Forum: GNU Radio gnuradio land speed record?

I'm curious what people do with the wideband capability of the 
gnuradio/usrp and what is the widest bandwidth signal one can really 
process with available computers?

For reference I have a ~2.4 GHz core 2 duo laptop. For a 200 kHz FM 
demodulator I consume about 40% of one cpu. That's pretty much the 
simplest useful thing anyone can do so that maps to my laptop might be 
able to process 1 MHz bandwidth continuously.

Similarly, my hard drive can't really keep up with 32 Mbyte/s recording. 
So if samples are 16-bit and you really can't afford lost data it seems 
like recording is limited to maybe 10 MHz or so bandwidth.

However, with gigabit Ethernet you can send 100 Mbyte/s or more. What's 
the most anyone has recorded or processed continuously? What level of 
compexity was the processing?

Thanks,
Clark

On 07/30/2010 09:33 AM, Clark Pope wrote:
> Similarly, my hard drive can't really keep up with 32 Mbyte/s
> recording. So if samples are 16-bit and you really can't afford lost
> data it seems like recording is limited to maybe 10 MHz or so
> bandwidth.
>
> However, with gigabit Ethernet you can send 100 Mbyte/s or more.
> What's the most anyone has recorded or processed continuously? What
> level of compexity was the processing?


With RAID arrays or SSDs, it isn't that hard anymore to sustain 100 MB/s
recording to disk.  With 4 and 6 core systems and the i7 architecture
you can get more than 5X the performance of your laptop.

There are a lot of applications using the full 25 MHz of RF bandwidth.
You just need to pay a lot of attention to efficiency of your program
and algorithms.

Matt

> With RAID arrays or SSDs, it isn't that hard anymore to sustain 100 MB/s 
> recording to disk.  With 4 and 6 core systems and the i7 architecture you can 
> get more than 5X the performance of your laptop.
>
> There are a lot of applications using the full 25 MHz of RF bandwidth. You 
> just need to pay a lot of attention to efficiency of your program and 
> algorithms.
>

For 'speed record' type information, you might be interested in SORA, a
software radio project from Microsoft Research.  They use different
hardware and custom software, but the fundamentals are the same.

As Matt points out, efficiency is a function of engineering. Using 
modern
processors, 64-bit architecture, multicore, software LUTs, and a variety
of other optimizations they were able to fully process 802.11g signals 
of
20 MHz bandwidth and sustain reception of 54 Mbps signals including
Viterbi decoding, etc.  I see no reason this couldn't be done with
USRP(2) / GNU Radio... but looking at Microsoft's author list they had a
lot of developers working pretty hard on it!

There's not a ton of detail in the original paper, and what code is
available is almost certainly not something you want to look at without
reading the license very carefully, but here's the link to the project
website:

http://research.microsoft.com/en-us/projects/sora/

and the original paper:

http://research.microsoft.com/apps/pubs/default.aspx?id=79927

Dan

On 07/30/2010 01:19 PM, Daniel Halperin wrote:
>> With RAID arrays or SSDs, it isn't that hard anymore to sustain 100
> hardware and custom software, but the fundamentals are the same.
> available is almost certainly not something you want to look at without
> reading the license very carefully, but here's the link to the project
> website:

I don't think you need to read the license carefully to realize you do
not want to download the code.

1) No commercial use clause.
2) You cannot distribute derived works.
3) You grant MS the right to use your modifications to the code.

Even if you are in an academic situation, you need to think about your
future in the commercial space before looking at the code.

Philip

On 07/30/2010 01:01 PM, Matt Ettus wrote:
>
>
For reference, an early version of my IRA radio astronomy receiver was
able to do single-channel
  at 16MHz bandwidth, using a Core 2 Extreme 9770 and 8GB of memory and
a USRP1.

The IRA code does a number of things

     o compute total power over the entire bandwidth
     o compute an FFT with 1Hz resolution (that's a 16 million point 
FFT)
     o do a SETI analysis of the resulting FFT
     o do a transient signal analysis of the total power data
     o run a narrowband interference filter based on an FFT filter
     o run a GUI with lots of bells 'n whistles




--
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
http://www.sbrac.org

----------------------------------------
> > process with available computers?
> >
> You just need to pay a lot of attention to efficiency of your program
> and algorithms.
>
> Matt

Good point, if you remove the CPU bottleneck and go straight to storage 
you can do 100 MByte/s. Now is gigE the best for that or would SATA be 
better? Seems like once you buy the networking and storage equipment 
you've blown your budget relative to the usrp price.

Yeah, I'd be interested in what those 25 MHz apps are. Maybe we need a 
contest for widest bandwidth, practical, most useful application on 
gnuradio. One ground rule though would be that the cost of the 
processing device has to be less than the USRP2, for example.

Thanks,
Clark

----------------------------------------
> >> process with available computers?
> > just need to pay a lot of attention to efficiency of your program and
> 20 MHz bandwidth and sustain reception of 54 Mbps signals including
>
> and the original paper:
>
> http://research.microsoft.com/apps/pubs/default.aspx?id=79927
>
> Dan
>

Thanks that's a good data point! So a huge corporation with infinite 
resources tops out at about 20 MHz sustained processing of what I would 
call a real world practical signal. -Clark

----------------------------------------
> >
> a USRP1.
>
> Discuss-gnuradio mailing list
> Discuss-gnuradio@gnu.org
> http://lists.gnu.org/mailman/listinfo/discuss-gnuradio


Now was the FFT continuous, i.e. you got one output bin for every input 
sample or you did snapshots? Thanks

On Fri, 30 Jul 2010, Clark Pope wrote:
>> and the original paper:
>>
>> http://research.microsoft.com/apps/pubs/default.aspx?id=79927
>>
>> Dan
>>
>
> Thanks that's a good data point! So a huge corporation with infinite resources tops out at about 20 MHz sustained processing of what I would call a real world practical signal. -Clark


I think that should be thought of as a lower bound not an upper bound 
:).
They accomplished their goal of handling dot11g and haven't (that I'm
aware of) moved on to anything harder (yet).

Aside: reading the paper doesn't come with any licensing terms, and
there's even ASM code in the appendix for a SIMD-enabled FIR filter at 
the
end.

Dan

>> > On 07/30/2010 09:33 AM, Clark Pope wrote:
>> > recording to disk. With 4 and 6 core systems and the i7 architecture you can
>>
>> reading the license very carefully, but here's the link to the project
>
> Thanks that's a good data point! So a huge corporation with
> infinite resources tops out at about 20 MHz sustained
> processing of what I would call a real world practical signal.

Their infinite resources are in fact limited by their management's 
mindset and inability to think clearly about the
consequences of their business model, the impact of the Internet, etc. 
A guy like Ballmer is just a much a reason why
Linux exists as is Linus himself.  His rhetoric about GPL being a 
cancer, Linux developers are communists, etc. has
provided "infinite inspiration" to the guys with limited resources.

SORA may be a useful data point, but advise to carefully consider the 
source.

-Jeff

----------------------------------------
> >> Date: Fri, 30 Jul 2010 10:19:35 -0700
> >> >>
> >> > algorithms.
> >> Viterbi decoding, etc. I see no reason this couldn't be done with
> >> and the original paper:
> Their infinite resources are in fact limited by their management's mindset and inability to think clearly about the
> consequences of their business model, the impact of the Internet, etc. A guy like Ballmer is just a much a reason why
> Linux exists as is Linus himself. His rhetoric about GPL being a cancer, Linux developers are communists, etc. has
> provided "infinite inspiration" to the guys with limited resources.
>
> SORA may be a useful data point, but advise to carefully consider the source.
>
> -Jeff
>

This is true. And I seem to recall Gates saying basically that the 
reason windows is coded so inefficiently is because the hardware will 
always increase faster than Microsofts bad coding. I have an internet 
explorer process running right now that's use 80 megabytes of memory! 
Years ago we ran an entire graphical interface and gui in 64 kByte on 
the commodore 64.

Clark Pope wrote:
>>     
>>> recording. So if samples are 16-bit and you really can't afford lost
>>
>  
> Thanks,
> Clark                 
> _______________________________________________
> Discuss-gnuradio mailing list
> Discuss-gnuradio@gnu.org
> http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>   

I use the 25MHz sample-rate to test algorithms. In one application I
transmit during 0.1ms and then use 10ms to process the data. This is for
doing research on algorithms. It's still different from pure off-line
processing as I can do feedback, MAC-algorithms and so on.

I would actually like to have more than 25MHz maybe 100MHz to be able to
work with bandwidths such as 40MHz.

BR/
Per

On Jul 30, 2010, at 1:30 PM, Philip Balister wrote:
> I don't think you need to read the license carefully to realize you do not want to download the code.
> 
> 1) No commercial use clause.
> 2) You cannot distribute derived works.
> 3) You grant MS the right to use your modifications to the code.
> 
> Even if you are in an academic situation, you need to think about your future in the commercial space before looking at the code.


On Jul 30, 2010, at 3:48 PM, Daniel Halperin wrote:
> Aside: reading the paper doesn't come with any licensing terms, and there's even ASM code in the appendix for a SIMD-enabled FIR filter at the end.

Old topic, but just to be complete (since I was looking into it anyway):

The license is found when you try to download the SDK:

< 
http://research.microsoft.com/en-us/downloads/35a929d6-0cb0-4318-968a-69d05c9bbc65/ 
>

Adding to what Philip wrote above: The code runs on Windows only, and 
requires the use of their special "radio control board" -as well as- a 
USRP1/2 or WARP as the radio front-end to do actual over the air 
communications. - MLD