Rich, I'm not directly involved in the design of our power converters, but do work with the guys that are and have some exposure to their problems. For reference, the equipment is "Telecom" AC-DC "rectifiers" converting AC mains to DC output at nominally 48V. (actually 53V or so). This is accomplished by rectifying the input voltage, boosting it to about 420V with pfc and then converting the DC back down to 48V using an "LLC" converter and a transformer. The boost stage operates at about 70kHz and the DC- DC in the 150kHz - 400kHz range. Power levels are typically 1-3kW and efficiency about 92%. For both the boost and DC-DC stage we use N channel MOSFETs, the DC-DC ones having a DC rating of 600V. (The boost ones can be higher - 800V?). These are prettty heavy-duty items, packaged in T0-247 style packages. Rds_on is around the 100mOhm area. AFAIK we have never used P channel MOSFETS in complementary style output arrangements but have used parallel MOSFETS in the boost stage. Matching does not appear to be too much of an issue due to the positive temperature coefficent of the MOSFET resistance.. A good source of datasheets / part numbers (& application notes) would be the IR, ST, etc. websites. These can also supply data on suitable drivers for the MOSFETs, including high side capable drivers which use a capacitively coupled bootstrap circuit to provide a gate drive supply above the input voltage. Possibly something like the STW26NM60 could be a starting point ? (Taken from one of our older units so their may be a more modern, improved version available). Richard P 2008/8/11 Rich : > Richard, Thank you for the excellent advice. I am so pleased that there are > many persons on the list that are more knowledgeable than I. I did not > design the transformer. Years ago I designed a lot of E - I (refers to > shape) laminated core transformers, computing the window, core material wire > size (round or square) and so on. But I did not design high frequency > torroid and I would have to invest some significant time to study up, so I > had a magnetics engineer design and build the transformer. From what you > say, I need to focus more on power FETs. I can parallel them for the the > current load. > > I have had experience designing oscillators and power amplifiers including > high frequency power amplifiers. One caveat here (100 KHz) is controlling > crossover distortion, which, of course, is more challenging at higher > current loads and unstable heating. So some basic design principles are in > order. These currents at this voltage is a bit beyond what I normally > design. I have designed high current or high voltage but not both together, > so I need to identify the semiconductors first. > > It seems that you recommend power FETS, and that sounds plausible. I will > try to find power FETs that I can match for parallel operation. Unlike > bipolars in parallel where base resistors are required, that is not the case > with FETs; no gate resistors are needed because the gates do not draw > current, so the design is easier. > > I have not been able to find a 300 Volt @ 16 Amp (minimum spec) device. I > would prefer a 400 V @ 20 Amp N- channel and matching P-channel. Otherwise > I will have to use 2 N-channel FETs and drive them with a phase inverter on > the lower section. It would be so much easier if there is a COS pair at the > 300 V@ 16 Amp minimum, spec. > > ----- Original Message ----- > From: "Richard Prosser" > To: "Microcontroller discussion list - Public." > Sent: Sunday, August 10, 2008 4:04 PM > Subject: Re: EE > > >> Rich, >> >> 1. You need to include ferrite losses in your calculations. The loss >> will be highly dependent on the ferrite grade used, along with the >> drive level (how close you are to core saturation). At 100kHz you may >> also want to look at Litz wire. (Since you're using a toroid, I'm >> guessing the core is not gapped. Core saturation and material magnetic >> tolerances need to be considered.) It's not a trivial design task. >> >> 2. The transformer _may_ be more efficient with a sinewave input but >> the linear sinewave generating process will not be. For overall >> efficiency a squarewave generation followed by a tuned winding should >> be OK if you get things right. A tuned winding may give you soft >> switching capability and lower switching losses. >> >> 3. 100kHz is starting to push things for IGBT switches. MOSFETS are >> more common at this frequency. We use 600V types to switch a 420V bus >> in a half or full bridge configuration at power levels to about 2kW. >> Frequency 150kHz - ~400kHz output. Highside MOSFET driver chips for a >> squarewave are very easy to locate. (PWM can be more difficult if you >> want continuous 0 - 100% pwm). >> >> 4. Overall, a lower frequency is likely to be easier to design & get >> working but will take more space.. >> >> Richard P >> >> >> 2008/8/11 Rich : >>> Hi Sean: >>> Thank you for the comments. The project application is proprietary but I >>> can discuss the power supply. The transformer is designed for 220 VAC @ >>> 100 >>> KHz primary and the secondary delivers 3000 VAC 100KHz @ 1 Amp. >>> It is not a laminated core transformer or a "Metglass" core but a ferrite >>> core. It is a torroid. I want to drive it with a sine curve because it >>> is >>> more efficient. I had the transformer designed for 100 KHz operation, >>> not >>> 60 Hz, because the package must be kept to a small footprint and mass, so >>> I >>> thought to reduce the magnetics. >>> >>> My application does not require a ferroresonant design, and that is not >>> in >>> my power supply design specification. As for primary impedance, the 16 >>> ohms >>> suggested by Russell is perhaps unlikely. At 50 Hz or 60 Hz you can use >>> Ohms law to approximate the primary or secondary impedance by Z=Vp/Ip = >>> 220/13.5 = 16+ ohms. But at 100 KHz the skin effect modifies the winding >>> resistance R, and the impedance is more closely approximated by the >>> square >>> root of [R + (2 pi f)sqLsq] or Zp= sqr root of Rsq + Xsq. It is easier >>> to >>> just measure the impedance. I apologize for the equations. >>> >>> Since the volt amps of the secondary equal the volt amps of the primary, >>> the >>> primary current will depend on the current that the secondary draws. >>> VIp=VIs. Given that Is = 1 Amp and Vs= 3KV, solving for Ip I get about >>> 13.6 >>> amps. So I think I am stuck with the primary drive that I have >>> described, >>> unless I am not seeing this correctly. >>> >>> About creating a 50 ohm tank, that is not a hard spec. My thinking >>> proceeds >>> as follows. Your critique is welcome. If I drive the transformer with a >>> square wave it will work, not as efficiently, but it will work. However, >>> I >>> will also get all of the even harmonics. That is not good, even though >>> they >>> could be filtered. The output impedance of, say bipolars, will be a >>> function of the junction characteristic and the output circuitry and will >>> probably be lower than the transformer primary impedance. Normally, >>> driving >>> with a lower impedance is desirable but in this (what I consider higher >>> power application) I am expecting some aberration of the sine wave that >>> is >>> driving the transformer. The matched impedance should mitigate the >>> aberration of the curve. >>> >>> I have not designed anything with IGBTs. I can design a circuit with >>> bipolars to produce a fairly clean sine wave. But the problem with >>> bipolars >>> is the low beta. That means the base drive is going to be high. I can >>> use >>> power Darlington, but I will have to accept that they will run a bit >>> hotter >>> due to the higher VCEsat for darlingtons. HexFets might work because the >>> multiple FETS in parallel reduce the on resistance. In a switching >>> application I could expect ringing from the internal contact wire but >>> this >>> is a linear application and using a HexFet might not be a good choice. I >>> thought of using a complementary symmetry N-Channel and P-Channel power >>> FET, >>> if I could find a pair without resorting to parallel operation, which >>> could >>> work >>> >>> I am in the early stages of the design so I can be flexible. The driving >>> factor is the application, which has defined the transformer. So the >>> circuit needs to service the transformer. I will not commit to a >>> prototype >>> circuit until everything is worked out on paper. I will appreciate any >>> caveats that are pointed out because I cannot say that I have never done >>> anything stupid. In fact I better not embarrass my self by saying how >>> many >>> stupid mistakes I have made, The only redeeming factor is that I am >>> careful >>> to make them once. >>> >>> Thank you for your input, Sean. I look forward to your response. >>> >>> >>> >>> ----- Original Message ----- >>> From: "Sean Breheny" >>> To: "Microcontroller discussion list - Public." >>> Sent: Sunday, August 10, 2008 2:52 AM >>> Subject: Re: EE >>> >>> >>>> Hi Rich, >>>> >>>> Is there a reason why you are using such a high frequency to transfer >>>> power? What does the secondary voltage/current look like? What kind of >>>> core material does your transformer use? >>>> >>>> When the coupling between primary and secondary is close enough to 1 >>>> and the impedance of the load is significantly smaller than the >>>> inductive reactance of the windings, then a transformer looks "ideal" >>>> in that the input impedance looks like the square of the turns ratio >>>> times the load impedance. >>>> >>>> I do not think that you want to create a 50 ohm load - you have >>>> already specified the input voltage and current so your input >>>> impedance is determined (do you really have a setup which will draw a >>>> constant 14A at 230V?). >>>> >>>> I also doubt that you want to create an LC circuit. That is usually >>>> done when filtering is desired and the coupling between pri and sec is >>>> much less than 1 (so that the Q of the LC circuit isn't too spoiled by >>>> the load resistance). In "wireless" power transmission it is also >>>> sometimes done to get around some of the effects of having poor >>>> coupling coefficient (due to distance between the pri and sec). >>>> >>>> IGBTs or any other switching element for that matter will NOT of >>>> itself create a sine wave. If you go that route, you will either be >>>> driving the transformer with a square wave or you will have to do PWM >>>> which is of a significantly higher freq than the sine wave you want to >>>> generate. You could also go back to making an LC circuit and create a >>>> class C amplifier, but that will likely make things bulky and somewhat >>>> inefficient. >>>> >>>> Could you say a bit more about your application? The focus on >>>> transmitting kilowatts at 100KHz seems odd. >>>> >>>> Sean >>>> >>>> >>>> >>>> On Sun, Aug 10, 2008 at 1:23 AM, Rich wrote: >>>>> I wonder if I am mistaken here. The transformer primary is an >>>>> inductive >>>>> element. True, there is distributed coil capacity, and a frequency >>>>> component that makes the system complex. So there will be some >>>>> "natural" >>>>> resonance that may not be at 100 KHz and may have a very low Q, >>>>> regardless >>>>> of the fact that the transformer was designed to operate at 100 KHz. >>>>> But >>>>> by >>>>> adding the properly selected components, could one not design starting >>>>> from >>>>> the LCR characteristic of the primary winding to create a tuned circuit >>>>> at >>>>> 100 KHz and 50 ohms Z? Also, would it be true that if the Q is high >>>>> the >>>>> stability of the oscillator must be high in order to stay within the >>>>> resonant bandwidth? So, a high Q may not be desireable and the R >>>>> component >>>>> would be a factor? All comments, criticisms, "Oh what stupidity" >>>>> comments >>>>> are welcome. >>>>> >>>>> >>>>> ----- Original Message ----- >>>>> From: "Vasile Surducan" >>>>> To: "Microcontroller discussion list - Public." >>>>> Sent: Saturday, August 09, 2008 9:07 AM >>>>> Subject: Re: EE >>>>> >>>>> >>>>>> Nice way of computing... A square signal bumped with a positive glich >>>>>> at the end of the rising edge will increase the current with 10-20% at >>>>>> the same secondary load. >>>>>> Does the primary impedance seen by the driver will be different ? >>>>>> >>>>>> Vasile >>>>>> >>>>>> On 8/9/08, Apptech wrote: >>>>>>> > Where did you get 16 ohms from? >>>>>>> >>>>>>> R = V/I. >>>>>>> The driver sees the load reflected via the transformer. If >>>>>>> 230 VAC causes 14 amps to flow then the AC is seeing 230/14 >>>>>>> ~= 16. >>>>>>> >>>>>>> In the absence of load the actual transformer impedance will >>>>>>> be seen but it will very usually b swamped when loaded. >>>>>>> >>>>>>> >>>>>>> Russell >>>>>>> >>>>>>> >>>> I have to drive a transformer primary at 230 VAC @ 14 >>>>>>> >>>> Amps @ 100KHz. I have not yet measured the primary >>>>>>> >>>> impedance, but I will. >>>>>>> >> >>>>>>> >> The impedance you see will be about 16 ohms - ie the >>>>>>> >> load, >>>>>>> >> more or less regardless of the transformer's unloaded >>>>>>> >> impedance. >>>>>>> >> >>>>>>> >> >>>>>>> >> Russell >>>>>>> >>>>>>> -- >>>>>>> http://www.piclist.com PIC/SX FAQ & list archive >>>>>>> View/change your membership options at >>>>>>> http://mailman.mit.edu/mailman/listinfo/piclist >>>>>>> >>>>>> -- >>>>>> http://www.piclist.com PIC/SX FAQ & list archive >>>>>> View/change your membership options at >>>>>> http://mailman.mit.edu/mailman/listinfo/piclist >>>>> >>>>> -- >>>>> http://www.piclist.com PIC/SX FAQ & list archive >>>>> View/change your membership options at >>>>> http://mailman.mit.edu/mailman/listinfo/piclist >>>>> >>>> -- >>>> http://www.piclist.com PIC/SX FAQ & list archive >>>> View/change your membership options at >>>> http://mailman.mit.edu/mailman/listinfo/piclist >>> >>> -- >>> http://www.piclist.com PIC/SX FAQ & list archive >>> View/change your membership options at >>> http://mailman.mit.edu/mailman/listinfo/piclist >>> >> -- >> http://www.piclist.com PIC/SX FAQ & list archive >> View/change your membership options at >> http://mailman.mit.edu/mailman/listinfo/piclist > > -- > http://www.piclist.com PIC/SX FAQ & list archive > View/change your membership options at > http://mailman.mit.edu/mailman/listinfo/piclist > -- http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist