Przemyslaw Dmochowski | 26 Nov 2009 16:08
Picon
Favicon

Question about SPI transfer rate and OMAP processing power

Hello everyone,

I work on an SDR that would work from a battery and without a PC.
I read the posts about SPI transfer rates and I was wonder whether
you've done some work to test real processing power of the BeagleBoard. 
What I am interested in is the maximum spectrum bandwidth that can be
achieved with the combination of USRP and BeagleBoard assuming the data
transfer between them is done over SPI interface.
My applications I am thinking of at the moment are not very demanding
(AM/FM UHF analog radio) but in the future I would want to have 5Mbits/s
digital transmission (probably FSK) over the radio using OFDM.
I don't own the BeagleBoard at the moment and I would appreciate  your
comments on the signal processing capabilities of the USRP-BB
configuration. Do you think the 5Mbits/s using OFDM will be possible?

Philip Balister | 27 Nov 2009 15:32
Favicon

Re: Question about SPI transfer rate and OMAP processing power

On 11/26/2009 10:08 AM, Przemyslaw Dmochowski wrote:
> Hello everyone,
>
> I work on an SDR that would work from a battery and without a PC.
> I read the posts about SPI transfer rates and I was wonder whether
> you've done some work to test real processing power of the BeagleBoard.
> What I am interested in is the maximum spectrum bandwidth that can be
> achieved with the combination of USRP and BeagleBoard assuming the data
> transfer between them is done over SPI interface.
> My applications I am thinking of at the moment are not very demanding
> (AM/FM UHF analog radio) but in the future I would want to have 5Mbits/s
> digital transmission (probably FSK) over the radio using OFDM.
> I don't own the BeagleBoard at the moment and I would appreciate  your
> comments on the signal processing capabilities of the USRP-BB
> configuration. Do you think the 5Mbits/s using OFDM will be possible?

Basically, we need to get more signal processing code running on the 
OMAP3 to answer this question.

I suspect that processing will need to be divided between an FPGA, the 
DSP and the GPP for 5 Mbits/s OFDM.

Philip

Sylvain AZARIAN | 27 Nov 2009 15:47
Picon

Re: Question about SPI transfer rate and OMAP processing power

Hello,


The SPI transfert rate seems also quite slow for such processing... 
We are working on same type of systems at Supélec ( see tools4sdr.com ) and currently using GnuRadio & URSP. Even the USB bus is too slow...

In fact the clue is to have the decimation done between the ADC and the main processor.. In the USRP, the FPGA implements a CIC filter to decimate by at least 4 the incoming data flow...

sylvain

2009/11/27 Philip Balister <philip-7I+KllWURj5BDgjK7y7TUQ@public.gmane.org>
On 11/26/2009 10:08 AM, Przemyslaw Dmochowski wrote:
Hello everyone,

I work on an SDR that would work from a battery and without a PC.
I read the posts about SPI transfer rates and I was wonder whether
you've done some work to test real processing power of the BeagleBoard.
What I am interested in is the maximum spectrum bandwidth that can be
achieved with the combination of USRP and BeagleBoard assuming the data
transfer between them is done over SPI interface.
My applications I am thinking of at the moment are not very demanding
(AM/FM UHF analog radio) but in the future I would want to have 5Mbits/s
digital transmission (probably FSK) over the radio using OFDM.
I don't own the BeagleBoard at the moment and I would appreciate  your
comments on the signal processing capabilities of the USRP-BB
configuration. Do you think the 5Mbits/s using OFDM will be possible?

Basically, we need to get more signal processing code running on the OMAP3 to answer this question.

I suspect that processing will need to be divided between an FPGA, the DSP and the GPP for 5 Mbits/s OFDM.

Philip

Philip Balister | 27 Nov 2009 16:02
Favicon

Re: Question about SPI transfer rate and OMAP processing power

On 11/27/2009 09:47 AM, Sylvain AZARIAN wrote:
> Hello,
>
> The SPI transfert rate seems also quite slow for such processing...
> We are working on same type of systems at Supélec ( see tools4sdr.com ) and
> currently using GnuRadio&  URSP. Even the USB bus is too slow...

SPI is not super fast, but it is an easy way to attach an fpga to an 
embedded system. (OK, maximizing the bi-directional transfer rate is a 
bit tricky ) Improving the transfer rate can be done by using the GPMC 
interface on the OMAP3. I have done some preliminary work by looking at 
signals in the GPMC bus. I have the basics going, but the driver needs 
work to approach the theoretical transfer rates.

>
> In fact the clue is to have the decimation done between the ADC and the main
> processor.. In the USRP, the FPGA implements a CIC filter to decimate by at
> least 4 the incoming data flow...

We need something with a larger FPGA, then more processing can be done 
in the FPGA, for example if the synchronization is done in the FPGA, 
symbols, rather than samples, can be sent over the bottleneck.

Philip

PS: I'll be at the SDR Forum Technical Conference next week if anyone 
wants to talk about stuff in person.

>
> sylvain
>
> 2009/11/27 Philip Balister<philip@...>
>
>> On 11/26/2009 10:08 AM, Przemyslaw Dmochowski wrote:
>>
>>> Hello everyone,
>>>
>>> I work on an SDR that would work from a battery and without a PC.
>>> I read the posts about SPI transfer rates and I was wonder whether
>>> you've done some work to test real processing power of the BeagleBoard.
>>> What I am interested in is the maximum spectrum bandwidth that can be
>>> achieved with the combination of USRP and BeagleBoard assuming the data
>>> transfer between them is done over SPI interface.
>>> My applications I am thinking of at the moment are not very demanding
>>> (AM/FM UHF analog radio) but in the future I would want to have 5Mbits/s
>>> digital transmission (probably FSK) over the radio using OFDM.
>>> I don't own the BeagleBoard at the moment and I would appreciate  your
>>> comments on the signal processing capabilities of the USRP-BB
>>> configuration. Do you think the 5Mbits/s using OFDM will be possible?
>>>
>>
>> Basically, we need to get more signal processing code running on the OMAP3
>> to answer this question.
>>
>> I suspect that processing will need to be divided between an FPGA, the DSP
>> and the GPP for 5 Mbits/s OFDM.
>>
>> Philip
>>
>


Gmane