<marcan>
whitequark: ack, still tying up loose ends, noted
<whitequark>
marcan: no rush, just following once-weekly schedule :)
carl0s has joined #glasgow
Exec1N has joined #glasgow
Stormwind_mobile has quit [Ping timeout: 268 seconds]
Stormwind_mobile has joined #glasgow
<whitequark>
electronic_eel: question: would you be up to making a Glasgow ADC/DAC addon board that used the main connector(s)?
<whitequark>
maybe something like 10 Msps, with 4 channels of each [could be two boards?], ideally one differential ADC channel, with a narrow range (0-1 V is probably fine)
<whitequark>
the aim is to be able to play with analog interfaces like VGA and composite video, pretty much in line with my work to date
<whitequark>
this is something I have (or at least can acquire) enough skill to do, but for personal reasons right now it is hard for me to design and prototype PCBs specifically
<whitequark>
actually, this might not be possible
<whitequark>
I've only really seen SPI ADCs, parallel ADCs, and high speed serial ADCs
<Stormwind_mobile>
If you make the differential ADCs go down into the 100mV range full scale, the system can suddenly become attractive for scientists.
<whitequark>
ehhhhh
<whitequark>
not really, i think
<whitequark>
there's a bunch of boards catering specifically to scientists that would be much more convenient
<whitequark>
redpitaya, for example
<Stormwind_mobile>
I'm looking right now for a multichannel thermocouple interface, that ideally talks to LabVIEW. NI offers something like this, of course, 16 channels for 1395$.
<whitequark>
I feel like you're better off populating a board with some chips that can do it--basically just an AFE--and then plugging it into glasgow so you can drive it easily
<whitequark>
skipping the whole "add a microcontroller, and some port, and some protocol ... " thing
<Stormwind_mobile>
Just a spontaneous thought, if it's not difficult to implement, and doesn't hurt, not much effort on the line, is it?
<whitequark>
I am not opposed to it
<whitequark>
more that trying to be everything for everyone can only be stretched so far
<Stormwind_mobile>
I admit I don't know much about how Glasgow works, and if it's a good idea in the end.
<whitequark>
not many people are using glasgow right now, but once that changes, i am sure plenty of people will find manny aspects frustrating
<whitequark>
would you like me to tell?
<Stormwind_mobile>
Is there already an estimated price for a Glasgow board?
<whitequark>
esden indicated he is not comfortable giving an estimate yet, so essentially, there isn't
<Stormwind_mobile>
Well, if theres the option to hide all that complexity behind an API or a script, I could see opportunities there. I wouldn't try discussing this in chat, though, and I need to know much more before I make bold proposals.
<Stormwind_mobile>
Fair enough.
<Stormwind_mobile>
Who of you is planning to attend 36c3?
<whitequark>
wait, what complexity?
* tnt
raises hand
<whitequark>
(and why isn't it useful to discuss it in chat?..)
<Stormwind_mobile>
So what are the frustrations you wanted to tell about?
<whitequark>
oh, I don't know yet, that's the whole point
<whitequark>
people using tools in the wild and out of sterile lab environment always find the weirdest bugs
<whitequark>
or not even bugs
<Stormwind_mobile>
Well, I have a habit of making very aspiring propositions, talking about them in big steps, and losing people listening to it. I don't exactly know how I do it, but it still happens frequently Xd
<whitequark>
but for example, i'd expect people to be, at least, frustrated with the deployment strategy, esp on Windows; and with general slowness, again esp on Windows
<Stormwind_mobile>
And there are a lot of aspects that would require pages of text.
<whitequark>
i expect people will be frustrated that /usr/bin/glasgow starts up for 10 seconds on a raspberry pi 3
<whitequark>
or that it wouldn't let you saturate the full USB 2 bandwidth on a 10-year-old PC, or even 5-year-old
<Stormwind_mobile>
But then again, that's where I might be of use: I also have a habit of poking around in software or a software/hardware system, and finding funny things fast, that developers probably didn't ever think about.
<Stormwind_mobile>
Sure, Glasgow isn't a commercial product, and the resources for dealing with corner cases will be limited. So I'm not sure how useful it is knowing about corner cases when there are no resources to address those.
<whitequark>
indeed it is sometimes not. on another project of mine that contributed to burnout; that said, the reporter considered themselves an especially important person, and wasted no time demanding that his obscure bugs or idiosyncratic preferences be supported
<whitequark>
which made that a lot worse than just having a list of mostly-unactionable bugs
<whitequark>
anyway, I'm curious about your proposal still
<whitequark>
sometimes, listening to people describing their ideal workflow without pesky constraints of "reality" can be enlightening
<whitequark>
they call this "ideal final result" in TRIZ
<whitequark>
electronic_eel: actually, what i request might not even be possible. looks like no one makes QSPI ADCs. AD has some that are "QSPI-compatible" but I'm pretty sure they just mean using a QSPI-capable host in SPI mode
<whitequark>
oh wait, it would be possible to use a delta-sigma modulator for this, wouldn't it
<Stormwind_mobile>
What's the intended use for the ADC board you have in mind?
<whitequark>
playing with things like VGA, composite video, and whatever other old analog tech we have
<Stormwind_mobile>
Indeed I'm very consciously trying to "not be *that* guy", just boldly demanding the world turning around him. Makes no sense to me why someone would do that, anyway.
<whitequark>
yeah, I'm pretty sure you'll do fine. just... sharing that story, since it is both relevant, and has left its mark.
<Stormwind_mobile>
*nods*
<Stormwind_mobile>
Practicing self care and knowing, as well as protecting one's boundaries - lessons for life.
<Stormwind_mobile>
What makes QSPI the interface of choice?
<whitequark>
SPI tops out at maybe 30 MHz, generally
<whitequark>
even 8 bit at 10 Msps is 80 MHz
<whitequark>
QSPI reduces that to 20
Stormwind_mobile has quit [Ping timeout: 240 seconds]
Stormwind_mobile has joined #glasgow
craigo has quit [Ping timeout: 250 seconds]
Exec1N has quit [Ping timeout: 268 seconds]
carl0s has quit [Remote host closed the connection]
<sorear>
10 Msps seems a little low for VGA, 28 mhz pixel clock?
<whitequark>
oh yeah, VGA won't work that way
<whitequark>
sure
<whitequark>
composite only then
<whitequark>
or component
<sorear>
probably fine if you don't need the full 640/720 columns
<electronic_eel>
about an analog addon board: unfortunately there are a lot more different features required for analog than for digital interfaces
<electronic_eel>
high precision but slow / high speed high resolution / high speed low resolution - and all that multiplied by different voltage levels, differential / single ended and so on
<electronic_eel>
so an addon becomes quite application specific
<electronic_eel>
one that would work for vga and composite video would be of very limited use for rf work
<electronic_eel>
or impedance spectroscopy or ...
<whitequark>
yes, that i understand
<whitequark>
i think sdr is not a niche i want to push glasgow into, at the moment. the needs on every level are just too different
<electronic_eel>
so i wouldn't call it "analog addon board" but "analog video addon"
<whitequark>
right
<whitequark>
i think even cvbs in+out would be really fun. not sure how interested you'd be though.
<electronic_eel>
right now I want to get the test jig and selftest to work, after that continue working on the protection board
<electronic_eel>
I don't want to promise stuff too much into the future
<electronic_eel>
leads to long todos, pressure, it not being that much fun anymore
<whitequark>
ack. forget about it then. i'll throw it together myself or just ignore it for a few months.
<electronic_eel>
"i'll throw it together myself" - you seem to be really set on getting this capability for glasgow
<sorear>
seems to me that the interesting constraint is how fast you can run the glasgow <-> AFE connection
<sorear>
"high resolution high speed" is not a design point, there's a necessary tradeoff
<whitequark>
you can run it pretty darn fast over LVDS
<whitequark>
like hundreds of mbps i think?
<sorear>
oh I understood "that used the main connector(s)" to mean "not LVDS"
<electronic_eel>
the problem will be the speed to the host i think
<whitequark>
electronic_eel: well, yes and no. i find analog video captivating for some reason. but mostly, i find my inability to quickly throw together a PCB irritating, and try to get rid of that, as i know exactly why it happens. not successfully so far tho
<whitequark>
that's a huge part of the reason for why your SMT protoboard interested me so much
<electronic_eel>
did you already try it?
<sorear>
I'm assuming that the AFE would be used with multiple applets, and the basic Glasgow pattern of "do as much as reasonable, esp. everything real-time, on the FPGA side" would continue to apply
<whitequark>
sorear: oh, yeah. I was thinking of QSPI ADCs specifically since for this application that would be a perfect interface. (i think?) if you just wanted a pipe as fat as possible on revC you'd use the LVDS port.
<sorear>
what's the theoretical limit for LVDS right now
<whitequark>
electronic_eel: (speed to host), well, you get 300 Mbps, which is quite a bit
<whitequark>
electronic_eel: (try it) yes but in a redux sort of way. i soldered OPM and some pin headers to it. it does not behave quite the way i intuitively expect it to
<whitequark>
btw did you design it for PbSn? AgCu? both?
<electronic_eel>
whitequark: the question is how much quality you want and how much processing can be done on the fpga
<electronic_eel>
I usually use PbSn for prototyping
<electronic_eel>
as I wrote on the github page it takes a bit of a different soldering technique
<electronic_eel>
my main usage is to prototype small modules / sections of a bigger circuit and check they behave as planned before making a more complex board
<whitequark>
yeah, I expected a learning curve, and I found it
<whitequark>
no surprises there
<electronic_eel>
I find it takes too long making complex boards this way, compared to a low-density pcb done in kicad and having it fabbed
<sorear>
do they make ADCs which can be configured to trade off resolution and sampling rate on the fly?
<electronic_eel>
sorear: delta sigma works this way by concept
<electronic_eel>
but for higher speeds other architectures work better
<sorear>
I am much more familiar with ADC/DAC theory than I am with available parts and boring details of circuit design
<sorear>
s/boring/practical/
<sorear>
very high speed ADC (>1 Gsps) have pretty inflexible architectures but we have no reasonable way to transfer that much data to the hx8k so it's moot
<sorear>
but for something like SAR you could have a variable number of approximation steps per sample
<sorear>
*waves hands more*
<whitequark>
i think this is often configurable
<electronic_eel>
sorear: when you build a board, you can't just take the ADC-IC, you have to add a analog frontend to make it usable. Usually a low pass filter stage, for some applications diff amp, pga,...
<electronic_eel>
you'd have to adapt all that too to make it usable
<whitequark>
in theory you could have a board with a bunch of alternative assembly variants
<whitequark>
bring your own LPF values
<whitequark>
but... i'm not an analog designer at all. i mean i can follow an appnote, and that's about it, never had more success with analog design
<sorear>
also strongly reminiscent of conversations we've had about scope frontends back when FREESAMPLE was last active
<electronic_eel>
you'd need different opamp models, dc blocks and so on
<sorear>
I'm assuming less "assembly variants" and more "analog muxes"
<whitequark>
i guess i can understand a circuit someone else's made if it has like 2 moving parts
<electronic_eel>
so either modules to plug in or relays
<sorear>
since the point of this exercise is to be a tool you can reconfigure without soldering
<whitequark>
no, i mean just solder what your specific application needs
<whitequark>
i know you could make a reconfigurable AFE. it would be huge and expensive and complex
<whitequark>
and probably not that flexible anyway
<whitequark>
so do just one step above "design a PCB per application": "Assemble a PCB per application", with the same topology or a few alternative ones
<sorear>
jumpers, maybe
<electronic_eel>
I'm not sure one pcb with lot's of unpopulated parts is the best way to get that - I suggest to do a base board with plug-in-headers then
<electronic_eel>
so when you need a totally different frontend, you can design just that and plug it into the existing adc board
<sorear>
everything in this project has been trying to strike a tradeoff between flexibility, robustness, and cost
<electronic_eel>
one thing i could use a vga input board for would be a proper open source ip kvm for vga
<electronic_eel>
i mean there are several ip kvms out there, but most of them need some ancient java module, require some windows client or similar
craigo has joined #glasgow
<electronic_eel>
i already experimented with some video grabbers and converters, but they are all unusable because they are much too slow to react when the resolution changes
<electronic_eel>
and since the graphics mode changes quite a lot on boot and you need to react fast to get into some adapter bios or such, they are useless for this application
Exec1N has joined #glasgow
<whitequark>
hm
<whitequark>
oh, ip kvm
<electronic_eel>
I've seen that the new beaglebone ai has some video processing hardware on it and partly considered to try it for this task - but I ditched it when i considered how much other stuff i want to try out...
<electronic_eel>
I'd have to add some adcs to the beaglebone ai too, there is just the digital part on it
Exec1N has quit [Ping timeout: 265 seconds]
Exec1N has joined #glasgow
Exec1N has quit [Ping timeout: 250 seconds]
nicexe has joined #glasgow
<Stormwind_mobile>
Are there FPGAs with integrated, fast ADCs? That would be the best case scenario, wouldn't it?
<Stormwind_mobile>
The arrow FPGA contest also offers an "AnalogMAX" devboard. Sounds like it might be worth looking into. Not really a Glasgow topic, though.
<tnt>
Stormwind_mobile: Xilinx RFSoCs
<tnt>
multi gsps ADCs
<tnt>
Smallest board is 9k$ though
<mwk>
some gowin fpgas have adc inside, I think
<mwk>
also um
<mwk>
every virtex4+/series7/ultrascale fpga has builtin adc