azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing |,, | Logs:
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
futarisIRCcloud has joined #scopehal
<azonenberg> Sooo it looks like my probe issue was the resistor
<azonenberg> I'm still getting some issues but it's way better than i had before
<azonenberg> In the current configuration i'm going to say it's definitely suitable for recovery of gigabit serial data, you probably wont be doing in-system eye measurements with it but i'm not sure most probes can do that
<azonenberg> i really do want to get a ZS1500 to compare against though
<azonenberg> and one of those pico probes
<tnt> I have a couple of 1152A, I'd be curious to compare against that. (I'd need to get glscopeclient running first though :p)
<azonenberg> tnt: 1152a? are those active probes of some sort?
<tnt> azonenberg: yeah, they're an older model.
<azonenberg> ah ok, so about the equivalent of the lecroy zs2500
<azonenberg> i was going to get a zs1500 because my scope is only 1 GHz b/w
<tnt> yeah, mine is only 500 MHz, I got that probe because I was able to get it cheap on ebay :p Technically my scope doesn't support it, but a bit of hacking and I was able to make it work just fine on there.
<azonenberg> That might not be of much help then
<azonenberg> because my probe seems pretty flat out to 500 MHz
<azonenberg> it's ~500 - 1G i have problems with
<azonenberg> then past 1G it actually seems to peak a bit then level off
<bvernoux> just seen latest news about the probe
<bvernoux> the signal is really better
<bvernoux> I suspect it could be better with coplanar waveguide ...
<azonenberg> Yes
<azonenberg> I think i've gone as far as i can with the current PCB layout
<azonenberg> For a coplanar waveguide is there any rule as to how wide the ground has to be?
<azonenberg> the closed-form models all assume an infinite ground
<azonenberg> but i'm not sure how it falls off as the ground gets smaller
<azonenberg_work> Right now what i'm looking at doing is an 0.33mm wide pcb trace with 0.125mm horizontal spacing to the top layer ground (min oshpark clearance) and 0.18mm of FR408 prepreg to the layer 2 ground plane
<azonenberg_work> saturn PCB gives 50.7 ohms for that
<azonenberg_work> Then for an 0402 passive if i have ground on layer 3 (h=1.382 mm) but not 2, saturn gives me 62.9 ohm impedance
<azonenberg_work> meanwhile, if i put my main reference plane on layer 3 the whole way along, a 1mm wide conductor is a near perfect match (51 ohms)
<azonenberg_work> I think that's the way to go, because the 0402 is going to have an impedance step regardless (it's a resistor, duh)
<azonenberg_work> and the 1.1mm MMCX center conductor will be nearly perfectly matched (49 ohms)
m4ssi has joined #scopehal
<bvernoux> for me just follow the PCb calculator like KiCad or other
<bvernoux> or even better the one from PCB Toolkit v7.08
<bvernoux> especially to match impedance
<azonenberg> yeah i used saturn pcb
<bvernoux> Conductor Impedance -> Coplanar Wave
<azonenberg> yeah thats what i did, the problem is that it isnt a full field solver
<azonenberg> and doesnt model non-infinite ground width
<azonenberg> i just wasnt sure how big a deal that was
<bvernoux> IIRC the computation is quite good and shall be checked with a VNA anyway at end
<bvernoux> the main issue is accuracy of manufacturer to match the spec
<bvernoux> which can be specified to respect 50 Ohms Impedance
<bvernoux> but it cost more and cannot be done at OSHPark:(
<azonenberg> Yeah thats why i wanted to go with a coplanar waveguide specifically
<azonenberg> because with a nice wide pcb trace and ground plane, small etch changes will have much less of an effect on impedance vs microstrip
<azonenberg> microstrip is fine if you're using a real controlled impedance fab (although i'd probably still go GCPW for this probe to reduce loss) but here, i need to design for loose tolerance
<bvernoux> the best for impedance is RO4350 but I Imagine it is costly
<bvernoux> IIRC Harmon Instrument has done some nice test with just OSHPark
<bvernoux> even better with flexible PCB
<bvernoux> as the permissivity is better than FR408 and also for coplanar it is better thinner ...
<bvernoux> but IIRC their process is not frozen
<bvernoux> but there is no spec on substrat dielectric constant today
<azonenberg> yes, i wouldnt use oshpark flex for anything i cared about specs of
<azonenberg> i'd go multech and get controlled impedance
<bvernoux> it is hard in fact
<bvernoux> with flex PCB spec
<bvernoux> W=6mils, H=4mils(which is specified 4mil +/-0.4mil) and G=6mils there is a Z of 56Ohms
<bvernoux> problem is we cannot increase H
<bvernoux> and we cannot increase W/G as the ratio betwen W/H shall be < 2 else the simulation is wrong ...
<bvernoux> the best we can have is W=G=7.5 and H=4mils for a Z0=50.2
<bvernoux> and so with the tolerance of +/-0.4 mil for H we have a theoretical variation of Z0 47.3 to 52.8 Ohms which is not so bad
<bvernoux> I have choosen a Er of 4.6 also so it shall be fixed with the good one
<bvernoux> IIRC it is something like 3.8 TBD
<azonenberg> i think polyimide is more like 3.6
<bvernoux> in that case W/G shall be 8mils
<bvernoux> but the computation of Z0 is outside range of W/H ...
<bvernoux> even 9mils for 3.6
<bvernoux> with Z0 47.6 (H=3.6) to Z0 53.4 (H4.4)
<bvernoux> with a W/H = 2.045 ..
<azonenberg> how's that look?
<bvernoux> nice
<bvernoux> add some via on GND too
<bvernoux> also it will be better to avoid the big via
<bvernoux> for the pin
<bvernoux> maybe using a PIN with SMD
<bvernoux> conductor gap shall be equal to conductor width too
<bvernoux> by computation something like 7mils to 9mils depending on Er
<bvernoux> and H
<bvernoux> it is designed for OSH FlexPCB ?
<bvernoux> I think the best will be conductor W/G=9mils in that case
<bvernoux> to have something like +/-3 Ohm errors around 50 Ohms
<bvernoux> for Z0
<bvernoux> to be validated with VNA
<bvernoux> and also by comparison like you do
<azonenberg> This is designed for oshpark rigid
<azonenberg> this isn't a flex probe
<bvernoux> also a very nice tool => QucsStudio-2.5.7
<bvernoux> it is even better than Qucs
<bvernoux> ha ok so with FR408
<azonenberg> Yes
<azonenberg> I'm putting ground on layer 3 of 4
<azonenberg> so one prepreg and one core between me and the solid plane, then ground on the top layer
<bvernoux> ha with 4 layers PCB
<azonenberg> it's a 2-layer board built on a 4-layer stackup
<azonenberg> i'm only using layers 1 and 3
<bvernoux> ok
<azonenberg> so basically copper, prepreg, core, copper, prepreg
<azonenberg> This gave 50 ohms for 1mm wide top-layer traces with 0.125mm clearance between top-layer ground and signal
<bvernoux> I recommend checking with QucsStudio 2.5.7 => Line Calculator
<azonenberg> The mmcx center ground is 1.1mm wide
<bvernoux> it is more advanced
<azonenberg> which saturn pcb calculates as about 49 ohms on the same config
<bvernoux> it take into account also the length of conductor and lot of other things
<azonenberg> I deliberately chose these design rules to be relatively tolerant to overetch at the fab because i know oshpark has poor line width control for this kind of stuff
<azonenberg> So i targeted a config where a moderate overetch won't change impedance too badly
<bvernoux> yes it is very interesting
<bvernoux> especially to check on 3 boards to see differences
<bvernoux> I will buy a batch of 3 also ;)
<bvernoux> could you provide the different value you have used
<bvernoux> G/W/H/L/T
<bvernoux> T=Thickness of copper which should be 35um
<azonenberg> My design rules are 1mm wide conductor, 0.125mm clearance to top layer ground, 1.382mm to inner layer ground, FR408
<azonenberg> and yes, 35um conductor
<bvernoux> taking into account those tests
<bvernoux> as ER seems to be 3.3 instead of 3.66
<bvernoux> I find Z0 = 45Ohms
<azonenberg> yeah i used the 3.66 value for this
<azonenberg> So we'll see how it works :)
<bvernoux> I obtain that
<bvernoux> maybe I have missed something
<bvernoux> L is not 100mm IIRC
<bvernoux> probably something like 60mm
<azonenberg> yeah but that wont change impedance
<azonenberg> just loss values
<bvernoux> yes
<azonenberg> roughness is also way more than 100 nm i suspect
<bvernoux> that parameter change only conductor losses
<azonenberg> This might just be differences in the solvers qucs vs saturn uses
<azonenberg> In any case, i think i was say more than 5 ohms off the ideal impedance
<azonenberg> way*
<azonenberg> So this should be a lot closer even if it's not a perfect match
<azonenberg> also, i expect the conductor to be slightly overetched which will bring the impedance higher
<bvernoux> yes
<bvernoux> also avoid the pin through hole
<bvernoux> there is reflection ...
<azonenberg> that's the ground
<bvernoux> even if that shall not change lot of things
<azonenberg> we shouldnt have much current there, no?
<bvernoux> the pin is not ground it is signal
<bvernoux> the pin header through hole
<azonenberg> yes, that is ground
<bvernoux> ha ok
<azonenberg> the bare pcb trace coming off the resistor is signal
<azonenberg> i cut a sewing needle to about 1cm long and solder it edge-launch to that trace
<azonenberg> then secure with UV epoxy
<azonenberg> the pin header isn't a pin header either, it's actually a tinned copper lead from a PTH LED
<bvernoux> yes I do not know why I was thinking it was linked on bottom to the pin ;)
<bvernoux> so yes GND
<bvernoux> so it is ok
<azonenberg> that i bend to make a homebrew kinda-springy ground
<bvernoux> what the resistor value ?
<azonenberg> this is my current probe tip, the ground is in line with the signal
<bvernoux> I can check what happen with it in Qucs
<azonenberg> so its a little hard to se, but you can see the needle
<azonenberg> The original resistor was a Yaego 953 ohm that had poor high frequency performance
<azonenberg> the one I am using now is a Vishay FC series 1K
<azonenberg> it's flip chip style mounting without wraparound leads
<azonenberg> i really wanted 950 ohms but couldnt find anyone stocking it
<bvernoux> you are using a 0603or less ?
<azonenberg> 0402
<bvernoux> ok
<azonenberg> Vishay models the 0402 flip chip package as 26.2 fF shunt capacitance pad to pad
<azonenberg> then the nominal resistance value in series with 1.89 pH
<bvernoux> yes very nice
<bvernoux> as on previous resistor nothing was specified anyway ;)
<azonenberg> The 1K is flat out to ~600 MHz then starts dropping off a bit, at 1K ohms it's about 900 ohms
<bvernoux> as they are not for RF
<azonenberg> So you will get a little bit of peaking around 1 GHz which is consistent with the data i am seeing
<azonenberg> however i also am seeing loss from about 500 - 900 MHz which i suspect is PCB related
<azonenberg> hence the respin
<bvernoux> the 0402 Flip Chip version is even better in datasheet
<azonenberg> That's what i am using
<bvernoux> ha ok
<bvernoux> and it is something like 10USD/resistor
<azonenberg> not even closer
<azonenberg> mouser has them for 2.05 USD @ qty 1, 1.70 @ qty 10
<bvernoux> anyway at end it is still cheap with good performance to be confirmed with your latest design
<azonenberg> (no stock at digikey)
<bvernoux> ha good
<azonenberg> Now i'm working on respinning the enclosure
<azonenberg> the first round needed some machining in order for the MMCX mating half to fit (the female connector on the PCB fit flawlessly but the male connector on the cable collided with the enclosure)
<bvernoux> the enclosure seems already very nice
<azonenberg> Yeah i am overall pretty happy with it, it fits the Pico probe positioners beautifully
<azonenberg> My plan is to keep the top half as-is but respin the bottom half with a bit more clearance
<azonenberg> no other changes for the moment
<azonenberg> it's starting to look like a proper piece of test equipment though, so i'm pretty happy
<bvernoux> I need to buy some Pico prob positioners ;)
<azonenberg> i just need it to *work* like one :)
<bvernoux> I do not remember the ref
<azonenberg> pico TA102
<azonenberg> you can buy them from a bunch of different places for 25 USD each
<bvernoux> I plan to buy things on Farnell so I will check
<azonenberg> My enclosure is cylindrical and too big to fit into the round holes, but i have a rectangular cutout region exactly the thickness of the cutout on the hole at the tip of the positioner
<bvernoux> I try to avoid at maximum mouser/digikey because it is awfull with tax/import in France
<azonenberg> So it slides between the open jaws, then the cylindrical parts of the enclosure keep it from sliding front to back
<bvernoux> ha nice
<azonenberg> my first iteration was boxy, the rounding was just to make it more comfortable to hold freehand
<azonenberg> it serves no other purpose
<azonenberg> Also, i see them at Newark and Allied, not farnell
<azonenberg> also
<bvernoux> yes Farnell does not have it :(
<azonenberg> has them for 18 GBP though
<bvernoux> like lot of other stuff digikey ... are too often mandatory
<azonenberg> Since they're still EU, for now, can you avoid import tax for that?
<bvernoux> yes until the brexit ;)
<azonenberg> Hence "for now"
<bvernoux> it is why UK is not good guys as they avoid tax ;)
<bvernoux> they import from china and resell to EU without any tax ;)
<bvernoux> they are the bad boys in EU but they are not the only one hehe
<azonenberg> Ok so, respun pcb ordered from oshpark, i have enough resistors and mmcx's to make another two probes
<azonenberg> now to fix the enclosure
<bvernoux> do you plan to do some test with flexPCB ?
<azonenberg> I have a separate flex pcb that i already did some time ago from a real fab with impedance control
<azonenberg> that i want to try populating with the FC resistors instead of the cheap yaggos
<azonenberg> yaegos*
<azonenberg> have not done that test yet
<azonenberg> they're a solder-in design with castellations at the tip then a tiny fr4 stiffener under the MMCX
<azonenberg> gregdavill tweeted some nice pictures of them a few weeks back
<azonenberg> i have 500 of them and they work great for lower speeds but with the cheap resistor performance in the GHz range was pretty bad
<bvernoux> my idea is to use that also for LA as they are very thin
<azonenberg> Yes
<bvernoux> so for such use case I do not need something > 200MHz (today)
<azonenberg> I can mail you a few bare probes if you want to try characterizing them
<bvernoux> ha yes will be great I can check S11/S21 with my miniVNA Pro from 10KHz up to 230MHz
<azonenberg> Shouldnt be any problem with import taxes as the bare PCBs cost me 500 USD per 500 boards
<azonenberg> so i'll declare them as commercial samples (which they are) with a value of $1 each
<bvernoux> give me your paypal account I will make a payment as such things cost money + shipping
<bvernoux> woo you have bough 500
<bvernoux> I imagine it is the only way to have them cheaper else it is something like 30USD/probe (by5 or 10) ?
<azonenberg> the NRE was like $350 or something
<bvernoux> ha ok
<azonenberg> at that point it was silly to not buy a lifetime order :p
<azonenberg> then i populate as the need arises
<azonenberg> I have yet to try one with the vishay FC resistors
<azonenberg> note, they use a right angle smt mmcx
<azonenberg> not the edge launch one i use in my newer designs
<bvernoux> ha yes especially if that eavily improve them > 125MHz or even better up to 500MHz
<azonenberg> And the impedance match on the center conductor may not be perfect
<azonenberg> i dont think i put a ground cutout under the contact
<azonenberg> i couldnt tell them from a direct SMA connection with my 350 MHz scope
<bvernoux> would be funny to add a little LNA with filter at input too as option ;)
<azonenberg> I actually do plan to make an active probe at some point
<azonenberg> Just not yet
<azonenberg> Right now i want to see how far i can push passive probes
<bvernoux> yes step by step
<azonenberg> I'm pretty happy with the design i have now as far as debugging goes, it's easily enough to read gigabit data non-intrusively
<azonenberg> With a probe that costs $4 per PCB plus $5 in components, plus a sewing needle and piece of wire
<azonenberg> Lol
<azonenberg> What's the IEC refdes for a sewing needle?
<bvernoux> yes clearly a must have for a very very cheap price
<azonenberg> would it be P for plug since it's a male connector? :p
<bvernoux> especially when compared to R&S, Agilent ... probes which cost more than 10x or even 100x that...
<azonenberg> Yep
<azonenberg> Basically at this point i am trying to compete with the pico 910 series at a lower cost
<azonenberg> even if i can't hit 4-10 GHz bandwdith, a 1 GHz passive probe for $30ish in parts will be awesome
bvernoux has quit [Quit: Leaving]
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
m4ssi has quit [Remote host closed the connection]
<_whitenotifier-3> [scopehal] azonenberg pushed 4 commits to master [+0/-0/±5]
<_whitenotifier-3> [scopehal] azonenberg a828c4e - Fixed handling of optional input channels
<_whitenotifier-3> [scopehal] azonenberg 6377e34 - USB activity decoder: correctly include entire preamble
<_whitenotifier-3> [scopehal] azonenberg bd2036b - ClockRecoveryDecoder: initial gating support
<_whitenotifier-3> [scopehal] azonenberg ac74973 - EyeDecoder2: better handling of gated clocks, added vertical interpolation to prevent "scan line" visual artifact when roundoff error makes some histogram bins empty
_whitelogger has joined #scopehal
bvernoux has joined #scopehal
bvernoux has quit [Read error: Connection reset by peer]
bvernoux has joined #scopehal
futarisIRCcloud has joined #scopehal
<azonenberg_work> apparently pico makes a lower end probe that's more along the lines of what i'm trying to make
<azonenberg_work> considering picking up one to compare performance
<azonenberg_work> They claim 2 pF input capacitance which sounds awfully high