04:46 <azonenberg> So i think i found another way to improve response of the probe on top of what i'm doing already

04:48 <azonenberg> I can shorten the stub on the probe tip

04:48 <azonenberg> By around 0.4 mm

04:49 <azonenberg> It looks like that will buy me about 100 MHz of bandwidth

04:55 <azonenberg> So the next version I've got in my kicad design should, if my simulations are correct, still have peaking but it should be at closer to -18.5 to -19 dB instead of the -17.3 i have now

04:55 <azonenberg> and possibly slightly lower frequency, maybe 3.2-3.4 GHz vs the 3.63 it's at now

04:57 <azonenberg> Ultimate -3 dB bandwidth i'm not super sure about, but very roughly i'd say around 4.1 GHz should be doable

04:58 <azonenberg> interestingly the pico 6 GHz probe actually has around 1.5 dB of peaking at 3.1 GHz. So that's suggesting i'm starting to approach their performance

04:59 <azonenberg> The curve shape looks almost the same just shifted 6 dB down

04:59 <azonenberg> Gonna order this new probe early next week probably

05:03 <azonenberg> lain, miek: https://www.antikernel.net/temp/probe-evolution.png

05:03 <azonenberg> Cyan = current kickstarter probe design

05:04 <azonenberg> Green = pico 1.5 GHz piece of junk

05:04 <azonenberg> purple = pico 6 GHz probe (20:1, all others are 10:1)

05:04 <azonenberg> Black = today's prototype

05:05 <azonenberg> Red = simplified sim of today's prototype not including pcb or copper losses

05:05 <azonenberg> Blue = sim of my proposed next prototype, trying to reduce the peaking by a bit without losing too muc hbandwidth

05:05 <azonenberg> (also Implant monochroma wbraun tnt have a look)

05:07 <azonenberg> this is with the tip mounted ground, i'm not yet sure how far i can push it with the z-ground and other less ideal grounding accessories

05:09 <monochroma> :O so smooth

05:10 <azonenberg> monochroma: So basically it looks like there's three different major effects in play here

05:11 <azonenberg> The first one is reflections between each pair of resistors due to having a stub with a high impedance termination at both ends, you get standing waves. As the stub gets longer, or you add more resistors, you get high frequency loss

05:11 <azonenberg> Second effect is resonances in the stub from the DUT to the first resistor. So basically the length of the probe needle plus the trace from the needle to the first resistor. I shrunk that trace by 400um in my pending prototype

05:12 <azonenberg> Third is non-flat response of the resistors themselves, in order to reduce effect #1 i went from six resistors to three in the black trace, then up to four in the blue one. This inherently required changing resistor values

05:13 <azonenberg> the 200 ohm in particular has significant peaking at higher freqs because the impedance of the parasitic 26 fF capacitor in parallel with the resistive element becomes nontrivial compared to a 200 ohm resistance

05:22 <azonenberg> https://www.antikernel.net/temp/tip-stub-effects.png this is the resonant behavior of the probe tip in isolation

05:22 <azonenberg> Blue line is an idealized simulation with a zero-length probe tip and an ideal 450 ohm resistor with no parasitics, and a lossless line for the probe body

05:23 <azonenberg> Red is the same idealized simulation but with an 8.1mm long 226 ohm stub, which is the best fit so far to my experimental data for the probe tip

05:27 <azonenberg> https://www.antikernel.net/temp/resistor-spacing-effects.png this is comparing a series string of ideal 200-100-100-50 ohm resistors with no probe tip stub and no stub between resistors (blue) to 0.6mm stub between resistors (red)

05:30 <azonenberg> https://www.antikernel.net/temp/resistor-parasitic-effects.png and this compares the same ideal series string in blue, to the same string but with actual S2P models for each resistor in red

05:31 <azonenberg> Summarizing... probe tip stub creates a ~5 dB resonant peak at 4.8 GHz, gap between resistors creates loss that increases with frequency (2 dB @ 6 GHz for 0.6mm gap on 4 resistors)

05:32 <azonenberg> and parasitic shunt capacitance across resistors creates peaking that increases with frequency, about +0.7 dB at 6 GHz

06:00 <_whitenotifier-f> [starshipraider] azonenberg pushed 1 commit to master [+13/-0/±2] https://git.io/JUlcY

06:00 <_whitenotifier-f> [starshipraider] azonenberg 14d94df - More simulations of probe effects

07:05 <azonenberg> monochroma: https://www.antikernel.net/temp/notch-4.png

07:06 <azonenberg> So this is my latest experiment in sonnet. Red trace = current prototype, blue trace = proposed filter to null out that hump

07:06 <azonenberg> I'm going to tweak it slightly as the filter's peak seems to be just a little left of my actual peak but i think i'm on the right track

07:12 <azonenberg> https://www.antikernel.net/temp/stub-filter.png here's the layout

07:12 <azonenberg> intuitively, the peaking is mostly caused by the probe tip which is a high impedance section of line, so i'm going to add a low-impedance section of line to cancel out that effect

07:12 <azonenberg> probably right after the attenuator resistors

07:12 <azonenberg> in order to keep the number of variables low, i plan to add this filter to the existing v1.1 layout with no other changes

07:13 <azonenberg> And see how flat it ends up being in real life

07:13 <azonenberg> The filter is just a 13.6mm long section of 0.8mm wide transmission line

07:36 <azonenberg> https://www.antikernel.net/temp/probe-with-filter-2.png final layout

08:22 <_whitenotifier-f> [starshipraider] azonenberg pushed 1 commit to master [+1/-0/±0] https://git.io/JUllE

08:22 <_whitenotifier-f> [starshipraider] azonenberg 31e8cc9 - Added simulation of filter to oshpark probe

08:23 <_whitenotifier-f> [starshipraider] azonenberg pushed 1 commit to master [+0/-0/±2] https://git.io/JUllz

08:23 <_whitenotifier-f> [starshipraider] azonenberg e5f97cf - Added filter to v1.2 probe to cancel out peaking from tip. Moved resistors 400um closer to tip.

23:53 <azonenberg> https://www.antikernel.net/temp/probe-filter.png did some more sims and it seems like while my filter's cutoff freq is good, i made it a little bit too strong. Blue trace shows after some tweaking

23:54 <azonenberg> went from 800 to 600um width

23:54 <azonenberg> blue trace is simulated freq response of the entire probe with the filter

23:57 <azonenberg> The red trace is a bit optimistic and doesnt model dielectric or conductor losses, but the peak is at the right spot

23:58 <azonenberg> adding in losses i'm expecting this new rev to be quite flat out to ~4 GHz