[hpsdr] NVIDIA Jetson Nano SBC - YouTube video

Chris Smith chris at vspl.co.uk
Mon Apr 29 10:52:14 PDT 2019


Joe

The GNURadio approach looks like it has potential.

The other approach i.e. using several independent processing modules a la ghpSDR, which I followed for a while then abandoned when linHPSDR became available, is a little ‘messy’ and network heavy unless the various levels are served by separate networks. Perhaps I should revisit ghpSDR3?

The graphical approach tends to give me severe headaches as I have never had to imagine any processing in that way having come from a purely software background with a smattering of hardware thrown in. I did have to develop a real-time data acquisition system (software & hardware) a couple of decades ago which used an X Window System toolkit that took graphical input to define the GUI. However I found a way to circumvent the front-end and do everything using Xlib etc..!

Thanks. All worth another look.

Chris G4NUX


> On 29 Apr 2019, at 14:59, Joe Martin <k5so at k5so.com> wrote:
> 
> Chris, 
> 
> One of the approaches that does use multiple cores efficiently, and is under continuous development with an active community, is the GNURadio-companion flowchart programming approach using Linux-based PCs.  I use this approach for my radio astronomy wide bandwidth data acquisitions to good advantage with the Ettus USRP devices but the approach can just as easily be used for other SDR communications platforms too in principle.  
> 
> While not looking to find the cheapest solution to my requirements, but rather, looking for the highest performance I could obtain (without spending a personal fortune) I settled on the Ettus USRP radios (the X310 with TwinRX daughterboards, specifically, covering 10MHz-6GHz continuously and directly) matched with a powerful AMD 2950X Threadripper CPU that contains 32 soft cores running at 3.4 GHz.  I know, I know, most hams would gasp at this level of commitment but I’m serious about performance so I don’t want to play with barely capable hardware such as the cheap single board units mentioned already.  
> 
> The GNURadio flowchart program approach is a graphical programming approach in which specific tasks are implemented in discrete, highly optimized graphical “blocks”, with each block utilizing its own core for processing, up to the limit of cores that are available in the CPU.  This approach provides incredible performance gains wrt speed and throughput as, in my particular case, all 32 cores are used simultaneously to process the data flow using parallel processing with the 32 cores to receive and process wide bandwidth (high rate) data.  I’m currently using 50 MHz BW, dual channels, Rx streams to the PC and it works just fine.  Others use both TX and RX similarly.  
> 
> I don’t know if GNURadio blocks are available for use with the Nvidia CUDA core hardware or not (I haven’t seen it, but I’m new to GNURadio) but I thought I’d mention the GNURadio-companion flowchart graphical programming approach that is incredibly powerful for use with USRP devices and with the very cheap RTL dongles too for that matter.  The RTL dongles are receive-only so that’s a disadvantage for ham communications but the USRP devices can have both RX and TX functionality depending upon which daughterboard you use in the USRP and which USRP that you select to use.  If GNURadio-companion doesn’t have a source block for your desired SDR then perhaps someone would be kind enough to create one so that all the incredible power of the multitude of processing blocks/functionalities inherent in GNURadio can be brought to bear on it.  
> 
> The problem in utilizing GNURadio for typical ham use is twofold as I see it though:  
> 
> 1) The most useful platform to work with GNURadio is Linux.  That’s a stumbling block for many, but it needn’t be in fact.  I was a dedicated Microsoft Windows operator for years because I thought Linux was too complicated, but I finally learned it isn’t at all!   
> 
> 2) To usefully implement GNURadio functions for your radio you really must be willing to dig into the programming aspect of things and many hams are afraid to try that.  GNURadio-companion is a graphical approach, different from classical programming, in which you simply connect graphical blocks together with lines to get the functionality of the blocks implemented in your “program”, so it isn’t really programming at all in the classical sense from the user’s perspective.  You can create and use GNURadio-companion flowchart programs without know how to program in any computer language at all.  
> 
> Just thought I’d mention it, not that I think the GNURadio approach is viable for everyone.  It isn’t.  But it’s incredibly powerful and doesn’t take a lot of effort to implement basic functionalities.  It is certainly one example of how multiple cores are currently being utilized though, which is to the point of your question I think. 
> 
> 73, Joe K5SO
> 
> 
> 
>> On Apr 29, 2019, at 7:13 AM, Chris Smith <chris at vspl.co.uk> wrote:
>> 
>> ***** High Performance Software Defined Radio Discussion List *****
>> 
>> Scott
>> 
>> Apart from the cost advantages that Shirley mentions, as far as I’m concerned the main advantage is relief from the tyranny of Microsoft. Any boards/SBCs which can run a version of Linux can run linhpsdr.
>> 
>> I use linhpsdr from John Melton exclusively to control my HPSDR Atlas kit. It runs very well on my 4-core 2.4GHz Intel board but a Raspberry Pi 3B+ struggles even at 9600. I had audio reports which likened my voice to that of a Darlek speaking under water!
>> 
>> I have a Jetson TK1 which I bought when they were first announced and there was a flurry of interest in the possible use of the (192) CUDA cores. However, unless I’ve missed a thread, there was never any development effort in that direction and the TK1 has languished on the shelf apart from a few months running BOINC/SETI at home which I had a hard time building for it from sources. The supply of suitable data sets dried up and that died too.
>> 
>> The Jetson Nano looks like a decent unit but here again, unless there is some development effort from those who have far greater knowledge than I do, the CUDA cores are likely to be a GoFaster bolt-on which will simply draw lots of current while not actually contributing anything useful.
>> 
>> Anybody out there know of any developments using CUDA cores for SDR? I imagine the effort required to do the functional decomposition to spread the processing over a number of cores would be immense. Would there be any advantage to such parallel processing as far as SDR was concerned? I haven’t seen any discussion on this subject on this reflector.
>> 
>> Just my 2 cents worth hoping to stimulate some discussion...
>> 
>> Cheers & 73
>> 
>> Chris G4NUX
>> 
>> (Like Sid Boyce a Microsoft free zone though some of my computing tasks are performed on Apple kit)
>> 
>> 
>>> On 29 Apr 2019, at 09:01, Shirley Márquez Dúlcey <mark at buttery.org> wrote:
>>> 
>>> ***** High Performance Software Defined Radio Discussion List *****
>>> 
>>> Cost. The Odroid is $79 with 4GB RAM (the only version currently
>>> available) and the NVidia is $99 if you can get one at list price.
>>> (Supply is currently limited and resellers that actually have stock
>>> are marking them up. You can get them at list price directly from
>>> NVidia but their shipping charges are higher than some other sources.)
>>> A Mini ITX motherboard plus an i7 CPU will cost much more. Even
>>> combining that motherboard with a cheap Intel processor like a Pentium
>>> will cost quite a bit more.
>>> 
>>> Some ballpark numbers (price reference: Newegg):
>>> Motherboard: $80 (Mini ITX isn't a bargain, Micro ATX can be had for around $55)
>>> Current i7 processor (i7-8700): $300
>>> Current Pentium processor (G4400, about to be discontinued): $65
>>> Alternate more recent Pentium processor (G5400): $75
>>> 4GB DDR4 RAM (one DIMM): $20
>>> 
>>> So even with the cheap processor it's $165, and the case and power
>>> supply are also going to be more costly than the ones for the single
>>> board computers.
>>> 
>>> The Jetson Nano looks really promising for DSP applications. Although
>>> its four Cortex A57 cores at 1.43 GHz are nothing special (the
>>> Odroid's four A73 cores at 1.8 GHz has that easily beat), the 128 CUDA
>>> compute units on the GPU offer massive floating point math capability.
>>> 
>>> On Sun, Apr 28, 2019 at 11:18 PM Scott Traurig <scott.traurig at gmail.com> wrote:
>>>> 
>>>> ***** High Performance Software Defined Radio Discussion List *****
>>>> 
>>>> What is the advantage of these boards over simply using an i7 mini-ITX board?
>>>> 
>>>> 73,
>>>> 
>>>> Scott/w-u-2-o
>>>> 
>>>> _______________________________________________
>>>> HPSDR Discussion List
>>>> To post msg: hpsdr at openhpsdr.org
>>>> Subscription help: http://lists.openhpsdr.org/listinfo.cgi/hpsdr-openhpsdr.org
>>>> HPSDR web page: http://openhpsdr.org
>>>> Archives: http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/
>>> _______________________________________________
>>> HPSDR Discussion List
>>> To post msg: hpsdr at openhpsdr.org
>>> Subscription help: http://lists.openhpsdr.org/listinfo.cgi/hpsdr-openhpsdr.org
>>> HPSDR web page: http://openhpsdr.org
>>> Archives: http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/
>> 
>> _______________________________________________
>> HPSDR Discussion List
>> To post msg: hpsdr at openhpsdr.org
>> Subscription help: http://lists.openhpsdr.org/listinfo.cgi/hpsdr-openhpsdr.org
>> HPSDR web page: http://openhpsdr.org
>> Archives: http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/
> 



More information about the Hpsdr mailing list