[hpsdr] Phoenix

[ad0es]

Hi,

One benefit that I see is that with the FFT, FIR, etc processing being 
done on the GPU/CPU as opposed to
the FPGA the algorithms used can be quickly changed without having to 
re-program an FPGA.  This also allows
a lot more users to be able to play with the algorithms. Few people have 
the ability to program FPGA hardware.

Another that I plan to work on is diversity reception with 2 ADCs. With 
all the I/Q data sitting in the Cuda/GPU space
at the end of the DSP pass, you could continue to process the 2 streams 
for diversity.

Another is the shear number of virtual receivers you can support.  I 
once had 57 running in a test.

I also see the possibility of doing digital mode decoding of the I/Q 
data with the Cuda/GPU.

And how about Automatic Signal Classification?  Have Cuda code look for 
all the Olivia signals it can find...

Where I first started the sw I was using a GTX780 which has 2304 cores.
I did the 57 receiver test with a gtx1080 which has 2560 cores.
But now the "standard" is more like an RTX3070 which has 5888 cores.
Step up to an RTX3090 and get 10496 cores.
And you can put several in the same box to double your processing power!

The common denominator in this is Cuda processing.  You can do MUCH, 
MUCH more than just FFT/FIR processing
with this combination of hardware.

Steve, AD0ES

On 4/15/23 6:22 AM, Helmut Oeller wrote:
>        *Just my 2 cts: When Phil, VK6PH, presented his DFC idea at the SDR
>        Academy 2016 (HR Friedrichshafen) he spoke about a firehose to
>        fill the total amount of digital data (1. Nquist zone) directly
>        into a PC for processing. This is certainly an interesting
>        approach for (presently) frequencies up to 50 MHz. But e.g.
>        compared with HERMES I can't detect any benefit of the important
>        RF performance as the ADC/DAC is the limiting factor (IM,blocking,
>        phase noise etc.)
>        
>
>