[pminmo]

External drivers for power fet's.
I was looking to this thread for long time and some opinions for miscrostepping failure where due to gate capacity.

Unfortunately the gate capacitance theory had nothing to do with the problem with the actual part. It's a verified chip issue by the manufacturer when using motors with faster LR motor constants. Problem is the motors that cnc people would be using have those time constants.

[krazatchu]

Hey ppl....

I have a friend who is considering using the A3967 part....
Does anyone know if this part suffers from the same problems as the A3986 ??

Thanks!
Michael

[tiwanacote]

.

[scubasteve_911]

I'm designing a pcb for a 3-axis controller and interface to the parallel port so that it can be used with mach3. I will post schematics and gerber files when I am finished. I may have 3-4 spares because I am sending it out with my work's pcbs this week.

Steve

[MechanoMan]

Unfortunately the gate capacitance theory had nothing to do with the problem with the actual part. It's a verified chip issue by the manufacturer when using motors with faster LR motor constants. Problem is the motors that cnc people would be using have those time constants.

Actually it has nothing to do with the LR constant. It's simply motor R and input voltage.

For a nonmoving servo:
min Iout = (min duty * Vin)/R

A motor designed for 3.3V at 2.8A has a 1.179 ohm coil.

If the blanking interval creates a 5% minimum duty and you have a 48V supply, it can create an output of 2.036A @ 2.4v. So the fullstep is totally achievable. But even 4 microsteps from 0 to 90 deg would require a 38% point which would be 1.07A. The driver can't do that. 5% will produce 2.036A and it can't physically go lower, therefore the smaller microsteps will fail to resolve. Any pulses asking for under 2.036A will get 2.036A. The problem will lessen with reduced Vin. The hypothetical 5% min duty would work with 4 microsteps from 0 to 90 deg if the source voltage were 24v or below.

krazatchu, look up the min duty on the A3967. I recall looking up the other Allegro parts and there was one with programmable blanking interval, which sounded perfect except the min blanking interval was STILL too high to be useful for even 1/8th microstepping.

[pminmo]

Actually it has nothing to do with the LR constant. It's simply motor R and input voltage.

For a nonmoving servo:
min Iout = (min duty * Vin)/R

A motor designed for 3.3V at 2.8A has a 1.179 ohm coil.

If the blanking interval creates a 5% minimum duty and you have a 48V supply, it can create an output of 2.036A @ 2.4v. So the fullstep is totally achievable. But even 4 microsteps from 0 to 90 deg would require a 38% point which would be 1.07A. The driver can't do that. 5% will produce 2.036A and it can't physically go lower, therefore the smaller microsteps will fail to resolve. Any pulses asking for under 2.036A will get 2.036A. The problem will lessen with reduced Vin. The hypothetical 5% min duty would work with 4 microsteps from 0 to 90 deg if the source voltage were 24v or below.

Off the top of my head I'd like to politely disagree. Your analysis is for a completely resistive device. Inductor currents don't change instantaneously. A motor coil is an LR component. The delta change in current depends on the LR time constant of the device.

[NickSP]

Hello All,

I just started a week ago trying to design a driver for a high torque bipolar stepper. I was considering the A3986 initially, but as I was reading through this thread saw the problem this chip has, or at least most of you have experienced.

Anyone has had a better luck with this chip? Any other chip out there similar to the functionality of the A3986? My bipolar stepper motor specs 4.9A


thanks,
NickSP

[eSilviu]

Hello,
I’ve build a driver based on A3986. My motors are FL57STH76 (nema 23, bipolar parallel: 0.5ohm, 1.6mH).
The best I've get from this chip is:
- microstep16, minor degree error (as mentioned in all this thread) - but this error means nothing comparing with mechanical backlash and flexible coupling angle error;
- 30V supply voltage
- 3A motor current (even at 3A, motor can not run 1 hour without getting very hot, so the "4A" specified in datasheet are not for continuous work) Ref input of the 3986 can (and will) be used to lower the motor current at low speeds or when motor it's not spinning (kind of "enable" that will work for microstep too)
- 300kustep/sec (that's about 1800fullstep/sec or 5600rpm) with no load on motor. 100kustep/sec with light load, or 70kustep/sec with motor connected on cnc. Ramping is necessary if you want to spin motor faster than 12kustep/sec.
- motor current has the correct shape (as in a3986's datasheet) for:
15kustep/sec for 15V supply voltage
20kustep/sec for 20V supply
40kustep/sec for 30V supply
This depends on back-emf voltage from the motor (inductance of the coils).
- very low noise from the motor (you can't hear the PWN noise)
(driver board is 4/6cm; without filtering cap, no optical isolation, all SMD components)

Has anyone obtained better results from this chip?

[originator]

I built 3 machines with the A3986 and I would have not know there were any problem except for this thread speaking about it, since I never bother to check with the tests other than making parts. I use a rebuilt bench top the most at 1/16, and the play in the machine plus backlash is greater than any error this chip will ever produce. I suppose for those that have real tolerance issues maybe there should be such concern. The only thing I'd complain about is mosfet heat, which a small fan takes care of. I can run the machines all day with 4 amp motors, IRF540's. I suggest build a driver and run it on a part then decide.

[pminmo]

I posted long ago that the accuracy isn't cumulative, and for some machines (i,e, wood routers) the absolute accuracy probably isn't an issue, it would depend on transformation method of rotary to linear motion. On precision machines, this isn't the part to use.

The other issue is the inaccuracy also compounds resonance tendancies, so it becomes more problematic that a precentage of motor/power supply/mechanics won't play well together.

[pminmo]

- 3A motor current (even at 3A, motor can not run 1 hour without getting very hot, so the "4A" specified in datasheet are not for continuous work)

The chip has poor current regultion, which most of this thread discussion was focused on the innacuracy. There are multiple possible ramifications of poor current regulation, one could be excess motor heating.

[eSilviu]

The chip has poor current regultion, which most of this thread discussion was focused on the innacuracy. There are multiple possible ramifications of poor current regulation, one could be excess motor heating.

I'm sure this is not the problem. I can see (and measure) current with an oscilloscope, and it is as it should be.
Maybe heat is result of the PWM frequency (I use 10k for Rosc = maximum driver frequency). At about 330kustep/sec, motor has a e.m. resonant point: in about 5 sec will get verry hot, and I can feel a strange sensation in my hand, if I hold my hand on it (ultrasounds, for sure!).
But in practice I will never achive 330kusep/sec.

One more thing about precision:
I use 3mm trapezoidal thread: = 200full steps/rev = 3200ustep/rev
this means 15um/step or 0.9um/ustep16.
I really don't care if the motor introduce in system an error of 1/2 step = 8um Even tool play is larger than this (at least in my case). So I think driver is usable even for a mill (until a better driver will be produced)

[originator]

When talking about error, you have to consider the cumulative effect per step/micostep, including the machine components friction errors which has non cumulative error in backlash, drive train binding, inherent motor flaws, coupler binding, deflection, etc.

I think a fair assumption is that any machine/drive chain has a tendency towards an inherent non cumulative error which corrects it self at some rate as the drive train cyclically accumulates the error, with a bias towards coming up short of the desired position then when the bind reaches a certain point, the error corrects or even overshoots briefly.

Inducing a random error per step/microstep with a bias towards overshoot (as the 3986 does contribute) may actually have an effect of reducing the machines errors, although this may appear ridiculous at first glance. The linearity of energy transfer from motor>drive chain>tool is really just an average of errors on all machines. For the average user this chip is a very convenient and practical driver. I use it on aluminum and Delrin all the time.

[jalessi]

Originator,

Please post some pictures of the parts you are producing.

Jeff...

[pminmo]

microstep16, minor degree error (as mentioned in all this thread)

I'm sure this is not the problem. I can see (and measure) current with an oscilloscope, and it is as it should be.

If you have angle error, you have poor current regulation.

[eSilviu]

If you have angle error, you have poor current regulation.

The problem is more subtile than that. There are many factors involved in angle error (supply voltage, motor load, noise on current sense path, sync. rectification etc. ...)
Maybe all this problems can be solven, but this require time. But performance/design effort is verry good as it is now, so I think I will not try to improve my design ("no time, to many projects")

[jimmy4832]

Hello All,

I just started trying to design a driver for a 10A high torque bipolar stepper. I was considering the A3986 initially.
How to chioce the PowerMOS ?


thanks,
Jimmy

[Zentarus]

Hello,

I have finished mine, and it's working very well. I have never seen a stepper controller running like that, very impressive !
Ultra high speed (Step 800 KHz), very high current ( 8 Amps ), no heat (CMS MOS can be used).
Olso, I have no problem of microstepping precision, even at very low speed.
I have noticed that most of the design seen on the web could not run properly. So i have made mine, and i can confirm that the chip do an excellant job for microstepping.

[dandumit]

maybe you can give us some more details about board layout and component that you have used.

[Zentarus]

maybe you can give us some more details about board layout and component that you have used.

I have used Motorola transistor MTB75N06 with Rg = 47 Ohms, CBoot = 56nF LCC, Creg = 470nF, Rosc = 10K
I have used a RC ( 500 Ohms and 1 nF COG ) before inputing in the sense pin.

But the most important, an excellant grounding point done on the negative pole of a 1000uF/50V low ESR capacitor, ( like all the desin on the web ) and also, the sense layout MUST BE CONNECTED DIRECTLY TO THE RSENSE ! ( And this point everybody miss this and connect it on the source of the first mosfet meet and this is a big mistake ).