Seeking the solutions of high speed interconnection for the long distance transmission of 3.3v/24MHz

[X.Y.]

Seeking the solutions of high speed interconnection for the long
distance transmission of 3.3v/24MHz signals.

We need to connect a CMOS image sensor and a FPGA chip. The distance
between them is approximately 1 meter. The output signal of the sensor
is 3.3v and 24MHz. However, the length of the original cable of it is
only 1 centimeter, so it should be prolonged. If they are connected by
a cable directly, the signal would attenuate greatly which will cause
the system doesn't work. In my opinion, an interconnecting circuit
should be added between the sensor and the FPGA chip to process the
signal before and after the transmission. By now, the methods I have
found are as follows:
1, Using LVDS signal Transceiver and Receiver
The drawback of the method is: there are more than 10 bits signals,
which need many of this kind of chips. It will take too much place in
PCB and it also cost much.
2, Using chips of Serializer and Deserializer.
Both Maxim and National Semiconductor have such chips such as MAX9247
and DS90C241. It is a good method. Unfortunately the solution is fired
by my tutor because these chips are expensive.

My question is whether there are any other solutions? More cheap will
be better. Thanks a lot.

 

[John Larkin]

Seeking the solutions of high speed interconnection for the long
distance transmission of 3.3v/24MHz signals.

We need to connect a CMOS image sensor and a FPGA chip. The distance
between them is approximately 1 meter. The output signal of the sensor
is 3.3v and 24MHz. However, the length of the original cable of it is
only 1 centimeter, so it should be prolonged. If they are connected by
a cable directly, the signal would attenuate greatly which will cause
the system doesn't work. In my opinion, an interconnecting circuit
should be added between the sensor and the FPGA chip to process the
signal before and after the transmission. By now, the methods I have
found are as follows:
1, Using LVDS signal Transceiver and Receiver
The drawback of the method is: there are more than 10 bits signals,
which need many of this kind of chips. It will take too much place in
PCB and it also cost much.
2, Using chips of Serializer and Deserializer.
Both Maxim and National Semiconductor have such chips such as MAX9247
and DS90C241. It is a good method. Unfortunately the solution is fired
by my tutor because these chips are expensive.

My question is whether there are any other solutions? More cheap will
be better. Thanks a lot.

24 MHz isn't very fast. A 20-conductor ribbon cable (with alternating
grounds) should be fine, and source termination sounds like a good
idea. If the CMOS sensor drive strength is known to be weak, buffer
each logic signal.

John

 

[Joel Kolstad]

1, Using LVDS signal Transceiver and Receiver
The drawback of the method is: there are more than 10 bits signals,
which need many of this kind of chips. It will take too much place in
PCB and it also cost much.

There are plenty of quad LVDS drivers and receives in 16 pin TSSOP packages...
you'd need 3 of them. This is not exactly "a lot" of board space, is it?
National Semiconductor has some that do 28 bits at a swat.

2, Using chips of Serializer and Deserializer.
Both Maxim and National Semiconductor have such chips such as MAX9247
and DS90C241. It is a good method. Unfortunately the solution is fired
by my tutor because these chips are expensive.

Is it really that expensive compared to the cost of your sensor and your FPGA?
It almost sounds as if your "tutor" has a particular solution in mind he's
trying to get you to produce rather than letting you just meet some real specs
(such as an *overall system* price) however you'd prefer.

My question is whether there are any other solutions? More cheap will
be better. Thanks a lot.

-- John's approach of just sending the data over a ribbon cable -- with
interspersed grounds if you like, or even twisted-pair ribbon cable (although
this is spendy) -- will probably work fine.
-- Presumably you have an ADC for your CMOS sensor. How about replacing it
with one containing a built-in LVDS interface? (e.g., AD9219)
-- You might find it cheaper to use an "application specific" SerDes such as
those meant for DVI (see, e.g., Analog Devices' "HDMI transmitter/receiver
selection tables")
-- Since it is a camera, you might use the Camera Link interface
(www.alacron.com/downloads/vncl98076xz/CameraLinkSPEC.pdf). It is just LVDS,
but the point here is to tell your tutor that -- at least if this is an
industrial application -- if Dalsa and Pulnix and Coreco and others can afford
LVDS, so can you. Using Camera Link has the significant upside of being "plug
and play" with many commerial framegrabbers as well.

---Joel

 

[X.Y.]

To John Larkin: Thanks for your reply. Does the ribbon cable with
alternating grounds means that the cable should be connected to
signal, ground, signal, ground...? And, I think the CMOS sensor drive
strength is weak because it is used in a mobile phone and has a power
of only 90mW. It's IOH is 8mA and IOL is 15mA. Then, should I use a
buffer or bus transceiver chip on the output of the sensor?

To Joel Kolstad: Thanks for your recommendations. Maybe my tutor just
wants to shrink the cost. We do not have an ADC for the CMOS sensor.
We just buy it and its output is already digital signal. Thank you!

 

[John Larkin]

To John Larkin: Thanks for your reply. Does the ribbon cable with
alternating grounds means that the cable should be connected to
signal, ground, signal, ground...?
Yes.

And, I think the CMOS sensor drive
strength is weak because it is used in a mobile phone and has a power
of only 90mW. It's IOH is 8mA and IOL is 15mA. Then, should I use a
buffer or bus transceiver chip on the output of the sensor?

It would probably be OK without it, but if you have the room and the
money, buffer it. If it drives the cable directly, you could also get
ground bounce that affects the imager circuits. Buffered or not,
source terminate (100 ohms in series with each live ribbon cable lead
at the driver end) for miminum current glitching; that way, each
driver pin sees a 200 ohm load. Don't terminate the far end.

John

 

[X.Y.]

It would probably be OK without it, but if you have the room and the
money, buffer it. If it drives the cable directly, you could also get
ground bounce that affects the imager circuits. Buffered or not,
source terminate (100 ohms in series with each live ribbon cable lead
at the driver end) for miminum current glitching; that way, each
driver pin sees a 200 ohm load. Don't terminate the far end.

John

To John Larkin: Thank you for your help. We had design a PCB board to
test the method you told us and we could get it next Tuesday (April,
24th). However, we do a simple experiment to have a simple qualitative
analysis. We use a scope of 2.5G/S and 10GSa/S. The conclusion is
fairly positive. The figures can not be posted here, so they are
posted on my blog. So please visit the site: http://xieyu1219.blogspot.com/
to see it. Thank you! If we have further information, we will also
tell you.