>> The design is here:
>>
> picasaweb.google.com/schouse/Electronics/photo#5094485161260312626
>
> Hi Mohit,
>
> I doubt very much this circuit will work at 80KHz.
> You'll need much stronger turnoff than a 10K resistor
> on the gates of the MOSFETs.
>
> Also I question your use of the entire N-channel part
> of the circuit - throw it out and use a single
> schottky rectifier there.
>
> Next is your output filter - there's way too much
> capacitance. You may not need very quick reaction
> time, but it's a waste either way. Try 22 uH and 2.2
> uF. Then you can use a ceramic capacitor. Your ripple
> will be low enough. Use a bobbin ferrite inductor -
> shielded preferably.
>
> Use a P-channel MOSFET with lower resistance. You
> don't need the 100 volt rating of the IRF9540n.
>
> Cheerful regards,
>
> Bob
>

> Driving a Peltier cooler which I thought could be treated as a  
> current sensitive process ?
> Why does ripple matter ?  In fact, why not have just one switch and  
> and a method to
> measure total current flow ?  The commercial SMPS is perhaps overkill  
> in terms of cost.
>
> Cedric
>
> On Aug 3, 2007, at 9:49 AM, Bob Blick wrote:
>
>
>> The design is here:
>>
> http://picasaweb.google.com/schouse/Electronics/
> photo#5094485161260312626
>
> Hi Mohit,
>
> I doubt very much this circuit will work at 80KHz.
> You'll need much stronger turnoff than a 10K resistor
> on the gates of the MOSFETs.
>
> Also I question your use of the entire N-channel part
> of the circuit - throw it out and use a single
> schottky rectifier there.
>
> Next is your output filter - there's way too much
> capacitance. You may not need very quick reaction
> time, but it's a waste either way. Try 22 uH and 2.2
> uF. Then you can use a ceramic capacitor. Your ripple
> will be low enough. Use a bobbin ferrite inductor -
> shielded preferably.
>
> Use a P-channel MOSFET with lower resistance. You
> don't need the 100 volt rating of the IRF9540n.
>
> Cheerful regards,
>
> Bob
>

> I agree with Bob's comments here and in his next related email (wonders
> never cease :-).
>
> I'd like to add a little more.
> I may add a few parts from Bob/Mohit's next interchange.
>
> IF you don't need the Peltier to be electrically grounded then you can
> use an N Channel "low side" FET switch which generally gives you better
> specs (especially Rdson) per $.
>
>> I doubt very much this circuit will work at 80KHz.
>> You'll need much stronger turnoff than a 10K resistor
>> on the gates of the MOSFETs.
>
>>> What do you suggest? Should I be opting for
>>> dedicated MOSFET drivers?
>>
>> Either a much lower resistance(470 ohms), a push-pull
>> circuit, or a driver IC.
>
> Even 470 ohms is a bit high if you were using a resistive only drive.
> Without looking up FET spec, gate capacitance is liable to be in the nF
> range so the gate time constant per nanofarad would be about 0.5 uS
> meaning actual switching time could be 1 uS + per nF. This is too much
> to degrade the circuit by when there are other problems which are harder
> to address.
>
> A very simple and cheap gate driver is as per attached gif.
> This shows an N Channel low side drive but would be OK for the P Channel
> circuit you are using. The high side drive is via an emitter follower
> that does not saturate and low side drive is via a saturating open
> collector with less pull down (here = turn on). For more low side drive
> D1 can be replaced by a PNP transistor with base to Q8 base, emitter to
> Q8 emitter and collector to ground. A very low cost, mind bogglingly
> simple but effective circuit.
>
> The gate-source zener diode shown here is very highly recommended in
> real world designs (although some competent designers will swear they
> are unneeded and others will tell you that they can cause oscillation
> problems. I have had total success using them. Locate very close to FET
> with as short a lead length as possible.
>
>> Also I question your use of the entire N-channel part
>> of the circuit - throw it out and use a single
>> schottky rectifier there.
>
> I agree.
> For REALLY high current systems - say 100A up, a synchronous FET as
> shown here will outperform a Schottky if you are willing to spend the
> effort of designing it to do so. For currents around 20A you need a
> really good FET to do better than the Schottky AND you must pay
> exquisite care to your drive timings and levels so that it's a diode
> when you want it to be and not one when you don't. The saving grace is
> (or may be) the FET body diode which "steps in" if you get it wrong, or
> id switching fails for any reason, but with much higher losses both due
> to voltage drop and charge storage issues during switching.
>
> In most real world situations up to tens of amps a TO220 Schottky diode
> of appropriate rating is liable to be cost competitive with a FET of
> equivalent capability and have none of the switching and control issues.
>
> If you do insist on using a synchronous FET rectifier as shown then the
> gate drive circuit will need much improvement. It is essential that the
> FET is turned on and off reasonably precisely. Unless you went severely
> out of your way to non-overlap the switching waveforms the 10k pulldown
> shown here would leave your FET on or partially on when your pass FET is
> turned on and magic smoke can be expected to be released.
>
> Using a copy of the attached gate driver circuit here also would work
> wonders. Cost is very very low.
>
> If you must go to 80 kHz or above (and I'm not sure why you'd need to)
> watch skin depth in the inductor winding. Use of a "Litz wire" winding
> (simply multiple turn s in parallel) will help deal with skin effect and
> make winding easier. Even copper crow bars are quite hard to wind coils
> with :-).
>
> You mention difficulties sourcing parts. From your ad I assume that you
> are in India. Digikey has reasonable prices for high capacity Schottky
> diodes and other FETs but importing these may be problematic. Digikey
> purport to operate an Indian front end http://www.digikey.co.in/ but I
> suspect this may be about as real as their Taiwanese equivalent :-(.
> Certainly, any component searching takes you directly to a suspiciously
> US looking page in $US.
>
> If you do find it advantageous to pursue a synchronous rectifier then
> you will need a superior driver plus having a lower rated Schottky there
> as well (as I think someone else suggested) to handle the switching
> edges if you don't get things quite right. This almost constitutes diode
> abuse :-) (thump the Schottky solidly at each switching edge, let it
> recover for the majority of each cycle) but seems to be what other
> people do. [I have a cheapo Chinese sourced laptop 12V-16V converter and
> that's what they do too][It died, but not the diode].
>
> Core selection and inductor design is a far blacker art than may be
> apparent if you want to do it efficiently. The alternative is to use a
> far larger than needed core that always runs cool with large diameter
> wire. If you want an optimally sized and priced core it's not quite so
> easy unless you know secret short cuts. Such as ...
>
> I recommend that you look at the Micrometals website http://www.micrometals.com
> and download their core calculator. It deals with nice things like
> partial core DC saturation effects on inductance, core thermal lifetimes
> (a non trivial matter), skin effect, core materials and size and more.
> You can arrive at some quite sobering realisations using this
> calculator. Designing a semi saturated core, as most buck regulators
> will have, is a far less trivial task than might be expected, except
> when you use this calculator :-). Micrometals cores are amazingly cheap
> for what you get and have superior thermal performance to wannabe clones
> of theirs. uM have trademarked the core colour codes but that just gives
> the Asian clone makers something to make their products look more
> authentic :-). [[I have no financial arrangements whatsoever with uM
> except as a happy customer]]. Their cores are powdered iron (not
> ferrite) and have some amazingly good characteristics per dollar and per
> size compared to what one may expect from ferrite in many cases. [[One
> day I hope to make some of my own using local beach ironsand "just for
> fun", but until then I'm very happy with their products.]]
>
>> Next is your output filter - there's way too much
>> capacitance. You may not need very quick reaction
>> time, but it's a waste either way. Try 22 uH and 2.2
>> uF. Then you can use a ceramic capacitor. Your ripple
>> will be low enough. Use a bobbin ferrite inductor -
>> shielded preferably.
>
> I imagine the over filtering may be aimed at a low ripple for the
> Peltier. If you care muchly then a multistage filter may help. Again, a
> micrometals (or other) powdered iron core can do this well AND the
> toroidal core is inherently shielded (although perhaps not quite as well
> as a homogeneous ferrite core is for low mu cores).
>
>> Use a P-channel MOSFET with lower resistance. You
>> don't need the 100 volt rating of the IRF9540n.
>
> Agree.
> But, as above, if you can run the Peltier above ground an N Channel
> device will be even better.
>
> I believe (and may be wrong) that the main reason for running Peltiers
> on low ripple is that the device is effectively thermally cycled at each
> current cycle and the thermal stresses lead to degradation of the
> material. If this is so, then it follows that a ripple frequency well
> above any thermal response frequency, may be OK. But, don't take my word
> for it.
>
> Ripple can be greatly reduced at a small cost in efficiency and $ by
> running a tracking linear regulator on the buck regulator output. This
> can be as simple as a FET and probably a very cheap opamp which derives
> a reference point just below the lowest ripple voltage and removes any
> voltage input above that point. Half an LM358 and a suitably current
> rated FET with low Vdsmax should do. (The FET floats and never "knows"
> what the actual supply voltage is. Having enough gate drive voltage may
> be an issue if Vout is close to Vin.
>
> Enough for now :-)
>
>
>
>        Russell
>
>
> ------------------------------------------------------------------------
>

> Mohit Mahajan wrote:
> .
> > Actually the peltiers are driven by an H-bridge, so that they can cool
> > or heat (user menu driven). The DC-DC converter I'm trying to build will
> > power the H-bridge. The H-bridge will only reverse the voltage across
> > the peltiers. No PWM here. Because if there was PWM in the H-bridge I'd
> > have to do the filtering there, with capacitors across peltier and to
> > ground that could take voltage both ways. I'm really not able to
> > (easily/cheaply) find large value ceramic capacitors suitable for such a
> > filter. I'm also not comfortable with reversing voltage across
> > capacitors even if they were ceramics.
>
> You should really look at how other people do it, and not reinvent the
> wheel. Doing PWM with the H-bridge is the right way. After the H-bridge,
> use two inductors(or a dual inductor) and two capacitors. Each capacitor
> returns to ground, so it does not need to be bipolar.
>
> You mentioned the TC4420 driver, those are good. If you get the one with
> both input polarities it's easy to wire them to logically prevent
> "both-on" situations.
>
> I think either Maxim or LT have documents about Peltier drivers to
> support their specialized parts. Although I am not recommending their
> parts, I do recommend you read their documentation and look at their
> component and topology choices.
>

>
> Mohit Mahajan wrote:
>> Hi Bob,
>>
>>  > Doing PWM with the H-bridge is the right way. After the
>>  > H-bridge,use two inductors(or a dual inductor) and two capacitors.
>>  > Each capacitor returns to ground, so it does not need to be bipolar.
>> I just checked this:
>> http://www.ti.com/litv/pdf/slua202a
>> You mean something like figure 3 on page 4?
>
> Yes.
>
>>  > You mentioned the TC4420 driver, those are good. If you get the
>>  > one with both input polarities it's easy to wire them to
>>  > logically prevent "both-on" situations.
>> The PIC takes care of this. It has the Extended CCP with a half-bridge
>> output that could be used to drive a full-bridge circuit with
>> programmable dead band between the switching.
>
> Sure, still no reason not to have short-circuits impossible - it costs
> you no extra parts. You'll be glad you did it when you make a
> programming error during a debug session.
>
> Cheers,
>
> Bob

>>> A very simple and cheap gate driver is as per attached gif.
>
>> I would be very tempted to use Microchips gate drive chips.
>> MCP1401/2 inverting/non-inverting gate driver in SOT23-5 package
>> 500mA drive capability.
>
> The temptation may be easy to resist :-).
> Digikey know not of the product, nor do the other 20 sellers at
> http://www.findchips.com :-(
>
> The products sound useful if minimum component count is desired but
> performance wise they sound about as capable as the simple circuit I
> posted. It's not "my circuit" -  it has been around auld lang syne and
> it may even have been used by Noah for his FET drivers on the Ark's
> ventillator motors (of which he probably had quite a few).
>
> Microchip show samples (of the IC, not of Noah's cct) and you can
> follow the "buy online now" link to a putative Indian seller, but if
> you don't log in it's not obvious who the seller is.
>
> Microchip show budgetary price as $US0.44 in volume and, as the
> circuit as proposed would ideally have two of these that's probably
> $US1 in moderate volumes for drivers alone.
>
> Given Mahit's general comments about pricing and what he could obtain
> where he lives, I suspect that my sub $US0.10 solution (possibly
> little more than that for two of them) may prove attractive.
>
> The discrete circuit does not have, as shown, some of the specs of the
> uChip parts but it would be very satisfactory in this application.
> I've used it as shown for 20 kHz 8 bit PWM Addition of a speedup
> capacitor alone increases bandwidth into the low MHz region. Output
> voltage range is limited only by the transistors used and drive
> current capability can be "butal" if you want it to be. (For "brutal"
> you'll need to up the transistors from the BC337/BC327 I usually use).
> As is 500 mA is easy. Input levels are by design. Logic input will
> withstand whatever voltage the diode is rated at that you put on the
> input if you choose to put one there :-). Otherwise still quite a lot.
>
> The capacitive drive rise times are essentially another way of stating
> the current drive capabilities and 34nS to raise 1000 pF to 10V say is
> equivalent to about 340 mA constant current so is consistent with the
> 500 mA drive claim.
>
>
>         Russell
>
>
>> MCP1401/2 inverting/non-inverting gate driver in SOT23-5 package
>> 500mA drive capability.
>> logic level input.
>> output supply range 4.5-18V
>> high capacitive load drive capability - 470pF in 19nS, 1000pF in
>> 34nS (typ)
>> matched rise and fall times.
>> latch-up protected, will withstand 500mA reverse current on output.
>> logic input will withstand negative swing to -5V
>