These aluminum GWS prop adaptors are ridiculous. Lately they started stripping before getting anywhere near the required tension. Down to threadlock on plastic again.

Forget about those tiny set screws. Use TRex-450 bolts. Time to move to China to make our own.


Managed to get the Marcy 1 radios to stop locking up permanently. The ground station watchdog timer wasn't being reset. Definitely a point for microcontroller operating systems. That got it down to 1 lockup which instantly recovered around every 15 minutes. These are probably from the reset pins glitching & we know tying those to Vdd adds noise to the thermopile.

So cannibalized a motherboard to make inductors for Marcy 1. These are made with 20" of wire. The larger the spaces between the windings the better.

Inductors on Vdd & GND got it just a hair quieter than 1 inductor on Vdd.

Wound Vdd, GND, & signal on 1 core & that got the best results & least weight. It's probably down to the EMF through the air.

& the readout at flight RPM finally shows a waveform of sorts. Modulating PWM really kills it.


Now to get roll & pitch out of it you need to lowpass filter it, stretch it to a constant period, average several periods over time, & take the magnitude at the N, E, S, W phases. Already tried getting the magnitude & phase of the peak. That doesn't work because you've got an irregular horizon with peaks that always show up at the same position regardless of attitude.

Think we got roll & pitch curves. On the ground there's a lot of crosstalk. In the air it resembles more normal attitude changes.

Holding it 5ft off the ground, on the golf course, you get decent attitude sensing & there's a 1V thermopile change. Got a mag gimbal lock when she pitched up.

The completely processed thermopile waveform with spikes for azimuth.

Main problem is the DC offset & amplitude are completely random. The difference between max & min temperature for a given angle of bank is not constant & we don't have enough room for a Z pair. Best idea is a set of thermopiles on the ground station for measuring temperature range in addition to the sonar.

Finally, She's too heavy with the inductor & POV to do much flight testing. Don't know whether to start a lighter flight computer now or master attitude sensing on the ground.


Thinking things over, the attitude estimate & feedback on Marcy 1 is going to be so slow you're once again looking at filling in blanks in the data. Neural networks got abandonned because no matter how many algorithms you throw at a piece of garbage data, you're still reading the same garbage data.

They improve stability as long as the input data stays in a certain pattern. An unexpected gust of wind or descent causes these algorithms to destroy the data & your autopilot actually takes longer to compensate than it would using PID loops. We stopped neural networks when uBlox 5 came out.


Headed over Los Gatos at night. You get a nifty view of the valley. Unfortunately only been flying the camera after Major Marcy operations & those have only happened at night. The view would have been better in daylight.

Views: 491

Comment by Mark Colwell on March 19, 2010 at 2:58pm
Looks like you used plenty of thread lock.. Are props still braking off?
Comment by Jack Crossfire on March 19, 2010 at 3:43pm
3x3 props still break off, but not as quickly as 3x2.


You need to be a member of DIY Drones to add comments!

Join DIY Drones


Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service