Well now we are at the point of our
Saturday discussion. How fast? How Clean, How Many? None of us knew, and we
still don’t. Cecil did know and all of you have pointed to the solution -
lvds. The buss stuff really all started to unravel when Phil_C said he needed an
FPGA on the Radio board due to the probable lack of capability of the buss to
carry the A/D and other high speed sigs to and from the Lionheart. Bill –
KD5TFD and I sort of said the same thing Saturday: “If we can’t
leverage the Lionheart for all buss mounted FPGA projects now and in the
future, what is the point of the buss?” It would be nice to see any pin
pair be able to pass any frequency quietly. If it’s not possible then it’s
not possible. Also according to Cecil, an FPGA interfaced to an FPGA can do
lvds without any additional components but we still need twisted pair, or coax.
Cecil suggested a .100 pin header to
which we would attach a twisted pair to a single pair block going from board to
board. It is a good idea for a couple of pairs of pins, but if we need 16 pair or
probably more, it is a little impractical, but do-able.
I made a couple of suggestions for mass
termination or even a ‘flexible buss’ using twisted pair IDC cable which
didn’t garner any comments. This stuff does have flats and twists so also
may be impractical.
Call the ball!
email@example.com [mailto:firstname.lastname@example.org] On Behalf Of KD5NWA
Sent: Tuesday, January 24, 2006
Subject: Re: [Xylo-SDR] RFC -
I've built Computer systems using 2X EuroCards using double connectors
and no problems whatsoever back in early 80's, at that time the clock speed was
50MHz. Even though it was a two layer motherboard the signals were absolutely
clean with no ringing and coupling between lines, part of that was because of
massive ground planes where possible and the signals were isolated in groups
with common timing with three ground pins between groups. The system clock was
isolated all by itself away from other signals. Power was on the outsides of
the connectors with massive pours of copper to keep the power bus impedance
low. It was not a radio however.
That being said if you have 50MHz to 1000MHz signals, LVDS will keep those
signals very clean if you use tightly twisted wires such as the ones found
inside a CAT5e cable. I would assume that there would not be too many such
signals that have to move from card to card.
My suggestion is as before if you have a very high frequency signal that you
want to distribute between cards, do it on a connector on the top of the card,
using LVDS signalling and twisted cables, or use coaxial cables with SMT
connectors, but be aware that the output of a FPGA or logic is not too happy
with coaxial cables due to impedance mismatch, which causes ringing and phase
jitter, you know it as high SWR and it's reflections. LVDS is meant to use
tightly twisted wires with 110 Ohm impedance and a terminating resistor, which
is exactly what CAT5e cabling is, you add the resistor on the receiver side if
it does not have one built in.
The original box was fine with me, all we needed was an additional 1/8"
between the cards and the front plate, which I feel board mounted connectors
could tolerate. But if you are going to use a tiny MicroATX PC case, that is
fine too since there are many choices on the cabinet, I pointed one yesterday
that was reasonable in size and very inexpensive ($27). The one I pointed out
allowed easy access to the boards and had plenty of room for other components
such as ovens or GPS satellite receivers.
If we allow for three pairs of signals that can be programmed as a LVDS driver
to a Berg header on the top of the FPGA card, that will allow us if needed to
pipe a high speed clock out of or into the FPGA in case the bus has problems
with noise radiation.
At 05:38 PM 1/24/2006, you wrote:
Or should I say Motherboard Woes.
Cecil, Bill, Steve and I had a pretty long Teamspeak session last Saturday
discussing buss speed. What speeds CAN we put on our buss?
None of us really knew! Cecil knew a lot about LVDS and twisted pair, but we
dont have a twisted pair or coax buss! In fact we tried to figure a way to
get it into our tiny, cramped, custom designed enclosure, which prompted my
departure from the enclosure, buss connectors, etc.
However, the question remains! How DO we design our buss to go the 133 mhz PCI
standard or increasing speeds in the later PCI standards. (800 mhz and beyond?)
Can we even go 24 mhz to the FPGA, divide by 2 and send it back to the Janus?
My thought was to try to copy the PCI standards (Not the connector,
just the board spec for the standard). I found again, as I knew before, PCI
standards group is a private industry culb with a high price to join. When you
get right down to finding out more than the buss labels, there aint much
but general info. If you take a look at a mother board or even a riser you sure
cant see the innards! There aint no schematics ANYWHERE I can find!
I did find some info in a pdf from a PCI Bridge Chip manufacturer, on how to
design with their chip which yielded some light on the board design for 133
I didnt have time to read all the stuff and it was a bit beyond me,
however I did see one highlight as to how the PCI buss needs to be in order to
support high speed design.
The clocks and other Critical signals go on the top surface of the
This is followed by a groundplane layer.
Followed by the Power supplies layer
Followed by the bottom Signal layer (which I took to mean slow speed
That was pretty sketchy but it was the description and perhaps a
start. Is this doable for our MB?
I would hate to design and produce a motherboard which couldnt even pass
I2C stuff without causing noise everywhere else. This needs to be versatile and
Any suggestions as to where to go with a quiet, high speed MB design?
Xylo-SDR mailing list
To post msg: Xylo-SDR@ae5k.us
Subscription help: http://lists.ae5k.us/listinfo.cgi/xylo-sdr-ae5k.us
Xylo-SDR web page: http://xylo-sdr.ae5k.us
Forum pages: http://www.hamsdr.com/hamsdrforum/
I fail to see why doing the same thing over and over and getting the same
results every time is insanity: I've almost proved it isn't; only a few more
tests now and I'm sure results will differ this time ...