Re: [libdc] dc1394 on usb, format7 framerate

[Philippe De M. <ph...@ma...>]

Hi Damien,

On Mon, Oct 10, 2016 at 11:09:56PM +0900, Damien Douxchamps wrote:
> Hi Philippe,
> 
> On Thu, 2016-10-06 at 14:35 +0200, Philippe De Muyter wrote:
> > Hi dc1394 experts,
> > 
> > I try to use a iidc/usb camera on linux on a USB 2.0 host, and I have
> > questions about the framerate setting.  The libdc1394 faqs explains
> > only the real firewire cases.
> > 
> > - is there a way using libdc1394 to know we are actually working with
> > a usb camera ?
> 
> I had a quick look at the code and there seems to be no way to do so at
> the moment. It would relatively trivial to do, for example by setting
> flags in the currently unused dc134_camera_t->flags. I will look into
> that.
> 
> > - is there a libdc1394 function yielding the actual bus speed (not
> > dc1394_video_get_iso_speed, which only gives firewire speed)
> 
> This would be a libusb functionality. Of course it could be also probed
> by libdc...
> 
> > - would that help to compute the packet size needed to obtain
> > a desired framerate ?
> 
> Sorry, not sure here...
> 
> > - is there another way using libdc1394 to set the framerate for
> > format7 modes ?
> 
> I usually use the "framerate" feature. You can see it in coriander if
> it's supported by your camera (my FireflyMV USB has it).

Now that you mention it, I finally see the "framerate" feature in
coriander (it's the last entry in the Controls tab).  I must admit
that I hadn't noticed it before.
> 
> You could also enable the timestamp feature that some PGR cameras have,
> although their meaning for USB devices is not obvious (it's related to
> the bus clock on 1394 devices). PGR has great manuals though...
> 
> > I have some experimental results with a PointGrey Chameleon3 CM3-U3-13Y3C :
> > Here are the times to transmit 75 1280x1024 Y422 (also called UYVY) frames,
> > thus with a frame size equal to 1280x1024x2 = 2621440 bytes.  Camera
> > answers : unit_bytes = 328, max_bytes = 6560.
> > 
> > packet_size	time for 75 frames
> > 328		1m17.667s
> > 656		0m39.398s
> > 984		0m26.611s
> > 2624		0m10.300s		0m10.361s		0m10.401s
> > 2952		0m9.273s		0m9.431s		0m9.276s
> > 3280		0m8.926s		0m8.504s		0m9.185s
> > 3608		0m8.639s		0m8.537s		0m8.691s
> > 3936		0m8.470s		0m8.651s		0m8.466s
> > 4264		0m8.488s		0m8.637s		0m8.507s
> > 4592		0m8.503s		0m8.485s		0m8.736s
> > 4920		0m8.512s		0m8.495s		0m8.550s
> > 5248		0m10.105s		0m10.232s		0m10.170s
> > 5576		0m9.594s		0m9.549s		0m9.715s
> > 5904		0m9.092s		0m9.113s		0m9.151s
> > 6232		0m8.678s		0m8.662s		0m8.658s
> > 6560		0m8.391s		0m8.409s		0m8.748s
> > 
> > Note the strange behaviour beginning at 5248
> 
> That's a bit odd. Padding bytes would not explain such a large
> difference. Note the decreasing pattern from 2624 is similar from the
> what is observed at 5248.

Meanwhile, I have measured again the transfer rate with all the possible packet
sizes for different pixelformats, thus with different 'total_bytes' for a frame.
I've come to the conclusion that 'unit_bytes' for each pixelformat for my camera
is the packet size that will produce a '1 fps' transfer rate.  If I configure
a 'n * unit_bytes' packet size, I get a 'n fps' transfer rate, up to around a
~5084 packet size.  Above that packet size the transfer rate is reset to a lower
value.  This is probably limited by the USB 2.0 bus speed.

Now, using coriander, I see that in order to use the 'framerate' feature, I must
set the packet_size to its highest value (or perhaps the 'recommended' value),
Alternatively, if I reduce the 'packet size', I see that the framerate feature
value follows the change in packet size.

I'll try using the "framerate" feature, instead of reducing the packet size.

Thanks for you answer

Philippe

-- 
Philippe De Muyter +32 2 6101532 Macq SA rue de l'Aeronef 2 B-1140 Bruxelles