Model 51 - bobbybudnick/SFS-Mobile-Internet-Device GitHub Wiki

Always pc stick cmos failures
every pc stick recently has a failed cmos battery
this problem makes the mid look like trash with a slow boot
the cmos batteries are impossible to maintain
only a brand new pc stick with warranty will be functional
cmos battery must last for at least a 3 hour interval
this life simulates a short trip to demonstrate the device
the pc stick must be replaced if this requirement can not be met
resulted in failure necessitating device replacement and model 50 surgery

Model 51 design principles
the new computer is a kind of hybrid between mini pc and pc stick
the oled screen requires amplifier and ground loop isolator
hub still required due to the amplifier needing a usb port
the fpv hdmi cables make a large difference in device length with this design
the overall footprint only increased around 1/4" between model 50 and 51
there is equal exposed length of computer between top and bottom of screen
enough connectors stick out of bottom to support weight of the device itself
the only good place for the cellular device was along the left side
the cellular device had to be flipped around to access the power buttton
thermal tape used between computer and battery
mounting tape with cover on to keep computer attached and aligned with screen

Model 51 power considerations
does receive standby power when battery is fully charged
however this power is 5v so the device will not power on and stays red
unplug battery to switch to full voltage
does charge at high speed 9v with micro usb and usb tester still works to monitor

Model 51 graphics situation
need to go from beowulf to chimaera to get newer kernels
need to go from kernel 5.10 to 5.18 for the intel driver to load
all scaling fractional or not is a little slow with the intel driver and xorg
mostly kickoff and the plasma panel are slowest because they are translucent
however a solid plasma global theme is no faster so the effect just seems hidden
at least scaling 1.5x1.5 is needed to enable a 4:3 800x600 game window to fit
this allows for 816x1440 virtual resolution from 960x544 native
strangely kickoff is faster to respond to the touchscreen than the mouse

Model 51 audio capabilities
intel hdmi audio is standard
however an integrated usb audio device drives the analog audio connector
the audio device has an analog microphone input on the same connector
this could in theory be used for the microphone pickup
the analog audio out is loud enough nearly to drive a headphone speaker directly

Model 51 cpu power control
intel-rapl:0 is the power control for cpu
cpu power can be set to battery saver in bios but does not change rapl wattage
rapl wattage seems locked to 10 and can not be changed with echo either
temperature seems fairly stable around 40c with battery cpu setting
constraint 0 is long term
constraint 1 is short term
may be ok for now because power is needed for gaming anyway

Kickoff and onscreen keyboard
falkon/konsole/xvkbd recommended on kickoff
xvkbd is trash with no way to turn off suggestions/poor backspace placement
xvkbd is forced into a semi usable mode solely with kwin with no arguments/config
the other keyboard is garbage because it messes with window geometry
the entire project would be out of luck if not for plasma keyboard

Model 51 power saving modes
not counting battery pd transition time of a few seconds
regular boot time 60 seconds to desktop with usb mic as only peripheral
restore from hibernation time 35 seconds with usb mic and 1 gb swap with 8gb ram
of course boot time only is not the only advantage of hibernation
perhaps the primary advantage is restoration of system state
browser tabs and the main thread and support services can already be running
thus hibernation is not too bad of an option
with a commercial usb battery hibernate is relevant due to charging power loss

Model 51 instability and usb power saving
there is glitching in the amoled screen and audio noise when in linphone echo test
just like in old days linphone echo test is an excellent stability benchmark
daisy chain hubs not always reliable and is against spec with passive hubs
the previous proposal to move the amplifier directly to battery output is best
already the ground loop isolator is in place which will not hurt anything
this method 1 proposal will reduce system usb power usage by the amplifier
this method 1 proposal will reduce system usb power usage by eliminating the hub
this method 1 proposal will increase usb stability by eliminating the hub
however no relevant errors were logged in dmesg during this glitched time
method 2 would move to analog input/output and could have a similar effect
method 2 would remove the amplifier entirely
method 2 would also remove the hub and free up onboard usb-c
method 2 would work even if usb mic needed but may have questionable volume
method 3 would remove the hub and wire 5v parallel to usb accessory 5v
method 3 should reduce usb power usage and interference significantly
method 3 definitely needs the amplifier parallel to accessory 5v only
method 1 better than method 3 due to no more wiring/less system usb power usage
method 2 would have best stability but bad volume as noted so try last
however all three methods would remove the hub
method 1 need test if type a battery output works when usb-c pd is working - fail
method 1 phase 1 would be test just moving the amplifier to direct battery power
method 1 phase 2 would be hub elimination
testing reveals that 5v does not work alongside usb-c pd on the energizer battery
since method 1 fails the test then method 3 is the next to be used
method 3 still has noise and glitching
however repositioning the analog audio cable further from cellular has reduced it
now there is no more screen glitching and minimal audio noise
note that glitched configuration part 1 is type-c to amplifier
note that glitched configuration part 2 is type-a to hub to accessory and cellular
note that glitched configuration part 3 is type-a to screen

Linphone outgoing audio failure
outgoing audio fails about a minute and a half into a call
happens with ice and correct asoundrc and tcp and udp and tls known not to work
turn on ice/keep turn off/enter stun.linphone.org for stun server and it works
outgoing audio still worked 3 minutes into a call
this testing was done on cellular and may be unnecessary on cable

Latest on linphone ui
with the greater scaled resolution more is visible but not all
it is the best compromise to fix the window along the right edge
this causes the left edge to slightly hang off but all elements can be clicked
unfortunately this does cover up network status on left

Increase boot speed for hibernation
/etc/initramfs-tools/initramfs.conf set "MODULES=dep" reduce initrd 54M to 11M
update-initramfs -u to commit and controls loading initial ramdisk
/etc/default/grub and reduce menu timeout from 5 to 2
update-grub to commit and controls grub os selection countdown

Model 51 bill of materials
mele pcg02 pro
waveshare 5" oled
energizer 18w usb-c battery
1 foot right angle usb-c to right angle usb-c
standard ground loop isolator with housing removed
fpv twin 90 degree angle hdmi
standard high quality type a to type a extension for cellular
standard 8 inch right angle type a to right angle micro usb for screen
standard 1 foot right angle to right angle headphone cable
standard micro amplifier with speaker attached
standard zte mf861 cellular device
short right angle type a extension with spliced power for amplifier and c adapter

Further power savings proposal
a lock and unlock gesture could be used to xset dpms force standby
this should work if screen touch still functions after standby
the built in kde power management could due this automatically also
in general device should be hibernated for transport
in general device should have enough battery to sit for break periods screen on
another option would be redesign to expose the screen power button