azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing | https://github.com/azonenberg/scopehal-apps | Logs: https://freenode.irclog.whitequark.org/scopehal
juli968 has quit [Quit: Nettalk6 - www.ntalk.de]
<azonenberg> And finally. the slowly increasing edge i think is an indicator of dielectric loss in the line
<azonenberg> the inverse of what you would see in a TDR trace
<azonenberg> Actually on that note
<azonenberg> i think i'm gonna retool my channel emulation filter to be able to use S11 as well as S21
<azonenberg> so i've traced the ripples to an impedance mismatch at the SMA. Which, despite my ground plane cutout, is too low impedance
<azonenberg> I confirmed the direction and location of the mismatch by adding a 400 fF lumped capacitor from signal to ground at the SMA
<azonenberg> Which made things drastically worse
<azonenberg> same ripples, same place, but muuuuch bigger
<azonenberg> So I guess the next experiment will be trying to shave down the center conductor of the SMA a bunch
<azonenberg> of the SMA footprint*
<azonenberg> see if i can match it better
<azonenberg> as much as it pains me to say this, I think i will need a respin
<azonenberg> Unless i can figure out how to cost-effectively hand trim and vna test every single probe
<azonenberg> which on one hand sounds awful, on the other hand it would let me salvage north of $1K of boards
<azonenberg> and if they still meet spec...
Degi_ has joined #scopehal
<azonenberg> One takeaway from this is that the sma center pin is WAY too long in the footprint
Degi has quit [Ping timeout: 245 seconds]
Degi_ is now known as Degi
<azonenberg> there is no way it needs to stick out that far
<azonenberg> right now the wide area goes almost the whole way to the end of the connector
<azonenberg> So altering the footprint will mean that even if there is a mismatch the magnitude is much smaller and it will be electrically short up to much higher frequencies
<azonenberg> Red is stock AKL-PT2 pcb, no overmold
<azonenberg> blue is after hacking it with a scalpel
<lain> aha
<lain> I see o.o
<azonenberg> the ripples changed sign and the TDR shows a slight positive reflection, meaning i bumped the impedance up too far
<azonenberg> but it's now a 6 GHz probe
<azonenberg> vs a 4.3
<azonenberg> for that alone it seems worth improving
<azonenberg> So... selling the current boards as-is is out the window
<azonenberg> I will definitely be respinning, and i hope to be able to use the mill to get the current rev boards to be within acceptable ripple tolerances
<azonenberg> Soo the question is... what to do about the overmold? it seems like it causes a significant Eeff shift
<azonenberg> Which messes up my 50 ohm line so it's not 50 ohms anymore
<azonenberg> even if i match the connector to it, there will still be a mismatch when it jumps to the coax
<monochroma> laser ablation systems are getting cheap
<azonenberg> I can try to guesstimate Er of the rubber and tell the fab to control the long line to slightly not-50 ohms
<azonenberg> Such that I can be matched to 50 once overmolded
<lain> azonenberg: given measurements in free air vs. overmolded, you should be able to caluclate the Er
<lain> I think
<azonenberg> the other option is to give up on the whole molding and stick with bare flex pcb
<azonenberg> which will be a lot easier
<azonenberg> so the question is really i guess how much the rubber jacket adds to it
<azonenberg> And the second question is, is it worth trying to rework and sell some of the current batch of 250 PCBs? Or am I better off just writing them off and respinning immediately
<azonenberg> This batch of bare boards was $1020 plus import duty so probably $1250 or so (about $5 a pop)
<azonenberg> Which is enough that just throwing them out is something I don't want to do casually
<monochroma> could sell em as "early access prototypes", their specs are well known, and still useful for a lot of things
<azonenberg> You mean as-is, no rework? Yes, that is another option
<azonenberg> I mean, ~0.6 dB of ripple and 4 GHz bandwidth isn't THAT bad
<azonenberg> I'm leaning towards killing the overmold though
<azonenberg> it's a lot of work, i'd need at least 2 more revs of the mold, and there's a lot of unknowns around dielectric properties of the rubber
<azonenberg> and i'm beginning to question the benefits
<monochroma> well, how do you keep em from shorting against each other when you have multiple in a confined area
<monochroma> that was the original reason
<azonenberg> Yes
<azonenberg> Except now they're under coverlay and not bare topped anymore
<azonenberg> so only the tip is conductive
<monochroma> ahh
<monochroma> i thought you were keeping them bare
<azonenberg> this is what the tip looks like with the overmold
<azonenberg> without silicone it's coverlay flex all the way to the sam
<azonenberg> sma*
<monochroma> and it doesn't effect the signal much?
<azonenberg> The coverlay or the rubber?
<monochroma> coverlay
<azonenberg> not that i can see
<azonenberg> it's less lossy than enig
<monochroma> that was my understanding of keeping it open, was that any mask covering was going to cause issues
<azonenberg> yeah i found a reasonably low loss coverlay that should be fine for my purposes
<azonenberg> it's not great but good enough
<monochroma> oh? what's the downsidE?
<azonenberg> i mean there's still dielectric loss
<azonenberg> immersion silver + open air miiight work but i'd need to figure out how to fabricate an air gap inside the overmold
<azonenberg> that was on my list of possibilities for future R&D
<monochroma> yeah, probably worth coming back to later, but sounds like you have a basically complete product now with the coverlay and the SMA layout changes
<azonenberg> Exactly
<azonenberg> So i guess at this point the game plan is going to be selling off at least some of the current inventory as soon as i can write a manual
<azonenberg> either as-is or with some quick hand bodging to thin out the SMA center pin and match impedance better
<azonenberg> then if there's sufficient demand respin a new version that doesnt need any bodging
<monochroma> hmmm maybe mask and chemical etch ?
<azonenberg> not through enig
<azonenberg> With a 250 um endmill it should be possible to do fairly well with hand trimming
<azonenberg> My planned sale price of "not more than $100" estimated labor costs for the overmolding
<monochroma> just need a little aqua regia :P
<azonenberg> If i skip the overmold and hand bodge the sma, about the same time per probe should be doable
<azonenberg> BOM is only $5 PCB + $6 resistors + $10 SMA = $21 ish
<azonenberg> so i could sell them for $100 sustainably spending up to $79 of time assembling and reworking each board
<azonenberg> So I guess i'm going to desolder the SMA from the board i just hack-bodged then hop on the mill with one of the other... 7? populated PCBs and see what i can do with milling it
<azonenberg> Then decide best course of action based on results from testing that
<azonenberg> Gut feeling is that if i made up a jig of some sort to hold the pcb, i could probably rework each one in a couple of minutes
<monochroma> yeah
<azonenberg> But at this point i'm pretty confident the overmold is going bye-bye at least for now
<azonenberg> So I guess the plan is going to be writing up a datasheet based on the assumption that it's bare, include preliminary characterization data from one of the milled boards
<azonenberg> then open up a public beta selling off as-is units
<azonenberg> Get feedback from users then do a final v0.6 board spin fixing the impedance mismatch in gerbers
<azonenberg> i knew lots of serdes silicon had this feature but i've never seen it done in ascii art before lol
<monochroma> HEH! that's cool
_whitelogger has joined #scopehal
<d1b2> <j4cbo> what kind of mill?
<azonenberg> Sherline... 12 inch I think? 3-axis manual mill
<azonenberg> with a stereo microscope set up for precision work
<azonenberg> i have a 1/8" collet in the spindle for using micro-scale endmills and PCB drill bits
<azonenberg> my most common tooling on it is a 250 μm carbide endmill
<azonenberg> but i have mills down to i think 100 μm
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
<azonenberg> ok now i am even more confused
<azonenberg> Latest measurements show a few small ripples on an unmodified probe, but it's actually quite good
<azonenberg> And it hits 6 GHz BW after de-embedding the fixture
<azonenberg> actually not even
<azonenberg> Red is latest test
<azonenberg> Blue is AKL-PT1 v1.3, cyan and pink are PicoConnect 921 with one and two ground needles making contact with the DUT
<azonenberg> Fixture de-embedded from all measurements using the same .s2p
<azonenberg> I think i may have been confused and looked at the version with the overmold on or something
<monochroma> :o
<azonenberg> This is good enough to sell as is. I need to characterize a few more and test consistency
<azonenberg> but basically i think i don't need any rework or a respin or anything
<azonenberg> I think i wrecked four boards barking up the wrong tree looking for an impedance mismatch that was probably just a bad solder job on that first prototype
<monochroma> is the original board that kicked this off still intact?
<azonenberg> Um, i can check
<azonenberg> i think i salvaged the SMA off it to rework the other ones with
<monochroma> ahhh
<azonenberg> since i only had like three left and didnt want to wait for more to come in
<azonenberg> and that solder joint is now suspect :p
<azonenberg> So i destroyed the evidence
<monochroma> yeah
<azonenberg> Buut basically right now it looks like if i don't overmold, the PT2 as it stands is a 6 GHz probe that just needs a datasheet written and more units assembled and it's ready to offer to the public
<azonenberg> maybe even closer to 7 GHz
<azonenberg> note the vertical gridlines are only 0.5 dB spacing
<azonenberg> -23.5 dB would be 3 dB down from the DC gain
<azonenberg> and guesstimating a bit i think it's still a GHz or so from hitting that
<azonenberg> i'd spec it at 6 GHz as i have no way to characterize further
_whitelogger has joined #scopehal
juli968 has joined #scopehal
bvernoux has joined #scopehal
<bvernoux> azonenberg, I just saw your latest video about probe testing up to 4GHz
<bvernoux> very nice
<bvernoux> You speak about you are waiting a new scope which works up to 16GHz correct ?
<azonenberg> Yes
<azonenberg> It's a LeCroy SDA 816Zi
<azonenberg> Which I will be upgrading next year to the SDA 816Zi-B
<azonenberg> I am removing the overmold from the AKL-PT2 though
<azonenberg> I spent a lot of time on it, but it just caused too much of an impedance shift
<azonenberg> I'm sure I could respin the board with a different trace geometry and make it work, but it seemed like a lot of effort for minimal benefit
<azonenberg> So it's just going to be bare flex PCB
<bvernoux> woo SDA 816Zi-B have impressive specs
<bvernoux> 80GS/s per chan
<azonenberg> That's interleaved, so only on 2 channels
<azonenberg> The original 8Zi does not have the interleaver so is 40 Gsps only
<azonenberg> but on all channels simultaneously (my current scope is 40 Gsps interleaved, 20 Gsps on all channels)
<azonenberg> The 8Zi can be upgraded to 8Zi-B which includes an improved frontend that adds interleaving capability and some other improvements
<azonenberg> as well as a faster PC and more capable software
<azonenberg> You can also do bandwidth upgrades, it's a modular design that lets you swap parts in and out without having to replace the whole scope
<bvernoux> yes at least to make the jump to Windows10 instead of Win7 (as IIRC it was provided with Win7)
<azonenberg> Lol
<azonenberg> no, mine is a 2009 model year. It runs vista
<bvernoux> the best will be native linux but it seems Lecroy does not do that ...
<azonenberg> But AFAIK the chipset on that motherboard is too old (core2quad) to run win10
<azonenberg> So I will be doing the win10 upgrade next year as part of the upgrade to the 8Zi-B
<bvernoux> ha strange I have a old PC Core2Duo which work fine with win10
<bvernoux> ok it is a bit slow ;)
<azonenberg> It might be possible, but realistically it lives on an airgapped lab network anyway
<azonenberg> every piece of gear on that subnet has an unauthenticated scpi interface
<azonenberg> an old unpatched windows OS is hardly an additional security risk :p
<bvernoux> yes it is not so critical when it is in lab with restricted access to internet ...
<azonenberg> So I'm just going to keep vista on it until next year
<azonenberg> Anyway, right now it's at lecroy's factory getting refurb'd
<azonenberg> the dealer turned it on to test before shipping to me after it had sat in their warehouse for a while
<azonenberg> and it wasn't working right
<azonenberg> It's fixable, and I'm going to get a factory refurb'd scope at the dealer's expense, but will delay things by another week or so
<bvernoux> Does they will provide high speed probe with it too ?
<bvernoux> as they are crazy expensive ;)
<azonenberg> The dealer is not. But I traded my 2 GHz scope back to LeCroy
<azonenberg> In exchange for that I'm getting a rack kit for the SDA, a single 13 GHz active diff probe (D1330-PS)
<bvernoux> ha ok nice
<azonenberg> And upgrading the SDA's trigger board to the full 14.1 Gbps SERDES pattern trigger
<bvernoux> 13GHz actif diff probe is nice for reference probe
<bvernoux> to check the diff with your own diff probe which are WIP IIRC
<azonenberg> Which will let me trigger on an 80-bit NRZ pattern, a single 64/66b symbol, or up to eight 8b/10b symbols
<azonenberg> So i can trigger on the start of a 10Gbase-R frame, a PCIe TLP, etc
<bvernoux> you video https://www.youtube.com/watch?v=Jjh-g0YiAsY was like an electronic RF marathon => 4hours
<azonenberg> That was my first time experimenting with live streaming
<azonenberg> No editing, all real time
<bvernoux> it is a real pain to solder those SMA highFreq I have the same it is awfull
<bvernoux> each time it take lot of time to solder them correctly
ericonr has quit [Ping timeout: 258 seconds]
ericonr has joined #scopehal