07:09 <azonenberg> ok WOW. Simulation isn't finished but i think i might have found my v1,4 AKL-PT1 design, no filtering needed

07:10 <azonenberg> Switching from Vishay FC series 0402 resistors to CH series 0201 in an attempt to make the attenuator string electrically smaller (as an added bonus CH series has a 250 ohm so I can combine the 200 and 50 into a single 250 plus a 50)

07:10 <azonenberg> and making a few other small tweaks to ground layout

07:10 <azonenberg> So far the tip region is looking to be flat +/- 0.5 dB from DC out to past 16 GHz

07:11 <azonenberg> now, this isn't game over - this is before adding loss from the probe body, and the actual tip pin is difficult to simulate (I have a rough approximation I'll bolt onto the end of this sim when it's done and see what the end result looks like)

07:11 <azonenberg> But it's looking like a huge improvement

07:11 <monochroma> :D

07:12 <azonenberg> in particular, wrt getting rid of the 5+ GHz peaking Shahriar complained about in the review

07:12 <azonenberg> So, it turns out there isn't exactly 5 GHz peaking

07:12 <azonenberg> there's two separate effects going on

07:13 <azonenberg> https://www.antikernel.net/temp/tip-comparison.png

07:13 <azonenberg> ok so, for starters, red is a simulation of the tip end of the v1.3 probe as-is. ENIG loss not modeled so the oshpark prototype will have increasing attenuation as frequency increases

07:14 <azonenberg> (these sims all include FR408 dielectric loss, but for the tip region only - the whole probe is too big to fit even in sonnet gold at usefully small mesh sizes)

07:14 <azonenberg> Note how it rises to 5 GHz, then dips

07:14 <azonenberg> then plateaus around 9 GHz and rises again?

07:15 <azonenberg> https://www.antikernel.net/temp/tip-nofilter.png here's the corresponding layout

07:16 <azonenberg> Any red flags jumping out at you in the layout? If not, you missed the same thing I did :p

07:17 <azonenberg> monochroma: sooo, see that little "tail" in the ground just above the leftmost resistor (COMP3)?

07:17 <monochroma> yeah

07:18 <monochroma> the dark red trace, right?

07:19 <azonenberg> The dark red hatched area

07:19 <azonenberg> the solid outlined boxy things are vias

07:19 <azonenberg> i simplified rows of closely spaced vias into single bars of via

07:19 <monochroma> ahh

07:19 <azonenberg> uses less mesh cells to model and as long as they're << λ apart it's a very close approximation

07:20 <azonenberg> anyway, i'm referring to that whole chunk of metal north of COMP3

07:20 <monochroma> ah gotcha

07:20 <azonenberg> where it goes up then right instead of continuing straight up

07:20 <azonenberg> https://www.antikernel.net/temp/tip-nofilter2.png

07:20 <azonenberg> In the current view it's a little hard to see but once you know what to look for there's definitely more current there than there should be

07:20 <monochroma> :O

07:20 <azonenberg> suggesting that little bit of ground is resonating

07:21 <azonenberg> Soo

07:21 <azonenberg> you know what happens when you cut off the ground vertically and don't have that tail?

07:22 <azonenberg> You get the blue trace

07:22 <azonenberg> in the original graph i linked

07:22 <azonenberg> it stays about the same until 4 GHz then keeps on climbing all the way up to -11 dB at 16 GHz

07:22 <azonenberg> The difference between red and blue is entirely caused by resonances in that tiny stub

07:25 <azonenberg> (I've been running sims almost nonstop for weeks making tiny tweaks to the layout to track down some of these things)

07:26 <azonenberg> anyway so basically that little stub acted like a weak notch filter and concealed the true extent of the peaking

07:26 <azonenberg> All of these sims also remove the deliberately added filter I had on v1.3 to weaken the peaking as well

07:27 <azonenberg> Since i figured that was also concealing the real problem and i wanted to get to the bottom of it

07:28 <azonenberg> Anyway, then blue to purple kept that stub gone and switched from 200-200-50 ohm FC series to 250-200 ohm CH series. The CH series resistors have looser tolerance so the DC gain accuracy drops from like 0.1% to 2%, but much flatter frequency response

07:28 <azonenberg> And I can make the attenuator string smaller by going from 3 to 2 resistors

07:29 <azonenberg> Then purple to cyan stays with CH series but shrinks from 0402 to 0201

07:29 <azonenberg> I'm now running a sim of the cyan layout using RO4350B instead of FR408HR to see how much things change with less dielectric loss

07:30 <azonenberg> then will concatenate this with a separate sim of the probe body to get end to end board level response. Then I can try and estimate contributions from the probe tip

07:30 <azonenberg> this sim includes the PCB footprint for the tip sockets but not the needles themselves

13:00 <_whitenotifier-5> [starshipraider] azonenberg pushed 2 commits to master [+50/-1/±9] https://git.io/JtjUJ

13:00 <_whitenotifier-5> [starshipraider] azonenberg 9a7d30f - Continued AKL-PT1 simulations

13:00 <_whitenotifier-5> [starshipraider] azonenberg 4083bd6 - Lots more AKL-PT1 simulation work