1. Origins: A Corporate Gamble That Paid Off
The Lynx story begins in the mid-1990s, when General Atomics
Aeronautical Systems, Inc. (GA-ASI) was still an insurgent company
building its reputation on the original Predator. The firm made an
unusual decision: it would fund the development of a high-performance
SAR entirely with corporate money, partnering with Sandia National
Laboratories for the radar design and image-formation algorithms. At the
time, tactical SAR systems were heavy, expensive, and tightly
controlled by established prime contractors. GA-ASI wanted a radar small
enough for the I-GNAT and Predator airframes, capable enough to compete
with manned reconnaissance sensors, and affordable enough to sell
commercially.
Sandia brought deep expertise. The laboratory had been building and
flying SAR systems since the 1980s, developing key algorithmic
innovations including Overlapped-Subaperture (OSA) processing and
Phase-Gradient Autofocus (PGA)—techniques that allowed extremely fine
resolution images to be formed in real time despite platform motion
errors. Armin Doerry, a distinguished radar engineer at Sandia and a
co-developer on the original Lynx team, would later receive SPIE Fellow
recognition for work that spanned much of the growth arc of modern
airborne SAR. His publications on SAR performance limits, motion
compensation, and impulse response analysis provided the theoretical
bedrock on which the Lynx architecture rested.
Flight tests commenced in July 1998 aboard Sandia’s DOE DeHavilland
DH-6 Twin-Otter. By March 1999 the radar was flying on a GA I-GNAT UAV.
Within two years, Sandia had built two units and GA was constructing a
third. The image quality goals were met or exceeded in manned flight
tests, and by the early 2000s the radar was heading toward military
qualification under the AN/APY-8 designation.
2. System Architecture: State of the Art, Circa 1999
The original Lynx is best understood as two assemblies: a Radar
Electronics Assembly (REA) housed in a VME-bus chassis, and a Gimbal
Assembly containing the antenna, front-end microwave components, and
motion-measurement hardware. Combined weight was approximately 125
pounds (57 kg), with minor variation depending on cable assemblies for
different platforms.
RF Front End
Lynx operates in the Ku band between 15.2 and 18.2 GHz—a frequency
chosen for its favorable compromise between atmospheric propagation,
achievable resolution for a given antenna size, and low probability of
intercept in the dense electromagnetic spectrum below X band. The
transmitter is a traveling-wave tube amplifier (TWTA) producing 320
watts at 35-percent duty factor, averaged across the operating band. The
receiver employs a low-noise amplifier delivering an overall system
noise figure of approximately 4.5 dB—reasonable for the era, though
today’s GaN-based front ends can cut that by a decibel or more.
The antenna is a vertically polarized horn-fed dish with a 3.2-degree
azimuth beamwidth and a 7-degree elevation beamwidth, mounted on a
custom three-axis gimbal designed and built by Sandia. All front-end
components ride on the inner gimbal, which enables the wide squint-angle
range (±45 to ±135 degrees off the velocity vector in stripmap mode)
that gives operators the flexibility to image on either side of the
aircraft across a wide field of regard.
Signal Processing
The digital waveform synthesizer generates a linear FM chirp with
42-bit parameter precision at 1 GHz. The analog-to-digital converter
operates at 125 MHz and provides 8-bit data, which is presummed and
preprocessed before transmission across a RACEway bus to the signal
processor—sixteen nodes of Mercury Computer Systems 200-MHz PowerPC
processors, implementing a scalable architecture for image formation.
Four additional nodes handle motion measurement, radar control, and
optional data recording.
Image formation in all SAR modes uses stretch processing (de-ramping
the received chirp prior to digitization), followed by the OSA algorithm
and PGA autofocus. This pipeline produces either complex (undetected)
images for coherent change detection or detected images for the operator
display. The motion compensation philosophy is to correct as early as
possible in the signal path: transmitted waveform parameters and pulse
timing are adjusted in real time to collect optimal data on the desired
space-frequency grid, minimizing subsequent interpolation. Residual
spatially variant phase errors are compensated during OSA processing,
and any remaining unsensed-motion artifacts are cleaned up by autofocus.
Navigation and Motion Measurement
The inertial measurement system centers on a Litton LN-200 fiber
optic IMU mounted directly on the antenna back, augmented by a
carrier-phase GPS receiver from Interstate Electronics Corporation. A
Kalman filter fuses these measurements to estimate position and velocity
with the accuracy required for sub-foot SAR resolution. This
architecture—IMU on the antenna itself—was a critical design choice: it
eliminates the flexure-related errors that plague systems where the IMU
sits in the airframe and the antenna is separated by compliant
structure.
Lynx Original (Block 10/20) Key Specifications
| Parameter | Spotlight | Stripmap | GMTI |
|---|
| Resolution | 0.1 to 3.0 m | 0.3 to 3.0 m | — |
| Slant Range | 4–25 km (3–60 km derated) | 7–30 km (3–60 km derated) | 4–25 km |
| Min. Detectable Velocity | — | — | 5.8 kts (at 35 m/s near range) |
| Min. Target Cross-Section | — | — | +10 dBsm |
| Angular Coverage | ±50–130° | ±45–135° | ±135° (270° total) |
| Frequency | Ku band, 15.2–18.2 GHz |
| Transmit Power | 320 W (TWTA), 35% duty factor |
| System Weight | ~125 lb (57 kg) |
| Noise Figure | ~4.5 dB |
3. Operational Evolution: Block 20A, Block 30, and Beyond
The Lynx that flies today on MQ-9 Reapers worldwide is not the radar
that first imaged the Belen railroad bridge over the Rio Grande in 1999.
GA-ASI’s Reconnaissance Systems Group has incrementally upgraded the
system through several block configurations, adding capabilities that
the original architecture anticipated but did not initially implement.
The Block 20A introduced Dismount Moving Target Indicator (DMTI)
mode—capable of detecting individual people on foot, not merely
vehicles—and a Maritime Wide Area Search (MWAS) mode that extends
coverage to over 80 kilometers of ocean surface for vessel detection.
VideoSAR, demonstrated in 2013, provided continuous real-time
high-definition SAR surveillance in full 1080p video format, bridging
the gap between the traditional SAR “snapshot” and full-motion video
from EO/IR turrets. The Block 30 Lynx saw deployment in NATO exercises
including Unified Vision 2012 at Ørland, Norway, where the radar
provided uninterrupted all-weather ISR coverage across land and maritime
domains alongside allied ISR assets.
Automated exploitation tools have also evolved. Amplitude Change
Detection (ACD) and Automated Man-Made Object Detection (AMMOD)
algorithms allow rapid comparison of SAR image pairs, highlighting
differences that human analysts might miss in the torrent of data from a
long-endurance mission. The Claw sensor-control and image-analysis
software, developed by GA-ASI’s Reconnaissance Systems Group, provides
the integration layer that cross-cues the Lynx radar with the aircraft’s
EO/IR turret—a capability demonstrated repeatedly in combat operations
and exercises.
Today, GA-ASI reports that the Lynx Multi-mode Radar is deployed by
the U.S. Air Force, U.S. Army (on Gray Eagle and the former Sky Warrior
Alpha), the Royal Air Force, the Italian Air Force, the French Air
Force, the Iraqi Air Force (on Peace Dragon manned ISR aircraft), and
the U.S. Department of Homeland Security. It has flown on platforms
including the MQ-9 Reaper, MQ-1C Gray Eagle, C-12, U-21, DH-7, and King
Air 200.
4. Combat Lessons: The Houthi Problem
The Lynx radar’s operational record cannot be discussed without
confronting the losses. The MQ-9 Reaper, and by extension the Lynx
sensor suite it carries, was designed for permissive
airspace—environments where the platform could orbit at medium altitude
with impunity while the radar and EO/IR sensors surveilled the terrain
below. That assumption has been violently challenged over Yemen.
Since 2023, Houthi forces have shot down multiple MQ-9 Reapers
operating over the Red Sea and Yemeni airspace. At least three were lost
in 2024, a fourth was mistakenly downed by U.S.-backed Kurdish fighters
in Syria, and losses continued to mount through 2025 into 2026. By
April 2026, an estimated 24 U.S. MQ-9s had been lost in the broader
conflict, many to surface-to-air systems that would have been considered
low-tier threats a decade ago.
These losses expose a structural limitation that transcends the radar
itself. The MQ-9 platform is slow (240 knots), non-stealthy, and
predictable in its orbit patterns. The Lynx radar’s 25–30 km operational
slant range at full resolution requires the aircraft to fly within the
engagement envelope of even modest SHORAD systems. A sensor that could
achieve comparable image quality at 80 or 100 km of standoff—or that
could operate from a faster, more survivable platform—would change the
calculus fundamentally.
“The era of permissive-airspace ISR orbits is ending. The sensors on
the next generation of unmanned platforms must see farther, process
faster, and adapt to contested electromagnetic environments.”
5. Where Current Technology Could Take a Lynx-Class Radar
If you were to redesign the Lynx today, starting from its mission
requirements but unconstrained by its 1990s component technology, the
result would be transformational across every subsystem. The physics of
SAR have not changed, but the engineering available to exploit that
physics has advanced enormously. What follows is a
subsystem-by-subsystem assessment of where mature, available technology
could take a Lynx-class radar while maintaining or reducing the original
system’s size, weight, and power envelope.
5.1 Phased Array Antenna Replacing the Mechanical Dish
The single most impactful upgrade would be replacing the mechanically
gimbaled horn-fed dish with an Active Electronically Scanned Array. The
original Lynx antenna—a vertically polarized dish on a custom Sandia
three-axis gimbal—was elegant for its era, but a mechanical gimbal
imposes fundamental constraints: slew rate limits the speed of mode
transitions, the gimbal mechanism is a reliability liability with
bearings, slip rings, and resolvers that wear over time, and the entire
front-end assembly (TWTA, LNA, antenna, IMU) rides on the inner gimbal,
driving weight and moment-of-inertia upward.
However, an AESA introduces a different constraint that must be honestly confronted: field of regard.
A planar AESA can electronically steer its beam approximately ±45 to
±60 degrees from the array normal before the projected aperture becomes
too small and the element pattern rolls off, causing unacceptable gain
loss and beam broadening. The original Lynx’s gimbaled dish, by
contrast, could look anywhere from ±45° to ±135° off the aircraft
velocity vector in stripmap mode, and the GMTI mode scanned ±135° (270°
total). This gave the operator port-and-starboard coverage and
forward-sector GMTI scanning—capabilities that a fixed side-looking AESA
panel simply cannot match.
Several architectural solutions exist, each with tradeoffs. The simplest is a single-axis mechanical repositioner
that physically rotates the AESA face between port and starboard
orientations. The Saab Gripen E’s ES-05 Raven AESA uses exactly this
approach: a roll-repositionable antenna on a single-axis barrel-type
joint that provides a full ±180° field of regard—roughly twice what a
fixed array provides—with connectors adapted from oil-drilling
technology to handle the rotation. The Eurofighter’s Captor-E uses a
dual swashplate repositioner for an even wider field of regard. For a
Lynx-class system, a simple single-axis repositioner that flips the
array between port-looking and starboard-looking orientations would
restore bilateral coverage at the cost of a brief mechanical transition
time (seconds, not the milliseconds of electronic steering) and a modest
weight penalty for the rotary joint. This is far simpler and lighter
than the original three-axis gimbal—one axis of rotation vs. three, no
slip rings for high-power RF (the TWTA is eliminated), and no continuous
tracking requirement since the repositioner only needs to slew between
two or three discrete positions.
A more ambitious approach is dual-face or conformal arrays:
two AESA panels mounted on opposite sides of the aircraft (or a single
conformal array wrapped around the fuselage), providing simultaneous
port and starboard coverage without any mechanical motion. Leonardo’s
Osprey radar is offered with up to four antenna faces for full 360°
coverage on helicopter platforms. For a UAV like the MQ-9 or Gray Eagle,
two panels—one port, one starboard—would restore the original Lynx’s
bilateral capability while adding the ability to image on both sides
simultaneously, a capability the gimbaled dish never had. The cost is
doubled antenna hardware, increased power, and additional thermal
management. For platforms where SWaP and cost are tightly constrained,
the single-face-with-repositioner approach is likely the pragmatic
choice; for larger platforms or satellite applications, dual faces
become feasible.
GA-ASI is already moving toward AESA integration. The EagleEye
radar’s planned AESA antenna, which the company projects will more than
double the radar’s range from roughly 80 km to over 160 km, was in lab
prototype by late 2024 with flight tests planned for 2025. The
field-of-regard question will be a critical design decision for
EagleEye’s operational configuration. For a Lynx-class system, the
AESA’s advantages remain compelling even with the field-of-regard
constraint: electronic beam steering with zero mechanical inertia within
the array’s scan volume (enabling near-instantaneous mode-switching
between SAR, GMTI, and maritime search), graceful degradation as
individual T/R modules fail, low probability of intercept through beam
agility and waveform diversity, and the elimination of the heavy
three-axis gimbal mechanism—a substantial weight and reliability win
even if a single-axis repositioner is retained.
5.1.1 Ku-Band Element Spacing and Packaging Constraints
Operating at Ku band introduces a specific and challenging constraint
that does not arise at the X-band frequencies used by most current UAV
AESA radars (ICEYE, Capella, PicoSAR). To avoid grating lobes when
scanning—spurious beams that radiate energy in unintended directions and
create false targets or ambiguities—the element spacing must not exceed
approximately one-half wavelength (λ/2). At 16.7 GHz (the center of the
Lynx band), λ/2 ≈ 9 mm. This is an extremely tight pitch for packaging
the T/R module electronics behind each element. For comparison, an
X-band array at 10 GHz has λ/2 ≈ 15 mm—nearly 70% more room per element,
and even at X band the packaging challenge is considered significant.
At 9-mm spacing, each element’s unit cell has an area of
approximately 81 mm²—barely larger than a fingertip. The T/R module for
each cell must contain a GaN power amplifier, a low-noise amplifier, a
T/R switch, a phase shifter, a variable attenuator, and associated DC
bias and control circuitry. This drives the design toward 3D multilayer
packaging techniques where the T/R electronics are stacked vertically
behind the radiating element rather than arranged laterally beside it.
Ku-band AESA T/R modules using 3D multilayer technology with LTCC
(low-temperature co-fired ceramic) substrates and vertical
interconnection have been demonstrated, and Analog Devices has noted
that the half-wavelength constraint “creates particularly challenging
designs at higher frequencies where the length of each unit cell becomes
smaller,” driving demand for highly integrated beamformer ICs that
combine multiple channels per chip. The INDIGAM project’s single-chip
front end—integrating PA, LNA, and switch on one GaN die—is precisely
the kind of integration required to fit within a 9-mm Ku-band lattice.
5.1.2 Mutual Coupling and Scan Blindness
The tight element spacing at Ku band also exacerbates inter-element
mutual coupling. When elements are closely spaced, the electromagnetic
fields radiated by one element couple into adjacent elements, altering
their impedance and radiation patterns as a function of scan angle. At
certain combinations of frequency and scan angle, this coupling can
produce scan blindness—a condition where the active reflection
coefficient of the array elements approaches unity and the array
effectively ceases to radiate. Scan blindness is driven by surface wave
modes in the antenna substrate that are excited at specific scan angles,
and the problem becomes more severe as the substrate permittivity
increases and the element spacing approaches the critical threshold.
Mitigation techniques are well documented in the literature and
include electromagnetic bandgap (EBG) structures between elements to
suppress surface wave propagation (demonstrated to reduce mutual
coupling by 15–20 dB), decoupling surfaces placed above the array
aperture, wide-angle impedance matching (WAIM) layers, and careful
selection of element geometry (tightly coupled dipole arrays, or TCDAs,
deliberately exploit inter-element coupling to extend bandwidth and scan
range). A 2024 IET paper demonstrated a Ku/Ka-band shared-aperture
phased array achieving ±60° scanning at Ku band with inter-element
isolation exceeding 15 dB using a hybrid decoupling approach. These
techniques add design complexity but represent solved engineering
problems, not fundamental physics barriers.
5.1.3 Endfire Steering for 360° GMTI Coverage
An intriguing possibility for restoring the original Lynx’s 270° GMTI scanning coverage is to operate the array in an endfire
mode for the GMTI function. In endfire operation, the beam is steered
parallel to the array face (90° from broadside) rather than
perpendicular to it. While endfire operation dramatically reduces the
array’s effective aperture and therefore its gain—because the projected
area seen by a target at endfire approaches zero for a planar array—GMTI
does not require the high SNR needed for fine-resolution SAR imaging.
GMTI operates against moving targets with relatively large radar
cross-sections (the original Lynx specified a minimum detectable target
of +10 dBsm) and uses coherent integration over multiple pulses to build
detection sensitivity. A GMTI mode that accepts 10–15 dB less gain than
the broadside SAR mode could potentially scan through endfire,
providing forward-sector and even rear-sector moving-target detection
from a single planar array—approaching 360° azimuth coverage for the
GMTI function while reserving the high-gain broadside sector for SAR
imaging.
This approach would require careful antenna element design to
maintain acceptable pattern and impedance behavior near endfire—a regime
where mutual coupling effects are most severe and where conventional
patch elements have null radiation. Tightly coupled dipole array (TCDA)
designs, which inherently have broader element patterns than patch
arrays, are better suited to wide-angle and near-endfire operation. The
tradeoff is clear: endfire GMTI would provide degraded sensitivity
compared to broadside operation, but it would cover angular sectors that
a fixed planar array otherwise cannot reach at all, potentially
eliminating the need for a mechanical repositioner for the GMTI mission
while retaining the repositioner only for bilateral SAR coverage.
Leonardo’s ECRS Mk2 AESA radar for the Eurofighter Typhoon, which
completed its first flight trial in September 2024 and is currently
optimizing SAR and GMTI modes through 2025, demonstrates the state of
the art: simultaneous target tracking and electronic attack from a
single aperture, with digital beamforming enabling multiple independent
beams. That paradigm—a single array supporting multiple concurrent radar
functions—is precisely what a next-generation Lynx needs.
5.2 GaN Transmitter Replacing the TWTA
The original Lynx’s 320-watt traveling-wave tube amplifier was the
highest-power, most efficient transmitter available in a Ku-band
airborne package in the late 1990s. But TWTAs are bulky, require
high-voltage power supplies (typically several kilovolts), have finite
tube life, and are single-point failures. Replacing the TWTA with
distributed GaN-based solid-state power amplifiers integrated into each
T/R module of an AESA array transforms the transmitter from a single
fragile component into a distributed, fault-tolerant power source.
The key enabler is gallium nitride semiconductor technology. GaN HEMT
power amplifiers deliver substantially higher power density, wider
bandwidth, and better efficiency than the GaAs devices that dominated
radar front ends for decades. At Ku band, GaN-based single-chip front
ends (SCFEs) integrating the power amplifier, low-noise amplifier, and
T/R switch on a single die have been demonstrated by European consortia
including the ESA-funded INDIGAM project, which achieved 10 watts of
transmit power with 30-percent power-added efficiency and a receive
noise figure below 2.8 dB at 13–16 GHz. South Korea’s ETRI and Wavice
achieved domestically fabricated GaN MMICs for AESA radar and SAR
satellite applications in 2025, establishing performance parity with
U.S. and European foundries.
GaN-on-SiC stacks exploit silicon carbide’s superior thermal
conductivity to draw heat from junction hot spots, enabling higher duty
cycles and longer pulse widths without liquid cooling. Raytheon’s
PhantomStrike AESA, designed for small platforms including unmanned CCAs
and now selected for the DARPA ACE autonomous F-16, demonstrated that a
fully air-cooled GaN AESA radar is feasible—eliminating the liquid
cooling plumbing that would otherwise add weight and complexity. For a
Lynx-class system with, say, 200–400 T/R modules each producing 5–10
watts at Ku band, the aggregate radiated power would exceed the original
TWTA by a significant margin while distributing the thermal load across
the entire array face, enabling passive or forced-air cooling within
the original power budget.
5.3 Improved Low-Noise Amplifier
The original Lynx’s receiver chain delivered an overall system noise
figure of approximately 4.5 dB. Every decibel of noise figure
improvement translates directly into detection range: a 1.5-dB
improvement yields roughly 10 percent more range for the same target
cross-section, or equivalently allows detection of smaller targets at
the same range. Modern GaN LNAs at Ku band have demonstrated noise
figures below 2.8 dB in the INDIGAM single-chip front end, and dedicated
GaAs pHEMT LNA designs from companies like Altum RF push noise figures
below 2.0 dB at X and Ku bands in compact QFN packages. Integrating a
modern LNA into each T/R module’s receive path, positioned immediately
behind the antenna element to minimize transmission line losses, would
cut the system noise figure by 1.5–2.0 dB relative to the original
Lynx—a meaningful sensitivity improvement that translates to extended
range, improved minimum detectable velocity in GMTI, and better
performance against low-observable targets.
5.4 Navigation-Grade IMU and Multi-Antenna GNSS
The original Lynx’s Litton LN-200 fiber optic IMU was a
tactical-grade unit—excellent for the late 1990s, but the state of the
art has moved considerably. Honeywell’s HG3900, an all-silicon MEMS IMU
currently in design verification with initial production planned for
2027, claims near-navigation-grade performance (matching or exceeding
fiber optic gyroscope and ring laser gyroscope stability) in a
dramatically smaller, lighter, lower-power package. For applications
requiring proven navigation-grade performance today, EMCORE’s
closed-loop FOG products (acquired from KVH Industries) deliver bias
stability below 0.05°/hr in compact form factors, with the
navigation-grade DSP-1760 gyroscope offering the world’s smallest
precision FOG with photonic integrated chip technology for improved
reliability. Honeywell has even demonstrated a hybrid quantum-enhanced
FOG with drift below 0.1 m/hr over a 24-hour mission.
Upgrading from the LN-200 to a navigation-grade IMU would improve the
SAR’s ability to maintain coherent apertures during long spotlight
dwells, reduce the burden on autofocus algorithms, and improve
geolocation accuracy for targeting applications—all without increasing
weight.
Equally important is the GPS subsystem. The original Lynx used a
single Interstate Electronics Corporation carrier-phase GPS receiver. A
modernized system should employ a multi-antenna GNSS array—three or four
antennas distributed on the aircraft structure—to provide direct
attitude determination in addition to position and velocity.
Multi-antenna carrier-phase GNSS attitude systems can achieve heading
accuracy of 0.01–0.05 degrees with baselines of one meter or more,
providing an independent check on the IMU’s attitude solution and
dramatically improving heading accuracy during straight-and-level flight
where gyroscope drift is otherwise unobservable. The system should
receive multi-constellation signals (GPS L1/L2/L5, Galileo, GLONASS,
BeiDou) for resilience against single-constellation jamming or spoofing.
Anti-spoofing capability—using angle-of-arrival consistency checks
across the antenna array to detect illegitimate signals—is no longer
optional in contested environments where GPS warfare is an active
threat.
The multi-antenna GNSS solution also eliminates the need for a
dedicated heading sensor or magnetic compass, simplifying the navigation
architecture while improving accuracy. Sony has demonstrated
time-multiplexed single-receiver architectures that switch rapidly among
multiple antennas, achieving attitude determination without requiring a
separate receiver per antenna—a significant SWaP advantage for UAV
integration.
5.5 Simultaneous SAR and GMTI
One of the most consequential limitations of the original Lynx is
that SAR and GMTI are time-sequential modes: the operator selects one or
the other. In the original architecture, this was unavoidable—a single
dish antenna with a single receive channel cannot simultaneously form a
high-resolution SAR image and perform the multi-channel clutter
cancellation required for GMTI. An AESA with digital beamforming changes
this equation fundamentally.
With multiple simultaneous receive beams, a digital AESA can
partition its aperture and processing to perform SAR imaging and GMTI
concurrently. The IEEE literature documents this explicitly:
multichannel SAR systems with digital beamforming can simultaneously
generate ambiguity-free scene images and detect ground moving targets
using adaptive sum-and-difference beam techniques. A 2016 IEEE paper
specifically described a processing architecture for simultaneous SAR,
GMTI, ATR, and tracking from the same radar data. Leonardo’s ECRS Mk2 is
demonstrating exactly this class of concurrent multi-function operation
in its 2025 flight optimization phase.
For the operator, simultaneous SAR/GMTI eliminates the tactical
penalty of mode-switching: moving targets detected by the GMTI beam are
automatically correlated with the SAR scene context, and the SAR image
provides the stationary-scene backdrop against which GMTI tracks become
immediately interpretable. This fusion of stationary-scene imagery with
moving-target tracks, delivered in near-real-time, is what commanders
actually need—and what the original Lynx, despite its individual mode
excellence, could not provide simultaneously. Space-time adaptive
processing (STAP), enabled by the multichannel digital receive
architecture, would also substantially lower the minimum detectable
velocity below the original Lynx’s 5.8-knot threshold, bringing
slow-moving dismounts and vehicles on rough terrain into the detection
space.
5.6 Modern Signal Processing and On-Board AI
The original Lynx’s sixteen 200-MHz PowerPC nodes delivered perhaps
10–20 GFLOPS of sustained throughput for image formation. A single
current-generation embedded GPU—say an NVIDIA Orin or a ruggedized
A100—delivers hundreds of TFLOPS, a four-order-of-magnitude improvement.
This computational abundance enables real-time Video SAR at full
resolution across wider swaths, on-board deep-learning inference for
automatic target recognition (ATR) and change detection, adaptive
waveform optimization based on terrain and target characteristics,
multi-aperture and multi-baseline interferometric processing that would
have been computationally prohibitive in 1999, real-time coherent change
detection without ground-station processing, and the simultaneous
SAR/GMTI processing described above.
The shift from fixed-function DSP pipelines to software-defined,
GPU-accelerated architectures also makes the radar fundamentally more
adaptable. New modes and exploitation algorithms can be deployed as
software updates, not hardware redesigns—a paradigm that GA-ASI has
embraced with EagleEye’s AI/ML-enhanced target detection running on
board the aircraft.
The explosion of deep learning applied to SAR imagery represents the
most consequential change since the radar itself was invented. A
February 2026 survey in Sensors documented the state of the art
across SAR despeckling, segmentation, classification, and detection,
finding that convolutional neural networks are the predominant
architecture but that generative adversarial networks and graph neural
networks remain significantly underutilized and offer substantial room
for improvement. For a modernized Lynx-class system, AI/ML enables
automated GMTI track correlation and behavior analysis, coherent change
detection with learned scene models that dramatically reduce false-alarm
rates, cross-cueing between SAR, GMTI, and EO/IR sensors based on
learned target signatures, and synthetic training data generation using
GAN-based SAR image synthesis. GA-ASI has stated that EagleEye
incorporates real-time AI/ML software running on the aircraft for
improved target detection range, and the broader industry is racing to
embed autonomous exploitation at the sensor edge.
5.7 Wideband Waveform Generation and Phase Noise Reduction
This subsection addresses what may be the single most
under-appreciated performance limitation of the original Lynx: the phase
noise introduced by its waveform generation and upconversion
architecture.
The original Lynx waveform generator operated from a 100-kHz
oven-controlled crystal oscillator (OCXO) that fed a digital
accumulator, which in turn used a sine/cosine lookup table to generate a
coherent chirp waveform at baseband—essentially a classic DDS
architecture with 42-bit phase precision. The choice of a 100-kHz
reference was deliberate and reflects a fundamental tradeoff in
oscillator physics: lower-frequency crystal oscillators exhibit
inherently better phase noise performance than higher-frequency sources,
because the Q factor of a quartz resonator scales inversely with
frequency. Higher-frequency oscillators available in the late 1990s had
even worse phase noise, so the designers accepted the penalty of
subsequent frequency multiplication as the lesser evil—starting from the
cleanest possible source and multiplying up, rather than starting from a
noisier high-frequency oscillator that required less multiplication.
A critical and often overlooked feature of the Lynx DWS was that the
chirp accumulator’s initial value and increment were not fixed
parameters—they were set adaptively as part of the motion measurement
and compensation process. The Kalman filter fusing the LN-200 IMU and
carrier-phase GPS measurements provided real-time estimates of the
aircraft’s position and velocity, which were used to adjust the
transmitted waveform parameters on a pulse-by-pulse basis. This is what
the original Sandia paper meant by “transmitted waveform parameters are
adjusted, as well as pulse timing, to collect optimal data on the
desired space-frequency grid.” By modifying the chirp start frequency
(initial accumulator value) and chirp rate (accumulator increment) based
on the instantaneous platform motion, the system performed motion
compensation at the earliest possible point in the signal chain—before
the signal was even digitized on receive. This minimized the need for
subsequent data interpolation in the image formation processor and was a
key reason the Lynx could achieve 0.1-m resolution from a relatively
unstable UAV platform with only a tactical-grade IMU.
The problem lay in what happened after the DDS: the baseband chirp
had to be transformed into a wideband Ku-band signal through two
distinct operations. First, frequency multiplication was used
to expand the chirp bandwidth to the desired value (the DDS output
bandwidth was far too narrow for fine-resolution SAR). Second, frequency shifting
(mixing with a local oscillator) was used to translate the multiplied
chirp to the desired Ku-band center frequency of 15.2–18.2 GHz. These
are fundamentally different operations from a phase noise perspective.
The multiplication stages were the primary offenders. Each frequency
multiplication by a factor N degrades phase noise by 20 log₁₀(N) dB—this
is a fundamental consequence of the phase being multiplied along with
the frequency. If the original DDS chirp bandwidth was multiplied by a
factor of, say, N = 100 to reach the required 1.5 GHz of chirp bandwidth
for 0.1-m range resolution, the phase noise degradation from
multiplication alone is:
20 log₁₀(100) = 40 dB
The frequency shifting stages (mixers) are less damaging in
principle—an ideal mixer translates the signal to a new center frequency
without multiplying the phase. However, the local oscillator used for
the shift was itself generated from the same OCXO through its own
multiplication chain (the STALO module), adding that chain’s multiplied
phase noise to the output. The LO path from 100 kHz to the ~16 GHz STALO
frequency involved multiplication factors on the order of 160,000×,
contributing 20 log₁₀(160,000) ≈ 104 dB of phase noise degradation on
the STALO signal. While the dechirp process partially cancels the
STALO’s phase noise (since the same STALO drives both transmit and
receive paths), the cancellation is imperfect for targets at non-zero
range delay. The net effect is that both the chirp multiplication chain
and the STALO multiplication chain contributed significantly to the
system’s residual phase noise floor.
The combined result was that even starting from the best available
low-frequency oscillator, the cascaded multiplication stages—in both the
chirp bandwidth expansion path and the STALO frequency generation
path—elevated the phase noise to levels that, in many operational
scenarios, became the predominant source of self-noise after dechirp,
setting the floor for the system’s achievable signal-to-noise ratio
regardless of the thermal noise performance. This was the unavoidable
cost of the 1990s-era design: every clock source has phase noise, and
the architecture required multiplying that noise up by large factors to
reach the operating bandwidth and frequency.
In a stretch-processed (dechirp-on-receive) SAR like the Lynx, phase
noise has a particularly insidious effect. The dechirp mixer multiplies
the received chirp echo against a replica of the transmitted chirp. If
the transmitter and receiver reference the same oscillator (as they do),
phase noise that is common to both paths cancels in the dechirp
process—but only for targets at zero range delay. For targets at
non-zero delay Ï„, the dechirp operation samples the oscillator phase
noise at two different times separated by Ï„, and the residual phase
noise after dechirp is determined by the oscillator’s phase noise power
spectral density (PSD) evaluated at offset frequencies corresponding to
1/Ï„. Doerry analyzed this in detail in a Sandia report on radar receiver
oscillator phase noise, showing that for SAR modes the residual phase
noise manifests as a paired-echo sidelobe structure around each target,
with the sidelobe level set by the integrated phase noise PSD over the
relevant offset frequency range. For coherent change detection—where two
images are interferometrically compared—phase noise in the two images
adds non-coherently, requiring an additional 3 dB of margin. In many
operational scenarios, this oscillator-derived phase noise was the
predominant source of self-noise after dechirp, setting the floor for
the system’s achievable signal-to-noise ratio regardless of the thermal
noise performance.
A modern direct digital encoding architecture eliminates both
multiplication chains at their root, while preserving and enhancing the
adaptive chirp capability that was one of the Lynx’s most innovative
features. Rather than generating a narrow-bandwidth chirp at baseband
and multiplying to expand its bandwidth, a modern high-speed DAC
(running at 10+ GS/s with 14–16-bit resolution) can directly synthesize
the full-bandwidth chirp waveform—1.5 GHz or wider—at an intermediate
frequency, requiring no bandwidth multiplication whatsoever. The chirp
bandwidth is determined digitally by the waveform memory and DAC sample
rate, not by analog multiplier chains. The adaptive chirp
parameters—start frequency, chirp rate, and pulse timing—are set
digitally on a pulse-by-pulse basis by the motion compensation
processor, exactly as in the original Lynx, but without the phase noise
penalty of multiplication. The 42-bit phase precision of the original
accumulator is easily exceeded by modern digital implementations.
For frequency translation to Ku band, the wideband IF chirp is mixed
with a low-noise microwave oscillator that can be generated with far
less multiplication than the original STALO architecture—a modern
low-noise dielectric resonator oscillator (DRO) or sapphire-loaded
cavity oscillator operating directly at 10–15 GHz requires no
multiplication at all, or at most a single ×2 step from a high-quality 8
GHz source (adding only 6 dB of phase noise). With the latest
RF-sampling DACs operating at 12–20 GS/s (such as the Analog Devices
AD9082/AD9084 family or Texas Instruments DAC38RF8x), the chirp can be
synthesized directly at frequencies approaching X band, further reducing
the translation requirement.
The fundamental tradeoff that drove the original
design—lower-frequency oscillators have better phase noise—still
applies. Any clock source has phase noise, and the DAC’s output phase
noise is ultimately limited by the jitter of its sampling clock. But the
critical difference is that a modern architecture does not multiply
that clock phase noise by factors of 100–160,000. A 1-GHz clock with
–155 dBc/Hz phase noise at 1 kHz offset (readily achievable with current
OCXO technology) feeds the DAC directly; the output chirp inherits that
phase noise floor without multiplication. The improvement over the
original Lynx’s cascaded chain is on the order of 20–30 dB—the
difference between phase noise being a binding performance constraint
and phase noise being a secondary contributor well below the thermal
noise floor.
In a digital AESA architecture, the improvement goes further. Each
T/R module contains its own DAC-fed waveform generator locked to a
common distributed clock. The phase coherence across the array is
maintained by the digital clock distribution network, not by analog
phase-locked loops and frequency multipliers. The motion-adaptive chirp
parameters propagated from the navigation Kalman filter can be
distributed digitally to all elements simultaneously, maintaining the
“compensate as early as possible” philosophy across the entire array.
The practical impact on SAR image quality is substantial. Lower phase
noise translates directly to lower sidelobe levels around bright
targets (improving the ability to detect weak targets near strong ones),
better coherent change detection sensitivity (enabling detection of
subtler scene changes), improved GMTI performance (lower phase noise
reduces the clutter residue that limits minimum detectable velocity),
and wider usable dynamic range in the SAR image. For the original Lynx,
phase noise was arguably the binding constraint on CCD performance and
on the achievable image quality in cluttered scenes. Eliminating the
analog multiplication chain is not merely an incremental improvement—it
removes a fundamental architectural limitation that has constrained
stretch-processed SAR performance for decades.
Beyond phase noise, wideband digital waveform generation provides
additional benefits. Wider instantaneous bandwidth directly translates
to finer range resolution; a system with 2–3 GHz of bandwidth at Ku band
could achieve centimeter-class range resolution. Wideband agile
waveforms provide electronic protection: a radar that hops across a
3-GHz-wide band on a pulse-to-pulse basis is far harder to detect, jam,
or geolocate than one transmitting a fixed chirp on a predictable center
frequency. And digitally generated waveforms can be adaptively
optimized in real time—adjusting chirp bandwidth, window functions, and
spectral notching to avoid interference or to tailor resolution to the
target environment.
5.8 Longer Range, Wider Swath
The combined effect of the upgrades described above—higher aggregate
transmit power from a GaN AESA, lower noise figure from modern LNAs,
wider bandwidth, and smarter signal processing—yields dramatic
improvements in the two parameters that matter most operationally:
maximum standoff range and area coverage rate. The original Lynx
achieved 25–30 km slant range at full resolution in weather; a
modernized system with an AESA could operate at 80–160 km, keeping the
platform outside the engagement envelopes of most SHORAD threats. Wider
swath coverage follows from the AESA’s ability to form multiple
simultaneous receive beams (using the scan-on-receive technique) to
capture returns from a broader ground strip without sacrificing range
resolution. Where the original Lynx stripmap mode yielded a 934-meter
view size at 0.3-meter resolution, a digital beamforming architecture
could extend the ground swath to several kilometers at the same
resolution—a step-change in area coverage rate that transforms the
sensor from a soda-straw into a surveillance tool.
5.9 Wiring, Cabling, and Integration
An often-overlooked but operationally significant improvement area is
the cable harness. The original Lynx used custom cable assemblies that
varied by platform, contributing to the weight variance noted in the
specification (about 125 lb with “some variance due to different cable
assemblies for different platforms”). The REA communicated with the
gimbal assembly via dedicated RF, digital, and power cables routed
through the airframe—a design driven by the VME-bus era’s separate-box
architecture.
A modern redesign would consolidate the radar into fewer physical
modules with high-speed serial interconnects (replacing the RACEway bus
and VME backplane with modern standards like 10/40 Gigabit Ethernet or
PCIe-over-fiber), reducing both the number and weight of inter-module
cables. The elimination of the gimbal removes the most complex cable
routing problem entirely—no more slip rings, no more flex cables to the
inner gimbal. Power distribution benefits similarly: the TWTA’s
high-voltage supply (several kilovolts for the tube) is replaced by
lower-voltage DC distribution to the T/R modules, simplifying the power
conditioning electronics and reducing cable insulation weight. Modern
flex-rigid printed circuit interconnects can replace many discrete
cables within the electronics assembly, further reducing weight and
improving reliability by eliminating connectors—each of which is a
potential failure point in a vibration environment. The net effect is a
cleaner, lighter, more reliable installation that is easier to integrate
across different platforms—precisely the “relatively generic packages”
philosophy that Sandia’s original designers aspired to, but now
achievable with far fewer compromises.
5.10 Maintaining or Reducing SWaP
The critical constraint on all of the above is that the modernized
system must fit within the same or smaller size, weight, and power
envelope as the original Lynx. The original system weighed approximately
57 kg (125 lb) and drew power consistent with a medium-altitude UAV’s
payload budget. The good news is that the weight and power reductions
from eliminating major legacy components—the TWTA and its high-voltage
power supply, the three-axis gimbal with its drive motors and resolvers,
the VME chassis with twenty PowerPC boards, and the associated
cabling—create substantial margin for the new components. A GaN AESA
with integrated T/R modules, a single GPU-accelerated signal processor
board, a modern navigation-grade IMU, and multi-antenna GNSS receivers
can be packaged in a form factor that is lighter and more compact than
the original two-box architecture. The thermal management challenge
shifts from cooling a single hot TWTA to managing distributed
lower-level heat across the array face—a problem that GaN-on-SiC’s high
thermal conductivity and air-cooled AESA designs have shown to be
tractable. The target should be 45–55 kg total system weight with
reduced prime power draw—achievable with disciplined systems engineering
and the SWaP advantages of modern semiconductor and packaging
technology.
5.11 Data Links and Distributed Processing
The original Lynx offered two image-transmission paths: an NTSC video
link (treating the radar as “just another sensor”) and a digital data
link for full-resolution NITFS 2.0 imagery. Today, the requirement is
for wideband, LPI/LPD data links that can deliver Video SAR, multi-mode
MTI tracks, and ATR products simultaneously to multiple consumers—ground
stations, other aircraft, and command nodes in a joint all-domain
command and control (JADC2) architecture. On-board AI processing is
critical here: rather than downlinking raw SAR imagery at full
resolution (which saturates any practical data link), the modernized
system would downlink ATR detections, GMTI track files, change-detection
alerts, and operator-selected image chips—compressing the data
bandwidth requirement by orders of magnitude while delivering
higher-value intelligence products.
5.12 Self-Calibration, Built-In Test, and Fault Detection
A significant production bottleneck of the original Lynx was that
each deliverable unit required hand calibration by flight test. The
system’s analog RF chain—TWTA, upconverter, receiver, and mechanically
aligned antenna—exhibited unit-to-unit variations in gain, phase, and
frequency response that could only be fully characterized in the actual
flight environment, where vibration, temperature gradients, and
aerodynamic loading affected performance. This drove delivery schedules
outward and required Sandia and GA engineering support for each
installation.
A digital AESA architecture fundamentally changes this paradigm. Each
T/R module contains its own digitally controlled phase shifter and
variable-gain amplifier, enabling closed-loop self-calibration by
injecting known reference signals and measuring the response of every
element. The array can characterize and correct its own amplitude and
phase errors at power-up, periodically during operation, and after
environmental changes—compensating for hardware imperfections,
temperature drift, and aging without external test equipment. IEEE
research has documented self-calibration algorithms for digital
beamforming arrays that compensate not only for internal errors but even
for external cover effects such as painted radomes. Current AESA
production experience shows that element-level testing and phase
calibration account for over 40 percent of total manufacturing cost;
self-calibration dramatically reduces this by replacing chamber-based
element-level characterization with in-situ automated procedures that
run in minutes rather than days.
Equally important is continuous built-in test (BIT). The digital AESA
can monitor the health of every T/R module in real time—detecting
failed elements, degraded power output, elevated noise figure, or
phase-shifter errors—and automatically reconfigure the beamforming
weights to maintain performance despite element failures. This provides
the operator and maintenance crew with a precise fault map without
requiring specialized test equipment, transforming maintenance from
periodic scheduled inspections to condition-based servicing. The array
degrades gracefully: losing 5 percent of elements (20 out of 400) costs
only about 0.45 dB in array gain, whereas a TWTA failure is catastrophic
and grounds the sensor. The net production and sustainment impact is
substantial: eliminate flight-test calibration from the delivery
process, reduce field maintenance burden, and increase operational
availability from the 70–80 percent range typical of complex analog
radar systems toward the 90+ percent characteristic of modern digital
arrays.
5.13 On-Board Data Storage and Recording
The original Lynx provided optional data recording on four additional
PowerPC nodes, with storage capacity limited by late-1990s media
technology. Modern ruggedized NVMe solid-state drives provide 4–16 TB
per unit in compact M.2 or U.2 form factors, with sustained write speeds
of 3–7 GB/s—sufficient to record full-bandwidth complex (I/Q) phase
history data in real time for post-mission analysis and SAR image
reprocessing.
For a modernized system with 200-MHz sample bandwidth, 16-bit
digitization (vs. the original 8-bit), and a 5-kHz pulse repetition
frequency, the raw recording data rate is approximately 4 GB/s per
channel. A 16-TB SSD provides roughly 67 minutes of continuous raw
recording—sufficient for multiple collection passes within a single
sortie. The jump from 8-bit to 14–16-bit ADC resolution alone represents
a significant dynamic range improvement of 36–48 dB over the original
Lynx’s 48 dB, enabling detection of weak targets in the presence of
strong clutter without saturating the receiver.
On-board recording of complex phase history data is essential for
three advanced capabilities that the original Lynx could not fully
exploit. First, coherent change detection requires library images
registered to sub-pixel accuracy against new collections; the original
Lynx transmitted complex images to the ground station for CCD
processing, but on-board storage with GPU processing enables both
real-time and post-mission CCD without ground-station dependency.
Second, tomographic 3D SAR processing requires precisely co-registered
multi-pass data that must be stored and recalled for joint inversion.
Third, post-mission ATR training benefits enormously from having raw
phase history data available: new target signatures discovered during
exploitation can be traced back to the original collections for
algorithm refinement and retraining.
5.14 3D SAR Imaging, Tomographic Processing, and Layover Elimination
One of the fundamental limitations of all conventional
two-dimensional SAR—including the Lynx—is the layover phenomenon.
Targets at different heights but the same range and Doppler frequency
collapse into the same image pixel, creating geometric distortions that
severely degrade image interpretability in urban areas, mountainous
terrain, and complex infrastructure. The original Lynx’s coherent change
detection mode represented an early approach to extracting additional
information from registered image pairs, but true three-dimensional
reconstruction was beyond its capabilities.
Tomographic SAR (TomoSAR) resolves this limitation by constructing a
synthetic aperture in the elevation direction through multiple passes at
different spatial baselines. For a total perpendicular baseline span of
100 meters (achievable with 5–8 passes at different flight altitudes or
lateral offsets), a Ku-band system achieves elevation resolution of
approximately 2.25 meters—sufficient to separate ground-level and
rooftop scatterers on multi-story buildings and eliminate layover
ambiguity. A 2025 paper in the ISPRS Journal of Photogrammetry and Remote Sensing
proposed multi-frequency TomoSAR (MF-TomoSAR) configurations that
achieve comparable 3D imaging quality with fewer passes by exploiting
frequency diversity—directly applicable to a wideband Ku-band AESA that
can tune across the full 15.2–18.2 GHz Lynx band.
For an airborne UAV SAR with an AESA containing multiple receive
subarrays, single-pass 3D imaging becomes feasible without multiple
orbits. Distributing receive phase centers across the array aperture
creates interferometric baselines within the array itself. With a
Ku-band array of 20–30 cm vertical extent and 4–8 receive subarrays,
single-pass elevation resolution of 5–15 meters is achievable—coarser
than multi-pass TomoSAR but immediately valuable for real-time terrain
modeling and layover mitigation during the mission.
Multi-pass 3D reconstruction with precise flight path control yields
high-quality 3D point clouds that can be processed into building models,
terrain maps, and infrastructure damage assessments. The computational
requirements for tomographic inversion—historically a barrier to
operational deployment—are readily addressed by the GPU signal processor
described in Section 5.6. Deep-learning approaches to TomoSAR,
including compressed-sensing algorithms and neural network-based
elevation estimation, further reduce the number of passes required for
acceptable 3D reconstruction quality. A July 2025 comprehensive review
of SAR tomography documented 30 years of progress from the first Ku-band
laboratory demonstration in 1995 to current deep-learning-assisted
spaceborne and airborne systems. The combination of on-board complex
data recording (Section 5.13), GPU processing, and AI-assisted
tomographic inversion positions a modernized Lynx-class system to
deliver 3D scene reconstruction as a standard operational product rather
than a specialized post-mission analysis exercise.
“The combination of on-board complex data recording, GPU processing,
and AI-assisted tomographic inversion positions a modernized Lynx-class
system to deliver 3D scene reconstruction as a standard operational
product.”
Projected Performance: Modernized Lynx-Class System vs. Original
| Parameter | Original Lynx (1999) | Modernized Concept (2026 Tech) |
|---|
| Antenna | Mechanical dish, 3-axis gimbal | GaN AESA, electronic steering, no gimbal |
| Transmitter | 320 W TWTA, single tube | Distributed GaN T/R modules, >1 kW aggregate ERP |
| LNA / Noise Figure | ~4.5 dB (system) | <3.0 dB (GaN/GaAs LNA per element) |
| Spotlight Resolution | 0.1 m | 0.05–0.1 m (wider bandwidth) |
| Max Operating Range | 25–30 km (full spec) | >80 km; >160 km with AESA |
| Swath Width (Stripmap) | 934 m (0.3 m res) | Multiple km (digital beamforming) |
| SAR/GMTI | Sequential modes only | Simultaneous (multichannel DBF) |
| Min. Detectable Velocity | 5.8 kts | <3 kts (STAP processing) |
| IMU | Litton LN-200 (tactical grade) | Navigation-grade FOG or MEMS |
| GNSS | Single antenna, GPS only | Multi-antenna, multi-constellation + attitude |
| Signal Processor | 16× Mercury 200 MHz PPC (~10–20 GFLOPS) | GPU-accelerated, >100 TFLOPS |
| Video SAR | Not available (added later) | Native mode, HD real-time |
| On-Board ATR / AI | None | Deep-learning inference at edge |
| Electronic Protection | Limited (fixed waveform) | Wideband agile waveforms, beam agility |
| Cabling | Custom per-platform, VME/RACEway bus | High-speed serial, flex-rigid PCB, fewer connectors |
| Weight | ~57 kg (125 lb) | 45–55 kg target |
| Calibration | Flight-test hand calibration per unit | Automated AESA self-calibration at power-up |
| Fault Detection | External test equipment required | Continuous BIT, per-element health monitoring |
| On-Board Recording | Optional, limited capacity | 16 TB NVMe SSD, ~67 min full I/Q recording |
| 3D Imaging / Layover | None (2D SAR only) | TomoSAR (~2.25 m elev. res.), single-pass array InSAR |
| Integration | Permanently installed in fuselage | Modular pod, mission-configurable |
| Urban Performance | Degraded (single-channel, high clutter/multipath) | MIMO-STAP multipath suppression, TomoSAR layover resolution |
5.15 Podded Design: Modularity and Mission Flexibility
The original Lynx was permanently installed in the Predator/Reaper
fuselage and nose radome, with the REA in the airframe and the gimbal
assembly in the nose or belly. This meant the aircraft carried the full
57-kg radar on every mission, whether the SAR was needed or not. In the
counterinsurgency operations that dominated MQ-9 employment in
Afghanistan and Iraq, a large fraction of missions were EO/IR-only
overwatch and close air support, where the Lynx radar was dead weight.
That 57 kg of unused payload capacity could have carried additional fuel
(extending endurance), more weapons, a communications relay package, or
an electronic warfare system.
GA-ASI actually demonstrated the alternative approach during NATO
Unified Vision 2012: the Block 30 Lynx was housed in a pod underneath a
King Air 200 surrogate aircraft, with the antenna, waveform generator,
and processor in the pod and the operator interface in the cabin. This
pod-based configuration proved that the Lynx could operate independently
of the host platform’s internal avionics, connecting through a standard
power and data interface.
A modernized Lynx designed as a self-contained pod would offer
several advantages. First, mission configurability: the pod is installed
only when the SAR mission requires it, freeing the payload stations for
other stores on non-radar missions. Second, cross-platform portability:
the same pod, with a standardized mechanical and electrical interface,
could be mounted on the MQ-9, Gray Eagle, MQ-9B SkyGuardian, manned ISR
aircraft like the King Air or C-12, or even next-generation CCAs—without
the platform-specific integration that the original Lynx required
(recall the “some variance due to different cable assemblies for
different platforms” noted in the specification). Third, maintenance and
logistics: a defective pod can be swapped on the flight line in
minutes, with the replacement pod’s self-calibrating AESA verifying its
own health at power-up, rather than requiring the aircraft to go through
depot-level radar maintenance. Fourth, technology insertion: the pod
can be upgraded independently of the aircraft, allowing radar technology
refreshes on a faster cycle than airframe modernization programs.
The pod form factor also naturally accommodates the AESA
field-of-regard solutions discussed in Section 5.1: a belly-mounted pod
with a single-axis repositioner can rotate the array face between port
and starboard, and the pod’s external mounting provides better thermal
access for radiating waste heat than a buried fuselage installation. For
the spaceborne variant discussed in Section 7, the pod concept maps
directly to a satellite payload module with standard bus
interfaces—further reinforcing the modularity argument.
5.16 Urban Environment Performance
The original Lynx had notably poor performance in urban
environments—a significant operational limitation given that the wars in
Afghanistan and Iraq frequently required ISR over cities, villages, and
built-up areas. The problems were twofold: high clutter returns from
buildings and infrastructure degraded both SAR image quality and GMTI
detection performance, and multipath propagation from building surfaces
created ghost targets and false GMTI detections.
In SAR mode, urban terrain produces severe layover: buildings, walls,
and elevated structures at different heights but the same slant range
collapse into the same image pixel, creating a confused tangle of
overlapping returns that obscures the scene structure. The original
Lynx’s two-dimensional SAR had no ability to resolve this ambiguity (the
TomoSAR capability described in Section 5.14 directly addresses this).
Building facades also produce extremely bright specular returns
(double-bounce scattering between the ground and vertical walls) that
can saturate the receiver or mask weaker targets of interest nearby—a
problem exacerbated by the original system’s 8-bit ADC dynamic range and
elevated phase noise floor.
In GMTI mode, the urban environment is even more challenging.
Building surfaces create multipath clutter: radar energy reflects off
walls, roofs, and the ground in multi-bounce paths that arrive at the
receiver with Doppler shifts that mimic moving targets. This
“spread-Doppler clutter” fills the velocity space that GMTI processing
must search, raising the false alarm rate and masking genuine movers.
The original Lynx’s single-channel exo-clutter GMTI had no ability to
distinguish direct-path returns from multipath returns—they were simply
additive clutter. The original system’s +10 dBsm minimum detectable
target specification was achievable in open terrain, but in urban
canyons the effective detection threshold was substantially higher due
to the elevated clutter floor.
The modernized system addresses these limitations through multiple
reinforcing improvements. The 14–16-bit ADC (vs. 8-bit) provides 36–48
dB more dynamic range, enabling the receiver to handle bright building
returns without saturating while still detecting weak targets. The
dramatically lower phase noise from the direct digital waveform
architecture (Section 5.7) reduces the paired-echo sidelobes around
bright scatterers, improving the ability to detect weak targets near
strong urban returns. Multichannel STAP on a digital AESA can
distinguish direct-path returns from multipath by exploiting the
different spatial signatures of the two propagation paths—MIMO-STAP
techniques have been specifically demonstrated for multipath clutter
mitigation in urban environments. TomoSAR processing (Section 5.14)
resolves the layover ambiguity by separating scatterers at different
heights, transforming a confused 2D urban SAR image into a 3D point
cloud where buildings, vehicles, and ground surfaces are geometrically
distinguishable. And AI-driven clutter classification, trained on urban
SAR/GMTI datasets, can learn to recognize and suppress the
characteristic signatures of multipath ghost targets—a task that is well
suited to deep learning but impossible with the rule-based processing
available in the 1990s.
The net effect is that a modernized Lynx-class system would be
operationally useful in urban environments where the original was
marginal—a critical capability gap given that future conflicts are
increasingly likely to involve urban terrain.
6. The Eagle Eye Bridge
GA-ASI’s Eagle Eye radar, which came off the production line on 31
July 2024, represents the clearest near-term path from Lynx’s legacy
architecture toward the modernized concept described above. Designed as a
drop-in enhancement for the U.S. Army’s Gray Eagle Extended Range and
the new Gray Eagle 25M, EagleEye delivers SAR/GMTI, Video SAR, change
detection, and Maritime Wide Area Search in a package that GA-ASI
describes as offering “multiple times the range” of previous radars.
The planned AESA upgrade—funded by GA-ASI corporate investment,
paralleling the original Lynx development model—targets more than
doubling EagleEye’s 80-kilometer range. As Jeff Hettick, GA-ASI vice
president of Agile Mission Systems, framed it, the increased range and
optimized multi-mode performance allow the aircraft to operate well
outside the weapons engagement zone of most threat systems, adding
survivability for the platform and the sensor alike. The Army National
Guard has ordered 12 Gray Eagle 25M aircraft with EagleEye as part of
the initial configuration.
The broader trajectory is clear: the Lynx Multi-mode Radar continues
to serve on MQ-9 Reapers and other platforms worldwide, providing proven
ISR capability that allies and combatant commands depend on daily. But
the technology baton is passing to EagleEye and its successors, which
will carry Lynx’s DNA—the video-camera-like user interface philosophy,
the deep Sandia algorithmic heritage, the multi-mode flexibility—into an
era where standoff range, electronic resilience, and autonomous
exploitation are not optional features but survival requirements.
7. Adaptation to LEO Satellite Deployment
The modernized Lynx-class system described in Section 5, with its GaN
AESA, GPU processing, and 45–55 kg total mass, falls squarely within
the parameter space of current commercial SAR microsatellites. ICEYE’s
X-band SAR satellites weigh approximately 85–92 kg with peak transmit
power of 3–4 kW; Capella Space’s Acadia-class satellites operate at
X-band with sub-0.25 m resolution; and China’s Taijing-4(03), launched
in January 2024, deployed a Ku-band phased-array SAR payload weighing
less than 80 kg with better than 1-meter resolution. The question is not
whether a Lynx-derived sensor could fly in orbit, but what engineering
challenges must be addressed to make the transition from an airborne to a
spaceborne environment.
7.1 The Thermal Problem: Dissipating Heat in Vacuum
This is the central challenge. In the atmosphere, the original Lynx
dissipated its TWTA and electronics heat through a combination of
conduction to the airframe and forced convection from the aircraft’s
slipstream. In LEO, there is no convective cooling—heat must be rejected
entirely by radiation to space and conduction to spacecraft thermal
mass. The problem is acute for a GaN AESA because the aggregate DC power
input to the array can be substantial.
Consider a 400-element Ku-band AESA with 5–10 W RF output per element
at 30% power-added efficiency. At 10 W per element and 30% PAE, each
element dissipates approximately 23 W of heat (10 W / 0.30 = 33 W DC
input, of which 23 W is waste heat). For 400 elements operating
simultaneously, the total thermal dissipation is approximately 9.3 kW—an
enormous heat load for a microsatellite. However, spaceborne SAR
systems do not operate continuously. The duty cycle is the critical
design variable: ICEYE satellites operate with peak RF power of 3–4 kW
but image for only a fraction of each orbit (typically 10–60 seconds per
imaging pass, with a total imaging duty cycle of perhaps 5–15% of the
orbit). The Taijing-4 Ku-band SAR similarly pulses its array for brief
imaging windows, storing solar energy in batteries during the sunlit arc
and discharging during the imaging pass.
For a Lynx-derived spaceborne system, the thermal design strategy
would involve several elements. First, limiting the number of
simultaneously active T/R modules: rather than activating all 400
elements at full power, the digital AESA can operate with a subset (say
100–200 elements) for wider-swath, coarser-resolution modes, reserving
full-array operation for high-resolution spotlight passes of limited
duration. Second, thermal mass buffering: the array structure and
spacecraft bus absorb heat during the brief imaging pulse (tens of
seconds), then dissipate it radiatively over the remainder of the orbit
(approximately 90 minutes for a 500-km LEO). A 2024 MDPI study on
thermal optimization for a 310-kg X-band SAR satellite achieved peak T/R
module temperatures below 29°C during imaging by integrating thermal
interface materials directly into the antenna structural design,
achieving what the authors described as “zero-consumption thermal
design” for the SAR antenna. Third, deploying radiator panels on the
anti-sun side of the spacecraft, sized to reject the time-averaged
thermal load rather than the peak instantaneous load.
The thermal challenge is real but tractable. GaN-on-SiC’s inherently
high thermal conductivity (SiC substrates conduct heat approximately 3×
better than GaAs) is a significant advantage: it allows higher junction
temperatures (up to 200°C for brief pulses, vs. 150°C for GaAs) without
reliability degradation, providing margin for the constrained cooling
environment of space. The key insight is that the spaceborne thermal
problem is dominated by duty cycle management, not by the peak power
capability of the array.
7.2 Power Budget and Solar Array Sizing
A spaceborne SAR’s power architecture operates fundamentally
differently from an airborne system. The aircraft provides continuous
prime power from the engine-driven generator; a satellite must harvest
solar energy, store it in batteries, and discharge at high rates during
imaging. For the 400-element array at full activation drawing
approximately 13 kW of DC (33 W per element × 400), even a 30-second
spotlight dwell requires 13 × 30 / 3600 = 0.11 kWh of battery energy.
With a 25% orbit duty cycle for imaging (multiple passes per orbit), the
total energy requirement per orbit might be 0.5–1.0 kWh, well within
the capability of modern lithium-ion battery packs (ICEYE satellites
carry battery packs providing roughly 1.6 kWh of usable capacity). The
solar array must be sized to replenish this energy plus housekeeping
loads over the sunlit portion of the orbit. For a 500-km dawn-dusk
sun-synchronous orbit with approximately 60 minutes of sunlight per
95-minute orbit, a solar array generating 600–1000 W average would
sustain the imaging duty cycle.
7.3 Radiation Environment and Component Hardening
LEO satellites operating at 400–600 km altitude are exposed to
trapped radiation in the South Atlantic Anomaly, solar particle events,
and galactic cosmic rays. Every electronic component in the modernized
Lynx must be assessed for radiation tolerance. The good news is that GaN
devices exhibit inherent radiation hardness superior to silicon and
GaAs: the wide bandgap (3.4 eV vs. 1.1 eV for Si) means that
radiation-generated electron-hole pairs cannot produce the parasitic
short circuits that cause single-event effects (SEE) in narrower-bandgap
materials. EPC Space has delivered rad-hard GaN power FETs rated above
1,000 kRad(Si) total ionizing dose (TID) with SEE immunity to linear
energy transfer (LET) of 83.7 MeV·cm²/mg—far exceeding the LEO
environment. Thousands of rad-hard GaN devices are already operating in
orbit.
The more challenging radiation concern is the signal processor.
Current GPU-class processors (NVIDIA Orin, etc.) are designed for
automotive and industrial environments, not space radiation.
Space-qualified alternatives include Xilinx Kintex UltraScale FPGAs
(used in the S-STEP SAR satellite), radiation-tolerant ARM processors,
and emerging rad-hard GPU-class devices from companies like Innoflight.
The software-defined nature of the GPU processing architecture allows
migration to whatever space-qualified processing hardware is available,
with algorithmic adjustments for the different computational profile.
FPGA-based signal processors are already standard in operational SAR
satellites (ICEYE, Capella, S-STEP) and can implement the OSA image
formation algorithm and deep-learning inference networks in
radiation-tolerant fabric.
7.4 Orbit Geometry and SAR Performance at LEO Altitude
Moving from airborne to spaceborne operation changes the SAR geometry
fundamentally. At a 500-km orbit altitude (vs. 5–10 km for a UAV), the
slant range increases by two orders of magnitude, the platform velocity
jumps from 35 m/s to approximately 7,500 m/s, and the synthetic aperture
time for a given azimuth resolution shortens dramatically. The SAR
radar equation (1) shows that SNR degrades as R³—a factor of
10&sup6; for a 100× range increase. This is partially compensated by
the much larger synthetic aperture (higher platform velocity yields
longer effective aperture for a given dwell time), longer coherent
integration, and the ability to use larger antenna apertures in space
(where weight is constrained but physical size less so). Current
spaceborne SAR microsatellites achieve 0.25–1.0 m resolution at slant
ranges of 500–800 km with peak transmit powers of 1–4 kW, confirming
that the physics closes for a Ku-band AESA in the Lynx power class.
A Ku-band choice (vs. the X-band favored by ICEYE and Capella) offers
both advantages and challenges for spaceborne operation. The shorter
wavelength provides finer resolution for a given antenna size and
bandwidth, but Ku-band suffers greater rain attenuation than
X-band—roughly 2–5 dB for moderate rain at Ku-band vs. 0.5–1 dB at
X-band. For many applications (urban monitoring, infrastructure
assessment, defense ISR), the resolution advantage outweighs the rain
penalty, particularly for a system designed to operate at multiple
selectable resolutions.
7.5 A GA-ASI / GA-EMS Joint Program
The most compelling aspect of a spaceborne Lynx-derived SAR is that
General Atomics already possesses all the necessary corporate
competencies—split across two affiliates that have never, to public
knowledge, combined their capabilities on a SAR satellite program.
General Atomics Electromagnetic Systems (GA-EMS), based in San Diego
alongside GA-ASI, has built an extensive space systems portfolio. The
division describes itself as having “rich heritage and extensive
expertise in the design, integration, test, and manufacture of
satellites, payloads, and space systems at any scale.” Specifically,
GA-EMS brings the GA-75 spacecraft bus—a 75-kilogram-class modular,
configurable half-ESPA platform designed to support ISR and
communications payloads, compatible with multiple launch vehicles and
capable of packaging two spacecraft per ESPA port. Two GA-75 spacecraft
with GA-EMS Optical Communication Terminals (OCTs) are scheduled for
launch in 2026 under a Space Development Agency contract for Tranche 1
LEO airborne-to-space demonstrations. The EWS weather satellite
program—a $380 million contract restructured in 2024 to include two
operational weather satellites with five years of on-orbit services
through 2030—demonstrates GA-EMS’s ability to deliver complete
spacecraft with integrated sensor payloads on operationally relevant
schedules. GA-EMS supplies EO/IR missile warning, tracking, and fire
control payloads for Lockheed Martin’s SDA Tranche 2 Tracking Layer (18
satellites, launch in 2027). And in September 2025, GA-EMS and Kepler
Communications successfully demonstrated bi-directional air-to-space
optical communications between a GA-EMS OCT mounted on an aircraft
(using a GA-ASI LAC-12 turret) and an SDA Tranche 0-compatible satellite
in LEO—proving the cross-division integration that a joint SAR
satellite program would require.
GA-ASI, meanwhile, brings the Lynx/EagleEye SAR design heritage, the
Sandia-derived image-formation algorithms, the Claw exploitation
software, the VideoSAR and GMTI processing chain, and 25 years of
operational experience in UAV radar integration. The company’s
Reconnaissance Systems Group has the radar RF engineering, signal
processing, and AI/ML exploitation expertise that would be required for
the SAR payload itself.
A joint GA-ASI/GA-EMS program would combine these complementary
capabilities into a vertically integrated SAR satellite system. The
division of labor maps naturally: GA-EMS provides the spacecraft bus
(GA-75 or a scaled variant), power system, thermal management, attitude
determination and control, optical inter-satellite links, and launch
integration; GA-ASI’s Reconnaissance Systems Group provides the Ku-band
GaN AESA SAR/GMTI payload, the GPU or FPGA signal processor running
Sandia-heritage algorithms, the on-board AI exploitation chain, and the
ground segment software including Claw. The self-calibrating AESA
architecture described in Section 5.12 eliminates the flight-test
hand-calibration bottleneck that plagued the original Lynx—essential for
a constellation where you need to build and launch 12–24 satellites on a
production timeline, not hand-tune each one.
The corporate structure makes this feasible: both divisions are
affiliates of privately held General Atomics, answering to the same
ownership (the Blue family). There are no inter-company contracting
barriers of the kind that complicate collaborations between independent
primes. The precedent for cross-division integration already exists: the
September 2025 air-to-space optical communications demonstration
explicitly combined a GA-EMS OCT with a GA-ASI airborne turret, with
GA-EMS vice president Gregg Burgess publicly describing the connection
between the two divisions’ capabilities.
What would such a satellite look like? An 80–120 kg microsatellite
built on a scaled GA-75 bus, carrying a deployable Ku-band GaN AESA (2–3
m² deployed antenna area), with GA-EMS optical communication terminals
for SDA-compatible data relay, FPGA-based signal processing implementing
the OSA/PGA algorithms with on-board AI inference, 16 TB of
radiation-tolerant solid-state storage, and the self-calibrating array
that eliminates per-unit flight calibration. A constellation of 12–24 in
phased orbital planes could provide sub-hourly revisit of any point on
Earth—persistent, all-weather, day/night SAR and GMTI coverage that is
immune to the SHORAD threats that have destroyed 24 MQ-9 Reapers over
Yemen. The data flows through SDA-standard optical crosslinks to ground
stations or directly to tactical users via the same Claw exploitation
software that MQ-9 operators already know.
General Atomics would not be entering the commercial SAR
constellation market cold. ICEYE (54+ satellites, €250M+ revenue in
2025, a €1.5B backlog) and Capella Space have proven the business model.
Rheinmetall took a 60% stake in ICEYE in 2024, explicitly to create a
cross-domain provider bundling orbital imaging with ground-based
systems. A GA SAR constellation would target the defense and
intelligence market segment where the company’s existing
relationships—U.S. Air Force, U.S. Army, allied air forces—provide
immediate customer pull, and where the integration of spaceborne SAR
with airborne Reaper/Gray Eagle ISR would offer a multi-domain sensing
architecture that no competitor currently provides.
“A GA-ASI/GA-EMS joint SAR constellation would offer a multi-domain
sensing architecture—integrating spaceborne SAR with airborne Reaper and
Gray Eagle ISR—that no competitor currently provides.”
8. The Institutional Question
The Lynx radar’s development history illuminates a pattern that
recurs throughout GA-ASI’s corporate strategy: the company funds
critical technology development internally, maintains tight integration
between radar and platform, and delivers capability faster than
traditional DoD acquisition programs. Lynx was developed entirely on GA
corporate funds. EagleEye’s AESA upgrade is likewise a company
investment. This model has produced results—but it also means that the
most ambitious modernization possibilities (a fully digital AESA with
GaN, on-board GPU processing, and AI-driven autonomy) depend on the
alignment between corporate investment horizons and military procurement
timelines.
The U.S. Air Force, meanwhile, plans to retain 140 MQ-9 Reapers
through 2035 while divesting older airframes and seeking a “more
survivable, flexible, and advanced platform.” The Collaborative Combat
Aircraft (CCA) programs—including GA-ASI’s own YFQ-42A Gambit—will
likely require a new class of conformal, low-observable sensors that
share Lynx’s mission but not its form factor. The question for the next
decade is whether the investment in sensor modernization will keep pace
with the investment in airframes, or whether the services will fly
advanced platforms with sensors designed for a threat environment that
no longer exists.
9. Conclusion
The Lynx SAR/GMTI radar stands as one of the most successful tactical
sensor programs in the history of unmanned aviation. Conceived as a
corporate-funded venture between a then-upstart drone maker and a
national laboratory, it delivered a multimode, fine-resolution imaging
radar at a fraction of the weight and cost of contemporaneous
systems—and it worked from the first flights. Its architecture embodied
elegant engineering choices: the IMU on the antenna, the
stretch-processing pipeline, the View Manager interface that let
non-radar operators exploit the sensor like a video camera.
But the world the Lynx was designed for—permissive airspace,
low-threat targets, bandwidth-limited data links, and human analysts
with time to study SAR imagery—is receding. The technologies available
to transform a Lynx-class system are mature: GaN AESA arrays, GPU signal
processors, deep-learning ATR, and wideband agile waveforms. The
demonstrated losses of MQ-9 platforms to contested air defenses make the
case for standoff-range, electronically resilient sensors with
self-evident urgency. GA-ASI’s EagleEye represents a credible bridge,
but the full potential of modern radar technology applied to the UAV SAR
mission remains largely unrealized.
Armin Doerry, in a career spanning nearly three decades of SAR
development at Sandia, often returned to a fundamental point in his
publications: the performance limits of a SAR system are dictated by
physics, no matter how bright the engineer. That remains true. But the
distance between what physics allows and what the fielded systems
deliver has never been wider. Closing that gap is the engineering
challenge of the next decade.
