Jack Crossfire's blog View Details
Posted by Jack Crossfire | May 21, 2017 @ 01:10 AM | 4,071 Views
After much conceptualizing of how to make the handle/camera mount out of carbon fiber, it was decided to just extend the coroplastic.

While the coroplastic on the Ruckus came from storage boxes from 17 years ago which had long been discontinued, it turned out Home Depot still had coroplastic sheets. It cost $14, a lot more than it did 17 years ago. After much experimenting with string & adhesives to bond it, OSH turned out to still have reusable zip ties. All the coroplastic would be held together by zip ties.

After many agonizing measurements, the new cargo container was done. Every effort was made to document the relative sizes of the 3 vehicles before the Ruckus was torn down. At 10" x 7" x 5", the lunchbox has 315-350 sq in, roughly equal to the H King. The H King can't haul as much mass as the lunchbox.

While the H King looks much bigger than the Ruckus, it's only 1" wider & can't be filled any higher than the Ruckus. The extra material is manely the handles & more durable bumpers. Experience proved the container constantly runs into things & people, so it needs strong bumpers.

The new coroplastic was 4mm thick while the old coroplastic was 3.1mm thick. It's so much more massive, the bumpers may be replaced by plastic sheets.

Since the Ruckus had a dying servo & differential, the decision was made to tear it down & reuse its electronics in the H King. There were a lot of memories in the 17...Continue Reading
Posted by Jack Crossfire | May 20, 2017 @ 02:03 AM | 4,202 Views
The H King wheels from edge to edge came in at 8.25" x 12.5". The Ruckus wheels from edge to edge were 7.5" x 10"

It was only slightly bigger than the Ruckus despite going from 1/16 to 1/12. Being 2 wheel drive, there was no shaft down the middle, freeing up a large space. The wheels were all on ball bearings. It was such a better value than the Ruckus, 2015 feels like the dark ages.

Applying lessons learned, there would only be a cargo container & a handle/camera mount with no material over the wheels. The LEDs & radios would all go on top of the cargo walls. The electronics would go under the battery. The cargo would expand from 244 cubic in to 318 cubic in. The old cargo was 15" x 5" x 3.25". It would increase to 4.25" wide.

The handle/camera mount is key. Designing this feature is the great challenge & going to keep the Ruckus on the road for much longer. The scale is big enough to get the camera a little higher, but this requires reinforcing the side panels.

Since cargo is increasing from 3.25" to 4.25" wide, the camera platform either needs to get wider or the handle needs to bend when the camera is on it.
Posted by Jack Crossfire | May 18, 2017 @ 12:29 AM | 3,587 Views
Now the long since deleted interview from the New York University Astronomy Society. You might think it strange to hear so many female voices in an interview about rocket engines, but it was an astronomy meeting, not an engineering meeting. Nothing has changed in the female preference for basic sciences that lead to teaching careers, even 20 years after the lion kingdom witnessed the same dynamic.

Tom Mueller interview (52 min 16 sec)

Tom Mueller is probably just as important to the Falcon 9's development as Musk. Musk consulted him on at least 1 idea & it makes one wonder who originally started the discussion about propulsive landing. It's hard to imagine coming up with the idea of a staged shutdown to reduce the number of valves on the engine. He said it was done on smaller engines but never scaled up. There are no goog search results on what a staged shutdown is.

There were a few other nuggets of information, like how something as small as the injector was a big factor in the engine's total efficiency & how they got the injector efficiency as high as possible. You would think blowing up hardware on the test stand is only what NASA & Blue Origin do, but for the 1st time, Muell revealed SpaceX blew up a lot of hardware in getting staged shutdown to work. We can infer combustion is still too complicated for even the latest advancements in computer modeling to simulate.

There was also the nugget about Goog proposing satellites for internet back haul. The reason is the latency is too long for user interfaces, but mirroring large amounts of data for a data center wouldn't need short latency & it's still faster to mail very large amounts of data than send it through the ground.

Posted by Jack Crossfire | May 15, 2017 @ 11:55 PM | 3,891 Views
Insprucker dropped a few knowledge nuggets during today's launch. They pushed the oxygen loading back to 35 minutes before launch & fuel loading to 60 minutes before launch. When he said that, the race was on to see if it would explode again, but there were notes on the internet about them putting in an extra helium tank & lowering the pressure from the rocket which exploded.

After stage separation, the audience was cheering about something followed by "1st stage lost at sea". It might have been an attempt to soft land the 1st stage in the ocean.
Now that Musk released a video of the fairing steering itself, we know not all the recovery hardware was omitted. Based on the cheering, it must have landed right before the 1st stage crashed. They're not showing complete fairing videos because of BEZOS.
Posted by Jack Crossfire | May 13, 2017 @ 07:39 PM | 3,706 Views
It was disappointing to read the SLS had encountered similar failures as the X-33. According to kiwipedia, the hydrogen tank for the X-33 failed pressure tests shortly before it was canceled. The oxygen tank for the SLS similarly failed weld tests before ever reaching pressure testing. More specifically, the oxygen tank required a thicker wall than any previous friction stir weld & they weren't able to develop a suitable welding tool.

Probably the final nail in the coffin was the LOX dome being dropped by incompetence & damaging tooling. The delay will give bean counters more time to kill it. NASA has a strange mandate to pursue cutting edge methods in some places like the welding, yet stick to age old propellants like hydrogen.

Ironically, it may be the conservative use of aluminum that kills the SLS. In the 20 years since X-33, progress on composite tanks continued at a snail's pace. They could have probably done it more easily with composites than aluminum, now. The powers that be gave Earth the bare minimum amount of materials so if we worked right at the limit of physics & did it absolutely perfectly, we could just barely colonize another planet.

Unfortunately, the room temperature layups with vacuum bags of your hobby project aren't perfect enough to do the job. The carbon fiber needs to be heated & evacuated in an autoclave to reach the perfect mass required. Then, the entire tank needs to be built in 1 piece.

In the 8 Mike Griffin years, they managed to produce 1 new vehicle, an Ares I test article. In the 8 Charlie Bolden years, they produced absolutely nothing. Even in that time span, without counting the moon program, the trend was down. Also noteworthy, 50 years after building the Saturn V, they were completely unable to build an equivalent rocket.
Posted by Jack Crossfire | May 13, 2017 @ 05:41 PM | 4,419 Views
It would be worth releasing the lion kingdom's fully functional BLE stack for the RPZW as open source, but it would make rent go up again when a CEO sold a startup based on it. The mane idea is it starts a shell using Popen, then feeds the input characteristic to stdin while feeding stdout to the notification handle. Stderr is dumped to the console. This restricts the python to just a bluetooth layer. The shell must be started as 'bash' not 'bash -i' or it'll generate SIGINT.

The system is based on modifying bluez/test/example-advertisement & bluez/test/example-gatt-server
1st, you have to run example-advertisement once. This runs continuously as connections come & go. Then, you have to run example-gatt-server to wait for a connection. After the connection ends, it has to be restarted.

There is no callback in the bluez dbus interfact to detect when it's connected or disconnected. Instead, 99% of the time, you get a StartNotify when it's connected. StopNotify callback when it's disconnected. When that doesn't happen, you always get a StartNotify callback when it's reconnected.

When StartNotify is called twice in a row, issue os.system('killall -9 python3') to cause the python script & Popen subprocess to die. This won't kill a subprocess started by the Popen subprocess. That requires giving another process name to killall. An outer loop can restart it, where StartNotify then calls Popen. Besides this method, your BLE client can always send codes to the characteristic when it connects. There are many tricks to redirecting I/O from the subprocess.

There are still problems when starting it from /etc/rc.local instead of ssh. Having more than 1 characteristic also doesn't work.
Posted by Jack Crossfire | May 11, 2017 @ 11:10 PM | 4,924 Views
When last we visited the LED backlight replacement, it needed a more precise way to control brightness than varying the DIM pin voltage. The proper way to use the FB pin voltage requires 2 voltage regulators: 1 to provide 5V to DIM & 1 to provide variable voltage to FB. The decision was made to try dividing the 5V with a 1k pot.

This worked quite well. The current drawn by FB was low enough that the pot accessed the full range of brightness by outputting 1.4V to 4.6V. It was much finer control than the DIM pin. The brightness was also more independant of input voltage, though not completely. So the proper way of using the DF6113 is 5V on the DIM pin/ENA wire & dividing the 5V with a 1k pot for the FB pin/DIM wire.

1/16" balsa was hot glued in to get the LEDs to press against the glass. It needed slightly more than 1/16" of filler.

It's now much brighter than even a modern phone. What a bobby dazzler, like the lion kingdom bought one. There's no artifacting from the LEDs. It's finally bright enough to use minimum waveform intensity. Soldered to the power switch, since that cuts off the 14V. The 15V connector on the back is always on & is really 14.7V. The power switch has 14V. The manes voltage is always on. Tweeked the pot so the 3 LEDs in series got 8.7V.
Posted by Jack Crossfire | May 04, 2017 @ 05:44 PM | 3,841 Views
The ECX Ruckus blew its 2nd differential in 15 months & the servo started jittering. Time to desperately look for more expensive vehicles. Managed to get in 1.5 super fast miles.

# dist sec min/mile
1 211 46 6:6
2 201 46 6:0
3 202 46 6:0
4 212 47 6:9
5 210 45 6:1
6 208 48 6:9
7 201 45 5:57
8 210 48 6:7
9 202 51 6:40
10 201 46 6:5
11 201 46 6:4
12 201 46 6:8
13 207 53 6:52
Posted by Jack Crossfire | May 03, 2017 @ 01:26 AM | 4,609 Views
In the very old days, every LCD display including every laptop had plain old, miniature fluorescent lights for backlighting, right down to the traditional fluorescent light glasswork & mercury. If you're lucky enough to have a piece of 20 year old vintage equipment, they'll need replacement. They slide into the LCD panel for easy replacement. Since this LCD panel is $300, it's a good idea to replace just the tubes.

New fluorescent tubes would be over $50 today, so an even better idea is to get an LED replacement kit from China for $8 delivered. The LEDs would be brighter. After 3 weeks, the LED replacement arrived.

They're nothing remarkable for an LED afficionado. The LEDs are packed as close as possible. The important note is the strips must be shortened for the display. The LEDs are wired in groups of 3 which have clearly visible traces joining each group of 3. They must be cut between the groups.

This driver board is a simple buck converter based on the DF6113. It takes 10-24V & outputs a fixed current. It amounts to a maximum voltage of 3V per LED. Input current at full power, for the full 24" strips is 1A at 10V.

Very important: When the strips are cut, the brightness increases because it's a constant current regulator. Also, the constant current varies based on the input voltage until it hits the maximum current. Then it becomes fixed for all input voltages.

The board is wired according to the "ENA+DIM inverted analog...Continue Reading
Posted by Jack Crossfire | Apr 30, 2017 @ 09:21 PM | 4,222 Views
Iron Man's next objective was to refly a Falcon 9 in 24 hours. We can infer the 1st 24 hour reflight wouldn't be with a customer payload, but some kind of demo mission. They would have to

transport it from LZ-1 to the hangar with the landing legs intact
stack a new 2nd stage, new payload fairing, & payload
transport it to LC-39

Static firing the engines would have to be skipped. All the data which is analyzed manually would have to be automatically analyzed. It was a horrible waste to refurbish an old shuttle carrier, but still have to detach the landing legs. These can all be overcome with more construction.

The mane problem is the sensors. Sensors dogged the shuttle program for its entire life. Today's sensor glitch was an instant 24 hour delay.
Posted by Jack Crossfire | Apr 29, 2017 @ 02:17 AM | 3,847 Views
The Gopro made another trip to the city, yielding another haul of side firing remote cam videos. The mane interest was capturing the mane in an epic mile before it was cut off. The side firing cam definitely gives better results than the rear firing cam.

Lion hits a 6m55s mile (7 min 42 sec)

Lion cools down (3 min 41 sec)
...Continue Reading
Posted by Jack Crossfire | Apr 29, 2017 @ 02:12 AM | 4,070 Views
After 12 desperate years, we finally have a cheap, miniature way to bridge a USB host to Wifi & bluetooth. All those gadgets which only provide a USB device controller for I/O can now be wireless.
Unlike the previous pi's, it really can function in the given form factor instead of requiring a bag full of peripherals. They finally fixed the problem of having to desolder 40 pins to access the GPIO header or having to desolder a bunch of headers to gain access to the USB & power signals. Wires can be soldered directly to large test pads on the back.

The heart of the beast is the BCM4343X, now sold as the CYW4343X. It's a bare die & a sign of how all chips are going to be packaged in the future. It provides all the wireless functions. The same chip is used on the raspberry PI 3B.

The bluetooth side of the BCM4343X is accessed by a UART. The kernel uses hciattach to attach the UART to the networking stack. Then, bluetoothd converts the socket interface into a d-bus interface. The python examples use the d-bus interface.

Current consumption is <300mA at 5V. Clockspeed is 700Mhz to 1Ghz. The kernel can be cross compiled, but all other C programs have to be compiled on the pi to satisfy dependencies. The mane program which needs to be compiled is bluetoothd. This takes 30 minutes & an 8GB flash card.

You must install the full raspian distribution, not the lite version to do anything useful. The 1st partition is a vfat boot partition. The 2nd...Continue Reading
Posted by Jack Crossfire | Apr 16, 2017 @ 05:16 PM | 3,760 Views
Minitaur Explores the Outdoors (2 min 58 sec)

Ghost Robotics had another video, shortly after the last blog post. It's pretty slow on uneven terrain. Legged robots are once again busted.

Meanwhile, the lunchbox had a pretty successful 19 mile drive. It used 3455mAh from battery 1 on the bottom. 4374 from battery 2 on the top. It carried 12oz of fluid for 12 miles & the speaker for 19 miles. With the speaker & a drink, it was the mobile party originally envisioned. It reignited the idea for a new wheeled vehicle with a low center of gravity & much higher payload.

Electric skateboards now abound, but still can't steer remotely & they still cost over $600. They still work, despite rigid axles & no shock absorbers. With the Inboard M1, they started putting the motors inside the rear wheels instead of coupling by exterior belts. The hub motors still provide enough thrust to move a human around & take the minimum space. They might use a planetary gear but the millenial reviewers are completely unconscious about how anything works.

The skateboard is the model for the smallest possible drive train. Steering still requires a big servo outside the wheels. With no shock absorbers or tilt steering, the payload could be right down against the wheels.
Posted by Jack Crossfire | Apr 14, 2017 @ 11:57 PM | 4,477 Views
The closest you'll get to a vintage computer might be the FT2232. It's 1 of the older FTDI chips, but apparently still in production, widely available, & quite capable. It's a dual serial interface. The high end is the FT4xxx line which are quad serial interfaces.

The trick with the FT2232 is to not use the commercial drivers, but use libftdi. Most of the commercial drivers don't support it but libftdi does. Also, libmpsse provides limited abstraction of libftdi into SPI, I2C, & GPIO interfaces, but it doesn't allow simultaneous GPIO control & hardware protocols which the FT2232 does allow & it has many bugs. The trick is to refer to libmpsse to get the command structure for the hardware protocols, but directly call libftdi commands in your own program to control the GPIOs.

It's not clear how he figured out the command structure besides sniffing packets as they went through libusb. FTDI doesn't document it anywhere. Fortunately, it's not a very complicated protocol. The chip doesn't have any concept of GPIO, SPI, I2C modes the way libmpsse does. Libmpsse just uses mode settings to determine what commands to send, but the chip can accept any command in any mode.

The hardest part is figuring out the pins labeled GPIOH & GPIOL are ambiguous. What's really happening is all the pins labeled ADBUS are GPIOs written in a traditional 8 bit GPIO register & all the pins labeled ACBUS are written in another 8 bit register. They can be...Continue Reading
Posted by Jack Crossfire | Apr 07, 2017 @ 12:15 AM | 4,414 Views
It was a long & hard process between 1 month of commutes, but Heroineclock I was finally shut down after 13 years of blinking & replaced by a much better Heroineclock II, factoring in much more experience.

Introducing Heroineclock II (4 min 0 sec)

All the Chinese options which appeared in the last 13 years were studied & found to be too expensive. The largest Chinese digits were 12" tall & $50. EL wire would cost a fortune. Projectors were too dim & small. Superbright red LEDs had dropped from 22c to 15c & become much brighter.

With the new experience, power consumption dropped from over 20W for Heroineclock I to 7W, yet light output increased greatly. The audio synthesizer was much more sophisticated, with 3 square wave oscillators & decay timing. The clockspeed was increased from 20Mhz to 40Mhz to allow the improved audio. The new speaker driver was a real PWM driven class D with the 0 crossing at 50% duty cycle. The speaker brought the total current with all the LEDs & sound to 0.6A.

The flaky buttons & switches which plagued Heroineclock I were replaced with an IR remote. It was a 1st experience decoding an IR remote & it revealed IR receivers don't output RS232 but PWM. The decoding had to be really schmick for it to handle the rapid button presses involved in setting a clock.

The shift registers which drove Heroineclock I's LEDs were replaced by direct wiring to 1 GPIO per segment....Continue Reading
Posted by Jack Crossfire | Apr 03, 2017 @ 01:20 PM | 7,804 Views
The answer is yes, you can change the voltage of a laptop power supply, within limits. It was long dreamed of changing a 12 year old, 20.5V, 6A deal into a more useful voltage. 1st was discovering that it could be opened. 2nd was discovering there was a common circuit in most power supplies. It could be probed for by providing DC to the output instead of connecting the manes voltage. With the output powered, look for 2 inputs of an op-amp. 1 input should be around 2.5V, the other input should be a ratio of the output voltage.

The ratio is determined by a resistor divider. In this case, it was an irrational number around 15 for the high side & a complicated network for the low side. Decreasing the high side would raise the op-amp input & lower the output voltage.

Whacked in a parallel 47k which reduced the output to 16V, but what the internet didn't say was on a scope, it was a sawtooth. With load, it was 8-16V. Without load, it was 14-16V. Whacked in a 100k & 1 meg in parallel to finally yield a voltage of 17.9, right on the edge of where it became unstable but .1V low enough to power the Accucel 8150. Another 1 meg in parallel threw it off.

The problem is it needs a minimum voltage to power itself. It tries to bootstrap itself when it turns on, but if the minimum voltage isn't reached, it indefinitely tries to bootstrap itself. That's what generates the sawtooth waveform.

The 2.5V reference is generated by a Xena diode labeled 1273. It's...Continue Reading
Posted by Jack Crossfire | Mar 30, 2017 @ 07:36 PM | 4,346 Views
They pushed the reentry velocity a bit higher on this one. The grid fins were molten hot right when the video died. It looked like it burned up, but it reappeared after a successful landing. The 2 lower grid fins got a lot hotter than the upper grid fins, showing an attempt to use the fins to pitch the rocket. You'd think it would be more important to slow it down by spreading the heat between all 4 grid fins than change the angle of attack by sacrificing 2. The rear end probably got hotter.
Posted by Jack Crossfire | Mar 21, 2017 @ 12:16 AM | 5,205 Views
So today, the Alphabet corporation finally closed down the party by disabling the last of the downloadable API's on youtube. 4kvideodownloader & movgrab no longer work for any videos. No more 4k video unless your connection & browser are fast enough to play it in realtime. No dynamic range compression besides what a piece of hardware connected to your headphone jack can do. There's a slight chance someone may find another back door. There should be another back door as long as playback is supported on a browser, but the current generation doesn't have any interest in making things work.

With cleverly hacked software, 4k was glorious on 7 year old hardware that couldn't play it otherwise & an internet connection only fast enough for 640x480.

Still remember 8 years ago, when it was as simple as renaming files in the browser cache & the entire video was a single file. They played cat & mouse games for 8 years, constantly obfuscating the API in new ways. It's not whether something can be done anymore, but whether anyone has the interest or the intelligence.
Posted by Jack Crossfire | Mar 15, 2017 @ 12:52 AM | 5,209 Views
A vision of the pad if they tore down the shuttle service tower. The transporter erector starts to echo futuristic architectural elements of the Mars rocket service tower. The shuttle service tower will never be torn down because it has a lightning pole & locations to put cameras. NASA had to tear the 39B tower down because they envisioned a much bigger rocket.
Posted by Jack Crossfire | Mar 13, 2017 @ 01:55 PM | 4,850 Views
Finally conceeded that the hardware in a cheap ESC isn't fast enough to make a continuous sine wave stepper controller. That's why we do minimal tests. But it can step in 120deg increments through the sine wave by only modulating N's. The reduced precision was probably good enough, but the FETs still got too hot when the motor was stalled.

The scope showed all the voltage drop in the FETs, so they were putting out 12W. The motor has to provide a minimal resistance greater than the FET resistance. Then ohm's law transfers all the voltage drop to the windings & the ESC can reach its ratings.

A custom, super motor driver is required, with active cooling & FETs placed farther apart with heat dissipating pads. The supply of Hobbyking 35A's will probably never be used, so the parts could be recycled. It would still be expensive. The ATMega pinout could be rearranged for hardware UART & encoder feedback. 2 Hobbykings would be combined on a single board, with all the FETs on the center of the board under a fan.