Current Server Time: 22:25 (Central Europe)

#apertus IRC Channel Logs

2017/03/17

Timezone: UTC


00:02
Alvis_
left the channel
00:06
Bertl
off to bed now ... have a good one everyone!
00:06
Bertl
changed nick to: Bertl_zZ
00:11
Alvis_
joined the channel
00:18
BAndiT1983
changed nick to: BAndiT1983|away
00:23
arpu
left the channel
00:36
arpu
joined the channel
01:31
Elbehery
left the channel
01:54
usmankhan
left the channel
02:16
Alvis__
joined the channel
02:16
Alvis_
left the channel
02:23
jucar
joined the channel
02:43
Rex0r
joined the channel
02:59
RexOrCine
joined the channel
02:59
RexOrCine
left the channel
03:06
RexOrCine
joined the channel
03:17
usmankhan
joined the channel
03:25
usmankhan
left the channel
03:48
usmankhan
joined the channel
03:51
Spirit532
left the channel
04:07
usmankhan
left the channel
04:21
jucar
left the channel
04:27
usmankhan
joined the channel
04:37
usmankhan
left the channel
06:02
Bertl_zZ
changed nick to: Bertl
06:02
Bertl
morning folks!
06:12
Alvis__
left the channel
06:13
usmankhan
joined the channel
06:14
usmankhan
Hello Bertl, good morning
06:15
usmankhan
I am wondering if we can use a multiplexer to manage all sensors using a single adc (I am guessing our switching frequency is high enough so this won't be a problem)?
06:22
Bertl
we probably could use a 4 or 8 channel ADC, yes
06:23
Bertl
i.e. a device with integrated mux
06:24
Bertl
the main question for me is, do we need to digitize the reference signal (DAC) as well to know what we want to follow?
06:25
Bertl
or do we add some kind of low ohm power switch and use the switching output for the reference voltage as well
06:35
Bertl
http://vserver.13thfloor.at/Stuff/AXIOM/BETA/axiom_beta_sensor_cmv12000_zif_v0.18_r1.2.pdf is one example of the sensor frontend
06:35
Bertl
on the second page, you see that we have a bunch of voltage related connections
06:36
Bertl
first there is the block of *_CV, *_UV, *_OV pins
06:37
Bertl
they are designed to provide some means to check if a given voltage for a rail is within limits
06:38
Bertl
*_CV (check voltage) is compared to the low (min) and high (max) voltage for that rail and *_UV (undervoltage) and *_OV (overvoltage) is set accordingly
06:38
Bertl
the we have the rails themselves like N_VW, N_VN, N_VE and similar for S_*
06:39
Bertl
and what you do not see in this schematic (only in the interface schematic) is that one of each of the power rail pins is not a supply, but a feedback connection instead
06:40
Bertl
i.e. the actual voltage on the SFE (sensor frontend) power rails is connected back to the power board for verification
07:13
usmankhan
left the channel
07:13
RexOrCine
Bertl are you able to make an estimate on timescale Re V2 Beta SW?
07:13
Bertl
not at the moment
07:16
RexOrCine
Is there a chance that a Beta II would ship without it?
07:17
Bertl
well, if the current firmware/software can be polished up enough, why not
07:17
Bertl
although I would prefer to see a solid redesign on a 'product'
07:18
RexOrCine
I'd imagine you'd have a better idea Re timescale going into Q3 or over the summer.
07:18
Bertl
yes, I hope so
07:19
RexOrCine
Thanks.
08:01
se6astian|away
changed nick to: se6astian
08:01
Bertl
morning se6astian!
08:06
Spirit532
joined the channel
08:07
se6astian
good day!
08:10
RexOrCine
What's going on today?
08:11
Bertl
weird stuff, obviously :)
08:12
RexOrCine
Allied Vision? yeah.
08:57
pusle
joined the channel
09:10
BAndiT1983|away
changed nick to: BAndiT1983
09:34
sagnikbasu95
joined the channel
09:35
sagnikbasu95
Hi Bertl. Could you please check this link : https://github.com/sagniknitr/Real-time-sobel-filter-in-FPGA/blob/master/FPGA_acceleration.md
09:42
Bertl
I see where you're going with that ... let me add a few comments here:
09:43
Bertl
first, we should make the kernel generic, i.e. NxN or NxM with 'arbitrary' multipliers
09:43
Bertl
i.e. have a factor Fxy and use that for weighting
09:45
Bertl
then, fetching data from DDR memory is not an option anyway, because we don't have the bandwidth for that
09:46
Bertl
so what you call accelerator store will most likely be a dual port memory inside the FPGA
09:47
Bertl
pipelining is essential to reach the required throughput
09:48
sagnikbasu95
ok..so instead of storing in memory..we will directly use signals from the cmos sensor...so how many lvds banks from the cmos sensor will be connected to the FPGA?
09:49
sagnikbasu95
I read in the datasheet there were 7-8 lvds banks
09:52
Bertl
either from the sensor, or from the DDR memory when it is retrieved to be displayed via HDMI, but it will not be random acccess, it will be fetched in some kind of chunks
09:53
RL
joined the channel
09:53
Bertl
i.e. I'd say it safe to assume, that the data arrives in 'pixels' per line, for each line from top to bottom
09:58
Bertl
so there needs to be a line buffer with at least N-1 lines to apply an NxN kernel
10:03
sagnikbasu95
ok..got it...one question: In the axiom beta wiki , there was a color cinversion block mentioned in the signal datapth..i assume this is for RGB/YUV conversion ?
10:03
sagnikbasu95
*conversion..
10:05
Bertl
I guess you are referring to the Image Processing Pipeline, yes?
10:05
Bertl
https://wiki.apertus.org/images/c/cc/Image01.png
10:08
LordVan
joined the channel
10:10
sagnikbasu95
yes exactly..
10:10
Bertl
the problem there is that it is somewhat outdated (i.e. it was copied from the Axiom Alpha)
10:10
Bertl
the entire HDMI Encoder part was a solution from Analog Devices back then
10:11
Bertl
but now, in the Beta, we generate the HDMI ourselves
10:11
Bertl
so currently, it is 'limited' to RGB, but in the future, we will also do YCrCb for which we will require a color space converter
10:12
Bertl
(this is mainly to get higher frame rates by utilizing subsampling)
10:12
Bertl
but none of this should affect the kernel or filter, as it doesn't make much sense to put it after any color conversion
10:13
sagnikbasu95
yes of course..
10:14
sagnikbasu95
Thanks Bertl
10:14
Bertl
no problem
10:14
Bertl
so to sum up, the main challenges for the filter are:
10:14
Bertl
- high bandwidth throughput
10:15
Bertl
- line store with generic number of lines
10:15
Bertl
- pipelining to reduce resources and improve performance
10:16
Bertl
the generic weighting should also be done in such a way that it can be optimized out for a given kernel
10:17
Bertl
i.e. fields with weight 0 will not get computed at all and fields with weigth 1 will not be multiplied just added
10:18
Bertl
I would also suggest to start with a simple design and build a more generic/complex one based on the results and ideas
10:21
sagnikbasu95
left the channel
10:36
intracube_afk
changed nick to: intracube
10:48
usmankhan
joined the channel
10:49
usmankhan
I would like to point out to this useful resource for carrying out pipelined convolution: https://daim.idi.ntnu.no/masteroppgaver/013/13656/masteroppgave.pdf
10:49
usmankhan
See page 37 for convolution implementation
10:50
Bertl
excellent! please add it to the apropriate lab task
10:51
usmankhan
Sure! just to be clear, I am talking about section 4.4.4
12:04
usmankhan
left the channel
12:38
dimaursu16
left the channel
12:48
dimaursu16
joined the channel
13:11
Bertl
off for now .. bbl
13:11
Bertl
changed nick to: Bertl_oO
13:12
dimaursu16
left the channel
13:28
tyrone
joined the channel
13:42
RL
left the channel
13:43
ljmocic
joined the channel
13:48
LordVan
left the channel
14:01
anuditverma
joined the channel
14:01
anuditverma
left the channel
14:27
niculescu_vlad
joined the channel
14:42
dimaursu16
joined the channel
14:55
ljmocic
left the channel
15:39
tyrone_
joined the channel
15:40
tyrone
left the channel
15:50
RexOrCine
left the channel
15:52
se6astian
changed nick to: se6astian|away
15:57
jucar
joined the channel
16:06
arpu
left the channel
16:11
arpu
joined the channel
16:22
jucar
left the channel
16:42
tyrone_
left the channel
16:47
tyrone
joined the channel
17:05
tyrone
left the channel
19:05
se6astian|away
changed nick to: se6astian
19:22
BAndiT1983
changed nick to: BAndiT1983|away
19:23
derWalter
joined the channel
19:30
derWalter
left the channel
19:42
slikdigit
joined the channel
20:11
Rex0r
left the channel
20:12
RexOrCine
joined the channel
20:26
comradekingu
left the channel
20:27
comradekingu
joined the channel
21:06
se6astian
off to bed
21:06
se6astian
good night
21:07
se6astian
changed nick to: se6astian|away
21:51
niculescu_vlad
left the channel
21:57
slikdigit
left the channel
22:12
slikdigit
joined the channel
23:56
slikdigit__
joined the channel
23:57
slikdigit
left the channel