<bvernoux>
at the end we are all reinventing the wheel ;)
<bvernoux>
it is such a mess with all the SMA connectors that I have spent time to find the best compromise with one footprint ;)
<someone--else>
I blame eBay with its $10 Southwest connectors!
<bvernoux>
yes me too ;)
<bvernoux>
I have a lot here but they could be not so bad if center pin is soldered
<someone--else>
how many have you got? '=D
<bvernoux>
about 20
<bvernoux>
2.92mm ;)
<bvernoux>
to be accurate Southwest Microwave 1092-02A-5
<someone--else>
I have 15 or so
<bvernoux>
I have recovered that with measurement with my microscope as you do not know what you will have from ebay ;)
<bvernoux>
but I have about 50% which are not usable ...
<someone--else>
the seller turned out to have at least two different types with different pin/dielectric diameter
<bvernoux>
especially towards center pin which shall be replaced...
<someone--else>
for different substrate thicknesses I presume
<someone--else>
hmm, I scored new unused ones
<bvernoux>
in my case I was lucky as I bought all to the same guy and they are all 1092-02A-5 to avoid mess ;)
xzcvczx has quit [Ping timeout: 265 seconds]
<bvernoux>
anyway I bought them for less than 5USD/unit so I cannot really complain ;)
<bvernoux>
I'm just a bit disappointed that I need to solder the center pin to have accurate/reproductible measurements else there is crappy things which appears
<someone--else>
checked the p/ns: have some 1092-01A-5 and some 1092-02A-5
<someone--else>
yep, the price is unbeatable
<someone--else>
I had good experience, no problems with contacts
<bvernoux>
Do you know if we can replace the center pin ?
<someone--else>
pcbs are enig
<bvernoux>
It seems such option was possible
<someone--else>
it is easily replaceable if you have a replacement one
<someone--else>
don't know if that's obtanium
<bvernoux>
ha ok the center pin is just so thin ;)
<someone--else>
or unobtanium
<bvernoux>
especially for 2.92mm connectors ;)
<someone--else>
if you unscrew the sma connector part (4 hex screws), then there is a separate teflon+center pin assembly which is replaceable
<someone--else>
sma/2.92
<someone--else>
it's very modular
<someone--else>
I mean the center pin which contacts the pcb anyway
<bvernoux>
I do not really need to go up to 40GHz so far (as I do not have VNA or Spectrum Analyzer ... which can reach that freq) but it is always good for test or maybe in future ;)
<someone--else>
I think it's the same between sma/2.92/etc
<bvernoux>
ha yes great the question is where to buy that center pin
<bvernoux>
especially a new one genuine ;)
<someone--else>
southwest microwave, probably
<bvernoux>
as it is the main drawback on ebay the center pin is often damaged(soldered/desoldered multiple time ...)
<bvernoux>
I have pushed my measurements pictures here
<someone--else>
I use JLC7628 all the time, no problems
<someone--else>
impedance was within +-10% so far
<bvernoux>
ha ok
<bvernoux>
it is interesting to have feedback that they respect their tolerance especially for impedance
<bvernoux>
+/-10% is very acceptable especially for the price
<someone--else>
lemme take a look into s2ps
<someone--else>
+-10% is the normal thing for every pcb fab I think, but in jlcpcb case there is no guarantee that it will remain within this tolerance
<bvernoux>
yes if you can find something interesting ;)
<someone--else>
I guess if you use jlcpcb, you test the impedance yourself and discard the batch if it's not good enough '=D
<someone--else>
more expensive fab will do the same, but by themselves
<bvernoux>
I do not plan to use them for serious RF stuff anyway ;)
<bvernoux>
I shall play more with s2p to TDR stuff
<bvernoux>
still need to find a convenient way to do that
<bvernoux>
are you using scikit-rf ?
<bvernoux>
the limitation of my VNA is I cannot exceed 1601pts in fact so the limited number of points (big steps especially from 30KHz to 6GHz) do not help to analyze TDR
<bvernoux>
I have not done an other calibration just to check the sweep spot between 2GHz & 3GHz where there is very strange -3dB attenuation ...
<bvernoux>
I'm potentially planning to buy a LibreVNA at least it will be 30x faster and can be used with 10kpoints... even if performance is worse especially >3GHz compared to my "boat anchor" HP8753D ...
<someone--else>
seems my hypothesis about the cpwg impedance is not correct
<someone--else>
if this was the case, frequency of the dip for 31/76 mm lines would have scaled with the length
<someone--else>
which is not the case
<someone--else>
so, something else must be going on
<someone--else>
tdr resolution depends more on the max sweep frequency, I think
<bvernoux>
yes correct and my HP8753D is pretty limited ;) (it is a 30years old HW which is amazing even today but clearly too loudy a bit slow and too big for the lab)
<someone--else>
I wonder if I should simulate this..
ericonr- has joined #scopehal
ericonr has quit [*.net *.split]
bvernoux has quit [*.net *.split]
bvernoux has joined #scopehal
ericonr- is now known as ericonr
bvernoux1 has quit [Ping timeout: 265 seconds]
someone--else has quit [Quit: Connection closed]
bvernoux has quit [Quit: Leaving]
someone--else has joined #scopehal
<azonenberg>
someone--else: Did you see the twitter thread from me and derek kozel about our SMA test board?
<someone--else>
I think so
<azonenberg>
tl;dr we have small mismatches at a bunch of connectors, currently investigating root cause. TDR transform (and actual TDR measurement) far more useful than looking at frequency domain s-parameters for this kind of work
<azonenberg>
also oshpark overetched our microstrip and gave us 55 ohm Z0 instead of the design 50 ohm
<azonenberg>
dielectric seems OK, EM sim of the as-fabbed trace geometry gives impedance very close to the TDR plots
<someone--else>
I agree on the microstrip but I still think there's a problem with cpwg dimensions within the connector footprint
<someone--else>
at least if the github design is what was actually fabbed
<azonenberg>
Very possible
<azonenberg>
Current status is us going back and forth w/ Sonnet support to reconcile differences between my sim models and the measurements
<azonenberg>
this overetch is the first discrepancy we've root caused
<azonenberg>
but e.g. the samtec microstrip connector has a huge dip down to 35-40 ohms at the launch
<someone--else>
not sure about samtec, it is probably a different issue
<someone--else>
as for cpwg - simulation might perhaps not reflect these small geometry differences because of the grid size
<someone--else>
I just plugged the numbers into a cpwg calculator and arrived at impedance values which seemingly matched the tdr dips
<azonenberg>
yeah it could be mesh size and but could also be just me not running the sweep high enough
<azonenberg>
I'm gonna do some more tweaks though and see how much closer i can get
<azonenberg>
And re-import with a finer mesh size on the as-fabbed gerbers
<someone--else>
I think adjusting cpwg dimensions via a standard calculator might be an easy fix for the next rev
<someone--else>
in case new sim doesn't give conclusive results
<azonenberg>
at this point i'm not trying to fix the launch
<azonenberg>
i'm trying to fix the simulation to match reality
<azonenberg>
Once i get good agreement between sim and measured data i can use the sim to tune the launch
<someone--else>
that would be useful
<azonenberg>
Once i have sim aligned with reality i like to use it for debug too
<azonenberg>
e.g. finding a structure that's resonating that i didn't intend to
<someone--else>
though I gave up when I tried to make 2 simulators agree on probe pcb design :-)
<azonenberg>
the current density plots are great for that
<azonenberg>
lol
<azonenberg>
what kind of probe?
<someone--else>
500 ohm Z0 with vishay resistors
<someone--else>
I tried to verify my very optimized and wideband design with a different simulator.. nope, not even close
<azonenberg>
lol
<azonenberg>
how much BW did you get out of it?
<azonenberg>
my best design to date is the AKL-PT2 which is 6 GHz and primarily limited by losses in the flex material but also has repeatability issues w/ SMA soldering hence exploring connectors with alignment pegs
<someone--else>
I think it depended somewhat linearly on the resistor-tip-ground connection loop area; smallest practical dimensions yielded something about 10-15G
<someone--else>
of nonwiggly/linear phase response
<azonenberg>
Oh i was measuring to -3 dB
<someone--else>
actual -3db bw could be made higher at the expense of phase response
<azonenberg>
Nice
<someone--else>
but I don't know if I should believe these results
<someone--else>
didn't yet get to fabbing any probe hw
<azonenberg>
Oh
<azonenberg>
this was simulation only, you didnt actually build it?
<azonenberg>
you're in for a rude surprise :P
<someone--else>
the plan is to build it after doing more urgent stuff
<someone--else>
finishing the scope/tdr design seemed a better idea for now
<someone--else>
since the probe rabbithole is apparently rather deep
<azonenberg>
Yes lol
* azonenberg
looks up from 6 GHz active diff probe design
<someone--else>
in the end I think usable bandwidth is more or less determined by divider/tip dimensions and trying to be clever with the design might not be worth it
<someone--else>
so the plan is to fab the simplest smallest geometry what's still usable
<someone--else>
and measure
<azonenberg>
One of the limiting issues I had was length of the tip actually
<azonenberg>
as the tip is ~impossible to manufacture matched to a 500 ohm Zin
<azonenberg>
so it's always an unterminated stub that will reflect
<someone--else>
yep
<azonenberg>
LeCroy's ultra high BW probes get around this by using a distributed resistor in the tip itself
<azonenberg>
by machining it out of carbon fiber
<someone--else>
ah, I also simulated the distributed resistor/lossy tline design
<azonenberg>
It's not the entire tip, photos make it look fairly short
<azonenberg>
but it's literally in the needle rather than on the pcb
<azonenberg>
so you keep the stub ultra tiny
<someone--else>
seems to result in very similar bandwidth for similar dimensions, but by adjusting the length of the resistor you can adjust the damping/lpf response to remove the high-frequency wiggles
<azonenberg>
the other thing i found WRT tuning flatness is that a single resistor will have peaking due to capacitive coupling across the resistive element (this is worse w/ higher value R's)
<someone--else>
I can actually prototype this, too out of 50 ohm/square ITO
<azonenberg>
but a series string of multiple resistors has reflections between them
<azonenberg>
so it's a tradeoff
<azonenberg>
my current best actually fabbed design is 200-200-50 Vishay CH02016
<someone--else>
I've been thinking about maybe doing a custom diff amp for this
<someone--else>
there is a SiGe:C fT/fmax = 350/450 GHz process with $10k-ish MPW budget
<someone--else>
this would blow the socks off of any off the shelf parts
<someone--else>
and maybe the divider resistors can be integrated on die
<someone--else>
the only unknown so far is if their PDK supports cheap layout/verification tools or not
<someone--else>
don't want to spend a million on cadence
<azonenberg>
An open hardware active diff amplifier would be a total gamechanger all right
<azonenberg>
Did you see my mechanical design concept?
<someone--else>
yep, looks great
<azonenberg>
Right now my current amplifier board is targeting the ADL5569
<someone--else>
also a track/hold amp - there's a lot of literature with 0.5mm2 THAs on similar processes with 30-70ghz bw
<azonenberg>
which is the most broadband diff amplifier i found that had the properties i needed
<azonenberg>
Namely
<azonenberg>
1) capable of being DC coupled at input and output with the same common mode offset (the ADL5580 is great and goes to 10 GHz but has different VICM and VOCM which makes it a nightmare to actually design for in a DC coupled application)
<azonenberg>
in my case the ADL5569 nominally has a vcc/2 common mode so i shift everything so that i ruin at +/- 2.5V and that puts Vcm at ground relative to the dut
<azonenberg>
2) as much BW as possible
<azonenberg>
3) good matching in single ended output configuration
<azonenberg>
a lot of diff amplifiers have good performance for inP-to-outP and inN-to-outN
<azonenberg>
but inP-to-outN and inN-to-outP are significantly mismatched
<azonenberg>
and they expect the symmetry of the output will make up for this
<azonenberg>
but if you terminate one leg of the output and run the other single ended it destroys your CMRR
<azonenberg>
Also working on a new import of the gerber geometry as fabbed with a much smaller mesh size
<azonenberg>
25 microns in X and 5 in Y
<azonenberg>
currently have it down to 18.3K mesh cells and 2.58 GB of simulator memory
<azonenberg>
i need to get it under 2GB to run under my current license
<someone--else>
imagine being able to design all these issues away with a couple of schematic tweaks :-)
<azonenberg>
This is also using "normal" rather than "fine" meshing and maxing at 16 GHz, i'd like to go to 26.5 to match derek's s-params
<azonenberg>
and lol yes
<azonenberg>
BTW, any thoughts on sky130 for probe work?
<azonenberg>
it's not gonna be super high fmax but would a low ghz amplifier be plausible in 130nm cmos?
<azonenberg>
or is a III-V process pretty much a must
<someone--else>
also I suspect these off the shelf diff amps use a lot more ancient SiGe processes compared to the IHP 130nm
<someone--else>
sky130 would probably yield something useful, but a bipolar process would be dramatically faster. MPW costs are similar I think
<azonenberg>
well its more if you do it soon you can probably get google to fund the mpw cost :p
<azonenberg>
Tooling wise is probably where you will be screweed
<azonenberg>
the open source eda tools are already limited wrt design of digital standard cell logic
<azonenberg>
analog bjt layout is probably going to be way beyond their capabilities
<someone--else>
I remember there were some shenanigans with their io buffers/debug infra inserted between your design and the pads, don't know if this is optional now
<someone--else>
well, there are open-source custom schematic/layout/sim tools, but the question is would they accept anything designed in such way and is PDK compatible with these tools
<someone--else>
didn't ask yet
<azonenberg>
woo 13.83K sections / 1.4 GB
<azonenberg>
gonna go run this and see what i get
<azonenberg>
oh wait i forgot to change the sweep to 26 GHz
<azonenberg>
i merged the vias that were further from the initial launch transition into bars which in past simulations has given acceptable fidelity
<azonenberg>
as far as the io infrastructure, i think they are going to relax that
<azonenberg>
there are teams interested in building ddr phys and high speed serdes on sky130
<someone--else>
mesh size looks good to me
bvernoux has joined #scopehal
<azonenberg>
someone--else: re the diff probe, i have two mechanical design versions printing, one is about a week behind the other
<azonenberg>
the first one is probably too floppy but will let me fit test some other parts of the geometry and proof of concept the shielding
<azonenberg>
second one is closer to what i actually plan on using
<azonenberg>
among other things i shrunk the adjustment screw from a 4-40 to a M2
<someone--else>
mechanical design looks great, I think it's going to work (with a couple of revisions perhaps)
<azonenberg>
Yeah. the stiffness of the sls plastic is the big thing that is a bit uncertain
<azonenberg>
wrt spring dimensions
<someone--else>
but the electrical part (long uncoupled diff pair legs) bothers me a bit
<azonenberg>
Yes. I want to make that as short as possible but there are mechanical constraints
<azonenberg>
It's a pretty moudlar design though so i can easily retool just the tip area to fix that
<azonenberg>
right now connectorized but once fully debugged i want to go flex-rigid
<azonenberg>
i'm not sure if you saw the AKL-AD1 / AKL-PD1 or if you weren't following my work back then
<azonenberg>
but it was a solder in diff probe connected to an earlier amplifier design via dual SMPMs
<someone--else>
I remember that
<azonenberg>
The amp had poor CMRR and the solder-in tip was an almost perfect... i forget if quarter or half wave, dipole in the 2.4 GHz ISM band
<azonenberg>
soo yeah
<someone--else>
ah, right
<azonenberg>
also it looks like if i bump up to 26 GHz the sim uses too much ram still, going to have to merge more of those vias
<azonenberg>
oh well
<azonenberg>
i think the extra resolution from the higher bw will be worth it
<azonenberg>
hmm wow
<azonenberg>
seems like something happens right as i go from 17 to 18 GHz that requires much finer meshing and the ram usage skyrockets. guess i'm stuck at 17 GHz for now
* azonenberg
needs to buy sonnet pro
<someone--else>
I'm pondering going back to the uni for a bit, since they have *all* of the Cadence tools available for almost nothing :-)
<someone--else>
Since IHP PDK is probably Cadence-only after all (or ADS but that's not really cheaper)
<azonenberg>
what about Tanner? it seems to be the budget option for commercial EDA targeting mixed signal
<someone--else>
probably, but the question is if IHP supports it. website so far only mentions cadence and ads and mysterious texeda for layout/sim
<someone--else>
the irony is there's little need for all these expensive tools for simple designs like a diff amp
<someone--else>
schematic capture, spice sim and a simple layout tool is enough
<someone--else>
final drc can be run by the fab
<azonenberg>
Yeah
<azonenberg>
I think we're headed in that direction though. Especially for older/less bleeding edge nodes like 130nm CMOS
<azonenberg>
but if google/skywater set a good precedent i think we might see more open PDKs eventually
<azonenberg>
and just as important, open IC design tools advancing and being taken more seriously
<someone--else>
the simple open-source tools mostly there, question is fab support (apart from sky130)
<someone--else>
and not having to sign scary ndas
<azonenberg>
The one thing i'm pretty sure is missing WRT more advanced processes is all the additional DRCs
<azonenberg>
i dont know if e.g. magic is even capable of expressing a TMSC 28HPL rule deck
<azonenberg>
or if all of the more advanced checks require a more sophisticated drc engine
<someone--else>
sure
<someone--else>
I mean surely not
<someone--else>
but for small analog designs this might be less of an issue
<someone--else>
for large/expensive ones tool costs are probably worth it
bvernoux has quit [Quit: Leaving]
electronic_eel has quit [Ping timeout: 268 seconds]
electronic_eel has joined #scopehal
<azonenberg>
An hour and 44 mins per frequency bin o_O
<azonenberg>
56 minutes to calculate coupling and only six and a half to solve the matrix
<azonenberg>
This is going to take literally all day to run. and probably a good chunk of tomorrow too at this rate
bvernoux has joined #scopehal
bvernoux has quit [Quit: Leaving]
<_whitenotifier-a>
[scopehal] azonenberg edited issue #491: Channel emulation: allow any S-parameter to be selected, rather than only S21 - https://git.io/JGgE5
<_whitenotifier-a>
[scopehal] azonenberg pushed 2 commits to master [+0/-0/±4] https://git.io/JZcuX
<_whitenotifier-a>
[scopehal] azonenberg 8c24520 - DeEmbedFilter: added support for all four S-parameters rather than only S21. Fixes #491.
<_whitenotifier-a>
[scopehal] azonenberg 86b0bb4 - DemoOscilloscope: added 200 and 500 Gsps sample rates for nicer looking output when doing channel emulation at high frequencies
<_whitenotifier-a>
[scopehal] azonenberg closed issue #491: Channel emulation: allow any S-parameter to be selected, rather than only S21 - https://git.io/JGgE5