<miek>
azonenberg: it's half of an HP 71612C 12.5Gb/s Error Performance Analyzer, bit of a boat anchor :p
<miek>
sold as not working cause they can't test it and doesn't have the display, but it should run standalone with gpib control & maybe i can get it working. it was *dirt cheap* though, the connectors on the front are worth more than i paid. should be interesting.. :)
<azonenberg>
lol nice
<miek>
i probably have to spend more on adapters/terminators just to turn it on safely, lol
<azonenberg>
What's the connector interface?
<azonenberg>
SMA? 3.5mm?
<miek>
it's mostly 3.5mm with some SMA, but i'm not shelling out for 3.5mm savers
<miek>
that would definitely be many many times what i paid for it!
<azonenberg>
what about using SMA savers?
<azonenberg>
SMA should be fine for 10G data rates
<miek>
yeah, that's that plan - i'll just get some quality SMA ones
maartenBE has quit [Ping timeout: 256 seconds]
maartenBE has joined #scopehal
kbeckmann has quit [Ping timeout: 240 seconds]
electronic_eel has quit [Ping timeout: 240 seconds]
electronic_eel has joined #scopehal
promach3 has quit [Read error: Connection reset by peer]
<azonenberg>
i've started VNAing all of my SMA-SMA cables in order to de-embed them from measurements etc
<azonenberg>
There's a surprising amount of variation
<azonenberg>
as much as 0.8 to 1.4 dB at 3 GHz
<NeroTHz>
azonenberg, how are you testing this? Did you verify this is not repeatability/bending?
<azonenberg>
I'm working on collecting more data now
<NeroTHz>
would be interesting to plot that vs just mating one or two cables 10 times, and checking that, as well as just checking for drift in the VNA
<azonenberg>
but they were measured through the same (de-embedded) SMA F-F couplers on the same table at about the same orientation
<azonenberg>
Yeah. That's going to be the next piece of data
<NeroTHz>
these are minicircuits cables right?
<azonenberg>
I have other stuff but these are all FL086-24SM+
<azonenberg>
They seem to be well matched in length, When I plot S21 phase the traces stack ~perfectly
<azonenberg>
and when i look at group delay any variation is below the noise floor
<azonenberg>
So the variation i'm seeing is only in amplitude
<NeroTHz>
yeah, I don´t think you have the dynamic range to do group delay measurements tbh
<NeroTHz>
you need a *lot* of points for group delay though
<azonenberg>
These traces are 10K points from 300 kHz - 6 GHz
<NeroTHz>
though you seem to have a lot of points
<NeroTHz>
yeah okay that is fine
<NeroTHz>
what is the IF meas bandwidth?
<azonenberg>
There's a "bandwidth" setting i left at the default of 1 kHz
<NeroTHz>
that can help to lower noisefloor of groupdelay (combined with lots of trace filtering, after all your group-dealy is going to be an almost linear or quadratic curve)
<NeroTHz>
it can be beneficial to measure groupdelay by looking at S11 phase
<NeroTHz>
iirc
<NeroTHz>
on most not-ultra-high-end-VNAs
<NeroTHz>
the thing is that your group delay comes from comparing the reference signal phase at port 1 to the measured phase at port 2, but your VNA likely only has one LO path, so it needs to really quickly switch between the two measurements and just assume that the LO doesn´t have phase noise
<NeroTHz>
which is a big contributor to noise in GD measurements
<azonenberg>
no thats now how it works AFAIK
<azonenberg>
First off, this is a 4-RX VNA
<azonenberg>
There's no muxing on the rx path
<azonenberg>
Second, the VNA is not directly giving me group delay, I'm deriving it from the measured S-parameters. It's -dPhase/dFreq IIRC?
<azonenberg>
it's just a display mode i can set in the s-parameter viewer in sonnet (which i like much better than the VNA software)
<NeroTHz>
yeah, ofcourse. but phase noise in your signal is going to show up straight in that dphase, and so you get a lot of noise as dPhi/dfreq goes up and down fast with the internal phase noise
<NeroTHz>
I´m looking for a VNA block diagram now to explain what I meant (still not sure I´m right though)
<azonenberg>
what i mean is it's measured from S21 only
<azonenberg>
as opposed to port 2 phase - port 1 phase
<NeroTHz>
yeah, but the VNA measures S21 by looking at what goes out of port 1 (ref) and what goes in at port 2 (meas)
<azonenberg>
Anyway so my next experiment is using cable #3 (I serial numbered them all so i can tell them apart)
<azonenberg>
I mated it once and am doing several acquisitions. Remotely from my desk in the other room
<azonenberg>
so no chance of me bumping and bending the cable between captures as i grab the mouse
<azonenberg>
This should be a good measurement of VNA repeatability
<NeroTHz>
each port is that, and your VNA uses the reference at each output (here alpha-a) to know how much signal went in and with what phase, and then compares that the the b-wave (alphab) at port 2, to figure out phase delay and amplitude shift
<NeroTHz>
and the thing is that usually, you measure those two things with a mixer driven by an LO, and if you have phase noise on the LO, that translates to phase-noise in the measurement. but I was wrong, I don´t think it switches, I think it might only do that when using external frequency converters because of how the multiplication factors happen to work out?
<azonenberg>
so not a huge amount of variation but the entire curve is displaced vertically from the previous batch of measurements
<NeroTHz>
I´d say not a huge variation, but combined with the .1-ish dB from movement, you already have a good bit of your variation between different cables located
<azonenberg>
Variation at 3 GHz within those four measurements is -0.9346 to -1.0024 dB, or 0.0678 dB
<azonenberg>
Then adding in the movement the range is -0.9346 to -1.1093 or 0.1747 dB
<NeroTHz>
still, quite a lot of change, what length of cable is this?
<azonenberg>
The range observed among the 7 cables at one measurement each was -1.4024 to -0.8392
<azonenberg>
Or 0.5632 dB
<azonenberg>
I'm using 600mm "standard" grade test leads from Pico (N-SMA). They claim 2 degree or 0.1 dB variation at 6 GHz for a 10cm diameter bend compared to straight
<azonenberg>
Their "premium" grade cables are stiffer and designed to hold their position (i guess kinda semi-rigid but not quite?)
<azonenberg>
those have 3.5mm connectors and claim 0.8 degree / 0.05 dB stability for the same bend, and also bend less during use because they're stiffer
<azonenberg>
Thinking about picking 'em up at some point although they're a lot more expensive ($749 vs $279 per cable)
<azonenberg>
i suspect most of the variation i'm seeing is probably bending in the minicircuits cables, which are not labeled as phase stable
<azonenberg>
i was not expecting that much amplitude variation, that's for sure
<NeroTHz>
Junkosha has some really nice phase-stable cables
<azonenberg>
The pico ones are Jyebao
<azonenberg>
no idea if thats a remotely decent brand or not, i've never heard of them :p
<NeroTHz>
MWX221 cables are what you are looking for I think? at 15 GHz, <2 deg phase shift when wrapped 360deg around 60mm mandrel
<NeroTHz>
yeah I´ve never heard of Jyebao either
<azonenberg>
So... MWX221-00610AMSAMS
<azonenberg>
sma to sma M-M, 24 inch, $279 at Gap Wireless. That's actually less than i expected
<NeroTHz>
I think so
<azonenberg>
it's on par with the lower end pico cables
<azonenberg>
They do not appear to have SMA-N cables so i'd need to get a SMA-N adapter for the VNA
<azonenberg>
but that's no biggie
<NeroTHz>
ah you have N connectors on the VNA?
<azonenberg>
Yes the VNA ports are N. I currently have rosenberger M-F couplers on there as savers for the vna-side ports
<azonenberg>
then i have the pico / rebranded jyebao SMA-N cables off that
<NeroTHz>
We got a quote for that cable recenetly for some decently phase-stable cable, and it was nice quote - MWX221-02500AMSAMS/B
<azonenberg>
anyway that will be a good thing to do, but the other thing is that i obviously have variation in the DUT cables too
<azonenberg>
So there's only so much i can do to figure out the "actual" s-parameters of a cable when it's not the same as i bend it :p
<NeroTHz>
huh odd, they seem to have N-connectors in the datasheet though
<azonenberg>
Yes
<azonenberg>
They have N-N
<azonenberg>
I did not see any SMA-N versions. maybe available special order
<NeroTHz>
ah yeha, that might be it
<NeroTHz>
I recenetly got very furstrated about cable-combination possibilities
<NeroTHz>
well, I actually still am frustrated
<azonenberg>
yeah i've been buying a lot of lower end cables from cdint.com who will custom make pretty much any combo you want
<azonenberg>
but the coax they use isn't all that great
<azonenberg>
they're my lossiest cables
<NeroTHz>
yeah, this was for VNA test cables, so gore phaseflex or equivalent
<NeroTHz>
our newest VNA uses 1.85 mm connectors, but all our VNA extenders use 2.92, and to avoid always having to add adapters to the test-cables, we wanted to order custom 1.85f->2.92m cables
<NeroTHz>
but none of the 4 brands we asked can make those (because the cable for 1.85mm is different in diameter than that for 2.92mm, and so they can´t put those two on the same cable)
<azonenberg>
oh so it's actually physically incompatible
<azonenberg>
it's not just that it's not a standard catalog part
<NeroTHz>
indeed, we were prepared to pay a custom-cable-premium but gore was like ´unless you buy 5k, ain´t gonna happen, because we need to tool for a custom 2.92 connector body´
<azonenberg>
i was gonna say i figured a shop with your kind of pull/budget would be able to get any cable you wanted
<azonenberg>
even if it wasnt a stock SKU
<NeroTHz>
we did too but reality hit us in the face :p
<azonenberg>
lol
<azonenberg>
So you just cannot get 2.92 and 1.85 connector bodies meant to fit the same coax?
<NeroTHz>
you can, but not on the specific phase-stable coax we need. For generic types of coax it is possible
<NeroTHz>
but since this is VNA test-cable, we *need* phaseflex or similar, and they only have a handfull of ´models´ of cable there
<azonenberg>
Yeah makes sense
<azonenberg>
yeah ultimately i just want to be able to de-embed cable loss or see how degraded a signal will be after a given length of cable
<NeroTHz>
they have a few types <18 GHz, and then they have one type up to 27 GHz, one type up to 42 GHz, one type up to 50, and one up to 67 GHz, and all are different size, and so they only have connectors for very specific sets of combination
<azonenberg>
So realistically 0.2 dB of variation is not a horrible deal
<NeroTHz>
yeah, only if you have high-q measurements it really matters
<azonenberg>
yeah
<azonenberg>
the immediate use case is measuring some realistic cables then stacking a couple of them to emulate a really long channel
<azonenberg>
Then testing my CTLE on it
<azonenberg>
i think i fixed the equalizer
<azonenberg>
But i need a realistic channel to test it with. The scope frontend's 4 GHz antialiasing filter is not a realistic channel
<azonenberg>
because it's super flat then has huge rolloff
<NeroTHz>
you can get one of those PCB test boards that are just really long traces
<azonenberg>
That costs money
<NeroTHz>
I think they show up on ebay for checap now and then
<azonenberg>
VNAing cables i already have is good enough for this
<azonenberg>
i can just slap five copies of the same long cable end to end
<NeroTHz>
true
<azonenberg>
in simulation
<azonenberg>
and get s-parameters for a really crappy link
<azonenberg>
Quantitative results dont matter, i just need a plausible worst-case link to test it on
<azonenberg>
The other thing i plan on doing soon is measuring actual insertion loss of my scope front end
<azonenberg>
or estimating at least
<azonenberg>
basically, set the VNA to output a pure tone from ~DC to 6 GHz, measure amplitude on the scope
<azonenberg>
de-embed the cable and whatever rolloff is left must be the scope
<azonenberg>
then make a simple piecewise linear curve i can use to flatten out the frontend response a bit past the 4 GHz official passband
<azonenberg>
Do i expect to get ultra high quality measurements out of that? of course not
<azonenberg>
can i squeeze a little more performance out of the hardware i already have? probably
<azonenberg>
bvernoux: highly experimental and questionably accurate data... but this is the frequency response of my 4 GHz scope frontend after de-embedding the cable from the VNA lead to the scope
<azonenberg>
nominal 4 GHz actually has a -3 dB (in this methodology) bandwidth around 4.5 GHz
<azonenberg>
and right around 5 GHz it flattens out for a bit before dropping again
<azonenberg>
I don't know how much to trust this data
<bvernoux>
woo very nice
<azonenberg>
I wish lecroy just published specs
<bvernoux>
we can say it is very go up to 4.5GHz
<azonenberg>
basically i measured the amplitude of a nominal 0 dBm tone coming out of my VNA signal generator sweeping from 0 to 6 GHz
<NeroTHz>
How did you verify outpu tlevel of the VNA?
<azonenberg>
NeroTHz: I didn't. Which is why i say "questionably accurate"
<NeroTHz>
ah :p
<azonenberg>
pico has a signal generator utility
<azonenberg>
and i ordered the vna with a traceable calibration
<azonenberg>
Let me go look at the cal data and see how they cal'd the output
<azonenberg>
I really need a proper traceably cal'd siggen for this sort of experiment
<bvernoux>
I have a signal generator for that
<NeroTHz>
yeah but usually the output level isn´t critical I think
<azonenberg>
Exactly
<bvernoux>
will test it ;)
<azonenberg>
which is why i am wondering if they calibrated the level at all
<bvernoux>
as it was calibrated and very accurate something like +/-0.1dB
<NeroTHz>
my guess is no, or maybe just a ´it passes the 2 dB ripple spec´
<azonenberg>
But it's the best data i can get with the instrumentation i have right now
<azonenberg>
ok so first off, the lecroy cal certificate for channel 2
<azonenberg>
It lists gain at 10 mV/div on channel 2 in 50 ohm mode as -0.87 +/- 0.50 dB. I measured -1.19 by this methodology at a gain slightly above 10 mV/div so the number is plausible
<azonenberg>
actual -3 dB bw is not cal'd, just "better than -3 at 4 GHz"
<azonenberg>
The pico cal cert says nothing about output voltage flatness
<azonenberg>
But mostly what i was interested in here was the major falloff from 4-6 GHz on the scope. And the measured data is at least somewhat plausible for de-embedding the frontend
<azonenberg>
basically the way i'm looking at this is, the waveform after inverting this transfer function is likely to be closer to the actual values than before
<azonenberg>
how much closer? no clue
<NeroTHz>
the only ´right´ way to cal the level is with a power meter that you can use to verify output level
<NeroTHz>
siggen could also work here, but not ideal
<NeroTHz>
but yeah, it should probably be good enough, indeed, the rolloff is what one would expect
<azonenberg>
Yeah. And i mean i'm already trying to push the lecroy >1 GHz past the datasheet bandwidth spec
<azonenberg>
so calibration is kinda out the window at this point
<azonenberg>
"better than nothing" is the goal lol
<azonenberg>
This actually seems to be working, the eye looks more open
<azonenberg>
not that i couldn't do protocol decoding on it before, but this is an improvement
<NeroTHz>
that is nice
<azonenberg>
pretty sure you're not supposed to be doing this :p
<azonenberg>
de-embedding the scope frontend antialiasing filter to push a 4 GHz scope to capture a 10.3125 Gbps signal
<bvernoux>
NeroTHz, I have bought that too ;)
<azonenberg>
but i mean it's 40 Gsps so i have plenty of sample rate before nyquist becomes an issue
<bvernoux>
NeroTHz, an old HP 53150A is not too bad ;)
<miek>
you're *supposed* to use the normal scope UI too, but oh well :p
<azonenberg>
miek: lol
<azonenberg>
lecroy doesn't sell a 64/66b or 10Gbase-R protocol decode for the waverunner 8000 series scopes that max out at 4 GHz and have BNC inputs on the front panel
<NeroTHz>
hehe :p
<azonenberg>
Wonder why
<azonenberg>
:p
<bvernoux>
NeroTHz, the good things in those old HW HP is the HP 53150A have a very very stable OCXO the best I have in my lab
<NeroTHz>
ah that is nice
<NeroTHz>
I have little need for a super accurate time-base, mostly just phase noise that matters for us