flood's input lag measurements

[spacediver]

sounds almost like an oscilloscope that measures light instead of voltage!

[flood]

except it isnt

an oscilloscope would easily allow me to see ns.

anyway for a baseline measurement with led controlled by arduino shining directly at a photodiode, i get something around 5-10us

i think it's limited by the op amp's slew rate...

oh yeah anyone can replicate what i'm about to do here for less than $50... all that's really needed is a crt

[spacediver]

so how does it work? Arduino registers the button press that turns on LED, and the diode measures a signal and reports back to arduino and the time difference is measurement lag?

[flood]

so how does it work? Arduino registers the button press that turns on LED, and the diode measures a signal and reports back to arduino and the time difference is measurement lag?

well its not completely done but yea like that except the arduino does the "button pressing" itself...

i should make a switch to turn it off though otherwise the mouse would fly around too much while the arduino is running

[spacediver]

i should make a switch to turn it off though otherwise the mouse would fly around too much while the arduino is running

you mean the mouse cursor? Why not just restrict "mouse events" to button presses, so you don't have to worry about the cursor moving around?

[flood]

hard to make whole screen bright with button presses

also

there's this which, although i can't follow his logic really, seems to claim that raw input performs suboptimally when theres more than 1 polling event per frame

http://www.overclock.net/t/1405271/rega ... timization

so the plan is to have the mouse shake up and down by 1 pixel continuously (i.e. every 1ms) , then during the event, move it 100 pixels left or right, then measure how long it takes.

[spacediver]

wait, you're using a game to do the measurements?

why not just do a single white frame sandwiched by black frames? Surely that's a much more controlled experiment, and you can set it up so that the white frame is rendered as soon as the button is "pressed". You could also do your "cross" polling experiment the same way, having the white frame render upon the first horizontal displacement.

[flood]

don't worry i'm going to do that and more

[spacediver]

don't worry i'm going to do that and more

cool

I might follow in your path once I've done my little line spread function project with the DSLR.

I took some measurements today in the lab for stimulus timing precision (my current research requires precisely timed stimuli, where 8-16 ms makes a difference). The measurements (using Matlab + psychtoolbox) involved looking at the timestamps of flip commands, and I was able to get the standard deviation of 200 stimulus presentations down to under half a millisecond. It would be interesting to verify this by actually measuring the light from the display directly. I'm using the VIEWPixx (120 hz, scanning backlight).

[flood]

FIRST DATA:
cs1.6 with default 100fps cap:

time (us): 8804 cycles: 13938
time (us): 11776 cycles: 18645
time (us): 10300 cycles: 16298
time (us): 6640 cycles: 10512
time (us): 9852 cycles: 15591
time (us): 10200 cycles: 16150
time (us): 6892 cycles: 10903
time (us): 5508 cycles: 8716
time (us): 11068 cycles: 17525
time (us): 6688 cycles: 10579
time (us): 10828 cycles: 17131
time (us): 9064 cycles: 14347
time (us): 4716 cycles: 7468
time (us): 7032 cycles: 11129
time (us): 8064 cycles: 12759
time (us): 11616 cycles: 18392
time (us): 6672 cycles: 10555
time (us): 2540 cycles: 4008
time (us): 9600 cycles: 15199
time (us): 9688 cycles: 15345
time (us): 7300 cycles: 11556
time (us): 7008 cycles: 11093
time (us): 10016 cycles: 15849
time (us): 10620 cycles: 16812
time (us): 12504 cycles: 19790
time (us): 9208 cycles: 14567
time (us): 6868 cycles: 10869
time (us): 10096 cycles: 15984
time (us): 11352 cycles: 17966
time (us): 4044 cycles: 6396
time (us): 4084 cycles: 6461
time (us): 11396 cycles: 18039
time (us): 2172 cycles: 3427
time (us): 7740 cycles: 12248
time (us): 11684 cycles: 18492
time (us): 10228 cycles: 16192
time (us): 5300 cycles: 8390
time (us): 9296 cycles: 14707
time (us): 9572 cycles: 15145
time (us): 9532 cycles: 15093
time (us): 8420 cycles: 13333
time (us): 12252 cycles: 19396