New Products Flash Sale
Jack Crossfire's blog View Details
Posted by Jack Crossfire | Apr 13, 2009 @ 05:12 AM | 2,852 Views
Seems U need a lot more than 16 MIPS to get 450khz sampling so those $5 parts wouldn't fly. Fortunately, got it up to 260khz with a lot of assembly language tricks. Basically, accounting for every single clockcycle, unrolling the main loop, stripping the USB stack to the minimum, polling instead of interrupts, replacing subroutines with a big loop. The USB stack is down to 850 bytes.

Got analog crosstalk at 260khz, so it ended up taking nops to slow it down to 200khz. Some say it's worth building 2 USB devices at 260khz to see if sonar is feasable at all.

So with the nyquist now 50khz, the Maxbotix EZ-1 replaced with a bare transducer, only 2 dimensions being measured, the result was yet another incremental improvement but still worthless.

The next step would be repeating basic range tests at different sample rates.

Can't get good results unless your nyquist is several multiples of the ping frequency & we're only at a 50khz nyquist. Like a 44100 CD trying to generate a 14khz tone, sometimes the samples line up, sometimes they beat, & the beats make it sound grainy.
Posted by Jack Crossfire | Apr 11, 2009 @ 09:28 PM | 2,676 Views
Another set of batteries & sonar is definitely busted. Even with all the progress in omnidirectional sensing without propellers, with propellers & a large array, range was still a narrow cone directly above the array. The large airframe & the lack of accelerometers & magnetometers made positioning too sloppy to stay in the small cone. Altitude was nowhere close.

What happened with the array if single sensors had great range? Suspect the spacing in the array made every sensor have to look farther sideways. Another possibility is peripherals in the recycled Maxbotix EZ-1 interfering with the tone generation. Finally, there's the 30khz sampling creating aliases of propeller noise & destructive interference between aliases that wasn't there in 150khz sampling. A tone at 40 & 20 that didn't self destruct at 150khz would interfere at 30khz.

The best results were a pair of 13 second hovers directly over the array. Used really high gains & no altitude feedback. Almost had to flip it over to get feedback tight enough.

Your best chance with sonar is the smallest airframe possible or a blimp. Maybe an E-sky with SMT board. It would be a great replacement for an Evolution Robotics sensor if it was a ground vehicle or a blimp.

More likely you'll get another electronics upgrade. Sonar is that important 4 U.
Posted by Jack Crossfire | Apr 11, 2009 @ 01:58 AM | 2,921 Views

It takes a lot of current to feed an ADC because they need to recharge a sample hold capacitor. This makes direct IMU or thermopile connections noisy & op-amps a good idea. Vref's for ADC's need a lot of current too, so you're better off stockpiling quad op-amps. No surprise that the cheap, crummy, obsolete Picopilot has an op-amp buffering the ADC inputs but we don't because we're too poor.


Also, for any conversion from an analog value to a geometric solution, we found it better to smooth the raw analog values than the geometric solution.


What a freakshow that was. Another subfreezing La Ninia April. The first dead laptop battery in flight. Uncalibrated feedback gains. Massively underestimated propeller noise in the derivative. Only a tiny box in which sonar position was valid. Not enough control to stay in the tiny box.

50% bandwidth is required on position & 25% bandwidth is required on velocity.
Posted by Jack Crossfire | Apr 10, 2009 @ 01:52 AM | 2,709 Views
1) U can make more thrust by crumpling dollar bills & throwing them from the back of a spaceship than U can by burning them in a combustion chamber.

2) Unless U plan on flying them into orbit for less than the cost of throwing away crumbled dollar bills, they're a lot of work with no gain 4 U.

3) We glaze over when hearing mixture ratio, molar mass, fractional dilution, distillation ratio, & freezing point depression.

But, Henry Pohl's lecture had a lot more basic engine theory than J. R. Thompson because he was an engineer while J. R. Thompson was a manager.

Did U know the OMS fuel injector was made by UV lithography? They burned photo resist patterns onto extremely thin sheets of metal to etch away thousands of holes. Then they stacked thousands of those sheets to create a 3D manifold. Drill bits weren't accurate enough.

The minimum pulse of the vacuum started engines was 60ms because residual fluid would evaporate & freeze after firing in a vacuum. Eventually the freezing would overcome the amount of heating required to ignite the fuel & the engine would be dead. So they needed 60ms of firing to reverse the amount of freezing in shutdown.

The OMS engines need 20s of firing to reach optimum efficiency. They can't be used for shorter burns because they would waste too much fuel. Also, the 0.61m/s/s created by the OMS engines prevents anyone from sleeping or using the toilet.

The RCS engines R film cooled. They inject a lot of...Continue Reading
Posted by Jack Crossfire | Apr 09, 2009 @ 03:39 AM | 2,989 Views
We've gotten so much mileage out of empty plastic bottles, Calif* is probably going to outlaw recycling to keep the landfill union in business.

Definitely more accurate at 10Hz because we have fewer echos, but also a lower threshold is calculated because there R fewer echos. A bit more accurate if U play with the threshold calculation. Suspect an adaptive threshold based on position noise would be beneficial. Maybe a neural network.

Got it to where it would fly if the propellers were totally silent & it didn't bank too much, but we know the propellers make a lot of noise & change the threshold.
Posted by Jack Crossfire | Apr 08, 2009 @ 02:33 PM | 2,509 Views
U once had a boss who spent forever & everyone else's time building a corny rack & pinion machine to show how a touch pad could be used to control camera zoom & how the touch pad could be used in 3 different zoom modes. Touch pad position either controlled absolute zoom, relative zoom, or constant zoom rate.

Guess in his mind, showing the mechanical mechanism of a lens somehow conveyed these zooming modes & this would somehow make Japanese investors invest in a touch pad instead of a rocker switch. This was long before the iPhone, when no-one ever heard of such a concept.

No-one ever suggested he just write a simple piece of software to actually magnify a photo because they'd get fired for making him look like an idiot & also because of the top rules of the white collar world:

#1 Don't suggest anything that makes the boss's idea look completely idiodic.

#2 If U suggest it, it'll get thrown out because the boss didn't think of it. If U don't suggest it & the boss thinks of it, it'll get accepted. So the best way to get something accepted is to not suggest it & wait for the boss to think of it.

We were actually in the room when the VP proudly explained how the clunky rack & pinion demo would bring touch pad zoom to the cameras of the world, wondering why the fools didn't magnify a photo in software.

Well not only a huge waste of everyone's time, but the Japanese eventually canceled that project. Fortunately for the world, a manager at Apple did magnify a photo instead of build a rack & pinion machine.
Posted by Jack Crossfire | Apr 08, 2009 @ 01:17 PM | 2,556 Views
Clock synchronization with USB is a real buster. There's a lot more jitter than RS232. Some say it's the audio capture load. Some say it's bugs. Fortunately, no danger of the dry weather for 2 weeks.

To get clock drift, U need long histories of clock readings from the 2 computers to take the readings which R most similar, but the longer the history, the slower the drift update. U need to replace the history as fast as possible to get good gaussian drift results.

Also tried waveform derivatives & there was a small improvement at longer distances.

Angling the sonar array 45' didn't work. Too many reflections from the floor.

Everything's moving towards 1 flight with the array, then a move towards reflected sonar altitude & a dipole. It's a 4 wire connection to an op-amp board, assembly language implementation of the DSP, reconfiguration for higher samplerates, UART interface to the sonar board.

Of mane concern is transporting the sonar array, so we'll probably use a bigger array on pods & dump the aluminum.

Component testing has gotten really anal with the last few airframes. It takes too long to set up, we don't officially have any time off anymore, & you're now experts on what is required to fly, so the systems have to work the first time & U don't need as much flight time as U used to.
Posted by Jack Crossfire | Apr 07, 2009 @ 03:31 AM | 2,989 Views
So how do U plug sonar positioning into a GPS autopilot? Convert the sonar results to fake latitude & longitude of course. & for all U GOOG shareholders, add offsets so they can be plotted in Goog Earth.

Unfortunately, 20Hz is real problematic. The indoor reflections bleed into the next frame. Fortunately, got the CPU usage back to 1%. The ping rate isn't exactly what's configured, so U get 25Hz & multiple pings in the same frame. Measurements are starting at random offsets.

Also, try derivatives instead of peak levels.


Now as much of the Obooobu stamp of defense cuts as U can make out from the AP coverage:

F-22: GONE
Italian Marine 1: GONE
Missile Defense: GONE
Airborne laser: GONE
Predator & Reaper: increased by $2 billion
F-35: 16 more planes
Mortgage bailouts: 80,000,000 more mcmansions

Fortunately, there R plenty of jobs 4 U unemployed laser gun experts in N. Korea. FYI, we own a small number of United Technologies shares, so buy buy buy.
Posted by Jack Crossfire | Apr 06, 2009 @ 01:00 PM | 3,101 Views
Finally got a 3D position out of sonar. All those small errors really add up. It's 1ft off in the air. On the ground, it's hopeless. It is 20Hz at least, it's better than GPS, & it's indoors. It would definitely improve with a larger array & 5x faster sampling. What do U expect for $20?

So it's taking 42% of an 800Mhz CPU to do sonar navigation. Sometimes clock drift works & sometimes it doesn't. The triangulation algorithm is the least efficient possible. If the copter clock is ahead of the ground clock, fuggedaboutit. As Kris Kraft would say, U need to start flying it regardless of the imperfections, to know if sonar is completely nuts.

There was once a large book of software anomalies for the space shuttle. The risk & cost of software changes was too high to bother fixing it.

So, in a 10 minute flight, Vika 2 needs to process 55 megabytes of telemetry to stay in the air. In comparison, your GPS vehicle with 14 satellites crunches 1 gigabyte in a 10 minute flight.
Posted by Jack Crossfire | Apr 05, 2009 @ 09:27 PM | 2,833 Views

Can't believe Heroinesapien fizzled out right after release the way she did. U people forgot about her within a month of 2008 CES. At least we know she could walk on carpet if properly hacked.

Heroinesapien walking on carpet (2 min 15 sec)

Another note on Heroinesapien: The servos are timed so if they don't reach their desired position in a certain time, they just stop. A little smarter than RC servos which just burn out.


Got the latency figure for ground based autopilot & it's 60ms. It takes 60ms to transmit an event from the copter, to the ground station, & get the response back. So really the feedback is 16Hz & the other 34Hz R redundancy for interference. Got that by comparing transmitted packet numbers with returned packet numbers.


Also, with the increased workload on the groundstation, we're now aligning the sonar clocks to within 21 useconds or 0.27" of distace error. Pretty horrible.

Remember using the Maxbotix to sense altitude? It was killed because the Maxbotix didn't have an accurate enough trigger, was too expensive, the shrouds defeated XY sensing, & U thought U could synthesize pings on the flight computer to save a few milligrams.

Well, the answer is no. Your flight computer doesn't have enough clockcycles to synthesize pings. It can only reach 8khz. It turns out, when 8khz pulses are applied to the transducer leads, current...Continue Reading
Posted by Jack Crossfire | Apr 04, 2009 @ 11:04 PM | 2,815 Views
At least all those laid off Boeing employees have jobs in N. Korea.

Now the headline on CNN:
> The NCAA men's basketball Final Four couldn't come at a better
> time for the city of Detroit.

With a budget of $654.1 billion, your defense department is going to pay most of the cost of new mortgage, SUV programs & that oh so precious NCAA men's basketball final. The prime cuts will be:

airborne laser: gone
Air Force refueling tanker: gone

Our gut reaction is most of the 1 man UAV startups will be gone & the budget is going to focus on increasing the number of tried & true predators & global hawks to replace manned planes. Hummingbird A160, MQ-8B firescout, X-45 UCAV, X-51 waverider: fuggedaboutit.
Posted by Jack Crossfire | Apr 04, 2009 @ 07:06 PM | 2,771 Views
The only thing we need from Canadia is Motor-7mm 1.7Ohm. Can't believe no-one besides Canadia sells a stupid motor. If there was any other place to get it from, or humans were more creative, this story would already be over.

Kris Kraft notes

Kris Kraft thinks like most of the world exept US. Mainly, the rest of the world thinks about doing while US thinks about politics. He wished China landed on the moon tomorrow because it would make U think about something besides Michelle Obama's dress.

He didn't say anything U don't already know from watching the many TV reruns of Apollo footage. He thinks the space shuttle could be economically viable if it wasn't for politics, job losses, constituents, lobbyists, unions, & basically the USA. The shuttle program was required to buy components from every state. Sheesh.

The only reason the shuttle happened was because it was the only reason the Soviets were willing to negotiate in 1 meeting in 1979. It was really Peanut Head's call, but don't bother searching for Jimmy Carter names on the Goog.

Thinks the shuttle should fly with less of the redundant systems functioning, because that's what redundant systems R for. The main engines could be run at lower power, the TPS could be more robust, & the hydrolics could be made electronic. It could have 100 redundant computers with today's technology. Mainly, it should continue evolving like the 60 year old B-52.

Thinks a lunar vehicle should go between low Earth orbit & the moon, & rely on the shuttle to move people between orbit & the ground. The space station should be moved to a 28' inclination with many incremental propellant burns & become the docking station for the lunar vehicle.

Like J. R. Thompson, said the Orion would be just as problematic & expensive as the shuttle, with a fraction of the capability, because it's the management, not the vehicle. But doesn't think Orion is ever going to happen because of the cost of mortgage programs.
Posted by Jack Crossfire | Apr 03, 2009 @ 04:59 AM | 2,906 Views
Vika 3

Deriving position from realtime digital audio processing is really hard, so lets start thinking about actuators & motor mounts for Vika 3.

Building a micro copter out of Canadia parts would cost over $100, so you're going to be building parts from scratch. Canadia charges $15 for a crummy coil of wire. People R mounting the motors with rubber bands on sticks. Aft thrust vectoring is the hard part.

Ground based autopilot then

Originally, no-one knew if ground based autopilot would work. The latency, full duplex bitrate, & interference effects were unknown. Today we know the maximum feedback rate is 55Hz, & there is enough bandwidth. The latency is 1 calculation U need a day off from your boss to do.

Ground based autopilot now

When the timing software was added, it brought the maximum sampling rate from 47khz to 40khz per channel. Since no alias frequencies show up at 40khz, U need to run it at 30khz per channel. Now the ground station is only 739kbit. Really U expected it to go to 30khz anyways.
Posted by Jack Crossfire | Apr 02, 2009 @ 02:04 PM | 6,254 Views
He sounds skinnier & more serious than he looks. Says "umm kay" a lot & is very beaurocratic. Probably doesn't trust anything that bleeds for 5 days & doesn't die.

Clean pads don't save money. The time required to move a rocket back to the hanger & fix it costs more than fixing it on the pad. U need to do as much on the pad as possible.

U need the largest mission control possible. U need as many people on the ground as possible to do what doesn't have to be done in space. This may be more important for manned missions.

The space shuttle IMU has to be reset from the ground once a day. It doesn't have star trackers, accelerometers, or magnetometers. If the update has an error, as happened many times, it'll burn up all its fuel & lose radio contact. IMU's have been replaced by GPS everywhere except the shuttle because it wasn't supposed to fly after 2010.

U need a functioning IMU to align the antennas & U need aligned antennas to keep the IMU functioning. The shuttle can't land without ground communication either. Ground communication is a huge failure point.

Now our highlights video of STS-119, since all the other copies of these highlights were blurry, improperly exposed, jerky & unedited.

best of STS 119 (2 min 25 sec)

The modern consumer electronics industry is basically a space program. The gadgets U read about in the news aren't made by single companies anymore. They're just too complicated,...Continue Reading
Posted by Jack Crossfire | Apr 01, 2009 @ 04:29 AM | 2,750 Views
So U want to easily swap between USB & UARTs for easy conversion between a ground based autopilot & an embedded autopilot eh. Well, fuggedaboutit. All the FIFOs, single byte reads, blocking, timed waits, & polling that UARTs did 4 U have to be written manually for USB.

At least U finally got the XBee, 72Mhz, & audio all crammed into a single cable. Try doing 1.1Mbits over an Ardu nacho taco dilla. Feedback frequency is also a bit more stable over USB instead of serial dongles.

To make it fit in the available clockcycles, U ended up with 6 endpoints, with all the IO except the audio directly between the peripherals & the endpoint DMA area. There's no way to do the audio without memcopies.

E: Ad=01(O) Atr=02(Bulk) MxPS= 64 Ivl=0ms
E: Ad=81(I) Atr=02(Bulk) MxPS= 64 Ivl=0ms (audio)
E: Ad=02(O) Atr=02(Bulk) MxPS= 32 Ivl=0ms (to xbee)
E: Ad=82(I) Atr=02(Bulk) MxPS= 32 Ivl=0ms (from xbee)
E: Ad=03(O) Atr=02(Bulk) MxPS= 16 Ivl=0ms
E: Ad=83(I) Atr=02(Bulk) MxPS= 16 Ivl=0ms (from 72Mhz)

To Xbee had to be big enough for a complete uplink packet, to keep the XBee from splitting it into 2 802.15.4 frames & destroying the full duplex rate. U could have done all this on the latest $1 ARM chip without any optimizations at all of course.

Unfortunately, not enough clockcycles to do 50khz audio & radio on the same chip. It's stuck at 47khz. Theoretically, sonar is optimum at 50 & 30 because these are 10khz away from...Continue Reading
Posted by Jack Crossfire | Mar 31, 2009 @ 01:12 PM | 3,242 Views
He rambled a lot about management styles & project planning. That's what software is for nowadays.

The SSME was tested to destruction. That's why it's never failed, so far. No other part of the shuttle was tested to destruction, so the rest of it has had problems as managers speculate on the lethal conditions.

The turbine blades & the bearings crack in every flight & R reused after cracking. They couldn't get them to stop cracking, so instead of fixing the problem, they tested to destruction to know when the cracks were lethal & told the astronauts they were riding cracked blades.

Unfortunately, it's real expensive to test to destruction. U ever blow up a bag of LiPo's to figure out what their true C rating is?

The A1 test stand has no roof. They can't fix an engine if it's raining. The 1 time they put a roof on it, it trapped pure O2 from a leaky valve & made all the wires on the test stand burst into flames when the engine started.

He thinks NASA should have continued improving & testing the shuttle after 2003 as if it wasn't being phased out. Now there R no more improvements, the experts have left the program, & you're flying extension after extension on a vehicle which is no longer tested.

He thinks NASA emphasized pushing the limit of technology too much & should have derated the performance to increase margin. The shuttle was no more dangerous than any other system, if its performance was reduced.

NASA can keep making new vehicles all it wants & they'll be just as dangerous as the shuttle. By pushing the limits of performance & not testing components to destruction, they'll make Ares-1 just as problematic & expensive.
Posted by Jack Crossfire | Mar 30, 2009 @ 02:48 PM | 3,073 Views
Dreamed Russian Heroine became a rich banker & formed her own olympic swimming program. She hired some coaches & athletes. She was much older, sitting in a chair next to the pool, & her athlete lover swam over.

We figured the program was just a way of providing a constant supply of men to make love with. She still recognised us but was much more interested in her athlete.

In another dream, we joined a startup full of Russian programmers & she came in, yelled & screamed about stalking her. It was on some university in a classroom.

Vacuum tubes forever

Hopefully U all saw this Gizmodo revelation about vacuum tubes.

Electrons on Parade pt 1 - RCA (9 min 31 sec)

U can't blog about sonar without bringing up vacuum tubes. The answer is yes, BYU used vacuum tubes in its recording studio in 1997. More specifically, a pair of tube leveling amplifiers. Vacuum tubes still are the only way to get certain sounds just like a Steinway is the only way to get certain sounds.

So U want to simulate vacuum tubes in software, eh.

The truth about digital audio

Sampled audio at 48khz can't reproduce a 440Hz sound. It can alternate between 440.366972Hz & 436.363636Hz to try to fake it, but it can't do the 440Hz your great great grandparents heard on their LP's. To make discretely sampled audio do every frequency that analog did, it would have to sample millions of Hz at 24 bits. It literally amounts to digitizing the molecular structure of the complete physical tape.

Wouldn't be surprised if no-one in the current generation ever heard true A440.

Kinetic energy busters

The kinetic energy credit card had some more problems. It took -2v to rewind the first mass, so the second mass needed 3v. The 3v mass would finish its movement before the -2v mass rewound. Since each mass movement took less time to complete than the previous one, a new mass would be required for every movement. There was no way to defer kinetic energy repayment.
Posted by Jack Crossfire | Mar 29, 2009 @ 06:32 PM | 3,267 Views
3 years ago, U envisioned the kinetic energy credit card. It accelerated payloads to 17,500mph while deferring the equal & opposite reaction for later. At 17,500mph, it released the payload & then repayed the equal & opposite reaction by returning to Earth & discharging the rest of the debt into the ground.

It used 2 moving masses in real long tubes. 1 moving mass shot aft to accelerate the payload forward. When it reached the end, it reversed direction & another moving mass shot aft with enough velocity to pay back the direction reversal of the first mass + additional payload acceleration. They kept alternating & accelerating exponentially to build up payload velocity.

In reality, it would have used millions of oscillating masses, they would have exceeded the speed of light real fast, & the tubes would have been really really long vacuums, maybe several miles long.

It could work with a ground based vehicle. It would allow the vehicle to travel a short distance without requiring traction.
Posted by Jack Crossfire | Mar 29, 2009 @ 06:06 PM | 3,121 Views
There is no direct servo control via IR on the Heroinesapien. The 1 kid who went through all the trouble of uploading a teardown to the internet didn't label the fricking chips.

U would have to splice 3 wires for each servo to control them manually. U would have to bias the potentiometers without turning on the H-bridge driver. All the leg wires R hard soldered to the board, so U couldn't just make a board with connectors & plug it in.

Walking requires changing the center of gravity by moving the arms & body motor, then moving the leg motors....Continue Reading
Posted by Jack Crossfire | Mar 29, 2009 @ 01:18 AM | 3,299 Views

Automatically add someone famous to the subscribers list of every account. You'll have them falling over themselves with glee, taking their comments seriously, & referring your social network to the world. Gizmodo originally did that.

Make a fake IQ test which always gives rediculously high numbers.

Most importantly, use a strange font that no-one's seen in years.

Sonar busters

Phased array sonar is busted. The ping is too repetitive & the waveforms aren't similar enough for absolute differences or correlations of the waveforms to work. Was hoping correlation could get a hedge from multiple distance readings. It would work if the frequency was super low & the microphones real close together.

It detects phase in a single period nicely. That could be applied to velocity or error correction.

All U can get are independant distance readings from the microphones.