[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Discuss-gnuradio] Gigabit ethernet
From: |
Stephane Fillod |
Subject: |
Re: [Discuss-gnuradio] Gigabit ethernet |
Date: |
Tue, 23 Nov 2004 10:25:55 +0100 |
User-agent: |
Mutt/1.5.6+20040907i |
Hi David,
On Tue, Nov 23, 2004 at 02:32:40AM -0600, David Carr wrote:
> Has anyone looked into using gigabit ethernet is an interface to
> software radio hardware?
What a coincidence, I just added a page to the wiki no later than this
week-end. Have a look at http://comsec.com/wiki?USRPnotUSB and let me
know what you think.
REM: btw, how to do we pronounce "USRP"?
> There are some definite benefits:
> -max data rate ~750Mb/s, the rest is lost as signaling overhead
> -can do nifty tricks like multi-casting to multiple hosts or sending
> different streams to different hosts. One could then throw multiple
> machines at high-bandwidth problems
> -gigabit ethernet interfaces are becoming common and cheap
> -hardware is not that expensive (PHY ~$15)
> -can remotely locate hardware
and much more, especially with cable and tapping at the feed point.
> Downsides
> -have to implement PCI or gigabit MAC in FPGA
> -not trivial to implement (either of above)
> -latency (could be mitigated to < 1ms -- comparable to USB)
>
> It seems like to best approach to this would be to use an external PHY
> and implement the MAC in the FPGA. The PHY would use the GMII or RGMII
> interfaces. Implementing the MAC is difficult but doable and there
> exist things like the 10/100 opencores MAC to use as a guide.
A simplyfied MAC can be implemented, ie. only full-duplex = no
collision, or even, single speed (=no speed negociation), which are quite
complex stuff otherwise. There are PHY that are SGMII, on top of a SerDes link.
I don't know about what's available for MAC as openIP, but I've heard
manufacturers are about to embed GigE MAC (diffused) in their FPGA.
Such a GNU Radio device would need a small flash/eeprom, at least for a
minimal boot to request through BOOTP/TFTP firmware and FPGA bitstream.
A standalone device would be better though.
> How much
> network stack would we need? I think that we'd use the jumbo frame size
> and UDP.
Yes, at GigE speed, jumbo frame (<9.6KiB) has still less latency than regular
100Mbps frames. Talking about protocol, UDP is mostly recommanded and maybe
RTP which comes in handy for timestamps, etc.
FPGA wise, the UDP/RTP is just a context prefixed to the payload, no
clever logic here.
Cheers,
Stephane