MakerBot Industries

But, Protonite, in this case, the "LOW" level should have been only temporary, isn't it?

I couldn't detect any state like that with the oscilloscope during my firsts tests (It's still possible that I missed that "moment").

It seems that my delay/and 0xFF transmit before 0xD5 start mitigates a little bit the problem, but I still see "good packets" with no reply from the extruder.

May be the path is a resend of packet when we have no reply from extruder, and a more robust protocol.

Try adding a pullup to one line and a pulldown to the other so that when both transmitters are tri-state the receivers see a logic 1. Modern RS485 drivers guarantee the receiver gives logic one when its inputs are open circuit, but I think MB uses the old 5V ones which do not.

I just checked, the SN75176 really has NO failsafe options and only 2kV ESD protection. Thanks for the hint. Didn't even consider the makerbot-team would dare using such an outdated transceiver…
I'll replace my transceivers with something like the SN65HVD22 or the SN65HVD12 as soon as possible.

That reminds me. See if R16 on the motherboard, the 180 Ohm resistor, is populated. A quick hack to add termination would be to snip a Cat5 cable, add some resistors and plug it into one of the empty extruder ports on the motherboard end.

Even better, pg 12 of the SN75176A datasheet sez both ends of the cable should be terminated with it's characteristic impedance. For Cat5e that's 100 Ohms (see http://en.wikipedia.org/wiki/Category_5_cable#Electrical_characteristics_for_Cat.5e_UTP).

Try adding a 100 Ohm resistor to the extruder's R1 and 100 Ohm to the motherboard's R16.

You don't need termination with the baud rates and cable lengths involved. The reflections will have died down several orders of magnitude less time than a bit.

The problem may well be that the line floats the wrong way between packets though so that first character is corrupted. A pull up on one line and pull down on the other will fix that.

But it's in a floating high-Z state between packets so other transceivers on the bus know it's safe to transmit. Adding the termination as documented will stop the differential pair from glitching relative to one another via inductive and/or capacitive coupling effects.

Another variation:

• On one end only (the motherboard Cat5 connector hack would be easiest way to try), terminate with two 50 Ohm resistors in series and hook up a larger resistance (say 1K Ohm) from the center tap point to ground. That will keep things matched and the diff pair tied together and within supply-reference bounds in the high-Z state without adding (many) ground loop paths.
• At all other points (ie. the extruder(s)) only add the 100 Ohm termination across the diff pair.

Just found this technical reference for a proper pull-up/down failsafe bias implementation: http://www.national.com/an/AN/AN-847.pdf

The transceivers "know" nothing, they are not intelligent and they cannot detect a tri-state condition. They simply convert the differential signal to a TTL level.

Adding termination ensures both lines will exactly the same voltage when floating. For the 75176 this leads to an undefined output but UARTS require the receive line to idle high. This is fixed by adding a weak pull up and pull down to make sure the differential lines float to slightly different levels. This is known as failsafe biasing and modern transceivers include it, but for the 176 it needs adding externally. If you are using termination then the three resistors have to be chosen to give the correct impedance and the correct voltage to bias the receiver to a defined level. In this application termination is not needed so just a 4k7 pull up on one line and 4k7 pull down will work fine.

We are in agreement on the failsafe biasing.

I get nervous when proper termination gets ignored and/or forgotten. I've seen too many IC designs fail for similar reasons.

Still, for simplicity, and to wrap it all up:
I would like to point out for the makerbot industries that they should swap that old fashioned trasceiver for a newer model immediately, so the next production run of mainboards and extruder boards does not have to deal with problems like that ever again!
It does not cost more, no changes to the layout necessary, it's just a drop in replacement that solves some common problems. Would be really nice if you could organize a parts upgrade coordinated with all the places like thingiverse and reprap and wherever there are different partlists for the same product around. (I dont have the global view of that structure yet…)

About the price: I did not check the prices actually, but using a 10Mbit transceiver as the one that is currently used definately is overkill for our application and the faster slew rates actually produce more noise (EMI), so you should probably try to settle for something <= 1MBit. That would help to lower the price too, as the max speed is one of the main price driving factors of these transceiver circuits.

If you need more advice and someone looking into the vast arrays of transceiver options, pointing you to a replacement part number, just let me know and i will step up to the task.

I think you will find that slew rate limited and failsafe drivers will actually be a lot more expensive.

Okay, i did check some prices and you are right, SN75176A costs about 0.36$ at volume price and one of the cheapest new generation transceivers from TI (SN65HVD3082E) with all the failsafe options (short, open, idle), a 0.2Mbps signaling rate and 15kV EDS rating costs around 1$ at the same volume price (price for 1k Units, per piece, taken directly from ti.com).
One of the other big features of this particular newer driver is that its a lot less power hungry. Instead of having a max Icc rating of 50mA it only asks for 0.9mA. The way to go green, if you need another argument… ;-)

Now, you can make up your own mind if you would rather pay a dollar more and have a hardware that is almost foolproof / failsafe and uses a lot less power, or if you want to continue on the "as cheap as possible" path.

Well for my own stuff I have used the 3.3V slew rate limited drivers for anything in the last ten years. But, in this case, adding the bias resistors will fix all the problems and is the simplest retrofit option. There is nothing wrong with using the SN75176 provided it has the extra resistors.

Hey Guys,

Thanks for the great info here. We'll get this implemented ASAP and if it doesn't get into the next Gen3 run, it will definitely make it into the Gen4. From what I've read here, the best suggested fix is to switch to a better chip?

As the designer of the Gen3 stuff, I apologize for any mistakes. I am not an EE by any means. We just wanted to get a cheap 3D printer out there so people could have access to it. On the plus side, we're hiring an EE this week and combined with the excellent community support shown here then next versions should be rock solid.

Cheers,
Zach