Transcription
At 10:26 local time, over the Sea of Azov, a single Ukrainian Luti drone was screaming toward a target that Russia couldn't afford to lose.
This Russian IL38N aircraft, the only platform Russia had in the Black Sea that could detect their new weapon, and satellite imagery from 6 hours ago showed ground crews already removing engine covers. The IL38N would be flight ready by 1300. This was their only shot to take it out before getting airborne. The clock was already running out.
At 10:31, the Luty crossed the Ukrainian coastline and descended to 50 m above gray winter water. Ahead lay 280 km of open sea. At 220 kmh, the crossing would take 76 minutes. 76 minutes of silence. The drone's composite airframe shattered radar energy in every direction except back toward Russian stations. like throwing a handful of glitter instead of a mirror to see yourself.
At 50 m altitude, the Earth's curvature did the rest. Radar waves travel in straight lines, but the Earth doesn't cooperate. The drone was flying below the horizon, invisible to every Russian radar within 200 km. For over an hour, it might as well not existed. Everything seemed to be a routine operation until it got overground again.
11:47 when the drone made landfall east of Perosco Atars, 70 km to YaK, 19 minutes to impact, the invisible phase was over. Now came the part where Russia got a vote. 3 seconds after crossing the coastline, that silence ended.
If you look here at this cellular tower outside the village of Stoot Terovskia, you'll see what looks like standard telecom equipment. A gray cylinder bolted to the frame, roughly the size of a coffee thermos. Inside are four omnidirectional microphones and a processor running classification algorithms. Clever. But clever only works if your enemy doesn't know about it.
This is Russia's answer to radar evading drones. You can't see what flies below the radar horizon, but you can hear it. The Luthy's Turbo Jet doesn't care about stealth. The engine produces 85 dB at 100 m, the sound of a motorcycle at highway speed. Compressor blades create a high-pitched whistle between 4,000 and 8,000 hertz. Combustion adds a lower rumble between 200 and 500 hertz. In cold December air, that sound carries far. Dense air transmit acoustic energy more efficiently than warm air.
The microphone picked up the first traces at 11:47. Signal strength 53 dB, about as loud as your coworker who swears nobody heard him rip one in the break room. faint, but well above the 35 del ambient noise of a rural Russian winter morning, Russia had built microphones sensitive enough to catch Steve from accounting red-handed.
The algorithm had found its needle. The processor performed a fast furer transform, breaking the incoming sound into its component frequencies, like separating a chord into individual notes. The algorithm matched the spectral fingerprint against its database. That whistle wine of compressor blades, that combustion rumble, the specific ratios that distinguish turbo jets from propellers. Classification returned in 1.3 seconds. UAV confidence 87% bearing 245°.
One microphone gives directions. It doesn't give distance. The drone could be 2 km out or 10. like knowing someone shouting from the east but not knowing if they're in your yard or on the next block. But Russia has over 200 of these sensors along the Krasnadar coast spaced 3 to 5 km apart. Within 8 seconds, three more microphones reported. Bearings 248, 251, 243°. The fusion algorithm computed where those lines intersected. Basic trigonometry, but performed thousands of times per second. A track appeared. Hostile UAV 3 position accurate to 50 meters, updating every two to three seconds. Deluti had no idea, and it could have cared less if it did. The drone carried no acoustic warning system because such a device doesn't exist for aircraft this size.
The Skynode Sflight computer showed all systems normal. Target coordinates locked. Work tracking its every move might as well have been tracking a ghost that didn't know it was already dead. 67 km to Y, 18 minutes to impact. Still enough time if nothing else went wrong. The microphones could follow the loot across Ksnodar. They couldn't shoot it down. For that, Russia needed a different weapon.
At 11:52, 5 minutes after the microphones found the drone, the track data reached a TORM M2 battery Strovskia, 18 km southeast of the drone's position, well within the systems range. The TORM M2 is Russia's short-range answer to exactly this threat. Eight missiles on a track chassis. Radar that can track 48 objects simultaneously. On paper, the Luty was already toast.
The battery commander had solid coordinates from the acoustic network. He slewed his fire control radar to the general direction and searched for the target. Nothing. He had position data accurate to 50 m. His radar could track a bird at 25 km, but the screen showed empty sky. The problem was the elevation angle. The tour's fire control radar needs to look slightly up at its targets, at least 1.5° above the horizon. Any lower and ground reflections would overwhelm the return. Radar waves bounce off terrain, creating false echoes that the processor can't separate from real targets.
The system was designed for aircraft and helicopters, threats that fly at hundreds of meters. Nobody planned for a drone at 30 m. At 18 km, a target flying at 30 m sits at an elevation angle of 0.1°, essentially on the horizon. The radar was staring straight into ground clutter, and the drone was buried in it.
The commander did the math. For his radar to see a target at 30 m altitude, that target needed to be within 1.2 km. At 180 km per hour, the Luty would cross that window in 24 seconds. He'd have one chance if the drone flew directly toward his position. It wasn't. The Luty was heading northeast, crossing in front of him, not approaching. He tracked the acoustic data as it updated. Range 16 km, then 15. The drone was getting closer, but the geometry wasn't improving. The track would pass 4 km north of his position, never entering the narrow window where his radar could acquire.
At 11:55, the acoustic network showed it at 12 km, sparing northeast at 175 kmh. The tour's radar still showed nothing but frozen farmland and interference patterns. At 11:57, the track crossed behind his engagement envelope entirely. Range opening. The drone was now heading away. This one flew under the radar, literally.
47 km to Y, 12 minutes to impact. The Luty had slipped past the acoustic network through terrain masking and the Torah through geometry and physics. But something else had been hunting it since before it crossed the coast. Something that didn't care about elevation angles or radar horizons.
40 km to the west, that weapon was already airborne. The Jiren 2 launched from a position near Timuk at 1050. 3 minutes after the Luty took off, Russian standing orders kept armed interceptors orbiting whenever satellite warning indicated Ukrainian launch activity. A looty leaving a Zaparicia runway triggered the alert. The Garin was airborne before the Ukrainian drone reached the coast.
But this wasn't a normal Garin. If you look here at the top of the fuselage where the Delta wing meets the engine housing, you'll see something that doesn't belong on an attack drone. a 9K 333 Verba missile in a modified launcher tube held by aluminum brackets bolted to the airframe. Russia had turned a one-way attack drone into an airborne SAM platform. It was the drone equivalent of duct taping a shotgun to a lawnmower. Ugly, improvised, and just stupid enough to work.
The Guarante is an Iranian design built under Russian license. The Shared 136 with cerillic stencils. Delta wing. Pusher propeller four-cylinder piston engine producing 50 horsepower. Cruise speed 185 km/h. Normal endurance 4 plus hours. The armed variant is different. The Verba system adds 12 kg. That extra weight cuts loiter time to 90 minutes. And if you look here at the mounting hardware, you'll see two black boxes near the launcher tube. These are servo motors. One controls the protective cover over the missile seeker. The other actuates the firing trigger. Cables run into the fuselage taped to the skin. This means the operator doesn't fire with a simple button press. He sends commands through a mesh network, a radio system where signals hop between relay drones and ground stations. Each node passing data to the next until it reaches the drone. First command, open the cover. Second, align the drone. Third, fire.
By 11:47, when the LUTI made landfall, it had been airborne for roughly 60 minutes. 30 minutes of fuel remaining, one missile, one target hiding a 400 meter acoustic blur. The Garin received acoustic track data. Hostile UAV 3 bearing 095, range 38 km, heading northeast. The operator adjusted course eastward, descending to 600 m, and began searching. 47 km to Y. 15 minutes of fuel remaining. The clock was running for both sides now.
But finding a drone at 30 m altitude in an irrigation canal isn't like finding an aircraft against open sky. The Verbus Seeker needs contrast. Hot target against cold background. Against the sky, a 400° exhaust contrasts with minus15° air. Easy acquisition at 6 km. Against the ground, everything changes. Frozen farmland at minus3°. Plowed fields warming to + 5°. A complex thermal tapestry that all three seeker channels must somehow distinguish from the target.
The operator searched for 6 minutes. His camera showed frozen fields, bare trees, the dark line of a canal. The acoustic network said the target was somewhere in a 400 meter ellipse, but somewhere isn't a firing solution. 12 minutes of fuel. The canal ended ahead. The Luty would have to climb. When it climbed, everything would change.
At 11:59, the canal ended. Open farmland rose toward the northeast. No terrain to hide behind. The Lut's profile called for minimal exposure. Stay low. Stay fast. Never silhouette against the sky for more than seconds. The Skynode S identified farm buildings 2 km ahead. Metal roofs, thermal clutter. If the drone reached those buildings before climbing, the background might mask its signature. It almost worked.
At 12:07, a ridge line blocked the path. No route around. The drone pulled up to 80 m, crossed the ridge, and dove for the far side. 3 seconds. Exhaust silhouetted against gray December sky. The Garin's camera caught it perfectly. Breoint, moving fast, unmistakably hot.
The operator's hands moved. First command opened the launch cover. The servo motor actuated, protective cap swinging clear. The tri channel seeker began scanning. Second command aligned. The Garin banked, pointing its spine toward the target bearing. The seeker found the exhaust immediately. Ultraviolet strong. Infrared strong. Mid infrared strong. All three matched an engine profile. Not a flare, not a thermal reflection. A real target. Third command fire. The solid fuel motor ignited. 20 gs of acceleration. The Garren shuttered as the missile left the airframe. The operator fought to stabilize as it screamed toward its target. Range at launch 7.2 km. Closure rate 850 m/s. Time to intercept 11 seconds, maybe less because of this feature.
The Verba uses proportional navigation. The missile doesn't chase the target. It flies to where the target will be, constantly adjusting angle to cut off escape. Simple geometry, extremely difficult to defeat without counter measures. The Luty had none. No chaff, no flares, no jamming. The drone was built for range and payload, not survivability. But it had something else.
The moment the Luty crested the ridge, the Skynode S executed the next waypoint. Nose down, throttle idle, diver terrain on the far side, a stream bed that would provide acoustic shadow for the next leg. The drone wasn't evading the missile. It didn't know the missile existed. Pre-programmed terrain masking. coincidence, but a very effective coincidence.
The seeker held its lock. All three channels tracked the descending exhaust. Guidance commanded a dive to match, but now the missile was flying toward the ground at 600 m/s. The magnetic proximity fuse is designed to detonate near metal structures. But that sensitivity creates problems near terrain. Frozen ground, buried pipes, ironrich soil. All of these are generating faint magnetic signatures the fuse had to distinguish from valid targets.
9 seconds after launch, missile at 30 m, still descending. Seekers still locked. 11 seconds, missile at 18 m. Luty at 25 m. Pulling out of its dive. The fuse triggered, not on the drone, on an abandoned tractor in the stream bed. The 1.5 kg warhead detonated on frozen earth. Fragments scattered across the field. The looty 40 meters ahead and climbing shuddered from the shock wave. No damage.
The operator watched the explosion. Wrong position. Wrong angle. No debris trail. Miss. His only missile. 9 minutes of fuel remaining. The target was gone.
In CrossNadar, the operator stared at his screen. 11 months of intercepts. And this was his first miss. Someone would want an explanation. He didn't have one. He chose pursuit. Throttle forward. Camera tracking. The Garin couldn't catch the Luty. It was 35 kmh too slow, but it could relay position to base defense.
26 km to Yank, the Luty had survived acousta detection through terrain masking and survived the Verba through luck and programming. Ahead lay the hardest part, finding one specific aircraft among dozens with an AI that had never seen this target from this angle. The looty had to find the IL38N. Ignore the standard IL38 parked 80 m away. Hit precisely enough that the airburst warhead destroyed the sensor section, not just damage the airframe.
If you look here at the two aircraft, you'll see why this is difficult. Both have four turborop engines. Both have high-mounted wings. Both have long fuselages in similar gray paint. From 35 m altitude at a slant angle, they're nearly identical. The difference is on top of the forward fuselage. The IL38N carries the NV9 elint pod, a boxy structure roughly 2 m long, mounted on external struts above the cockpit. Elint stands for electronic intelligence. Pod contains passive receivers that intercept radar and radio emissions, classifying signals by frequency, pulse pattern, and origin. The standard IL38 doesn't have it. That pod marked the target. Destroy the wrong aircraft and Russia loses an $8 million trainer. Destroy the right one and Russia loses its only Black Sea submarine hunter.
The neural network running on Skyote S was trained for exactly this discrimination. Here's how it works. Ukrainian intelligence gathered thousands of images, satellite photos, reconnaissance footage, publicly available pictures, commercial aviation photography. Each image was labeled. This is an IL38N. This is a standard IL38. This is something else. The algorithm learned patterns, not explicit rules, patterns. It learned that IL38N aircraft have a rectangular shape above the cockpit that standard variants don't. It learned the aspect ratio of that shape, the shadow it casts at different sun angles, its position relative to the cockpit windows.
During flight, the camera feeds frames to the processor at 30 per second. Each frame goes through feature extraction. The algorithm identifies edges, shapes, shadows, and geometric relationships. It looks for four engine aircraft first, then high-mounted wings, then distinctive nose profile. Then it searches for the e-lint pod. The system draws a bounding box around candidate shapes, measuring dimensions pixel by pixel. It compares aspect ratios against its training data. It weighs confidence based on how closely the observation matches known examples. Final classification IL38N. Overall confidence 94%.
At 1215 plus 8 seconds, the satellite link carried an encrypted burst. Strike authorized. 14 seconds to impact. But at 12:15 + 12 seconds, the navigation computer detected an obstacle not in its reference imagery. A ground power unit beside the target, exhaust stack extending 4 m into the approach corridor. The satellite photos were 18 hours old. Someone had moved equipment. The AI recalculated 3° right, steeper descent. That correction caused 12 m of precision.
At 1215 plus 19 seconds, the IL38 filled the camera frame. NV9 pod clearly visible. Classification confidence 97%. At 1215 plus 21 seconds, the radar altimeter measured 6.1 m above a surface. The surface was the IL38N's upper fuselage. The warhead behind the Lut's fiberglass cone was built for exactly this moment.
If you look here at the internal structure, you'll understand why conventional explosives wouldn't work. A standard bomb detonates on contact, creating localized damage at a single point. Against a 40 m aircraft, you might destroy one section while leaving critical systems intact. The security service of Ukraine needed to perform a labbotomy. The airbururst warhead contains a central explosive charge surrounded by 2,000 steel fragments cubes approximately 8 mm per side packed in a polymer matrix. A radar altimeter in the nose transmits pulses downward, timing the returns to measure height. When the altimeter reads 6 m, the fuse triggers, but the detonation isn't spherical. The explosive is shaped. Curved surfaces direct blast energy downward, not outward. The fragments accelerate in a focused cone toward whatever lies beneath. Velocity 1,800 m/s. Time to cross 6 m, 3.3 milliseconds. The impact pattern roughly 4 m in diameter. Fragment density 40 impacts per square meter. Aluminum aircraft skin offers no resistance.
At 1215 plus 21 seconds, the warhead detonated. The fragment cone struck slightly after of the planned aim point over the wing route rather than directly over the elint pod. The obstacle avoidance had shifted the impact zone. It didn't matter. The NV9 pod took hundreds of direct hits. Housing ruptured. Antenna arrays shredded. Processing units destroyed. The pressurized forward compartment containing the NV4 mission computer gone. The NV4. The brain. The system that fused radar tracks. MAD contacts, SonoB buoy data, and electronic intercepts into a single picture that tracked 32 targets simultaneously. Destroyed in 3.3 milliseconds.
Secondary impact struck the number two engine the cell. Oil lines severed, fuel lines ruptured, fire within seconds. The IL38M didn't explode dramatically. It simply stopped being a functional aircraft. The Nolla radar still existed, but nothing processed its returns. The Mad Boom still extended from the tail, but nothing read its data. The Sonoboy receivers still worked, but nothing fused their inputs. The aircraft wasn't damaged, it was lobbomized. Bye for now.