headers
devin kelly | 18 Jun 2010 16:09
Picon

Beagle Board Performance

Hello,

I'm interested in doing a project with the Beagle Board and the USRP1, along with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only USRP2s), so I need a few questions answered before I commit to this platform.

What type of performance can I expect from the Beagle Board?  I'm mostly interested in doing FFTs with some comparatively small processing on that data, is this possible? 

What sort of decimation rates can I use and expect near real time performance for this application?  I have a few (albeit several years old) machines that struggle with just taking an FFT with a decimation rate of around 8.  This is on the USRP2 as well.

Any advice at all is appreciated.
Devin

Permalink | Reply |
headers
Philip Balister | 18 Jun 2010 16:46

Re: Beagle Board Performance

On 06/18/2010 10:09 AM, devin kelly wrote:
> Hello,
>
> I'm interested in doing a project with the Beagle Board and the USRP1, along
> with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only
> USRP2s), so I need a few questions answered before I commit to this
> platform.
>
> What type of performance can I expect from the Beagle Board?  I'm mostly
> interested in doing FFTs with some comparatively small processing on that
> data, is this possible?

It is possible. There is a Google SoC project to improve fftw support 
for the beagle. See http://elinux.org/BeagleBoard/GSoC/2010_Projects.

>
> What sort of decimation rates can I use and expect near real time
> performance for this application?  I have a few (albeit several years old)
> machines that struggle with just taking an FFT with a decimation rate of
> around 8.  This is on the USRP2 as well.

Decimation rates of 8 will be hard. Especially getting data from USB.

Philip

>
> Any advice at all is appreciated.
> Devin
>
Permalink | Reply |
headers
devin kelly | 21 Jun 2010 15:01
Picon

Re: Beagle Board Performance


What about a more reasonable decimation rate, like 32 or 64... Could I collect samples 50%-75% of the time and then process that data the rest of the time?

Thanks again,
Devin

On Fri, Jun 18, 2010 at 10:09 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
Hello,

I'm interested in doing a project with the Beagle Board and the USRP1, along with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only USRP2s), so I need a few questions answered before I commit to this platform.

What type of performance can I expect from the Beagle Board?  I'm mostly interested in doing FFTs with some comparatively small processing on that data, is this possible? 

What sort of decimation rates can I use and expect near real time performance for this application?  I have a few (albeit several years old) machines that struggle with just taking an FFT with a decimation rate of around 8.  This is on the USRP2 as well.

Any advice at all is appreciated.
Devin

Permalink | Reply |
headers
Sylvain AZARIAN | 21 Jun 2010 15:10
Picon

Re: Beagle Board Performance

The decimation process is done in the FPGA inside the USRP1.

Why are you trying to redo that outside ? It is intended to reduce the bandwith over the USB...

Regards
Sylvain Azarian
www.tools4sdr.com

2010/6/21 devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>

What about a more reasonable decimation rate, like 32 or 64... Could I collect samples 50%-75% of the time and then process that data the rest of the time?

Thanks again,
Devin


On Fri, Jun 18, 2010 at 10:09 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
Hello,

I'm interested in doing a project with the Beagle Board and the USRP1, along with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only USRP2s), so I need a few questions answered before I commit to this platform.

What type of performance can I expect from the Beagle Board?  I'm mostly interested in doing FFTs with some comparatively small processing on that data, is this possible? 

What sort of decimation rates can I use and expect near real time performance for this application?  I have a few (albeit several years old) machines that struggle with just taking an FFT with a decimation rate of around 8.  This is on the USRP2 as well.

Any advice at all is appreciated.
Devin


Permalink | Reply |
headers
devin kelly | 21 Jun 2010 19:12
Picon

Re: Beagle Board Performance

Sylvain,

Sorry for not being clear.  I was just trying to say that, the higher the decimation rate the more data I have to process, I wasn't talking about decimating twice. 

So would I'd like to do is record samples for 0.5 to 0.75 seconds (for example), then do some processing on those samples for the remaining, 0.25 to 0.5 seconds, and repeat.  The processing I'm thinking of is taking an FFT, making a decision, then recording it to non-volatile storage.  So my question is, to do processing like I described, what sort of decimation rates can I choose?

Devin

On Mon, Jun 21, 2010 at 9:01 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:

What about a more reasonable decimation rate, like 32 or 64... Could I collect samples 50%-75% of the time and then process that data the rest of the time?

Thanks again,
Devin


On Fri, Jun 18, 2010 at 10:09 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
Hello,

I'm interested in doing a project with the Beagle Board and the USRP1, along with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only USRP2s), so I need a few questions answered before I commit to this platform.

What type of performance can I expect from the Beagle Board?  I'm mostly interested in doing FFTs with some comparatively small processing on that data, is this possible? 

What sort of decimation rates can I use and expect near real time performance for this application?  I have a few (albeit several years old) machines that struggle with just taking an FFT with a decimation rate of around 8.  This is on the USRP2 as well.

Any advice at all is appreciated.
Devin




--
http://users.wpi.edu/~dkelly/
Permalink | Reply |
headers
Sylvain AZARIAN | 22 Jun 2010 11:15
Picon

Re: Beagle Board Performance

Devin,


Then you are not taking the problem from the right perspective...
First, you have to know the bandwidth you need for your application ( mainly "fmax"). This will define your sampling rate.
Then, you have to set your decimation factor depending on the amount of data resulting from the previous choice. On a "fast" pc running Linux, an avg around 30 msps can be achieved. Not tested with Beagleboard yet.

Then.... processing : you need to have one thread reading the USB stream from the USRP and filling a buffer. you need a second thread performing the fft and doing the decision.
the good question now is : what is the frequency resolution you need ( this will set the fft size .... and time needed to compute that).

I would suggest you start by estimating FFT size, then evaluate the time you have to compute that.
But maybe I am completely wrong and in fact you do not need real time processing... ????

sylvain

2010/6/21 devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Sylvain,

Sorry for not being clear.  I was just trying to say that, the higher the decimation rate the more data I have to process, I wasn't talking about decimating twice. 

So would I'd like to do is record samples for 0.5 to 0.75 seconds (for example), then do some processing on those samples for the remaining, 0.25 to 0.5 seconds, and repeat.  The processing I'm thinking of is taking an FFT, making a decision, then recording it to non-volatile storage.  So my question is, to do processing like I described, what sort of decimation rates can I choose?

Devin


On Mon, Jun 21, 2010 at 9:01 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:

What about a more reasonable decimation rate, like 32 or 64... Could I collect samples 50%-75% of the time and then process that data the rest of the time?

Thanks again,
Devin


On Fri, Jun 18, 2010 at 10:09 AM, devin kelly <dwwkelly-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
Hello,

I'm interested in doing a project with the Beagle Board and the USRP1, along with using gnu radio.  My lab doesn't have a Beagle Board or USRP1 (only USRP2s), so I need a few questions answered before I commit to this platform.

What type of performance can I expect from the Beagle Board?  I'm mostly interested in doing FFTs with some comparatively small processing on that data, is this possible? 

What sort of decimation rates can I use and expect near real time performance for this application?  I have a few (albeit several years old) machines that struggle with just taking an FFT with a decimation rate of around 8.  This is on the USRP2 as well.

Any advice at all is appreciated.
Devin




--
http://users.wpi.edu/~dkelly/

Permalink | Reply |

Gmane