[hpsdr] Thetis for Atlas based HPSDR

[ad0es]

Hi,

I'm taking a simpler approach than Minerva, KISS!

First my board will be receive only. It makes the circuit about 1/4th as 
complicated, as
well as keeping noise issues simpler. The cost of Cuda processing 
(compute cost, not $$$)
an xmit stream is as great as a single recv stream, its basically just 
the same math in reverse order.
I can justify the use of 10-20 or more receivers, but you only need one 
xmitter, while that xmitter
costs you half of your Cuda resources. In other words, 50+ receivers, or 
maybe 10 or less receivers
and one xmitter?

Second I am doing this on a standard PCIe board, not the M2 version 
designed to fit on a Jetson.
Layout and fab should be much simpler.  Its really just a smallish FPGA 
to move the samples from
ADCs to computer, 2-4 ADCs, and simple front ends consisting of input 
protection, grounding relay,
attenuator, and possibly an LNA.

When I started the sw I was using a GTX750ti cuda board with 640 cuda 
cores.  I was using a GTX1080 with
2000+ cores when the project "stopped".  Today the equivalent RTX3080 
has over 10,000 cores.
The "future friendly"  aspect of the cuda approach can't be overstated. 
If the FFT etc. were in the FPGA
the capabilities would be forever frozen in the board design. And you 
can throw multiple Cuda boards
at the problem if needed...

Steve AD0ES

On 12/1/22 7:28 AM, Helmut Oeller wrote:
> ***** High Performance Software Defined Radio Discussion List *****
>
> Hi,
>
> chick Minerva is still sitting in its nest since Phil, VK6PH,
> introduced  his DFC approach on the HamRadio Friedrichshafen 2016. No
> surprize for me: Eliminating the FPGA, spending a fast PCIe interface
> and let a Jetson board do the job is in my eyes no big benefit at least
> under economic aspects. RF performance is determined by ADCs, frontend
> circuits, phase noise of clocks for sampling and PLLs. Time was moving
> on and very sophisticated SoC FPGAs are available today saving a maximum
> of hardware, but that's presently the playground of the professional
> developers for 5G stuff etc.
>
> Take it easy and care for your (approved) HPSDR hardware.
>
> 73, Helmut, DC6NY
>
> Am 01.12.2022 um 10:59 schrieb Marc OLANIE:
>> ***** High Performance Software Defined Radio Discussion List *****
>>
>> Hi all
>>
>> I was really convinced that the Minerva project was in "slow 
>> evolution", this orbituary  is a real shock for me
>> Minerva is, imho, the best solution to decrease the cost of digital 
>> processing, to imagine modular SDRs with multiple frontends, to get 
>> rid of bus limitations thanks to PCIe performance, and especially to 
>> see a successor to the previous hpsdr generation. (thanks to the 
>> bicoin miners who did a lot to lower the cost of gpu :- )) )
>> Unfortunately I have no skills in development (vhdl or other), and 
>> I'm not very good in electronics. But couldn't we imagine a 
>> collaborative effort to produce a series of small platforms, with a 
>> comparable spirit to what allowed Steve Haynal to realize the 
>> Hermeslite V1 project?
>> Too much work has been invested in Minerva to let it die in general 
>> ignorance.
>>
>> Mark f6itu