<SingularitySurf>
Hey thanks for all your hard work on open source tools (esp. scopehal/glscopeclient ofc)!! Would be awesome to maybe have a cheapish 1Gsample "acquisition board" at some point so everyone can build their own flavor of "backhaul" behind that! Unf. I think I lack the tools and skills to be of much use for that right now..
<SingularitySurf>
And stay safe everyone! :)
SingularitySurf has left #scopehal [#scopehal]
<awygle>
lain: do PCIe cards have to do fcc emissions testing?
<awygle>
I thought they fell under the "not a thing, just a part of a thing" rule
<lain>
awygle: ah yeah I was referring to integrated/pre-assembled systems, not sure how bad interference is in a typical pcie card
<awygle>
Mk
Degi_ has joined #scopehal
Degi has quit [Ping timeout: 250 seconds]
Degi_ is now known as Degi
<Degi>
Huh, it took whole day to design the adapter. I think I'll finish it up tomorrow
zigggggy has quit [Remote host closed the connection]
_whitelogger has joined #scopehal
bvernoux has quit [Ping timeout: 250 seconds]
_whitelogger has joined #scopehal
zigggggy has joined #scopehal
_whitelogger has joined #scopehal
<azonenberg>
So i'm thinking what i might want to do next is plugin support
bvernoux1 has quit [Quit: Leaving]
_whitelogger has joined #scopehal
wbraun has quit [Ping timeout: 272 seconds]
wbraun has joined #scopehal
LeoBodnar has quit [Ping timeout: 256 seconds]
LeoBodnar has joined #scopehal
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
_whitelogger has joined #scopehal
futarisIRCcloud has joined #scopehal
<Degi>
Versa doesn't have PCIe x4 and costs like 200 €
<Degi>
(well now this project costs close to 200 € too but at least it has x4)
<Degi>
And the versa has a 45k, not 85k
<electronic_eel>
Degi: how do you plan to mount the scope components in the end?
<electronic_eel>
the fpga will probably be on a pcie-card, but how will the scope channels connect?
<Degi>
Hm I think I'd make a custom PCB for that. The end goal is having the ECP5 with the ADC connector and PCIe connector on one PCB
<Degi>
The ADC+AFE will be on one module with 60 or 120 pins (I forgot the connector model)
<Degi>
QTH-030 or QTH-060
<electronic_eel>
but do you want to build a case which houses your pc and the scope channels in one thing?
<Degi>
Nah its just a PCIe addon card
<electronic_eel>
but will the scope channels then just hang out of the back or what?
<Degi>
Maybe a PCIe backplate with SMAs mounted to it, like a GPU
<Degi>
I mean if you want it on your desktop, you could use SMA cables to something like a SMA patch panel (or BNC? not sure which connector we want to use)
<electronic_eel>
ok, but then the back of your pc will face to your bench because you screw in the probes there?
<Degi>
Hm I mean I didn't really think about using this with probes, but maybe an option to have a 5.25" drive cover with SMA connectors could be made
<electronic_eel>
I don't know how your electronics bench is set up, but I wouldn't like to have to work with that
<electronic_eel>
the rack/half rack solution azonenberg plans sounds much more practical to me
<Degi>
I mean it could be offered in a custom housing where the SMAs come out the front, but my general goal was a card which you can put into every PC and then connect it to some apparatus
<electronic_eel>
how about having the scope channels in a rack / half rack box like azonenberg plans, have some line driver thing or a smaller fpga there and then send the data over some wire (usb, sata, sfp+,...) to the card in the pc?
<Degi>
But yes, that's a good point on how to use it as a scope
<Degi>
I mean you could stick the card into a server (which was my original plan), but yes, a sender card could be done too (just amplify the channels of the ADC to 6.6 V pk-pk LVDS and then send that over a long cable, though I'm not sure how long because the skew needs to be less than 500 ps or so)
<Degi>
Like I have no idea of housing design to be honest (most of my projects have no housings)
<electronic_eel>
proper cases are a lot of work, and need a bit different skillset than the electronics inside - so I can fully understand to build something without a case
<electronic_eel>
but if you want a proper tool that you want to use every day, I think you need some kind of case
<electronic_eel>
unfortunately cases (and therefore your project) look ugly if you don't spend some time and care on them
<electronic_eel>
that can ruin the fun in showing your otherwise successful project to others
<electronic_eel>
idea: re-use the case azonenberg choses for the scope channels, just mount a different digital/transceiver board inside
<Degi>
Hm maybe a small SMA patch panel box, which you can connect with 1-2 meters of SMA cables to your PC, could be sufficient as a desktop unit? Otherwise I could stick the ADC boards into a small desktop box which then connects with some very wide bus to your PC but that's a bit impractical.
<Degi>
2 m SMA extender cable on ebay is 1.38 $
<Degi>
And maybe a rackmount SMA patch panel (but I'm not sure whether most people who use scopes have a rack nearby their work bench? The closest thing I have to a rack is a metal shelf where the mounting holes are approximately spaced like in a rack.)
<Degi>
Huh the ECP5 SERDES has a 150 mV receiver sensitivity... Hm so the mainboard needs to be transmitting with at least 0.67 V that the received signal is within spec (13 dB loss)
<azonenberg>
i'm choosing ethernet as the interface specifically because pcie cards are a pain to use unless you have the back of a pc right there
<azonenberg>
with ethernet you just need a cat5 or multimode drop somewhere nearby, run a cable to it, the pc can be anywhere
<azonenberg>
if you don't mind low WFM/s rate you can even use wifi
<azonenberg>
also very easy userspace app development with no need to spin custom drivers
<azonenberg>
pcie is high bandwidth, but at the expense of basically everything else - one bug in the driver and you kernel panic
<azonenberg>
not portable, has to be installed in a specific spot on a full sized desktop pc or rackmount server
<azonenberg>
Ethernet goes up to 40 Gbps with relatively easily available FPGAs (kintex-7 -2 or -3 speed grade) which is faster than usb3, lacks the need to deal with patents/ID registration like usb and pcie
<azonenberg>
40 Gbps is a match for pcie 2.0 x8 line rate, and faster than 3.0 x4
<azonenberg>
100G is a thing, if you've got the money for it
<azonenberg>
ethernet can be bridged between arbitrarily fast/slow PHYs down to 10baseT, run over a vpn, etc. So much more flexibility
<Degi>
Oh neat
<azonenberg>
Basically for general purpose high speed data capture it seemed like the best option
<azonenberg>
10G cards are coming greatly down in price too
<Degi>
Yes that's true. I mostly chose PCIe because that fits the ECP5
<Degi>
And because it doesn't require any additional hardware.
<Degi>
(And because ECP5 has FOSS toolchain. I really hope better FPGAs get that soon too)
<azonenberg>
Yeah. I'm a xilinx shop because i have a lot of xilinx hardware already, and kintex7 has 10g transceivers
<azonenberg>
lack of foss toolchain is annoying but fundamentally, my first priority is to get my work done. Doing it with foss tools is a bonus
<azonenberg>
Same reason i bought sonnet instead of using openems, but do my boards in kicad
<azonenberg>
when the foss tool is usable, i use it. When not, "don't get my work done, but use foss tools" is not an option
<azonenberg>
Dont have time to do a proper design review now, @ $dayjob
<Degi>
Okay
<electronic_eel>
what is the cutout for?
<Degi>
The one in the middle? There's the JTAG header on the ecp5
<electronic_eel>
ah, so the ecp5 board is directly screwed in?
<electronic_eel>
into the smas
<Degi>
Yes
<Degi>
The 3D rendered image is a bit outdated though, some soldermask has been removed around the connector.
<miek>
how are you going to tighten up the sma connectors?
<Degi>
Probably with tweezers
<Degi>
Or I'll tighten them before soldering and then later I can figure out how to remove them
<miek>
no chance of doing that with tweezers, and there's no compliance anywhere so alignment has to be perfect. also that's like > $100 in sma connectors
<Degi>
Oh I got them for like 20 €
<Degi>
Sure, on mouser that's 100+ but not those not-really-gold plated connectors on ebay
<Degi>
How do you mean that there's no compliance? The only connector that's probably going to be a problem is J14 on the ECP5 EVN (the one surrounded on all sides)
bvernoux has joined #scopehal
<Degi>
There's enough space around them to turn them step by step
<Degi>
Hm besides the surrounded one, which seems kinda like a problem
<Degi>
But that one can be rotated with two sticks. They don't need much torque to screw in after they've been used once.
<awygle>
40 GbE is definitely faster than 2x8
<awygle>
"no compliance" meaning no play in the connectors. they have to line up perfectly.
<awygle>
as opposed to e.g. SMP with bullets which can stand some small amount of deflection
<awygle>
assemble then solder is your best bet, the mounting holes will have some play. bit weird though.
<awygle>
also, hello, good morning
<Degi>
Hm SMP seems a bit pricy
<awygle>
oh yeah don't actually _use_ SMP
<awygle>
just saying
<awygle>
that's what miek meant
<Degi>
I'll just use pre-screwed SMA
<awygle>
solder to EVN, screw the mates to the SMAs then solder them to your adapter, that's my suggestion
<miek>
i think it would make a lot more sense to spin a board with the ecp5 on it, rather than trying to use the EVN
<awygle>
i agree with that. or just buying the Versa
<awygle>
but :shrug: if you want to make this work, that's how i'd do it
<Degi>
I mean the versa has only x1, that's my problem with it. Hm making a board would be an idea, though then I'd need to figure out where to solder the BGAs etc since I don't have a reflow furnace at home.
<Degi>
If I'd spin my own board now, it'd cost me more than 80 € (PCB + mask + solder paste) and I'd rather do that when more design details are clear
<Degi>
Hmm I could probably reflow on an uniformly heated plate.
bvernoux has quit [Quit: Leaving]
<zigggggy>
anyone run ubuntu server?
* zigggggy
pokes azonenberg
<Degi>
Are there any plans yet for clock generation for the HMCAD?
<azonenberg>
Degi: my FPGA board is going to contain an input for an external reference clock, an internal TCXO or OCXO (probably TCXO to keep costs down since it's an entry level unit) then a LMK04806 PLL or similar
<azonenberg>
also, i would recommend using sma cables rather than mating like you have there if at all possible
<azonenberg>
Also, re BGAs, they're not that hard to do yourself. Hardest part is placement
<azonenberg>
a cheap walmart toaster oven is the way to go
<azonenberg>
mine cost me maybe 30-50 USD and has a convection fan, which greatly improves results
<Degi>
Hm why would SMA cables be better?
<Degi>
Also how hard is it to reuse BGAs?
<azonenberg>
Reballing is a pain and generally best avoided. If it's a suuuuuper expensive part maybe
<azonenberg>
Depends on how much practice you have, if you have the right parts, and how much you value your time vs the chip
<Degi>
Like 50 €... So I can't just apply a film of solder paste and oven it and scratch excess off, right?
<azonenberg>
No. You can't stencil paste either, you need too much solder for that to be viable
<azonenberg>
you need to smear flux over the whole thing then individually place a solder ball on each pad
<azonenberg>
(after carefully removing any residue of the old solder balls)
<Degi>
I guess if you apply paste, the balls get connected with each other? Hm I mean for 381 Bballs that's not too bad...
<azonenberg>
generally this is done with some sort of stencil or jig
<Degi>
I guess I can't just not reball it because then some contacts won't have solder contact with the PCB, right?
<azonenberg>
If you pull the chip it has to be reballed before you can reuse it. There's no other viable option
<Degi>
Huh the LMK0480x parts look nice jitter-wise...
<azonenberg>
Yes. I'm actually also building a dedicated rackmount clock distribution platform
<azonenberg>
as soon as the AFE is done i'll be moving back to work on it
<azonenberg>
1U system with ethernet control, GPS 10 MHz + PPS inputs, internal OCXO for use without GPS sync
<Degi>
Nice
<azonenberg>
five front and five rear single-ended PECL outputs, with individually adjustable dividers off a 2.5 GHz VCO + individually adjustable phase in 25ps steps
<azonenberg>
(front and rear share same phase and divider value due to limitations of the PLL)
<Degi>
Or something from the 090-443* series for higher end versions? (The 090-44310-32 looks cheapest)
<azonenberg>
i.e. 1A (front left) and 1B (rear left) are the same frequency, and can either be at zero phase or an adjustable phase, but the two can't be adjusted independently
<Degi>
Yes, that's a pretty nice chip.
<azonenberg>
additionally a bunch of outputs driven via an FPGA that can have arbitrary duty cycle and divide, but more jitter than the PLL outputs and no phase control
<azonenberg>
And no i'm not going to put a rubidium standard on there
<azonenberg>
GPSDO's get almost as good for a tiny fraction of the price
<Degi>
Hm right
<azonenberg>
LeoBodnar makes a GPSDO i'm interested in using, we've talked about him making a module form factor i can integrate into my system
<azonenberg>
and just have an antenna port
<Degi>
Well rubidium has the upside of turning on faster I guess, but leaving the clock gen on 24/7 shouldn't be that much of a problem...
<azonenberg>
You can still get 10 MHz off the OCXO or GPSDO during startup
<azonenberg>
it just won't be slaved to GPS time/frequency yet
<monochroma>
ribudium takes a while to lock, it's not instant on
<monochroma>
it has to warm up etc
<Degi>
Hm right, to vaporize it...
<azonenberg>
for my purposes, i need stability and low jitter but don't care much about the absolute frequency
<azonenberg>
so i may not even use the GPS lock functionality, although i'll provide it for those who do
<Degi>
I think for me low drift would be important
<azonenberg>
my main goal is to be able to synchronize the timebase of all of my scopes, siggens, etc
<Degi>
Neat
<azonenberg>
I'm also going to have a deskew calibration input port on it
<azonenberg>
connect any set of cables from output ports to the deskew input, one at a time, and push "deskew"
<azonenberg>
and it will adjust phase so the rising edge hits the end of all of the cables simultaneously
<Degi>
Neat
<Degi>
Hm yes, that PLL phase adjustment is rather nice.
<azonenberg>
regardless of electrical length of the wires
<azonenberg>
glscopeclient is also going to support trigger deskew for multiple scopes eventually
<Degi>
I got a PLL chip cause it can do LVDS, but this one is so much better and the ECP5 can do LVPECL 3.3V as well...
<azonenberg>
the idea being you touch a probe from each scope to an arbitrary signal and it'll tweak trigger phase until they line up
<azonenberg>
to compensate for propagation delay of trigger out on the first scope to trigger input on the second
<Degi>
Yes that's nice
<azonenberg>
It's planned, but i havent had time to sit down and write the code yet
<Degi>
How's the electrical length dependency temperature wise? Like if I heat an ordinary SMA cable by like 10 °C does that make any difference?
<azonenberg>
So right now while multi scope "works" it's bare bones and you're on your own for the setup
<azonenberg>
very likely yes
<azonenberg>
The deskew can be repeated at arbitrary times, of course, but the normal solution to this sort of thing is to try and maintain stable temperature in your lab :p
<Degi>
Well my floor is like 3 °C colder than desk level
<Degi>
Hm are any optical probes planned, to record light?
<azonenberg>
You mean like fiber inputs? or for free space optics like a tv remote?
<Degi>
Either is fine
<azonenberg>
for the latter, you can just build a circuit with a photodiode and probe it using a normal electrical probe, or put a sma output on it
<azonenberg>
for fiber input, i think you can just use a SFP/SFP+ module with a linear output stage rather than a limiting amplifier
<Degi>
Hm yes, guess that would work
<azonenberg>
They exist, but you have to look a bit
<Degi>
Hm do SPFs have by default a limiting amplifier / comparator?
<azonenberg>
Yes. So you need to explicitly look for a linear one if you want to record the shape of the waveform and not just recover data
<azonenberg>
It looks like the FS 10Gbase-LRM SFP+ has a linear RX interface but is 1310 nm wavelength not 850
<Degi>
Oh well, mouser has photodiodes with 0.4 ns risetime for 15 €
<azonenberg>
That's $21 and has all of the electronics integrated. Supply 3.3V and get a waveform out
<Degi>
Well as long as its reactive to 1064 (and maybe 532)
<azonenberg>
I think you can find linear RX's for 850 too, just might have to look harder
<Degi>
Well as long as it has less than 90 dB attenuation at 1064 nm, I'd probably get an useful signal, at least once
<Degi>
You can get a 100 lot for 200 $ of the SFPs... huh
<azonenberg>
well ebay'd connectors, not so sure :p
<azonenberg>
but the big problem is mechanical
<Degi>
Yes I hope I won't need to unscrew it
<Degi>
(I mean its better than ebay'd connectors + ebay'd cables which I heard have like 30 ohm impedance? You know the 1.5 $ ones)
* azonenberg
looks over at CCSMA-MM-086-6 cables in bin...
<Degi>
I have like one copper pipe shielded SMA cable from university, I bet that has low loss
<Degi>
And they look actual gold plated, compared to the 1.5 $ cables where the "gold" oxidizes
<lain>
beryllium copper more likely
<electronic_eel>
semirigid is mostly used for isolation
<Degi>
Thinking about it, I have probably a gram of beryllium in my bedroom, distributed in electronics...
<azonenberg>
actually my current "high end" rf cables are minicircuits FL086-24SM+
<lain>
azonenberg: how are you liking them btw? :3
<Degi>
Also I have a SMA with an approx 5 mm thick coax line
<azonenberg>
lain: so far they've performed great. I don't have anything that can push their limits yet
<azonenberg>
i recall them being ~0 loss for everything i've used them on so far
<Degi>
Neato
<Degi>
Is the insertion loss of coax cables the S12?
<azonenberg>
S12 or S21, depending on which end of the cable you call port 1 vs 2
<azonenberg>
in theory they should be identical
<Degi>
Oh nice, so 1.5 dB loss at 10-18 GHz...
<azonenberg>
To be technical, actually, S-parameters are the signal going through
<Degi>
I have several meters of some 2 mm coax here which gets unusable at 1 GHz...
<azonenberg>
and insertion loss is the signal that doesn't make it
<azonenberg>
i.e. assuming no reflection, if you have S21 of -1 dB the insertion loss is +1 dB
<Degi>
Ah right
<azonenberg>
i prefer s-parameters because the sign and meaning are unambiguous
<azonenberg>
also "return loss" vs S22/S11 doesn't say which end the reflection is coming from
<Degi>
Hm this solid copper coax has rather weird looking SMAs... The insulator in the middle has a metal ring around it, maybe for increased frequency range.
<monochroma>
might not be SMA
<monochroma>
there are a bunch of high frequency connectors that look similar to SMA but are very much not
<lain>
Degi: that almost sounds like triax
<Degi>
The metal is connected to the shield though
<Degi>
Well it does mate with a standard SMA connector
<monochroma>
...
<monochroma>
if you don't know for sure if it's sma, don't mate it
<Degi>
Hm yes, that's true
<Degi>
Huh, this thick SMA cable is labelled "times microwave SF-142B" and apparently that has 0.76 dB/feet at 21 GHz, that's nice