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
<azonenberg> The order of v0.3 probe boards got here
<azonenberg> The new MMCX footprint looks much nicer than the old one
<azonenberg> I'm going to solder one up with the five resistors in a few, then try with no cap and with a discrete compensation cap
<azonenberg> the distributed cap is on the v0.4 board which is still at fab
m4ssi has joined #scopehal
<azonenberg> hmmm
<azonenberg> There is still some other effect going on that i am not modeling
maartenBE has quit [Ping timeout: 245 seconds]
maartenBE has joined #scopehal
m4ssi has quit [Remote host closed the connection]
bvernoux has joined #scopehal
<azonenberg_work> bvernoux: did you see my tweet thread from yesterday?
<bvernoux> ha yes
<bvernoux> very interesting
<azonenberg_work> That weird step we've been tracing is an impedance mismatch for sure
<azonenberg_work> but why the heck is the observed impedance not matching everything i calculate it should be??
<bvernoux> I do not understand how do you have checked that you have Impedance of 30 Ohms in fact
<bvernoux> instead of 50Ohms expected ...
<azonenberg_work> I don't know for sure that I do
<bvernoux> as it is very strange if it is true
<azonenberg_work> What I can say is, the shape of the waveform is consistent with a length of 30-ohm trace equal to the length of my pcb from resistor to mmcx
<azonenberg_work> followed by a 50 ohm line to the scope
<bvernoux> do you have some capture with the probe and without on same signal with scope ?
<azonenberg_work> additionally, when i machine away part of the trace to increase the impedance
<azonenberg_work> the height of the step was reduced
<azonenberg_work> Those are experimental results I can report as fact
<bvernoux> maybe the most impact is on inductance/capacitance too ...
<azonenberg_work> I am not confident enough to say that the actual impedance is indeed 30 ohms
<bvernoux> as I doubt it do so much difference on impedance
<azonenberg_work> yellow = probe, pink = sma feedthrough
<bvernoux> they are not in phase too
<bvernoux> they are intended to be synchronized ?
<azonenberg_work> That's because the cables are equal in length
<bvernoux> ok
<azonenberg_work> but the probe adds another ~70 mm
<bvernoux> so yellow is the new probe and the other is PicoProbe ?
<azonenberg_work> no
<azonenberg_work> i dont have a pico probe
<azonenberg_work> pink is a direct SMA to my scope
<azonenberg_work> with no probe
<bvernoux> ha ok
<azonenberg_work> i'm probing a tiny test PCB with two back to back edge launch SMAs
<azonenberg_work> and the probe tip is going between sma connector ground and pin
<bvernoux> what is very nice is the amplitude
<bvernoux> so the resistor does not impact too much measurement about that
<azonenberg_work> i mean i have the scope calibrated for 10:1 attenuation
<azonenberg_work> but yeah
<bvernoux> or something like 10mV max
<azonenberg_work> The main problem is that 50 mV hump
<azonenberg_work> it needs to go
<bvernoux> do you have eye diagram for both to check ?
<bvernoux> as it seems clearly better than previous probe
<azonenberg_work> i have done eye plots
<azonenberg_work> the new probe is a bit better but not by much
<bvernoux> yes the bump could be removed I think
<azonenberg_work> however, when i did the milling to narrow the center conductor
<azonenberg_work> the hump height was decreased by 7%
<bvernoux> very interesting
<bvernoux> 03 is the latest modification ?
<bvernoux> as it clearly improve the signal
<azonenberg_work> Yeah
<azonenberg_work> the high frequency rolloff i saw on the 1-0-1-0 pattern was reduced
<azonenberg_work> the hump height on the rising edge was reduced
<bvernoux> I suspect capacitance and maybe other parameters are improved
<azonenberg_work> This is my main evidence pointing to the impedance being wrong
<bvernoux> inductance/impedance ...
<azonenberg_work> My design parameters here were 125um gap to top-layer ground, 1.17mm gap to inner layer ground, and 1.1mm wide center conductor
<azonenberg_work> Later simulations suggested 1.0mm would be closer to 50 ohms but it was too late and the pcb was at fab already
<azonenberg_work> My latest version that i sent out last night reduces the center conductor to 0.75mm and increases the gap to 200um
<azonenberg_work> plane spacing stays the same
<azonenberg_work> Sonnet models that as having a 60 ohm impedance
<azonenberg_work> vs the ~45 of what i built this time
<bvernoux> maybe Sonnet is not very good to compute that strange
<azonenberg_work> i deliberately went pretty aggressive with the impedance bump to make sure i see results
<azonenberg_work> it might be an overshoot vs ideal match
<bvernoux> or there is other parasitics things not taken into account
<azonenberg_work> So, the two possibilities i'm considering is that there's some other parameter i didnt model
<azonenberg_work> or that you need a thick-metal model to accurately model GCPWs
<bvernoux> we will have the truth with S-parameters ;)
<azonenberg_work> the free edition of sonnet does not support thick metal
<azonenberg_work> when i buy the full version that won't be a problem
<bvernoux> ha ok
<bvernoux> you plan to buy the full version
<bvernoux> ?
<azonenberg_work> I thought that a 30um conductor with 125um gap would not have major thickness variations
<bvernoux> yes me too I was not thinking it will change so much things
<azonenberg_work> Yes... most likely the "gold" version which has a 2GB ram cap, allows 3 conductor layers, unlimited ports, unlimited ideal components, thick metal, dxf import/export, etc
<azonenberg_work> MSRP is 12.5 kUSD however the sales guy said that's negotiable for small businesses
<bvernoux> especially here we have something like 500MHz Signal about 2ns ...
<azonenberg_work> bvernoux: well the actual signal has much higher freq components
<azonenberg_work> it's a SGMII idle pattern on a 7 series GTP with a ~40ps rise time
<azonenberg_work> the 1 GHz scope front end is rounding that off
<bvernoux> ha ok so it is more 1GHz in fact
<azonenberg_work> there is definitely frequency content out well past 1 GHz
<azonenberg_work> I almost certainly cannot afford the "pro" version of sonnet which is $CALL_FOR_PRICE
<azonenberg_work> "level 3 gold" is in the same price range as my new scope
<bvernoux> maybe Sonnet guys can make a huge effort to support open source things ;)
<azonenberg_work> And if i can get them to take a bit off for me being a one-man shop it will be within the plausible range
<azonenberg_work> Lol
<bvernoux> with a special price
<azonenberg_work> We'll see, i'm going to be getting a 30-day demo of pro in august-ish when i'm moved into the house
<azonenberg_work> so that will give me a chance to run some much more detailed analysis
<azonenberg_work> The other option i have available, should it be necessary, is to buy a 30-day timeboxed version of pro (the normal license is perpetual for that version)
<bvernoux> yes maybe you could wait a bit I do some measurements
<bvernoux> to start the 30days trial ;)
<bvernoux> as we will have more details on what happen
<azonenberg_work> Which is 1.5 kUSD
<azonenberg_work> the way i see it, if i use it for more than a few months then i'm better off buying a perpetual license of gold unless i actually need the pro features
<bvernoux> yes
<azonenberg_work> Any edition below gold is pretty useless to me with the tiny ram caps
<azonenberg_work> And nothing would stop me from buying a short term pro license on top of my copy of gold if i needed the extra features for a short time
<bvernoux> I think you plan to simulate lot of other RF things too the hard things it to group all in the 30days window ...
<azonenberg_work> Well yeah
<azonenberg_work> as much as i would love to use OpenEMS for all this work
<azonenberg_work> it's just not mature enoguh
<azonenberg_work> the UI isnt there
<bvernoux> yes OpenEMS is too much coding ...
<azonenberg_work> (literally, the UI is nonexistent - you have to write python to do anything)
<azonenberg_work> maybe in 5-10 years it will
<azonenberg_work> but right now it simply isnt an option
<bvernoux> yes and it is prone to lot of trial and errors ...
<azonenberg_work> Anyway, so my current plan is to wait 2-3 weeks until the latest probe iteration comes back from fab
<azonenberg_work> Test that
<bvernoux> I started checking what we can do but like you when I have seen the python script ...
<azonenberg_work> then probably send you one of those boards for VNA analysis
<azonenberg_work> along with the other pile of stuff i have queued up for you
<bvernoux> anyway congrats for the latest update and very good results
<azonenberg_work> I keep thinking i'm ready for you to do the measurements
<azonenberg_work> then the board comes back and i find problems :p
<bvernoux> yes no problem I'm not limited by time to do measurements and even try things on the boards to see if that improve things
<azonenberg_work> I can't believe i just sent out my fifth respin of a board that is literally a resistor and a coax connector
<azonenberg_work> lol
<bvernoux> yes but just res & coax change lot of things ;)
<bvernoux> I was clearly not thinking it will have such big impact especially on freq < 1GHz ...
<bvernoux> and it is very good to have an excellent probe for > 1GHz
<bvernoux> do you have checked the probe tip sold by Pico ?
<bvernoux> they are quite cheap in fact I will buy some to check if that change anything
<bvernoux> gold plated Pico probe tips
<bvernoux> I plan to buy 1 for test for 33euros ;)
<bvernoux> if that improve something it is not too expensive
<bvernoux> as the PicoConnect seems to be clearly just plastic probe with well tuned resistor with coax connector ;)
<bvernoux> especially when I see your results
<azonenberg_work> i think the pico's use a printed resistor
<azonenberg_work> i have not tested the pico probe tips yet
<bvernoux> yes it seems printed resistor could be last step
<azonenberg_work> i already rfq'd my fab
<azonenberg_work> 3 kUSD MOQ
<bvernoux> anyway the hint with multiple resistor seems also very good
<azonenberg_work> if i do it, it will be on a volume order or something
<azonenberg_work> yes
<azonenberg_work> thats definitely a step up
<azonenberg_work> especially at >1 GHz
<bvernoux> 3KUS MOQ for printed resistor ?
<azonenberg_work> if i understood correctly yes
<bvernoux> woo it is expensive :(
<bvernoux> and for lot of PCB ?
<azonenberg_work> and this was a chinese fab that is normally quite affordable
<azonenberg_work> I didnt even ask after that
<azonenberg_work> lol
<azonenberg_work> its not happening
<bvernoux> yes
<azonenberg_work> not unless i do a kickstarter or something for a volume order
<azonenberg_work> way too expensive for a high-risk prototype
<azonenberg_work> if i was going to make thousands or hundreds and sell them, sure
<azonenberg_work> but not unless i know the probe is otherwise going to work
<bvernoux> yes
<azonenberg_work> my plan is to prototype on oshpark until i get it working
<azonenberg_work> then do a small volume order at Multech with impedance control and tighter tolerances, maybe sell a few prototypes of that
<azonenberg_work> Then consider making higher volume if there's demand
<bvernoux> yes very nice plan
<azonenberg_work> The other thing i need to do is spend more time thinking about the enclosure
<azonenberg_work> the current enclosure needs some index features ot keep the halves from sliding when i'm gluing it shut
<azonenberg_work> to*
<bvernoux> will be amazing to have some good quality probe for at least 5x time less money (or even 10x time if possible) than PicoProbe ;)
<azonenberg_work> the PCB provides stability in two axes
<azonenberg_work> but the third axis has some wiggle room
<bvernoux> I have a fab which do very nice Aluminium custom case if you are interested
<bvernoux> at an amazing price ;)
<bvernoux> especially for low or big volume
<azonenberg_work> I'd be concerned about coupling effects
<azonenberg_work> so i was going to do nonconductive polymer
<bvernoux> yes it is why aluminum could be problematic ...
<azonenberg_work> My prototype enclosures are all SLS nylon
<bvernoux> also nylon shall be anti static if possible
<bvernoux> i do not know properties of SLS nylon
<azonenberg_work> let me measure :p
<bvernoux> anyway we can quickly see if the enclosure has a bad effect ;)
<azonenberg_work> it's carbon doped so it absorbs the laser
<azonenberg_work> But i dont know if it's conductive from that or not
<azonenberg_work> Definitely not conductive whatsoever
<azonenberg_work> offscale >10M ohms across even a short span
<bvernoux> it seems there is solution for that
<bvernoux> ABS-ESD7 ...
<azonenberg_work> Nice, good to know
<azonenberg_work> For prototypes i'm not too worried
<bvernoux> yes
<azonenberg_work> anyway we can always consider a metal enclosure down the road for shielding
<azonenberg_work> But my concern would be coupling noise into the ground from holding it
<azonenberg_work> i think if anything, plastic with a metal-paint liner or something
<azonenberg_work> or dissipative plastic etc
<bvernoux> yes it is for final step ;)
<bvernoux> it is just good to have it in mind to avoid ruining efforts because of an enclosure
<bvernoux> especially if the probe can go up to 6Ghz ;)
<bvernoux> on par with PicoProbe
<bvernoux> for me even just 4GHz will be amazing
<bvernoux> mainly to check If I can do VNA measurements of antenna ... with this probe
<bvernoux> without too much variations in measurements
<azonenberg_work> Yeah
<azonenberg_work> my qucs sim of the current probe suggests flat +/- 0.25 dB out to 2.5 GHz
<azonenberg_work> +/- 1 dB out to around 4
<azonenberg_work> and -3 dB rolloff just shy of 5 GHz
<azonenberg_work> so all i need to do is make reality match that :p
<bvernoux> yes
<azonenberg_work> Because that is performance i'd be quite happy with
<azonenberg_work> i think going past that will need printed resistors
<azonenberg_work> But i need to build freesample at that point because even if i make a 10 GHz probe i have no way to use that performance :p
<bvernoux> yes
<bvernoux> next step is to have a way to validate the 10GHz BW ;)
<bvernoux> I have found some good scope on Ebay up to 40GS
<bvernoux> but without any probe ;)
<bvernoux> so clearly good cheap "enough" probe is a must have
<azonenberg_work> high sample rate != high bw
<azonenberg_work> my lecroy is 20 Gsps realtime / 200 Gsps ETS
<azonenberg_work> but only 1 GHz bw
<bvernoux> yes
<bvernoux> in theory 20GSPS shall be good with 2GHz BW
<azonenberg_work> yeah, lecroy makes this same scope with 2.5 GHz bw
<azonenberg_work> there's also a 4 GHz version but i think that's 40 Gsps
<bvernoux> I'm pretty sure they just sw lock other version ;)
<bvernoux> with a SPI/I2C filter ...
<azonenberg_work> very likely
<azonenberg_work> the 40 Gsps is a hardware mod i'm sure
<azonenberg_work> they probably DNP half the ADCs
<azonenberg_work> then the MSO mode is a hardware mod too, unlike hte wavesurfer 3000 series
<azonenberg_work> there isnt even a connector for it
<bvernoux> ha bad :(
<azonenberg_work> anyway, once i build freesample i probably wont be buying any new scopes for a bit
<azonenberg_work> i want a decent AWG soon though
<azonenberg_work> Know of any DC - 1 GHz AWGs that are sanely priced? :p
<bvernoux> they are crazy expensive :(
<bvernoux> I do not understand why measurement instruments are always so expensive with spec which have do not changed since 1980 for some ...
<bvernoux> example of AWG or even VNA ...
<bvernoux> or Scope ;)
<azonenberg_work> Lol
<azonenberg_work> yeah, i just dont have the time to replace every piece of test equipment from scratch
<azonenberg_work> probes and scope software first
<bvernoux> I imagine the WavePulser cost more than 20KUSD
<azonenberg_work> oh easily
<bvernoux> way more ;)
<azonenberg_work> that's why i said i wanted it, not that i was gonna buy it :p
<bvernoux> but yes for a very good AWG with lot of memory for complex signal it is crazy expensive
<bvernoux> as it requires ultra fast DAC with lot of bits and memory ...
<bvernoux> mine is 14bits from 1980
<bvernoux> Agilent E4432B
<bvernoux> but the AWG part is defective ... so I can build only basic signals and anyway complex signal is quite limited by Memory