It's common to add resistors in series at outputs in order to reduce crosstalk between lines that run parallel to one another in a cable, connector, or traces on a board. Crosstalk is proportional to the rate of change of voltage and current in the parallel lines and does not necessarily have anything to do with the transmission line model. It can be a killer in devices that use clocked logic. In Domanik's simulation, it could be extended to include capacitance and mutual inductance between the signal lines. Resistors slow down the that critical slew rate in addition the damping of transmission line reflections. I'd definitely not be tempted to omit those resistors, and would tend to think about increasing them toward the 60Ω value. Brett probably had a good practical reason for including them.
Ive worked with a variety of twisted pair and the typical value is about 100 ohms +/- 10%. To get a lower value requires alternate grounds between each signal and much closer conductors (ie thinner insulation). I assumed it was probably not lower and close enough for a rough model.
I dug out my original notes from a few months back and made some minor changes to the simulation. Source impedance = 35 ohms, rise time = 3ns instead of 2ns. Originally I found some info in the specs that gave me a hint at the value (might have been the output short circuit current Ios). To get closer the prop output was measured with a quality 10ns of RG58 50 ohm coax. I toggled a prop output driving the coax and measured, with a scope, the magnitude of the notch in the returning reflection, then trial and error with various resistors until the reflection matched the source at 50%. 50 ohms minus the 17 ohm resistor equals the prop source impedance (35 is a good enough approximate number). Also I measured the rise time of the prop at about 1 volt/nanosecond at the source.
In tinkering with the simulation Ive found a 140 ohm in series most of the way to the end of the line has some very nice characteristics, as seen in LCD4. Not for your situation though. The 60 ohms should work okay. I've adjusted several of the simulation values looking for problems and the circuit seems to be very forgiving.
As Heater mentioned going from 3.3V to 5V might be okay BUT take a look at the undershoot on the high-low transition. The -1.5V undershoot might be enough to damage the input transistors. This is something I didn't realize until now.
So if the impedance of a twisted pair cable is around 100 ohms, what is the impedance of PCB tracks spaced 0.1" apart like in the LameStation?
93.7 ohms ... Seriously though, the impedance is probably between 85 and 140 ohms. Qualify that with the number of variables that affect the result: 1) trace width 2) dielectric constant of the FR4 3) distance from adjacent traces 3) distance from ground/power plane 4) inner or outer PCB layer 5) sandwiched between planes? 6) speed of light in your neighborhood. It's all relative. There are lots of resources on the internet with various formulas and online calculators to help you come up with an exact answer.
The KS0108 LCD latches the 8-bit address and display data on the falling edge of the E signal line (Enable). E is next to the P0 data signal on the connector. I'd be wary of crosstalk between P0 and E, that a falling edge on P0 could cause a glitch on E and latch false data. However, I'm also a bit confused by the LCD specs. For example the setup and hold times in the LCD AC specifications are more on the order of 200ns, quite a bit slower than the 50ns implied by a 20MHz clock rate. Does the Lamestation really toggle that pin that fast? The reason given for the fast update rate is that it needs to create the gray color by toggling the affected pixels rapidly between black and white.
The KS0108 LCD latches the 8-bit address and display data on the falling edge of the E signal line (Enable). E is next to the P0 data signal on the connector. I'd be wary of crosstalk between P0 and E, that a falling edge on P0 could cause a glitch on E and latch false data. However, I'm also a bit confused by the LCD specs. For example the setup and hold times in the LCD AC specifications are more on the order of 200ns, quite a bit slower than the 50ns implied by a 20MHz clock rate. Does the Lamestation really toggle that pin that fast? The reason given for the fast update rate is that it needs to create the gray color by toggling the affected pixels rapidly between black and white.
The Samsung datasheet for the KS0108 processor has a max clock of 20 u sec and a min of 2.5usec. Maybe a mix up in terms.
William, a more specific title might attract the attention of the Lamestation developer. Your question seems to be more about that than about the resistors per se.
There seem to be several different clocks for the KS0108 controller. The one for the MPU interface is basically 1MHz or less for the E latch signal.
I'm not convinced that reflections and transmission lines have anything to do with it. The Lamestation display plugs right into the motherboard, and the whole connection path can't be much more than an inch. The LCD is relatively slow and any reflections should have time to die out before the enable latch line makes its high to low transition. EMI suppression yes, resistors could help with that and maybe that's all there is to it.
You are the first person that dares to say that the damping resistors are not required on LameStation.
This is what I was waiting for, so I can get rid of 12 resistors from my PCB.
The costs of soldering 12 extra resistors is prohibitive...
but wait, Domanik says 60ohm resistors are required.... oh no.
You are the first person that dares to say that the damping resistors are not required on LameStation.
This is what I was waiting for, so I can get rid of 12 resistors from my PCB.
The costs of soldering 12 extra resistors is prohibitive...
but wait, Domanik says 60ohm resistors are required.... oh no.
The costs of soldering 12 extra resistors is prohibitive....
...seriously? A dollar twenty is too much?
I've been watching this thread with some interest. Not so much for technical information, but for the debug approach.
I've seen theorizing, modeling, datasheet perusal, and a bit of "welllll....(fill in the blank)". Unless I missed the memo, nobody actually o'scoped out the signal lines with and without the resistors to observe any effect.
Ummm... aside from Tracy's info, I believe Domanik mentioned that you could use resistors to reduce ringing OR by careful construction of the board with grounding between each lead (see #22 "add more ground wires to the cables") -- you might eliminate ringing without the resistors. (I really am not 100% sure).
That would certainly eliminate the cost element of 12 resistors. But one really has to face the fact that design of new board is trial and error. We can speculate endlessly, but one just might have to build, fail, and build again to get the right board.
Please don't come back and say I misguided you! It would be an experiment on your part, a departure from a design that is known to work with that LCD. The main reservation I have concerns the proximity of the data0 signal to the E latch signal, but with a short connection path even that should not an issue. What are you going to do if ghosts appear? You're making a prototype, right?
A dollar twenty? More like few cents tops for the 12 resistors and a bit more for the soldering. What is the prohibitive cost? Through-hole or SMT?
My curiosity got the best of me so I simulated 3 possibilities: 1) no resistor 2) 15 ohms 3) 60 ohms
The PropOut 1, 2, 3 are: 40 ohm source impedance, 2ns rise/fall time, 23ns pulse width, 50ns cycle (20MHz), 0V to 3V.
The transmission line is 110 ohm, 0.8ns (6"), non-lossy.
Parasitics added: capacitance 5pf, .04uh inductance, .5 ohm series resistance.
LCD input pin of 6pf (estimated). Don't know if there's any termination resistance- 300 ohm kills most ringing & overshoot.
In #3, 60 ohm resistor, a close up look at node "A" shows the notch due to the reflection half way up the rising edge.
I'm guessing, in the original design, 15 ohms was added because it made it work.
Nice simulation, but your estimate for the Prop's rise time is too slow. I've measured it a few times with the 1GHz scope at work and they have a ~0.8ns rise-time when loaded by only the probe or a 50ohm transmission line. (even 74LVC at 5v is ~1.1ns, and 74AC is 1-2ns)
I've personally seen some ringing when driving 6" traces on a PCB. Wasn't enough to cause problems with digital circuits, but did screw up an analog circuit. So, 6" or longer traces is when I'd consider a trace/wire "long"
Nice simulation, but your estimate for the Prop's rise time is too slow. I've measured it a few times with the 1GHz scope at work and they have a ~0.8ns rise-time when loaded by only the probe or a 50ohm transmission line. (even 74LVC at 5v is ~1.1ns, and 74AC is 1-2ns)
I've personally seen some ringing when driving 6" traces on a PCB. Wasn't enough to cause problems with digital circuits, but did screw up an analog circuit. So, 6" or longer traces is when I'd consider a trace/wire "long"
Marty
Hi Marty, Thanks for the excellent feedback. I used a 500MHz scope, passive probe ~10pf, 10meg so it smooths out the high end ringy-dingies. I assume your 0.8ns rise time was measured at the prop pin and 20% to 80% without a circuit trace attached? Funny things is I'm working with some 74LVC so the 1.1ns RT is interesting too; didn't know it was that quick. This thread has prompted me to look closer at some of my previous assumptions about matching impedance's. Thanks again.
Ahaha, I never did see this topic all those months ago.
If anyone out there still cares (probably not), the damping resistors allowed us to seriously speed up the LameStation's display driver and made it much more stable.
Hi Brett
Although I didn't participated from the original debate, I'm still curious about what was the final result of it.
Then you comes and says that the resistors improved driver's stability.
There comes my question about it:
Was the observed improvement the allowance of sending faster data to the display, i.e., it has improved the relationship between outputing new data and toggling E, to latch it at the display?
As for the data that was being sent, was it totaly based on a set of values, internaly stored at Propeller's memory, or, at least partly, it relies on the actual (external) data bus contents, whose value in turn, could be affected by any ringing present at the data lines?
Sorry for any mistyping but I'm using an Android device and its difficult to manager typing and correction in parallel.
The resistors limit the slew rate of the signal as otherwise the fast edges can cause crosstalk as well as EMI although I don't believe that Prop I/O signals can operate at that high a frequency that ringing could cause a problem interfering with the signal itself that would require "damping" or series termination resistors.
Alternatively if the LCD was 5V and was meant to read back data then having resistors designed in just means you can decide later whether to make them a higher value for current limit or shorted using a convenient low value resistor. If I design in a "resistor" although it is only the copper pattern, I always have this option otherwise it could mean a pcb redesign.
None of these resistors will help if the LCD logic high input threshold was too high for 3.3V although in my experience many LCDs are TTL threshold so that >2V = logic high.
@Domanik: well done with the sims!
EDIT: man, this is a long thread, I see there are some good answers such as Tracy's but did I hear correctly that the cost of resistors was "prohibitive"!!!!!??????? whaaaat?
A reel of 5,000 resnets (4in1) might cost $20, so that's 20,000 resistors for $20 or ten for a cent.
But looking now at the schematic and type of LCD I know for sure that you don't need these resistors. The R/W line is tied low so you only ever write to it so there is no need for current limit resistor either. These LCDs are SLOW and the inputs are TTL compatible too.
Comments
Thank you for your spice work.
How do you estimate the transmission impedance to be 110 ohms and
the Propeller source impedance to be 40 ohms?
I dug out my original notes from a few months back and made some minor changes to the simulation. Source impedance = 35 ohms, rise time = 3ns instead of 2ns. Originally I found some info in the specs that gave me a hint at the value (might have been the output short circuit current Ios). To get closer the prop output was measured with a quality 10ns of RG58 50 ohm coax. I toggled a prop output driving the coax and measured, with a scope, the magnitude of the notch in the returning reflection, then trial and error with various resistors until the reflection matched the source at 50%. 50 ohms minus the 17 ohm resistor equals the prop source impedance (35 is a good enough approximate number). Also I measured the rise time of the prop at about 1 volt/nanosecond at the source.
In tinkering with the simulation Ive found a 140 ohm in series most of the way to the end of the line has some very nice characteristics, as seen in LCD4. Not for your situation though. The 60 ohms should work okay. I've adjusted several of the simulation values looking for problems and the circuit seems to be very forgiving.
As Heater mentioned going from 3.3V to 5V might be okay BUT take a look at the undershoot on the high-low transition. The -1.5V undershoot might be enough to damage the input transistors. This is something I didn't realize until now.
There seem to be several different clocks for the KS0108 controller. The one for the MPU interface is basically 1MHz or less for the E latch signal.
I'm not convinced that reflections and transmission lines have anything to do with it. The Lamestation display plugs right into the motherboard, and the whole connection path can't be much more than an inch. The LCD is relatively slow and any reflections should have time to die out before the enable latch line makes its high to low transition. EMI suppression yes, resistors could help with that and maybe that's all there is to it.
You are the first person that dares to say that the damping resistors are not required on LameStation.
This is what I was waiting for, so I can get rid of 12 resistors from my PCB.
The costs of soldering 12 extra resistors is prohibitive...
but wait, Domanik says 60ohm resistors are required.... oh no.
...seriously? A dollar twenty is too much?
I've been watching this thread with some interest. Not so much for technical information, but for the debug approach.
I've seen theorizing, modeling, datasheet perusal, and a bit of "welllll....(fill in the blank)". Unless I missed the memo, nobody actually o'scoped out the signal lines with and without the resistors to observe any effect.
Again, seriously?
That would certainly eliminate the cost element of 12 resistors. But one really has to face the fact that design of new board is trial and error. We can speculate endlessly, but one just might have to build, fail, and build again to get the right board.
A dollar twenty? More like few cents tops for the 12 resistors and a bit more for the soldering. What is the prohibitive cost? Through-hole or SMT?
Nice simulation, but your estimate for the Prop's rise time is too slow. I've measured it a few times with the 1GHz scope at work and they have a ~0.8ns rise-time when loaded by only the probe or a 50ohm transmission line. (even 74LVC at 5v is ~1.1ns, and 74AC is 1-2ns)
I've personally seen some ringing when driving 6" traces on a PCB. Wasn't enough to cause problems with digital circuits, but did screw up an analog circuit. So, 6" or longer traces is when I'd consider a trace/wire "long"
Marty
If anyone out there still cares (probably not), the damping resistors allowed us to seriously speed up the LameStation's display driver and made it much more stable.
Although I didn't participated from the original debate, I'm still curious about what was the final result of it.
Then you comes and says that the resistors improved driver's stability.
There comes my question about it:
Was the observed improvement the allowance of sending faster data to the display, i.e., it has improved the relationship between outputing new data and toggling E, to latch it at the display?
As for the data that was being sent, was it totaly based on a set of values, internaly stored at Propeller's memory, or, at least partly, it relies on the actual (external) data bus contents, whose value in turn, could be affected by any ringing present at the data lines?
Sorry for any mistyping but I'm using an Android device and its difficult to manager typing and correction in parallel.
Henrique
Alternatively if the LCD was 5V and was meant to read back data then having resistors designed in just means you can decide later whether to make them a higher value for current limit or shorted using a convenient low value resistor. If I design in a "resistor" although it is only the copper pattern, I always have this option otherwise it could mean a pcb redesign.
None of these resistors will help if the LCD logic high input threshold was too high for 3.3V although in my experience many LCDs are TTL threshold so that >2V = logic high.
@Domanik: well done with the sims!
EDIT: man, this is a long thread, I see there are some good answers such as Tracy's but did I hear correctly that the cost of resistors was "prohibitive"!!!!!??????? whaaaat?
A reel of 5,000 resnets (4in1) might cost $20, so that's 20,000 resistors for $20 or ten for a cent.
But looking now at the schematic and type of LCD I know for sure that you don't need these resistors. The R/W line is tied low so you only ever write to it so there is no need for current limit resistor either. These LCDs are SLOW and the inputs are TTL compatible too.