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
<_whitenotifier-8> [scopehal-apps] lainy pushed 1 commit to master [+0/-0/±1] https://github.com/glscopeclient/scopehal-apps/compare/120c03f08deb...0b831752249f
<_whitenotifier-8> [scopehal-apps] lain 0b83175 - Install Catch2 with homebrew in macOS CI build
Degi_ has joined #scopehal
Degi has quit [Ping timeout: 256 seconds]
Degi_ is now known as Degi
<_whitenotifier-8> [scopehal-apps] lainy pushed 1 commit to master [+0/-0/±1] https://github.com/glscopeclient/scopehal-apps/compare/0b831752249f...978194dcf115
<_whitenotifier-8> [scopehal-apps] lain 978194d - Fix macOS CI build missing OpenMP
<_whitenotifier-8> [scopehal-apps] lainy pushed 1 commit to master [+0/-0/±1] https://github.com/glscopeclient/scopehal-apps/compare/978194dcf115...57f5f7b9ea0d
<_whitenotifier-8> [scopehal-apps] lain 57f5f7b - Try again to fix macOS CI build
<_whitenotifier-8> [scopehal-apps] lainy pushed 1 commit to master [+0/-0/±1] https://github.com/glscopeclient/scopehal-apps/compare/57f5f7b9ea0d...8bbb0fe835b2
<_whitenotifier-8> [scopehal-apps] lain 8bbb0fe - Yet another macOS CI fix attempt
Bird|ghosted has quit [Remote host closed the connection]
Bird|ghosted has joined #scopehal
<azonenberg> ok so probe update
<azonenberg> i ran some TDR measurements on a SMPM test board that had been literally sitting unpopulated on my bench since december
<azonenberg> (yay for being crazy busy)
<azonenberg> And it seems that a) oshpark is overetching my transmission lines and b) they were too small even if perfectly etched
<azonenberg> The PT5 is currently designed with 0.4mm wide traces
<azonenberg> the test board in question had 0.45mm traces (overetched to about 0.4 on the top and 0.46 at the bottom, i don't know the cross section profile yet) and TDR showed it in the 55-58 ohm range
<azonenberg> Which means my 0.4mm nominal (probably overetched) traces are even higher than that
<azonenberg> This doesn't quite match some of my simulations which makes me wonder if the dielectric config at oshpark isn't quite what i've been simulating
<azonenberg> I'm probably going to run some cross sections later this week
<tnt> azonenberg: couldn't see anything relevant in the validation layer.
<tnt> UNASSIGNED-GPU-Assisted-Validation(ERROR / SPEC): msgNum: -359834619 - Validation Error: [ UNASSIGNED-GPU-Assisted-Validation ] Object 0: handle = 0x56247d540870, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0xea8d5c05 | Setup Error. Detail: (GPU-Assisted validation requires fragmentStoresAndAtomics and vertexPipelineStoresAndAtomics. GPU-Assisted Validation disabled.)
<tnt> So OTOH I think it couldn't run a bunch of the shader based validation because of this.
bvernoux has joined #scopehal
bvernoux has quit [Ping timeout: 250 seconds]
<azonenberg> yeah
<azonenberg> so a lot of things around bad pointer detection etc dont work as a result
<d1b2> <david.rysk> tnt: what intel gpu do you have?
<d1b2> <david.rysk> if it's an older one, I can dig out my thinkpad x230t and do some testing
bvernoux has joined #scopehal
<d1b2> <tnt> @david.rysk Yeah, it's an older one but I think the x230t is even older. What CPU does it have ?
<d1b2> <tnt> But it also happens on newer one apparently. (Else I wouldn't bother)
<d1b2> <david.rysk> i7-3520M / Intel HD Graphics 4000
<d1b2> <david.rysk> vulkan wouldn't work at all without updating mesa to bleeding edge (but this was before 23.0.0 was released)
<azonenberg> Ok sooo let's see how this goes. About to submit a new AKL-PT5 board design for fab
<azonenberg> first new probe design of 2023 i think since work has been so busy
<azonenberg> it deletes the whole filter network and widens the trace from 400 to 450 μm
<azonenberg> and i think i can also cut about 1mm off the long axis of the board
<azonenberg> This means that the transmission length is now 5.9 mm, so if i have any more problems with impedance my null should be when that's a quarter wave long at a velocity factor of about 0.62. Which comes out to a period of 32 ps one way
<azonenberg> so if we have a null from reflections it would be at a period of 128 ps or 7.81 GHz
<azonenberg> actually no i can actually shorten it by 2mm
<azonenberg> That gives about 4.8mm overall length
<azonenberg> That gives 24 ps for one way, 96 ps for the whole wave, and a potential null at 10.42 GHz
<azonenberg> So even if i didn't make the transmission line any better this tweak should be shifting badness significantly higher in frequency. I wonder why I didn't do this before?
<azonenberg> I think there's potential to make it even smaller but this is a nice start
<d1b2> <tnt> @david.rysk Yeah mine is i7-4600U, so Intel® HD Graphics 4400 , a generation after.
<_whitenotifier-8> [starshipraider] azonenberg pushed 1 commit to master [+7/-0/±4] https://github.com/azonenberg/starshipraider/compare/f2079bf67807...733cc64f9dd8
<_whitenotifier-8> [starshipraider] azonenberg 733cc64 - Updated design files for AKL-PT5 v0.11
<d1b2> <tnt> FWIW So ngscopehal, default window size, showing the full 2uS of the demo scope is ~ 900 fps
<d1b2> <tnt> Something is wrong with the filter graph window though. It beomes all black when I hover over a node and it shows the tooltip.
<d1b2> <tnt> Shouldn't the eye pattern be interpolated ?
<azonenberg> @tnt there is a known issue with the filter graph window related to the ImGuiNodeEditor library
<azonenberg> there's an open github issue w the library for it
<azonenberg> for now the workaround is to dock the graph editor rather than having it in a popup
<azonenberg> as long as it's docked it draws fine
<azonenberg> wrt eye interpolation, no. Each sample in the incoming waveform is rendered as a single point
<azonenberg> With a real signal that is asynchronous to the sampling clock this is usually fine
<azonenberg> but in the demo scope there is a perfect trigger with no jitter and no asynchronicity
<azonenberg> which leads to htese artifacts
<azonenberg> you can up the sample rate or use a sinc upsampling filter to reduce it
<azonenberg> the proper solution is for the demo scope to add noise to the waveform then simulate an actual level trigger on the noisy waveform
<azonenberg> but it was meant as a proof of concept for driver testing etc, not to be a physically accurate model of a real scope
<azonenberg> (I'd love to make a more accurate scope simulator at some point though)
<_whitenotifier-8> [scopehal-apps] Walkerru forked the repository - https://github.com/Walkerru
<d1b2> <tnt> @azonenberg Ah thanks. Makes perfect sense !
<d1b2> <tnt> There is definitely a couple of other gotcha with my tiling window manager (even in non-tiling mode).
bvernoux has quit [Quit: Leaving]