum-edge-cut formulation by a minimum-node cut and generalizes from one target to multiple targets using two strategies: an individual approach, which solves a separate cut problem for each target and unions the robot placements, and a holistic approach, which merges all targets into a common sink before solving a single cut problem. The holistic approach always used fewer or equal robots in the reported simulations and was most beneficial in medium-density obstacle environments (Dudash et al., 2024).

A mobile-agent realization is given in “Free-Space Ellipsoid Graphs for Multi-Agent Target Monitoring” (Ray et al., 2022). The paper decomposes free space into ellipsoids, builds a weighted connectivity graph over overlapping ellipsoids, and uses the same ellipsoids both for high-level coordination and as state constraints in MPC. Tracking agents have second-order unicycle dynamics,

$(\mathbf{W}_{RF}, \mathbf{W}_{BB})$9

and the case study uses four tracking agents monitoring fifteen dynamic targets. The paper reports an average MPC solve time of 50 ms and treats assignment, path planning, and control as a unified obstacle-aware monitoring problem. In this usage, TMT-like functionality is realized by mobile tracking agents rather than by fixed terminals (Ray et al., 2022).

5. Astronomical, detector, and beamline exemplars

The astronomy literature provides a large-scale monitoring-system analogue rather than an explicit TMT usage. TMTS—the Tsinghua University–Ma Huateng Telescopes for Survey—is a multi-tube telescope system consisting of four 40 cm optical telescopes at Xinglong Station of NAOC, each with a 4096 × 4096 CMOS camera, total field of view of about 18 deg$\mathbf{w}_d$0, readout time < 1 s, a broad Luminous filter covering about 330–900 nm, and a $\mathbf{w}_d$1 limiting magnitude of about 19.4 mag for a 1-minute exposure. Its observing strategy is to stare at LAMOST areas for the whole night with cadence of about 1 minute, yielding uninterrupted light curves especially suitable for short-period variability (Lin et al., 2021). In the first two survey years, TMTS monitored 251 LAMOST/TMTS plates with valid sky coverage of 3,857 deg$\mathbf{w}_d$2 and produced 10,856,233 uninterrupted light curves with at least 100 epochs, leading to 1,107 unique stars with $\mathbf{w}_d$3 hr, including 1,076 $\mathbf{w}_d$4 Scuti stars and a rare-object tail containing BLAPs, ZZ Ceti stars, an ELMV, ultracompact binaries, and cataclysmic variables (Lin et al., 2023). In the first 3-year CV survey, the scale increased to 19,099,266 uninterrupted light curves over 449 LAMOST/TMTS plates and $\mathbf{w}_d$5, with a catalog of 64 CVs/CV candidates and minute-cadence light curves revealing superhumps, QPOs, large-amplitude orbital modulations, and rapid periodic/spin modulations (Liu et al., 2024). TMTS also supports discovery-to-follow-up workflows: cross-matching the Gaia EDR3 white dwarf catalog with the TMTS source catalog using a matching radius of $\mathbf{w}_d$6 produced TMTS light curves for almost 3000 white dwarf candidates, from which TMTS J17184064+2524314 was identified as a ZZ Ceti star and later characterized with follow-up photometry, TESS, spectroscopy, and asteroseismology (Guo et al., 2024). The machine-learning classification study later classified 11,638 variable stars into 6 main types using XGBoost and Random Forest classifiers with accuracies of 98.83% and 98.73%, respectively (Guo et al., 2024). Functionally, these papers show a survey-scale field-monitoring terminal architecture rather than a single local terminal.

Fixed-target and beamline instrumentation provide a more literal terminal-like pattern: a permanently installed device monitors a physically critical target under hostile conditions. In “A photogrammetric method for target monitoring inside the MEG II detector” (Cavoto et al., 2020), a CMOS-based, high resolution, high radiation tolerant and high magnetic field resistant photo-camera is mounted inside the detector at about 0.8 T, focused on a dot pattern printed on the target about 1 m away. The target-monitoring fit uses a 7-parameter optical mapping $\mathbf{w}_d$7, a rigid transform $\mathbf{w}_d$8, and a deformation model $\mathbf{w}_d$9 based on Zernike polynomials, with a relative fit

$M$0

Bench-top performance reached $M$1 and $M$2, with an estimated $M$3 resolution along the target normal. In “Optical Transition Radiation Monitor for the T2K Experiment” (Bhadra et al., 2012), the monitored quantity is the proton beam immediately upstream of the production target in a highly radioactive environment. The monitor uses optical transition radiation from a thin foil at 45° to the beam, a passive four-mirror optical transport system, and a remotely located CID camera to move light out of the harsh environment. It measures proton beam position and width with a precision of better than 500 $M$4m, satisfying the T2K requirement. These systems are not called TMTs in the papers, but they are terminal-like in the strong sense of being permanently installed target-facing monitoring nodes with remote readout and calibration (Cavoto et al., 2020, Bhadra et al., 2012).

6. Common design principles, trade-offs, and unresolved issues

Across these literatures, several recurrent design rules appear. First, TMT-like systems often separate transmit/control from receive/monitor functions. In PMN-TMT and networked ISAC, BSs illuminate while TMTs receive and forward reflections to a controller; this avoids full-duplex operation at the BS and enables dense, flexible sensing geometry (Xie et al., 2021, Xia et al., 6 Oct 2025). In T2K and MEG II, the target-facing element is optically or photogrammetrically simple, while fragile electronics are moved away from the most hostile region (Bhadra et al., 2012, Cavoto et al., 2020). In power-grid monitoring, the analogous separation is between precomputed monitoring configurations and a runtime scheduler that activates them strategically (Sengupta et al., 2020).

Second, geometry and topology are decisive. The ToA CRLB in multi-TMT localization depends explicitly on the Jacobian $M$5 and the BS–TMT information matrix $M$6; robotic access monitoring depends on minimum separators and chokepoints; ellipsoid-graph monitoring depends on obstacle-aware connectivity rather than Euclidean distance; and power-grid identifiability depends on transformer–sensor reachability under two-hop constraints (Xia et al., 6 Oct 2025, Dudash et al., 2024, Ray et al., 2022, Sengupta et al., 2020).

Third, calibration and synchronization are not ancillary details but structural requirements. Wired synchronization among BSs and TMTs is assumed for ToA localization; global alignment in MEG II is guaranteed by corner cubes or reflectors on the target support; the T2K OTR monitor relies on a calibration foil, backlighting, and efficiency maps; and TMTS astronomy papers repeatedly couple wide-field monitoring to external catalogs such as Gaia and LAMOST for interpretation and validation (Xia et al., 6 Oct 2025, Cavoto et al., 2020, Bhadra et al., 2012, Lin et al., 2021).

Several limitations also recur. The term itself remains heterogeneous: some papers use TMT explicitly, others only admit a TMT-like interpretation. Communications papers often simplify to one TMT or one target before later multi-TMT or multi-target extensions. Grid-defense results assume a single-sensor attack model and do not optimize movement timing $M$7. Robotic optimal planning remains exponential in the worst case, with optimal RPT* practical up to about 40 vertices and bounded-suboptimal F-RPT* up to about 200 vertices within the reported time budgets. Single-view photogrammetry, as in MEG II, retains strong parameter correlations and would benefit from two cameras from different viewpoints (Xie et al., 2021, Sengupta et al., 2020, Lyu et al., 19 Jan 2026, Cavoto et al., 2020).

The main misconception dispelled by this literature is that a TMT must be a single hardware form factor. The published record instead points to a broader concept: a TMT is a target-facing monitoring endpoint whose value derives from placement, coupling to a target-specific geometry or process, and integration with upstream control, inference, or fusion layers. In some domains it is a passive sensing receiver, in others a planning abstraction, a survey platform, or a target-adjacent optical monitor. What unifies these uses is not hardware uniformity but the role of the terminal within a larger monitoring architecture.