
A reduced-order model for simultaneous tracking and navigation (RO-STAN) with low Earth orbit (LEO) satellites with erroneous ephemerides is developed. In RO-STAN, each satellite's ephemeris reference time error and in-track position and radial velocity orbit errors are captured by a single equivalent timing error state. An extended Kalman filter aids the vehicle’s inertial measurement unit (IMU) with navigation observables from a LEO receiver, while simultaneously estimating the equivalent timing error. Monte Carlo numerical simulation results with Orbcomm, Iridium, Globalstar, OneWeb, and Starlink LEO satellites are presented for: (1) “trusting” erroneous ephemerides; (2) using correct ephemerides; (3) utilizing standard STAN, which estimates the vehicle's states simultaneously with the satellites' position and velocity states and relative clock error states between each satellite and the vehicle-mounted receiver; and (4) RO-STAN. Experimental results are presented of LEO-aided tactical-grade IMU navigation on a ground vehicle and an unmanned aerial vehicle (UAV), exploiting 11 and 9 Starlink satellites, respectively, while: (i) employing two-line element (TLE)-initialized SGP4-propagated ephemerides, (ii) STAN, and (iii) RO-STAN. The ground vehicle traversed a 3 km trajectory, the last 2.25 km of which without GNSS; while the UAV traversed a 500 m trajectory, the last 370 m of which without GNSS. The 3-D position root-mean squared-error (RMSE) of the ground vehicle's and UAV's GNSS-IMU fused with an altimeter grew to 258 m and 46.51 m, respectively; while the RMSEs of (i)–(iii) were 150, 16.19, and 8.41 m, respectively; and 17.82, 14.35, and 8.15 m, respectively.
CubeSat swarms in low Earth orbit are moving from educational demonstrations to operational platforms for Earth observation, Internet of Things connectivity, and scientific data relay. As density rises, three problems once treated separately become coupled: keeping satellite-to-ground links under rain, cloud, and ionospheric impairments; sustaining inter-satellite communication through adaptive routing and spectrum access over a topology that changes within seconds; and avoiding conjunctions with debris or other satellites. We define a CubeSat swarm as a self-organizing cluster of three or more nanosatellites with collective behavior beyond any member, unlike large, loosely coordinated constellations. This tutorial introduces small-satellite practitioners to swarm intelligence, multi-agent deep reinforcement learning, and physics-informed channel modeling, and shows how they couple in a layered architecture of slow formation-keeping, intermediate routing-and-spectrum, and fast physical-layer adaptation loops. We review safety methods (passive safety with closed-form control in relative orbital elements, sequential convex programming, model predictive control, artificial potential fields, control barrier functions, and run-time assurance), routing and spectrum methods (contact graph routing, double and recurrent deep Q-networks, value decomposition, and graph-based policies), and International Telecommunication Union Radiocommunication Sector propagation models, with physics-informed attenuation decomposition as an open direction. Against a classical guidance, navigation and control baseline, we give selection guidelines by scenario and mission need and compare published quantitative results, including flight results. We cover flown formation missions (PRISMA, GOMX-4, CPOD, NASA Starling, NanoFF) and upcoming ones (SWARM-EX, VISORS), compare this tutorial with ten prior surveys, and set out open challenges separating today's demonstrations from operational autonomous swarms.
In complex electromagnetic environments, an electronic support measures (ESM) receiver intercepts radar pulses from multiple emitters, which are subsequently represented as an interleaved sequence of pulse descriptor words (PDWs). For downstream tasks, such as tracking and threat evaluation, the radar intelligence system must recover individual emitters from this interleaved signal without prior knowledge of their number or parameters, making deinterleaving a critical challenge. Towards this, the letter presents a novel unsupervised clustering-based three-stage radar pulse deinterleaving pipeline and its real-time deployment on a low-cost embedded processor. Stage I forms the initial clusters using the proposed Separability-Adaptive Distance (SAD), whose weights are derived from the data through three fused histogram-separability statistics, with an adaptive minimum cluster size preserving sparse low-PRF emitters. Stage II recovers pulse repetition interval (PRI) structure from time of arrival (TOA) differences, and Stage III reunites the fragments of individual emitters through an optimal-transport merge on log-PRI distributions. Importantly, the proposed pipeline operates with $O(N\log N)$ computational complexity, enabling real-time operation, while attaining state-of-the-art performance with an adjusted Rand index of 0.908 on a physics-constrained synthetic benchmark.