What AI Cannot Fix

C4ISR & Sensor Fusion

The Naval Tactical Data System's most intractable operational problems — sensor alignment bias, dynamic navigation error, and multi-sensor track association failure — were substantially ameliorated by the E-2C/D Hawkeye's network-derived registration and the Cooperative Engagement Capability's architectural innovation of distributing raw sensor measurements rather than processed tracks. Neither fully solved the problem, and neither transfers completely to the GPS-denied, multi-domain, multi-service environment that JADC2 requires. The AI system at the top of the processing chain receives a track picture corrupted by errors it cannot see, and generates confident engagement recommendations it cannot qualify.

Stephen L. Pendergast, LT USNR, IEEE Sr. Member (Ret.) | Senior Engineer Scientist July 2026

Sources: GAO, ETHW, IATA, FAA, OPSGROUP, USNI, CSBA, CRS, DOT&E

BLUF — Bottom Line Up Front

The U.S. Navy's Naval Tactical Data System, first deployed in 1961, pioneered multi-ship sensor fusion and encountered three fundamental problems that were never mathematically solved: data association ambiguity in dense contact environments, dynamic sensor alignment bias in azimuth and bearing, and inter-ship coordinate frame misalignment from navigation error. Two intermediate-generation systems substantially ameliorated these problems: the E-2C/D Hawkeye's network-derived registration through the Link 16 Joint Units network, which provided a geometrically distinct high-altitude sensor node enabling inter-platform bias estimation; and the Cooperative Engagement Capability — conceived by Johns Hopkins University Applied Physics Laboratory — which addressed the NTDS double-track problem at its architectural root by distributing unfiltered radar measurements rather than processed tracks, employing Extended Kalman Filter multi-sensor tracking, and implementing precision gridlock for sensor coordinate frame alignment. Together, these systems brought the naval battle group's air defense sensor fusion to fire-control quality within the GPS-available, line-of-sight battle group environment. JADC2 — the Joint All-Domain Command and Control system receiving $1.4 billion or more annually — must extend this solution across all services, all domains, and coalition partners in a GPS-denied peer conflict environment. GPS spoofing and jamming have increased 500% in 2024 alone by some measures, restoring the pre-GPS navigation errors that CEC's gridlock suppressed for airborne sensor contributors. The AI systems now being integrated into JADC2's sensor-to-shooter loops receive a track picture that may be systematically corrupted by these errors — none of which is visible in the track data the AI processes, and none of which current AI systems are designed or tested to detect.

In 1962, the prospective commanding officers of new guided missile frigates — informed that the Naval Tactical Data System was going to be built into their ships — came to the NTDS project office to deliver a message. As the Engineering and Technology History Wiki records it: "No damned computer was going to tell them what to do. For sure, no damned computer was going to fire their nuclear tipped guided missiles."¹ Secretary of the Navy John Connally and Chief of Naval Operations Admiral Arleigh Burke were pushing the NTDS project office to accomplish in five years what would normally take thirteen. They prevailed. The system was fielded. And it immediately encountered the fundamental technical problems that the commanding officers' instincts, if not their stated objections, had correctly anticipated: in a real operational environment with multiple ships, multiple sensors, noisy measurements, and dynamic navigation errors, building a common tactical picture that was reliable enough to act on was harder than any of the system's architects had projected.

Sixty-five years later, the Department of Defense is spending over $1.4 billion annually on JADC2 — Joint All-Domain Command and Control — a program whose ambition is to do what NTDS did for the naval task force, but across all military services, all domains, and in coalition with allies, connected through AI systems that transform sensor data into engagement decisions at machine speed. The commanding officers of 2026 are not objecting. They are requesting the capability. But the technical problems that made NTDS harder than expected — multi-sensor data association in dense environments, dynamic sensor alignment bias, and coordinate frame misalignment from navigation error — have not been solved in the intervening six decades. They have been masked by GPS, and they are about to be unmasked by the adversary.

NTDS: The Original Experiment in Multi-Ship Sensor Fusion

The Naval Tactical Data System was developed beginning in 1956, with major contracts awarded to Sperry Rand's UNIVAC division for the AN/UYK computing systems, Collins Radio for communications, and Hughes Aircraft for displays and radar integration.² The system achieved its first operational deployment in October 1961 aboard USS Oriskany (CVA-34) and USS King (DLG-10), connected by the AN/USQ-55 data link that would evolve into Link 11.³ The vision was straightforward in concept: each ship's sensors would contribute to a common tactical picture shared across the task force, each ship could see what every other ship saw, and weapons assignment could be coordinated across multiple shooters against a common threat picture.

The operational reality was more complex. As the Military History wiki documents, NTDS's two major problems were immediately apparent: "each ship had its own view of the battlespace, independent of the rest of the ships in the task force" — and the system designed to create a common view immediately encountered the technical barriers to doing so reliably.⁴ Those barriers were not software bugs or hardware failures. They were fundamental properties of the sensor fusion problem that the system architects had not fully anticipated.

NTDS's combat debut in Vietnam — aboard PIRAZ (Positive Identification Radar Advisory Zone) ships in the Gulf of Tonkin — demonstrated both its value and its limitations. The PIRAZ ships provided air traffic control, threat warning, and search and rescue coordination for strike packages operating over North Vietnam, crediting NTDS with saving many pilots' lives through precise radar coverage and data link coordination.⁵ But the operational environment also revealed the depth of the multi-sensor correlation problem. Multiple radar systems on different ships, operating with different measurement biases, different navigation errors, and different scan rates, produced contact tracks that were difficult to merge into a coherent common picture. The same physical aircraft appeared as multiple contacts in different ships' track files. The association algorithms of the era — rule-based correlation on range, bearing, and speed — produced systematic errors when contacts were closely spaced, when sensors disagreed, or when navigation biases separated the coordinate frames of different ships' measurements.

The Three Fundamental Problems

I. The Data Association Problem

The data association problem — determining which sensor measurement at the current scan corresponds to which previously established track — is mathematically equivalent to an assignment problem in combinatorial optimization. In a sparse contact environment with well-separated targets and low sensor noise, it is tractable. In the contact densities encountered in the outer air battle, with multiple threat aircraft, friendly strike packages, tankers, and escorting fighters simultaneously in the same radar coverage volume, it is not.

The Mahalanobis distance — the normalized statistical distance between a measurement and a track prediction, accounting for the combined uncertainty from process noise and measurement noise — provides the correct association metric under the assumption that measurement errors are zero-mean and Gaussian. It was the state-of-the-practice approach for track-to-measurement association and was incorporated into NTDS-derived systems through the 1970s, 1980s, and 1990s. Applied to the Linear Assignment Problem via the Hungarian algorithm, it provides the globally optimal one-to-one assignment of measurements to tracks under the Gaussian noise assumption.

Even optimal assignment leaves residual errors in three conditions that operational environments reliably produce. First, when two targets are separated by less than the sensor resolution cell, their measurements merge or cross-contaminate and no association algorithm can correctly assign the measurements independently. Second, when two tracks cross in the observation space, the post-crossing assignment is ambiguous — and the optimal algorithm, by choosing one assignment, is wrong half the time. Track crossing produces the "track swap" pathology, where IFF history, ESM emitter identification, and tactical attributes are assigned to the wrong physical object after the crossing. Third, when targets maneuver outside the assumed motion model, the track prediction diverges from the actual position, increasing the Mahalanobis distance to the correct measurement and potentially causing the track to associate with a nearby false alarm instead.

Joint Probabilistic Data Association, Bar-Shalom and Fortmann's theoretically optimal extension of the Kalman filter to uncertain association environments, addressed the crossing track problem by replacing hard association with probabilistic soft assignment weighted by posterior association probabilities.⁶ It was evaluated seriously for NTDS-derived combat direction systems in the 1990s. Its computational load was the disqualifying factor: in dense contact environments the joint event enumeration is equivalent to computing the permanent of a matrix — a #P-complete problem that scales factorially with cluster size. JPDA delivered optimal decisions too slowly for real-time combat direction system operation during high-intensity threat scenarios. The computational intractability was not a hardware limitation that Moore's Law would resolve — it is a mathematical property of the problem structure that makes exact optimal association computationally intractable in exactly the conditions where it matters most.

II. The Sensor Alignment and Azimuth Bias Problem

Radar azimuth measurement is subject to systematic biases that the Mahalanobis distance metric cannot correctly handle, because the metric assumes zero-mean noise and a bias by definition has a non-zero mean. The sources of azimuth bias in shipboard radar systems include: antenna mount azimuth encoder error, thermal effects on the antenna structure, antenna pedestal leveling error as the ship rolls and pitches, and for AESA systems the phase shifter calibration errors that accumulate with component aging and thermal cycling. These biases are dynamic — they change with temperature, operational conditions, and equipment state — and they are not directly observable from the radar's own measurements.

A 0.5° azimuth bias at 50 nautical miles produces a cross-range position error of nearly a nautical mile. If the Mahalanobis distance metric's measurement noise covariance R characterizes the random radar measurement error as having a standard deviation of 200 feet in cross-range, a bias-induced error of 4,000 feet appears as a 20-sigma event — far outside any reasonable gating threshold. The association algorithm correctly classifies this as a non-association, producing separate track files for what is a single physical contact observed by two differently-biased sensors. The "double track" problem that plagued NTDS operations throughout its history — the same aircraft appearing as two separate contacts in the common tactical picture — was in large part a manifestation of uncalibrated inter-sensor azimuth biases interacting with an association algorithm that had no mechanism to account for systematic measurement offsets.

III. The Navigation Error and Coordinate Frame Misalignment Problem

In the pre-GPS era, shipboard navigation rested on inertial navigation systems whose dominant error source was gyro drift — on the order of 0.01 degrees per hour for quality 1970s-era floating gyroscopes, integrating to approximately one nautical mile of position error per hour between celestial or LORAN fixes. When Ship A with a 1 nautical mile navigation error and Ship B with a 1.5 nautical mile navigation error both tracked the same contact, their measurements of that contact's position were expressed in coordinate frames offset by the vector difference of their navigation errors — up to 2.5 nautical miles in an unknown direction. At the contact ranges relevant to the outer air battle, this inter-ship coordinate frame misalignment was larger than the random sensor measurement noise by factors of ten or more, systematically corrupting every inter-ship association attempt regardless of algorithmic sophistication.

The track registration problem — estimating the inter-ship alignment parameters from the statistical properties of jointly observed contacts — requires simultaneously solving for both the alignment and the association, creating a bootstrapping problem: correct registration requires correct association, and correct association requires correct registration. Expectation-Maximization approaches that iterate between soft association and registration estimation handle this correctly in principle but introduce additional computational load and convergence uncertainty in sparse contact environments where the estimation is ill-conditioned.

What GPS Changed — and What It Left Unchanged

GPS achieving Full Operational Capability in 1995 was a transformative event for the navigation error problem. Modern ring laser gyroscope INS systems, achieving drift rates on the order of 0.0001 degrees per hour, accumulate position errors of hundreds of meters rather than nautical miles between GPS updates spaced days apart. The coordinate frame misalignment that dominated inter-ship association failures in NTDS — the systematic position offsets that no association algorithm could overcome — became smaller than the random sensor measurement noise with GPS-quality navigation. The double-track problem caused by navigation error largely disappeared from the surface ship common picture after GPS became standard.

For shipboard platforms, GPS denial is geometrically bounded by the line-of-sight horizon. A surface-based jammer at 10 meters above sea level has a geometric horizon of approximately 7 nautical miles to a ship's GPS antenna at 20 meters. GPS satellite signals arrive at high elevation angles where ship GPS antennas have their maximum gain, while jamming signals arrive near the horizon where antenna discrimination reduces effective jammer power. Modern ring laser and fiber optic gyro IMUs, achieving drift rates orders of magnitude better than 1970s SINS, require GPS updates only on the order of daily time scales to maintain navigation quality adequate for inter-ship track correlation. Ships close to hostile coasts face more jamming exposure, but the geometry substantially limits shipboard GPS vulnerability compared to airborne platforms.⁷

What GPS did not change was the azimuth bias problem. The sensor-to-platform lever-arm calibration — the precise alignment of each radar's phase center or antenna boresight to the ship's navigation reference frame — remains subject to measurement error at installation and subsequent change from structural flexure, thermal cycling, and equipment aging between calibration events. For a carrier operating at high tempo flight operations, flight deck structural flexure relative to the ship's keel-mounted IMU produces a dynamic sensor-to-platform misalignment that GPS-quality navigation cannot correct because it occurs within the inertial measurement interval. The AESA phased array calibration table that maps commanded phase shifts to actual beam directions degrades with component aging and thermal cycling between depot calibration events. These systematic sensor errors — smaller in modern systems than in 1960s hardware, but not zero — remain invisible to the association algorithm's Mahalanobis distance metric.

The Intermediate Generation: E-2C/D Hawkeye and the Cooperative Engagement Capability

The article's narrative would be incomplete — and unfair to sixty years of naval engineering — without acknowledging the intermediate generation of systems that substantially ameliorated the NTDS multi-sensor correlation problems between the GPS transition and the current JADC2 development. Two developments in particular represent the closest the Navy has come to correctly solving the sensor registration and association problems at an architectural level: the E-2C/D Hawkeye's network-derived registration of participating units in the Link 16 network, and the Cooperative Engagement Capability developed by the Johns Hopkins University Applied Physics Laboratory. Both deserve specific technical treatment because JADC2 inherits their achievements and their limitations simultaneously.

The E-2C/D Hawkeye as Registration Node

The E-2 Hawkeye's contribution to the sensor registration problem flows directly from its geometry. A surface ship's radar is limited to line-of-sight contacts — the geometric horizon restricts its coverage to roughly 20 to 30 nautical miles against low-altitude targets at typical mast heights. The E-2, operating at 25,000 feet, extends the radar horizon to approximately 350 nautical miles and provides 360-degree coverage of the airspace above the task force and the surrounding ocean. This altitude advantage is not merely a surveillance benefit — it is a sensor registration enabler.

When the E-2C's Block 2 aircraft received Joint Tactical Information Distribution System terminals and access to the Link 16 network in the 1980s, they became network registration nodes in a specific and technically important sense.^7a The Hawkeye's sensor observes the same contacts that surface ship radars observe, but from a different geometric vantage point at high altitude. The differences in measured contact positions between the E-2's radar and the surface ships' radars contain information about the inter-platform sensor alignment biases — the same registration information that ground-based track registration algorithms tried to extract from joint contact observations. Because the E-2's navigation was IMU-referenced and its radar geometry from altitude provided contact position fixes with a different error structure than surface ship sensors, the network of Link 16 joint units (JUs) could derive registration corrections by comparing E-2 contact positions with surface ship positions of the same contact — the network-derived registration that directly addressed NTDS's inter-ship coordinate frame misalignment problem.

The E-2D Advanced Hawkeye, entering service in 2014 with IOC in October of that year, extended this capability substantially.^7b The AN/APY-9 AESA radar — operating in UHF band specifically to improve detection of low-observable and sea-skimming threats — provides simultaneous tracking of over 3,000 targets. Integration of Cooperative Engagement Capability, Link 16, and Tactical Targeting Network Technology into a single architecture transformed the E-2D from an airborne early warning aircraft into a network-centric sensor fusion node capable of distributing fire-control-quality data across naval and joint forces. In Operation Epic Fury in 2026, E-2D Hawkeyes provided over-the-horizon targeting data to surface ships through CEC, enabling engagements against Iranian missile systems beyond the horizon of the engaging ships' own radars.^7c The E-2D is now, as one analysis noted, assuming expanded theater-level command and control responsibilities in joint operations as the aging E-3 Sentry fleet diminishes — making its sensor registration and fusion role even more critical to the force.

CEC: The Architectural Breakthrough That Link 11 Could Not Achieve

The Cooperative Engagement Capability represents the most technically sophisticated solution the Navy has fielded to the multi-sensor association and registration problems that plagued NTDS. Conceived by the Johns Hopkins University Applied Physics Laboratory in the early 1970s under the Battle Group Anti-Air Warfare Coordination program, developed through at-sea experiments beginning in 1990, entering engineering and manufacturing development in 1995, and achieving first fleet deployment in 1998, CEC addressed the NTDS correlation problem at its architectural root rather than through algorithmic refinement of the same track-distribution architecture.^7d

The key innovation is what CEC does not distribute: processed tracks. Where Link 11 and the NTDS data link architecture distributed track files — position estimates, velocity estimates, and track attributes generated by each ship's individual tracking software — CEC distributes unfiltered radar measurements directly associated with tracks, in near real time, to all participating units simultaneously.^7e Each CEC-equipped platform runs identical tracking algorithms on the raw measurements from all platforms in the network, producing a truly common track rather than attempting to reconcile separately-maintained track files that have accumulated different processing histories and different systematic biases.

The significance of this architectural choice cannot be overstated. The NTDS double-track problem — the same aircraft appearing as two separate contacts in different ships' track files — was fundamentally a consequence of separately-processed tracks diverging because of sensor bias, navigation error, and association history differences. When two ships' processing chains start from different raw measurements, apply different association decisions, and accumulate different filter states, their tracks of the same contact become progressively less correlated over time. No post-hoc reconciliation algorithm, however sophisticated, can recover the information lost in separate processing. CEC avoids the problem at its source by ensuring that all platforms process the same measurements through the same algorithms — the only way to guarantee that all platforms maintain the same track.

CEC's Gridlock Function: Solving Sensor Registration at the Architecture Level

CEC's gridlock function is the operational implementation of the sensor registration solution that NTDS-era track registration algorithms attempted to achieve analytically. The APL Technical Digest description of CEC processing subfunctions explicitly lists "gridlock" alongside track filtering, track divergence and convergence testing, and sensor interfacing as core CEP functions. Gridlock provides precision synchronization of sensor coordinate frames across all CEC-equipped units — not by estimating and correcting for registration biases after the fact, as classical registration algorithms do, but by maintaining a shared measurement reference frame in real time through the CEC data link. The result, per GlobalSecurity.org's documentation of the CEC system: "very accurate gridlocking between CUs" that supports fire-control-quality engagement even when the engaging unit does not itself hold a radar track on the target. This is the capability that NTDS engineers were trying to achieve through the alignment and registration algorithms of the 1970s, 1980s, and 1990s — CEC achieved it not by solving the registration estimation problem more cleverly, but by sharing the measurements before registration errors could accumulate.

The Cooperative Engagement Processor at each CEC unit runs the tracking algorithms — based on Extended Kalman Filter estimation that correctly handles the non-linear measurement geometries of multi-platform radar observation — and performs explicit track divergence and convergence testing as part of its processing cycle. This divergence testing is the operational implementation of the Bar-Shalom innovation sequence quality assessment: the system explicitly monitors whether tracks from different sensors are diverging from each other in ways that indicate association errors or sensor registration failures, rather than silently accumulating these errors into degraded track states. The CEP's "track divergence and convergence testing" subfunction is, in effect, an automated track quality monitor that CEC maintains continuously across the network — the diagnostic tool that NTDS lacked and that current JADC2 AI interfaces do not propagate to the reasoning layer.

The CEC data link itself — a high-data-rate, directional, jam-resistant, line-of-sight network that operates on a pairwise basis between units — was specifically designed to carry the bandwidth required for unfiltered measurement distribution within the stringent time budget that fire-control-quality tracking demands. A 2024 Chinese military report documented awareness of CEC's architecture and outlined plans to use integrated jamming platforms to degrade radar-sensor accuracy and disrupt CEC data fusion, specifically to isolate platforms and reduce cooperative engagement effectiveness in contested environments — confirming that peer adversaries understand CEC's architectural importance to naval air defense.^7f

CEC's Integration with NIFC-CA and the Path to JADC2

The Naval Integrated Fire Control-Counter Air architecture — which uses the E-2D as the sensor fusion and fire control node linking F-35C stealth sensors through the Hawkeye to CEC-equipped surface ships for over-the-horizon engagement — represents the mature expression of CEC's potential. The F-35C, operating in a low-observable profile that precludes active radar emission, uses its passive sensors and data link to relay targeting data to the E-2D, which converts it to CEC-format fire-control-quality tracks and distributes them to surface ships for engagement. The engaging ship fires on a target its own radar has never seen, at ranges its own radar could not reach, based on fire-control-quality tracks that CEC's gridlock and multi-sensor EKF fusion have made reliable enough for missile engagement.^7g This capability — demonstrated operationally, not merely prototyped — represents the highest current expression of the multi-sensor registration and association problem's partial solution.

CEC's limitations, however, define the gap that JADC2 is attempting to bridge and that makes the architectural challenges described in this article so significant. CEC is a line-of-sight system within the battle group — its data link cannot relay beyond the geometric horizon without an airborne relay node, limiting the network to the battle group's organic sensor coverage. CEC addresses the naval air domain specifically; it was not designed for the multi-domain, multi-service, coalition integration that JADC2 requires. And critically: CEC's gridlock and sensor registration function within the GPS coordinate framework. GPS denial degrades CEC's inter-unit registration quality for exactly the same geometric reason it degraded NTDS registration — airborne platforms with degraded navigation inject coordinate frame errors into the shared measurement space that the gridlock function cannot fully correct without external position reference. The partial solution that CEC achieved for the naval battle group in a GPS-available environment does not fully transfer to the GPS-denied, multi-domain, multi-service environment that JADC2 faces in a peer conflict.

JADC2 explicitly identifies CEC as one of the existing systems that will contribute to its architecture — Project Overmatch's network of networks includes CEC-equipped units as sensor nodes in the broader joint network. But incorporating CEC into a larger network that includes GPS-denied airborne contributors, F-22/F-35 incompatible data links, coalition partner systems, and space-based sensors introduces the original NTDS registration problems into the inter-network interfaces that CEC's architecture does not address. The problem CEC solved within the battle group reappears at the boundaries of the battle group, at exactly the seams where JADC2's architecture is least mature.

The NTDS-to-CEC-to-JADC2 progression is therefore not a linear improvement but a successive extension of the same problem to larger and more complex sensor networks. CEC solved the within-battle-group registration problem for the naval air domain with GPS-quality navigation. JADC2 is extending sensor fusion to all domains, all services, and coalition partners, in a GPS-denied peer conflict environment. The solutions that CEC embedded in its architecture — raw measurement distribution, gridlock, EKF multi-sensor tracking, divergence testing — need to be extended to the JADC2 scale, adapted for GPS-denied conditions, and preserved at the interface between the tracking layer and the AI reasoning layer that CEC's architecture predates.

JADC2: NTDS at Scale Across All Domains

JADC2 is the Department of Defense's multi-decade program to connect sensors, shooters, and decision-makers across all military domains and services into a unified AI-enabled network. Each service contributes its own program: the Air Force's Advanced Battle Management System (ABMS), the Army's Project Convergence, the Navy's Project Overmatch, and the Space Force's National Defense Space Architecture. The January 2023 GAO report on JADC2 found that DoD had released initial strategy guidance but had not yet defined which existing systems would contribute to JADC2, what future capabilities needed to be developed, or the scope, cost, and schedule of the overall effort.⁸ The 2024 defense budget requested $1.4 billion in research and development funds for JADC2-related programs, with services' contributions growing by over $400 million from 2022 to 2024.⁹

The Navy's Project Overmatch — receiving $226 million in FY2023 and $192 million in FY2024, with $716.7 million planned across the five-year defense program — is described by its program manager, Rear Admiral Doug Small, as "creating a more interoperable force, allowing more pieces of the Navy — more ships, more aircraft, more unmanned systems — to connect with one another." The program's June 2025 announcement of a Maven capability, developed with Palantir, described "a unified tactical display that provides insight on vessels across the world" using artificial intelligence as the fusion and presentation layer.¹⁰

JADC2's sensor integration scope dwarfs NTDS's. Where NTDS connected shipboard radars within a task force via Link 11 at data rates of 1,364 bits per second per track, JADC2 is designed to fuse data from: shipboard AESA radars; airborne early warning aircraft; F-35 and F-22 sensor networks (which the GAO documented as operating on incompatible tactical data links¹¹); space-based sensors from the National Defense Space Architecture; ground-based missile defense radars; unmanned maritime surface and subsurface platforms; and coalition partner systems operating under different data standards and security classification levels. Each sensor contributes measurements in its own coordinate frame, with its own error characteristics, its own update rate, its own latency through the data link, and its own systematic biases.

The F-22/F-35 Interoperability Problem as a Modern NTDS Echo

The F-22 Raptor and F-35 Lightning II were designed with incompatible tactical data links — the F-22 uses IFDL (Intra-Flight Data Link) and the F-35 uses MADL (Multifunction Advanced Data Link), and the two systems cannot communicate directly. As a result, an F-22 and an F-35 operating in the same airspace cannot share track data without an intermediate relay node, introducing latency and potential association errors at the relay point. This is the modern equivalent of NTDS's inter-ship coordinate frame misalignment: sensors that should be contributing to a common picture are operating in different communication frames that require mediation, and that mediation introduces both latency and potential data integrity issues. JADC2's explicit goal of addressing this incompatibility is the same goal that NTDS was designed to solve for radar data — connecting sensors that were designed without interoperability in mind.

The GPS Threat: Return of the Navigation Error

The adversary electronic warfare strategy that most directly unmasks the NTDS-era problems in JADC2 is GPS denial and spoofing — and that strategy has moved from theoretical concern to documented operational practice at an accelerating rate.

The FAA documented a 65% increase in GPS signal loss events per 1,000 flights in the first half of 2024 compared to the same period in 2023.¹² OPSGROUP, the international aviation safety organization, estimated a 500% increase in GPS interference events over the course of 2024.¹³ The European Aviation Safety Agency issued a safety bulletin in July 2024 warning of the increasing frequency and multitude of impacts from GPS interference.¹⁴ Estonia's air navigation authority recorded over 600 dangerous GPS incidents in a single month following Russian jamming operations in the Baltic region. Aireon's analysis of aviation GPS anomaly trends through January 2025 showed consistent increases in spoofing indicators — duplicate position reports, position errors exceeding 20 nautical miles, and IPC flag events — distinct from the jamming indicators that had been more stable.¹⁵

The geographic distribution of interference is expanding beyond active conflict zones. The FAA issued NOTAMs in early 2026 advising caution over Mexico, Central America, parts of South America, and portions of the Pacific Ocean, citing potential military activity and GNSS disruption.¹⁶ The Korean Peninsula and areas around Beijing are documented interference hot spots. The Indian-Pakistani border region has emerged as a new GPS denial zone. These are not random electronic noise — they are deliberate denial and deception operations conducted by state actors for strategic purposes.

For airborne military platforms operating in a peer conflict environment, the GPS denial threat is qualitatively more severe than for surface ships. An airborne platform at 20,000 feet has line-of-sight to a surface jammer at roughly 170 nautical miles — ten times the surface horizon. Carrier aircraft operating in the contested airspace around a peer adversary's A2/AD system are exposed to GPS jamming throughout their flight profile. Tactical-grade airborne IMUs, constrained by aircraft size, weight, and power budgets, achieve drift rates of 0.1 to 1 nautical mile per hour — two to three orders of magnitude worse than the navigation-grade ring laser gyro installations on surface ships. An aircraft GPS-denied for 30 minutes in a maneuvering profile may accumulate navigation errors of 1 to 3 nautical miles — recreating the pre-GPS NTDS era navigation uncertainty that systematically corrupted inter-platform track association.

The vertical channel compounds this further. Airborne inertial navigation suffers an inherently unstable vertical channel — aircraft climbing and descending experience vertical accelerations indistinguishable from gravity variations without external reference. GPS stabilizes the vertical channel; GPS denial leaves altitude errors growing at rates that depend on maneuvering history and accelerometer quality. An F/A-18 that has been GPS-denied for 15 minutes in a tactical maneuvering profile may have altitude errors of hundreds of feet and position errors that translate into targeting errors of comparable magnitude.

China's A2/AD GPS Targeting

Chinese People's Liberation Army doctrine explicitly identifies adversary GPS dependency as a primary vulnerability to be exploited in the opening phase of a high-intensity conflict. PLA publications describe comprehensive electronic attack plans targeting GPS reception for both space-based and airborne assets in the Western Pacific theater. The scenario that JADC2 is designed for — a high-intensity conflict in the Western Pacific — is precisely the scenario where GPS denial is most capable, most prepared, and most deliberately targeted. JADC2's sensor fusion architecture was developed primarily in a GPS-available environment. Its performance in the GPS-denied environment that peer adversary doctrine is designed to create has not been publicly characterized by any DOT&E assessment or GAO report reviewed for this article.

The JADC2 Multi-Sensor Correlation Problem: Old Wine in New Architecture

JADC2's vision of fusing sensors across domains, services, and coalition partners introduces versions of all three NTDS problems simultaneously, with geometrically and organizationally greater complexity.

The data association problem operates at larger scale. A JADC2 common tactical picture that incorporates contacts from shipboard radars, airborne AEW, F-35 sensor fusion, unmanned maritime platforms, and space-based sensors contains contacts whose spatial density in contested environments may exceed what any classical assignment algorithm can process in real time. The computational complexity that made JPDA intractable for NTDS-scale contact populations in the 1990s applies at proportionally greater severity for JADC2-scale multi-domain fusion. Modern GPU acceleration has shifted the tractable cluster size substantially, but the fundamental #P-complete scaling of joint event enumeration is unchanged. Dense decoy environments — drone swarms with Luneberg lens reflectors designed to inflate the contact population — exploit this scaling property directly, potentially consuming the fusion system's computational resources during the engagement window.

The sensor alignment problem takes on additional dimensions with heterogeneous multi-modal sensor fusion. Fusing a shipboard AESA radar track with a carrier aircraft's EO/IR track of the same contact requires correctly characterizing the cross-sensor covariance — how errors in the radar's measurement are correlated with errors in the EO/IR system's measurement. These cross-sensor covariances, which describe the statistical relationship between two fundamentally different measurement modalities with different geometry, different error sources, and different reference frames, are among the least well-characterized parameters in any real sensor fusion system. They are not in the calibration data packages for individual sensors, because they are properties of the system-of-systems rather than any individual sensor. In an operational deployment, they are estimated implicitly from track quality statistics — the innovation sequence that Bar-Shalom identified as carrying quality information that higher-level systems typically discard.

The navigation coordinate frame problem is most acute in the GPS-denied airborne sensor scenario, as described above. But it extends to coalition partners as well. An allied navy operating NTDS-derived or NATO-standard sensor systems with different navigation reference architectures may introduce systematic coordinate frame offsets into the JADC2 common picture that the fusion system cannot distinguish from genuine contact position disagreements. Coalition interoperability exercises document navigation frame alignment as a persistent source of track correlation errors that require manual reconciliation during exercises — a labor-intensive process that is not available at machine speed in a real engagement.

What the AI Layer Cannot See

JADC2's AI systems — including the LLM and neural network components being integrated into Project Overmatch through partnerships like Palantir's Maven capability — receive as input the output of the sensor fusion layer: a track picture comprising position estimates, velocity estimates, and associated covariances for each contact, along with attribute data including IFF status, ESM emitter identification, track quality indicators, and tactical context. This is the same information format that NTDS delivered to combat direction operators, now presented to an AI reasoning system rather than a human operator.

What the AI cannot see in this input — because it is not there — is the history of association decisions that produced each track, the sensor alignment calibration state at the time of each measurement, the navigation error accumulation since the last GPS fix for each contributing sensor platform, or the confidence of the track registration algorithm's convergence. A track that has been coherently maintained for 200 scans from a well-calibrated radar with GPS-quality navigation and a correctly estimated azimuth bias looks identical in the track file to a track assembled from the residual association errors of a computationally degraded fusion algorithm processing a decoy-inflated contact environment with GPS-denied airborne sensor contributions. Both tracks have a position, a velocity, a covariance, and attribute data. Neither carries a reliable indicator of its quality as a representation of physical reality.

The AI system generates its engagement recommendation from whichever track picture it receives, with the confidence that its inference statistics produce. If the track picture is corrupted by systematic sensor alignment biases — as it will be in GPS-denied environments where airborne navigation has degraded — the AI's engagement recommendation is based on a falsified representation of the contact's position. If the track identity has been corrupted by a crossing-track swap in the association algorithm — as it will be in dense threat environments with deliberate decoy employment — the IFF and ESM attributes the AI is reasoning about belong to the wrong physical object. The AI cannot know either of these things from the track data it receives, because the information that would reveal them — the association history, the innovation sequence, the registration algorithm's confidence — is not propagated to the AI's input layer.

Bar-Shalom's Innovation Sequence and What JADC2 Discards

Yariv Bar-Shalom's foundational work on tracking and data association identified the innovation sequence — the series of differences between predicted and actual measurements across scans — as the primary carrier of track quality information. A track with consistently large innovations has a model that is wrong: either a bad association or an incorrectly estimated motion model. Using the innovation sequence to detect association errors and flag degraded track quality is mathematically correct and operationally valuable. Modern tracking systems generate this information. JADC2's sensor-to-AI interface typically does not propagate it to the AI reasoning layer — the interface presents processed track state and covariance, discarding the diagnostic information that would tell the AI whether the track is reliable. This is not a design choice made consciously by JADC2 architects — it reflects the historical convention of presenting track data rather than filter internals to decision systems, a convention established before AI systems were expected to make autonomous engagement decisions based on track quality they cannot independently assess.

The Comparison: NTDS Problems versus JADC2 Problems

Problem Dimension	NTDS Era (1960s–1990s)	JADC2 Era (Present)	Trend
Data association scale	N tracks, M measurements, single domain; shipboard radar contacts within task force	Multi-domain, multi-service, coalition; airborne, surface, space, unmanned; decoy-inflated contact populations	Substantially worse
Association algorithm capability	Rule-based correlation; Mahalanobis LAP; JPDA computationally intractable in dense environments	Deep MOT, GLMB filters, GPU-accelerated; better in benign environments, same fundamental scaling limits in dense adversarial environments	Better in typical case; same fundamental limit under adversarial stress
Navigation error — ships	SINS drift ~1 nm/hr; inter-ship misalignment dominant correlation failure; pre-GPS era 1960s–1994	Modern RLG/FOG IMU; daily GPS update adequate; ships geometrically protected from surface jamming; IMU bridge to GPS denial manageable	Substantially better for surface ships
Navigation error — airborne	Periodic INS updates; tactical IMU quality limited; 1-3 nm/hr drift typical	GPS normally available; GPS-denied: tactical IMU 0.1–1 nm/hr; 500% increase in GPS interference 2024; airborne platforms directly threatened by surface jammers at 170 nm LOS	Worse in peer conflict GPS-denied environment
Sensor alignment bias	Mechanical antenna biases; encoder errors; calibrated periodically; known problem	AESA phase calibration drift; carrier structural flexure; multi-modal cross-sensor covariance uncharacterized; coalition partner alignment unknown	Similar magnitude; different sources; harder to characterize across heterogeneous sensors
Sensor interoperability	Homogeneous radar sensors; Link 11 data link; single domain; U.S. only task forces	Radar, EO/IR, ESM, SIGINT, space-based; F-22/F-35 incompatible data links; multi-service; coalition partners with different standards	Substantially more complex
Decision system	Human combat direction operators with track picture; expert judgment about track quality	AI systems receiving processed track picture; no access to association history or filter internals; cannot assess input quality from output data	AI blind to input quality degradation that operators could detect contextually
Adversarial exploitation	Electronic jamming of individual radars; navigation disruption limited by available technology	GPS spoofing/jamming at scale; deliberate track swap attacks via decoys; prompt injection through sensor data manipulation; A2/AD doctrine explicitly targeting GPS dependency	Substantially more sophisticated and deliberate
Consequence of error	Track quality degradation; operator confusion; delayed engagement decisions; fratricide risk managed by human judgment	AI engagement decisions on corrupted track picture; fratricide mechanism through confident wrong recommendations; no human deliberation in compressed decision timelines; accountability vacuum	Categorically more dangerous with autonomous engagement authority

The Testing Gap That No Program Has Closed

DODD 3000.09 requires that autonomous weapons systems be tested to "function as anticipated in realistic operational environments against adaptive adversaries taking realistic and practicable countermeasures."¹⁷ GPS jamming and spoofing are realistic countermeasures that peer adversaries have demonstrated in operational environments. Dense decoy employment designed to inflate contact populations and exploit association algorithm failure modes is doctrinal for peer adversaries in anti-access/area denial scenarios. Dynamic sensor alignment errors from carrier flight operations, airborne platform GPS denial, and coalition partner coordinate frame misalignment are operational realities of the environment JADC2 is designed for.

No publicly available GAO report, DOT&E assessment, or JADC2 program documentation reviewed for this article describes AI system testing in a GPS-denied, high-jamming, decoy-dense environment representative of peer adversary A2/AD operations. The GAO's January 2023 JADC2 report found that DoD had not yet defined which existing systems would contribute to JADC2 or specified future capability requirements — a finding that precedes any meaningful test and evaluation of system performance under degraded conditions.¹⁸ The Air Force's ABMS, evaluated separately by GAO in 2020, was found to lack the planning and cost analysis necessary to ensure it would not slip behind schedule and incur cost overruns — a finding about program management that does not address operational performance under degraded navigation conditions.¹⁹

The NTDS program encountered its multi-sensor correlation and alignment problems through operational experience — in Vietnam, in fleet exercises, and in the analytical work that motivated the development of JPDA, IMM, and track registration algorithms over the subsequent two decades. The problems were discovered, characterized, and partially addressed through the iteration between operational use and technical research that normal program maturation provides. JADC2's AI systems are being deployed into the processing chain before the equivalent iteration has occurred for the GPS-denied, adversarially complex operational environment — and the AI layer introduces an additional abstraction that makes operational discovery of input quality problems less likely, because the AI produces confident outputs regardless of whether its inputs are reliable.

The sensor fusion problems that made NTDS harder than expected were never solved. They were masked by GPS for thirty years. The adversary knows this. The question is whether JADC2's architects do. — Author's assessment, derived from NTDS operational experience and current JADC2 program documentation

What Should Be Required

The technical requirements implied by this analysis are demanding but specific. They are not arguments against JADC2 — the operational need for multi-domain, multi-service sensor fusion is real and urgent. They are arguments for an honest accounting of what the technology can and cannot do in the operational environment for which it is being developed.

First: GPS-denied operational testing must be a mandatory program requirement. DODD 3000.09's "realistic operational environment" standard cannot be met by testing in GPS-available conditions for a program whose operational scenario explicitly assumes GPS denial. Every AI system in the JADC2 sensor-to-shooter loop must be evaluated against a track picture generated by a sensor fusion layer operating with GPS-denied airborne contributors, realistically-biased sensor alignment errors, and a contact population representative of peer adversary decoy employment doctrine. The tested reliability must meet the system's specified engagement performance threshold under these conditions, not under nominal GPS-available conditions.

Second: Track quality metadata must be propagated to AI reasoning layers. The innovation sequence, association history, registration algorithm confidence, and sensor alignment calibration status — the information that carries the quality of the track picture — must be included in the data provided to AI systems making engagement recommendations. An AI system that cannot receive track quality information cannot assess whether its inputs are reliable enough to support an autonomous engagement decision. The Bar-Shalom innovation sequence is generated within the tracking filter and is available — the engineering decision not to propagate it to higher-level systems was made before AI systems were expected to make autonomous engagement decisions, and it should be revisited.

Third: Adversarial exploitation testing must be included. DODD 3000.09's requirement for testing "against adaptive adversaries taking realistic and practicable countermeasures" must include testing specifically against decoy-induced contact density, deliberate track swap attacks through coordinated close-approach geometries, and GPS spoofing designed to corrupt airborne platform navigation without triggering spoofing detectors. These are not exotic threats — they are documented adversary capabilities that are central to peer adversary A2/AD doctrine.

Fourth: The interoperability testing gap must be closed before not after integration. The F-22/F-35 data link incompatibility, coalition partner coordinate frame misalignment, and multi-modal cross-sensor covariance characterization represent known interoperability problems that introduce systematic errors into the JADC2 common picture. These errors must be characterized through joint testing before AI systems are integrated into the fusion layer, not discovered through operational experience after AI-enabled engagement decisions have been made based on a track picture whose interoperability errors were not characterized.

Conclusion

The Engineering and Technology History Wiki preserves a remark from the NTDS program's earliest operational days that remains instructive sixty-five years later: the title of one chapter is simply "No Damned Computer is Going to Tell Me What to Do." The commanding officers who said that were wrong about NTDS — the system provided genuine operational value and was the direct ancestor of every subsequent naval C2 system including Aegis. But their instinct that the technology needed to earn operational trust through demonstrated performance in realistic conditions before receiving autonomous authority was correct — and it is an instinct that JADC2's acquisition trajectory is not honoring.

The three fundamental problems of NTDS-era multi-sensor fusion — data association ambiguity, sensor alignment bias, and coordinate frame misalignment from navigation error — were not solved. They were masked by GPS. Modern IMUs have extended the time between required GPS updates for surface ships to manageable intervals, and surface ship GPS denial is geometrically bounded. For airborne sensors in a peer conflict GPS-denied environment, the original NTDS-era navigation errors return in their full severity, inflating the inter-platform association failures that GPS had suppressed.

JADC2's AI systems receive a track picture that carries these errors invisibly. The processed track state and covariance presented to the AI reasoning layer does not include the innovation sequence, association history, or registration confidence that would characterize track quality. The AI cannot distinguish a coherent, well-supported track from a chimera assembled from the residual association errors of a degraded fusion algorithm working with GPS-denied airborne sensor inputs and a decoy-inflated contact population. Its confident engagement recommendation is the same in both cases.

No publicly available program document, GAO report, or DOT&E assessment documents testing of JADC2 AI systems in this GPS-denied, adversarially complex environment. The same institutional failure that put EMALS in CVN-78 before land-based testing was complete — the organizational priority of fielding capability over demonstrating that the capability works in the operational environment — is operating in JADC2's AI integration program. The discovery that the AI works differently in GPS-denied adversarial conditions will come operationally, at operational cost, from an engagement decision made on a track picture whose corruption the AI could not detect. The NTDS program learned its multi-sensor correlation lessons through decades of operational experience. The pace of JADC2 development is not providing an equivalent learning period before autonomous engagement authority is conferred.

Notes and Sources

Engineering and Technology History Wiki. (Last updated May 12, 2021). "First-Hand: No Damned Computer is Going to Tell Me What to DO — The Story of the Naval Tactical Data System, NTDS." [Secretary Connally; Admiral Burke; five-year versus thirteen-year development; commanding officer resistance.] ethw.org
Wikipedia contributors. (2025). "Naval Tactical Data System." [Sperry Rand UNIVAC computers; Collins Radio; Hughes Aircraft; AN/UYK-7 computer characteristics; 30-bit word; core memory; 600K ops/second.] wikipedia.org
Grokipedia. (January 17, 2026). "Naval Tactical Data System." [October 1961 first deployment USS Oriskany and USS King; Link 11 data link; NATO exports by December 1966.] grokipedia.com
Military History Fandom Wiki. "Naval Tactical Data System." [Two major problems: isolated ship views; need for common picture.] military-history.fandom.com
Engineering and Technology History Wiki. (May 12, 2021). "First-Hand: The Naval Tactical Data System in Combat — Chapter 7." [PIRAZ operations; pilot rescue missions; operational use in Vietnam; IFF monitoring.] ethw.org
Bar-Shalom, Y. & Fortmann, T.E. (1988). Tracking and Data Association. Academic Press, Orlando. [JPDA development; joint event enumeration; Mahalanobis distance; Gaussian measurement noise assumptions; LAP formulation.] ISBN 0-12-079760-7.
GlobalSecurity.org. "E-2C Hawkeye." [Block 2 aircraft with JTIDS/Link 16 installation; Link 11 NTDS data link on all E-2C aircraft; JTIDS enables network-derived registration through joint contacts observable from altitude geometry distinct from surface ship geometry.] globalsecurity.org
Wikipedia contributors. (2025). "Grumman E-2 Hawkeye." [E-2D first flight August 2007; IOC October 2014; AN/APY-9 UHF AESA radar; integrated CEC, Link 16, TTNT; T4O copilot tactical role; VAW-125 first delivery March 2014.] wikipedia.org
Army Recognition. (April 2, 2026). "U.S. Navy Command Aircraft E-2D Hawkeye Central to Air Operations in Epic Fury Over Iran." [AN/APY-9 360° coverage; CEC fire-control-quality track distribution; over-the-horizon targeting; theater C2 role as E-3 Sentry fleet declines; JADC2 integration.] armyrecognition.com
Wikipedia contributors. (2025). "Cooperative Engagement Capability." [JHU/APL conception early 1970s; Battle Group AAW Coordination; first at-sea experiment 1990; Navy acquisition program 1992; EMD 1995; first fleet deployment 1998; Raytheon Systems Co. development with JHU/APL.] wikipedia.org
GlobalSecurity.org. "Cooperative Engagement Capability." [Raw measurement distribution rather than processed tracks; identical algorithms at each CEP producing common air picture; jam-resistant pairwise data link; gridlock and fire-control quality composite tracks; remote missile engagement capability.] man.fas.org; and JHU/APL. (1995). "The Cooperative Engagement Capability." Johns Hopkins APL Technical Digest 16, No. 4, pp. 377–396. [CEP processing subfunctions: track filtering, track divergence and convergence testing, gridlock, sensor interfacing; stringent time budget; unfiltered measurement sharing.] jhuapl.edu
Grokipedia. (January 14, 2026). "Cooperative Engagement Capability." [2024 Chinese military report: plans to use integrated jamming platforms to degrade radar-sensor accuracy and disrupt CEC data fusion, isolating platforms and reducing cooperative engagement effectiveness.] grokipedia.com
Army Recognition. (May 2026). "E-2D Advanced Hawkeye." [NIFC-CA architecture; F-35C passive sensor data relayed through E-2D via CEC to surface ships for over-the-horizon engagement; Northrop Grumman; 70 E-2Ds supporting global operations; Japan, France, Egypt, Taiwan operating E-2 platforms.] armyrecognition.com
Author's assessment based on GPS line-of-sight geometry: surface jammer at 10m ASL, ship GPS antenna at 20m ASL, horizon ~7 nm; ring laser gyro drift rates 0.0001 deg/hr; fiber optic gyro drift rates 0.00001 deg/hr achievable for navigation-grade systems. See also: Honeywell Aerospace. (November 14, 2025). "In GPS Denial: Addressing the Jamming and Spoofing Challenge." [GPS jamming and spoofing now threatening 1,500+ flights daily.] aerospace.honeywell.com
Government Accountability Office. (January 13, 2023). "Battle Management: DOD and Air Force Continue to Define Joint Command and Control Efforts." GAO-23-105495. [JADC2 lacks definitions of which systems contribute; no scope, cost, or schedule defined; ABMS two efforts; F-35 prototype plans shifted.] gao.gov
Center for Strategic and Budgetary Assessments. (2024). "JADC2 Policy Brief." [FY2024 JADC2 budget request $1.4B R&D; service contributions grew $400M from 2022 to 2024; CSBA program analysis.] csbaonline.org
Defense Visual Information Distribution Service. (June 18, 2025). "Commercial Tech Partnerships Drive Unprecedented Progress for Project Overmatch and Navy Capability." [Maven capability with Palantir; unified tactical display; Open DAGIR ecosystem; Project Overmatch FY2023 $226M, FY2024 $192M, five-year $716.7M.] dvidshub.net
Wikipedia contributors. (2025). "Joint All-Domain Command and Control." [F-22 IFDL and F-35 MADL incompatible data links; ABMS; Project Convergence; Project Overmatch; National Defense Space Architecture.] wikipedia.org
GlobalAir.com. (April 14, 2026). "FAA flags global surge in GPS jamming and spoofing, updates its playbook." [65% increase GPS signal loss per 1,000 flights first half 2024 vs 2023; eight hot spots; FAA GPS Interference Resource Guide v1.1; NOTAMs covering Pacific, Mexico, Central America.] globalair.com
OPSGROUP data cited in: Aireon. (May 2025). "Observations of Trends in GPS Anomalies Affecting Aviation." [500% increase in GPS interference events 2024 per OPSGROUP; 80% increase in outage events 2021–2024 other models; EASA safety bulletin July 2024.] aireon.com
EASA. (July 2024). GPS Safety Bulletin. Cited in Aireon White Paper, op. cit. [Frequency and multitude of impacts increasing; potential impacts including navigation errors, diversions, potential controlled flight into terrain.]
Aireon, op. cit. [GPS anomaly metrics August 2024 through January 2025; spoofing indicators: position errors >20 NM, duplicate reports, IPC flag; jamming indicators: Low PIC, FTC0.] aireon.com
FAA GPS Interference Resource Guide v1.1. (December 2025; updated April 2026). Cited in GlobalAir.com, op. cit. [NOTAMs issued January 2026 for Mexico, Central America, portions of Pacific Ocean; Korean Peninsula and Beijing areas documented hot spots.]
Department of Defense. (January 25, 2023). DoD Directive 3000.09, Autonomy in Weapon Systems. [Testing requirement: "function as anticipated in realistic operational environments against adaptive adversaries taking realistic and practicable countermeasures."] esd.whs.mil
GAO-23-105495, op. cit. [DOD has not defined which existing systems contribute to JADC2; scope, cost, schedule undetermined; House report directed DOD to report on these elements.]
FedScoop. (April 17, 2020). "Air Force network of sensors needs better analysis and guidance, watchdog finds." [GAO 2020 report on ABMS; "nontraditional approach"; lacks planning and cost analysis; risk of schedule delays and cost overruns.] fedscoop.com
Bar-Shalom, Y. & Kirubarajan, T. (2001). Estimation with Applications to Tracking and Navigation: Theory, Algorithms and Software. John Wiley & Sons. [Innovation sequence as track quality indicator; IMM estimator for maneuvering targets; track registration and bias estimation theory.] ISBN 978-0-471-41655-5.

Search This Blog

State of The Nation

From NTDS to JADC2: The Unsolved Problems of Multi-Sensor Fusion