#milkymist IRC log for Tuesday, 2012-04-10

wolfsprauljpbonn: which system are you on?00:17
jpbonnWhich milkymist?  It's some pre-release thing sebastian gave me - so I'm not sure.00:26
wolfspraulah ok, I thought you meant an flterm build problem on your host :-)00:26
jpbonnoh, no.  I can download things but I haven't been able to get anything to run.  I thought I'd try something known.00:27
cladamwa[M1r4 latest rough placement] AD version: http://downloads.qi-hardware.com/hardware/milkymist_one/pcb/r4/041012/MILKYMISTONE.pdf01:46
cladamwawpwrak, good morning !01:47
cladamwathis version concludes:01:47
cladamwa1). LEDs-M are under related CONs. 2). expansion board limited area is placed. 3) most added parts are moved/placed into pcb design01:48
cladamwawpwrak, but I've not go very details on each parts. so for you reference. Since I've not confirmed all footprints. so now house stops now once I feedback them.01:50
wpwrakah, it's starting to look more complete :)01:53
wpwrakis there a way to move that fat-looking cap next to tge button ?01:55
wpwrak(under the expansion board area)01:55
cladamwawpwrak, regards to the illustration of LED for each CON. Since in AD , there's colorized area you can see, so I'll copy each one to wiki page to review, then you will easy to get understand01:56
cladamwayeah...that's also house asked me yesterday.01:56
wpwrakyes, i see them. the little blue parts01:57
wpwrakdoesn't DVI have one too ?01:57
cladamwaso C178's height(7.5mm) actually smaller than the limited height(11.04mm), but yes, we need to find another option placement.01:58
wpwrakah, it's obscured by the DVI's ground01:58
cladamwa(little blue ones) yes, they are LEDs.01:58
wpwrakthe ~10-11 mm are the absolute maximum height for anything under the expansion board. if it's taller, the board outline needs changing, which would be bad01:59
cladamwadon't care the DVI's gorund now. house will cut/seperate them.01:59
wpwrakall the "unused" space will be available for the expansion board itself. e.g., component at the bottom, or also through-hole pins01:59
wpwrak(DVI ground) naw, that was just why i didn't see the LED. now i found it :)02:00
cladamwayeah...good idea02:00
wpwraklooks pretty good02:00
cladamwaso for current rc3's area on IR, SW3, etc, they are good for the components at the bottom of expansion board design.02:01
wpwrakyes, they all stay nicely out of the way02:02
cladamwaalright, since I need to still go for very details on footprints, and those parts house already placed to see if any issues behind them.02:02
cladamwaso anytime you discover any problems on this 041012 version, let me know firstly, since I need to sum up all feedbacks to house and meet them again.02:02
cladamwabtw, the un-used LED outside boards, we need to think where we want to placement. :-)02:03
cladamwawpwrak, did you see them ?02:04
wpwrakC260 is a bit tall. 2.5+/-0.3 mm. but i guess it would be hard to move that one02:06
wpwrakthey can go wherever it's convenient. maybe try to keep them together. "interesting" locations would be: on the edge of the PCB, on the "east" side of the expansion board02:08
cladamwayeah..hard to move as it must be close to tranceiver.02:08
cladamwawait, you meant C260 or C178 ?02:09
wpwrakor, also on the edge for the main PCB, on the "south" side, in the "southwest" corner under the power section02:09
wpwrakerr, C24002:09
cladamwaaha...that's C240 actually have good history.02:10
wpwrakoh, it even has a history ;-)02:11
cladamwain rc1, it doesn't have it. In rc2 we started to added and tried to fix dram problems. but in the end, Sebastien found that it's s/w or f/w problem which is not the C240/100uF problem at all. :-)02:12
cladamwaso actually it's no need there, but somehow the level can be stable well after added it. :-)02:13
wpwrakhmm :)02:14
cladamwaoriginal issues: http://www.milkymist.org/wiki/index.php?title=RC1_signal_integrity_measurements02:15
cladamwathat's discovery after rc1 board. :-)02:16
wpwraknice scope :)02:17
cladamwain rc1 without C240, rc2 has it.02:17
wpwrakso better not to touch it. unless we're prepared to verify signal integrity again02:18
cladamwayeah..so let's don't touch C240 even we want to get more height. :-)02:20
wpwrakwe need about 2 mm of minimum clearance for U1 and U18 (1.6 mm) anyway02:22
cladamwayou meant 2 mm between GND and AUDIO_AGND ?02:26
wpwraki mean vertical clearance02:28
wpwrakthe ICs have a height of 1.6 m each02:29
wpwrakerr, 1.6 mm02:29
cladamwa8:10 now collides expansion board edge, did you see that ?02:30
cladamwabtw, seems house already moved jtag/serial headers lower.02:30
wpwrakah yes02:32
cladamwaand 8:10 socket didn't move.02:34
cladamwamaybe house made mistake: http://en.qi-hardware.com/wiki/File:M1r4_expansion_board_frame.png02:36
cladamwaor maybe the data i gave them was wrong ? : http://en.qi-hardware.com/w/index.php?title=File:Xbrd-top.pdf&page=102:38
cladamwahmm...i need to check. :-)02:38
wpwrakcan you measure the distances in AD ?02:38
cladamwasure, i'm checking... ;-)02:40
cladamwaexpansion length in AD is 85.08mm same as your pdf.02:42
cladamwacould you give me your latest drawing again ?02:43
cladamwai listed it on: http://en.qi-hardware.com/wiki/Milkymist_One_Layout_Criteria/zh_tw#M1R4_.E4.BD.88.E7.B7.9A.E6.BA.96.E5.89.87.E8.88.87.E6.B3.A8.E6.84.8F.E4.BA.8B.E9.A0.8502:44
cladamwamaybe i forgot to update your latest drawing.02:44
wpwrakhmm , 85mm does indeed seem too long02:44
wpwrakcan you check the distance between J19 and the "eastern" PCB edge ?02:45
cladamwaJ19's eastern edge to expansion's eastern edge is 83.82 mm.02:48
cladamwa83.83 mm is to expansion's eastern edge not r4's pcb eastern edge. :-)02:49
wpwrakso let's make it 82 mm then02:50
cladamwajust rough 1.82 mm clearance. :-)02:51
wpwrakupdated. here's the new version: http://downloads.qi-hardware.com/people/werner/tmp/xbrd-top.pdf02:54
wpwrakthanks for catching that one !02:54
cladamwano, you're welcome. thanks for updated. :-)02:56
cladamwawpwrak, since i think i need to make clear LED's colorized placement for you. so this review for 041012 i think we will need like 2 days at least to review and I'll record feedback to house here: http://en.qi-hardware.com/wiki/Milkymist_One_R4_Layout_History02:59
wpwraki can see the LEDs. the locations look good.03:01
cladamwaso please be paid patience for me, sorry. meanwhile I stopped KiCad's milkymist/board-m1 editing. https://github.com/milkymist/board-m103:04
wpwrakyeah, looks good. no rush :) i won't get much done until i've gotten rid of that flu anyway03:06
cladamwahouse did a very outlook color footprint for led on "cut-out layer" for mask layer. :-)03:08
cladamwaso I'll show you. :)03:08
cladamwa[041012 design sources] http://downloads.qi-hardware.com/hardware/milkymist_one/pcb/r4/041012/r4_design_20120410.tar.bz203:10
cladamwawpwrak, J24 (internal usb e/f ports) placement is okay to you ?03:11
wpwrakgood question :) we still don't know what will connect to it. so maybe yes, maybe not.03:12
wpwraki think wolfgang had some ideas about it (?)03:13
cladamwa(J16 & J20) house made mistakes on placing toward board inside, I'll feedback to them to shift to outside of board edge a bit. The distance should be the same as rc3's outlet.03:15
cladamwawolfspraul, did you follow our discussion, sorry to bother you. :-)03:17
wpwrak(j16, j20) yeah, they look a bit suspicious03:20
cladamwasure, i'll feedback them. :03:22
wpwrakgoing to suspend mode now ...03:39
cladamwahope you get rid of flu soon.03:46
kristianpaulIf you have some desing guidelines for next FN GUI, i may get colaboration from a local university at Ecuador willing to help to a floss-related project..04:30
kristianpaulwolfspraul: fedex package seem will arrive tomorrow, fyi04:34
wolfspraulkristianpaul: nice!07:18
wolfspraulwpwrak: oh wow, flu. I hope you have a 14h sleep and feel MUCH BETTER thereafter... rest well07:18
lekernel_mwalle: also have a look at DCM_CLKGEN08:10
Fallenoulekernel_: you don't check the output of LOCKED signal from DCM  because you have STARTUP_WAIT=true so that fpga won't release global reset untill DCM is locked, right ? (trying to understand)08:16
wolfspraulcladamw: yes I saw the discussion above08:45
wolfspraulam I supposed to say something? :-)08:45
cladamwhehe... maybe Werner thought you have had idea on how to connect that internal usb. ;-)08:46
cladamwbut it's okay that J24(internal usb 2*5 pins) let it placed there firstly. :-)08:46
cladamwsince I'm still reviewing one-by-one part's connection and their footprints, and needs feedback to house, so any speed-up can be joined later. :-)08:50
wolfspraulI certainly don't want to "spit in the soup" from the outside08:51
wolfspraulI am sure the layout folks first-hand, and then you and werner and maybe others will try to have a nuanced look at the internal usb situation08:51
wolfspraulI can always contribute new ideas like "let's remove the 2 internal usb", and see what happens then :-)08:51
Fallenoulekernel_: I found this in some Xilinx doc : After configuration, the RST input should be asserted until the CLKIN input stabilizes09:01
Fallenou(highly recommended).09:02
Fallenoumaybe it could be a good idea to put DCM in reset at startup and then release it09:02
FallenouI will see if I can add this09:02
Fallenouand also add a synchronizer to the reset button input to avoid meta-stability issues09:04
lekernel_the CLKIN is long stabilized before the FPGA has read its bitstream09:07
Fallenouok, good to know09:10
lekernel_if you're looking for relatively simple FPGA tasks, you can try adding ethernet to -ng :)09:10
Fallenouactually I was just speaking to MatthiasM (you may know him, from ##fpga) and he told me about meta stability issues, asynchronous reset issues and DCM things09:15
Fallenouso I just checked to see if we were doing the right things09:15
Fallenoumy main focus remains the mmu :)09:15
wpwrakwolfspraul: the USB question was about where we should place that header. you were once thinking about some cables you could source easily, that would connect on one side to the connector and on the other to the USB device. would the current tentative connector placement suit them ? (or have you dropped the idea of these cables already ?)10:19
wpwrakwolfspraul: the question is basically how we're going to use the connector. particularly if we plan to ship m1r4 with an internally-mounted RF dongle (for kbd+mouse), we should begin to develop a plan :)10:21
wpwrakthe currently suggested placement would require either a cable or a board that 1) is parallel to the main pcb, 2) meanders out of the nordeast corner, avoiding DMX and USB (both connectors are taller than the stacking height of the usual header), and 3) then perhaps ends in a USB receptacle parallel to the main PCB10:28
wpwrakmaybe we should move that connector a bit out of its corner. in there, there's very little you can do with it with the case in place. about the only choice is a cable. pcbs would already be messy.10:34
wpwrakcladamw: how about moving it right behind the "northern" USB connector ? same orientation10:36
wpwrakthat would mean moving the IC "east" of U17 (or using an SMT variant of the connector) and maybe also pushing U17 a bit to the left10:37
wpwrak(in the absence of a specific plan with a cable or such, i simply assume that we'll make some small PCB in the end)10:43
wolfspraulgod that sounds scary10:45
wolfspraulwerner knows how to force me to think through all the details in the end...10:45
wolfspraulwe can also postpone the internal usbs to a later revision... we go from 2 to 4 already10:46
wolfspraulbut ok ok, I will read through this and look at the board etc :-)10:46
wolfspraulthough I don't think I can have any more wise or better idea than the people who did this before me10:47
cladamwwpwrak, "northern" USB is J20(c/d port), so as column orientation J24 >> J20 >> J16, is it your meanings ?11:03
wpwrak(scary) the meandering PCB ? yeah. if we can place the connector a bit better, that should be friendlier11:03
cladamwhmm...sounds two opinions are different, seems should I postpone J24 question now ? :-)11:05
wpwrakcladamw: no, J24 between "DMX TX" and J2011:05
wpwrakcladamw: so the board that plugs into it could have a USB A receptacle facing "west", with the USB dongle residing over the DMX TX/RX section11:07
wpwrakwe'll have to see what this does in terms of interference. having a strong RF source right next to the rest of the board may not be ideal. but we'll see.11:08
cladamwoah... as this concern ? >> "maybe we should move that connector a bit out of its corner. in there, there's very little you can do with it with the case in place."11:08
cladamwman ! i should have read carefully backlog but if a simple illustration would be good. My brain now is hard to digest.11:13
cladamw(led's cut-off layer) wpwrak btw, see this: http://en.qi-hardware.com/wiki/File:M1r4_apa1606surck_cutoff_layer.png11:14
cladamwwpwrak, this will cut off mask. but I think a sector-shape would be rather than rectangle one. how do you think ? Are you free from bad flu ?11:15
wpwrakcladamw: (led cut-out) not sure if the shape makes a big difference in practice. as long as the opening is about as wide as the LED, it's probably fine11:17
wpwrakcladamw: (flu) naw, that will probably be around for a few days. today it's less unpleasant than yesterday, though.11:18
wpwrakred outline would be the connector. green outline would be the dongle. (more or less - may be a bit longer, considering the receptacle)11:21
cladamwwpwrak, the area of cut-out yes if makes it purple rectangle bigger than the LED's length should be okay. so i'll let house to add length a bit.11:23
cladamwwpwrak, thanks the drawing, now i realize the dongle idea. phew ~11:25
cladamwhmm...not sure if house will get crazy(not sure route easy or not) to we let J24 placed there. ;-)11:27
wpwrakthey should save the craziness for DVI ;-)11:52
wpwrakanyway, afk for a couple of hours11:58
lekernelmwalle: sent you 2x USB transceivers12:11
wolfsprauldoes 'dongle' mean that the internal usb are effectively removed?12:36
wpwrakwolfspraul: what gives you that kind of crazy idea ? :)13:13
wpwrakwolfspraul: i think you're in the wrong business. you should have become a surgeon, specializing in amputations ;-)13:14
wpwrakthe idea is to move the connector to a place where it's easy to attach an adapter board. that board would then have a regular USB A receptacle, or maybe two13:14
wpwrakinto this, you could plug, say, the rii RF dongle13:15
lekernelso, now you're talking about adding USB receptacles to the internal ports?13:15
wpwraklekernel: well, you need to connect the USB device somehow :)13:17
wpwraklekernel: and that would be a separate board. completely optional. if you wish to enjoy the unobstructed view of the M1's internals, just don't use it :)13:17
lekernelafaik USB thingies that are supposed to be used internally do not have a USB plug on them13:20
wpwrakthere are only very few that are designed for internal use. i'm more thinking of things that are a permanent part of the system, like the RF dongle for the keyboard+mouse combo13:22
wpwraklekernel: see, they do have FPGAs at revision: (in german) http://www.heise.de/newsticker/meldung/Revision-Demoparty-Buntes-Treiben-mit-Mikrocontroller-und-FPGA-1518081.html13:31
lekernel...for the Commodore One Extender Board ...13:32
wpwrakwell, maybe we could find an old ZX81 case for the M1 ;-)13:34
wpwrakthe size should be about right13:34
lekernelnah, don't bother :) Milkymist should be about making powerful and accessible tools. the demoscene is about pushing the limits of some obsolete/low-specs hardware using arcane knowledge. hardly the same thing.13:56
lekerneland Milkymist as an FPGA project is not too badly referenced, I'm pretty sure lft and other talented demosceners have stumbled on it at least once...14:00
lekernelwolfspraul: just wondering... how hard/expensive is it to source medium-sized HDTV LCDs? and what sort of electrical interface do they have? LVDS like laptop panels?14:06
wpwraklekernel: (demo scene) ah well, maybe another year :) i think there is some overlap, since that's about art through technology as well14:07
lekernelwpwrak: yeah, but it's very different aesthetics14:08
wolfspraulI spent some time on that recently actually :-)14:08
wolfspraulended up buying an AOC e2243Fw, which has a 21.5'' panel at 1920x1080 pixels, 12V power supply (about 20W)14:08
wolfspraulthat one supports vga and I think also dvi-i (digital) in14:08
lekernelwolfspraul: I'm talking about the LCM only (with the idea to hook it up directly to the FPGA)14:09
wolfspraulthat's a whole unit, but I simply took the lcd module out of it14:09
wolfspraulwhat do you mean with 'directly'?14:09
wolfspraul'module' is relative, it always goes further and further inside14:09
lekernelwithout the control chip that takes DVI14:09
wpwrakregarding the internal USB ... so far our working hypothesis was that we provide a connector and then wolfgang pulls a suitable cable out of his magic hat. shall we replace this with the working hypothesis that we'll make a little adapter board that provides at least one USB A receptacle ? (the board would be primitive - just two connectors, no other components. 1 or 2 layers)14:09
wolfspraulthere is not just one chip14:10
lekernelor VGA, etc. and scales the picture and does stuff14:10
lekernel(OSD, contrast/brightness, etc.)14:10
wolfspraulyou want to drive all thousands of wires directly?14:11
wolfspraulwait let me see I have that e2243Fw disassembled right here :-)14:11
lekernelthe thousands of wire are usually handled by chips soldered on the glass panels14:11
wolfspraulit cost an amazing 110 USD btw14:11
wolfspraulfor the whole monitor including case14:11
lekernelthen they have some simpler serial interface, which in the case of laptop panels is LVDS based14:11
wolfspraulsometimes yes, on smartphone lcds14:12
wolfspraulwell there are probably lots of varieties in the end. most larger screens have a large but very thin pcb behind the screen14:12
lekernel(the silicon is actually mounted on the glass directly)14:12
wolfspraulnot always, but ok14:12
lekernelanything that is PCB we probably don't need14:13
lekernelexcept for the backlight inverter maybe14:13
wolfspraulwait let me get that beauty over real quick14:13
wolfspraulI was planning to take pics but overloaded14:13
lekernelof course, we can make our own backlight inverter, and drive the power transistors straight from the FPGA even, but... :-)14:14
wolfspraulthe truly mainland module makers can only go up to about 100 dpi btw14:14
wolfspraulthe higher stuff comes from foreign controlled corps or korea/japan/taiwan14:15
wolfspraulso in that AOC monitor, the LCD module inside comes from L&T Display Technology in Fujian14:15
wolfsprauland there's a long and thin pcb that drives the wires14:16
lekerneldoes it have some sort of metallic cover behind it?14:16
wolfspraulyes the module is enclosed14:16
lekernelmetal or plastic?14:16
wolfspraulinside are the various filter layers etc. - a lot of layers14:16
wolfspraulmetal I think14:16
wolfspraulI specifically wanted to look at a low-cost high-volume mainland made thing14:17
wolfspraulI was more interested in understanding the supply chain14:17
wolfspraulif you look at high-end japanese models - it's all different14:17
wolfspraulthe japanese are doing all their own stuff, all the way into the ics14:17
wolfsprauldon't think that it's all the same - it's not14:18
wolfspraulof course eventually you hit physics, then it's all the same again :-)14:18
lekernelbut I guess there are common sizes or electrical interfaces?14:18
wolfspraulthe interface to the lcm module could be lvds, yes. guessing.14:18
wolfspraulnot at all14:19
wolfspraulbacklight changed from ccfl (?) to led14:19
lekernelalso we'd need one that can let high frequency magnetic fields through. those metal shields sound nasty.14:19
wolfspraulthen the different manufacturers may integrate all sorts of things14:19
wolfsprauldon't underestimate this14:19
wolfspraulthey are typically added into the wires14:19
wolfspraulso you have at least 10-20 different types of wires, even in notebook panels14:20
wolfspraultv - dozens more14:20
lekernelled is great, probably less potential EMI problems than CCFL14:20
wolfspraulI know entire companies just making endless varieties of the internal cables and connectors going to lcm modules :-)14:20
lekernelare there LCM panels without a metallic shield behind?14:22
wolfsprauldon't recall14:24
wolfspraulyou can take one apart and see when it stops working :-)14:24
wolfspraulhere are some connectors, from notebooks http://en.qi-hardware.com/wiki/Notebook_connectors#LED-backlit_LCD_panel14:24
wolfspraulbut 'module' is relative, really14:24
wolfspraulthe module maker making the lcm may get some drive circuitry from yet another company14:25
wolfspraulglass cutting14:25
wolfspraulI see 2 larger ICs here, let me take a pic14:25
wolfspraulon the last ic before the wires I see SM4025 DA1150 SMAT08214:28
wolfspraulon the pcb before that (the one that goes from vga/dvi-i to probably lvds), I see a novatek nt68668aufg14:29
wolfspraulso it's like this: vga/dvi-i -> novatek nt68668 -> cable that looks like notebook lvds cables -> long and wide and thin pcb with 2 ics, maybe rows & cols?, then tons of wires out from that pcb around the corner into the lcm module14:31
wolfspraula bit difficult to take a good picture because it's big, at night, reflections, hard to see text on ics, etc.14:32
wolfsprauldo you need pics?14:32
wolfspraulI could take the lcm module apart, but most likely at least the first one would break in the process14:32
wolfspraul100 USD experimentation costs :-)14:33
wolfspraulmost likely inside there you only find a lot of thin layers and sheets and then glass, I think the crystals are between 2 sheets of glass14:33
wolfspraulthe problem with 'sourcing' is at which level you want to source14:36
wolfspraulyou can easily buy an aoc e2243Fw retail, anywhere14:36
wolfspraulmoq: 114:36
wolfspraullead time: 1 day14:36
lekernelunless you want to do chip-on-glass processes yourself14:36
wolfspraulthat's just one tiny part of the work14:37
lekernelwhich by the way has its advantages too, since we can perhaps use the TFT panel's wires instead of hacky PCB printed coils behind the whole module14:37
wolfspraulif you want to talk to the company that makes the 'module', maybe you can get some samples, but any moq will be at least 1k, maybe 5k maybe more14:37
wolfspraulif you go in further, even higher moq14:37
lekernelwell that's the question I'm asking :)14:38
wolfspraulyou start by buying a few retail units, take them apart14:38
lekerneland do we want to strip LCMs out of e2243Fw's ?14:38
wolfspraulthey tell you a lot about the supply chain14:38
lekernelor fit a M1 board inside them?14:38
wolfspraulall of those companies can be contacted, but work is adding up quickly14:39
wolfsprauland someone has to pay :-)14:39
wolfspraullet's say nobody but Apple works that way :-)14:39
wolfspraul(and the crazy Milkymist One folks, of course)14:39
wolfspraulmost others, even the big guys, don't actually dig in that deep, they just buy with the trend...14:40
wolfspraulso what we should do is: first experiment with prototypes we buy retail14:40
wolfspraulthen come up with a realistic manufacturing plan :-)14:40
lekerneland then figure out that this weird LCM that happened to be in that device we bought is obsolete, not sourceable anywhere and incompatible with everything else?14:41
wolfspraulI'm wondering whether those L&T guys also make the 2nd pcb here on my monitor14:41
wolfspraulno no14:41
wolfspraulall wrong14:41
wolfspraulthat's how it appears if you just source from a 1-person office in HK14:41
wolfspraulbut that is 100% the laziness of the buyer only14:41
wolfspraulI am just explaining to you the innards of an AOC monitor14:41
wolfspraulit sounds simple, but who is doing that?14:41
wolfspraulyou think 1 person at all Media Markts in Germany take apart 1 AOC monitor to understand anything?14:42
wolfsprauleven if they sell hundreds of thousands of them, so what14:42
wolfspraulI could open the module, and I wouldn't be surprised if there are more company markings inside14:42
wolfspraulno secrets, just laziness14:42
wolfsprauland things don't "just change"14:42
wolfspraulexcept when idiots are buying, sure14:42
wolfspraullet me google those row & column driver markings real quick14:43
wolfspraulwith a little homework I could find out more14:43
wolfspraulthe reference design of the first pcb is surely from novatek14:43
wolfspraulthey are big14:43
wolfspraulthen maybe the row & col drivers are also from them14:43
wolfspraulthe other day I wanted to find out some info about my "asus" notebook14:44
wolfspraulyou would think asus is already an "oem"?14:44
wolfspraulthat thing is designed and probably also made by compal!14:44
wolfsprauland maybe compal outsources some lines of models again, possible14:45
wolfspraulsay someone has an idea to give a mainland team a chance14:45
wolfspraullet them do their own line :-)14:45
wolfspraulbut these things do not just change, that's not fair to say14:45
wolfspraulit's just that 99% of buyers don't care to understand the origins/supply chain14:45
wolfspraulthey just want to press a button and see the stupid thing turn on...14:46
wolfspraulabout 'incompatibilities', yes the lvds cables are being changed all the time14:47
wpwrakonly 99% don't care about the supply chain ? that would leave a while 1% that does ! i think you're off by a few orders of magnitude ;-)14:47
wolfspraullook at the url I pasted, that's just a random selection of about 10 or so14:47
wolfspraulcould be. I think 99.9% even of Dell employees have absolutely no clue where and how those "dell" computers actually come from14:47
wolfspraulthey are really just trading and customer support outsourcing firms14:48
wolfspraul(they outsource the customer support again, not that they take a single call...)14:48
wolfspraulI could go visit those L&T guys in Fujian, no problem14:48
wolfspraulthey would receive me, a potential customer14:48
lekerneldon't they have digikey-style distributors?14:48
lekernelany LCM manufacturer14:49
wolfspraulI just tried to unwind it a little :-)14:49
wolfspraulare the novatek ics on digikey?14:49
wolfspraulwe can check14:49
wolfspraulthe glass cutters? no14:49
wolfspraulglass cutting & crystals between glass is typically done in 1 factory I think14:49
lekernelwhy bother with novatek ICs? we have a FPGA14:49
wolfspraulbut this is constantly moving and changing/optimized14:49
lekernelthe only ICs we are going to have are the chip-on-glass ones and perhaps the LED/CCFL controller14:50
wolfspraulI see *hundreds* of wires going around the corner of my "lcm module"14:50
wolfspraulcan't even count14:50
wolfspraulwhat I suggest is this: take cheap chinese (mainland) models14:50
wolfsprauland then just dig in14:50
wolfspraulstay away from korean & japanese stuff14:50
wolfspraulbecause they are regularly making really innovative high-end stuff, and then you really have zero documentation14:51
wolfspraulwhereas the chinese are typically starting with us ics and reference designs14:51
wolfsprauland then relatively simple stuff14:51
lekernelok, good14:51
wolfspraulhannsg.net is another one14:51
wolfspraulthose are already the 'brands'14:51
wolfspraulthen behind that you have more companies14:51
lekernelwolfspraul: where do those hundreds wires go?14:51
wolfspraularound the corner and into the module14:52
wolfspraul"module" is a layer of sheets & glass14:52
wolfspraulso I think on this one here, no cog14:52
wolfspraulbut that's a detail, then the chip is around the corner14:52
wolfspraulwith cog you must have a border somewhere14:52
wolfspraulI think14:53
wolfspraulI am not an lcm manufacturing expert, not at all. just a few snippets over the years.14:53
lekernellol, the cheapest LCM with a decent resolution (1024x768) is $336 on digikey14:53
wolfspraulno no14:53
wolfspraulthat's why I say14:53
wolfspraulaoc or hannsg, retail 120 USD14:53
wolfspraulyes it's silly to buy all that plastic and metal etc.14:54
wolfspraulbut if you go deeper into the supply chain, moq goes up14:54
lekernelok. guess Adam will have fun gutting HDTVs in a year or so :)14:54
wolfspraulthe moment you order 100, 500, more, you can alreayd start to optimize14:54
wolfspraulplease keep the totals in mind14:54
wolfspraulsay the plastic costs 5 USD14:54
wolfsprauleven 1014:54
wolfspraulso we talk about a volume of 100 units14:55
wolfspraul100*10 = 1000 USD14:55
wpwraklekernel: digi-key have a number of parts where their intention of not selling them is fairly clear ;-)14:55
wolfsprauland for that you want to go through all the trouble of contacting another lower company and sourcing a less standardized product that may not be in stock etc?14:55
lekernelwolfspraul: and where do the other end of those wires go?14:55
wolfspraulI think they run to the crystals, no?14:55
wolfspraulI don't know, guessing14:55
lekernelso, one end goes to the crystals and the other inside the module. doesn't seem very clear....14:56
wolfspraulwell I don't see the full route now14:56
wolfspraulfor that I have to take it apart and most likely destroy14:56
lekernelhow does this LCM connect to the board that has DVI on it?14:56
wolfspraulbut obviously it must make sense electrically14:56
wolfspraulok let me try to take a few pics agian :-)14:57
wolfspraulI can tell you already from my observation the last cable to the lcm 'module' is not standardized well at all15:05
wolfspraulthat's because the different brands/makers like to mix in stuff15:05
wolfspraulon this monitor there is also a second ic next to the novatek and a smaller 8-10 wire fpc merging into the cable going to the module15:06
lekernelok, well I know someone who's spent the last few years designing control ASICs for TV LCMs. will ask :)15:06
wolfspraulyou could check out novatek a little, then you get an idea15:07
wolfspraulpic1 starts with the vga & dvi-i connector, and novatek all-in-one ic15:08
lekernellooks similar to laptop LCMs15:08
wolfspraulthen the big cable going from there to the module from L&T15:08
wolfspraulon the way a smaller cable coming from an unknown ic is mixed in15:08
lekerneland the twisted pairs smell like LVDS15:08
wolfspraulthen it goes to the thing and long pcb I described, from where the hundreds of drive wires go into the module15:08
wolfspraulbut there is no standardization there15:09
wolfspraulthat's why Apple is pushing embedded displayport, I think15:09
lekernelthe only bothersome thing here is the metal shield15:09
wolfspraulit might well work even after removing it15:09
wolfspraulthen you can see through except for the crystals that are colored :-)15:09
wolfspraulcould be, need to try if I'm in the "let's waste the next 120 USD" mood :-)15:10
lekernelwe'll replace it with a PCB15:10
lekernelwith printed coils15:10
wolfspraulso big?15:10
wolfspraulah ok15:10
wolfspraulsure sounds great15:10
wolfspraulI'm up for such ideas15:10
wolfspraulthe work into the supply chain is not crazy hard15:10
wolfspraulit's just a lot of legwork, like if you have to buy 20 things for a difficult recipe15:10
wolfspraulthat last missing spice can drive you nuts sometimes :-)15:11
wolfsprauland companies like those L&t guys - the ones making the module, can be quite happy to talk to people who want to do something different15:11
wolfspraulso even if you tell them "but our volume may only be 100 units", they may still do it15:11
wolfspraulthey call it "sample" and pull it off15:11
wolfspraulbut it all starts with you knowing exactly what you really want to build, because those corps are all *manufacturers*15:12
lekernelseems it works :)15:12
wolfsprauldid you see my pics?15:16
wolfspraulI don't think you want to touch that long & thin pcb15:16
wolfspraulthat is essentially what you mean with chip-on-glass15:16
lekernelno, we can hook up to the LVDS cable15:16
lekernelmuch simpler15:16
wolfspraulon smaller panels you integrate that more and put those row & column drivers on the glass15:16
wolfspraulyes correct15:16
lekerneland compatible with more displays15:16
wolfspraulah no15:16
wolfspraulnot really15:16
wolfspraulbecause as I said many times now, you find *lots* of varieties of the pins on those cables15:16
wolfspraulthat's 12 connectors already there15:17
wolfsprauljust mechanically different15:17
lekernelwell it's just a matter of cable with dozens of pins :) I very much prefer that to varying analogue properties of thin film transistors and mechanical differences with thousands of pins involved15:17
wolfspraulcan you take a look at those pics?15:18
lekernelyes, done15:18
wolfspraula while ago I already tried to dig in there a little, and of course you find the same patterns15:18
lekernelbefore I mentioned those twisted pairs :)15:18
lekernel[17:00] <lekernel> looks similar to laptop LCMs15:18
wolfspraulno I mean the second url with different connectors15:18
wolfspraulbecause it's not as easy as "standard lvds"15:18
wolfspraulthere is no standard for the pin assigmnents15:18
wolfspraulbut they all keep adding extra wires15:19
wolfspraulfor an integrated ir control15:19
wolfspraulintegrated webcam15:19
wolfspraulintegrated whatever15:19
wolfspraulintegrated captouch buttons15:19
wolfspraulyou name it15:19
lekerneloh but that's for the backlight, no?15:19
lekerneland accessories15:19
wolfspraulsometimes backlight goes over that cable, sometimes separate cable15:19
wolfspraulccfl often separate cable15:19
wolfspraulled often integrated with the lvds signals15:19
wolfspraulsometimes backlight has 2 wires, sometimes 415:20
wolfsprauland with high-end japanese, as I said, they will do everything differently :-)15:20
lekernelanyway, this isn't something to be afraid about imo15:20
lekernelworst case it's a minor PCB respin15:20
wolfspraulbut maybe I should repeat embedded displayport a bit more, then you will be afraid :-)15:20
lekernelin the LCM itself?15:20
wolfspraulbecause apple is trying to cleanup that lvds cable mess15:20
wolfspraulto the module15:21
wolfspraulbut right now it's all lvds, no worries :-)15:21
wolfspraulbut a LARGE VARIETY of pin assignments15:21
lekernelas long as this thing can operate in backwards DVI compatibility mode, displayport is not too much of a mess15:21
wolfspraulyou can buy a cheap aoc or hannsg monitor if you want to try15:22
wolfspraulI should try to remove the back metal, good point - thanks!15:23
wolfspraulI was happy that this thing still worked after disassembly15:23
lekernelthough their price has doubled by the time they have reached German retail stores :)15:23
wolfspraulnot at all15:23
wolfspraul20% vat, ok15:23
wolfspraulhere http://www.alternate.de/html/product/AOC/e2243Fw/695998/?15:26
wolfspraulthat's the e2243Fw in my pics15:26
lekernelah, not too bad indeed15:26
wolfspraulcurrently not in stock, but "on delivery" which is usually accurate15:26
lekernelanyway, I'm going to the US for a month on Thursday, so these things have to wait15:27
lekerneland I'll probably want to start with the coil array first anyway15:27
wolfspraulsame monitors there of course :-)15:27
wolfspraulhere http://www.amazon.com/21-5-Wide-LCD-Glossy-Black/dp/B004G603S815:28
wolfspraulthose are the smallest, actually larger is very easy15:29
wolfspraulhigher dpi is more difficult, I think the mainland companies cannot go much above 100dpi right now15:29
wolfspraulnot that I think we care, just saying15:29
lekernela bit large for the luggage, especially considering I'm going to visit a dozen cities or so :)15:29
wolfspraulthe really high-end stuff is all out of japan15:29
wolfspraul(in that sector)15:29
wolfspraulyeah unfrotunately the chinese don't have smaller ones :-)15:30
lekerneland it's not like I have nothing to do already15:30
wolfspraulin fact those are the smallest15:30
lekernelno but it's great15:30
lekernelwe'll want a nice, large, visible table15:30
wolfspraulyou can have 27'' easily15:30
wolfsprauland larger15:30
lekernelthat's why I'm asking about TV LCMs and not laptop LCMs15:30
wolfspraulthe size is not the cost issue here, it's the capability of the factory to make high resolutions15:30
wolfspraulyes the chinese are perfect for large and cheap panels15:31
wolfspraulI haven't showed you the large ones yet :-)15:31
wolfspraulbecause I tried to buy the *smallest* :-)15:31
lekerneldon't forget that we'll have to stuff the backside with a large PCB too :)15:31
lekernelhopefully 4 layers will be enough15:31
wolfspraulBenQ and Acer also seem quite active, they are Taiwanese15:32
lekernelmaybe even 215:32
wolfspraulbut if you want to get a quick view, check aoc & hannsg, their websites are decent too if you stay in the chinese interface and use chromium with translation15:32
wolfspraulaoc - http://www.aocmonitor.com.cn15:33
wolfspraulhannsg - http://www.hannsg.net15:33
wolfsprauland yes, as I explained above. those are 'retail' brands15:35
wolfspraulso later we may want to work with the companies behind them, but that is just work, completely no problem15:35
wolfspraulbut that should be economically driven, not technically driven15:35
wolfsprauleconomically at the beginning we may be better off buying the whole thing for 120 USD and taking the parts out that we want15:35
wolfspraulbecause those 120 USD were put together at an efficiency that we don't want to fiddle with at the beginning15:36
lekerneland machine and reuse the case? ;)15:36
wolfspraulotherwise it will be 200 USD in no time, and more15:36
wolfspraulok, just googled15:40
wolfspraulL&T is a joint venture between TPV and LG15:41
wolfspraulso korean15:41
wolfspraulforced to work with a chinese 'partner' :-)15:41
wolfspraulstill trying to see whether l&t has a website15:41
wolfspraulthat is sometimes a bit difficult because one needs to know their chinese character name15:41
wolfspraulah here is something, the parent company first http://www.tpvholdings.com/html/corp_profile.php15:44
wolfspraulAOC is their brand15:44
wolfspraulso they enter this joint-venture with LG, subsidiary makes modules, AOC (TPV) sells monitors15:45
wolfspraulmeaning: if you want to customize something, contact tpv (I'm too lazy to look for direct l&t contacts now, found some staff at linkedin though)15:45
wolfspraulyou get the idea15:46
wolfspraulall possible, no magic15:46
wolfspraulbut customizing/modifying very efficient processes can easily get costly15:46
wolfspraulnice, and the profits are all going to the carribean http://www.tpvholdings.com/img/structure.jpg15:47
wolfspraulthat's where I could need an island too15:47
wolfspraul(for the unsuspecting among us, BVI = british virgin islands)15:49
wolfspraulcalling it a day, n815:50
mwallelekernel: what do you think about using two PLLs for all clocks?18:09
mwallelekernel: thx for the transceivers18:09
mwallelekernel: atm im running two PLLs: Vco=600MHz with two outputs phy_clk (25MHz) and 24MHz which is feeded into the second, operating at Vco 720MHz, which generates 72MHz and 80Mhz18:12
mwalleunfortunately i'm not able to synthesize the 60Mhz for the vga pixelclock18:13
lekernelwhy not :)18:19
lekernelactually -ng uses a PLL for clocking already18:19
lekernelthe VGA pixel clock should use a DCM_CLKGEN with the SPI-like interface to change the frequency (it used not to work with first versions of ISE, but according to xilinx it is fixed now)18:19
mwallelekernel: why are the DCMs pinned by constraints? just to make the life of the placer easier?18:24
mwalleatm im just removed these constraints, because my fpga_editor doesnt start18:26
lekernelfirst versions of ISE had bugs and failed to place them18:48
lekernelalso to guarantee some timing relationships18:49
lekernel(regarding IO)18:49
wpwraklekernel: just finished watching Bret Victor's talk. really good stuff. and yes, that's a direction in which i'd like to see the M1 UI go as well22:50
--- Wed Apr 11 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!