#apertus IRC Channel Logs

off to bed now ... have a good one everyone!

First I'd like to learn a little more about your project - technical details.

welcome here

Are you working on the camera?

A note about something you said on your web-page: My design is 8-layer PCBs, and they aren't much more expensive than 2 or 4 layer PCBs, but much better performance.

Are there any "real" prototypes yet?

there are ! but only for engineers, I'm just helping wherever I can

So, maybe they don't need electronics engineer types any more?

I think we do :) you'd have to talk about this with Bertl and first with se6astian

The camera I designed was less demanding (but lots cheaper to build)... only 2595x1944 resolution.

I'm more interested in what else is inside the camera.

indeed, that and the informations the sensor manufacturer sent us are our only sources

Okay, great

yes indeed for now it's working off an FPGA board from Xilinx

Getting the data into a PC in real time is extremely challenging!

So you'll have a custom PC card that plugs into a PCIe slot in the PC ???

or rather, I should say Axiom

I was about to say that not all camera owners want or can have a computer sitting near the camera

I just found your site so I can't keep Apertus and Axiom straight in my mind.

Okay Apertus is both the company and association running the operations and gathering the followers, volunteers and, hopefully, the manufactering (in very general terms) . Axiom is the name of the camera

I hope too, and what kind of fast video input ? are you thinking about digital like SDI or consumer HDMI or Displayport ?

What kind I don't care... I can design hardware! :-)

I mean, it's a crucial point to me, it makes it a very different product

for now, all computers can have (unless one can design a pci card :-) are a very selective range of often ecpesnive products that seem to have thought to fit in very specific niches

Yes. I assume you want to make the price of this camera reasonable (for a 4K rez device).

hm, that info is on the website, that's for sure

http://axiom.apertus.org/

Okay, so $4K would make you smile. :-)

Some places on the page say "this is a concept", and other places say "we've shot footage". What's the actual status?

I don't think an initiative of our magnitude can achieve $4k

But that wasn't the manufacturer, so it could be wishful thinking.

there is one electronics engineer actively working, he's Bertl, present here but away it seems

something to get images out

https://www.apertus.org/alpha_prototype few pictures there

that design was considered for a long while : very high speed connection to a computer but was considered a flaw by a lot of potential users

For your sensor, flowing straight through would be quite a challenge!

and now yes it is supposed to hold images and video inside

Yes. I developed a lossless image compression scheme that helped me make mine work.

It is much easier to store the data in standard flash memory chips, which can then be removed and inserted into a PC.

In the end, I decided the best way is to create two separate sets of electronics, one for each application.

Would you say the large majority of people interested want this for cinema?

absolutely yes

Though you probably won't hear from that many people, the robotics people would go crazy for a cheap 4K camera with real time image flow through.

what are the exact usecase ?

The most common one is in factories, where the actions robots are taking must be corrected every 1/60 second (or 1/30 second) based upon what the robot sees.

hahaha

My scheme sent the horizontal rows of pixels out as the CCD generated them, so the algorithms could start thinking about what to tell the robotics motors before the whole frame even got sent out.

You just cannot wait for the image in robotics!

I hope you can stay here for a while, Bertl and se6astian will be able to answer your questions with more details

I can stay for a while.

Europe

13:53 here

Okay... so I'd be the "odd man out" as usual. hahaha.

exactly

Can't interface the sensor to a standard Xilinx prototype PCB.

https://www.apertus.org/sites/default/files/alpha-prototype-2d.jpg the purple board behind the lens is what we've designed

so i guess i'm seeing a small 60mm square PCB (yours) above a xilinx demonstration PCB ???

yes

ah, so about 88% for cinema-like apps, and maybe 12% for robotics.

but actually, you'll get a huge number of orders for factory automation and other robotics if you have a suitable model.

they don't talk much, they don't spend much time on the internet or anything, but they spend money like crazy.

are you thinking about this as buffer memory ?

FergusL: industrial vision is insane... i was asked if I can supply my camera for filming train crashes :) that is a demanding app

I do realise that

i think i saw that number somewhere (maybe lower rez though).

yes

yes?

maybe lower depth too (8-bits versus 12-bits).

Oh, I missed your question about buffer memory. Yes, I was thinking about writing directly into [very high speed versions of] those little flash memory thingies that DSLRs and video cameras accept.

usually all storage devices fake their write speed as it is not stated with: incompressible over whole capacity

Yes, I understand that, which is why it is necessary to either select certain special brands, or downgrade expected bandwidth.

Personally, I totally HATE lossy compression.

you're def the odd man out !

just got back to the channel

let me read up what you discussed so far :)

hi se6astian!

Hence I swore off lossy compression for anything I make. Hahaha.

not vendors do :)

Huh. Is the file wrong, or it really is JPEG lossy?

Lossy compression can get enormous compression ratios... but... well... drives me nuts.

Which is better, 5 megapixels lossless, or 10 megapixels lossy? Personally, I'd rather have the 5 megapixels.

Dual 3G-SDI

or USB3

SATAIII would be the most convenient one for film makers

there is no PC

well in SFP+ case there would have to be a PC

yes

so this is for AFTER taking shots.

no, its the recording media interface

record video -> dump to SATA SSD for example

without compression, or with?

raw

okay, let me do a quickie here: 4K x 3K = 12M * 1.5 bytes per pixel * 60 fps == bandwidth?

1,132,462,080 bytes per second

but yeah, at 24 fps i need to multiply by 24/60.

so those higher rates on the webpage (full rez at 75FPS) are not what you are really shooting for.

well first priority is making sure it works at standard framerates (24,25,48, etc.) but with enough overhead planned into the available bandwidths to go higher

cmosis recently announced a new version of the sensor that doubles the frame rates once more

so 150fps @ 12-bits @ 4Kx3K ?

but 4096x3072 x 10 bit x 300 FPS = 36Gbit/s which is rather insane :)

i have a super-simple lossless compression scheme i developed for my camera, but it only buys us about 20%.

like for 1 second

then it takes 10 seconds to read out the RAM

Are you planning to have RAM ?

DDR3

I was always thinking of supporting 8 flash memory devices in parallel to achieve that bandwidth (10Gbps).

it was designed for inside the FPGA.

it is especially designed for 12-bit pixel depth.

converts most 12-bit pixels into 8-bits, and the rare "failure to compress" pixel into 16-bits.

which is not as difficult as one could think as the memory blocks from Samsung, etc. are all the same in current SSDs, usb sticks, etc

oh, right. we could buy the die and bond them.

well, that is not as impractical as you may think.

depends on how much value gets attached to supporting high frame rates.

I understand. But it is worth looking forward in order to avoid having to develop a dozen completely separate implementations.

which does not mean nobody else would shoot high speed of course

exactly

So, where would you plan to put the solid-state hard disk?

Or "that's not your problem"?

I mean physically where? I guess the customer buys this piece, huh?

about your 12 bit -> 8 bit compression, where/how do you store the metadata which pixels get 8 and which 16 bits?

and we just have a standard SATA connector and a drive bay

I see

indeed!

how you encode difference of less than 8 ?

i don't bother with that.

and in my case, that extra 20% made all the difference (cuz i was squeezing 2592x1944 12-bits through 1Ge).

no. if the most-significant bits are 0001, 0010, 0011, 0100, 0101, 0110, 0111, 1000, 1001, 1010, 1011, 1100, 1101, 1110, 1111... then those are the most-significant 4-bits of the delta value (followed by 4 more bits).

ok

however, when the delta is more than +/- 120, then the first 4 bits are 0000 (which only happens when 16-bits are sent), and the next 12-bits are the absolute value of the pixel intensity.

or maybe it was even -118 to +118... i'd have to go back and look.

have you done any research whether the 8/12 bit is the best case? :)

if you're sensor only generates 10-bits, or generates 16-bits... the whole issue needs to be re-factored.

understandable

and the 20% was sufficient to save my ass.

as i recall, to decode takes about 4 assembly-language instructions!

well, a few more, but it only executes 4 as i recall.

have you a per line buffer to choose which to send?

same thinking :)

sounds good

hahaha. i have a terrible memory.

one reason i kept a couple extra values was to toggle back and forth between uncompressed and compressed... in the middle of the scan line.

Bertl is currently working on FPN measurements, correction software

when he will be here in the next 2-3 hours maybe

please talk to him as he knows best what a first/next step could be

as each coloumn has its own ADC

sounds great

yes. BTW, you'll find that you get better results if you measure and store the fixed pattern noise at 16 different intensity levels.

*email address removed* or *email address removed* or *email address removed*

BTW, you might as well assume 8-layer PCBs. the savings for less layers is so little.

okay, what i downloaded doesn't say much, except it appears 32 LVDS outputs.

thanks

64 LVDS

it sure would be wonderful (cost wise) to be able to adopt a cheapo FPGA like cyclone5 or something.

man, the faster (and fancier) FPGAs are totally absurd in price.

we used services from OSHpark which does group panelization of small orders for a very good price

its easy of course once you go high volume...

i bought low volume, for prototypes. let me check.

i'm trying to find price, but it was something like $30 each.

not bad

so actually 3.3$ per square inch

price is much higher at 1 to 5 boards.

you won't understand those PCBs unless I explain.

so for many industrial applications, only one gigabit ethernet cable could handle 4 cameras.

no memory there?

anyway

anything you want to ask before I leave?

should not take that long

who are the most important people for me to talk with?

he lives more in the US timezone :)

even thogh he is located like one hour away from where I live :)

i'm usually on phobos time! hahaha.

about the memory, those PCBs have a little static ram memory, but not enough to do sophisticated image processing.

ok see you later maybe then

oh, actually the ram is only 1MB.

also, there are a fair number of components on that large PCB that aren't normally stuffed, including those RAMs.

and like 10 binary counters to twiddle the addresses of those RAMs.

@se6astian: i found what i paid for the PCBs. 25 of the 150mm square controller PCB == $1200 : 50 of the 75mm square sensor PCB == $900 : 8-layer PCBs with 4-mil traces, 0201s, 0.35mm contact pitch, BGAs, some filled holes for BGAs, gold plated (all conductive surfaces).

not sure whether this will matter, but i have a 4-trace 5 giga-sample-per-second oscilloscope that might be useful for circuit verification and debugging.

hi. i've been looking through some FPGA specs at the LVDS specs. can i assume the sensor puts out an LVDS clock as well as the data channels?

good. i'm curious to figure out whether we can handle this sensor with a cheap FPGA (well, relatively cheap).

BTW, I typed in a little information for you above, about what i actually paid for those PCBs.

thanks, yes

I was hoping you had found a way to get cheap prototypes as well for 8 layers

well, i thought those were pretty cheap. oh well.

yup. how many prototypes do you envision?

i have a pretty amazing track record for needed zero revisions.

on each of the 64 LVDS channels.

we have two assembled prototypes of the 3 pcbs we ordered

since the image sensor is rather expensive we cant easily make more

yeah. is the sensor about $1500 ???

1800$

$1800 is the new [faster] one, or the old one?

oh, i guess LVDS.

price is the same for both versions

let me think... which to buy? hahaha.

one place says up to 875 Mbps LVDS.

question: how many of these do you GUESS you'll sell in the first year?

is the sensor cheaper in 100 or 1000 unit quantities?

but nothing like the price drops when ordering higher volume PCBs

but maybe 25% cheaper?

would be nice to suck some heat out of that baby to reduce noise.

okay, so sensor is $1500 even in pretty big quantities.

without making noise or vibrations

but maybe we could try to pull heat right out of the pins (that go into ground and power-plane layers).

looks like the cheapest FPGA that can handle 64 channels of LVDS costs about $160.

but its all a concept for now

the zynq we target will be around 200-250$ so thats almost in the same ballpark already

does the sensor maker advise about cooling, or just ignore the issue?

is that a heat-sink?

I'm working on a kids show, we're scheduled to shoot until February 2015. All of the sets are built to kid scale. A small camera head definitely has its place here.

Rig rather

cmosis has no advice regarding cooling

I hope we will have something useable until then

why socket rather than direct solder into PCB ?

Repair ability

The work flow for the show is kinda funny. We're shooting red but recording the hdsdi feel to atomos samurai prores, except for vfx and high speed shots, then it's internal r3d

i hope not! that's all cnc machined stuff, which should be easily within .001" == 0.025mm

It's the person who installs it

i suppose it might be possible to try to ?spring-load? the glass face against a machined surface.

but that's a bit gross, mechanically.

or you are just too afraid on general [financial] principles?

You won't notice with stills lenses though, only cinema lenses with very fine focus adjustment

I would need to read up in the datasheet but I am pretty sure it wont be healthy for the sensor

actually, with a pin-grid package, it can be soldered manually with a hot-air soldering pencil.

See how the epic does it

There is a screw that adjusts forward and backwards

how arri does it

i'd rather have any adjustability be on everything behind the sensor (light weight and not tweaked by lens weights).

frankly.

Yes but you need to design for that. Cooke s5 is as heavy as they make primes, beyond which you need lens support

http://www.cookeoptics.com/l/fiveilens.html

It needs to be able to support the lenses cinematographers use or it will never have any impact with them

@se6astian: looking at the side view of the camera (showing modules)... is this really what you're hoping to do (support and endless stream of 20mm ~ 30mm thick add-on modules)?

that is, assuming i'm understanding the photo.

that uses slots instead o stacks of modules

so like a long motherboard along the bottom, and push PCBs in from the top.

this problem always drove me crazy too! not an easy one to do optimally for every situation!

i guess they're open, and you just have an end cap.

it is such a royal pain in the PCB to try to make the PCBs (and case) so damn small.

the lens mount we use is already so big that a side-by-side 3d rig makes no sense even if you put them as close together as possible

of course some impossible-to-please types will always complain.

what diameter is that square body now? i'd love to bloat it out to 100mm diameter myself.

i vote for 100mm x 120mm... or larger! hahaha.

morning everyone!

hi Bertl

I'm an electronics designer, designed cameras before, and worked with big CCDs in astronomy (big telescope) context for about 25 years (off and on). I might be of some use... maybe.

bootstrap: yep, I've read up on the entire day now

sorry to put you through that!

and good morning Bertl ;)

yes, as mars_ pointed out, help and know-how is always welcome

bootstrap: as you already figured, we created a sensor frontend which connects via FMC to the zedboard

this FE uses 32 of the 64 sensor channels to transfer data at 300MHz (150MHz DDR)

bootstrap: link, because I've never used lens support with cooke s5s

this is one issue i don't know a lot about. what is required? 5kg, 10kg, ??? (without support).

well, i've been self-employed developing [mostly electronics, computer, software, optics] products since i got out of school, so this is right down my alley, and very interesting to me.

i've been thinking about something along these lines, but when i started my last camera this sensor didn't exist.

agreed :)

i think trying to make a 4K camera fit in a shirt pocket is... mixing two different things.

meaning... for holding while filming?

or whatever is the term now.

makes the size of the camera not so important.

bootstrap: stabilizers are their own beast

and god knows the number of times I work with directors who want "That handheld feel"

so either well mounted or with some kind of "steadycam" gizmo. or you disagree?

I have to fight the epic constantly

bootstrap: well, tbh I've been pushing for something like an Aaton form factor and so have any well known DPs that have been attracted to the project but it is like at odds with the other design goals of the project

but they are easier to machine :{

that's what plastic surgery is for! :)

hahaha... yeah, i guess we're too much like hardcore engineers to design something like that!

having irrgular pcb shapes pushes up the cost

bootstrap: and I hate it

Having to rig out cameras

rexbron: do you have a problem with an ergonomic housing which 'contains' the rectangular camera?

yeah, and rig out something else.

Bertl: depends on the design of the container

bootstrap: you want your CoG as low as possible for handheld

i don't have a wide enough variety of experience with modern high-end video equipment.

basically, the evf and top handel should be at or around eyelevel and the rest below that

the reason I ask is, that I do not see a problem with having both, a camera block and an ergonomic shoulder fitting whatever

Bertl: neither do I, provided the ergo rig works as well as it should :)

to make it an extention of your body and effortlessly controlled

in fact, the more i've thought about video, the more i want to mount the camera on a servo driven system that records the exact orientation (and moves in programmed ways).

moco cameras? or moco accessories?

I mean moco as programable moves, not motion capture of human data points

bootstrap: lol, it already is done

yes.

Cinematography is storytelling

I don't belive that any camera is "THE ONE TRUE GOD" camera, despite what manufaturers will try and tell you

which means modular.

which means not building 1000 gizmos into the camera.

at least, not 1000 mechanically controlled gizmos.

modular is a pain in my ass when I order a camera body and it shows up with the wrong lens mount or missing SDI because the owner couldn't afford gas for his new Lambo

bootstrap: you've never used a redmote

fuckign thing loses signal _ON THE DAMN CAMERA_

to attach all the gear

i guess part of what i am asking is... which gear MUST be physically on the camera, and which gear can be "elsewhere"?

all of it

no, CAN BE, but doesn't have to be.

true, I'd say SHOULD be

well... that means the cinematographer and his assistants must fly around on the hovercraft drone too.

bootstrap: if you are designing a camera for that application, sure

and the DP and director want to shoot handheld

or in studio mode

@Bertl: please enlighten us!

i'm not sure about that.

the main design focus is on 'openness'

does that mean lots of people will try to build this hardware?

(sec door bell)

but it just occurred to me that it seems IMPOSSIBLE for this project to create a fully/massively integrated camera with every little feature and convenience that multi-billion dollar manufacturers have developed in the past decades!

time to cook dinner, bbs

bootstrap: do you know of the Yolk Y2 concept camera

but now i'm looking at some photos of it.

what is that? 2 hard drives in the shoulder piece? hahaha!

lcd or evf is fine

what is evf?

great. read a bit then kick butt (especially mine).

the idea is to allow for all of those features to happen eventually

make the absolutely required functionality... than add bits and pieces (and wrappers) as necessary to customize.

the sensor can do up to 300FPS, so we plan on not artificially limiting it to less

now consider a typical camera you can buy today, using this very same sensor

for various reasons, it will be limited in many ways, but let's ignore that for now

now cmosis offsers a v2 sensor, same product name CMV12000, same interface, but higher datarates

more likely is that they will do a new camera, which is essentially the old one with the new sensor :)

so, we design the frontend/interface to be as flexible as possible

we target 4k at 12bit and high frame rates, so it is definitely not consumer

still, with open interfaces they can use the same or similar building blocks

or folks interested in still photography at high resolution can add a 60 megapixel sensor which doesn't do more than 15FPS

naturally we will have to create certain 'products' which basically consist of a default set of modules

i like that approach in theory, but i suspect in practice it won't work quite as perfectly as that.

for example, if some OTHER company comes out with the next "killer sensor", the electronics will need to be somewhat different.

true.

if in practice few people try to copy the whole design, you guys will be supplying the hardware (but not an infinite selection of accessories that other people hopefully make).

you're not planning to offer bare PCBs (or stuffed PCBs) are you? or are you?

sure, but i'm asking for your opinion of what will happen in practice.

that's what i think. i mean, i've got a 4-trace 5-giga-sample-per-second oscope... but not many people do.

the fact it breaks up signals into 64 LVDS channels helps A LOT (with signal speed), but it still isn't a hobby level circuit!

of course.

I was thinking the data would stream right into a PC, but se6basian said no.

second stage, storage via small sized SSDs inside the camera

what is an SDI recorder?

there are a number of readily available recorders which take N (3G) SDI inputs and record the data

often they are combined with preview or monitoring equippment

modern flash cards start to approach reasonable speeds, but having a complete array of them makes it rather complicated

bootstrap: just google for SDI recorder/monitor

i did that, just never noticed them before.

the first stage can also be used with SDI PCIe cards

what do you prefer for storage, if you could go right to your favored configuration?

two of them in a striping configuration

yes, that gives us about 1Gbyte/s

not sure whether that number is advertizing, or realistic sustained write speed.

for slo-mo... don't think so.

150 FPS seems to require about 23Gbps sustained.

depends on the resultion, bit depth, compression, etc

oh, one thing that might help (we shall see) is that i invented a super-simple to implement in FPGA lossless compression. only 20% compression, but... in my camera that made all the difference.

so it sounds like you prefer to store the images INSIDE the camera.

there is no RL in it :)

i guess you read how it works up above.

yes indeed.

as I said, there will be options

i.e. even if we design the 2nd stage to store data on SSDs

there is still the option to stream them via SDI or 10G ethernet

but then again, i hadn't seen this sensor at 150FPS then!

unless i screwed up my math.

if we find out what flash parts they have, we just lay out our own module.

things change!

it appears to have 8 or 16 flash memory chips on it (on a PCIe PCB).

one chip did like 20MB/s... today if it does double per die

it has hundreds of chips with volume higher than would fit your camera

i'm looking at a small PCIe card.

that claims those speeds.

ocz is sandforce, 500M burst, 150MB cont :)

do they have components on both sides of the PCB?

Bertl: on the photo is a daugther board, so double that

maybe double for both side flash chips :)

it is basically a SAS HBA and then just sata ssds put on one pcb

no, but the chips may be. if they can put flash chips in parallel, so can we.

i don't think it is sophisticated, just PARALLEL.

is the point massive throughput?

yes, i'm sure it will take a few minutes of thought. maybe even hours, or days.

but you need to realize, the point for this application is NOT as a hard disk drive, but to capture the camera output.

i don't know yet.

but if it was up to me, i'd prefer to find a way to shove 23Gbps down an thin little optical cable to the PC, and let the PC capture the data.

part of the question is frame rate. if the new sensors work at 150 FPS as seems to be the case, that's serious throughput!

a PCIe PCB with an external connector?

https://twitter.com/ApertusOSCinema/status/427147068176486400

i know. but frankly i've been itching to make a custom super-high-speed i/o card for PCs for ages! what an excuse!

go direct by pcie

i dunno. if we're talking 23Gbps, that's not PCIe... is it? well, maybe 16-channel would.

8x 5G would do :)