azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing |,, | Logs:
futarisIRCcloud has joined #scopehal
_whitelogger has joined #scopehal
Degi has quit [Ping timeout: 258 seconds]
Degi has joined #scopehal
<azonenberg> electronic_eel, Degi: so i'm thinking of having the refclk etc inputs have Vt set to +10 mV but be terminated to ground
<azonenberg> this way if the input floats there's a well defined state
<azonenberg> maybe i should go higher for the DC biased inputs?
<Degi> Well could put a dac there
<Degi> 10 mV should be fine if its 50 ohm terminated
<Degi> Maybe just have 10 mV of hysteresis?
<Degi> We can just make the input isolated
<Degi> Isnt the converter isolated anyways
<Degi> I think these standard round connectors as found on laptop chargers etc should be fine
<azonenberg> There is hysteresis
<azonenberg> Degi: thats the *problem*
<azonenberg> if the converter is isolated we don't have a protective earth connection
<azonenberg> Which means bad probing could make the chassis hot
<azonenberg> i think using an explicitly earthed supply is safer
<Degi> Hmh yes
bvernoux has joined #scopehal
<_whitenotifier-f> [starshipraider] azonenberg pushed 1 commit to master [+1/-0/±16]
<_whitenotifier-f> [starshipraider] azonenberg 101b53d - Lots of work on external inputs and MCU subsystem
<azonenberg> coming along nicely, still a lot more to do
<Degi> Hysteresis has equal duty cycle at the cost of phase shift, when no signal is connected, it should be either 0 or 1, that shouldnt be too much of a problem
<azonenberg> hmm
<azonenberg> so you think just bump up the hysteresis a bit
<azonenberg> that could work i guess
<azonenberg> Rhys = 5K gives ~20 mV
<azonenberg> that should be good
<azonenberg> how's the rest of the design looking?
<Degi> I dont have time right now and probably for the rest of the day, but I could look over it tomorrow
<electronic_eel> azonenberg: I didn't get the link you posted. I'll take a look at your schematics in the evening
<azonenberg> electronic_eel: ok. I'm uploading new schematic pdfs to the url periodically
<azonenberg> so that will always be fairly fresh
<azonenberg> so i'm thinking i'll probably want to delete the 10G SFP+
<azonenberg> it's pretty redundant and will add complexity
<azonenberg> just keep the 40G QSFP+ and use a 4x 10G optic and fanout cable for 10G
<monochroma> but what about all of those people who want to use 10GBASE-T ?!? (aka, bad people ;) )
<azonenberg> sucks to be them? :p
<azonenberg> also the xc7s6 is so cute and smol
juli969 has joined #scopehal
juli969 has quit [Quit: Nettalk6 -]
maartenBE has quit [Ping timeout: 246 seconds]
maartenBE has joined #scopehal
<azonenberg> MEAD boards are out for delivery
<azonenberg> No tracking updates on stencils yet
juli969 has joined #scopehal
<bvernoux> azonenberg, do you have news on the probes ?
<azonenberg> bvernoux: waiting on a new pcb rev with better impedance matching on the tip
<azonenberg> i believe last update from oshpark was expected to ship on the 24th
<bvernoux> ha ok
<bvernoux> azonenberg, I have found that JLCPCB have also special PCB with Dielectric of 4.05
<bvernoux> which is nice
<bvernoux> called JLC2313
<bvernoux> I will probably try it next time ;)
<azonenberg> that sounds like just a standard fr4 prepreg with 2313 glass weave
<bvernoux> as it is very cheap and could be not far from RO4350B
<bvernoux> azonenberg, yes but the dielectric constant is interesting ;)
<azonenberg> 4350B is much lower, it's like 3.6, right?
<bvernoux> yes
<azonenberg> its a close match to 370hr
<azonenberg> sorry to fr408hr
<bvernoux> but not the same price
<azonenberg> True
<bvernoux> and JLCPCB are 10x faster than OSHPark ;)
<azonenberg> 2313 is a fairly wide, glass heavy weave
<azonenberg> looking at some photos of it
<azonenberg> not like a 106 or 1080
<bvernoux> and we can choose other things that ENIG ;)
<azonenberg> although i like 2116 better
<azonenberg> (does anyone know what those numbers actually mean? who invented them? is there a standard somewhere?)
<bvernoux> 2116 has higher DIelectric
<bvernoux> 4.25
<bvernoux> vs 2313
<bvernoux> 4.05
<azonenberg> yeah because more glass
<azonenberg> but more importantly, 2116 is more uniform
<bvernoux> I see 2116 is only on 6 layers boards @JLCPCB
<azonenberg> a lot of the lower-Er FR4 materials have local areas of high Er over a string of glass then low between, so the average is low
<bvernoux> anyway can be nice for cheap 6layers RF stuff ;)
<azonenberg> but you get lots of local variation
<bvernoux> yes clearly not comparable to RO4350B which shall be stable over the full size
<bvernoux> but not same price and delay ;)
<bvernoux> I plan to build TRL board with RO4350B @PCBWAY
<bvernoux> to test it but first I'm waiting my latest design to be sure all is ok
<bvernoux> will probably test tapper ;)
<azonenberg> also i'm eyeing some conductive gasket material at digikey, 1944-1363-ND
<bvernoux> ha very interesting
<bvernoux> very nice for aluminum box ;)
<bvernoux> or other
<bvernoux> to avoid EMI leakage ...
<azonenberg> actually 1944-1275-ND looks like a better fit for an aluminum enclosure in terms of material compatibility
<bvernoux> ha yes great
<bvernoux> I have asked RO4350B stack-up details to PCBWAy
<bvernoux> as it is not available on their website
<bvernoux> I plan to build some PCB TRL, filters and other fun stuff to check RO4350B
<azonenberg> Meanwhile i'm thinking of building a SLS aluminum enclosure to test how MEAD works in a metal box
<azonenberg> the other option is to try nylon with spray-on metal coating
<bvernoux> I know that with AirSpy aluminum box have no bad effect
<bvernoux> from 25Mhz up to 1.6GHz
<bvernoux> But it really depends on design
<azonenberg> thinking more, i like the idea of just using 3d printed nylon with spray on silver acrylic coating
<azonenberg> as long as thermal performance is acceptable since the nylon wont be as thermally conductive
<bvernoux> yes huge advantage of aluminum is thermal dissipation
<azonenberg> I guess i'll wait till i have the board in hand and see
<miek> just make sure the surface prep is good :) in one of the probes i got, the metallisation had all flaked off and was shorting stuff inside:
<azonenberg> miek: SLS/MJF polymers generally have pretty rough surfaces from the sintered grains
<azonenberg> i would expect this to lead to very good coating adhesion
<azonenberg> more so than a smooth injection molded finish
<bvernoux> miek, what is this board ?
<miek> bvernoux: it's inside the scope end of an agilent 1156a active probe
<bvernoux> ok
<azonenberg> monochroma: sfp-sma parts came in today
<azonenberg> No stencil or boards yet
<azonenberg> I believe those are on the same panel as the v0.9 probe so ~24th shipping date
_whitelogger has joined #scopehal
<azonenberg> MEAD boards came in
<azonenberg> First impression: the MMCX connectors don't fit. I can probably sand/file/mill out the space
<azonenberg> but as it stands the cutout is a tiny bit too small
<monochroma> :O
<azonenberg> The boards fit the enclosure nicely
<azonenberg> The sff8087 connector footprint is wrong
<monochroma> D:
<azonenberg> investigating now but it looks like the step model i have - which i thought was for a different model connector - is correct
<azonenberg> and the drawing i did the footprint off of is wrong
<azonenberg> There is a differrent drawing on another page of the datasheet that looks to match this connector
<azonenberg> I can fix it for this board by cutting off a few of the shield pins and gluing it down to the pcb
<azonenberg> but definitely something to fix pre MAXWELL
<azonenberg> And re-verify
<azonenberg> alternatively i can try and source the actual connector this footprint matches :p
<azonenberg> but that one had a surface mounted shell and i wanted the PTH for robustness
kc8apf_ has joined #scopehal
megal0ma1iac has joined #scopehal
laintwo has joined #scopehal
<azonenberg> I think i can modify the footprint to be compatible with both
<azonenberg> After clipping three pins the connector fits
kc8apf has quit [*.net *.split]
laintoo has quit [*.net *.split]
Ekho has quit [*.net *.split]
megal0maniac has quit [*.net *.split]
kc8apf_ is now known as kc8apf
Ekho has joined #scopehal
elms has quit [Ping timeout: 260 seconds]
<electronic_eel> I think removing the 10G SFP+ connector from MAXWELL is no problem. There are 3 meter direct attach breakout cables from QSFP+ to 4x SFP+ available for like 40 bucks, so this is no issue for a product of this price category
<electronic_eel> you could add 10G ethernet switch gateware, to use the remaining 3 10G SFP+ to connect other gear ;)
<azonenberg> Lol
<azonenberg> that's... not an entirely bad idea :p
<azonenberg> but not in initial firmware for sure. lol
<azonenberg> and i have enough fiber to this bench that i have no immediate need for switching
<electronic_eel> unfortunately you don't need any extra hardware for it. otherwise it would be perfect feature creep
<azonenberg> well ok let me rephrase, i have ten LC ports across two 4+6 port wall plates
<azonenberg> there is no fiber in the conduit yet
<azonenberg> the connectors are just blanks right now
<azonenberg> the coils of fiber are sitting on the shelf next to the lab bench
<azonenberg> because i'm busy pulling to a different location and havent got to this one yet since none of my current lab bench gear needs >1 Gbps yet
<electronic_eel> usually the pc running scopehal will be within close range of the scope, so some sort of cheap direct attach connector will be enough for most users
<azonenberg> Yeah, in my case it will at least go through a switch
<electronic_eel> and not having a switch in between will reduce latency
<azonenberg> two switches*
<azonenberg> because my lab PC doesnt have a 10G NIC but the one in the office does
<azonenberg> the lab pc needs a total rebuild, it's a 2014 era system from when i was in grad school
<azonenberg> but since i have a xeon on the rack running xen i can push heavy lifting to, it hasnt been a priority
<electronic_eel> a dedicated 10G or 40G card in the pc, to connect to the scope/LA
<azonenberg> My problem is i dont have any pcie slots free in the lab pc :p
<azonenberg> and buying a new haswell mobo now seems kinda silly
<electronic_eel> I suggest Threadripper, they have tons of PCIe, to connect up a 40G card, a fast GPU and some NVMe
<lain> just make sure it's Zen2, not Zen or Zen+
<electronic_eel> only downside is the quite high idle power consumption, so you don't want them running 24x7
<azonenberg> yeah i leave all of my gear running constantly and modern low end intel (i3/i5) is super power efficient
<azonenberg> my entire lab is using a bit over 2 kW
<electronic_eel> that is the difference I'm talking about. with low end intel gear you have low idle power, but not with the threadrippers
<azonenberg> i see
<azonenberg> also although i'd have a tiny m.2 on the computer to boot from,, i generally have little to no local storage on my systems
<azonenberg> everything is on the network
<electronic_eel> the ryzen have reasonable idle power too, but I think all these pcie lanes and stuff in threadripper add up
<azonenberg> my main storage right now is a 1GbE attached NFS with dual 4TB spinning rust drives
<azonenberg> but longer term i plan to replace it with a ceph cluster of probably 3 servers crammed full of nvme and attached to 10G
<azonenberg> have some block devices for VM images so i can remove the drives from my xen server, then cephfs to replace the nfs i have now
<electronic_eel> I began looking through your MAXWELL schematics
<electronic_eel> STM32 reset is missing, probably should be connected to the JTAG connector to allow connecting under hard reset
<azonenberg> electronic_eel: i'll break it out to a push button but the xilinx jtag connector doesnt actually have a reset port i think? i havent hooked up boot0 or a bunch of other things yet either
<azonenberg> like, you know, power :p
juli969 has quit [Quit: Nettalk6 -]
<electronic_eel> will work fine without ;)
<electronic_eel> you usually don't need the hard reset on stm32
<electronic_eel> but you can misconfigure the chip so that you'll have to use it
<electronic_eel> to connect to jtag
<electronic_eel> so you could also hook it up to a pin on the xc7s6
<electronic_eel> and if your really need it, set this pin via jtag
<electronic_eel> will usually not be needed, but is something that should be accessible without a hardware mod
<azonenberg> no i cannot reset via the s6
<azonenberg> because the stm32 is actually in charge of the fpga reset :p
<azonenberg> and who knows what will happen if they both try to reset each other in turn lol
<azonenberg> I'll add a push button for emergency use
<azonenberg> you'll have to open the case
<electronic_eel> can you access a xilinx via jtag when it is not in reset?
<azonenberg> Yes, that's how the LA etc works
<electronic_eel> and do stuff like boundary scan?
<azonenberg> there is a jtag disable fuse but it's a hard fuse you need to blow
<azonenberg> not something you'd do by accident
<azonenberg> and i think that only disables bitstream readback, not boundary scan
<azonenberg> For this board i am also using separate jtag connections for each fpga and the stm32
<electronic_eel> ok, so then you could always control this reset pin to the stm32?
<azonenberg> So i can have multiple debug sessions active
<azonenberg> yes, i could i guess? but that seems like a lot of work vs a tiny push button i hopefully never need
<azonenberg> it would only be required if i softbricked the stm32 by disabling the jtag pin
<azonenberg> which i've only done once, to a f031
<azonenberg> i mean you have the case open if you are jtagging it anyway
<azonenberg> that won't be accessible externally
<azonenberg> the intended production firmware update flow is going to be IP based
<electronic_eel> if you use separate jtag connectors for the fpgas, why not also an extra one for the stm32?
<azonenberg> there is one
<azonenberg> it's just a xilinx pinout since that's what all of my jtag dongles use
<electronic_eel> and it doesn't have a reset pin?
<azonenberg> there is a "halt" pin that i think might be usable as a reset but i've never actually used it so i'd have to figure out which ftdi pin it's hooked to on the digilent dongles. my homebrew dongles don't have reset, just power/ground and the four wires
<azonenberg> all xilinx fpgas can be soft reset via jtag
<azonenberg> so there's no need for a dedicated pin
<azonenberg> there is a hard reset pin but its not normally brought out to the jtag port
<electronic_eel> I just googled xilinx jtag pinout and there seems to be a hard reset
<electronic_eel> what do you mean by not brought out to the jtag port?
<electronic_eel> can't your jtag adapter control it?
<electronic_eel> what kind of jtag adapter do you plan to use for the stm32 anyway?
<azonenberg> probably just one of my normal ftdi dongles
<azonenberg> the ones i made myself are literally four wire jtag plus vref and ground, there is no reset output
<azonenberg> the digilent dongles may have the HALT pin hooked up, i'm not sure
<azonenberg> but what i meant more is, normally xilinx fpga *boards* do not have PROG_B connected to the 2x7 header
<electronic_eel> why not hook up the stm32 reset to this pin?
<electronic_eel> you could always hook it up to your ftdi cable
<azonenberg> I could i guess
<azonenberg> I'd have to find the right dongle to drive it. but i'd still have a button
<azonenberg> The digilent hs3 i have might have the reset since i think zynq likes having an external reset
<electronic_eel> a ftdi dongle can always drive it via bitbang
<azonenberg> the hs2 has a 6 pin header just like mine
<azonenberg> that then connects via a passive adapter to the 2x7 connector
<azonenberg> (These are custom built ftdi dongles for xilinx debug, they're not just a breakout cable)
<azonenberg> i just checked for reference, the ac701 does not connect prog_b to the jtag header pin 14, it's NC'd
<electronic_eel> as I said, you could mod your ftdi dongle and drive it via bitbang. I think it is a cleaner design if you have a way to control it programmatically
<electronic_eel> next thing: you should spec the -H suffix for the NCP45525 load switches
<electronic_eel> there are also the -L models available, they are enable on low
<electronic_eel> I must also say that I'm a bit disapointed by the NCP45525 - if you use a fancy ic like this, they should have a integrated current limit
<electronic_eel> unfortunately I couldn't find one with integrated current limit that allows 12v and is still small
<electronic_eel> there are automotive ones, but they are huge
<electronic_eel> and tons of 5v stuff, nice and small
<electronic_eel> don't know why they don't make nice small 12v ones
<electronic_eel> I heard local regulation is a thing, and usually you don't power that with 5v
<azonenberg> and yeah but thats what the external fuse is
<azonenberg> imo a fuse is ok because something would have to go really wrong for it to pop
<azonenberg> maybe i could go PTC instead though, i'll have to think
<electronic_eel> yeah, it is not a big issue for MAXWELL. but I thought if you use a integrated switch solution, it should have it
<electronic_eel> I don't expect the pods to have shorts unless there is something really wrong
<electronic_eel> maybe you plugged in your external SATA disk box?
<electronic_eel> next thing: how do you plan to do the pod presence detect exactly? there is the 4k7 pulldown in the pod. if you use a pullup in the fpga, you get a voltage divider
<electronic_eel> do you want to use some diff input on the fpga as comparator? or can you control the pullup strength?
<electronic_eel> or would it be easier to lower the resistor value in the pod to something like 1k to be on the safe side?
elms has joined #scopehal
<azonenberg> my plan was to put a huge pullup on the host
<azonenberg> like 100k or something
<azonenberg> this should produce a voltage on the pin that's well below Vil
<azonenberg> a discrete resistor i mean, since the fpga side pullups are not super tightly controlled drive
<azonenberg> and i deliberately chose pinning that should be close enough to a real sas connector that it shouldnt break anything
<azonenberg> also we used the *internal* sas connector
<azonenberg> so a normal pc should only have this inside the case
<electronic_eel> ok, with a 100k this should work. I had the internal pullups in mind, and as you said, they are usually a bit variable and it would be too risky to rely on them
<electronic_eel> but I have to go to bed now. unfortunately MAXWELL is quite a complicated board and I wasn't able to look at more than a few pages this evening
<azonenberg> Yeah it's my most complex project ever
<azonenberg> Also, i plan to buy one of those benchtop PnP's to assemble it
<azonenberg> likely a TVM802A
<azonenberg> hand assembling it would take so long i fear the paste would be dry before i was done :p
<azonenberg> I'll probably go hybrid, hand placing the big ICs and one-off parts but PnPing the 0402s
<electronic_eel> I haven't followed the threads about these, the last time I looked it was more trouble than they helped for most kind of projects
<azonenberg> I havent made a final decision on the model
<azonenberg> but what it comes down to is, can it place 0402s faster than me
<azonenberg> if the answer is yes, then i'm going to get one
<electronic_eel> I think you need quite some time learning how to work with the machine and how to setup the software
<azonenberg> don't forget i used to work on a SMT line
<azonenberg> it was an old europlacer 928 which is a bit of a different animal from a modern benchtop
<electronic_eel> also I heard that the software that is supplied is a pain to work with
<azonenberg> but i dont think it will be too bad
<azonenberg> My thought was to, whatever unit i get, use it to place a selection of common "house parts"
<azonenberg> probably 33R 0402, 0.47 uF 0402, 4.7 uF 0603
<azonenberg> and maybe a handful of others
<azonenberg> anything else i'll hand place, i just want it to do the ~80% of parts that end up being the majority of total placements
<electronic_eel> ah, ok, that may work better than fully loading and reloading the small machine with reels for one run
<azonenberg> yeah that was never the plan
<azonenberg> i might also do a few smaller ICs, it has a tray you can put single loose components in and it will orient and place appropriately
<azonenberg> maybe if i get the TVS diodes on a reel i'd use it for those, etc
<azonenberg> Little things
<azonenberg> i'm not expecting a $2500 pnp to be able to place a ffg676 :p
<azonenberg> probably wouldnt even have enough vacuum to pick it up
<electronic_eel> my biggest fear about such a chinese machine is that you have to end up rewriting the whole software to make it usable
<azonenberg> So from what i hear, some of these have openpnp support which makes them more usable
<azonenberg> But i havent done the research yet
<azonenberg> at this point i've decided i'm reaching the point that a basic pnp is likely going to be needed moving forward
<azonenberg> and that's about the extent of it
<electronic_eel> if you are just building one or two prototype boards and it takes 30 seconds to manually teach in each resistor to be placed, you won't win anything
<electronic_eel> so the workflow from kicad into the software for the machine is important
<azonenberg> i should be able to just write a script to convert kicad csv output to whatever format it uses
<electronic_eel> this is something I'd look out for
<electronic_eel> but if the format is some cryptic windows database, hardlinked into the software?
<electronic_eel> what I learned from machine control systems is that the software engineers writing these software think completely different than me
<azonenberg> lol
<electronic_eel> their design choices could also be done by aliens if you ask me
<miek> aside from weight maybe, why would the ffg676 be hard for a cheap machine to place?
<azonenberg> i mean thats a general problem
<azonenberg> miek: weight, level of vacuum achieved, size/range of nozzles
<azonenberg> the EP928 had two 4-nozzle heads
<azonenberg> most of the little ones only have a single head with one nozzle
<azonenberg> so you have to use it for all parts, or swap nozzles halfway through the run
<azonenberg> (the EP928 was also like the size of a small car)
<miek> i thought even the cheapy pnps could swap nozzles automatically?
<azonenberg> That i dont know
<azonenberg> like i said i'm not at that level of due diligence yet
<azonenberg> i'm a long ways from being ready to buy
<miek> fair enough
<azonenberg> The CHMT36VA is another inexpensive model that looks like a good candidate
<azonenberg> sparkfun has eagle and kicad conversion scripts already made for it
<miek> nice. looks like colin o'flynn's got one of those
<azonenberg> Yeah
<azonenberg> and it seems like once you get them set up right they can do 0402s pretty reliably
<azonenberg> i wouldnt want to do 0201 with one, but that's fine as i dont do high volume or large designs with 0201s
<azonenberg> honestly, if i got a pnp that could do nothing but 33r and 470nf 0402s i'd be happy loil
<azonenberg> i.e. series terminators and local decoupling caps
<azonenberg> that's probably half of my total bom right there :p
<miek> heh