Design considerations for (2m) long signal lines

[Peter S. May]

I'm working on a microcontrolled LED lighting setup in which breakout
boards with eight outputs each would be connected to the microcontroller
board by cables (CAT-5 is what's on hand) at a length of somewhere
between 1m and 3m. The breakout boards would contain some sort of
D-type flip-flop or latched parallel-out shift register with outputs
connected to transistors that then drive the LEDs. Using a shift
register would have three signal lines (signal, shift clock, latch
clock) and two supply lines.

I need sort of a run-down of the design considerations I need to account
for, and I'm not sure where to look. Are decoupling capacitors (say,
..1uF) going to be enough to keep the ground stable? Will putting
Schmitt triggers on all the input signal lines of the breakout boards
help? Without resorting to differential signaling, what limits will I
run into for the data rate? Will a long cable make a significant
addition to propagation delay?

Thanks in advance for any insight...
PSM

 

[John B]

I'm working on a microcontrolled LED lighting setup in which breakout
boards with eight outputs each would be connected to the
microcontroller board by cables (CAT-5 is what's on hand) at a length
of somewhere between 1m and 3m. The breakout boards would contain
some sort of D-type flip-flop or latched parallel-out shift register
with outputs connected to transistors that then drive the LEDs.
Using a shift register would have three signal lines (signal, shift
clock, latch clock) and two supply lines.

I need sort of a run-down of the design considerations I need to
account for, and I'm not sure where to look. Are decoupling
capacitors (say, .1uF) going to be enough to keep the ground stable?
Will putting Schmitt triggers on all the input signal lines of the
breakout boards help? Without resorting to differential signaling,
what limits will I run into for the data rate? Will a long cable
make a significant addition to propagation delay?

Thanks in advance for any insight...
PSM

Sounds like a job for RS485 drivers and receivers. Look for MAX485 and
equivalents.

 

[Peter S. May]

Sounds like a job for RS485 drivers and receivers. Look for MAX485 and
equivalents.

Is there anyway to do this without overcomplicating things? It seems
like too simplistic an application to bother with any but the TX line,
which I'm guessing _could_ be done, but then how would I clock the latch?

 

[Eeyore]

I'm working on a microcontrolled LED lighting setup in which breakout
boards with eight outputs each would be connected to the microcontroller
board by cables (CAT-5 is what's on hand) at a length of somewhere
between 1m and 3m. The breakout boards would contain some sort of
D-type flip-flop or latched parallel-out shift register with outputs
connected to transistors that then drive the LEDs. Using a shift
register would have three signal lines (signal, shift clock, latch
clock) and two supply lines.

I need sort of a run-down of the design considerations I need to account
for, and I'm not sure where to look. Are decoupling capacitors (say,
.1uF) going to be enough to keep the ground stable? Will putting
Schmitt triggers on all the input signal lines of the breakout boards
help? Without resorting to differential signaling, what limits will I
run into for the data rate? Will a long cable make a significant
addition to propagation delay?

Use a low signalling rate and these won't be serious issues. There's no point in
using some over-specced protocol for this. I'd be inclined to use something
similar to I2C which only needs one data and one clock line and run it at maybe
10k baud or thereabouts. There are I2C 'expanders' available. Check out PCF8574
for example.

Graham

 

[MooseFET]

Is there anyway to do this without overcomplicating things? It seems
like too simplistic an application to bother with any but the TX line,
which I'm guessing _could_ be done, but then how would I clock the latch?

Yes, it is way to complicated and not likely to be needed. For modest
speeds and modest lengths of cable you can use the following:

The drivers are HC buffers with a resistor between its output and the
cable. The resistance of this resistor should be something like 1/3
the characteristic impedance of the line.

The receiver end has a pair of Schottky diodes clamping the signal to
the rails and then perhaps a small resistor (50R) to the input of a
74HC14.

For a lossless line with correct matching, the swing on the two ends
are equal. The 1/3 Z0 resistance would make the swing at the cable
end 3/4 of the Vcc if the far end was correctly terminated. The far
end is in fact allowed to swing by Vcc making the line appear to be
mismatched at the far end. You get some reflections on the cable as a
result but they are not enought to cross the boundary of the 74HC14.

 

[John B]

Yes, it is way to complicated and not likely to be needed. For modest
speeds and modest lengths of cable you can use the following:

The drivers are HC buffers with a resistor between its output and the
cable. The resistance of this resistor should be something like 1/3
the characteristic impedance of the line.

The receiver end has a pair of Schottky diodes clamping the signal to
the rails and then perhaps a small resistor (50R) to the input of a
74HC14.

For a lossless line with correct matching, the swing on the two ends
are equal. The 1/3 Z0 resistance would make the swing at the cable
end 3/4 of the Vcc if the far end was correctly terminated. The far
end is in fact allowed to swing by Vcc making the line appear to be
mismatched at the far end. You get some reflections on the cable as a
result but they are not enought to cross the boundary of the 74HC14.

Sorry, I think you misunderstood me. I just meant translate to RS485
levels for the long cables as below. OK, so it's not a cost effective
solution, but will be totally reliable and noise immune. It really
depends if the OP is just making a one-off.



|\ RS485 levels |\
Clock In | \------------------------| \ Clock Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Data In | \------------------------| \ Data Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Latch In | \------------------------| \ Latch Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

(created by AACircuit v1.28.6 beta 04/19/05 www.tech-chat.de)

 

[John Larkin]

I'm working on a microcontrolled LED lighting setup in which breakout
boards with eight outputs each would be connected to the microcontroller
board by cables (CAT-5 is what's on hand) at a length of somewhere
between 1m and 3m. The breakout boards would contain some sort of
D-type flip-flop or latched parallel-out shift register with outputs
connected to transistors that then drive the LEDs. Using a shift
register would have three signal lines (signal, shift clock, latch
clock) and two supply lines.

I need sort of a run-down of the design considerations I need to account
for, and I'm not sure where to look. Are decoupling capacitors (say,
.1uF) going to be enough to keep the ground stable?
Probably.

Will putting
Schmitt triggers on all the input signal lines of the breakout boards
help?

Absolutely. Even better if you put an r-c lowpass filter at the input
of each schmitt, to nuke any cable ringing. Use HC-type schmitts.

Without resorting to differential signaling, what limits will I
run into for the data rate? Will a long cable make a significant
addition to propagation delay?

If you run single-ended, unterminated, and use maybe 1 us tau r-c's in
front of schmitts, figure 200 kbps is safe, 2.5 us high/low levels. A
hundred feet of cable would be OK, if you don't have ground loop
problems.

If you need to go faster, more care will be required, preferably
terminated differential levels.

John

 

[[email protected]]

I'm working on a microcontrolled LED lighting setup in which breakout
boards with eight outputs each would be connected to the microcontroller
board by cables (CAT-5 is what's on hand) at a length of somewhere
between 1m and 3m. The breakout boards would contain some sort of
D-type flip-flop or latched parallel-out shift register with outputs
connected to transistors that then drive the LEDs. Using a shift
register would have three signal lines (signal, shift clock, latch
clock) and two supply lines.

I need sort of a run-down of the design considerations I need to account
for, and I'm not sure where to look. Are decoupling capacitors (say,
.1uF) going to be enough to keep the ground stable? Will putting
Schmitt triggers on all the input signal lines of the breakout boards
help? Without resorting to differential signaling, what limits will I
run into for the data rate? Will a long cable make a significant
addition to propagation delay?

Thanks in advance for any insight...
PSM

I think cat5 is around 5ns/m so that shouldn't be a problem and all
signals will be delayed the same anyway.

if have fast edges and the cable isn't terminated correctly you will
get reflections that might result in the receiver seeing multiple
clocks for every clock edge.

if you slow down the edges noise right as it goes through the
threshold. might result in the receiver seeing multiple clocks for
every clock edge.
using Schmitt triggers on the clock inputs should help on that

the data signal shouldn't matter, the value should have settled before
the clock edges

-Lasse

 

[Mike Silva]

Sorry, I think you misunderstood me. I just meant translate to RS485
levels for the long cables as below. OK, so it's not a cost effective
solution, but will be totally reliable and noise immune. It really
depends if the OP is just making a one-off.

|\ RS485 levels |\
Clock In | \------------------------| \ Clock Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Data In | \------------------------| \ Data Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Latch In | \------------------------| \ Latch Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

I'm with you on this. I don't see that the other methods suggested so
far offer a substantial savings at all, and I think they may be
considerably underestimating the data rate (which seems to be at least
100kbps based on the description). And once you double everything
(cable length, data rate) for the inevitable future expansions the
case gets even better.

 

[Rich Grise]

Sorry, I think you misunderstood me. I just meant translate to RS485
levels for the long cables as below. OK, so it's not a cost effective
solution, but will be totally reliable and noise immune. It really
depends if the OP is just making a one-off.

|\ RS485 levels |\
Clock In | \------------------------| \ Clock Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Data In | \------------------------| \ Data Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

|\ RS485 levels |\
Latch In | \------------------------| \ Latch Out
----------| \ | \----------
| / | /
| /------------------------| /
|/ |/

I'm with you on this. I don't see that the other methods suggested so
far offer a substantial savings at all, and I think they may be
considerably underestimating the data rate (which seems to be at least
100kbps based on the description). And once you double everything
(cable length, data rate) for the inevitable future expansions the
case gets even better.[/QUOTE]

Well, if you're going full-goose bozo, you might as well use twisted-pair
ribbon cable:
<http://www.mouser.com/search/Refine...23038+4294795225+254614+0&Ns=P_SField|0&Msb=0>
[Mind the wrap: I've put the brackets but it still might not be
clickable; just copy/paste if that's the case.]

But it still seems like overkill for 6 feet. I'd be happy to drive 6'(2M)
of ribbon cable with HC logic, interleaving signals and ground, albeit
it takes the same number of conductors, but they don't have to be twisted.
(and all the standard goesinta-goesoutta interfacing schtuff.)

Cheers!
Rich

 

[Mike Silva]

But it still seems like overkill for 6 feet. I'd be happy to drive 6'(2M)
of ribbon cable with HC logic, interleaving signals and ground, albeit
it takes the same number of conductors, but they don't have to be twisted.
(and all the standard goesinta-goesoutta interfacing schtuff.)

I understand what you're saying. I would only say that once he adds
the HC drivers on the PIC side and the Schmitt trigger (and maybe some
RC) on the shift-register/LED side, it's hard to see that as _much_
simpler. On top of that I did automatically double the specified
length and assumed a fairly high data rate (figuring he didn't want to
spend all of his time just outputting bits, but rather wanted to
squirt them out pretty quickly and get on to other processing tasks).
But it's only to the good to have the different suggestions all
spelled out, and he can decide which one he wants to try.

Like I said, I am genuinely interested in other people's suggestions
as well.

 

[MooseFET]

Sorry, I think you misunderstood me. I just meant translate to RS485
levels for the long cables as below. OK, so it's not a cost effective
solution, but will be totally reliable and noise immune. It really
depends if the OP is just making a one-off.

No, I didn't misunderstand. Note that the OP has 3 meters of cable,
and signals from a micro. The cable is short.