EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: charon1985 on May 14, 2016, 12:56:24 pm
-
Hi, i have made a IGBT/Mosfet driver circuit on protoboard with 1x TL494, 2x UCC37321p (9 Amp each), and a home made GDT that has 1 primary and 4 secondary winding. I plan to use this circuit to drive 4x IGBT IRGP50B60PD1-EP (50 Amp each, 600V, 150KHZ maximum). Gate voltage is 15V. 7 Ohm gate resistor, below 7 it starts ringing and appears strange spikes near the dead time.
I want to work with my circuit in the range of 65-120Khz.
10Amp at 310V (Colector - Emitor)
My question is how to improve the turn off time of the IGBT? The turn on is 180-200ns but the turn off is 800ns (on oscilloscope)
The PDF say for this IGBT:
Turn-On delay time 40 ns, Rise time 15 ns, Turn-Off delay time 150 ns, Fall time 15 ns
Or can i use it like it is now? i have 2x MAX4420 (6amp) and i thought to put them in paralel with the UCC37321P, is this a good idea?
Thank you for your help.
-
It's hard to understand your circuit from a textual description, can you provide a schematic?
-
Turn on, turn off time typically speced at 10% and 90% of final values,
your cursors do not appear to be there. Also the spec sheet is done at a
gate R of 3 ohms, yours 7, that will aggravate the performance. It may
be layout that caused you to go to 7 ohms, eg. parasitic L causing the
ringing.
Have you looked at the sim tools ?
http://www.infineon.com/cms/en/tools/landing/igbt.html (http://www.infineon.com/cms/en/tools/landing/igbt.html)
Regards, Dana.
-
Thank you Dana for your suggestion regarding sim tools, i did not know that they exist.
Regarding the parasitic inductance i was suspecting it to be the cause of ringing, but i thought to ask and see what other say. The circuit obviously is not optimized for high frequency...
But how about putting in parallel the two drivers, max4420 and ucc37321 ?
-
Post a schematic.
Also, IGBTs love having their gate pulled negative to speed up turnoff and reduce the chances of dv/dt-induced turn-on.
-
At what operating frequency?
What are the small signal parameters of the GDT?
Tim
-
Some AP Notes on driving -
MOSFET Ap Notes
http://www.infineon.com/dgdl/an-937.pdf?fileId=5546d462533600a40153559ea1481181 (http://www.infineon.com/dgdl/an-937.pdf?fileId=5546d462533600a40153559ea1481181)
http://www2.st.com/content/ccc/resource/technical/document/application_note/68/cd/c6/ab/ef/17/41/06/CD00003900.pdf/files/CD00003900.pdf/jcr:content/translations/en.CD00003900.pdf (http://www2.st.com/content/ccc/resource/technical/document/application_note/68/cd/c6/ab/ef/17/41/06/CD00003900.pdf/files/CD00003900.pdf/jcr:content/translations/en.CD00003900.pdf)
http://www.radio-sensors.se/download/gate-driver2.pdf (http://www.radio-sensors.se/download/gate-driver2.pdf)
http://www.ixys.com/documents/appnotes/ixan0010.pdf (http://www.ixys.com/documents/appnotes/ixan0010.pdf)
https://www.fairchildsemi.com/application-notes/AN/AN-6069.pdf (https://www.fairchildsemi.com/application-notes/AN/AN-6069.pdf)
http://www.ti.com/lit/an/slua054/slua054.pdf (http://www.ti.com/lit/an/slua054/slua054.pdf)
http://ww1.microchip.com/downloads/en/AppNotes/00898a.pdf (http://ww1.microchip.com/downloads/en/AppNotes/00898a.pdf)
http://www.onsemi.com/PowerSolutions/supportDoc.do?type=appNotes&category=809 (http://www.onsemi.com/PowerSolutions/supportDoc.do?type=appNotes&category=809)
Regards, Dana.
-
ovnr, the gate is puled negative (-15volt) for 50%time - dead time, the turn off slope that you see in the screenshots is going into the dead time. What to do with it?
T3sl4co1l, the frequency will be in the range of 65-120Khz, one screenshot is made at 62khz and the other is made at 100khz...
Thank you Dana.
I will reorganize the parts and connections on the protoboard to reduce parasitic induction.
-
I will reorganize the parts and connections on the protoboard to reduce parasitic induction.
I suspect that's contributing heavily to the problem.
Try making it on a PCB with nice short traces for the gate and source connections.
-
What are the small signal parameters of the GDT?
That's a bigass IGBT, past the point where it's reasonable to use a GDT. Maybe if you were doing 20kHz, but then the transformer would be huge anyway, and you're still better off with an isolated gate driver.
It should also have a negative gate voltage in the off state. The datasheet doesn't say so, but it is crucial for hard switching applications.
Tim
-
The small signal parameters of the IGBT operated in a non linear, e. switching environment, probably not
of much use ?
Regards, Dana.
-
The small signal parameters of the IGBT operated in a non linear, e. switching environment, probably not
of much use ?
Regards, Dana.
GDT, the Gate Drive Transformer.
Tim
-
What are the small signal parameters of the GDT?
That's a bigass IGBT, past the point where it's reasonable to use a GDT. Maybe if you were doing 20kHz, but then the transformer would be huge anyway, and you're still better off with an isolated gate driver.
It should also have a negative gate voltage in the off state. The datasheet doesn't say so, but it is crucial for hard switching applications.
AFAIK the trick is to discharge the capacitor between the base and emitter which is a bit more difficult to do with an IGBT than with normal transistors.
-
AFAIK the trick is to discharge the capacitor between the base and emitter which is a bit more difficult to do with an IGBT than with normal transistors.
Minority carriers are taken care of internally. Since as you say, obviously, you can't discharge that from an external pin. Methods include shorting it out with a resistor (Shorted Anode, I think?), varying bipolar hFE, and adding doping to control carrier lifetime (Pt/Au doping, electron irradiation). All of which increase Vce(sat) as a tradeoff for shorter turn-off times, hence why IGBTs are available in multiple speed grades for different applications.
With a gate charge of 300nC (max at 15V), you have an equivalent 20nF gate capacitance. Driving that is hard enough.
Tim
-
T3sl4co1l, the GDT parameters are:
- 5.31 mH inductance, i know i could go lower, but the resonant frequency of my installation will be around 65Khz.
- aprox 51 pF capacitance
- aprox 3.2 uH leakage inductance
- 0.6 cm2 for the core
- and 31 turns
The biggest Mosfet i ever seen operating at 300V were almost the size of my fist, they were working at 20-120khz and they were GDT driven. My igbt are musch smaller than those monsters. They were used in a semi-industrial machine, factory made.
-
So the equivalent circuit between driver and gates (4 in parallel) is 3.2uH in series, some ohms resistance, and about 20nF per gate.
The critically damped resistance is sqrt(L/C) or sqrt(3.2uH/0.08uF) = 6.3 ohms.
You should have more than 6.3 ohms in series with the drivers, but better is to have more than 25 ohms in series with each gate (which has a parallel equivalent of 6.3 ohms).
Observing ringing below 7 ohms is exactly consistent with all the component values. Good!
Now suppose you use 47 ohms per gate.
The RC time constant at the gate is 0.94us, so gate switching will take almost 2us. Using the 1/20th rule, you won't be able to operate above 25kHz, and that's probably with generous derating.
You can't do any better than half this, and even if you try, you get bad ringing. (You could try adding more LC poles to make a higher order network with a risetime of ~1/4 this, but that makes things more complicated, harder to design, and less tolerant of component variation.)
Does this make the point, that GDT isolation is absurd? ;)
I've seen plenty of badly designed industrial circuitry, so that's not a favorable comparison to make...
Also, having a series total of over 7 ohms equivalent at the driver, means from 15V, you don't need any more than (15V)/(7 ohms) = 2 amps of peak driver capacity. Why would you need three ~6A drivers in parallel?
Run the numbers. Basic Ohm's law, it doesn't lie. 8) 8)
Tim
-
There are not 4 transistors in parallel, is a H bridge, only 2 are open at any moment, the two UCC37321 are not in parallel, they are each at one end of the primary winding. I will post a screenshot with the output of the GDT to understand better. The two screenshots that i have posted are obviously from one IGBT transistor to see the switching performance...
I have tested the performance of other IGBT for example (on the same driving circuit and same GDT ), IRG7PH35UD1PBF 1200v and 50A and the turn off is 1800ns, which has a gate capacitance almost two times lower than 50b60.
I also tested some Mosfets (comparable gate capacitance) on the same circuit and GDT and the turn off are around 140-160-180ns
Although i think i should try with a higher load on the transistor because on the data sheet the switching speed is given for 33Amps and 390 Volt conditions. My test conditions are 30V and 1 Amp, because i don't have a bigger power supply.
-
Where do you get ringing under 7 Ohms? I would guess on the rising edge. If it's only on the rising edge you could put a fast diode anti-parrallel over 7 Ohm resistor. That should not affect the rising edge but would help discharge the gate capacitance. I've used this trick with IGBT driven solidstate teslacoils with good results. H-bridge around 100 KHz...
-
Thanky you Arjan, i will try that.
I think that parasitic inductance and the Figure 12 on the datasheet explains my problems
http://www.tme.eu/ro/Document/67a94254cd88110237c678f0ca0fdc77/irgp50b60pd1-ep.pdf (http://www.tme.eu/ro/Document/67a94254cd88110237c678f0ca0fdc77/irgp50b60pd1-ep.pdf)
In fact i have measured the consumed current by the load (a light bulb) and is 0.3 amp, i think that here is my problem, if i interpreted correctly the figure 12 on the datasheet.
-
Yeah, you need to minimize any parasitic inductance. A big source can be the leakage inductance of the GDT. Take great care in designing that transformer. The more turns/inductance the more leakage inductance, so keep that a minimum without saturation problems. I always got away with 10 turns on a 3e25 ferroxcube core of 0,6 cm2. Core material is important !
Twist the 5 wires together before winding the transformer. This significantly reduces leakage inductance. And ofcourse keep all the leads as short as possible.
Edit, i use TN29/19/15-3E25 , thats 74mm2
-
There are not 4 transistors in parallel, is a H bridge, only 2 are open at any moment, the two UCC37321 are not in parallel, they are each at one end of the primary winding.
Ok so the drivers are in series; still, you only need 2A capacity, perhaps a 4A capacity dual driver since output resistance will matter.
The transistors act in parallel, by transformer action. If you aren't seeing this, please look up how transformers operate!
Yeah, you need to minimize any parasitic inductance. A big source can be the leakage inductance of the GDT. Take great care in designing that transformer. The more turns/inductance the more leakage inductance, so keep that a minimum without saturation problems. I always got away with 10 turns on a 3e25 ferroxcube core of 0,6 cm2. Core material is important !
Twist the 5 wires together before winding the transformer. This significantly reduces leakage inductance. And ofcourse keep all the leads as short as possible.
Edit, i use TN29/19/15-3E25 , thats 74mm2
This helps, but not as much as is needed.
I can estimate the performance of such a transformer, since you've provided dimensions. :)
The circumference (wire length per turn) is around 34mm, so 10 turns gives 0.34m wire length.
The wire construction is multifilar twist, a variation on twisted pair transmission line. The characteristic impedance between pairs of wires will be on the order of 100 ohms -- probably higher (120-150?) due to the odd pairings and extra space between wires. (An improvement would be to twist each secondary with its own primary strand, and connect all the primary strands in parallel. Thus you use eight strands total. Pairs of these could further be married, by making quadrafilar sets, where the wires are arranged in opposing order: P, S, P, S. This leaves two pairs of quad twist, which can't really be paired up in any useful bundles; they can simply be loosely twisted or tied together to complete the winding.)
This is an early warning sign, because as we've established earlier, the gate circuit impedance is on the order of single ohms. The circuit impedance is below the transmission line impedance. Also, the transmission line electrical length, of about 1.7ns, means you don't have to worry about transmission line effects from the much slower (>10ns) drivers. Thus, we model the transmission line as an inductance, which is connected in series between the drivers and the gates.
The inductance of the line is about 1.7ns * 150 ohms = 250nH, give or take. This is per pair of strands, so acts in series with each gate.
Evidently, it would do better than OP's 3uH. Let's see how well.
The critical impedance is sqrt(250nH/20nF) = 3.5 ohms (per gate). All four in parallel would give less than 1 ohm (which would be well deserving of a >= 10A driver from 15V), but if the primary is only a single strand as well, it probably won't be this low.
We should expect an LC time constant (or similar RC time constant if critically damped) of around pi*sqrt(L*C)/2 or 111ns, so a gate risetime in the 200ns range would be possible.
For simple things like SSTCs, such a transformer would work okay.
A 29mm toroid is pretty big for a GDT, though -- building one from scratch will cost more in parts than HVIC type gate drive ICs, let alone in production. (Assuming such drivers are suitable, which again, they really aren't.)
The best solution really is isolated gate drivers, with +15/-5V drive outputs (or better).
Tim
-
Hi tim.
I'd be really interrested if you know some suitable partnumbers for isolated gate drivers. I played around with this topology 10 years ago and back then the propagation delay for those devices was just too long. I'm sure we have evolved since that time.
Basicly we had to drive the primairy of an drrsct 90 degrees out of fase of the curent flowing through the primairy.
The easy way was measuring the current with a current transformer. Same 2e25 core. buffer the signal (74hc) and put it through a ucc37221/37222 pair into the gdt. perfectly self tunning, 90 degrees out of fase and very little propagation delay.
Optocouplers we're too slow back then.
-
A fairly good description of the issues -
http://www.ixys.com/documents/appnotes/ixys_igbt_basic_i.pdf (http://www.ixys.com/documents/appnotes/ixys_igbt_basic_i.pdf)
https://www.fairchildsemi.com/application-notes/AN/AN-9016.pdf (https://www.fairchildsemi.com/application-notes/AN/AN-9016.pdf)
https://www.fairchildsemi.com/application-notes/AN/AN-9020.pdf (https://www.fairchildsemi.com/application-notes/AN/AN-9020.pdf)
Regards, Dana.
-
Turn on, turn off time typically speced at 10% and 90% of final values,
For IGBTs this can be different. For this particular one they define it as 90%–5% but it can also be 90%–20% due to the current tail.
The slow turn off is probably due to high Lg and Rg. Measure the gate of the transistor (on transistor legs) and you will know.
Another thing that can impact a turn-off process in this way is a common inductance in gate and power path. Make sure you decouple these two lops. Solder the IGBT as close to the PCB as possible.
-
I have have made the shortest connection possible on the protoboard, but the legs of the IGBT and the Gate resistors were not cut off, the gate resistor is 11 ohms (7ohm is even better but i think is not necessary), and here are the results:
First Screenshot, Turn On 120 ns -> improvement 80ns (200-220 previous)
Second Screenshot, Turn Off 720 ns, Load 0.15 A, -> improvement 100ns (820-840 previous)
Third Screenshot, Turn Off 400ns, Load 0.5A, -> improvement 420ns
I did not made a screenshot at 90-10% but Turn Off is 260ns, Load 0.5A
Minimum load in the final version will be around 2 amps, so its ok, it will go below 260ns (90-10%) at 2 amps...
Further improvement will be made in the Turn On and maybe Turn Off when the circuit is on a PCB board...
Thank you every one for the help
Thank you Dana for all the Help and Hints, by the way my real name is Dan.
Mosfets are turned on in 120nS and turned off in 180ns (approximately same gate capacitance)
-
I'd be really interrested if you know some suitable partnumbers for isolated gate drivers. I played around with this topology 10 years ago and back then the propagation delay for those devices was just too long. I'm sure we have evolved since that time.
Sure. Here's an example:
https://www.semikron.com/dl/service-support/downloads/download/semikron-datasheet-skhi-23-12-r-l5002371 (https://www.semikron.com/dl/service-support/downloads/download/semikron-datasheet-skhi-23-12-r-l5002371)
though I don't know how good the availability will be.
They rate it for a rather long rise/fall time, but I suspect that's because they provide it as-is with overly large Rg's, so you can connect your own in parallel as needed, or short them out completely.
Powerex makes one,
http://www.pwrx.com/Product/VLA500-01 (http://www.pwrx.com/Product/VLA500-01)
~$40 from Richardson.
Microsemi makes one, IXYS makes one, and a bunch of little players make their own too.
~$50 is good value for just lashing things together and being able to play with industrial (>$300) modules!
Or you can design and build your own, but you'll easily spend this much per board on the prototypes, and that's discounting your labor and NRE to zero.
Basicly we had to drive the primairy of an drrsct 90 degrees out of fase of the curent flowing through the primairy.
The easy way was measuring the current with a current transformer. Same 2e25 core. buffer the signal (74hc) and put it through a ucc37221/37222 pair into the gdt. perfectly self tunning, 90 degrees out of fase and very little propagation delay.
Optocouplers we're too slow back then.
Hmm, that doesn't make sense. And doesn't sound right, anyway.
Voltage 90 degrees out of phase with current means you have no load power. Tons of reactive power, but nothing performing work. It's tuned wrong.
You really need a proper controller for this application. It can be something like a dumb TL494 running wide open, just to get the alternating gate pulses -- and you modulate frequency to control phase and power.
You need three limits: you need a voltage detector on the resonant capacitor (more important for induction heating, which uses an identical circuit), so you don't overvolt and explode it; you need a current detector, so you don't explode the inverter; and you need a phase detector, so you don't fall through resonance if neither of the other limits pull in.
For voltage and current control, there's no phase loop whatsoever, so the propagation delay of the drivers doesn't matter at all.
For phase control, you measure the inverter output voltage and current, and detect the phase between them. Again, propagation delay doesn't matter.
Using the internal oscillator output would be silly, because of the uncompensated propagation delay. But even there, you can add a monostable delay to the detected current waveform zero-crossings to roughly compensate, or if you don't need much frequency range (which might be true for a SSTC, but likely not for an induction heater), just set the phase shift setpoint to an angle that accommodates the delay.
Tim
-
Thanks Tim.
A drrstc, dual resonant solid state tesla coil, is defficult to design under 50kHz. It consists of a primairy LC circuit that is loosly coupled to the secundairy LC resonator, k= max 0.2. Mostly due to size and insulation constraints.
Due to the low coupling you need a lot of reactive energy sloshing back and forth, because it takes several cycles to transfer energy from LC1 to LC2, unlike a normal transformer with k=0,999.
So current in LC1 is mostly reactive.
The output of the current transformer in LC1 gets clamped, sent through a fast smith trigger and straight to the ucc's
This ensures the igbt will turn on when the current is zero, this is the key.
This way they can switch much faster as the datasheet suggests and at full specked current.
Long propagation delay makes switching at 0 current rather complicated.
-
Ah, but current at the inverter, with respect to the inverter, must not be highly reactive. Indeed, that's what the capacitor is there for, absorbing the primary's inductive reactance! The reactance manifests as the resonant voltage between capacitor and primary; the inverter should see nearly unity power factor, because it's only providing "make up" power to the network. :)
If k ~ 0.2, then Q ~ 5, and you'll have say 300VAC output from the inverter, 1500V at the capacitor and primary, and equal current through all three. The inverter delivers approximately V*I watts, and the cap/coil resonate with V*I*Q apparent power.
Any phase shift or delay in a direct feedback circuit like that, would of course be problematic, but such circuits are a bad idea for a number of reasons anyway. They are prone to locking into unusual resonant modes (subharmonic and odd duty cycle resonances), essentially because the gain is extremely high (that's what a comparator is supposed to do!), and independent of frequency (i.e., the switching speed can be much faster than the propagation delay!). Which is at odds with what an oscillator is supposed to be: gain at just one frequency.
(TLDR: yes, I've familiar with the circuit, and I'm aware that it works. That doesn't make it any less cringeworthy...)
It's really quite hard to avoid the PLL-with-controller circuit I described, if you want any basic level of switching protection and operating stability.
Tim
-
Sorry Tim, im just a carpentar with a hobby in electronics. Mostly building tesla coils in all shapes,sizes and topologies.
I find it hard to follow you sometimes because you are much more knowledgeable then me.
I just follow what has done before and give my own twist to it. I did make a 125 KVA 9 meter sparking sparkgab coil before ie.
I guess what you're saying is that it's a bit ridiculous switching the reactive current through the H bridge because that will slosh back and forth between L and C anyway. You need to have the inverter just pushing the swing higher and higher.
What would you propose. something like this?
http://www.richieburnett.co.uk/indheat.html (http://www.richieburnett.co.uk/indheat.html)
I built it and works beautifully. The idea of using it in a teslacoil has been nagging me for quite some time. Do you think this is a good approach?
-
Sorry Tim, im just a carpentar with a hobby in electronics. Mostly building tesla coils in all shapes,sizes and topologies.
But even a carpenter must know geometry; or he can only follow plans that are complete with no missing dimensions! ;)
I guess what you're saying is that it's a bit ridiculous switching the reactive current through the H bridge because that will slosh back and forth between L and C anyway. You need to have the inverter just pushing the swing higher and higher.
You've misunderstood something fundamental about this circuit, I think--
The inverter is a constant voltage source.
The inverter, capacitor and primary are in series.
Thus, it is a classic series resonant circuit, with some equivalent loss on the primary inductor, which we can model as ESR. (It's more closely in parallel, because it's a transformer, even with the low k factor. Still, we can calculate an equivalent.)
For a series resonant circuit, the current is simply the current. The current is equal in all components. It can't be said to be reactive, because it has no reference to compare it to!
Now the voltage, that's the reactive part. The reactive voltage manifests across the coil and cap: for a given Q factor, this voltage is V_inv * Q, and the apparent power across the coil or the cap is approximately (V_inv * Q) * I, almost all reactive power!
If the inverter is driving exactly on resonance, the L and C voltages cancel out, connecting the inverter directly to the ESR. In other words, current is in phase with inverter voltage, which means no reactive power, all real power!
What would you propose. something like this?
http://www.richieburnett.co.uk/indheat.html (http://www.richieburnett.co.uk/indheat.html)
I built it and works beautifully. The idea of using it in a teslacoil has been nagging me for quite some time. Do you think this is a good approach?
This actually makes things slightly worse, but not enough to matter. Indeed, there are operating conditions where this circuit still exhibits zero phase shift (as seen by the inverter), or negative (i.e., capacitive)!
It's usually possible to pick values so that -- as you'd expect -- it behaves like a resonator in series with an inductor, and you can deliver high power at a modest inductive phase angle. The inductive angle helps with ZVS switching, making this popular for resonant switching supplies as well as SSTCs and induction heaters.
The biggest downside is, you need an inductor that handles about 1/3 the reactive power of the tank, so it tends to be large.
The series inductor also serves as a matching element, where the impedance of the parallel tank can be quite different (lower or higher!) than the inverter's expected load resistance (R ~= V_supply / I_inv). This is easier to see, if you imagine the tank capacitor being split into two parts: one associated with the series inductor, making a series resonant circuit; the other with the parallel tank, being supplied by the midpoint of the series circuit. I won't go into the math, but suffice it to say: you can use any resonant network as an impedance transformer, as long as the Q factor is high enough (and bandwidth low enough, same thing) to accommodate the impedance ratio.
Tim
-
Thanks tim, again for an another overwhelming answer. I need to look at this for a few days.
-
Some AP Notes on driving -
MOSFET Ap Notes
http://www.infineon.com/dgdl/an-937.pdf?fileId=5546d462533600a40153559ea1481181 (http://www.infineon.com/dgdl/an-937.pdf?fileId=5546d462533600a40153559ea1481181)
http://www2.st.com/content/ccc/resource/technical/document/application_note/68/cd/c6/ab/ef/17/41/06/CD00003900.pdf/files/CD00003900.pdf/jcr:content/translations/en.CD00003900.pdf (http://www2.st.com/content/ccc/resource/technical/document/application_note/68/cd/c6/ab/ef/17/41/06/CD00003900.pdf/files/CD00003900.pdf/jcr:content/translations/en.CD00003900.pdf)
http://www.radio-sensors.se/download/gate-driver2.pdf (http://www.radio-sensors.se/download/gate-driver2.pdf)
http://www.ixys.com/documents/appnotes/ixan0010.pdf (http://www.ixys.com/documents/appnotes/ixan0010.pdf)
https://www.fairchildsemi.com/application-notes/AN/AN-6069.pdf (https://www.fairchildsemi.com/application-notes/AN/AN-6069.pdf)
http://www.ti.com/lit/an/slua054/slua054.pdf (http://www.ti.com/lit/an/slua054/slua054.pdf)
http://ww1.microchip.com/downloads/en/AppNotes/00898a.pdf (http://ww1.microchip.com/downloads/en/AppNotes/00898a.pdf)
http://www.onsemi.com/PowerSolutions/supportDoc.do?type=appNotes&category=809 (http://www.onsemi.com/PowerSolutions/supportDoc.do?type=appNotes&category=809)
Regards, Dana.
this is another useful AN on this topic
http://www.ti.com/lit/ml/slua618/slua618.pdf (http://www.ti.com/lit/ml/slua618/slua618.pdf)