Jack Crossfire's blog View Details
Archive for December, 2006
Posted by Jack Crossfire | Dec 31, 2006 @ 05:17 AM | 11,593 Views
In a record 2 days of soldering and debugging, copterbrain is alive. Its bootloader is programming its flash and sending back debugging info.

Currently the servo inputs are sharing the programming pins because the programming pins are the only ones which can generate interrupts.

There actually is 1 more pin capable of generating interrupts, should the need for a collective pitch servo arise.

Copterbrain is our second use of a double edged clock for the parallel port communication, allowing much faster programming.

One trick to get copterbrain to communicate was to delay between toggling an output pin and listening to in input pin. Toggling output pins caused transients on the input pins.

The next step is to benchmark math routines. Probably going to be a fiasco.

Before flight:
Drop from the current, overclocked 48Mhz to the rated 40Mhz.
Move programming pins away from servo pins.
Posted by Jack Crossfire | Dec 30, 2006 @ 04:59 AM | 12,462 Views
So to build more confidence in the autopilot plan, weighed a mockup of as many of the electronic parts as possible. Solder and ribbon cables are going to weigh it down even more.

Managed to reverse engineer the software flow in the autopilot firmware and unfortunately it involves tons of + - * / sin cos tan asin acos atan of 32 bit floats. It's going to be a long run of point-to-point soldering and then math tests just to see if this MPU has enough horsepower.

This autopilot firmware uses quaternions and their bird nest of trig functions liberally. In a previous experience with quaternions, discovered they were replaced with much faster 4x4 matrices and linear functions in the 80's.
Posted by Jack Crossfire | Dec 28, 2006 @ 03:05 PM | 12,865 Views
For the morbidly fascinated, we have weighed the 4 legged monster and some parts of the autopilot. All autopilot components must come in below 160g to keep the Mega 16/15/4 from overheating.

Latest revelations with the autopilot show it can't fly on accelerometers alone. Accelerometers show 0 acceleration in a tilt because the copter always makes enough horizontal movement to keep accelerating directly opposite the rotor force. Accelerometers show horizontal acceleration once the tilt is complete, from horizontal translation.

It needs the gyros to determine the tilt and accelerometers to determine translation. When 0 translation is observed, the gyros are reset to 0 tilt.
Unfortunately the accelerometers drift, so 0 observed translation eventually becomes significant translation.

GPS is the final link which resets the accelerometers to 0 translation. So it's GPS resetting accelerometers and accelerometers resetting gyros. All this requires the copter to stop periodically to reset its IMU of course.

The trick with GPS seems to be using it to determine relative position instead of absolute position but there's nothing on the internet about comparing a current reading with previous readings from the same unit.
Posted by Jack Crossfire | Dec 27, 2006 @ 05:05 AM | 13,109 Views
Trying to learn about this stuff is like trying to figure out witholding allowances for maximum 401k contributors. The main problem in copter autopilots is knowing the orientation of the copter. The current state of autopilot seems to be:

1) Robbie Helicommand: uses downward video to hold position. Only low altitude due to requirement for high resolution ground features.

2) Micropilot 2128 Heli: $6000 Uses GPS and inertia. 1oz

3) CARVEC: "Low drift MEMS" Another inertia system.

4) FMA Copilot: IR. $70 Can't hold position. Uses only absolute roll & pitch but not gimbal rates. Videos show it has enough information to maintain attitude but quickly flies away without manual control to maintain position.

5) rotomotion.com: They use 2 accelerometers to detect movement and 2 gyros to detect rate of gimballing motion.

People are loath to give out any details of their autopilot algorithms.

GPS is too slow and inaccurate to hover a helicopter. One $50 GPS unit claims a 1Hz measurement within only 7 meters. Most of the work is definitely by inertial measurement.

Most autopilots use a combination of accelerometers and gyros. This allows the computer to differentiate translational movement from gimballing movement. This alone seems to be enough information for a helicopter to hold a position.

The autopilot algorithm involves plugging the sensor output into specific slots of a matrix, transforming the matrix by a complicated, fixed...Continue Reading
Posted by Jack Crossfire | Dec 25, 2006 @ 07:44 PM | 12,495 Views
So flew the fastest ever today. Got it probably faster than Mr. Arlton intended it to go. The tail starts wagging quite a bit when it's maxed out and when going fast enough, it seems to spin out of control.

Had the first experience with death stalls today. The idea is to get moving very fast from left to right. Then pull up hard when it's at 12 oclock. Without a heading hold gyro, it yaws either nose-in or nose-out, usually indeterminable.
This seems to be the cause of the 2 latest crashes. The only solution is to watch where it goes after the yaw and assume the nose is pointing into it.

So considered a Canon A530 or a new Sanyo HD1 for copter photography. Decided the best route is instead of buying more cameras, to focus on developing an autopilot system. Copter photography isn't going to get much better with humans at the controls, no matter how wide the lens is or good the motion tracking is.

Hopefully autopilot technology is advanced enough for a properly armored human to walk up to it and grab it in a hover, eliminating the need for landing gear....Continue Reading
Posted by Jack Crossfire | Dec 24, 2006 @ 09:48 PM | 11,956 Views
It happened in the same spot as last time. The sun was low, right behind the copter. Rudder control started behaving backwards but it looked nose out. The hands kept saying it was nose-in but the eyes kept saying it was nose-out. Didn't have time to pick a winner before it plunged into the U Know of Where.

After reshaping the tail boom, it was real unstable.
Posted by Jack Crossfire | Dec 23, 2006 @ 08:31 PM | 11,787 Views
The current flying field is 720x180ft at night
but only 418x201ft by day. The new field is 883x297ft or 883x594ft with the old man's plot & house. Quite an improvement. Went ahead and bought the 883x297 in May instead of waiting for US$ to fall through the floor like they eventually did. Too bad all the jobs are 3000 miles away but it beats saving $.
Posted by Jack Crossfire | Dec 21, 2006 @ 04:43 PM | 11,140 Views
There doesn't seem to be a comprehensive FPV HOWTO which lays out the best equipment and the technique for FPV. It seems to be word of mouth and a lot of bits of information. So far the bits of information converge on the following:

Flight by video with the most expensive equipment is difficult at best.
You can't really fly by video 100% of the time.
The most expensive systems can be flown purely by video, but landings and takeoffs are still by eye.
Cheaper systems allow momentary flight by video with frequent looks at the model to know where it is.
Expect to lose your model occasionally if you fly by video alone and go beyond visual range.
One guy is flying a helicopter by video alone but spent a hell of a lot of money to do it.
Fixed exposure and high resolution isn't required to fly by video. Some fly with only a horizon line to go on and very little detail above or below it.
The difference between wide angle lenses and normal lenses seems hardly noticable. They all lose the horizon.

Personal experience:
The 1mW 2.4Ghz video transmitter interferes with the 75Mhz receiver.
The 75Mhz transmitter interferes with the 2.4Ghz video receiver.
The 2.4Ghz signal can be jammed by security cameras.
You can't tell how high you are from video alone.
With a normal lens, the slightest cyclic input makes the horizon disappear.

Feel free to share any exceptions to these bits.
Posted by Jack Crossfire | Dec 19, 2006 @ 04:34 PM | 11,454 Views
Today's flight was to try once again at keeping an object in the viewfinder as long as possible without looking at the copter. The object was now the Magorator, so it could be viewed from higher altitudes. It was even colder than last night. A 10mph wind prevented any hits for 15 minutes. Then the wind died down and once again, rudder input kept it from moving left and right but it always slipped off the top or bottom after a few seconds.

In the longest holds, the copter naturally pirouetted around the Magorator but we didn't allow it to point nose-in. Also was hard to see the copter because of the need to point the Magorator towards the pilot. Was impossible to detect forwards and backwards motion through the viewfinder. Long periods of flying are going to be impossible without a very wide angle lens.
Posted by Jack Crossfire | Dec 18, 2006 @ 04:58 AM | 11,677 Views
Today's test flight could have happened in daylight but had to be delayed until midnight due to extremely high winds. By midnight there was no wind but it was <32` the coldest weather this copter has ever flown in.

The wind chill from rotor downwash was intolerable but it was so stable it could hover for long periods of time while taking a hand warming break.

The objective was to keep the flightbox in the viewfinder as long as possible without looking at the copter. Flying through this camera was like flying through a peephole. In the dark, with only the flightbox visible, meant reverting back to 3rd person flying every time the flightbox disappeared and that was pretty often.

Using the rudder was the only way to keep the flightbox from moving left or right. If it moved up or down it was impossible to get it back in view.

As pathetic as it is, all the experiments with collinear antennas were worthless. The stock whip antennas on both transmitter and receiver provided the best video when they were finally mounted on the copter. They don't have as much range as the collinears but omnidirection is the key to this copter.

Installed a much longer 75Mhz antenna but still had glitches. Interferance from the Astak continues to plague the 75Mhz receiver. Clearly at least 2 crashes so far have been due to radio problems. It looks like spread spectrum may be the only way out of this one.

Spread spectrum system: $350

The other problem is the stability....Continue Reading
Posted by Jack Crossfire | Dec 17, 2006 @ 03:11 AM | 11,400 Views
It looks like the end of landing platforms. The last test flight was a FPV test with the Astak camera and a pylon just for the transmitter antenna. The reception was horrible. It was a black screen most of the time. During the first landing attempt, pulled back on the cyclic after some failed attempts and the tail hit the ground, damaging the forward tail gears again.

So it was off the landing platform and back to skid mounted antennas. Also replaced the receiving collinear antenna with the original whip antenna and the reception improved drastically. Seems the collinear is too directional.

Then the GWS receiver started glitching again. It seems the 2.4Ghz transmitter interferes with the 75Mhz receiver. At another point 230ft away, the cyclic stopped responding. Seems the GWS locks the servo positions if it loses the carrier....Continue Reading
Posted by Jack Crossfire | Dec 16, 2006 @ 04:01 AM | 11,243 Views
In an attempt to try to lock the exposure of the Astak camera, opened up its optical block to search for some kind of shutter signal or offboard microprocessor. Sadly, it appears the exposure and conversion to composite video is done on the image sensor die itself. There's no way to lift a shutter pin and override it. So that leaves:

1) Get a really light, cheap, solid state, 320x240 camcorder to perfect the flying part and eventually get a Sanyo for HD.

2) Get a Sanyo now and hope there's no job loss or currency collapse.

3) Try aiming the Astak at the ground, debug the radio side, and wait for $700 to materialize before getting a Sanyo.

Mouthwash and toothpaste are getting mighty expensive in the Bermuda Trapezoid.

Speaking of the Bermuda Trapezoid, did the first complete discharge of both batteries since the rebuild and they both had less flight time than before the rebuild.

3300mAh: 23min
4000mAh: 28min

So clearly it isn't the battery but the country that's the problem.
Posted by Jack Crossfire | Dec 13, 2006 @ 04:35 AM | 11,806 Views
So this mount got 2 flights during a 45 minute window between squall lines. On the first flight, The Sanyo failed and lost all the video. The second flight was successful. Blade tracking was 3/8" off on flight 1 and 1/2" off on flight 2. Looked like a twin rotor. The Corona 120 definitely doesn't retain blade tracking between flights. Either the pushrods are too lose or the payload is affecting it.

Landing platform 8 was a much lighter version and only 1ft tall. It was possible to move everything in one trip without being a heroine. Reduced height didn't seem to degrade stability.

Unfortunately, the vibration was the worst ever recorded. Excluding vibration, the footage looked like it could almost be useful. The trick is to enter a stationary hover, settle the climb rate down, use small trim tab movements for tail control, and use the smallest cyclic inputs. With video assist, it should be possible to at least aim it at the right object.

14.55 A consumed. Not bad considering we added a bolt for this mount.
Posted by Jack Crossfire | Dec 12, 2006 @ 12:13 AM | 11,752 Views
So for this test flight, a stiffer pylon and wooden landing platform was built. There was no way to pad the camera without making the pylon too flexible, so the camera wasn't padded. Padding looks impossible because of the pylon's tendancy to oscillate if it isn't stiff.

As stiff as it was, it still had sympathetic oscillation though not as much as test #1. Took 1m25s to land. Wasn't as patient as last time because it was a long flight and the remaining charge was unknown. Nicked the tail rotor on that landing.

Unfortunately, the footage looks no more stable than other methods. For image stabilization, this pylon is going nowhere, but as inconvenient as it is, the landing platform allowed the highest payload for the least energy so it's probably staying. 14A for that flight.
Posted by Jack Crossfire | Dec 10, 2006 @ 08:25 PM | 12,092 Views
So had a few minutes between storms to sneak in the air with the Astak
camera. A black bar and a white bar and lots of interference was all it
showed. Not enough information to fly by. Could use the monitor and
apply rudder to stay pointed at an object but without a ground view it
slipped uncontrollably sideways. Also, usually couldn't keep it pitched
down enough to see the horizon. Maybe aim it down and forget about the
horizon next time.

Flight time with the Astak and no LEDs was 22min, only a 22% drop from
bare bones. Had frequent glitches with the Astak attached. Without the
Astak the glitches stopped.

Did some shots with the Canon pointed straight up. The flash lit copter
under the grey clouds above was definitely unique....Continue Reading
Posted by Jack Crossfire | Dec 10, 2006 @ 06:03 AM | 11,384 Views
With the weather completely closed in, decided to skip test flights of the Astak camera and move straight into video assist for HD. This requires extracting composite output from The Sanyo without its heavy cable and eliminating the Sanyo's battery. The Sanyo uses 5V, so it can
run on speed controller power, or can it?

Note 4 U Sanyo hackers:

The Sanyo uses a mode select pin to enable composite output. 1.25V on this pin enables composite output. The mode select and composite are both horrible, 2mm tall 0.8mm fins.
Posted by Jack Crossfire | Dec 09, 2006 @ 06:12 AM | 10,026 Views
Eliminating the 9V battery for the Astak camera was serious work. The voltage converter ended up gigantic but lighter than a 9V battery.

Although the MAX632 doubles voltage on paper, its output has way too much ripple for the camera. It takes at least 4700uF to eliminate the ripple but since our stock capacitor weighs more than a 9V battery, we have some horizontal bars until the next drive to Silicon Valley.

Replacement parts for the tail gears arrived after 8 days from the notorious "Marks Hobby Warehouse". Nothing beats a relaxing helicopter flight after 8 days terrified of $85 lost. At least it wasn't Euros....Continue Reading
Posted by Jack Crossfire | Dec 07, 2006 @ 02:57 PM | 10,285 Views
Today was the single segment collinear on both transmitter and receiver. Broke the transmitter out of the camera housing and attached to a 4 pin Molex and a much lighter collinear. Much more exciting to have a high quality 16:9 video source capable of filling the entire monitor's display. When the video came through the great beyond, it felt like we were receiving Tokyo transmissions, but of course it was only prerecorded footage from Tokyo.

The single segment collinears had the same range as the 6 segment, but much less polarization. 790ft was now probably an accessible distance for conveying orientation. As the test proceeded, reception deteriorated rapidly, even below 200ft. Seems the ancient lantern battery died. Wanted to get some full resolution captures of this footage to show the full potential of 2.4Ghz wireless. Wanted to try again with the stock antennas. But have to start heading back to Silicon Valley where the jobs are.

Long ago, overclocked an FM transmitter to go 400ft and what a thrill it was. Now FM systems are obsolete. Very soon, these 2.4Ghz systems are going to be obsolete and replaced by spread spectrum systems that go for miles without an FCC license. It's a shame to spend so much energy debugging this system only to have it join our FM transmitter from long ago.
Posted by Jack Crossfire | Dec 06, 2006 @ 03:21 AM | 10,488 Views
So today's test involved fabricating a single segment collinear transmitter antenna. With collinear receiver and transmitter antennas, the improvement was huge. The trend with collinears is not a clearer picture but a longer distance and more polarized crummy picture.

Undecided about reinstalling a microphone because audio gives you rotor speed but is unintelligable with the interference.

Sadly, every day that passes without the $85 parts shipment reduces the chance that we'll fly the Corona again and increases the chance that we'll switch to a Slow Stick. With this kind of parts turnaround, the Corona is basically useless. They don't have the manufacturing capacity to make enough parts to sustain a useful flying machine.

The Slow Stick is hardly the most versatile, but is the easiest, cheapest, least risky way to get aerial HD. If we were software managers things would be different. Maybe there would be a Trex 600 in the future, or a condominium. Unfortunately only 1 in 1000 become software managers....Continue Reading
Posted by Jack Crossfire | Dec 05, 2006 @ 02:41 AM | 10,369 Views
Today's test was the collinear receiver antenna with 4 segments. Range subjectively improved with this one.

Also went with a 1/2 wavelength transmitter antenna. The transmitter antenna makes no difference beyond a certain length.

The theory with the collinear is you can combine multiple antennas in a series circuit by putting loops between them, thus increasing amplification. This collinear was 4 segments. Increasing the number of segments increases the amplification but also makes it more directional.

Range tests are only possible at night because there isn't enough open space.