High side driving with 0 to 100% duty cycle

[Pudu]

Hi all,

I'm trying to optimize the performance of a class D modulator for a transmitter. This is basically a buck converter with active free-wheeling rectifier, or in other words, a plain simple half-bridge. The supply voltage to the half bridge is 50V. I need a 30kHz signal bandwidth. To make the lowpass filter reasonably simple and achieve enough suppression of the carrier, I'm using a 200kHz PWM frequency. I need the circuit to work as linearly as possible over the full 0 to 100% pulse width range, in order to get an output covering the full range from zero to as close to 50V as possible. A linearity like 0.1% would be completely satisfying. Somewhat worse is still acceptable.

Not finding any PWM IC that performs well enough, I made my own PWM circuit using fast comparators and op amps. That circuit is working reasonably well, although some improvement is still possible. It has some glitches - but it's currently assembled on a protoboard, and YES, I know very well about the limitations of protoboards, so please don't throw rocks at me! At least I used good parts layout and a good grounding scheme on my protoboard...

The reason of this post is asking if anybody can suggest a good way to drive the high side MOSFET.

My first thought was to use a half-bridge driver IC such as the IR21844, but a look into its datasheet was enough to discard it, because it has far too long minimum deadtime.

Scanning through parts catalogs, I found several candidate half-bridge buck drivers that have beautiful timing specs - but they all seem to be limited to rather low voltages. The highest I could find was one that can take 30V on the half bridge - but not 50V.

I then built a test setup using an IRS2110, driving its two inputs exactly in opposite phase. According to the datasheet, this should result in a short but nonzero deadtime at the outputs. But practical testing showed that this works well only from roughly 4% to 96% pulse width, at this frequency. This IC cannot correctly handle shorter pulses, or shorter pauses! When driving it with ever shorter pulses, it first freezes the output pulse width to roughly 100ns, and then suddenly drops to no pulses at all. Likewise when driving it with ever longer pulses, eventually it locks to about 97% pulse width, and then snaps to 100% duty cycle. The worst part is that the combined behaviour of the high and low side drivers is such that in a narrow input pulse width range, around 70ns, the two outputs actually are both on at the same time, for about 20ns! That's enough to drive fast MOSFETs into shoot-through. So it seems that the IRS2110 is unsuitable for applications requiring short deadtimes. From playing with it, it seems that the shortest deadtime at which this IC would work well is roughly 200ns, and that's too long to achieve the performance I'm trying to obtain. Given that the MOSFETs can be made to switch in about 30 to 50ns, it seems stupid to put in 200ns of deadtime just to accomodate a misbehaved driver IC!

The IRS2110, like all of those high voltage high side drivers I have seen, uses a pulsing system to bring the signal across the voltage barrier, and a flipflop to regenerate the output signal. Of course this sets limits on the timing performance it can achieve.

Since I anyway need some more gate drive current than the IRS2110 can deliver, I have been considering using two TC4422A drivers, and bringing the signal to the high side via a fast optocoupler. I happen to have some TC4422A's on hand, and also some HCPL7723 CMOS optocouplers, which seem well suited. But the circuit would get rather complex, for several reasons:

- The TC4422A doesn't have undervoltage lockout, so I would have to add it with additional parts, to keep the MOSFETs safe. This needs a voltage reference, a comparator and some gating circuit.

- The HCPL7723 is LED ON for OUTPUT LOW, so I need to invert the signal to avoid blow-through of the MOSFETs if the input-side power supply goes down. TC4421A drivers would solve this, but I don't have any.

- The optocoupler needs a 5V supply both on the input and the output, while the MOSFET driver needs 12V or so. And my PWM signal is generated from an 8V supply. So, I would need to add a 5V regulator to the floating 12V supply (which could double as the reference for the UVLO), another 5V regulator on the input side, and of course I would also have to insert such an optocoupler in the low side path, just to keep the two sides time-aligned.

It can all be done, but the resulting circuit is more complex and far less elegant than I would like.

Can anybody suggest a clever solution? Jellybean parts are strongly preferred, because in my country the electronic parts stores carry very little selection, the postal service takes 4 months to deliver any import (even letters!), and courier companies charge far too much, specially for small orders of electronic parts. Still, any idea is welcome - I have a small selection of interesting parts in stock, and some part donors in my junk box, so I can see if I can implement any idea you suggest, using what I have.

I'm grateful for any good idea!

 

[KlausST]

Hi,

first: 100% duty cylce may be a problem, because almost all high side drivers work with a charge pump. Thus 99% may be more easy to achieve. everything less is more easy.

Why don´t you use a complete circuit that includes fast drivers and a fast power stage. Something like DRV8432.
Or real audio class D stages. They are optimized for high precision, very good linearity. I think it´s not easy to achieve similar performance with discrete parts.
Additionally they include additional features like overcurrent protection, overtemperature protection...

Klaus

 

[Pudu]

Hi Klaus,

It's clear that if I use up to 100% duty cycle, I need a DC-DC converter to power the high side driver. I can either look for some ready-made, small DC-DC converter that has low enough input-output capacitance and good enough dV/dt tolerance, or more easily I can build my own. In the past I have made such DC-DC converters, to replace commercial ones that kept failing! The way I built them is with a single-transistor oscillator running at roughly 1MHz, using a toroidal core, with a physically separated secondary winding feeding a rectifier, capacitor and zener diode. It's very crude and not highly efficient, but works well to power a high-side driver, that typically requires less than 1 watt. The input-output capacitance is less than 1pF.

On the other hand, in my project I could indeed work with a maximum duty cycle of 95% or so. I would simply need a slightly higher supply voltage, like 53V, and the high side driver can then be bootstrap-powered, and the distortion problem when nearing 100% would be avoided. But in any case I do need to modulate down all the way to 0%, with decent linearity. To avoid the 0 to 5% range, I would need to add a high current, very low voltage negative supply, and that's not practical.

The DRV8432 is interesting indeed! Unfortunately it's too small for my project. I need a maximum sustained output current of 40A, so the switch current, when including the inductor's ripple current, is up to 50A or 55A. The DRV8432 handles only 9A sustained current per channel (curiously the datasheet says 9mA, but that must be a typo!), so even by combining all 4 channels (in phase, or in a 4-phase system), that IC is too small. And its efficiency at 200kHz is given as 94%. In my application that would lead to around 100W dissipation in that tiny chip! That would require an extremely good heatsinking solution.

Also it's very uncomfortable for me to work with such chips that have 0.65mm pin pitch. I far prefer old-fashioned through-hole parts, and when I do use SMDs, I try to stick to ICs that have 1.27mm pin pitch, transistors no smaller than SOT-23, and passive components in 1206 size. Everything smaller is really a pain for manual assembly. When I make printed circuits at home, I can also make them fine enough for these parts, but not for those 0.65mm parts. I can make double-sided boards, but I cannot make plated-through holes. And having PCBs professionally made isn't an option, because there is no company in my country making good ones, and having them made overseas incurs in either 4 month delays for postal shipping, or very high cost for courier shipping.

So I'm rather limited by external circumstances!

I looked around on the TI website, and indeed they have many nice ICs, but I couldn't spot any that's really good for what I need. Basically I just need a really fast, DC-true and time-true replacement for an IR2110! I don't think I can use a fully integrated solution, with the power stage built-in, because of the power level I need to handle. So my intention is to use two sets of four to five TO220-encased MOSFETs, physically arranged in an interleaved fashion on a double-sided board and attached to a heatsink, driven by two TC4422A chips placed very close to the FETs, and I just miss the interface between my logic-level PWM signal and the inputs to these driver chips, which also has to implement UVLO. The quest is to find the simplest, most practical circuit that does this. The typical high/low side drivers like the IR2110 would fill that hole, if they only were fast enough.

Can you suggest any ready-made class-D audio amplifier blocks, so that I could look at their datasheets and see if any could be used? The ones I have found are all too small in terms of power output, and typically they don't work over the 0 to 100% duty cycle range. In audio (speaker amplifiers) the most common way of operating is to have the amplifier idling at 50% duty cycle, and modulate the duty cycle up and down for the audio signal. Instead of driving it to the absolute saturation at 0% and 100% , they just limit the drive a little to avoid the extreme duty cycles. That works great for speaker amplifiers, but not for my application.

What I have lying on my desk, on that humble protoboard with all its parasitics, built with the IRS2110, could work as a speaker amplifier with a distortion below 0.1%, by keeping the duty cycle in the range of 5% to 95%. It just gets bad in the duty cycle extremes, and ready-made class-D audio amplifiers very likely suffer from the same problem.

The very reason why I'm building this from scratch is that I couldn't find ICs with enough performance, even for the individual blocks! For example I tried to use the LTC6992-1 as my PWM generator. That IC actually gets pretty well to the duty cycle extremes, but it's awfully nonlinear over the whole range, with several irregularities and zones of varying gain. It wasn't made for low distortion applications, of course, so I can't blame it - it's just not suitable for my use. I also tried to use an UC3638 motor controller IC as my PWM. It works well over the moderate duty cycle range, but gets very bad at extreme duty cycles. I tried several power supply controllers too, like the ubiquitous TL494, with poor results. They are just not made for low distortion and wide duty cycle range. That's why I ended up with my homebrew PWM based on fast comparators and opamps, which works well enough. Now I just need to interface it to the MOSFETs.

Does that mean that I have to implement my own high-side driver too, instead of using an IC? Maybe...

Manfred

 

[KlausST]

Hi,

To avoid the 0 to 5% range, I would need

Why? There is no need to avoid it. At least not for the drivers.

And having PCBs professionally made isn't an option, because there is no company in my country making good ones, and having them made overseas incurs in either 4 month delays for postal shipping, or very high cost for courier shipping.

Afaik there are cheap manufacturers shipping them with short delivery time all over the world. 1 week shouldn´t be a problem.

Audio Class D: a search at farnell (I didn´t read their datasheets)
TAS5261 (integrated power stage)
IRS2092 (external power stage)
International Rectifier -> now a part of Infineon may have other interesting parts. And application notes. A good source of information.

Klaus

- - - Updated - - -

Added:
What about digital PWM? --> good linearity.
What about "feedback" to improve linearity?

Klaus

 

[Pudu]

Klaus,

To avoid the 0 to 5% range, I would need
Why? There is no need to avoid it. At least not for the drivers.

In my original post I detailed the problem I found with the IRS2110 driver in the 0 to 5% range. This problem is inherent to the design of voltage-shifting high-side drivers, so I suspect that all such drivers suffer from it to some degree. So I either have to avoid that range, or I have to avoid those drivers!

Afaik there are cheap manufacturers shipping them with short delivery time all over the world. 1 week shouldn´t be a problem.

Unfortunately it's not so. The manufacturers indeed make and ship the boards quickly, but they have no control over the rest of the delivery process. Unfortunately in my country (Chile) the postal service is taking an average of 4 months to take a parcel from the airport and bring it to my P.O.Box. They claim it's due to customs processing, but even plain simple letters are taking 4 months to arrive, and letters do not go through customs, AFAIK! So it's clearly the Chilean postal service causing these ridiculous delays.

Last year I placed a total of about 25 small orders for electronic parts and some other things, in several countries including China, USA, Germany, England, Canada. The fastest took 9 weeks to arrive, the average was four and a half months, and the slowest still hasn't arrived, 6 months after it was shipped, and probably was lost or stolen. Several of those orders, placed after september 2017, still haven't arrived. The most recent order that has already arrived is from october of last year.

Courier services (Fedex, DHL, UPS) are available and take only about one week to deliver, but a small parcel sent to Chile by courier service costs typically around 200 Euros, divided into the shipping charge proper and the the service charge they raise when delivering. The taxes are in addition to this, of course. So I avoid courier shipping as much as possible. I'm doing electronics as a hobby nowadays, and I don't enjoy spending lots of money on shipping!

So my first option is to make do with the components I already have at hand, and those I can buy in Chile. RS-Components does offer its service in Chile, and is the best source for parts right now. My second option is to take advantage of some friend traveling abroad. Some time ago a friend went to Germany for a few weeks, and I bought some parts I needed from a distributor in the USA, and had them sent to his temporary address in Germany. That way I got them just three weeks after ordering. But I don't always have such nice friends traveling just at the right time!

And my third option is to order parts from China, the USA, or Europe, and then wait... and wait.... and wait! I do that with parts I will need for my next project, but preferrably not for my current one. And PCBs naturally are always for current projects! I don't know of any PCB service offering good quality that can deliver to Chile in a quick and not too expensive way.

The TAS5261 is much too small for my project. So I didn't look further into its datasheet.

The IRS2092 is interesting, so I had a long and hard look at its datasheet. But I don't think it would work well in my application, because of the modulation method it uses: It's a free-oscillating feedback amplifier. Near the middle of the output range, which is at ground level when using a split supply as shown in the typical application, such amplifiers work very well, but the closer they get to the extremes of their output swing, the more spurious components appear on the output. In an audio amplifier, driving a speaker, it doesn't matter if there is broadband dirt just above the audio range, at a level of maybe -30dB, when the amplifier is driven to its peak power. It's ultrasonic and inaudible, and weak enough to not damage the tweeter. But in my application this would be completely unacceptable, because all the crud that isn't suppressed by the lowpass filter would modulate my transmitter and appear as spurious signals causing interference to nearby channels. So I do need true, fixed-frequency PWM, at a relatively high carrier frequency like 200kHz, which allows moving all the crud far enough up in the spectrum to adequately suppress it in the lowpass filter.

The IRS2092 doesn't seem to be very well suited to driving it with a PWM signal instead of having it generate its own modulation. In addition its datasheet tells about a minimum pulse width, below which the output logic may malfunction. That's unacceptable if I want to modulate all the way from 0% duty cycle upwards.

I

nternational Rectifier -> now a part of Infineon may have other interesting parts. And application notes. A good source of information.

I checked again, their whole long list of gate drivers. All of their high-side drivers use that method with a flipflop driven by set/reset pulses internally generated from the input signal, and this method has the fundamental limitation that it cannot work linearly down to zero duty cycle. The best IC in their list, for my application, seems to be the IRS2011, which I would follow by two TC4422A power drivers, but it's still not good enough to make me hold much expectations. And I have none at hand to try it. The optocoupler route seems more promising, compared to that.

What about digital PWM? --> good linearity.

The gentleman who is writing the software of this radio's FPGA is working on that. Last thing I heard was that he doesn't see how to go beyond 100kHz along with 10 bit resolution. And that's a bit low. It seems the highest clock frequency he has in that FPGA is 125MHz. But he is resourceful and enthusiastic, so he might find a way to do it! He also mentioned sigma-delta encoding, with a 2MHz pulse frequency - but I shudder when I think about the switching losses this would cause in my 50V 40A MOSFETs!

In any case, if he comes up with a good software solution, it would remove the need to use a hardware PWM, but the need for a good, fast, DC-true and time-true high/low side driver would persist. And that's my main problem, not the PWM circuit.

What about "feedback" to improve linearity?

That's indeed an option I have been thinking about. But I worry about many things (perhaps too many!). For example loop stability issues, interference from the high power RF amplifier working right next to this module, and also about a very basic thing, best explained in an example: If I need 1% output voltage at a given point of the signal, and the driver isn't capable of producing anything shorter than 5% duty cycle, jumping directly from there to no pulse at all, then the feedback will produce the correct output level by making the system produce one 5% pulse and then skip 4 pulses. So the effective carrier frequency drops from 200kHz to 40kHz, which my lowpass filter can't suppress well enough, and my transmitter starts interfering on channels 40kHz above and below my own transmit frequency!

So I think it's actually better in this case to avoid feedback, and make an open-loop system that's as linear as possible, and live with that.

Manfred

 

[schmitt trigger]

Uncommon problems require uncommon solutions.

Let me see if I understand your request correctly:

Since you are already planning to use optocouplers and isolated gate supplies, would you consider to drive the “high “ side mosfet with a plain low side driver?

However, The whole isolated supply would be switching up and down, and I ignore whether that could cause EMI.

 

[Pudu]

Yes, if I end up using optocouplers then certainly I would use plain simple "low side drivers" on both sides.

The only such driver I currently have on hand for experimenting is the TC4422A. It's pretty good, but lacks undervoltage lockout, and also lacks an enable input. So yesterday I ordered some UCC37321 and UCC37322, which are basically the same thing but with an enable input, which allows implementing undervoltage lockout in a reasonably simple way. For the moment that seems to be the way to go for me. In the four months it will probably take for those chips to arrive (fabulous postal service!) I can play with the TC4422A drivers and try to avoid undervoltage situations.

Switching up and down the whole circuit comprising the output side of the optocoupler, the entire driver chip, the UVLO circuit, and the secondary side of the DC-DC converter, shouldn't cause any problem as long as the optocoupler can handle the high dV/dT, and the DC-DC converter has low enough input-output capacitance.

EMI isn't a big concern in this application, because this circuit will go inside a kilowatt-class shortwave transmitter. There will be extensive shielding and filtering, and any EMI to other devices will be dominated by the spurious signals and broadband noise from the transmitter. And during receive the whole transmit circuitry including this converter is shut down.

 

[KlausST]

Hi,

undervoltage lockout,

If you have a DCDC supply..why do you want undervoltage lockout?

Klaus

 

[Pudu]

I want UVLO to be safe. You surely know Murphy's Laws? I try to defeat Murphy in my designs - not always with success, but mostly!

Without UVLO it would be essential to start up and shut down the various power supplies and other circuits in a controlled, sequenced way. Doing this in such a way that it also works in the event of a power cut, fast power cycling, or other abnormal events, is much harder than including UVLO in the design.

Also I like to design circuits that don't suffer a chain reaction when something fails. Without UVLO, a failure of the DC-DC converter or the sequencing system could easily burn out the entire set of power MOSFETs, their drivers, and maybe something else.

So I think that UVLO helps in a simple and effective way to make a circuit more reliable, easier to design and modify, and also cut damage in the event of a malfunction. With drivers that have a shutdown input with hysteresis, UVLO can be implemented by a simple circuit comprising a zener and two resistors, maybe adding a clamping diode if the IC's input requires that. If the shutdown input has a sufficiently precise and stable threshold, the zener isn't even needed. And one can tailor the UVLO threshold for the power devices one will use.

Of course it's even more practical to use driver ICs that have built-in UVLO, and I usually prefer that method. But they are not available in every configuration one might need, and sometimes the UVLO threshold they have is unsuitable. For example, countless Chinese inverters have failed because they use IR2110 ICs to drive IGBTs. The UVLO threshold of an IR2110 can be as low as 7.0V. That's a bit low even for MOSFETs, and definitely much too low for a typical IGBT! So whenever the battery starts dropping in voltage while the inverter is delivering high load, those IGBTs go up in smoke. Also many of those inverters use much too small electrolytic capacitors in the bootstrap supplies for the high sides. The capacitors dry out over time, the ESR goes up, the capacitance goes down, the driving of the high side switches gets ever more marginal, and eventually the devices blow up.

I have a small collection of such inverters in my junk box, abandoned here by friends after asking me to repair them, and my reply that if repaired they will just fail again! Worst of all is that some of those inverters failed with such a combination of shorted and open IGBTs that they put 350V DC on the 220V AC output! A good friend of mine burned out $4000 worth of various equipment in his home and lab, because a poorly designed $500 inverter failed that way. Every small power transformer, fan motor, AC relay, etc, smoked off into Nirvana.

Of course those inverters don't have any protection against DC on the output...

High power switching circuits without UVLO are a play with fire - or at least with smoke!

 

[KlausST]

Hi,

I want UVLO to be safe. You surely know Murphy's Laws? I try to defeat Murphy in my designs - not always with success, but mostly!

How can you think that the DCDC converter may be faulty, but at the same time think that the UVL circuit is not faulty?

Klaus

 

[Pudu]

How can you think that the DCDC converter may be faulty, but at the same time think that the UVL circuit is not faulty?

Do you really expect an answer to that? It's a bit like splitting hairs.

Well... For one thing, the UVLO circuit would consist of two resistors, and possibly a zener diode and a maybe an additional switching diode. All of those parts running at very low dissipation, and thus cool. That's pretty reliable. Instead a DC-DC converter has more parts, and some of them are subject to some stress - specially if something else fails, or if there are transients coming from the external world.

And the other thing is that it would take both the DCDC and the UVLO failing, to burn out my FETs! That's far safer than a single failure causing a blow-up. You might consider it a form of redundancy.

Would you take this as a valid answer?

 

[Easy peasy]

A long time ago we developed and sold small gate driver board for 0-100% gate drive for use on an 800V bus in inverters, up to 50kHz, opto coupled ( HCPL2601 - 10kV/uS CM immunity ) along with small CM beads on the supply input ( +/-15V at 100kHz) and the signal input +/- 5V to the LED with a reverse diode on the LED and 12mA drive on the remote driver ( 10 way ribbon for reduced Zo on the cable).
the output of the opto was fed to a high speed Schmidt trigger ckt buffered to give +/- 6A drive at +15 (adj) and -5V to the fets and IGBT's we were driving.
Each board had local UVLO, and de-sat detect in case another device went short.
The same board was used for high side, low side and midpoint devices ( 3 level inverter ) to equalise delays... 50kW at that time

total delay 100nS rise and 50nS fall, rise time of output volts was in the order of 40nS rail to rail.

Easy enough to design and build these days with similar results...

- - - Updated - - -

HCPL-4503 - a very good opto too, 15kV/uS CM

 

[mtwieg]

But in any case I do need to modulate down all the way to 0%, with decent linearity. To avoid the 0 to 5% range, I would need to add a high current, very low voltage negative supply, and that's not practical.

Doesn't sound worthwhile for me. As someone who has spent years working with envelope-tracking RFPAs, modulating the RF power down to 0% isn't possible due leakage in the RFPAs. That will often define the lower end of your dynamic range, not the tracking supply.

Also it's very uncomfortable for me to work with such chips that have 0.65mm pin pitch.

Hate to say this, but this is something you'll have to get over. The new, fast devices are small for good reason. My lab's etching setup is fairly simple but we can do 0.65mm pitch and 10mil traces with decent results.

The very reason why I'm building this from scratch is that I couldn't find ICs with enough performance, even for the individual blocks! For example I tried to use the LTC6992-1 as my PWM generator. That IC actually gets pretty well to the duty cycle extremes, but it's awfully nonlinear over the whole range, with several irregularities and zones of varying gain. It wasn't made for low distortion applications, of course, so I can't blame it - it's just not suitable for my use.

Interesting, I had considered using that chip in the past, but never got around to it. Was the duty cycle at least monotonic and smooth over the operating range? What about response time?

I've used the UCC35705 for high frequency PWM modulators before. Aside from that I've always done my own via discrete devices.

You should seriously consider using GaN FETs. EPC's devices are especially suited for low voltage, but the discrete FETs probably aren't feasible for you if 0.65mm is a problem. However there are a couple half bridge modules including the gate drivers and GaN FETs which would get around that limitation of yours. I've used EPC devices extensively and they can do pulse widths down to 10ns. Check out these modules from EPC and this one from TI (I believe its internals are basically the same as the EPC module). A few of those in parallel should meet your power spec fine (especially if you interleave them). Both are usually in stock at digikey. I have no advice on the shipping issues, though...

Some of EPC's dev boards use the Si8610BC in place of an optocoupler on some of the boards, so I presume it should work for your applications.

 

[asdf44]

I'm also designing digitally controlled amplifiers and Class-D's so I have a few general comments.

-Look at this DC-DC - cheap/small/very low C
https://power.murata.com/data/power/ncl/kdc_nxj1.pdf

-Look at isolated gate drivers which work on a different principle than the non-isolated level shifted ones you've been referring too (many UVLO options available):
https://www.digikey.com/product-detail/en/silicon-labs/SI8274AB1-IS1/336-3542-5-ND/5804360

-Or gate drivers optimized for GAN which must be fast
LM5113-Q1
https://www.psemi.com/products/gan-fet-driver

FPGA PWM Generation
-See this thread I started on this topic:
https://www.edaboard.com/showthread.php?t=364633
-I settled on the basic shift register I proposed there plus a DDR register for 800Mhz resolution in a Xilinx 7 series
-Note that due to dithering (either by design or loop limit-cycling) output resolution can be higher than PWM resolution

GANs
-Agree with that suggestion, they're much faster
-Gan systems puts 100V devices in a non-exotic package

Lastly did you explain why you need such a wide duty cycle? Normally you have an output voltage requirement and work backwards - choose a DC bus so you never need to go beyond 90 and 10 duty cycle.

 

[Pudu]

Interesting posts, thank you all!

To Easy peasy:

The driver boards you mention there are pretty comparable to what I'm trying to come up with, except that mine have lower requirements in voltage and power, but higher speed requirements. I have the benefit that my power stage will be physically small and compact, and that the PWM will be closeby (as long as I can pull that off without too much interference from the power stage!), so that instead of wires from the PWM to the optocouplers I would just have board traces a few cm long.

The HCPL-2601 you mention is quite decent, but the HCPL-7723 I have at hand is much better! And it needs to be, for the speed I want...

Instead the HCPL-4503 is very slow, compared to the others! It would be absolutely unsuitable.

To mtwieg:

Doesn't sound worthwhile for me. As someone who has spent years working with envelope-tracking RFPAs, modulating the RF power down to 0% isn't possible due leakage in the RFPAs. That will often define the lower end of your dynamic range, not the tracking supply.

When I tried to develop an EER RFPA twenty years ago, I found the exact same! The drive power feedthrough was just too much. But technology has advanced a lot. 20 years ago I used VMOSFETs, and nowadays I'm using LDMOSFETs. They have very much lower Crss, along with much higher gain. The combination of these two facts has pretty much removed the problem of drive power feedthrough. The effective saturated gain of the LDMOSFET power stage is 26dB, the drive feedthrough attenuation is a good 20dB, and what's more important, I gave up the constant-amplitude drive of EER and instead I'm using proportional drive, so that the drive signal is about 6dB into saturation at all power levels, down to zero, plus/minus a few dB given by the gain nonlinearity of the LDMOSFETs. The end result is that drive power feedthrough has completely ceased to be a problem.

Hate to say this, but this is something you'll have to get over. The new, fast devices are small for good reason.

And I hate to admit that that's right... to a point, at least. Still I prefer working with somewhat larger devices whenever possible. I guess this happens to most aging people: We delight in the capabilities of new technology, but hate the difficulties that come with it.

Having to use 7 different glasses to see at various distances definitely doesn't encourage working with the tiniest parts.

My lab's etching setup is fairly simple but we can do 0.65mm pitch and 10mil traces with decent results.

Can you give me some advice? I'm using an inkjet photo printer on Pictorico Ultra Premium film, transferring to positive presensitized board with the ink side down onto the board, between glass plates, exposing with a metal halide lamp to get something close to a point source for improved sharpness. The etchant is ferric chloride. At 0.65mm pin pitch my results aren't fully reliable. Also I worry about corrosion of the copper over time, with unprotected tracks that fine. Any advice is welcome, as long as it can be reasonably implemented in a home lab.

Actually I had better accuracy many years ago, when I printed my layouts at 2:1 scale or even larger and reduced them photographically to size, using Agfa Litex film. But that time is over. I still have the enlarger and developing trays and so on, but I can't get that material anymore, let alone the developer. "Dangerous chemical", thus banned from postal shipping.

I have tried paste soldering with poor results at tiny pin spacings - too many shorts from excess solder, since it's so hard to apply just the right amount. So I tin the pads thinly, then apply flux and the parts, and reheat, but the results are still all over the place.

I really love good old honest and "huge" through-hole parts, that can be soldered by hand at a rate of one pin per second, with highly reliable results. Maybe I should leave the nanoseconds alone and satisfy myself with tens of nanoseconds...

Interesting, I had considered using that chip in the past, but never got around to it. Was the duty cycle at least monotonic and smooth over the operating range? What about response time?

Yes, it was monotonic, but not smooth. It tends to "hang" at two places (roughly 1/3 and 2/3 of the range). And each third of the range showed a different gain. And there was considerable frequency pulling.

I didn't measure the response time, since it's rather irrelevant in my application, as I can compensate for any (fixed) delay in the digital signal processing. But it must have been small, because the reconstructed signal after the power stage didn't seem to have any unexpected delay beyond what's caused by the low pass filter.

I've used the UCC35705 for high frequency PWM modulators before. Aside from that I've always done my own via discrete devices.

That IC doesn't seem well suited to my needs. It won't get close enough to 100% duty cycle, and while it's rated to get to 0%, there is no information about its linearity at very small duty cycles. Having a flipflop output, it very likely has a certain minimum pulse width, and snaps from there directly to zero.

Really it seems better to keep my opamp-and-comparator-based PWM.

You should seriously consider using GaN FETs.

I have looked at them before, but really haven't come yet across a need to use them. Same with SiC FETs. What I need to do now can be handled with conventional silicon MOSFETs, of the cheap kind, in TO220 cases. Several in parallel, of a low gate charge type, on a good board layout, are just fast enough. But I'm following GaN developments with an eye on eventually using them for RF. The core idea, somewhat crazy, is a class D amplifier covering to 30MHz, with a switching frequency in the UHF range. Conventional RFPAs in the HF range almost always need a bank of switchable low-pass filters. A class D design like this could do with a single low-pass filter for the whole DC to 30MHz spectrum.

But that's a future project. For the moment I want to get my LDMOSFET ET system into shape!

Some of EPC's dev boards use the Si8610BC in place of an optocoupler on some of the boards, so I presume it should work for your applications.

Interesting device, and funny, with that LED-simulating input! Yes, it might work for me, but according to its specs it's likely that a combination of a fast optocoupler and a strong gate driver IC is better in most regards (minimum pulse width, pulse width distortion, skew, driving current). But it's certainly a device to keep in mind.

I wonder how resistant they are to intense radiated EMI from nearby power stages and filters. Their transient immunity rating is excellent, but what happens if they get into the magnetic stray field from an inductor?


To asdf44:

-Look at this DC-DC - cheap/small/very low C
https://power.murata.com/data/power/ncl/kdc_nxj1.pdf

That one seems well suited for powering a high side driver.

-Look at isolated gate drivers which work on a different principle than the non-isolated level shifted ones you've been referring too (many UVLO options available):
**broken link removed**

That's really a very interesting series of drivers! With selectable UVLO, with half-bridge drivers that incorporate a very small deadtime, selection between low-jitter and de-glitched versions, and extreme dV/dt robustness! I think I will have to order some of these! I will devote some more time to studying them in detail, but it looks like these could indeed be a single-chip solution for my problem!

-Or gate drivers optimized for GAN which must be fast
LM5113-Q1

That one isn't well suited for the kind of silicon FETs I'm using, and also I wonder about the minimum pulse width requirement in the datasheet. Could it be that with shorter pulse widths it might lock up in the logic high state?

https://www.psemi.com/products/gan-fet-driver

And those seem much over the top for my needs!

FPGA PWM Generation
-See this thread I started on this topic:
https://www.edaboard.com/showthread.php?t=364633
-I settled on the basic shift register I proposed there plus a DDR register for 800Mhz resolution in a Xilinx 7 series
-Note that due to dithering (either by design or loop limit-cycling) output resolution can be higher than PWM resolution

I haven't yet programmed any FPGAs, so I don't know about the intrincacies of getting fast logic to work inside them. I hope I will eventually get into FPGAs, but not before I have completed my current project. There is a very capable guy doing the programming, I'm using his software and I don't see how I could directly contribute to it. My main contributions will mostly be hardware-related.

Lastly did you explain why you need such a wide duty cycle? Normally you have an output voltage requirement and work backwards - choose a DC bus so you never need to go beyond 90 and 10 duty cycle.

I agree on working backwards. I started at the antenna output of my transmitter.

I need zero to 50V at the output of my converter. In principle I can very well use a 55V supply and limit the range to 90% duty cycle, but to start from 10% duty cycle I would need to add a 5V negative supply, and that's a significant additional complication which I would like to avoid.

A minor factor involved is that of safety: High power LDMOSFETs are expensive. If anything goes wrong with a 0-100% PWM working off a single 50V supply, the LDMOSFET stage won't get anything outside the 0-50V range it can safely handle. But with a +55/-5V supply, the LDMOSFET stage might get anything inside that range, if the PWM stage fails. Specially -5V is very worrying!

So my position is: I can give up modulating to 100%, and satify myself with 95 or 96%, which seems quite easy to achieve. The existing 50V supply can be adjusted to supply 53V. But I would still like to get down to as close as possible to 0%, as linearly as possible, and then see what kind of signal quality I can generate with it.

For those interested in EER and ET transmitters: My preliminary testing with the current PWM system that has some trouble at both ends of the PW range produces an RF signal that has the 3rd order IMD products suppressed by 45 to 50dB, depending on band and other factors. That's already much better than conventional transmitters. But I want to get it as good as I can! The pulse width distortions at the extremes of the range cause envelope distortion that's visible on a scope, and it actually suprised me that the IMD was still that good. The fine print is that the high order IMD products are only 60 to 70dB down, over a wide bandwidth, and that's worse than what a conventional transmitter can do. That's why I'm trying to get the envelope modulation signal looking clean on the scope, before going further. From correlating the modulation waveform to the spectrum of the output I'm pretty sure that a good part of the high order IMD is coming from the envelope distortion at low amplitudes.

 

[asdf44]

Well life is such that requirements are always in competition.

Linearity seems to demand wider supplies (or a full bridge topology instead of half). Perhaps you can mitigate your voltage limit fears with a fast comparator circuit that disables gate drives if it sees anything outside 0-50V for example.

 

[mtwieg]

and what's more important, I gave up the constant-amplitude drive of EER and instead I'm using proportional drive

In that case feedthrough likely won't be the bottleneck. But if you are modulating the RF input power, then you don't actually need tto back the bias down to zero in order to get your RF output down to zero. Why bother?

Can you give me some advice? I'm using an inkjet photo printer on Pictorico Ultra Premium film, transferring to positive presensitized board with the ink side down onto the board, between glass plates, exposing with a metal halide lamp to get something close to a point source for improved sharpness. The etchant is ferric chloride.

We use a LV204 exposure system and Rota Spray etching tank from Mega Electronics. We use our own bath of NaOH for developing. Sensitized copper clad and ferric chloride from MG chemicals. For exposure we have used both plastic transparencies and heavy vellum tracing paper with an inkjet printer. Transparencies are easy to use but tracing paper gives better results (but are much harder to align for two layer boards).

I have tried paste soldering with poor results at tiny pin spacings - too many shorts from excess solder, since it's so hard to apply just the right amount. So I tin the pads thinly, then apply flux and the parts, and reheat, but the results are still all over the place.

With 0.65mm parts I expect to make a few bridges here and there. If I'm feeling lazy, I just dump a blob a solder on them, then wick away the excess to leave behind decent joins.

That IC doesn't seem well suited to my needs. It won't get close enough to 100% duty cycle, and while it's rated to get to 0%, there is no information about its linearity at very small duty cycles. Having a flipflop output, it very likely has a certain minimum pulse width, and snaps from there directly to zero.

I've used it at 500KHZ and get pulse width down to 10ns and duty cycle up to 95% (tie the DISCH pin low).

That one isn't well suited for the kind of silicon FETs I'm using, and also I wonder about the minimum pulse width requirement in the datasheet. Could it be that with shorter pulse widths it might lock up in the logic high state?

Exactly, how did you guess?

 

[Pudu]

Well life is such that requirements are always in competition.

Oh yes. And engineering even more so.

Linearity seems to demand wider supplies (or a full bridge topology instead of half).

A slightly higher supply is OK. But I don't want to go with dual supplies, and even less use a bridge and having to float the high power RF circuitry. I will see what's the best I can do with reasonably simple circuitry, and stick to that.

Perhaps you can mitigate your voltage limit fears with a fast comparator circuit that disables gate drives if it sees anything outside 0-50V for example.

Yes, but that means even more circuitry. At some point it stops being attractive. I have the hopes that if this radio ends up being good enough and simple enough , others will copy it, or take ideas from it. The more complex it gets, the less likely that is.

But if you are modulating the RF input power, then you don't actually need tto back the bias down to zero in order to get your RF output down to zero. Why bother?

To maintain linearity. I have tried approaches that combine envelope restoration in the mid and high part of the dynamic range with fixed-supply class AB at low signal levels, and have found it next to impossible to maintain good linearity on all bands and under all load conditions. Instead when fully modulating the supply voltage it's easy.

We use a LV204 exposure system and Rota Spray etching tank from Mega Electronics. We use our own bath of NaOH for developing. Sensitized copper clad and ferric chloride from MG chemicals. For exposure we have used both plastic transparencies and heavy vellum tracing paper with an inkjet printer. Transparencies are easy to use but tracing paper gives better results (but are much harder to align for two layer boards).

That exposure box uses diffuse light. That provides even exposure, but if you get any separation between the board and the artwork, traces will narrow and even disappear. For that reason I prefer using a point source of light. Many years ago I simply used the sun, but sometimes it was uncollaborative and I had to wait for a better day... So nowadays I use a 150W metal halide lamp, placed far enough away to produce even illumination over the board.

That Rota Spray tank looks nice! My etching system is far more homemade: A plastic tray, floating in warm water in the kitchen sink. A slight rocking motion provides agitation. When I make double-sided boards, I wind a few loops of plastic-insulated wire around the board, to act as separator and keep the bottom side from touching the tray. Despite the low tech system, the results should be similar.

I will try using tracing paper. Anyway the Pictorico transparency isn't far from tracing paper. It's translucent but not clear, and the printer prints pretty well on it, configured for high quality photo paper.

With 0.65mm parts I expect to make a few bridges here and there. If I'm feeling lazy, I just dump a blob a solder on them, then wick away the excess to leave behind decent joins.

My problem with that method is that if there is a short to any trace under the IC, it's over. I have to remove the IC to clear the short, and start over.

I can do it. But it's simply quicker, more comfortable and more practical to use through-hole parts whenever available and acceptable.

I've used it at 500KHZ and get pulse width down to 10ns and duty cycle up to 95% (tie the DISCH pin low).

So I should get a few and try them.

>That one isn't well suited for the kind of silicon FETs I'm using, and also I wonder about the minimum pulse width requirement in the datasheet. Could it be that with shorter pulse widths it might lock up in the logic high state?

Exactly, how did you guess?

40 years doing electronics can teach one to read between the lines of datasheets. And any such IC using a flipflop to reconstruct the output pulse can suffer from this sort of problem. Those that don't usually employ an internal circuit that strictly enforces a sufficiently high minimum pulse width to keep the flipflop behaving.

 

[mtwieg]

To maintain linearity. I have tried approaches that combine envelope restoration in the mid and high part of the dynamic range with fixed-supply class AB at low signal levels, and have found it next to impossible to maintain good linearity on all bands and under all load conditions. Instead when fully modulating the supply voltage it's easy.

I'm skeptical... it's certainly not easy either way (especially without feeback), but when you modulate the supply to such a degree you're going to get a great deal of AM-PM due to nonlinearities in the FET Coss. And at some point VDS will be so low that the FET is going to be operating in its triode region, which will cause lots of AM-AM distortion. Unless you're also varying the Vgs bias, which opens up a whole other can of worms....

That exposure box uses diffuse light. That provides even exposure, but if you get any separation between the board and the artwork, traces will narrow and even disappear.

Never had this problem, personally. The system draws a vacuum to keep things nice and tight.

I will try using tracing paper. Anyway the Pictorico transparency isn't far from tracing paper. It's translucent but not clear, and the printer prints pretty well on it, configured for high quality photo paper.



My problem with that method is that if there is a short to any trace under the IC, it's over. I have to remove the IC to clear the short, and start over.

I started using tracing paper based on advice I found on this website. Like it says, get the heaviest stuff you can find to avoid crinkling in the printer. The nice thing about vellum vs transparency is that the ink won't bead up due to surface tension.

40 years doing electronics can teach one to read between the lines of datasheets. And any such IC using a flipflop to reconstruct the output pulse can suffer from this sort of problem. Those that don't usually employ an internal circuit that strictly enforces a sufficiently high minimum pulse width to keep the flipflop behaving.

I don't see any reference to a FF in the datasheet for the LM5113. Though I certainly wouldn't be surprised if there is one in the high side level shifter. I believe I was the first one to report that silicon but in the LM5113, since then they've modified the datasheet to mention it, and released a new revision (LMG1205). But the new one is only available in a DSBGA package which even I don't want deal with....

 

[Pudu]

I'm skeptical... it's certainly not easy either way (especially without feeback), but when you modulate the supply to such a degree you're going to get a great deal of AM-PM due to nonlinearities in the FET Coss. And at some point VDS will be so low that the FET is going to be operating in its triode region, which will cause lots of AM-AM distortion.

The problem with fluctuating FET capacitances gets much smaller when using FETs that have small capacitances to start with. I'm making my tests with RD16HHF1 FETs, and the final project is planned around an MRF1K50N. Both of these have much lower capacitances, relative to operating impedance, than the FETs of many years ago.

The DSP platform and software I'm using supports adaptive predistortion for both amplitude and phase. I expect that I can use it to improve the final result, but for the moment I'm trying to optimize the system without using predistortion. Anyway the predistortion correction in this system is effective only over a 40kHz bandwidth.

Unless you're also varying the Vgs bias, which opens up a whole other can of worms....

That's not planned for now...

If you want to see some more about this project, you are welcome to visit its web page. Spectrograms of the first results I obtained, still using a relatively poor PWM, are included on that page. Since then it got better. Mostly the last quarter of the page is relevant to this thread:

https://ludens.cl/Electron/SDR/redpitayasdr.html

Never had this problem, personally. The system draws a vacuum to keep things nice and tight.

I see. I make a sandwich with the board and the artwork, between two glass plates held together by rubber bands. The glass isn't absolutely flat, so specially with larger boards the artwork can separate from the board by maybe 0.2mm. With point source illumination this doesn't matter much, but when I have tried to use low-pressure UV tubes, I got into trouble.

I started using tracing paper based on advice I found on this website. Like it says, get the heaviest stuff you can find to avoid crinkling in the printer. The nice thing about vellum vs transparency is that the ink won't bead up due to surface tension.

I will try. But I don't have any trouble with ink beading on the film I use. This film has an ink-absorbing coating, much like photo paper.

I don't see any reference to a FF in the datasheet for the LM5113. Though I certainly wouldn't be surprised if there is one in the high side level shifter.

Yes, the diagram of the chip's internals just shows a block for the level shifter, without any detail. All the datasheets for such high-side drivers I have seen that detail what's inside that block show a pulse-forming circuit that creates ON and OFF pulses, a capacitive pulse transmission over the isolation barrier that's separate for the ON and OFF pulses, and then a flipflop on the high side that gets these pulses at its set and reset inputs. So I'm simply assuming that the LM5113 isn't any different. If it was, then the datasheet would probably advertise that.

In such a circuit the pulse-foming circuit needs to make sure that the set and reset pulses can never be so close together that the reset is lost. So some chips internally enforce a minimum ON time, and if the input goes ON for less time, that pulse simply gets lost. That's the case with the IR2110 and similar drivers. If such a protective circuit isn't present in the chip, the user must enforce a minimum pulse length, and then it must be specified in the datasheet. And that is the case with this chip.

Anyway, the current datasheet from TI says in red letters "Not recommended for new designs."

I believe I was the first one to report that silicon but in the LM5113, since then they've modified the datasheet to mention it, and released a new revision (LMG1205). But the new one is only available in a DSBGA package which even I don't want deal with....

I only ever heard of this IC through your post... it never crossed my life before. It seems that nearly every piece of equipment that needs such a high/low side driver uses the IR2110 or one of its siblings. And those do work very well, as long as very short pulses or short delays aren't needed, and they are still going strong.

OK. Back to work now. I have been optimizing both my PWM and the driver circuitry, and things are improving, but I always think I can slim down those pulses just one more nanosecond... Let's call it Electronic Anorexia! :wink: