<azonenberg>
SMA test board soldered and VNA tested
<azonenberg>
It's... a lot better than i had before, but not ideal. VSWR about 1.25 at 1 GHz
<azonenberg>
and 1.35 at 3 GHz
<azonenberg>
very well matched at 2
<azonenberg>
S11 at 1 GHz is -18 dB which isnt horrible but not great
<azonenberg>
the overall shape of the curve matches my initial sonnet model well though and i'm redoing it with higher fidelity sim settings to see how much closer i get
<azonenberg>
after all the goal of this test was not a perfect match, it was testing how well simulation and reality line up
_whitelogger has joined #scopehal
<azonenberg>
So in this case it looks like the FR408 parameters from oshpark are a better match than darrell's numbers
<azonenberg>
maybe they have better control now
<monochroma>
doesn't oshpark have multiple PCB companies that they use?
<azonenberg>
monochroma: good point
<azonenberg>
i think they do
<azonenberg>
monochroma: so now i'm iterating a bit, fine tuning dielectric constant and thickness values in my sim seeing how close i can get the model to match the oshpark measurements
<azonenberg>
mostly looking at S11
<azonenberg>
my current sim is near perfect to 1.2 GHz and then starts to deviate a bit. the curve is shaped right but seems to be "stretched" a little in the frequency domain
<azonenberg>
the match is impressively good though
<azonenberg>
I also have a decent bit more loss than the sim would predict, i suspect due to the enig plating
<azonenberg>
but i'll run a sim with all nickel instead of copper and i bet that will nicely bracket my actual results
<azonenberg>
i want to get the frequency response to match first though
<monochroma>
ooooo
<azonenberg>
These SMAs and this model are encouraging though. 1.6 dB insertion loss at 6 GHz on oshpark GCPW
<azonenberg>
the connectors are a bit annoying to solder but i worked out a technique that's decent
<azonenberg>
once i have everything debugged i will probably respin the probe test board with these connectors
<azonenberg>
(oh, and the probe itself)
<monochroma>
azonenberg: oh hey, what tests did you perform on that ethernet transformer crosstalk characterization board?
<monochroma>
(sorry slightly offtopic, just popped into my mind that you have VNAs now that you didn't at the time)
<azonenberg>
i did 10baseT on a very short cable into the aggressor
<azonenberg>
then fed a weak sinewave into the victim and measured for any disturbance, noise, jitter, etc
<azonenberg>
the coupling was below the noise floor
<azonenberg>
i couldnt even observe any effects whatsoever
<_whitenotifier-c>
[starshipraider] azonenberg pushed 1 commit to master [+0/-0/±1] https://git.io/JfBfd
<_whitenotifier-c>
[starshipraider] azonenberg e580d06 - Updated oshpark-sma-test simulation to better match VNA measurements from live PCB
<azonenberg>
lain: awake?
<azonenberg>
I'm eyeing the samtec LSHM series for connecting the MEAD connector module to the carrier board
<azonenberg>
they're hermaphroditic connectors which is nice, no need to draw two footprints
<azonenberg>
no ground plane like q-strip, but still rated for 14 Gbps
<lain>
azonenberg: hi
<azonenberg>
which is overkill considering that MEAD's comparator has a max data rate of ~3 Gbps
<azonenberg>
what do you think?
<lain>
oh, that's fancy
<azonenberg>
and they're quite affordable. LSHM_120 (40 position) is $6.15
<azonenberg>
@ digikey qty 1
<azonenberg>
LSHM-120*
<lain>
oooh
<lain>
yeah that looks great, I was worried it'd be >$15 ea
<azonenberg>
vs the 60 position Q-strip, QTH-030, is $11.25
<azonenberg>
so you can get a mated pair of LSHM's for barely the cost of a single q-strip
<azonenberg>
now, no ground plane and generally not quite as high speed
<lain>
>Audible click when mated
<lain>
niiiiiice
<azonenberg>
QTH is rated for 25 Gbps
<azonenberg>
But LSHM looks like a very nice option. I might even want to use it for the AFE boards too
<azonenberg>
since BLONDEL is a cost optimized project and the data rates are more than adequate
<azonenberg>
LSHM with 5mm stack height has -3 dB insertion loss of 11.5 GHz
<azonenberg>
that shouldn't be a problem at all on a 3 Gbps comparator driving (for now) a 1 Gsps LA lol
<azonenberg>
even with the GTP/GTX based LA it's not going to hurt much
<azonenberg>
assuming we want 4x oversampling 12.5 Gsps on the GTX gives a max data rate of 3 Gbps at the comparator. Which is the comparator's limit, and massively below the connector limit
<azonenberg>
lain: were you around earlier today when i was talking with degi and electronic_eel about the LA stuff?
<azonenberg>
We've decided to split the LA pod design into two projects, MEAD and CONWAY (given that CONWAY was already being used as the name for the original LA pod, her coauthor was the perfect namesake for a sister design)
<lain>
wasn't around, no
<lain>
haha
<azonenberg>
same comparators, same sff-8087 interface to the host
<lain>
I see
<azonenberg>
but CONWAY will be for lower speed stuff and have high impedance inputs on 100 mil headers
<azonenberg>
MEAD will be 50 ohm inputs with a LSHM connection to one of several possible input modules
<azonenberg>
the default will be eight MMCX's
<azonenberg>
then in addition to the MSO feature on all of our planned scopes, there will be two new instruments
<azonenberg>
neither of which has a name yet
<azonenberg>
both 1U headless appliances. The first one will have ten MEAD/CONWAY ports on it giving a total of 80 channels, sampled with ISERDESE2's at 1.25 Gsps
<azonenberg>
Pretzel4Ever needed a lot of inputs for his research project and that seemed like a good excuse to design a massively multichannel LA using the existing pod design
<azonenberg>
The second one will have much less ports, probably two or three pods (MEAD only, CONWAY probably won't have the bandwidth) but sample them with kintex-7 GTX's at 12.5 Gsps
<azonenberg>
MEAD will be designed and built before CONWAY as it has a much simpler input stage, and we can develop the host side software using it. CONWAY will be interface compatible
<azonenberg>
Tentative names in my uncommitted notes file are MAXWELL for the 80 channel ISERDESE2 LA and DENNARD for the GTX based one
<azonenberg>
with specs TBD depending on what kintex we pick
<azonenberg>
There will also be an unnamed characterization board that mates a single MEAD/CONWAY pod to an INTEGRALSTICK for bringup testing
<azonenberg>
lain: sound reasonable?
<lain>
interesting
<lain>
I'm not a fan of rackmount for this kind of equipment but whatever, the equipment itself sounds good :D
<azonenberg>
well the pods wont be racked
<azonenberg>
they'll be 8087 cables out to whatever
<azonenberg>
we could totally design a non-racked chassis for it
<lain>
ye
<azonenberg>
i'm also wondering about the possibility of designing a higher sample rate LA with only one MEAD pod at 6 Gsps
<azonenberg>
it would use an xc7a200-2ffg676 which is far cheaper than a beefy kintex
<azonenberg>
one quad for XAUI to 10GbE SFP+
<azonenberg>
the other three quads will each use one channel for sync leaving nine inputs
<azonenberg>
so one 8-bit LA pod and one unused input
<azonenberg>
sorry i meant ffg1156 not 676, to get all sixteen gtp's
<azonenberg>
it wont be as fast as DENNARD but it doesnt need to be, and 6 Gsps @ 8 bits is a respectable level of performance for a LA
<azonenberg>
thoughts?
<lain>
yeah that'd be extremely handy and should be a lot cheaper I'd think
<azonenberg>
That's 48 Gbps and max RAM rate for DDR3L in Artix-7 -2 is 800 MT/s so a SODIMM gets you 51.2 Gbps before overhead
<azonenberg>
if we assume a decent not-toggle rate, RLE should bring our bandwidth requirement down to fit in a sodimm?
<lain>
yeah
<azonenberg>
although the safer option would be to provision two
<lain>
can always sample at 5 Gbps instead too
<azonenberg>
That's an option too, it would be a nicer number for fitting in with other test equipment at a round sample rate
<lain>
yeah and it's still plenty for 99% of things
<azonenberg>
yeah i mean 5 Gsps is enough to oversample DDR3 lol
<azonenberg>
you wont get timing information but you could do full bus decode
<azonenberg>
8 bits is enough, i think, to decode the full command bus
<azonenberg>
then you could have dq and dqs on analog channels of another scope and use the command bus decode to do read/write eyes etc
<azonenberg>
you might not be able to get bank addresses in that but it'd be fine for r/w distinction
<Degi>
Hm some kinda clock input would be neat too. Like if your logic outputs a clock, why not sample based on that
<Degi>
Unrelated idea: Remember the super fast SH buffers I linked a while ago? I bet they'd be neat for a sampling scope heh
<azonenberg>
Degi: my plan is to standardize that input channel 0 of each LA pod goes to a clock capable input on the fpga
<miek>
oh no, i really don't need to get into a new expensive connector standard :D
<bvernoux>
I have bought the stock ;)
<bvernoux>
sorry
<bvernoux>
;)
<bvernoux>
31.9USD for 10 units ;)
<bvernoux>
I have already bought 6 units from same reseller
<bvernoux>
I have done measurements on my JFW Industries Inc RF attenuator Model 50BR-001. 0 to 110db 50 Ohm 2GHz ;)
<bvernoux>
It works fine up to 3GHz ;)
<bvernoux>
and it is very hard to test 100dBm attenuation with IF BW 3KHz ;)
<bvernoux>
as I have a bug to change it and retrive the data over GPIB
<bvernoux>
It seems to be a limitation of memory to be confirmed in HP 8753D ...
<bvernoux>
It is first time I need to change IF BW so far ;)
<bvernoux>
to reach better dynamic range to measure <90dB attenuation
<miek>
i guess you've already got the source power cranked up? i know mine defaults to lower that max
<bvernoux>
yes I tried with +10dBm
<bvernoux>
but anyway after 70B Attenuation what is mandatory is to reduce IF BW to 100Hz to have less noise
<bvernoux>
I see it on screen signal is very nice
<bvernoux>
but I cannot capture it over GPIB ;)
<bvernoux>
It is a bit frustrating
<bvernoux>
sweep average work fine too as it is similar
<miek>
can you reduce the sweep range without having to re-cal?
<bvernoux>
no
<bvernoux>
all VNA I used so far requires a full call
<bvernoux>
cal
<bvernoux>
as the sweep step is different so calibration is required
<bvernoux>
even when you change Power level it is recommended to calibrate but that is still very good
<bvernoux>
as impact with Power Level is not huge as different power level are pretty linear (but it depends on HW) ...
<bvernoux>
azonenberg, I have seen your test with SMA connectors => https://twitter.com/azonenberg/status/1260815950296698880 It is crazy you have similar board& connector I have but you have only 1.6dB Attenuation @6GHz and I have about 2.8dB Attenuation @6GHz on my side
<bvernoux>
So far I do not understand why I have 1 dB more attenuation with what is expected (<2dB)
<bvernoux>
I doubt it is because I have potentially over heat the PCB/Connectors during Hot Air soldering ...