Jack Crossfire's blog View Details
Posted by Jack Crossfire | Jan 16, 2017 @ 10:46 PM | 2,102 Views
The 3rd LG Tribute disintegrated during a romp in the canyon. It lasted only a few weeks. Single women may have high medical bills, but their lack of exercise probably saves a lot on phones. Thus ended 3 years of Lucky Goldstar being the bottom end. They once adorned the Target shelves at $40. The Goldstars are now high end. The new low price leader is the $50 Kyocera at Target. For $10 less, there are Alcatels online only.

The low end reduced all their processors from 1.2Ghz to 1.1Ghz. They reduced all the cameras from 1920x1080 to 1280x720. It's 1 of those phases of economic stimulus where lower performance is rising in price to keep everyone employed. Mercifully, the Kyocera doesn't turn on when tapped. That bug definitely pushed the limits of millenial debugging ability.

Sadly, there's no way to get the contacts off the Tributes. Once their screens cracked, all the data was inaccessible.

Tehan falls (1 min 54 sec)

What a cracked LG gets you.
Posted by Jack Crossfire | Jan 15, 2017 @ 10:16 PM | 2,274 Views
Self Driving BASE 1 Rover - MountainTrack (5 min 1 sec)

This nugget caught the lion kingdom's attention last month. It seems to be doing the impossible: navigating a mountain dirt road with lots of shadows using only machine vision. For the 1st time in 11 years of UAV blog posts, there was no mention of GPS even being offered in the product. Of course, it didn't stay to the right like a running pacer has to do.

Knowing a convolutional neural network is just a storage & recall mechanism for images, it could have been done by training the neural network on many manual drives through the same 5 minute path. There's enough detail besides the path for the rover to ignore the shadows, determine where it is & steer to prerecorded headings. How the rover determined where it was in relation to the center of the path would be another trick. It could be a neural network trained on a bunch of offset camera views or it could be simple optical flow.

The key is the vertical detail from the trees & mountainside adding extra edges besides the sides of the path. It wouldn't do well on a bare path shaded by trees outside the camera view.

In other news, keeping with the modern slogan that the programming language is the computer, Tesla hired the guy who invented Swift & invented the LLVM as the VP of autopilot. 20 years ago, they would have hired a networking expert & 10 years ago, they would have hired a video expert. It would make sense for Tesla to invent its own programming language, since that's what everyone else is doing. Perhaps the world needs a programming language for neural networks instead of a library.

The LLVM has a good chance of replacing GCC. GCC has enjoyed a 3 decade reign, but ever since running it for the 1st time on an HP, it was slower than any commercial compiler.
Posted by Jack Crossfire | Jan 12, 2017 @ 06:41 PM | 2,513 Views
There was a lot of hype about hybrid apps in 2014. Based on the internet, the verdict is there's absolutely no advantage to either method. The lion kingdom's experience:

GUI & speed requirements ended up beyond what the standard hybrid libraries could do without major effort. The browser incompatibilities & replicated effort when trying to optimize the interface for each phone were just as bad as native.

4 months of work on a webview implementation yielded nothing, with so much expertise being required in web development to achieve minimal results that the IOS & Android developers ended up going back to other tasks on their own platforms until another developer could be hired for the hybrid part.

It ended up requiring more people than if the hybrid part was done natively.
Posted by Jack Crossfire | Jan 04, 2017 @ 10:05 PM | 2,190 Views
They're a dime a dozen, but still not sold commercially. Would feel sorry for anyone dumb enough to have to pay for one. The mane desire was to reuse foot pedals that were already around, but the Yamahas have an open circuit when they're active & closed circuit when they're inactive. The only solution was BJT's to invert the switches.

Making it work without batteries required 247k pullup resistors, so the pedals pulled down the bases when they were inactive & the camera bias voltage pulled up the bases when they were active. When the bases were pulled down, the resistors couldn't draw too much current to the point of triggering the shutter so they had to be large. It's an interesting use of a voltage to turn itself off.
Posted by Jack Crossfire | Jan 04, 2017 @ 12:03 AM | 2,467 Views
After many miles, the long awaited conversion to LF353 happened. It's ever so slightly subjectively clearer, but the speaker is too overloaded to hear any difference.
Posted by Jack Crossfire | Jan 02, 2017 @ 05:06 PM | 2,095 Views
As expected from the anomaly page


they had to abandon supercooling at least part of the propellants. The helium will be warmed up to prevent pockets of frozen oxygen from forming. More of the engine exhaust will be bled off to pressurize the tank instead of using helium. They confirmed voids in the COPV laminations pooled liquid oxygen. When the liquid oxygen froze, it expanded & burst the laminations. More specifically, it was a void between the aluminum inner liner & carbon fiber overwrap. To save money, the overwrap was launched with voids.

You normally think of a tank being a barrier between stuff on the inside & stuff on the outside, but the overwrap on a COPV is actually saturated with stuff from the outside. The total propellant load thus consists of everything the engines need + a little bit for the COPV's to soak up. Presumably, the oxygen impregnating the tank can solidify as long as it doesn't form large pools.

Along with more expensive struts, they'll need super high tolerances for the overwrap, so explosion by explosion, it becomes less off the shelf & more precision specked like a shuttle. NASA spent years troubleshooting voids between solid propellants & booster walls. They avoided the explosions, but it took immense manufacturing standards.

They mentioned other changes besides warming the helium, but not what they were. They revealed the other changes involved spending a longer time loading propellants. They implied the helium loading procedure was changed during the fateful day from what it was in previous launches.
Posted by Jack Crossfire | Jan 01, 2017 @ 02:21 PM | 2,398 Views
After 6 years on the wagon train taking heroic efforts to defeat noise, the preamp was redesigned again. The discrete transistor amplifiers are now an LF353. The discrete transistor was nifty, but impractical for fully balancing 2 channels. The LF353 subtracts the differential inputs & amplifies the result with minimal components. This eliminated the noise completely & finally allows the Zoom to record 4 channels. It also seems capable of sharing the power supply with the Zoom without any noise, but it's not soldered for it.

The mane problem is the pots being too close together. The mane cost in going to 2 balanced channels is 2 stereo pots & 8 wires to connect them. Previous work with digital pots ran into a lot of noise from the SPI & a display to show the current level. The balanced inputs may defeat the noise, but the analog controls have proven simpler.

The LF353 circuit had to deviate from the goog slightly. The 1k's had to be fixed. Only the 100k's could vary. It needed 1k's on the outputs to isolate it from Zoom noise. Spudger diodes prevent the 48V pop from blowing it up. The virtual ground is 5V. It needs 12V to be happy.

All support for 2V microphones & dynamic microphones is gone. It's now just a phantom powered 48V preamp.

The moral of the story is any microphone is going to need a balanced amplifier in its 1st stage, right down to your cell phone mems microphone. The balanced signal needs to be retained all the way until the line level or power supply noise will always be a problem.
Posted by Jack Crossfire | Dec 29, 2016 @ 06:24 PM | 2,127 Views
So the Zoom arrived in April, 2009. After 7 years without ever using the microphones, it was time to do the deed. All you need is to replace the microphones with RCA connectors. The gain switch set to low is the same as line level. It worked perfectly when taking input from a computer sound card, but had serious noise when recording the microphone preamp. Turning off the gain on the preamp made the noise go away. The line input didn't have the noise. The Zoom had fixed 2k resistors supplying phantom power to the RCA connectors, but pulling out a resistor didn't matter. The Zoom & preamp had independent power supplies on different grounds. The source of the noise remanes a mystery.

More importantly, it was impossible to monitor audio from a computer & microphone without some way of having independent monitoring volume for the 4 channels. Otherwise, the computer would have to be recorded at a very low level. The 4 channel experiment was dead.

The computer outputs TOSLINK, so it should be recorded by sniffing a digital signal. Another STM32 running the CP33 recording firmware would have to be hooked up to the point in the amplifier when the audio was in I2S format.
Posted by Jack Crossfire | Dec 26, 2016 @ 10:53 PM | 2,251 Views

Amusing Gootube channel about the Commodore 64. He does a better job covering it than others, in better quality & in more detail than average. Things would have been different if the younger lion kingdom had access to highest end supercomputers of the day, but there was no option of buying time on a Jeff Bezos compute node or knowing someone with a university computing account in those days. The only exposure to how multimedia programming was done was what consumers could afford, which was far inferior to the way the highest end systems worked.

You could be forgiven for thinking the way C64's did multimedia was the way it would always be done. The C64 was lightyears ahead in multimedia than anything else of the same price, for at least 6 years. It was invented by guys a lot smarter than you, surely pursuing the easiest architecture possible for programmers. In fact, it was the opposite of convenience but a need to optimize every single transistor down to the maximum a consumer could afford. It took a while to realize high end arcade games weren't using sprites & character sets in creative ways but using entirely new hardware. The 6502 assembly language remaned quite relevant to modern assembly languages.

It makes you wonder if at the rate human intelligence is going down, are future systems only going to have 64k again & are we going to have to use "multicolor bitmap mode" again. Future generations aren't always going to know the merits of addressing the screen in raster lines instead of character cells.

What about running into future limitations which might require multicolor bitmap mode. The C64's graphics were dictated by the minimum resolution to resolve text, making colors the constrained resource. Today's resolution is constrained by the minimum framerate to resolve motion & would have to be reduced by quite a bit before it couldn't resolve text.
Posted by Jack Crossfire | Dec 25, 2016 @ 01:26 AM | 3,081 Views
The Yamaha RH5M's are still the best of the best, despite Beats's $3 billion valuation & 30 years of advancements. They were discontinued long ago, yet there seem to be a few lying around for a price. The ear pads showed the 1st signs of crumbling after only 5 years. After 30 years, they were quite disintegrated. There wasn't any free stuff for XMas, but new ear pads arrived. The Yamaha's take Sennheiser HD25 ear pads.

They've seen a few resolderings & the oxygen free cable has gotten shorter. The limiting factor is the plastic becoming brittle. Being immersed in an oily environment like the mane & the face has kept them going longer than other plastic.
Posted by Jack Crossfire | Dec 22, 2016 @ 09:55 PM | 2,462 Views
Made a USB OTG cable using the instructions on


& plugged it in. It wasn't worth documenting in higher quality, since those who want to wait 2 weeks can order one for a buck.

The cheap LG-Tribute didn't support USB OTG at all & neither the LG-Tribute 2, or LG-Tribute HD have any support. All the Samsungs support it.

The mane reason is the next wishlist is automated music notation. It's a very old problem involving paltry amounts of data by today's standards. Helas, 30 years after its introduction, speedy entry on Finale is still the duck's guts & it's not on phones. There are still no other equivalent programs. All the phone apps do only manual notation. The phone would have been the ideal platform because it fits on the instrument.
Posted by Jack Crossfire | Dec 21, 2016 @ 10:02 PM | 2,210 Views
had been found. It only took 5 years. None of the commercial offerings were appealing.

It takes a minimum of desk space in front of the reading material.
A simple clamp makes up for the lack of a base.
The base has a hole for tripod mounting.
Different dowels can be attached to the base for stereo & portable applications.
A 10-24 bolt attaches it to a dowel.
A 10-24 wing nut allows adjustment.
To adjust the height, drill another hole or tilt it.
An air filter taped to the shock mount is an effective pop filter.
The shock mount has proven effective at removing typing noise, but noise still gets in from bumping the cable.

The nearest alternative was this $12 deal:

It would need more space & still not have a pop filter.
Posted by Jack Crossfire | Dec 20, 2016 @ 08:50 PM | 2,106 Views
Continuing to see deterioration in range. It went 10mph for 1.3 miles followed by 6.6mph for 4.5 miles. That took 2.9Ah or 0.5Ah per mile. The 8.5 mile dead battery was 3.8Ah or 0.44Ah per mile. The time spent at 10mph seems to be killing it.
Posted by Jack Crossfire | Dec 19, 2016 @ 07:20 PM | 2,369 Views
11.5 miles using battery 1 + battery 2, no payload: 2.3Ah each battery
7.5 miles using battery 1, no payload: 2.6Ah
6.5 miles using battery 2, 1qt egg nog for 3.25 miles: 2.8Ah
Posted by Jack Crossfire | Dec 14, 2016 @ 09:44 PM | 2,312 Views
Battery 1 died after only 8.5 miles & took 3.8Ah to recharge. So what was once a 5Ah battery is now a 3.8Ah. Some twists with this drive were 1 mile at 10mph, 1 mile at 8.25mph, many ascents & accelerations. The last .5 miles were hauling 18oz of cargo. There is a record of this battery going 12 miles on 3.8Ah, on flat ground, hauling a spare battery at 6.25mph.
Posted by Jack Crossfire | Dec 14, 2016 @ 01:16 PM | 2,018 Views

Basically, they're integrating the emitter & receiver on a single 4mm x 4mm $50 chip. The rotating platform, slip ring, optics, are still required, so at least $3200 for a 64 channel unit. The sensor chip is within the pricing of existing laser rangefinders.
Posted by Jack Crossfire | Dec 10, 2016 @ 10:40 PM | 2,075 Views
So the mane computer, the one which does 4k video, had always dropped connections with the Zoom 5345. The easiest way to reproduce the failure was trying to access, which always failed. After months of sporadic experiments, finally tried different kernel versions. It was the only computer still using 2.6, which it had used since 2010. On anything above 3.2.9, the network was perfect. The Zoom 5345 had a problem with Linux 2.6.

Tried Linux 4.8.9, but the NVidia driver for this kernel wasn't compatible with the obsolete graphics card. The graphics card is now 10 years old, but still plays 4k video in a 2.5k window. Downgraded the NVidia driver to the lowest version which could work with Linux 4.8.9 & the highest version which could work with the obsolete graphics card: NVidia 304, which required commenting out the MTRR functions to work with the new kernel. The driver compiled, but wasn't compatible with the X server. Downgraded the kernel to the lowest version which would work with the Zoom, the highest version which would work with the X server, & the lowest version which would work with the NVidia driver: Linux 3.2.9. The NVidia driver had to be downgraded to 295.

It worked, but finding components which overlap enough to keep Linux going on the current system is a lot harder than it used to be. Linux 3.2.9 came out in 2012. Eventually, routers won't support it either. Very cheap cable modems can't handle subtle changes in packet...Continue Reading
Posted by Jack Crossfire | Dec 10, 2016 @ 09:10 PM | 1,956 Views
Replaced a burned out LED & pointed the headlights forward.
Posted by Jack Crossfire | Dec 09, 2016 @ 11:49 PM | 2,087 Views
Iron Man released a video of a Merlin 1 nozzle being fabricated. It made sense for it to be formed around a plug on a lathe. The intriguing part was how everything else was jerryrigged. A bank of acetylene torches heat the work piece. The torch placement doesn't have to be very precise. The pressure from the flames pushes the metal down onto the plug. The torches have to be manually repositioned as the nozzle bends down & the torch fuel tanks have to be refilled. It seems to take many hours of painstaking manual labor, but launching 12 times per year only takes 120 engines, leaving 3 days to make each engine nozzle. It's far faster than braze welding thousands of tubes.

Judging from the thickness, the work piece already contains the cooling channels. It would have originally been a disk. 2 metal disks with radiating patterns of grooves would have been fused by a very large hydraulic press & heat. Then the bending of the fused disk around the nozzle plug would have retained the cooling channels. The method was probably 1st devised by the Russians.
Posted by Jack Crossfire | Dec 08, 2016 @ 12:43 AM | 1,649 Views
It's rarely covered anymore, but there were some nuggets about launches resuming on Dec 16.


Then a few days later, it was delayed until 2017.

Although they will never release complete details on the last explosion, they did reveal supercooled oxygen froze solid & caused the helium tank to burst. That could mean a lot of things. It could be oxygen got into the carbon fiber laminations before freezing & expanding, ripping apart the laminations.

The real bummer for space fans was delivered in Dec 2015, months before the troubles with supercooled oxygen emerged. It was Tom Stafford's letter complaining about the need to fuel the rockets with astronauts aboard & the lack of recirculation pumps to prevent temperature fluctuations at the engine inlets.


If they don't take off immediately after fueling, the oxygen will heat up, but the last explosion confirmed fueling is too dangerous to do with astronauts on board. Tom was prescient, indeed.

The only thing that's going to happen is the simplest solution: for flying humans, they're going to drop supercooled oxygen & board the rocket after fueling is completed. They'll have to give up reusability for human missions. They never released any performance figures for supercooled oxygen, but it was probably more than they could gain by shedding weight. There's a chance the falcon heavy...Continue Reading