Agras T100 Field Report: Antenna Geometry That Holds RTK Fix at 2.5 km Beyond the Valley

META: Dr. Sarah Chen documents how tilting the base-station antenna 12° and lifting the air-craft module 35 mm delivered 99.2 % RTK fix rate while scanning 110 kV lines across a 1 200 m canyon—no repeater, no cellular, no drift.

The morning the power company asked me to inspect a spur line that drops into a sandstone canyon west of Karamay, I packed the Agras T100 because its brochure promised “centimeter precision without continuous cellular.” What the brochure does not explain is how fast that precision collapses when the canyon walls pinch the sky down to a 30° sliver and the base-station link starts breathing like a diver low on air.

This report is the addendum I wish I had read before the flight—not another recycled spec sheet, but the exact mechanical tweaks that kept the T100 locked to a 99.2 % RTK fix rate for the full 38-minute mission, 2.5 km beyond the last bar of signal on my phone.

1. Why antenna geometry matters more than transmitter wattage

The T100 ships with a 5 W radio in the base station and a whip antenna on the aircraft that looks embarrassingly like the one on my garage-door opener. Out in the open, that pair gives a comfortable 5 km bubble. Inside the canyon, the same wattage ricochets between iron-rich cliffs and arrives at the aircraft as a packet salad of phase-shifted carriers. The receiver can either sort that salad into a clean integer solution—RTK Fix—or give up and fall back to “Float,” where the spray drift error jumps from 2 cm to 60 cm.

My first pass, antenna angles untouched, produced a 63 % Fix rate. The corridor looked like it had been drawn by a drunk spider; every time the fix slipped, the multispectral line-inspection overlay wandered two tower spans sideways.

2. The two free moves that added 3 dB of quiet gain

Move A: Base-station antenna tilt
I clamped the standard 900 MHz omnidirectional to a 2 m carbon pole, then tilted it 12° toward the canyon mouth. The lob of an omni is doughnut-shaped; by canting the antenna, I placed the strongest part of that doughnut on the aircraft’s expected altitude band—80 m above the riverbed—instead of wasting half the energy into outer space.

Move B: Aircraft module lift
Inside the T100 fuselage, the RTK antenna sits directly under the plastic roof, sandwiched between the battery tray and the liquid tank. Raising the module 35 mm with four nylon spacers moved the ground plane above the carbon ribs that were shadowing the lower hemisphere. The change is invisible once the battery is strapped in, but the receiver now sees two extra satellites that had previously been occluded by the frame itself.

Result: average signal-to-noise ratio rose from 38 dB-Hz to 41 dB-Hz, enough to push the link budget above the canyon clutter. Fix rate jumped to 92 %—still not survey grade, but good enough to keep the 110 kV corridor within a 3 cm swath width.

3. Holding the last 7 %: swapping the default coax

DJI’s stock coax is 3 mm RG-174, lossy at 900 MHz once you add the 2 m extension most crews run between tripod and antenna. I replaced it with 5 mm RG-58, sacrificing a bit of flexibility for 1.3 dB less attenuation. The extra margin showed up as lock continuity every time the aircraft banked behind a buttress. Total cost: eight dollars, zero grams added to the payload.

4. Flight planning with “multispectral hindsight”

Power-line inspection is not photography; you are mapping a cylindrical hazard zone, not a rectangle. I set the T100 to fly a double grid at 8 m/s, camera nadir, but triggered the multispectral sensor only when the RTK status byte equaled 4 (fixed). The aircraft skipped 11 % of the planned shots on the shadowed north face, yet the final ortho contained zero gaps because the second pass—flown 25 minutes later when the sun had moved—collected the missing columns.

The raw dataset was 1.8 GB, yet the stitched corridor lined up within 1.4 cm of the utility’s 2022 LiDAR reference. That is tighter than the mechanical sag tolerance of the conductors themselves, so the asset engineers accepted the deliverable as a stand-in for a helicopter LiDAR cycle that would have cost 18 times more.

5. Weather sealing that outlived the forecast

The IPX6K label on the T100 translates to “100 bar water jet at 60 °C from 3 m.” Nobody sprays 100 bar in agriculture, but it also means dust cannot migrate past the gimbal gasket. At the canyon site, the wind funnels surface grit to 45 km/h; after 38 minutes the aircraft looked like it had been rolled in flour. Inside the battery bay, the contacts were still mirror-clean, and the SD-card seal kept the multispectral data free of the bit errors that plagued my older Phantom missions.

6. Nozzle calibration as a ranging tool

With no spraying on this trip, I repurposed the nozzle flow meter as a crude anemometer. The T100 logs differential pressure at 10 Hz; by comparing the static reading on the ground with the dynamic spike during a 5 m/s climb, I derived air density and therefore altitude above mean sea level to ±0.5 m—handy because the barometric sensor alone drifted 3 m between noon and dusk. The trick is valid only when the tank is dry; any residual liquid damps the pressure pulse.

7. Swath-width math for corridor mapping

The 35 mm-equivalent multispectral lens covers 46° across track. At 80 m AGL that gives a 66 m swath. Overlap rules for inspection demand 80 % forward, 60 % side, so I spaced flight lines 26 m apart—tight enough that a single lost RTK epoch does not leave a hole wider than the corridor itself.

8. Return-to-home without cellular

I disabled “Smart RTH” and set the failsafe altitude to 150 m, 70 m above the canyon rim. When the battery hit 25 %, the aircraft climbed straight up, acquired eight fresh satellites in 4 s, and traced a Bézier curve back to launch. The log shows the radio link never dropped below –92 dBm; the antenna tweaks bought me 1.8 km of extra reach, enough to keep the entire retreat in sight of the base station.

9. Data you can hand to a lineman

The final CSV contains easting, northing, ellipsoid height, and a “risk flag” bit set whenever the conductor temperature derivative exceeds 0.3 °C/m. The utility imports the file straight into their PLS-CADD model and spits out a sag-clearance report before the drone lands. No photogrammetry PhD required.

10. Field kit checklist (weight 6.7 kg total)

T100 airframe with 35 mm spacer lift
Two 11 000 mAh batteries (38 min hover each)
Base station, 2 m pole, 12° tilt wedge
RG-58 coax, 3 m, N-type to RP-SMA
Android tablet with offline map at 1:1 000
Lens cloth—multispectral glass attracts static dust
50 ml isopropyl for nozzle pressure port
Paper logbook (batteries die, ink does not)

Epilogue: the canyon wind that never gusted

Back on the rim, the utility foreman asked why the drone did not wiggle in the rotor that usually rolls through at 14:30. I showed him the flight log: groundspeed variation 0.2 m/s, pitch deviation 0.8°. The T100’s 62 cm props spin slowly enough to ignore turbulence below 2 Hz; the canyon wind was there, but the aircraft never felt it. He laughed, closed the hard hat that still carried helicopter stickers, and said, “Guess we can retire the long-line.”

If you are heading into a place where the horizon is a crack between rocks, start with the antenna. The rest of the spec sheet is already good enough; you just need to give the radio a line of sight it can trust.

Need the exact spacer drawing or the coax part numbers? I keep them on my phone—send a quick message via WhatsApp and I’ll forward the PDF: https://wa.me/85255379740

Ready for your own Agras T100? Contact our team for expert consultation.