azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing | https://github.com/azonenberg/scopehal-apps, https://github.com/azonenberg/scopehal, https://github.com/azonenberg/scopehal-docs | Logs: https://freenode.irclog.whitequark.org/scopehal
azonenberg_work has quit [Ping timeout: 256 seconds]
azonenberg_work has joined #scopehal
Degi has quit [Ping timeout: 240 seconds]
Degi has joined #scopehal
electronic_eel has quit [Ping timeout: 260 seconds]
electronic_eel has joined #scopehal
juli965 has joined #scopehal
<azonenberg> LeoBodnar: ping
<azonenberg> So I see that the LeoNTP device has a single BNC output with PPS, 1 MHz, or 10 MHz outputs
<azonenberg> There's no way to get both out of a single unit?
<azonenberg> is that a limitation of your circuitry or the GPS module? And is the 10 MHz straight off the GPS module or filtered via some kind of PLL etc?
<azonenberg> it seems like right now if i want stratum 1 NTP, PPS, and a frequency reference that I will be best off using a LeoNTP for time+PPS and a GPS-CLOCK for the 10 MHz?
<azonenberg> Also yay it looks like the probe PCBs are coming early, expected today some time
<_whitenotifier-f> [scopehal] azonenberg opened issue #280: Allow sample rate/memory depth of SignalGeneratorOscilloscope to be changed - https://git.io/JUPVO
<_whitenotifier-f> [scopehal] azonenberg labeled issue #280: Allow sample rate/memory depth of SignalGeneratorOscilloscope to be changed - https://git.io/JUPVO
<_whitenotifier-f> [scopehal] azonenberg pushed 1 commit to master [+0/-0/±2] https://git.io/JUPwe
<_whitenotifier-f> [scopehal] azonenberg 9f362a5 - Refactored PeakDetectionFilter out to PeakDetector + PeakDetectionFilter
<_whitenotifier-f> [scopehal-apps] azonenberg pushed 1 commit to master [+0/-0/±2] https://git.io/JUPwT
<_whitenotifier-f> [scopehal-apps] azonenberg 8d3308a - Now render PeakDetector stuff even if not a PeakDetectionFilter
<_whitenotifier-f> [scopehal] azonenberg pushed 1 commit to master [+2/-0/±3] https://git.io/JUPoS
<_whitenotifier-f> [scopehal] azonenberg f39cd4e - Tek 5/6 spectrum view channels now display peaks. Hard coded to 10 peaks at 1 MHz minimum spacing for now. See #269.
<_whitenotifier-f> [scopehal] azonenberg edited issue #269: Support for Tek 5/6 series spectrum channels - https://git.io/JUaaE
juli965 has quit [Quit: Nettalk6 - www.ntalk.de]
<azonenberg> lain, monochroma, miek: So at this point i have most of the stuff need to view spectrum channels working
<azonenberg> But I don't have APIs for changing things like RBW, center frequency, span, etc
<azonenberg> The question is how we should fit that into the overall instrument object model
<azonenberg> In particular, how reasonable is it for a specan to be derived from Oscilloscope?
<azonenberg> Or in general how do we want to handle the case of things that are simultaneously specans and scopes?
<azonenberg> Conceptually, a spectrum analyzer is (to me) an oscilloscope where the X axis has units of frequency rather than time
<azonenberg> or more precisely an oscilloscope where at least one channel's X axis
<azonenberg> Note also that a RTSA may have the ability to give you time domain baseband I/Q data
<azonenberg> So the line between specans and scopes is not at all sharp
<azonenberg> also poke noopwafel, tnt, kc8apf)
<kc8apf> that's not entirely wrong. I'd need to think through how RBW effects on data map to something similar in an oscope.
<azonenberg> Well there would have to be new APIs for setting center frequency, span, and RBW
<azonenberg> The question is should we add these to the oscilloscope class or make a new one?
<azonenberg> there's no clear answer here
<azonenberg> No matter what, if we have a combined device like a tek 6 series, there will be cases where calling some APIs on some channels make no sense
<kc8apf> hadn't seen LeoNTP. Looks nice for the price. I want something with multiple outputs though. I want PPS, 10MHz, and adjustable output
<azonenberg> (internally I treat CH1 and CH1_SPECTRUM as totally separate channels, because other than sharing a frontend they have basically nothing in common)
<azonenberg> kc8apf: so in my case what I want is NTP, PPS, and 10 MHz with as few unique instruments as possible and without paying a fortune
<azonenberg> I think a LeoNTP for NTP+PPS and his GPSDO for 10 MHz is probably the best combination right now
<kc8apf> is the oscilloscope class intended to represent an actual oscilloscope or a more generalized display with X-Y charting?
<azonenberg> And that's the million dollar question :)
<azonenberg> What makes the most sense at an API level? Keeping in mind again that the lines between different types of instruments are not at all clear
<kc8apf> there are plenty of other devices that tend to use X-Y charting output
<azonenberg> Consider that some modern multimeters can have kHz to MHz sample rates, for example - it might make sense to have THEM represented as oscilloscopes too
<azonenberg> what if you want to do ffts and math functions on a fast DMM output?
<azonenberg> ultimately i think it is, no pun intended, a spectrum
<azonenberg> And a unified representation might make the most sense
<kc8apf> I expect trying to name things after physical equipment will always cause problems. Separating the concept of visualization from acquisition and analysis leads to much more generic naming
<azonenberg> The class will remain named Oscilloscope since that's what it mostly is used for
<kc8apf> what makes it an oscilloscope versus anything else?
<azonenberg> I'm thinking more at an API level. it doesn't matter what you call the class
<azonenberg> does it make sense for scopes and specans to be unified?
<azonenberg> Because i am seeing some convergent evolution happening here. Ultimately scopes, LAs, and specans are "things that take in a bunch of electrical data at high speed and plot it"
<azonenberg> A LA is conceptually a DSO that has 1-bit resolution, using comparators instead of ADCs for sampling
<azonenberg> A sampling scope isn't realtime but outputs either a time domain waveform or an eye pattern
<azonenberg> a specan can output either time domain I/Q or a frequency domain "waveform"
<azonenberg> I don't think there are any lines to draw that make sense
<kc8apf> At the API level, is this class what handles rendering of data or something more?
<azonenberg> No. Rendering is done by the WaveformArea class
<azonenberg> in glscopeclient
<azonenberg> libscopehal has no rendering at all, the Oscilloscope class is purely for acquisition
<azonenberg> New probe board came in
<azonenberg> probably going to populate and test over lunch
<kc8apf> For acquisition, I'd probably focus on classes that implement one channel. In that case, I'd have a Oscilloscope, LA, Spectrum Analyzer, etc classes.
<kc8apf> mostly due to the large amount of commonality in implementing specific implementations of each of those but also the moderate amount of non-overlap between them
<azonenberg> kc8apf: That doesn't make sense because of things like timbase and trigger config
<azonenberg> more importantly, the Oscilloscope-derived class is normally where the driver code that interfaces to the specific instrment is
<azonenberg> I don't want to have device-specific code in more than one class
<kc8apf> MSOs already start to break that model
<azonenberg> So LeCroyOscilloscope is the bridge from the Oscilloscope class API to the LeCroy SCPI protocol, etc
<azonenberg> As far as i'm concerned a MSO is just a DSO where some channels have lower bit depth than others
<azonenberg> and controls like gain/offset are no-ops
<azonenberg> and that's why i'm thinking adding frequency domain stuff to it isn't much of a stretch
<azonenberg> Each Channel has a Waveform object associated with it, and can have arbitrary X/Y units
<azonenberg> So having a Channel that has X=freq, Y=amplitude units is completely reasonable
<azonenberg> Since a Filter is a Channel that's not attached to an instrument, in fact, FFT's are already represented this way
<azonenberg> so the only difference is that now there can be frequency domain Channel's that are part of an instrument
<azonenberg> i have this now for the mso5/6, and it works. I just have to add commands for center/span and RBW i think
<kc8apf> sounds like you know what you want to do
<electronic_eel> are there kinds of instruments that aren't derived from this oscilloscope class?
<azonenberg> electronic_eel: so far power supplies and waveform generators
<electronic_eel> so why not call it acquisition_instrument instead of oscilloscope?
<azonenberg> however some may derive from both, like a scope with an integrated AWG. But these are fundamentally separate instruments in the same body
<azonenberg> oh and also (for now) multimeters
<azonenberg> VNAs are likely going to be different too because they have basically nothing in common with the others
<electronic_eel> why are multimeters different than scopes?
<azonenberg> That's why i said "for now"
<electronic_eel> don't they also deliver a datastream of inputs?
<azonenberg> Because i am thinking of unifying them
<azonenberg> Renaming the class is a decision for later
<azonenberg> For the time being the question is whether unifying SAs and scopes under one roof makes sense. ANd i think it does
<azonenberg> and yes actually there has been discussion with some of the xdevs folks about adding support for multimeters under the same framework
<azonenberg> the theory being that a multimeter is just a scope optimized for resolution rather than sample rate
<azonenberg> and with the various biasing etc circuits you need to measure current
<azonenberg> to measure resistance*
<electronic_eel> yeah, but does that matter much for grabbing and interpreting the data?
<azonenberg> well what matters more is how the acqusition model works
<miek> seems like the major distinction is the unit of data returned by the instrument. you've got scopes/specans/etc. that return a waveform, DMMs that usually give you a single value per query, SDRs/LAs that just give you a constant stream of data
<azonenberg> miek: that is kinda how i've been thinking of it too
<azonenberg> e.g. i think right now my R&S meters expect you to read points one at a time and we don't currently have support for roll triggers
<azonenberg> But even these lines are a bit blurry because some meters can return blocks of many samples
<azonenberg> which really doesn't seem any different from a waveform
<electronic_eel> don't better multimeters also have a buffer the sample into that you can read out later?
<azonenberg> Yes
<azonenberg> Which is why i'm saying those are basically just scopes with really high res ADCs and slow timebases
<electronic_eel> that looks very similar to a scope to me
<azonenberg> Exactly
<azonenberg> The xdevs guys were talking about wanting to do sub-Hz frequency FFTs on DMM waveforms
<azonenberg> i think one of them said he had a DMM that had high kHz or even low MHz ADC rates
<azonenberg> That is really just a slow scope
<electronic_eel> your fft algorithm doesn't care about the scale on the frequencies, doesn't it?
<azonenberg> Not... exactly
<azonenberg> the FFT itself does not
<azonenberg> however the X axis units for a Waveform are int64's
<azonenberg> and right now the base unit is Hz, so we cannot represent fractional Hz resolution in a frequency domain waveform
<azonenberg> I have a ticket open for changing the base unit to mHz or uHz to fix this
<azonenberg> Just a constant scaling factor we'll have to change in a few places (the tek MSO driver and the FFT class)
<electronic_eel> ah, that should be fixed for low-freq ffts. also maybe for function generators that usually are also able to do sub-hz stuff
<azonenberg> Yeah. Right now the base units are ps for time and Hz for freq
<azonenberg> most other units like dBm and volts are only really used for the Y axis at the moment and the Y axis for analog waveforms is 32-bit IEEE754 float
<azonenberg> so the choice of base unit doesn't matter much. But X axis units are fixed point because the data is discretely sampled at fixed intervals and you really want integer indexes you can iterate over
<electronic_eel> yes, having the x axis also in float would make it much slower
<azonenberg> We may have to revisit some of these design choices when we add X-Y plots though
<azonenberg> plotting volts against volts
<azonenberg> That architectural decision is the biggest blocker to proper x-y
<azonenberg> But that is not a super high priority right now
<azonenberg> Re changing the base unit to sub Hz, low frequency FFTs is the driving use case. But it will help with func gens for sure too
<azonenberg> in fact, i think my R&S PSUs can do really low freq AWG modulation
<azonenberg> for simulating weird ramp curves etc
<azonenberg> probably only a few hundred Hz given bypass capacitance etc
<azonenberg> But that is a feature the scopehal driver does not yet support
<electronic_eel> so you plan to have multimeters, power supplys, function generators and so on in glscopeclient - how about an interface to easily script test scenarios without having to recompile the whole thing?
<electronic_eel> as a replacement to the overpriced SMUs?
<azonenberg> That would be a nice long term function. Right now you have to write such stuff in C++
<azonenberg> in the examples directory there's a trivial curve tracer based on my R&S DMM and PSU
<azonenberg> sweep psu in 1 mV increments and measure current with the DMM to get higher resolution than the PSU
<azonenberg> then stop when you hit i think 20 mA or 5V
<electronic_eel> yes, exactly this kind of stuff
<azonenberg> Eventually either python bindings or a custom domain specific scripting feature, or both, would be nice
<electronic_eel> finally kill off labview
<azonenberg> also to be clear that will be in libscopehal
<azonenberg> glscopeclient, at least for the near term, is intended to only be used for instrments that output waveforms
<azonenberg> there will be separate UIs under scopehal-apps for things like PSUs
<azonenberg> i think there is already an incomplete psuclient app
<azonenberg> just a GTK dialog basically
<electronic_eel> hmm, why not integrate it into glscopeclient?
<electronic_eel> to get stuff like xy plots
<azonenberg> i have very basic Cairo based plotting, actually, already
<azonenberg> in psuclient
<azonenberg> But it would be interesting to have it integrated. I havent made a final decision yet. psuclient is pretty trivial right now and could be easily replaced
<azonenberg> My ultimate goal is for libscopehal to kill VISA though
<electronic_eel> yes, kill VISA and kill labview
<azonenberg> well as i see it libscopehal is the VISA killer and scopehal-apps is the labview killer
<electronic_eel> then scopehal-apps or glscopeclient also needs a scripting interface, not only libscopehal
<electronic_eel> because you want to control the graphical output to replace labview
<azonenberg> Hmm, yeah
<azonenberg> We'll have to think about that long term. There's a lot more basic infrastructure to work on first
<electronic_eel> yes, of course
<azonenberg> BTW, what is your current status wrt using glscopeclient? you said you got it compiling and running?
<electronic_eel> no, I'm finishing a pcb project of mine first. will get glscopehal to work when that is finished
<azonenberg> also longer term we might want a better name for glscopeclient. It's a very descriptive name that might be better off replaced with something more memorable :p
<azonenberg> it sounds like a unix utility lol
<azonenberg> Which isn't entirely untrue
<electronic_eel> glscopeclient | grep fft
<azonenberg> lol
<azonenberg> V1.2 probe is cooling after reflow now
<azonenberg> gotta hand solder the connector and then i can test :D
bvernoux has joined #scopehal
<azonenberg> bvernoux: perfect timing
<azonenberg> just finishing up soldering the v1.2 probe
<azonenberg> throwing it on the VNA momentarily
<bvernoux> ha nice
<bvernoux> it is like I have read your mind ;)
<bvernoux> I was checking the VNA2 ;)
<bvernoux> the guy has really done amazing job
<bvernoux> the Qt GUI is very nice
juli965 has joined #scopehal
<azonenberg> Soldering complete, on to test
<bvernoux> great
<bvernoux> do you have soldered the new connector for RF up to 26.5GHz ?
<bvernoux> similar to the one I'm using ?
<bvernoux> SMA
<bvernoux> as it clearly heavily improve transition from SMA to PCB
<bvernoux> But maybe not for the probe
<azonenberg_work> Yes that is what i am using for the probe
<bvernoux> ha great
<azonenberg> holy moley
<azonenberg> Purple = kickstarter version
<azonenberg> Blue = last prototype
<azonenberg> Red = latest prototype
<miek> nice!
<miek> was this on rogers or 408?
<azonenberg> It's not perfectly flat with the new filter but it's the best i have to date by FAR. And it's better than -3 dB to 5.54 GHz
<azonenberg> This is oshpark. not even impedance control
<miek> awesome
<azonenberg> With de-embedding i think this is a very usable 5 GHz probe
<azonenberg> Even without, it's not BAD but you will have a dB or two of gain error here and there
<azonenberg> Do you think maybe i overcorrected? should i try one more rev with the filter slightly weaker?
<azonenberg> or is this good as is
<azonenberg> I'm honestly tempted to leave it
<azonenberg> This is going to be hard to improve on
<bvernoux> great
<miek> i dunno, it's pretty darn good as-is. you could maybe tone the filter back a tiny bit to be flatter on average but it's probably not worth it
<azonenberg> Yeah. Next step is going to be testing S11 without the terminator
<azonenberg> see how the input loading looks
<azonenberg> then having a look at the rise time
<bvernoux> very nice it reach more than 5GHz
<azonenberg> Yeah. I think mission accomplished lol
<bvernoux> yes clearly
<bvernoux> and very linear +/-1dB !!
<azonenberg> I dont think it's quite as flat as the >$1K Pico probe but it's very close
<bvernoux> it is only 1 or max 1.5dB ;)
<azonenberg> Yeah. nothing you can't de-embed
<bvernoux> maybe the Pic probe is worse ...
<azonenberg> o
<bvernoux> Pico
<azonenberg> i'll check
<bvernoux> very nice improvement vs the Blue
<bvernoux> I do not even speak about the pink ;)
<bvernoux> pink is 1st version IIRC ?
<bvernoux> could you take a photo of the setup to measure it with the Pico
<bvernoux> ?
<azonenberg_work> Pink is the kickstarter version
<azonenberg_work> i'll do the pico probe shortly
<bvernoux> and it is just an OSHPark PCB
<bvernoux> amazing
<bvernoux> not even RO4350B ...
<azonenberg_work> grr probe moved during the zin measurement
<azonenberg_work> gotta red oit
<bvernoux> yes it is interesting to compare different probe with exactly same setup
<azonenberg> https://www.antikernel.net/temp/zin-v1p2.png this is S11 without the terminator
<azonenberg> https://www.antikernel.net/temp/zin-v1p2-2.png and input impedance
<azonenberg> i think i'm probably still more capacitive than i should be
<azonenberg> That's *nine ohms* at 1 GHz impedance
<bvernoux> yes and also at 5GHz (or maybe 20)
<bvernoux> the design is not with 50Ohms impedance for OSHPark ?
<bvernoux> as IIRC it was design for RO4350B ?
<azonenberg> Er of FR408HR and RO4350B are within a fraction of a percent
<azonenberg> they're almost perfect substitutes except 4350 has lower loss
<bvernoux> yes but layer has not same height ?
<bvernoux> or do you have chosen the same for RO4350B
<azonenberg> I'm using the same stackup as oshpark now for the rogers probes
<bvernoux> as there is lot of choice
<bvernoux> ha ok
<bvernoux> so it will be interesting to see what RO4350B will improve (or degrade who know)
<azonenberg_work> I am actually considering a production run on FR408HR at this point
<bvernoux> the most impact is towards ENIG
<azonenberg_work> And ENIG
<azonenberg_work> there's no reason to use a lower loss material now
<bvernoux> yes I doubt there is any advantage to test RO4350B + Silver ...
<bvernoux> as anyway Silver is not good in time ...
<bvernoux> with oxidation
<bvernoux> compared to ENIG
<bvernoux> ISIG will be a must but OSHPARK does not support it ;)
<bvernoux> I have rarely found fab supporting it anyway
<azonenberg> Wow
<azonenberg> This is interesting
<azonenberg> this is my probe vs the pico 921 on the same test fixture. The 6 dB vertical offset is because the 921 is a 20x probe and I'm a 10x
<azonenberg> They win on flatness out to about 2.5 GHz but i'm much better after that
<miek> huh, that step is weird. i wonder what causes that
<azonenberg> Good question
<azonenberg> but they win on return loss. They're much less invasive than me so i probably am too capacitive still
<bvernoux> haha yes pico is worse ;)
<azonenberg> they don't get below about 250 ohms input impedance
<bvernoux> I told you ;)
<azonenberg> while at the same 2 GHz point i'm around 10 ohms
<azonenberg> So basically at this point i have better S21 and they have better S11
<bvernoux> pico is good up to 3.5GHz ...
<azonenberg> So what i'm wondering is where is my big 2 GHz resonance coming in, and how can i get rid of it?
<bvernoux> maybe with rf absorber
<bvernoux> it is with the case ?
<azonenberg> no this is uncased
<azonenberg> Possibly related, when i simulate the simplified netlist model i get a ~3 GHz resonance that drops the input impedance down to about 100 ohms
<azonenberg> Which makes me think it's the tip needle itself
<bvernoux> you are not using same tip as Pico probe ?
<azonenberg> No i'm using a PMK tip. but i meant more like the whole assembly of tip to input
<azonenberg> maybe i need some kind of compensation or something, idk
<miek> that's what i'd suspect, all the other probes i see go to great effort to put a resistor as close to the tip point as possible to put a hard limit on impedance
<azonenberg> miek: I have a resistor as close to the tip as i can
<azonenberg> i think this resonance is before that
<azonenberg> i think it's the tip itself
<miek> yeah, but it's still far away relative to a lot of other probes
<azonenberg> what??
<azonenberg> that's the resistor placement vs the tip socket
<azonenberg> i literally cannot make it any closer
<bvernoux> cut the tip to test ;)
<miek> i know you can't, but i mean other probes embed a resistor into the needle
<azonenberg> The pico does not
<azonenberg> it's a replaceable tip that looks very similar to mine
<miek> isn't it a lot shorter? i'm just going off internet pics of it
<bvernoux> could be interesting to test their tip if it can be soldered just to check
<bvernoux> the ground plane is square
<bvernoux> it can do bad stuff
<bvernoux> on bottom
<bvernoux> and on top you could add GND too
<bvernoux> that could improve a bit the impedance
<azonenberg_work> i removed a lot of round metal near the tip to lower capacitance
<miek> bvernoux: pulling back all that metal was part of why it's doing so much better than the KS version :p
<azonenberg_work> actually no
<azonenberg_work> it seems that had minimal effects either way
<azonenberg_work> the single biggest gain was from moving the resistors closer together
<azonenberg_work> Also i just looked at the pico tips
<miek> i thought that showed a big change in the sim, separate from the resistors?
<azonenberg_work> They are almost indistinguishable from mine
<bvernoux> ha ok
<azonenberg_work> I think they do something to stop the tip from resonating though
<azonenberg_work> i'm not sure what that would be
<bvernoux> removing the GND near tip provide better results ?
<bvernoux> ok I was not aware of that
<bvernoux> I was thinking it was mainly resistor arrangement / values ...
<azonenberg_work> It seemed to maybe be slightly better
<azonenberg_work> but the resistor arrangement was the big improvement
<bvernoux> ok
<bvernoux> anyway I doubt I can really improve anything except maybe by testing an other tip
<bvernoux> especially the one from Pico ;)
<bvernoux> they sell tips IIRC
<azonenberg_work> ok so time domain results are in
<bvernoux> -I + YOU
<bvernoux> what I will say that the actual performance are already amazing
<bvernoux> it will be interesting if it is stable when assembling 2 or 3 probes
<azonenberg_work> So with the kickstarter version, typical 20-80% rise time is 122 ps, 10-90% is 180ps
<azonenberg_work> With this verison 20-80% is 85 ps, 10-90 is 126 ps
<bvernoux> ha yes
<azonenberg_work> But i'm still not thrilled about the resonance
<azonenberg_work> it's way too much loading for probing signals in the 2 GHz range
<bvernoux> you should test with the case too
<bvernoux> to check if there is an impact on that resonance
<bvernoux> or others ...
<azonenberg_work> I'll worry about that later. It's nonconductive and has a pretty good airgap above the board
<azonenberg_work> It should not impact much at all
<azonenberg_work> i havent seen any in previous testing
<bvernoux> what is the limit of your scope for ps ?
<bvernoux> interally
<bvernoux> ha it is in repetitive mode
<azonenberg> 20-80 is specced at 75 ps, 10-90 is specced at 100
<azonenberg> oh, and this is without cable de-embedding either
<bvernoux> so you are close to the limit of instrument
<azonenberg> Yes
<azonenberg> If i can just get this probe to load the DUT a bit i'll be satisfied
<bvernoux> yes more load on 2Ghz ;)
<bvernoux> it will also correct 5GHz
<bvernoux> the pico921 has exactly same effect in fact
<bvernoux> but with 250Ohm more
<azonenberg> Yeah so i'm thinking they probably did something to damp out that resonance
<azonenberg> on the far side of the tip
<azonenberg> i'm not sure what
<electronic_eel> or is it maybe just that they are a 20x probe?
<electronic_eel> and being a 20x probe just shifts the impedance up
<azonenberg> No. This loading is coming before the resistors
<azonenberg> increasing the resistor value means less DC loading
<azonenberg> This is a combination of input capacitance and some kind of resonances in the tip
<electronic_eel> is there a way to isolate this effect from the rest of the probe? like solder a 50 ohms terminator instead of the resistors?
<azonenberg> I'm thinking more like not solder the resistors at all
<electronic_eel> ok, you get a full reflection then, but you probably still see the effect of the probe
<azonenberg> well the resistor array is 500 ohm impedance
<azonenberg> that's effectively an open circuit anyway
<azonenberg> https://www.antikernel.net/temp/tip-resonance-sim.png this is my simplified netlist sim of the probe including tip
<azonenberg> the resonance is still there, although not quite as deep as it is IRL
<electronic_eel> is the calculated / inverse tdr in your vna fine enough to resolve the point where on the probe a impedance step is?
<azonenberg> No
<bvernoux> I have found this probe
<bvernoux> it is a 5GHz Passive probe
<azonenberg> Congrats, i think you've found the OEM for the LeCroy PP066
<azonenberg> They use Caddock low inductance power resistors, as does the competing version from... i forget if keysight or tek
<azonenberg> that is basically the same thing
<electronic_eel> the ground tip seems to very far away from the center pin
<bvernoux> and they clearly integrate the resistance in the tip ;)
<miek> apparently the lecroy name adds 50% BW :p
<azonenberg> electronic_eel: yes it is. which i find interesting
<azonenberg> miek: lol
<bvernoux> it is fun that you change the tip for 10:1 or 20:1 ;)
<electronic_eel> IIRC your testing showed very reduced bw with the gnd being a bit away as for the pmk accessories
<azonenberg> bvernoux: the "tip" is literally a resistor
<bvernoux> ha ok they have designed specifically
<bvernoux> +it
<azonenberg> no, i believe it is a Caddock MV311 resistor
<azonenberg> custom made for yokagawa in 450 and 950 ohm values
<azonenberg> (the normal series only goes out to 50)
<electronic_eel> the user manual shows how it goes together, they have two models of "resister"
<electronic_eel> 450 ohms an 950 ohms
<electronic_eel> I guess the resistor is not used as tip, but the actual tip is in the probe head you screw on
<electronic_eel> and a bit of magic is probably how far the gnd shield inside the probe head extends out over the resistor
<bvernoux> Pico916 for 132.87USD ;)
<azonenberg_work> I do not believe they extend over the shell at all
<azonenberg_work> over the resistor*
<bvernoux> but 182.98USD Shipping ;)
<azonenberg_work> that adds capacitance
<azonenberg_work> And the resistor is the tip i think, the thing that goes around it is just a plastic retaining clip
<azonenberg_work> at least thats what i suspect
<bvernoux> as anyway the probe act like an antenna depending on the size
<electronic_eel> ok, could also be that they had the resistors manufactured with stiff, sharpened gold leads
<azonenberg_work> electronic_eel: that might be. or maybe they have a tiny little socket that goes over it
<azonenberg_work> I'm not sure
<azonenberg_work> Anyway, i think at this point the next step is to go back to sonnet and start engineering a fix to the tip resonance
<azonenberg_work> I'll work on that after work today
<electronic_eel> unfortunately no s parameter graphs in the datasheet of the yokogawa
<azonenberg_work> yeah i have no idea how good it is. I'm using the pico 921 as the benchmark
<azonenberg_work> because it's the only probe i own right now that outperforms mine
<azonenberg_work> i've totally destroyed the pico ta061 even with the kickstarter version
<azonenberg_work> (it has impedance characteristics comparable to mine)
<azonenberg_work> but i have better s21 now
<azonenberg_work> also wait a minute
<azonenberg_work> how do these make sense
<azonenberg_work> the S11 graph is showing -3 dB S11 at 2 GHz
<azonenberg_work> meaning half of the power is reflected
<azonenberg_work> that doesn't seem to jive with 10 ohm impedance
<azonenberg_work> I feel like i'm misintepreting something
<bvernoux> ye something seem wrong
<azonenberg_work> These are graphs of the same s2p
<bvernoux> could you share the s2p ?
<bvernoux> to cross check with other tool to check the Impedance
<azonenberg_work> https://www.antikernel.net/temp/probe-v1p2-tipground.s2p (with 50 ohm terminator across tip)
<azonenberg_work> https://www.antikernel.net/temp/probe-v1p2-zin.s2p (probing across a coupler with no termination)
<azonenberg_work> But you can clearly see they have much less loading than me
<azonenberg_work> going off S11 only, there's obviously a problem
<bvernoux> probe-v1p2-tipground.s2p do not show any major issue anyway
<azonenberg> That is looking across a 50 ohm terminator though
<azonenberg> At some point soonish i'll do some tests with a thru line measuring loading on the dut
<azonenberg> See how it actually looks with a PRBS or something
<bvernoux> the impedance variation is from 45Ohm to 76Ohm(at 3.8GHz)
<bvernoux> yes it is more interesting with e THRU line
<bvernoux> a
_whitelogger has joined #scopehal
<azonenberg> So the S11 graphs for the simulated probe and the real thing are actually pretty close
<azonenberg> so this does validate my model i think
<azonenberg> I guess the next step will be seeing if i can do anything to reduce the dips
<azonenberg> I confirmed that if i make the tip zero electrical length in the sim the dip goes away
<azonenberg> So it's definitely related to the tip
<bvernoux> hehe
<bvernoux> coudld you try to cut the dip to check ?
<bvernoux> that will solve nothing but that will confirm reality with simulation
<azonenberg> I could try probing with just the tip socket i guess
<azonenberg> but unless i sanded the board down and cut off the socket i couldn't get rid of it entirely
<azonenberg> at this point the sim and real world data match close enough i'm confident it's close
<azonenberg> The tip is something i don't think i can do anything about
<azonenberg> because of mechanical constraints
<azonenberg> so the only question is if i can do something to dampen that resonance out without also destroying S21
<bvernoux> yes very good question
<bvernoux> maybe you can ask to Harmon what he think
<bvernoux> maybe he has a good idea to fix that
<azonenberg> So this is interesting
<azonenberg> It looks like a small resistance between the tip and ground has a significant damping effect at the cost of higher DC loading
<bvernoux> I do not have news from Darell Harmon since lot of time
<bvernoux> I hope he's fine
<azonenberg> but the loading at the resonant peak is much less since it damps out the resonance
<bvernoux> ha nice
<bvernoux> you have soldered it for test with results ?
<azonenberg> Not yet
<azonenberg> Also not sure if i have any 500 ohm FC0402 resistors but i guess i could try with a normal 500 and see
<bvernoux> I think the effect is here https://cdn.tmi.yokogawa.com/1/151/files/701974_L.png
<bvernoux> limited by the tip protection maybe too
<bvernoux> but here the resistor is inside tip IIRC
<bvernoux> but maybe you could add a material around the tip
<azonenberg> Anyway, i think i can actually do better
<azonenberg> gimme a sec, i want to see what happens if i add the ground closer to the tip
<bvernoux> IIRC your tip is "nude" ;)
<bvernoux> and quite long
<electronic_eel> bvernoux: do you think the plastic (?) around the tip is what solves the problem for yokogawa?
<bvernoux> it is maybe not simple plastic ;)
<bvernoux> I just search idea to check ;)
<bvernoux> as anyway with actual tip it is impossible to add resistor on the tip ...
<bvernoux> and it is quite long so it has an effect
<azonenberg> anyway, also interesting: the resistive damping idea also nulls out the peaking
<bvernoux> do you have a measurement and photo to see it ?
<azonenberg> I'm still playing in sim
<azonenberg> buit
<electronic_eel> bvernoux: yes, it could be something more sophisticated. but it is not some simple shielding either, otherwise there would be resonance
<bvernoux> ha ok it is in sim
<bvernoux> electronic_eel, yes shielding will add more issues ;)
<bvernoux> I think to something like rf absorber
<bvernoux> to be checked what the effect is
<bvernoux> as it does miracle in aluminum cavity ;)
<bvernoux> but it is not exactly the same use case here
<electronic_eel> what is also different is that their ground is really far away, so there is no danger of resonance there
<azonenberg> Red is the v1.1 probe, reverting today's changes
<azonenberg> blue is the same probe with 500 ohms from the tip to ground before the resistor string
<azonenberg> Still 4.75 GHz bandwidth
<azonenberg> actually closer to 4.9
<bvernoux> do you have a picture of the simulation how are the res
<bvernoux> ?
<azonenberg> This is an abstract netlist simulation, not full field solver geometry
<azonenberg> whoops
<azonenberg> Reuploaded
<azonenberg> wrong graph before
<azonenberg> So, this is after damping out the resonances with the 500 ohm resistor
<azonenberg> this is using a vishay S-parameter model of a real resistor, not an ideal one
<azonenberg> cuts about 3-4 dB off the resonance
<azonenberg> This gives us 250 ohm loading at ~DC which drops down to ~130 at worst
<azonenberg> so generally a lot flatter
<azonenberg> I think this is going to be the way to go
<azonenberg> It's more DC loading but that shoudlnt be horrible
<azonenberg> i can always make a 20x version if i need to
<azonenberg> Actually now that i think about it
<azonenberg> Pico has to be doing this
<azonenberg> There's no other explanation for how their *20x* probe has 500 ohm loading
<azonenberg> They've got a damping resistor at the tip
<azonenberg> I'm going to try reworking a damping R onto the v1.1 probe and see what happens
<electronic_eel> good that you always get 3 pcbs from osh
<Degi> Huhhh
<Degi> Smooth curve
<miek> finally peeled back all the layers to get the scope firmware out https://i.imgur.com/uOb7SGo.png :D
<Degi> Ohh neat, which scope?
<miek> hp 83480a sampling scope
<Degi> Nice
<azonenberg> Degi: well these are all sims of course
<azonenberg> electronic_eel: i was just gonna bodge onto the assembled one
<azonenberg> not do another
<azonenberg> So it looks like the Pico 923 is the equivalent to what i'm trying to build
<azonenberg> Annoyingly i do not see the FC0402 500 ohm resistor stocked anywhere
<azonenberg> I guess i could do a parallel string of two 1K's
<azonenberg> i think i have some 1K
<bvernoux> miek, nice done with Ghidra ;)
<bvernoux> it looke like it is 68k CPU ;)
<miek> yep, 68020 :)
<bvernoux> hehe exactly the same on my A1200 ;)
<bvernoux> I have programmed it during years ;)
<bvernoux> in asm + C
<azonenberg> here's the layout
<bvernoux> azonenberg, have very interesting just by changing the layout like that change a lot the issue
<azonenberg> This is a v1.1 board, the blue one with lots of peaking
<azonenberg> and damping resistors bodged on
<azonenberg> 1K each
<azonenberg> Firing up the VNA to see how it looks
<bvernoux> yes ;)
<bvernoux> you see VNA is a must have ;)
<bvernoux> before you was not sure to buy one ;)
<bvernoux> azonenberg, you could be interested by that thread https://github.com/jankae/VNA2/issues/2
<bvernoux> Aim is to build a nice 6GHz VNA fully open source
<bvernoux> but also a 2nd version up to 13.6Ghz ;)
<azonenberg_work> I'm actually drooling over the copper mountain VNAs now lol
<Degi> Copper mountain?
<azonenberg_work> they make rackmountable versions which is handy
<bvernoux> I have read very bad things about their VNA ;)
<azonenberg_work> 2 or 4 port out to 20+ GHz
<azonenberg_work> bvernoux: oh?
<bvernoux> yes
<azonenberg_work> the specs looked nice and their software runs on linux
<Degi> ECP5, nice"!
<azonenberg_work> it can't be worse than pico :p
<bvernoux> yes but IIRC they suxx
<bvernoux> anyway maybe it is only their "low cost" version too
<bvernoux> let me find the link
<azonenberg_work> The one i actually want is something like 50 kUSD
<azonenberg_work> Which is why i don't have it
<azonenberg_work> :p
<bvernoux> ha yes ;)
<bvernoux> Maybe Harmon will sell one less expensive
<bvernoux> he was planning a VNA 20GHz+
<azonenberg_work> Yes
<azonenberg_work> I want it
<bvernoux> he told me he has a 26.5GHz version ;)
<bvernoux> and it could be something like 10KUSD
<bvernoux> will be clearly a game changer
<Degi> Huhh
<Degi> Will it have like YIG filters?
<bvernoux> it will be compact
<bvernoux> something like 20cm x 30cm
<bvernoux> so it will requires ultra small YIG filters ;)
<bvernoux> Interesting point => Lower frequency models are fanless, all are very quiet.
<bvernoux> as my HP VNA is awfull ;)
<bvernoux> the same for my Agilent SigGen all do noise !!
<azonenberg> Ok so, with the damping in place
<azonenberg> the v1.1 probe has a little bit of waviness in the response but is -21 +/- 1 dB to 2.9 GHz
<azonenberg> then it dips down to -22.5 then back up again
<azonenberg> The -3 dB b/w is somewhere around 4.8-5 GHz
<azonenberg> So it did definitely help with the resonance
<azonenberg> now to see what happened with the input impedance...
<azonenberg> aaaand the resonance is actually worse??
<azonenberg> but the input impedance curve looks better
<azonenberg> now i'm even more confused
<bvernoux> do you have picture ?
<azonenberg> Blue is the reworked board with damping resistors
<bvernoux> yes it is better except between 2.8Ghz to 4.6Ghz
<azonenberg> Yeah but the S11 looks worse which confuses me
<bvernoux> I still have pain to see correlation between Impedance and S11 ;)
<azonenberg> this is S21
<azonenberg> red = v1.1 board, pink = v1.2 board, blue = v1.1 plus damping resistor
<bvernoux> yes it is strange it is so bad after 4GHz
<azonenberg> Well more to the point, the damped version has those weird oscillations in it
<azonenberg> But it's also higher amplitude in general than the version iwth the filter
<azonenberg> up to 4 GHz at least, which is about what i am targeting as my upper bandwidth
<azonenberg> it almost hits 5, not quite, so maybe 4.5 or so will be the official spec
<azonenberg> You know what, though? even though this probe with the damping resistors isnt perfect, it does have lower loading than the version with the fitler
<bvernoux> yes with weird oscillations which are strange of course
<azonenberg> And it outperforms my 1.5 GHz active probe after about 250 MHz
<Degi> Lol
<azonenberg> But the pico still wins. I guess that's not surprising as it's a 20x probe not a 10x
<bvernoux> yes 20x probe help
<bvernoux> its fun but we see the curve of Pico is very similar to the v1p1 damped
<bvernoux> but with amplitude of oscillation very attenuated
<bvernoux> if you try 1K res you will match Pico ;)
<azonenberg> Perhaps
<azonenberg> So this graph here... blue is the damped probe which is think is the best so far. The bandwidth is a little less than the one with the big tip resonance, but i need a probe that doesn't load my DUT into oblivion
<azonenberg> this is input Z
<azonenberg> So i think the blue curve has the best results in terms of loading of any probe i've done yet
<azonenberg> what confuses me is the S11 curves
<azonenberg> in S11 the damped curve is worse than the filtered v1.2 (cyan) except from about 4.25 GHz up
<bvernoux> yes it is like there's a filter ...
<bvernoux> this effect is very strange on drak blue
<bvernoux> dark
<azonenberg> yeah the resonance actually seems to have got worse
<azonenberg> Possibly related is the fact that I was using 1K ohm resistors for lack of a 500
<azonenberg> And the 1K is only about 200 ohms at 3 GHz
<bvernoux> ha yes it has bad characteristics
<azonenberg> in fact the 500 ohm has higher resistance than the 1K starting about 1.7 GHz
<azonenberg> And at 3 GHz, the 1K actually has less resistance than the 200 lol
<bvernoux> probably because there is parasitic on each res too
<bvernoux> which have a positive effect ;)
<bvernoux> for such case
<azonenberg> this is the parasitic C mostly
<azonenberg> I have 1K, 200, and 50 available but no 500
<bvernoux> yes the 1K shall be avoided it is clear here ;)
<azonenberg> My "500" is actually two parallel 1K's
<azonenberg> So i wonder if that's what's causing the input impedance to drop off as far as it does
<bvernoux> so you have 4x time bad effect vs 1 x 500Ohms ;)
<azonenberg> Let me try to find a 500 somewhere and see how that works
<bvernoux> this 1K is clearly not good for a probe
<bvernoux> not for such freq range
<azonenberg> Yeah
<bvernoux> anyway as parasitic you have L and C
<bvernoux> I must go
<bvernoux> see you
bvernoux has quit [Quit: Leaving]
<azonenberg> Wow
<azonenberg> So i just replaced the two 1K resistors with a single generic 500 ohm
<azonenberg> and i love it
<azonenberg> it doesn't get below about 115 ohms
<azonenberg> The S21 curve looks pretty nice too
<azonenberg> I'm gonna do one more measurement to confirm
<azonenberg> because i'm not sure i believe it
<azonenberg_work> Hmm, tried a different way to measure and now i got even more nonsensical results :p
<Degi> ll
<Degi> *lol
<azonenberg> ok so here's a different idea...
<azonenberg> And now i'm even more confused
<azonenberg> So now my probe is acting like it has negative resistance :p
<azonenberg> i probed a thru line with and without my probe across it
<azonenberg> at some frequencies, the impedance went UP