#apertus IRC Channel Logs

wb Bertl

but it's painfully slow and utterly stupid in its inner workings

I tried with a test image

comparison

hey

I have not read the pdf yet though.

Bertl: no idea.

I can write what you need.

actually that sound great! :)

He forwarded an email about PLL and I have no idea what it is. Apparently he could not tell me either.

I know about software, unfortunately not much about hardware.

do you have an idea what a PLL is?

I mean, I have from the name.

okay, let me give you a wikipedia supported crash course then

http://en.wikipedia.org/wiki/Phase-locked_loop

this is the basic building block for a PLL

you have a frequency at the input, a variable oscillator running inside producing the output and a feedback loop which adjusts the oscillator to 'match' the input somehow

And what is this output for?

The pdf Seb sent me was: http://www.xilinx.com/support/documentation/application_notes/xapp888_7Series_DynamicRecon.pdf

the output is used to control various building blocks in an FPGA

Ok. Looks complicated. :-)

114 pages.

Cool.

but it will help to read the MMCM/PLL sections

Basically though if I understand:

You want to have an easy script to set the input from a terminal, using this API.

not exactly

it is a little more complicated, but we'll get there soon

I was hoping for a shortcut. :p

and generate a 'desired' output (50-250MHz) in a stable manner

which allows us to 'reprogramm' the PLL from a bash script

I see.

think of it like reprogramming a device by setting specific registers

And you would like an higher level script, then?

I see.

I see that devmem seems to be a busybox command.

the last function 'pll_reg' will allow to read and write the DRP registers

basically, but if you want to 'test' just replace devmem by a printf or echo

alternatively you can use the busybox here:

http://vserver.13thfloor.at/Stuff/AXIOM/FAKE/

i.e. it accesses /tmp/mem instead of /dev/mem

hi Jehan_ZeMarmot

that is busybox with a modified devmem

But like what, virtual machines?

so 32/64 bit intel and arm

usually yes, they are elf binaries :)

Ok. I'll try to run the x86-64 then.

you need to make a symlink called devmem which points to this binary or you change the devmem to something like 'busybox.static.x86_64 devmem'

I see. Or simply an alias. :-)

FergusL: hi. Yes that appears to be what is on my passport. :-)

simple question from what I recalled :)

Bertl: what does "cmv" mean?

that is the first part of CMV12000 or CMV12K which is the sensor we use

(that's why I asked what script you would like to work on :)

I mean like the spec of the end high level script.

(maybe)

Script or C, or whatever, I don't really care.

Bertl: can you give me a command line example of this?

pll_out 1 150000000

What is 1?

(it has 6 outputs which are basically identical just different registers)

But isn't it basically what are already doing the bash script I see.

And write it to the computed address.

http://www.xilinx.com/support/documentation/application_notes/xapp878.pdf

(the latter one is the relevant one)

and the PLL needs to be configured in such a way that the VCO is within the given bounds

so basically between 800 and 1600 MHz

Though I'm just not sure what to read first. You sent me like 4 pdf.

the other stuff is for illustration

from the XC7Z020 data sheet you only need the one page (46)

you'll probably need code to do bitmanipulation, you might find something usefull here: http://vserver.13thfloor.at/Stuff/AXIOM/SCRIPTS/bit.func

the PLL config? yes

can't find a way to export properly from nthe goddamn Blender

Bertl: not really.

but the damn thing bakes or marks something

I'm not adding any metadata in my files with oiio

in Blender I have to disable CM to get the right picture

Or only PLL?

01:57 < Bertl> the PLL config? yes

basically MMCM and PLL are very similar except for the fractional output on MMCM

Ok.

no problem, just ask if something is unclear

in our case, the input is fixed at 100MHz (for the PLL)

and divide the result by M

the the PLL itself limits M to 1 - 128

N = 8 -> 800Mhz, M = 16 -> 50MHz out

The input is always the same, but we want to change the output, is that it?

so the script has to do two things: a) figure out the optimal N and M and b) configure the registers appropriately

Like a smaller N is better, or something.

so there can be only one N but many M (M0-M5)

in the case where we have different Ns the lowest N is probably the best

Ok. But I have skimmed through xapp888 and there is no mention of N and M.

I said there are 6 outputs not registers

So I am lost.

to be precise, there are at least 6 outputs, in the MMCM case there are 7 and there is also the feedback output

frequency multiplication (i.e. making that 'N' work) in hardware requires a feedback which 'divides' the output and feeds it back to the input

which controls the duty cycle _and_ the divider

never said it would be simple :)

divider setting. This divider is set with the CLKFBOUT_MULT attribute when instantiating the

if you feel uncomfortable with the PLL complexity, maybe consider interpreting the CMV registers instead

So that's also contradictory. You just told me N was the high time.

hmmm...

and similar for M

Wow. Maybe with examples, I'd understand better.

for example, to get a divide by 2, you set both high and low to 1

that will give 50% duty cycle and a total of 2 periods

___/``````\______/``````\___

if you set high to 2 and low to 1 then you get:

which has 66% duty cycle and divides by 3

what was the last line you got?

02:50 < Bertl> but there is also a trick to get 50% duty cycle with a division of 3

which basically changes when the low/high transision happens

similar applies to any off-by-one setting, e.g. 4+5

so the edge bit can be considered the lowest bit)

Like 1 "parameter" (high/low time for instance). They are registers?

Or is it 64? Or something else?

in this specific case, the DRP registers are 16bit

(the upper 16bits are ignored)

I see that the high/low times are flags in the DivReg register.

So DivReg is one of the 6 registers you told me about, right?

Ok.

But I think I'm starting to get it.

< Bertl> pll_out 1 150000000

I would basically update the 6 registers for this output.

well, in this example only two of them the feedback and the 1st output

pll_out 150000000 300000000 250000000

So I would probably just update DivReg?

you see the diagram on page 4 of the xapp888 ?

you basically have to figure out O0-O6 and M for a given set of 'desired' frequencies

which can be simply programmed into the registers

Now the question is: how do I figure out O0 to 6 and M for a frequency?

so if your M is set to divide by 8, the VCO frequency will be adjusted to 8 times the input frequency, in our case 800MHz

note that it would be good to output the selected frequencies (as result of M and Ox) for verification

Already 3PM. My stomach is too hungry. :-D

FergusL: email me with screenshots

hi Bertl

no

night everyone!

I'll have a look at this whole discussion and the pdf again and try to understand this all.

Hello

na

^^

morning everyone!

check the elphel development blog as to see what they do in response to our open modules concept

So multiple 393 boards can be linked together

With their model, I imagine you'd have to tell elphel which components you want to connect to the main 393 board when ordering, and they then build the camera according to your specifications? Those pictures don't make it look like you could just 'connect' the pieces together yourself.

i.c. Funny how Andrey keeps telling us that our design / plans won't work ;)

He has told us numerous times that we won't be successful, that our designs for a camera are unfeasible

Ask Sebastian about this in person perhaps...

I'd have to dig through emails sent to the apertus mailing list

This is a gift straight from Liam and Jack to us. Appreciate it and learn from it.

oops, sorry about that. copied wrong text from my clipboard

Andrey has posted a comment at the very bottom of the page

my bad, i feel so silly now :)

true, I can't disagree

- maybe because they don't want anybody else to use the interface

- maybe because it is better to do one big do-it-all thing and sell that

i agree.

Yeah, indeed. I still recommend you ask Sebastian more about this business with Andrey (on a private channel perhaps. He's much more informed about this than I am)

yeah

he's online now?

right, i.c

okay, it might make sense to motivate a few folks to watch/monitor certain channels/sources and do a compilation on a regular basis

yep, absolutely. I'll do my best to get others to follow suit

Bertl: Because that implies stuffing a matrix operation into the board (with the Bradford transform) so that folks can dial the WB.

and no, I don't think the sensor is in any way color aware

troy_s: so a general matrix would suffice for any white balance problem?

hey Bertl

well, no, but I'm working on the prototype, so consider me the technical prototype lead :)

Bertl: I wear a lot of hats in the camera department, 1st AC, DIT and sometimes DP.

rexbron: I'm not familiar with the camera department acronyms

Bertl: 1st Assistant Camera, Digitial Imaging Technician and Director of Photography. here is a link to my reel http://aehunter.net/video/Andrew_Hunter_Demo_Reel_Aug_22nd.mp4

works in chrome

mars_: yeah I guess mplayer will be able to handle it ...

sounds good!

and on set, reliability is paramount

Bertl: DP, AC and DITs get hired on reputation not with show reels

I see, so more to give an idea of the style?

Thats generally the idea :)

The footage is a mix of Red One MX, Blackmagic Cinema Camera and a little 7D

Bertl: allows for starting and stopping the rec function of a camera via a remote focus device. Since LANC is a serial protocol, it needed to be a little electronic adapter that listened for a switch on one side and sent out a lanc command on the other

I see LANC is a proprietary? interface created by Sony for their eqippment

Canon and Blackmagic both use LANC but call it something else

Bertl: refering to baud?

and that failed in what way?

Bertl: I think the adapter would get out of sync with the camera

okay, probably a minor issue then

Bertl: for you ;) for me, I just kinda shrugged and had to walk away from the project. I can't sell something I don't understand

did you open source the project?

I guess somebody would identify the issue pretty soon and fix it

especially if you explain what goes wrong where

I guess so, especially if it is special hardware

not super special, the initial dev board was an arduino :)

(also a first for me)

so if the code works for an unmodified arduino, I guess folks interested will have the hardware available

happens

ergonomics ;)

okay, I'm leaving shortly, but I should be back in 90 minutes

good evening

https://www.apertus.org/apertus-now-accepts-bitcoins

I dont want to announce this before I have seen it actually working at least once :)

how do you convert those donation to money? Do we even do that ?

ok, just saw that you can convert to local money. Weird stuff :-)

with quite huge value fluctuations currently

not that we will gather a huge amount of donations that way

back now ...

and the value is still increasing, so it makes perfect sense to allow that

did you ever take a look at the mining process

yep

5000$ http://www.butterflylabs.com/monarch/

Bertl: to pick up where we left off, I can offer real world production shooting experience as both a DP and AC.

(there will be a second one soon to shoot actual footage)

good n8!

morning [1]Sasha!

*additional