[hpsdr] Questions about Ethernet and Bandwidth

[John Miles]

> octets for the UDP header and 12 octets IP header.  Ethernet frame
> overhead is 38 octets.  Metis also has a packet header of 8 octets.  Each
> Metis packet holds 2x512 byte HPSDR frames.  That means that each packet
> is 1090 octets -- that's 1720 bits.  That means that you can fit 581,395
Metis
> packets per second down the pipe.  Each one of those packets holds 126 I/Q
> samples.  That's 73,255,770 samples per second.  Nyquist says that we
should
> have about twice the sample rate of the highest frequency we want to
> represent..  That means that you'll get about half the bandwidth as your
> sample rate.  That means that you'll be able to see about 36.627 MHz.

The good news is that an I/Q signal carries all of the information that's
preserved by sampling an equivalent real-only signal at 2x.   This makes
sense if you think about it -- you're getting two samples per cycle either
way.  From the FFT's point of view, you get a true double-sided spectrum
with a complex signal, not just two mirrored halves as with real data.  So a
73 MS/s I/Q signal actually gives you an honest 73 MHz of tunable spectrum,
if your CPU or GPU can drink from the fire hose!

Of course your I/Q downconverter has the same real-world constraints on
filter slopes that real mode signal chains have, so the 'honest' 73 MHz will
need to be somewhat narrower in practice.
 
-- john, KE5FX


 1314318304.0