Current Server Time: 15:23 (Central Europe)

#apertus IRC Channel Logs

2019/09/25

Timezone: UTC


04:21
intrac
left the channel
04:22
intrac
joined the channel
06:45
Bertl_oO
off to bed now ... have a good one everyone!
06:45
Bertl_oO
changed nick to: Bertl_zZ
07:12
LordVan
joined the channel
07:25
LordVan
left the channel
10:48
danieeel
changed nick to: danieel
13:42
BAndiT1983|away
changed nick to: BAndiT1983
13:57
BAndiT1983
changed nick to: BAndiT1983|away
14:28
namibj
The USB3 offloading seemed at least barely usable. Any idea whether it wold work with one of these: https://www.corning.com/optical-cables-by-corning/worldwide/en/products/usb-optical-cables.html ? Or something of that sort in general?
16:54
Bertl_zZ
changed nick to: Bertl
16:55
Bertl
morning folks!
16:55
Bertl
I'm pretty confident those active converters/extenders will work but not sure how they help except for permitting longer cable ranges
17:00
jhlink
What speeds are you looking at for USB3?
17:00
jhlink
or targeting namibj
17:02
jhlink
And why that cable in particular?
17:26
namibj
jhlink: Significantly faster than what the Axiom can/could push out via ethernet.
17:27
namibj
jhlink: THis cable is really light, and I'd consider it as a tether for a multicopter/hexapod, so that proper data recording can be stationary.
17:31
Bertl
I'd really like to know what chips they use for the conversion
17:32
namibj
Bertl: these ones use 5Gbit multimode fiber transceivers. Unfortunately the usb connection on the axiom isn't realized via an fpga transceiver, and can thus not easily be run over a suitable SFP/SFP+ transceiver. usb3 is quite tricky with timings and all that, so mcuh that I'd not want to try without specialized hardware like this cornings extender.
17:33
namibj
Bertl: if they'd not be that expensive I'd like to use one for analysis. Maybe one can get a broken cable on ebay? I understand they fake a couple hubs, likely to accumulate the allowed per-hub delay to get enough for whatever optical PHY they use for it.
17:34
Bertl
well, the phy is not that slow, so timing should not be a big issue
17:34
namibj
usb3 timing is evil.
17:35
Bertl
yeah, but a single hub should be allowed a way longer delay than is required
17:35
Bertl
anyway, what do they cost?
17:38
jhlink
I'm just a bit skeptical. Even if the USB3 maximum transfers speeds are 5 Gbits, would either the host or the Axiom be able to support the respective read/write operations?
17:39
jhlink
If the Axiom Beta runs storage of an SD card, and the maximum R/W times are less than 1 Gbits, then the bottle neck will be flash memory, not the communication protocol
17:39
Bertl
on the AXIOM side we can do up to 3.2Gbit/s with the USB3 plugin
17:39
Bertl
and there could be two of those plugins in use at the same time
17:39
jhlink
What about ethernet? I'm a little hazy here, but my speculation is that speeds are protocol dependent.
17:39
jhlink
Probably wrong, but open to corrections! :D
17:40
Bertl
well, we 'only' have a single 1Gbit/s ethernet connection on the Beta for now
17:40
Bertl
and the maximum realistic data rate there is about 800MBit/s
17:41
jhlink
What makes you say that?
17:41
namibj
Bertl: corning? 150 for 10 to 550 for 50
17:41
Bertl
jhlink: the tests we did run so far
17:42
jhlink
Goot it. :)
17:43
Bertl
namibj: so about 10 bucks for a cable, that's quite cheap
17:55
BAndiT1983|away
changed nick to: BAndiT1983
17:57
BAndiT1983
changed nick to: BAndiT1983|away
17:58
namibj
Bertl: you mixed them up. 150 EUR for 10m, 550 EUR for 50m..... Honestly, I wish tey'd be affordable, because then I could easily test...
18:00
namibj
Bertl: can one push raw ethernet frames at full link rate? Like, jumbo frames, for all I care? ON a direct link there should be no packet loss anyway, and it's easy/cheap to rig a fiber link for that.
18:00
namibj
However, now there is food on the table. Cya
18:12
Bertl
namibj: ah, well, that's as expensive as expected then :)
18:13
Bertl
@ethernet, in theory yes, but we didn't test that yet
18:13
BAndiT1983|away
changed nick to: BAndiT1983
18:13
Bertl
still it's only 1Gbit at most
18:15
felix_
iirc the hard mac in zynq7 doesn't support jumbo frames; not sure if the pins used for the phy could be rerouted to the fpga fabric instead of the hard mac
18:15
Bertl
yes, AFAIK they can be used from the fabric
18:16
felix_
iirc someone also started an own camera design with 10g ethernet; not sure how the project was called and if it's public though
18:17
Bertl
10g ethernet in a small package would be nice
18:18
Bertl
but I haven't seen any existing solutions yet which would be usable ...
18:19
felix_
the design uses a zynq with gtps and a TLK10031
18:27
Bertl
interesting, that one could actually work with normal LVDS pairs as well at half the data rate
18:29
Bertl
the price with 20 EUR is high but not that high
18:30
Bertl
and the quad channel version 'only' costs 55 EUR
18:32
RexOrCine|away
changed nick to: RexOrCine
18:59
BAndiT1983
changed nick to: BAndiT1983|away
19:02
namibj
Bertl: a gigabit is a gigabit ;).... There are reasons why I considered asking trenz about solding a 10ghz trensceiver version, where the only real difference would be the higher specs and a metal lid over the flip-chip, in one of their micro modules. It would have been smaller than the zynq, with 3 high-density connectors on the bottom. I decided against and went for a genesys 2, because of the higher
19:02
namibj
specs and lower price. It's not suitable for an axiom form factor though.
19:03
namibj
felix_: so let it have no jumbo support. I don't care about the measly increase in datarate, just about the lower scheduling overhead due to lower packet per second counts.
19:04
namibj
Bertl: you mention a theoretical feasibility to use 10g ethernet on the apertus board? Or am I missunderstanding you?
19:09
Bertl
well, we have two plugin slots with a total of 12 LVDS pairs capable of at least 1Gbit/s each
19:09
Bertl
so with the 'right' plugin 10G should be doable, at least in one direction
19:13
namibj
basically TE0745-02-30-2IA with the FB part swapped out for a FF part. Nice and compact, no more than 4/6 layers should be needed on the host. But as said, I didn't go for it, and I gotta ask the guy from Kron Technologies I met @camp for telling me which sub-500$ fpga board he mentioned that could handle 10G ethernet on SFP+.
19:15
Bertl
and that's how you ended up with the genesys 2?
19:15
Bertl
(because that's not sub 500$ :)
19:16
namibj
Bertl: yeah.... wait, this is unidirectional? Uh-oh, most transceivers afaik don't like this. But it might be possible nonetheless. Any idea for an affordable serializer to feed this though? otherwise one'd have to stick with 1G optics, and use a bundle of them with a 12-fiber cable. That would leave 2 for a return channel though, and require a switch on the remote end which you could tell to route
19:16
namibj
packets appropiately.
19:17
namibj
Bertl: I bought it last year ;). ALso I got academic discount.
19:19
Bertl
well, for raw unidirectional ethernet frames, you can just drop the received data
19:19
namibj
While I'm at this: could one theoretically transmit the sensor outputs single-ended to the fpga? So that it'd be possible to detach it to place it in the middle of a telescope tubu, with laptop-screen-edge-coax cables in a bundle for transmitting it without much optical obstruction? Or just general detached sensor?
19:19
Bertl
you might also make it 10G/1G bidirectional
19:20
Bertl
depends on the read out speed you want to achieve from the sensor
19:20
namibj
Bertl: it's about whether the optics themselves play with this, considering that you'd want to skip the useless fibers to stay within the cheap/light 12-fiber cables.
19:20
namibj
Bertl: 600MHz?
19:21
Bertl
the sensor can do 1 to 64 LVDS pairs for data
19:21
Bertl
so if you are happy with one data and one control channel
19:21
namibj
I know. I meant per channel. If I use it with the genesys 2, I'd try for 64.
19:21
Bertl
you can easily feed that through a cat 6 cable :)
19:22
Bertl
and at 600MHz you are probably fine for some distance
19:22
namibj
Oh, I am not after those distances. I mean, over like 50cm _max_.
19:22
Bertl
50cm is a lot for 67 LVDS pairs
19:27
namibj
The cable has low loss, well, somewhat. It's a fine coax: http://www.micro-coaxial.com/wp-content/uploads/2019/01/Sumitomo-HITACHI-micro-coax-cable-assembly-factory.jpg
19:42
namibj
Bertl: 15 might suffice if I'd push it, assuming it's for the 400mm f/2 design from https://www.telescope-optics.net/field_flattener.htm example 1 deign point 7 (the 7is circled in).
19:45
namibj
This has sufficient sharpness to get 90% ensquared energy, i.e., a planar wavefront would result in 90% of the photons getting to the sensor hitting a single pixel, as long as the angle was centered.
19:47
namibj
This is already at the edge of the sensor, btw., with 2 degree off-axis (= 4 degree diagonal FOV).
22:10
se6ast1an_
off to bed, good night
22:17
Bertl
nn
00:48
RexOrCine
changed nick to: RexOrCine|away