[hpsdr] Ozy with Gb Ethernet

[jeff millar]

Warren Walsh wrote:
  Are there any general
> guidelines that I should following? If anyone has any ideas already, I'd 
> be happy to hear them.

First a bit of qualification.  I'm currently managing the design an SDR with 100bT and a cyclone II.

A couple of points that I haven't see mentioned, yet.

   - Using Ethernet doesn't require using Layer 3 IP, UDP, TCP
   - Vanu (an SDR company) uses layer 2 packets on GigE
   - "ATA over Ethernet" also uses L2 packets for disk I/O
   - Layer 2 skips the complexity of IP, TCP, UDP
   - L2 protocols easily fit in a FPGA
   - Less than 2000 LE's for PHY to FIFO with control/status registers
   - Software generates L2 packets using the "raw Ethernet" facility
   - Linux can generate raw Ethernet packets easily
   - Use "Ethereal" for testing with raw packets
   - L2 packets can pass through a switch but not a router
   - Addressing based on MAC address, not IP address

Suggest using a dedicated Ethernet card for SDR so timing doesn't get messed up by the network stack.

On the subject of 1000bT vs 100bT.  The effective throughput of 1000bT probably runs about 50M 
bytes/second...considering various overheads.  Assume 2 bytes per sample, and that 25M samples per 
second or about 6-10 MHz of RF bandwidth...depending on the DSP approach taken.  100bT offers about 
600 KHz to 1 MHz of bandwidth.

The question becomes, what does software do with a 1000bT firehose of bandwidth?  I think we need to 
calculate the maximum data rate that the current SDR algorithms can support.  Since they generally 
hook to audio A/D with rates in the KHz range, it will probably take a lot of work by Moore's Law 
before processors can deal with 10s of MHz sample rates.

I think 100bT will support any processor out there for the foreseeable future.

jeff, wa1hco










 1173960698.0