azonenberg changed the topic of #scopehal to: libscopehal, libscopeprotocols, and glscopeclient development and testing | https://github.com/glscopeclient/scopehal-apps | Logs: https://libera.irclog.whitequark.org/scopehal
Degi has quit [Ping timeout: 246 seconds]
Degi has joined #scopehal
Stary has quit [*.net *.split]
Stary has joined #scopehal
<_whitenotifier-7> [scopehal-apps] azonenberg pushed 1 commit to master [+0/-0/±3] https://github.com/glscopeclient/scopehal-apps/compare/666107dd1070...b9f20b69be03
<_whitenotifier-7> [scopehal-apps] azonenberg b9f20b6 - Continued filter support. Fixed a bunch of bugs, can now pass inputs to filters. See #515.
_whitelogger has joined #scopehal
mithro has joined #scopehal
Ekho has joined #scopehal
bvernoux has joined #scopehal
_whitenotifier-e has joined #scopehal
<_whitenotifier-e> [scopehal-apps] azonenberg pushed 2 commits to master [+0/-0/±9] https://github.com/glscopeclient/scopehal-apps/compare/b9f20b69be03...920650e7482b
<_whitenotifier-e> [scopehal-apps] azonenberg ddd3def - Now using ImGui_ImplVulkan_RemoveTexture to delete texture objects. Refactored a bunch of code for more control over object destruction order.
<_whitenotifier-e> [scopehal-apps] azonenberg 920650e - Refactoring: removed no-longer-used descriptor pool object
<_whitenotifier-e> [scopehal-apps] azonenberg pushed 1 commit to master [+1/-0/±1] https://github.com/glscopeclient/scopehal-apps/compare/920650e7482b...e0d8103b9e52
<_whitenotifier-e> [scopehal-apps] azonenberg e0d8103 - Added imgui-node-editor submodule
<bvernoux> @azonenberg, I have found a regression on latest glscopeclient to load spi example which now crash and exit
<bvernoux> on older version it works perfectly it is very strange and the backtrace does not really help so far
<azonenberg> oh? where's it crashing?
<bvernoux> it's crashing during lof of spi example (your example)
<bvernoux> even if I'm doing it by hand in fact
<bvernoux> so either in commande line or manually
<bvernoux> command line
<bvernoux> I shall test on Linux too as so far it is reproductible on windows10
<bvernoux> @azonenberg, also I will do a talk at Hardwear.io NL on 27 Oct 2022
<bvernoux> I have found really funny things with standard simulation with LTSpice or QucsStudio
<bvernoux> Especially such simulation are useless when frequency exceed >1MHz without accurate model of capacitor/inductors ...
<azonenberg> Interesting. i'm trying to integrate the node editor you found and having build issues
<bvernoux> Yes I prefer you focus your time on ngscopeclient
<bvernoux> as it seems it is not far to be used with filters
<bvernoux> and probably after to load existing session and I could try spi example ...
<bvernoux> I have just finished to rebuild it now
<bvernoux> the build is really faster than before
<azonenberg> We actually can use filters right now in ngscopeclient
<azonenberg> what's missing is a lot of the rendering
<azonenberg> We can only *draw* waveforms of sparse/uniform analog/digital type
<azonenberg> no protocols, eye patterns, or spectrograms
<azonenberg> and of course file load/save is still missing
<bvernoux> latest version of ngscopeclient work fine
<bvernoux> no issue so far
<bvernoux> I do not how we can use filter
<bvernoux> right mouse button display only cursors & add markers which both work fine
<bvernoux> +see
<azonenberg> Right click on the channel button
<azonenberg> you can't do it on the main plot area anymore, because it's now possible to have many waveforms in a single plot
<azonenberg> so it would be ambiguous which you intended to work with
<bvernoux> ha yes
<bvernoux> very nice !!
<bvernoux> I'm testing Thresholds
<bvernoux> it is so fast
<bvernoux> oups drag & drop FFT on the ramp crash
<azonenberg> yeah ngscopeclient is massively better than glscopeclient performance wise. none of the actual filter code is different
<azonenberg> FFT is untested
<bvernoux> FFT was working fine
<azonenberg> ah it only died when you dragged it? interesting
<azonenberg> i know we dont have peak drawing working yet
<bvernoux> but it does not like to be drag&dropped on ramp
<azonenberg> interesting. ok i'll investigate
<bvernoux> yes it crashed instantly
<azonenberg> probably confused because of the different Y axis units volts vs dB
<azonenberg> but it should have detected that and not let you drop it
<bvernoux> I'm pretty sure it could be reproduced on your computer too
<bvernoux> yes I suspect it is because the scale is different
<azonenberg> maybe the problem is that it crashes if you try to drop on an incompatible area?
<bvernoux> and incompatible
<azonenberg> well what i mean is, i detect this
<azonenberg> and don't show the drop icon as available
<bvernoux> ha ok
<azonenberg> it's possible if you try to drop on an incompatible area, it tries anyway and hits a null pointer or something
<azonenberg> i'll look into it after i finish some of this other stuff
<bvernoux> the only missing things on FFT is the peak details
<bvernoux> but I suspect you have not implemented it so far as it was using 2D Txt before IIRC
<azonenberg> Yeah. Peak rendering is not currently implemented
<azonenberg> like i said only basic analog and digital waveform drawing works so far
<bvernoux> yes but it is amazing
<bvernoux> and so fast
<bvernoux> The FFT is really fast too
<bvernoux> yes Frequency also does not like to be drag&dropped on it's signal
<bvernoux> except some drag&drop which crash everything is nice and smooth
<bvernoux> Threshold drag&drop also crash
<bvernoux> when it come back to it's own window which same scale
<_whitenotifier-e> [scopehal-apps] azonenberg pushed 1 commit to master [+2/-0/±6] https://github.com/glscopeclient/scopehal-apps/compare/e0d8103b9e52...cff94b3233fd
<_whitenotifier-e> [scopehal-apps] azonenberg cff94b3 - Initial work on filter graph editor. Just an empty window, but we now compile the node editor code.
<azonenberg> yeah i fixed a few segfaults related to drag and drop recently
<azonenberg> seems we still have some left
<azonenberg> i'll work on that after i get the initial filter graph editor up and running
<azonenberg> bvernoux: you missed the dev call i think, but i mentioned we set a new record for streaming performance with ngscopeclient and the experimental thunderscope driver last week
<azonenberg> something like 7.6 Gbps of streaming waveform data :D
<bvernoux> yes unfortunately I missed the call
<bvernoux> ha great I will say amazing !!
<bvernoux> it was with a big 3D GFX card too ?
<bvernoux> and fast PC I Imagine
<azonenberg> 40M points * 4 channels * 22.5 WFM/s * 8 bit resolution
<azonenberg> or wait, let me make sure i'm reporting the right numbers
<azonenberg> ok so the record number was
<azonenberg> 5M points * 4 channels * 44.8 WFM/s
<azonenberg> that's 896 Msps out of a theoretical 1000 coming off the ADC on the scope
<azonenberg> times 8 bits is 7168 Mbps / 7.168 Gbps
<bvernoux> yes very nice
<azonenberg> iirc aleksorsist has an nvidia 3070, unsure what CPU but it was something fast
<azonenberg> with the scope connected to the PC via thunderbolt aka pcie
<azonenberg> this was just displaying waveforms, no filter graph processing yet
<azonenberg> but still, this is an enormous amount of data. idk if it's a record for pc based scopes but it's certainly up there
<bvernoux> yes just DMA PCIE => GFX card
<azonenberg> not direct from pcie to card, it still passed through system ram
<azonenberg> and the int8 to float32 conversion was done with AVX on the CPU
<bvernoux> so the CPU was pratically not involved in theory
<azonenberg> (we will likely optimize further by pushing that conversion to the GPU, but that's not done in the current driver)
<bvernoux> ha yes
<bvernoux> I was thinking about your latest probes
<bvernoux> as it is quite complex to simulate them
<bvernoux> Why do you not try with test fixture and your VNA to take fill caracteristics of each bloc from few KHz to 8.5GHz ?
<bvernoux> then you will have full modelisation with s2p that you can plug
<bvernoux> and so on for each step
<bvernoux> like that you could correct impedance or other parameters if they are bad
<azonenberg> Yes. De-embedding the individual components especially things like the tip resistor is extremely nontrivial
<bvernoux> with a TRL
<azonenberg> I'm actively working with a support engineer from Sonnet to develop good EM models of all the parts
<bvernoux> I have done few design of TRL and they are perfect for that
<bvernoux> you just need to solder the component and do full TRL calibration
<azonenberg> Ideally I want full EM models of the tip including coupling to other components etc
<bvernoux> yes but for that you need also full model of your PCB with connector
<bvernoux> as they have big influence on the results
<bvernoux> and for that TRL and VNA is a must have
<azonenberg> So one of many things i want to do is make a PCB with pads for solder in probes and a 2x thru for de-embedding
<azonenberg> that seems to be the way to go for modern high performance characterization work, that's what IEEE 370 etc is recommending
<bvernoux> yes it is the purpose of TRL
<bvernoux> it is used as test fixture at end
<bvernoux> It is an old version I shall push latest version (which is very similar)
<bvernoux> this example of TRL board can be modified to adapt it for other components to be fully caracterized
<bvernoux> here it is for Series/Shunt Fixture for 0402
<bvernoux> I have improved the connectors on latest version
<bvernoux> to be compatible and easier to solder any SMA connector
<_whitenotifier-e> [scopehal] miek opened pull request #717: Fix waveform drawing for the AgilentOscilloscope driver - https://github.com/glscopeclient/scopehal/pull/717
<_whitenotifier-e> [scopehal] azonenberg closed pull request #717: Fix waveform drawing for the AgilentOscilloscope driver - https://github.com/glscopeclient/scopehal/pull/717
<_whitenotifier-e> [scopehal] azonenberg pushed 3 commits to master [+0/-0/±3] https://github.com/glscopeclient/scopehal/compare/efc909a9568b...f9fa9c49e273
<_whitenotifier-e> [scopehal] miek 114eb39 - AgilentOscilloscope: fix waveform drawing
<_whitenotifier-e> [scopehal] miek 3279966 - AgilentOscilloscope: use standard sample conversion function
<_whitenotifier-e> [scopehal] azonenberg f9fa9c4 - Merge pull request #717 from miek/agilent-fixes Fix waveform drawing for the AgilentOscilloscope driver
bvernoux has quit [Quit: Leaving]