azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing | https://github.com/azonenberg/scopehal-cmake, https://github.com/azonenberg/scopehal-apps, https://github.com/azonenberg/scopehal | Logs: https://freenode.irclog.whitequark.org/scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
_whitelogger has joined #scopehal
bvernoux has joined #scopehal
<bvernoux> hello
<bvernoux> azonenberg, do you have some news on your simulation with SMA connectors I'm very interested by them ;)
<bvernoux> In order to avoid doing lot of SMA Test Fixture "blindly" until I obtain something well matched ...
<bvernoux> The idea is to test that with TRL board PCB containing all test fixture on same PCB ...
<azonenberg> bvernoux: i'm making progress but still chasing strange results
<bvernoux> ha
<bvernoux> last time I have found demo version has some "bugs"
<azonenberg> Still learning various quirks of the tool and/or counterintuitive RF resonance problems
<bvernoux> do you have seen answer from Sonnet guys on Twitter ?
<azonenberg> i found a segfault in the full version
<azonenberg> but they replied with a workaround in less than two hours
<bvernoux> as they have tested with latest version of Sonnet and that was working better ...
<azonenberg> and are working on a full patch to avoid the issue
<bvernoux> especially for the Impedance
<azonenberg> well i have the thick metal model in l2 basic
<azonenberg> So i've been using that a lot for coplanar waveguide stuff
<bvernoux> do you have seen this thread => https://twitter.com/SonnetSoftware/status/1189175278523817984
<bvernoux> the hint was to use thick metal option
<bvernoux> and they also have used latest version to have a correct Impedance vs Simulation on a very basic test with microstrip line
<azonenberg> i have not, will look later on
<azonenberg> But what i can say is, thick metal option is critical for CPWs with 100um gap on 35um thick copper
<bvernoux> Anyway your research are very interesting especially to use "cheap" SMA with correct tuning to have best performance
<azonenberg> yeah i am not saying my current footprint is optimal
<bvernoux> as using Harmon Instruments connectors are clearly too expensive ...
<azonenberg> Its the best i've come up with so far but i have been tweaking by hand, the next step is to run some sweeps to optimize
<bvernoux> yes very nice
<bvernoux> I'm mainly interested by the match on connector
<azonenberg> Right now i'm trying to simulate five copies of my equivalent circuit for the 200R resist
<bvernoux> I'm working on a TDR stuff with OSHPark 4layers PCB
<azonenberg> compared to a 1K. or the vishay s2p model
<azonenberg> including coplanar waveguide parasitics
<azonenberg> And having some problems i dont yet understand
<bvernoux> Yes I have seen it on Twitter as it does not match the datasheet of vishay
<bvernoux> but with SMA ;)
<bvernoux> not those costly connectors
<bvernoux> especially to go up to 6GHz ...
<bvernoux> or even 15GHz as SMA as specified (when perfectly matched ...) up to 18GHz or more
<bvernoux> so far I have not found any Application Note or something like that to provide a very good match with PCB ...
<bvernoux> also doing some test with flexpcb could be very interesting ;)
<bvernoux> as the Er is very low and flexpcb are cheap compared to RO4350 ...
<bvernoux> I have found some interesting test here https://github.com/loxodes/vna/tree/devel/breakouts/sw_edge_fr408_oshpark
<bvernoux> Done with OSHPark 4layers FR408
<bvernoux> but the connector is very expensive too
<azonenberg> yes i plan to design some higher end flex probes down the road too
<bvernoux> There is clearly very nice RF stuff to do without the need to use Rosenberger or other ultra expensive connectors especially if we do not need to exceed 10GHz ...
_whitelogger has joined #scopehal
<azonenberg> bvernoux: well yeah i'm just doing high speed digital, not mm-wave radar or anything (yet)
<bvernoux> but that can help and at end everything is RF ;)
<azonenberg> Lol
<bvernoux> I would like to have same document as this https://www.hasco-inc.com/content/Technical_Articles/Optimizing_Test_Boards.pdf but for cheap SMA ;)
<bvernoux> and up to 15GHz ;)
<bvernoux> anyway it is very interesting even with such expensive connectors
<bvernoux> Southwest Microwave 1492-02A-5 are only 100USD ;)
<azonenberg> oooh this is interesting
<azonenberg> So it looks like there's a trick to use up to 2 external s-parameter passives in an L2 Basic model as long as it's a 2-port network
<azonenberg> tl;dr add internal ports to the component pads
<azonenberg> sonnet's *netlist* solver lets you use s-parameter models
<azonenberg> So you can make a 6-port netlist with 2 boxwall ports and 4 internal ports then add two s2p models to that
<azonenberg> it means you have to do two simulations, and the current density plot doesn't work
<azonenberg> But its better than using an equivalent circuit at multiple GHz
<bvernoux> Do you have the new version of SOnnet or 15.53-Lite ?
<azonenberg> I bought L2 Basic of the new version, planning to upgrade to silver and eventually gold down the road
<azonenberg> May or may not end up with pro eventually
<azonenberg> i'm hugely more productive than with lite already
<bvernoux> ha great so yes you have the new version already
<bvernoux> I have just finished my first buy on JLCPCB ;)
<bvernoux> Mainly for test with different NFC antenna ;)
<bvernoux> very cheap and fast 24H to produce the PCB
<bvernoux> I have built a NFC Calibration Coil Compliant ISO/IEC 10373-6 ;)
<bvernoux> Such Calibration Coil cost more than 250USD ...
<bvernoux> just for a PCB respecting some rules ;)
<bvernoux> Fun things in my order I have 5x PCB => 220mm*151mm for 13.68Euros ;)
<bvernoux> it is clearly really cheap
<bvernoux> even if they are only 2 layers
<bvernoux> Just a very bad things the preview of PCB is totally buggy ;)
<bvernoux> But JLCPCB say it does not reflect the real board which will be built ...
<azonenberg> Ok so it looks like the flattest option with vishay FC is actually 2x 200R + 1x 50R
<azonenberg> The SMA match is still not perfect
<azonenberg> But here's the best i have so far. 200R + 50R, 0.5mm RO4350B between CPW trace and ground plane
<azonenberg> 0.1mm planar gap from trace to ground, 0.5mm wide trace
<azonenberg> via fence modeled as a solid wall
<azonenberg> Probe consists of the 3-resistor tip assembly, 40mm of CPW after that, then the SMA transition
<azonenberg> There's room to improve, but it's already flat +/- 1 dB out to around 6 GHz
<azonenberg> And -3 dB bandwidth is >10 GHz
<azonenberg> The actual probe tip is not modeled and will likely limit usable b/w to quite a bit less than this
<azonenberg> that last url was an old graph
<bvernoux> yes very nice
<azonenberg> yeah thats an old one
<azonenberg> sec
<bvernoux> ha ok
<bvernoux> S11 too if you have
<azonenberg> Reuploaded
<bvernoux> anyway variation of +/1dB for S21 is not too bad
<azonenberg> try s21-02 again
<bvernoux> ha ok
<azonenberg> That work?
<bvernoux> yes
<bvernoux> from -15 to -17.5dB a bit strange
<azonenberg> this has s11 too (right axis scale)
<azonenberg> I'm not modeling lossless conductors so i think some of this is conductor loss
<bvernoux> it is simulated with FR4 PCB ?
<azonenberg> No, RO4350B
<bvernoux> strange it loose so much
<azonenberg> and copper conductors. Both conductor and dielectric loss are included
<bvernoux> ha ok
<azonenberg> Plating is not modeled
<bvernoux> -3dB on 10GHz range
<azonenberg> Yeah
<bvernoux> it look like FR4 ;)
<azonenberg> Lol well the match to the SMA isnt perfect
<bvernoux> ha yes it include match to SMA
<azonenberg> theres -20 dB return loss at the connector right now from 6-10 GHz
<azonenberg> and this is modeling the sma footprint, not any extra capacitance etc from the sma itself
<azonenberg> But i think it's pretty close
<bvernoux> and TDR ?
<bvernoux> Time
<azonenberg> I dont have any time domain plots, sonnet is entirely a frequency domain tool
<azonenberg> but i guess i could export a s2p for the whole design and feed that into qucs
<bvernoux> It will be interesting to see it as it shall show the connector mismatch ...
<azonenberg> Yeah that is on my list of things to do when i get home
<azonenberg> (traveling for the holidays)
<azonenberg> anyway, even if i can't do better than this
<azonenberg> this probe would still be very usable out to the mid single digit GHz range
<azonenberg> vastly better than cheap oem probes :p
<bvernoux> you should have something like that https://www.edn.com/what-pcb-material-do-i-need-to-use-for-rf/
<bvernoux> we see clearly the difference between FR4 vs RO4350B
<bvernoux> loss shall be less than 0.2dB from 0 to 7 GHz and even more
<azonenberg> Yes there's a lot of imperfections
<bvernoux> In your simulation it take into account the SMA so yes it is different
<bvernoux> will be interesting to remove the SMA to check theory too
<azonenberg> I also modeled my conductors as copper, not plating
<bvernoux> ha ok
<bvernoux> could you display the Z fro whole freq ?
<bvernoux> for
<bvernoux> we shall see the mismatch with connector
<azonenberg> Of just the connector match? or the whole board
<azonenberg> the resistors dominate
<bvernoux> in your simulation there is also resistors ?
<bvernoux> I do not remember what there is in fact
<bvernoux> it is the whole RF probe ?
<azonenberg> Yes
<azonenberg> everything but the tip needle
<bvernoux> ha ok
<azonenberg> 2x 200R FC0402, 1x 50R FC0402, ~50mm of GCPW, and the SMA
<azonenberg> this is the line impedance for the sma connector match only, without resistors or the long waveguide
<bvernoux> ha interesting
<bvernoux> there is a big mismatch
<azonenberg> Yes. it's a lot better than i had before
<bvernoux> more than 8 Ohm
<azonenberg> But i havent fully optimized it
<bvernoux> before it was more than 10 ?
<azonenberg> i dont remember how bad the naive version was but it was a lot worse
<bvernoux> a must will be +/-5 Ohms ;)
<azonenberg> Going to do sweep optimization of ground plane cutout length and width, angles, etc
<bvernoux> especially over 10GHz
<bvernoux> will be intresting to check up to 15GHz too
<azonenberg> as well as the transition from sma to cpw
<bvernoux> as anyway SMA are specified as 18GHz or more
<azonenberg> i've optimized the cpw itself to be quite good
<bvernoux> can you show the stackup + view of the whole things ?
<azonenberg> 52.2 ohms at 100 MHz, 51.4 at 2 GHz
<azonenberg> then around 51.3 out to 10 GHz
<bvernoux> ha very nice
<azonenberg> I'm not going to try and optimize more than that because fab tolerances will be more than this
<azonenberg> And this design has nice round numbers
<bvernoux> yes fab tolerance will be +/-2 Ohms ...
<bvernoux> maybe better with your fab
<azonenberg> 0.5mm wide 1oz copper trace, 0.1mm gap to copper ground at left and right
<azonenberg> then 0.508mm ro4350b between signal and ground plane on internal layer
<azonenberg> then whatever cores/prepregs needed to pad the board out to 1.6mm finished thickness, but wont be electrically active
<azonenberg> maybe additional grounding or something
<bvernoux> it is only 2 layers ?
<azonenberg> it will likely be built as a 4L board
<azonenberg> signal on front, CPW from front to inner1
<bvernoux> yes but using only 2 layers ?
<azonenberg> then inner2 and back solid ground or empty, TBD
<bvernoux> ok
<bvernoux> I have a fun design using 4 layers ;)
<bvernoux> with trace on center
<azonenberg> oh cool
<bvernoux> will be fun to analyze
<bvernoux> as it shall be more robust to external things
<azonenberg> anyway like i said i'm not 100% on this design yet
<azonenberg> But its better than what i've had before
<bvernoux> but I think S11/S21 will be not very good ...
<bvernoux> For info BloopRadar use Southwest Microwave 292-06A-5
<bvernoux> they still cost > 76USD
<bvernoux> but it is for >27GHz ;)
<bvernoux> interesting point with OSHPark 4Layers
<bvernoux> The best for 50ohms is 13.4mils traces ;)
<bvernoux> see you later bye
bvernoux has quit [Quit: Leaving]