|Subject:||Re: [Discuss-gnuradio] unexpected underruns|
|Date:||Wed, 2 Aug 2017 11:10:58 -0400|
USRP sinkOn Wed, Aug 2, 2017 at 11:08 AM, Rui ZOU <address@hidden> wrote:My previous email shows the rate WITHOUTOn Wed, Aug 2, 2017 at 11:05 AM, Marcus Müller <address@hidden> wrote:
WAIT! Throttle? I didn't see that in either of the flow graphs you sent me first (twoparatx, onefile2tx)
Seriously?! Your GRC will even print a warning that you mustn't use Throttle together with hardware if you have both Throttle and a USRP sink.
Remove the Throttle, and try again.
On 08/02/2017 05:00 PM, Rui ZOU wrote:
RuiThe rate is similar to using file source, 0.78MS/s with throttle and 11.3MS/s when throttle bypassed.Changed to null source, the rate is still around twice the sample rate (390.625k) for throttle block.When the throttle block is bypassed, the rate jumps up to around 11.3MS/s.
******* MESSAGE DEBUG PRINT ********
(((rate_now . 781360) (rate_avg . 786529)))
******* MESSAGE DEBUG PRINT ********
(((rate_now . 1.16071e+07) (rate_avg . 1.12848e+07)))
On Wed, Aug 2, 2017 at 10:37 AM, Marcus Müller <address@hidden> wrote:
Ok, there's something fishy here. That rate (without the USRP Sink) is ridiculously low. Can you replace the file_source with a null_source? That way, we can rule out storage as the bottleneck.
The probe_rate does nothing but just count how many items fly by, and then send a message at its output port every update period. The message_debug just prints messages.
If it's not storage: are you perhaps running in a powersaver mode? Even so, the rate would be too low. I really don't know what's going wrong here. Run your flow graph, run "top" in a terminal, check that the python process and its spawned child threads really consume most of the CPU. If that's not the case, you've got something else eating away on your CPU.
On 08/02/2017 04:22 PM, Rui ZOU wrote:
Not sure if the debug setup is the expected since it's the first time I use the 'Probe Rate' and 'Message Debug' blocks whose functions are not very clear to me now just after reading the contents under the document tag. If there are other ways to learn about new blocks, please advise.
The rates I get when the USRP Sink disabled is around 0.78MS/s I guess from the debug output, shown below.
******* MESSAGE DEBUG PRINT ********
(((rate_now . 782916) (rate_avg . 784937)))
After enabling the USRP Sink, I got lots of 'L's and 2.4MS/s, shown below.
LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLLLLLLLLLLLLLLLLLLLLL LLLLLLLLLLL******* MESSAGE DEBUG PRINT ********
(((rate_now . 2.36319e+06) (rate_avg . 2.36319e+06)))
GRC and python files are attached.
On Wed, Aug 2, 2017 at 3:57 AM, Marcus Müller <address@hidden> wrote:
Huh, I really don't know what's happening there :/ I sadly don't have the USRP to test this live with me right now, but there's absolutely no timed commands involved¹
So, trying to weed out bugs:
* I've replaced the USRP sink with a "Probe Rate" block, connected to a "Message Debug"'s print port. I saw samples fly by with more than 7 MS/s, so there really shouldn't be a bottleneck here – can you try to do the same and see whether your system can get similar rates? 7MS/s is still far too little for my taste, but that is FM-Modulation-limited²
* Can you delete your subdev spec? in a 2-channel case, that should be the implicit one, anyways.
¹ "timed commands" are a USRP feature that allows certain things to happen at well-defined times. You get an L when a timed command reaches the USRP after the specified time has already passed. In your flow graph, all that could happen is that a sample packet reaches the USRP after it should – but that's unlikely, you'd get a "U" instead.
² at least on my machine, most of the time is spent in the FM modulator. Which is kind of annoying, because looking into that, what costs most time is the "keeping the phase within 0;2pi" floating point modulo operation. I might get the urge to fix that.
On 08/01/2017 08:31 PM, Rui ZOU wrote:
RuiHi Marcus,I have fixed the two parallel SISO by removing packeting encoding, using QT GUI instead of WX. But the 'L' indicator still comes on, even sooner than previous version. The GRC and generated python files are attached.
On Tue, Aug 1, 2017 at 12:04 PM, Marcus Müller <address@hidden> wrote:
Ah, cool, but then I wouldn't start by packetizing data.
Simply send your file GMSK-Modulated; drop the packet encoding; think about it: the MIMO coding (usually) happens *after* the data has been formed to logical data units.
A few notes on your flowgraphs: Don't use the WX GUI elements in new flowgraphs. We have deprecated them, since no-one can maintain them, and the Qt GUI sinks have shown to be both more stable and efficient. As far as I can foresee your application's needs, Qt has replacements for all the WX visualizations you'd need.
For the receiver, I'd guess you'd first simply start by just recording from to channels, and then experimenting with things like cross-correlation, and estimating the channel matrix based on your known transmit signal. I wouldn't be surprised if the channel is rather boring in your setup – I blindly assume you're doing this indoors, and that limits the path difference and the amount of change (and hence, the delay spread and the doppler spread) your signals are subject to, especially since your bandwidth is so low. Of course, having a flat channel is nice :) but it also means that it might be quite hard to get any actual MIMO gain, because the two RX antennas might be very correlated. If in doubt, increase bandwidth. Be agressive with roll-off / Bandwidth factors of your GMSK.Cheers,
On 08/01/2017 05:51 PM, Rui ZOU wrote:
RuiHi Marcus,My goal is to first build a 2-by-2 space multiplexing MIMO using two X310s and GNU Radio. As I'm new to all this stuff, I'm starting from building 2 parallel SISOs. If there are some good kick-start materials or any resources, they will be very valuable. Thanks.
On Tue, Aug 1, 2017 at 11:37 AM, Marcus Müller <address@hidden> wrote:
sorry, I might simply have missed those, and didn't find your first email when I saw your recent one! I apologize.
So, hm, interestingly, we have a severe bug in the packet_encoder block (its design is pretty bad, and that triggers an unexpected behaviour underneath). That might mean the packet_encoder is just consuming items as fast as it can, without actually producing packets. In other words, packet_encoder is broken; you can't use it right now.
The more appropriate way of dealing with data might be in the example flowgraphs that you'd find under /usr/[local/]share/doc/gnuradi
o/examples/digital/packet_loop back_hier.grc ; it's a lot more complicated, though, and you'd have to write a message / PDU source that gives you the data you want to transmit, rather than the Random PDU block!
I don't really know if that is the way to go. What is it, that you want to build? Maybe the mailing list can advise?
On 08/01/2017 05:26 PM, Rui ZOU wrote:
Here are the two flowgraphs I have used. I have tried to attach the two files in my first email. Probably failed in doing that. If still not seen, please let me know so I will try again. Thanks for your help.
Running the first flow graph will cause GRC stop responding instantly, while the second one can run for a little while and produce lots of 'L' before going not responsive.
On Tue, Aug 1, 2017 at 11:09 AM, Marcus Müller <address@hidden> wrote:
don't know, to me, it looks like replying didn't work out great, since my mail client showed your mail in a new thread. Really, replying to a mailing list mail should be nothing more than hitting the "reply" or "reply all" button.
Anyway, even the slowest PC/laptop/Raspberry Pi/… I could think of would be able to deal with these rates, so there's very, very likely something wrong with the GNU Radio flowgraph you're using. Maybe you'd want to share that!
On 08/01/2017 04:59 PM, Rui ZOU wrote:
RuiI use 390.625k as the sampling rate because this is the lowest I can get using the Ettus X310 without giving me a warning saying that the sampling rate cannot be provided by the hardware. The application is just transmitting a file using GMSK modulation on the two daughter boards of X310.Hi Marcus,Sorry for leaving the title empty, that's the first time to post to a mailing list. This is also the first time to reply, no sure if I replied correctly.
_________________ Discuss-gnuradio mailing list address@hidden https://lists.gnu.org/mailman/ listinfo/discuss-gnuradio
|[Prev in Thread]||Current Thread||[Next in Thread]|