How to calculate UC3842 components?

[technix]

I am in the middle of designing M103v2 power supply, but the cannot find a unified set of formula for UC3842. Help!

Attached is the schematic of one of my two UC3842 units. This is the fixed auxiliary rail regulator, outputting +/- 12V for analog section and +5V for digital section. The input voltage (at VIN net) is 400V, from a UCC28019-based active PFC circuit.

Questions:
1) How to calculate the components?
2) How to make this regulator adjustable without throwing power ratings out of the window? The primary regulator have an output range of 2.5-32.5V at 0-10A, with output power 50W maximum.

[T3sl4co1l]

R27, C16: Oscillator frequency. Arbitrary. Pick a frequency suitable for the project.  Typically 100-130kHz.

R26, R28, R32, C18: Opto feedback path. C18 = 0, resistors designed for proper voltage range.  The topology is wrong.  Normally, you wire the internal error amplifier as gain = -1, which would be done with resistors like so:



Which leads to the next parts,
D15: omit (replace with short), add resistor across opto (pins 1-2), add R+C across IC4 ("cathode" to "adj").  This provides compensation.  The supply is hopeless without the RC.

There's no aux winding.  Will R29 be able to supply operating current?  Normally, controller VCC is sourced from an additional winding on the transformer.  This also removes any need for limiting supply or output voltage (D9-D11 can be removed completely), and significantly improves efficiency (R29 can be about 100k 1W).

R30: the UC3842 output stage can supply about 1A, so from VCC around 15V, the resistance should be about 15V/1A = 15 ohms.  It can be smaller to improve speed slightly, or larger to reduce EMI.

R34 = 1V / Ipk

Ipk is determined by operating frequency, supply voltage, and output power.  If Vin = 300V and P = 50W, then Iin = 0.167A DC.  At 50% duty cycle, Iin = 0.333A average during the on-time.  The on-time is linearly rising, so Ipk is twice this, or 0.667A.  Add 20-40% more for tolerances, and you get maybe 0.8A, or 1.2 ohms.

R33 = 10k or optional.

R31, C17: typically 1k and 220pF.  Time constant determined by transformer's stray capacitance and transistor's gate capacitance.

Q6 drain: needs a snubber.  dV/dt preferred.  C = Ipk / (2*Vin / tr).  tr is the desired rise time, usually similar to the gate time constant (for IRF840, Cg ~= 6nF, and if R30 = 10 ohms, then tau = 60ns).  100pF would be typical here.  The diode can be UF4007, and the resistor should give an RC time constant shorter than the minimum on-time.  10kohms would be fine.

The RC also serves as damping for the unloaded ringdown of the transformer.  The RC values can also be chosen based on best damping there.

The transformer should have a primary inductance of L = Vin / (2*Ipk*Fosc), or about 2.2mH for this case.

The unloaded ringdown will be around 200pF (from Q6 Coss and reflected equivalent D12-D14 Cjo) and 2.2mH or 240kHz, which is pretty low, maybe low enough not to care.  The impedance is sqrt(2.2mH / 200pF) = 3.3kohm, so the snubber R being near this value will give better damping.

A smaller transistor could also be chosen, especially a newer one with significantly reduced Ciss and Coss.

Lastly, output side: don't split grounds. You're inviting ground loop and worse EMI problems than whatever you imagine you're trying to avoid in the first place.  Keep them all on common ground, and filter that ground.

With common ground, you can also jointly regulate supplies.  Change R35 to 20k, and add an 80k resistor from IC4 "ADJ" to C23 (+12V, but before the LC filter).  As shown, only the 5V supply is regulated, which means the 12V supply voltage will go crazy if the loads are mismatched.

Also, the LC filtering is not without scrutiny.  You are making a C-L-C resonant circuit, which needs to be heavily damped, otherwise the supplies will likely be worse (due to ringing) than the noise you filter from them.  The critical parameters are again R, L and C.  The two capacitors (C23 and C25; C24 and C26) act in series, so have half the capacitance and twice the ESR of a single capacitor (assuming identical components).  You might have 220uF and 1 ohm in this case; if the inductor itself also has an ohm or more DCR, then sqrt(220uH / 235uF) will be smaller than the total ESR+DCR, and the network will be overdamped (good).

Using less C, more L, or higher quality L and/or C (e.g., very low ESR aluminum polymer) would lead to problems.  The values as shown, assuming reasonable component choice, are probably good.

Tim

[technix]

R27, C16: Oscillator frequency. Arbitrary. Pick a frequency suitable for the project.  Typically 100-130kHz.

How good is 450kHz?

R26, R28, R32, C18: Opto feedback path. C18 = 0, resistors designed for proper voltage range.  The topology is wrong.  Normally, you wire the internal error amplifier as gain = -1, which would be done with resistors like so:



Which leads to the next parts,
D15: omit (replace with short), add resistor across opto (pins 1-2), add R+C across IC4 ("cathode" to "adj").  This provides compensation.  The supply is hopeless without the RC.

Thanks for the info here.

There's no aux winding.  Will R29 be able to supply operating current?  Normally, controller VCC is sourced from an additional winding on the transformer.  This also removes any need for limiting supply or output voltage (D9-D11 can be removed completely), and significantly improves efficiency (R29 can be about 100k 1W).

I want to live without the aux winding. One less winding to calculate...

R30: the UC3842 output stage can supply about 1A, so from VCC around 15V, the resistance should be about 15V/1A = 15 ohms.  It can be smaller to improve speed slightly, or larger to reduce EMI.

Thanks.

R34 = 1V / Ipk

Ipk is determined by operating frequency, supply voltage, and output power.  If Vin = 300V and P = 50W, then Iin = 0.167A DC.  At 50% duty cycle, Iin = 0.333A average during the on-time.  The on-time is linearly rising, so Ipk is twice this, or 0.667A.  Add 20-40% more for tolerances, and you get maybe 0.8A, or 1.2 ohms.

Great. Can I use 2512 chip resistors here? Also Vin = 400V not 300V, as I have an active PFC.

R33 = 10k or optional.

R31, C17: typically 1k and 220pF.  Time constant determined by transformer's stray capacitance and transistor's gate capacitance.

Q6 drain: needs a snubber.  dV/dt preferred.  C = Ipk / (2*Vin / tr).  tr is the desired rise time, usually similar to the gate time constant (for IRF840, Cg ~= 6nF, and if R30 = 10 ohms, then tau = 60ns).  100pF would be typical here.  The diode can be UF4007, and the resistor should give an RC time constant shorter than the minimum on-time.  10kohms would be fine.

The RC also serves as damping for the unloaded ringdown of the transformer.  The RC values can also be chosen based on best damping there.

Can I whack an antiparallel diode there (1N5408 maybe) instead of a snubber? Just like buck converters?

The transformer should have a primary inductance of L = Vin / (2*Ipk*Fosc), or about 2.2mH for this case.

The unloaded ringdown will be around 200pF (from Q6 Coss and reflected equivalent D12-D14 Cjo) and 2.2mH or 240kHz, which is pretty low, maybe low enough not to care.  The impedance is sqrt(2.2mH / 200pF) = 3.3kohm, so the snubber R being near this value will give better damping.

A smaller transistor could also be chosen, especially a newer one with significantly reduced Ciss and Coss.

Smaller transistor = $$$. IRF840, being an older but common design, is really, really cheap. Also since this transistor is also used on the main converter as well as the PFC converter, it is one component less to concern for me.

Lastly, output side: don't split grounds. You're inviting ground loop and worse EMI problems than whatever you imagine you're trying to avoid in the first place.  Keep them all on common ground, and filter that ground.

With common ground, you can also jointly regulate supplies.  Change R35 to 20k, and add an 80k resistor from IC4 "ADJ" to C23 (+12V, but before the LC filter).  As shown, only the 5V supply is regulated, which means the 12V supply voltage will go crazy if the loads are mismatched.

Then how do I do it with three voltages, each with drastically different current ratings?

+5V rail needs to be set at 5.2-5.3V and capable of a maximum output of 3A (as it may be used to charge an iPad and power an Raspberry Pi at the same time, which can pull up to 2A and 1A respectively)
+12V drives op amps and a fan, requires a maximum of 1A (fan + op amps)
-12V is only used by op amps so up to 200mA is required.

Also, the LC filtering is not without scrutiny.  You are making a C-L-C resonant circuit, which needs to be heavily damped, otherwise the supplies will likely be worse (due to ringing) than the noise you filter from them.  The critical parameters are again R, L and C.  The two capacitors (C23 and C25; C24 and C26) act in series, so have half the capacitance and twice the ESR of a single capacitor (assuming identical components).  You might have 220uF and 1 ohm in this case; if the inductor itself also has an ohm or more DCR, then sqrt(220uH / 235uF) will be smaller than the total ESR+DCR, and the network will be overdamped (good).

Using less C, more L, or higher quality L and/or C (e.g., very low ESR aluminum polymer) would lead to problems.  The values as shown, assuming reasonable component choice, are probably good.

Tim

About that LC filter for analog ground, if I use 680uH L's and 47uF tantalums, will it be better than the 470uH's and 220uF aluminum I have here?

[T3sl4co1l]

R27, C16: Oscillator frequency. Arbitrary. Pick a frequency suitable for the project.  Typically 100-130kHz.

How good is 450kHz?

I wouldn't recommend it, especially not with IRF840, and especially not if you're being shy with your transformer design.

I want to live without the aux winding. One less winding to calculate...

Calculate what?  It's just a ratio.  In fact, it will be fine to use exactly the number of turns as your +/-12V windings!

Great. Can I use 2512 chip resistors here? Also Vin = 400V not 300V, as I have an active PFC.

I gave the formulas, so you can substitute the different numbers and get the correct results.

How much power is a 2512 rated for?

The power dissipation will be up to P = R * Ipk^2 / 6.  ((Neat calculus exercise for the student: show why this is true.)

Can I whack an antiparallel diode there (1N5408 maybe) instead of a snubber? Just like buck converters?

I can think of about four things wrong with that suggestion, but the most important one should be obvious from looking at the circuit...

Hint: what does primary voltage necessarily rise to, when the transistor turns off?

Then how do I do it with three voltages, each with drastically different current ratings?

+5V rail needs to be set at 5.2-5.3V and capable of a maximum output of 3A (as it may be used to charge an iPad and power an Raspberry Pi at the same time, which can pull up to 2A and 1A respectively)
+12V drives op amps and a fan, requires a maximum of 1A (fan + op amps)
-12V is only used by op amps so up to 200mA is required.

A widely variable load is probably not a good idea, for this type of power supply, and it supposedly being a "auxiliary" supply.

Joint regulation gets you the best compromise, at least among the supplies being regulated.

You can't mix the -12V in, without using an inverting feedback stage, so you'll just have to ignore it. (Or use a much more complicated feedback circuit, but I doubt it's worth it.)

Probably none of the supplies need to be very accurate, and as long as there is a modest idle load, the worst fluctuations won't be dangerous.

Op-amps don't much care about their supply voltages, as long as they remain within ratings.  If you do require precise -12V and other supplies, you will need to consider a different topology.

About that LC filter for analog ground, if I use 680uH L's and 47uF tantalums, will it be better than the 470uH's and 220uF aluminum I have here?

I gave the formula and condition, so you can substitute for those values and see whether it is a good idea or not.

I would guess no.

You also didn't specify what is "better", say in terms of dB attenuation at whatever frequency.  Since you've given no noise or filtering spec, no reason to include themm I would just as well remove them, leaving the power raw.

Tim

[technix]

R27, C16: Oscillator frequency. Arbitrary. Pick a frequency suitable for the project.  Typically 100-130kHz.

How good is 450kHz?

I wouldn't recommend it, especially not with IRF840, and especially not if you're being shy with your transformer design.

So since 100kHz is a good idea I will stick to that, slightly higher maybe? (125-130kHz)

I want to live without the aux winding. One less winding to calculate...

Calculate what?  It's just a ratio.  In fact, it will be fine to use exactly the number of turns as your +/-12V windings!

UC3842 requires at least 16V to start up but the transistor have a maximum Vgs of 20V. Goldilocks much?

Great. Can I use 2512 chip resistors here? Also Vin = 400V not 300V, as I have an active PFC.

I gave the formulas, so you can substitute the different numbers and get the correct results.

How much power is a 2512 rated for?

The power dissipation will be up to P = R * Ipk^2 / 6.  ((Neat calculus exercise for the student: show why this is true.)

2512 size is rated for 1 watt if given proper thermal design (and since I will tuck all three current sense resistors, two for UC3842 and one for PFC, under the mains side heatsink I think I have adequate thermals for it)

Can I whack an antiparallel diode there (1N5408 maybe) instead of a snubber? Just like buck converters?

I can think of about four things wrong with that suggestion, but the most important one should be obvious from looking at the circuit...

Hint: what does primary voltage necessarily rise to, when the transistor turns off?

Then how do I do it with three voltages, each with drastically different current ratings?

+5V rail needs to be set at 5.2-5.3V and capable of a maximum output of 3A (as it may be used to charge an iPad and power an Raspberry Pi at the same time, which can pull up to 2A and 1A respectively)
+12V drives op amps and a fan, requires a maximum of 1A (fan + op amps)
-12V is only used by op amps so up to 200mA is required.

A widely variable load is probably not a good idea, for this type of power supply, and it supposedly being a "auxiliary" supply.

Joint regulation gets you the best compromise, at least among the supplies being regulated.

You can't mix the -12V in, without using an inverting feedback stage, so you'll just have to ignore it. (Or use a much more complicated feedback circuit, but I doubt it's worth it.)

Probably none of the supplies need to be very accurate, and as long as there is a modest idle load, the worst fluctuations won't be dangerous.

Op-amps don't much care about their supply voltages, as long as they remain within ratings.  If you do require precise -12V and other supplies, you will need to consider a different topology.

The +5V rail always have a MCU and a BLE module on there, so it always have a few milliamps. However the load can vary all the way up to a few amps. My main op amp is TL084, can I live without a regulated power rail for that?

Also if UC3842 don't like wildly varying outputs, how to construct an adjustable mixed-mode bench power supply using it as the primary regulator?

About that LC filter for analog ground, if I use 680uH L's and 47uF tantalums, will it be better than the 470uH's and 220uF aluminum I have here?

I gave the formula and condition, so you can substitute for those values and see whether it is a good idea or not.

I would guess no.

You also didn't specify what is "better", say in terms of dB attenuation at whatever frequency.  Since you've given no noise or filtering spec, no reason to include themm I would just as well remove them, leaving the power raw.

Tim

[T3sl4co1l]

UC3842 requires at least 16V to start up but the transistor have a maximum Vgs of 20V. Goldilocks much?

In fact, maximum 17.5V turn-on!

But as long as you stay above the maximum turn-off of 11.5V, you are fine.  It only goes up to 14.5-17.5 during startup, when powered by the high value bootstrap resistor.  The aux winding keeps it in the 11.6-20V range, which is quite easily done, even with a poor transformer.

The +5V rail always have a MCU and a BLE module on there, so it always have a few milliamps. However the load can vary all the way up to a few amps. My main op amp is TL084, can I live without a regulated power rail for that?

That sounds fine. I'd be way more concerned about the MCU and BLE getting safe voltages, than the op-amps, which are rated for +/-18V.

It may turn out that tighter regulation is desired on the 5V supply, in which case you can vary the ratio of the two resistors (the one from +5 to TL431 REF, and the one from +12) to adjust the weights of each.

You may also consider a single power supply, perhaps +12V, with a +5 and -12V (or -5V if you can tolerate it) DC-DC converters following.  Or use a commercial power supply instead of the offline supply.

And we still haven't even gotten to the EMI problem, which is likely to cause errors in your analog circuit, let alone disruption of nearby radio services, and liability if this is a marketed product.

Also if UC3842 don't like wildly varying outputs, how to construct an adjustable mixed-mode bench power supply using it as the primary regulator?

You don't. UC3842 is a poor choice for such an application.  Flyback is a poor topology for such a supply, anyway.

For a wide output range, you need an auxiliary supply, which might be a single-transistor oscillator, or another flyback supply (perhaps using a simpler e.g. TOPSwitch controller).  And then you can use a more fully-featured controller, like a forward converter (TL494 being an old example).

Tim

[technix]

You don't. UC3842 is a poor choice for such an application.  Flyback is a poor topology for such a supply, anyway.

For a wide output range, you need an auxiliary supply, which might be a single-transistor oscillator, or another flyback supply (perhaps using a simpler e.g. TOPSwitch controller).  And then you can use a more fully-featured controller, like a forward converter (TL494 being an old example).

Tim

What topology to use then? TOPSwitch = $$$, but TL494 is acceptably priced. Maybe I can build both the primary tracking regulator (32.5V/10A/50W max) and the secondary regulator (providing the fixed +5V & +12V voltages) using that? It seem to me that the dual error amps in TL494 can be fantastic for my purpose here as I can use them to keep both +12V and +5V rails safe (and let -12V rail to be)

And I think I need to scale my PFC capacity up to 150W or so, and raise the primary rail power cap to 100W as my transformer should be able to handle it. UCC28019 is awesome, BTW.

I was thinking about splitting grounds so I can have the analog and digital sections entirely separated, to the point of being able to move the analog and power sections, with their respective isolated controls, off to daughterboards so I can construct a simple plug-and-play multi-channel system. Main rectifier, PFC and digital section on the backplane, output strips on the daughterboards.

[T3sl4co1l]

What topology to use then? TOPSwitch = $$$, but TL494 is acceptably priced. Maybe I can build both the primary tracking regulator (32.5V/10A/50W max) and the secondary regulator (providing the fixed +5V & +12V voltages) using that? It seem to me that the dual error amps in TL494 can be fantastic for my purpose here as I can use them to keep both +12V and +5V rails safe (and let -12V rail to be)

And I think I need to scale my PFC capacity up to 150W or so, and raise the primary rail power cap to 100W as my transformer should be able to handle it. UCC28019 is awesome, BTW.

You can afford PFC on this project, but not another flyback controller (3842 or TOPSwitch or FPS or..)..?  The controller chip itself is a small part of the total cost, anyway; you'll likely spend more on the transformer, and you need two (aux and main).

I was thinking about splitting grounds so I can have the analog and digital sections entirely separated, to the point of being able to move the analog and power sections, with their respective isolated controls, off to daughterboards so I can construct a simple plug-and-play multi-channel system. Main rectifier, PFC and digital section on the backplane, output strips on the daughterboards.

I don't see any reason that should require isolation, though.  Or if it does, perhaps each channel would need isolation?  Which would perhaps be best handled as a per-module cost (i.e., put the isolator in the module) rather than a mainframe cost (you buy N isolated channels whether you need them or not).

Isolation really stinks if you need to isolate analog signals, too, in which case you'd rather do all the isolation digitally.

And if you only have digital and analog circuits, with common grounds between each domain, but isolation between the two... How?  Why?  The only reason I can see having both in a system, is because you eventually want a DAC/ADC between the two domains, single-point-grounding under the converters.  Which is a bad idea, a fantasy often promulgated by unscrupulous appnote writers.  There's no substitute for responsible signal handing; splitting grounds almost always leads to troubles worse than mere ground loop offsets.

Keeping everything on common grounds simplifies routing and EMI.  You may end up with some ground loops (that would otherwise be prevented with isolation), but these are usually a matter of routing signals appropriately.  For example, if your mainframe provides VREF to all modules, you should provide a VREF_GND which is essentially ground, but only connected near the reference and never carries load current, so it doesn't experience the voltage drop that circuit GND will.

The (VREF, VREF_GND) pair is then sensed differentially, where needed.  Since it's a DC signal, it can also be heavily filtered, to keep EMI out of the analog signal paths.

Or if you have analog signals (not DC, but modest bandwidth), common to all channels, the same approach works just as nicely, given that you've specified the current capability and bandwidth of that signal.  That is, the signal might be driven by an op-amp, so it will only be accurate for some 10s mA of load, and the minimum load (presented by all modules in parallel) should be some kohms of resistance, plus whatever load capacitance is required to filter to that bandwidth.

This works great for an architecture such as, input VREF --> master error amplifier --> slave amplifiers (usually transconductance mode) --> output.

If the system bandwidth is modest, and current sinking is available from the amplifiers, you can simply wire audio to the VREF input and you have yourself a big powerful laboratory amplifier, too!

You can apply lots of add-ons to such an architecture, like adjustable current limit by clamping the intermediate signal (master error amp output = load current setpoint), or using any selection of linear, switching or hybrid modules for whatever noise or efficiency goals.

Tim

[technix]

What topology to use then? TOPSwitch = $$$, but TL494 is acceptably priced. Maybe I can build both the primary tracking regulator (32.5V/10A/50W max) and the secondary regulator (providing the fixed +5V & +12V voltages) using that? It seem to me that the dual error amps in TL494 can be fantastic for my purpose here as I can use them to keep both +12V and +5V rails safe (and let -12V rail to be)

And I think I need to scale my PFC capacity up to 150W or so, and raise the primary rail power cap to 100W as my transformer should be able to handle it. UCC28019 is awesome, BTW.

You can afford PFC on this project, but not another flyback controller (3842 or TOPSwitch or FPS or..)..?  The controller chip itself is a small part of the total cost, anyway; you'll likely spend more on the transformer, and you need two (aux and main).

I can add another controller, just not a TOPSwitch which is difficult to source from within China. TL494 seem to be a better all-in-one chip for me, since it can support dual rail using two optocouplers (two error amps) and wildly varying output (multiple topologies, both switchers will see that, either from the user setting a voltage or plugging in an iPad)

I was thinking about splitting grounds so I can have the analog and digital sections entirely separated, to the point of being able to move the analog and power sections, with their respective isolated controls, off to daughterboards so I can construct a simple plug-and-play multi-channel system. Main rectifier, PFC and digital section on the backplane, output strips on the daughterboards.

I don't see any reason that should require isolation, though.  Or if it does, perhaps each channel would need isolation?  Which would perhaps be best handled as a per-module cost (i.e., put the isolator in the module) rather than a mainframe cost (you buy N isolated channels whether you need them or not).

Isolation really stinks if you need to isolate analog signals, too, in which case you'd rather do all the isolation digitally.

And if you only have digital and analog circuits, with common grounds between each domain, but isolation between the two... How?  Why?  The only reason I can see having both in a system, is because you eventually want a DAC/ADC between the two domains, single-point-grounding under the converters.  Which is a bad idea, a fantasy often promulgated by unscrupulous appnote writers.  There's no substitute for responsible signal handing; splitting grounds almost always leads to troubles worse than mere ground loop offsets.

Keeping everything on common grounds simplifies routing and EMI.  You may end up with some ground loops (that would otherwise be prevented with isolation), but these are usually a matter of routing signals appropriately.  For example, if your mainframe provides VREF to all modules, you should provide a VREF_GND which is essentially ground, but only connected near the reference and never carries load current, so it doesn't experience the voltage drop that circuit GND will.

The (VREF, VREF_GND) pair is then sensed differentially, where needed.  Since it's a DC signal, it can also be heavily filtered, to keep EMI out of the analog signal paths.

Or if you have analog signals (not DC, but modest bandwidth), common to all channels, the same approach works just as nicely, given that you've specified the current capability and bandwidth of that signal.  That is, the signal might be driven by an op-amp, so it will only be accurate for some 10s mA of load, and the minimum load (presented by all modules in parallel) should be some kohms of resistance, plus whatever load capacitance is required to filter to that bandwidth.

This works great for an architecture such as, input VREF --> master error amplifier --> slave amplifiers (usually transconductance mode) --> output.

If the system bandwidth is modest, and current sinking is available from the amplifiers, you can simply wire audio to the VREF input and you have yourself a big powerful laboratory amplifier, too!

You can apply lots of add-ons to such an architecture, like adjustable current limit by clamping the intermediate signal (master error amp output = load current setpoint), or using any selection of linear, switching or hybrid modules for whatever noise or efficiency goals.

Tim

So is this system design a good idea:
* Backplane provides the modules with some kind of mains power (rectified and PFC corrected? Or just filter it and allow the modules to handle PFC themselves?) a 5V rail straight from the auxiliary power (modules have to isolate this themselves,) a 3.3V I2C bus and open-drain interrupt signal (also modules need to isolate those themselves too.)
* The backplane contains the auxiliary power supply (outputs a fixed 5.3V, allowing a wildly variable load current,) digital control circuitry (main MCU, etc) and maybe a 2.5V reference (ADR03B)
* Each daughterboard have its own mains-rated power converter with both constant-voltage and constant-current capabilities. (here is where TL494 shines again, dual error amps means I can use it to implement a purely switch-mode output module with both CV and CC modes. My original design uses a tracking switch-mode section that provides a dual error amp linear regulator wired into a CC/CV mode with just enough power) Modules each contains their own local MCU (since somehow off-chip ADCs are more expensive than ATmega328P here, so I am using the weirdly cheaper network of microcontrollers)

[T3sl4co1l]

Mains filter should definitely be common: there's only one line cord, and only one point of filtering.

PFC likely has economy of scale, so you'd be better off with that common.  So, 400V or whatever common to all, that's not bad.

Can you reduce it further? Would you have a common, say, 30V rail for everything?  Might be worth isolating everything and running it from a modest rail.  Modules might then be buck type (or if higher output voltages are necessary, boost or SEPIC), linear or hybrid.

Running everything from 400V is fine, but incurs more EMI and noise problems (you have N * independent offline converters?), and more incremental cost (you're losing economy of scale).

Beware that the ATmega analog circuitry is shit.  The ADC is only 10 bits to begin with, and is only good for 8 at best (uncalibrated).  And that's using an external VREF.  The internal VREF is more like 5 bits...

I've also heard that the POR circuit doesn't work reliability (namely, under slowly rising or falling supplies), so you should probably put a reset generator IC alongside them.

The XMEGA seem okay, but I haven't done anything in-depth with them, nor heard any horror stories.  The 12-bit ADCs at least have okay specs.  The onboard ref still sucks, but what do you expect.

If you need 0.1% or better, and can't swing better than an ATmega, get a MCP3208 or whatever.  That's the first thing I think of, but if you can't find it for cheap around thereabouts, I'm sure there's a local equivalent that does.  Hey, if you're going to complain about prices, in the one country that produces any of this shit, that's your problem, deal with it...

Tim

[technix]

snip

Does this seem right? Can it handle a wildly varying output?

[T3sl4co1l]

Well, obviously it's not designed for it (R14 only covers a ~20% range -- which is more than enough trimming for the application, which is good), but you still have the problem that C14's voltage will follow C17's voltage. 

Why forward converter this time?

This brings other changes:
- IC3 --> UC3844 (set RTCT oscillator for twice the desired switching frequency)
- D6 --> ultrafast (or use RCD clamp snubber instead of primary reaction winding)
- R29 too large?
- D7, D8 will see over 40V peaks as shown, use UF4004's or whatever
- C16, R30 won't do anything as shown; they aren't really important or necessary for this type of circuit anyway.  C15-D6 serves the purpose of Vpk clamp snubber.
- C15 can probably be smaller (10n?), but be careful of resonance against transformer leakage
- R18 is too small for TL431 bias current (min 1mA)
- Remove R19, C10
- R17 --> 100 ohms or so
- Note that any additional load on +18V has to start up only after IC3 starts up.  Q2 should probably be repurposed as a switch gated by aux power.
- In general, L2 and L3 won't carry proportional currents or similar voltages.  L2 also sees 2.5 times more voltage.  C14 will probably develop around 40V, but be highly variable as well.

In this case, it would be acceptable to use a diode (with a current limiting resistor) to obtain aux power.  Remove D7 and replace L2 with 1-10 ohms. Reduce aux turns to suit.

This way, the supply (which is +400V stable, though with AC line ripple) appears across the primary while the transistor is on; the same voltage appears on the windings, divided by turns ratio.  A secondary of 4 turns would give about 16V peak, which will give somewhat less VDC after half wave rectification in series with a current limiting resistor.

The main secondary is probably fine, but expect to change L3 and C17 as needed.  And, in turn, R15 and C8.

Tim

[technix]

Well, obviously it's not designed for it (R14 only covers a ~20% range -- which is more than enough trimming for the application, which is good), but you still have the problem that C14's voltage will follow C17's voltage. 

Why forward converter this time?

This brings other changes:
- IC3 --> UC3844 (set RTCT oscillator for twice the desired switching frequency)
- D6 --> ultrafast (or use RCD clamp snubber instead of primary reaction winding)
- R29 too large?
- D7, D8 will see over 40V peaks as shown, use UF4004's or whatever
- C16, R30 won't do anything as shown; they aren't really important or necessary for this type of circuit anyway.  C15-D6 serves the purpose of Vpk clamp snubber.
- C15 can probably be smaller (10n?), but be careful of resonance against transformer leakage
- R18 is too small for TL431 bias current (min 1mA)
- Remove R19, C10
- R17 --> 100 ohms or so
- Note that any additional load on +18V has to start up only after IC3 starts up.  Q2 should probably be repurposed as a switch gated by aux power.
- In general, L2 and L3 won't carry proportional currents or similar voltages.  L2 also sees 2.5 times more voltage.  C14 will probably develop around 40V, but be highly variable as well.

In this case, it would be acceptable to use a diode (with a current limiting resistor) to obtain aux power.  Remove D7 and replace L2 with 1-10 ohms. Reduce aux turns to suit.

This way, the supply (which is +400V stable, though with AC line ripple) appears across the primary while the transistor is on; the same voltage appears on the windings, divided by turns ratio.  A secondary of 4 turns would give about 16V peak, which will give somewhat less VDC after half wave rectification in series with a current limiting resistor.

The main secondary is probably fine, but expect to change L3 and C17 as needed.  And, in turn, R15 and C8.

Tim

Forward topology: load varies a lot (200mA or so when idle with only the backlight of the LCD on, 3A when all systems go and charging an iPad Pro at full current). Flyback won't take that load variation too well. And the transformer design for flyback bites me hard.
R14: I just need to trim the backplane LV rail somewhere around 5V, so this range is okay. 5.3V is usually a good value for USB charger ports and that is why I trim for that. For primary output strips I use external error amps and ADC for voltage (and current) settings.
IC3: I will probably spin for the cheaper of UC3844 and UC3845. Won't hurt, I believe.
D6: US1M is SMT version of UF4007 (I default to SMT whenever possible.) So already ultrafast. RCD clamp have a lower efficiency as energy is burned in the clamp, but the reset winding returns the energy to the mains filter cap.
R29: 1V/(40W/400V*4) = 2.5ohm. Yup, should have spun for a 2.5 ohm one here.
D7, D8: Either SS16 or US1M should work. Same package anyway, so draw SS16 in but leave headroom for the higher voltage drop from US1M. Install whichever is cheaper that week and it will work.
C10, C16, R19, R30: I'll just leave the pads there, unpopulated (mark those as DNP.) Fairchild's AN included those though, so just as a safe move.
C15: I though about that... Going back to 10n
R17, R18: 100 ohms for each of them?
Q2: The PFC chip also shares this +18V rail, so does the output strips. So maybe I can leave D7 unpopulated and short out pads for L2 but keep the assisted Zener in there.

For switching other users of +18V I would use a MOSFET for that. Again, lower loss.

I am way oversizing L3 and C17 (as to keep the output ripple within the acceptable range for MCU and USB I only need 1/4 the capacitance of C17 and 2/3 of the inductance of L3) to keep the ripple down. I have a habit of doing this - in my previous designs even a MC34063-based boost SMPS that drives only one op amp and one MOSFET gate (linearly) have a specified 10mV maximum ripple on the 15V output - just a few hundreds of ppm.

[T3sl4co1l]

You don't have the option of reducing ripple by using overly large inductors: you're limited by the controller.  A peak mode controller goes unstable (chaotic, in fact) in CCM.  You don't have any slope compensation shown, but that only extends the operating range some, it doesn't fix it.

This is the main reason why peak current mode controllers are used for smaller applications, like < 100W.

Tim

[technix]

You don't have the option of reducing ripple by using overly large inductors: you're limited by the controller.  A peak mode controller goes unstable (chaotic, in fact) in CCM.  You don't have any slope compensation shown, but that only extends the operating range some, it doesn't fix it.

This is the main reason why peak current mode controllers are used for smaller applications, like < 100W.

Tim

Auxiliary rail maxes out at about 50W. NBD.

[T3sl4co1l]

Yeah, the only downside is using somewhat larger / more robust / more qty of electrolytics than you'd need otherwise, but they simply aren't very big to begin with (a buck or two worth), so it's not a big deal.

Tim

[technix]

Yeah, the only downside is using somewhat larger / more robust / more qty of electrolytics than you'd need otherwise, but they simply aren't very big to begin with (a buck or two worth), so it's not a big deal.

Tim

Let's just say except one dubious cap kit I bought long time ago, all through hole high capacitance electrolytes I have share a single brand: Nippon Chemi-Con. No worries here (and I just updated my PCB to use those Nippon Chemi-con ones despite preferring the surface-mount caps.)

I even got a few Nippon Chemi-con electrolytics for my mains filter cap. And that is NOT cheap.

[T3sl4co1l]

Brand doesn't matter, ratings matter. You can't use a "general purpose" (high ESR, 2000 hr 85C) part here, for example.

You don't need to use aluminum polymer here.  That would be overboard, no matter whose brand you get.

That leaves the lower ESR series, suitable for switching, and preferably with higher life.  That still leaves a lot of choice, even from a single manufacturer.  And it may be helpful to use two in parallel; there's nothing wrong with that.

Tim

[technix]

Brand doesn't matter, ratings matter. You can't use a "general purpose" (high ESR, 2000 hr 85C) part here, for example.

You don't need to use aluminum polymer here.  That would be overboard, no matter whose brand you get.

That leaves the lower ESR series, suitable for switching, and preferably with higher life.  That still leaves a lot of choice, even from a single manufacturer.  And it may be helpful to use two in parallel; there's nothing wrong with that.

Tim

I don't even have any general purpose aluminum electrolytic caps except that dubious through-hole cap kit I bought long ago which tops out at 470uF. I haven't used a single through hole electrolytic below 1mF in my design from the get-go (I use chip tantalums there)

[underwol2]

estoy comenzando a estudiar fuentes y me gusta estudiar y practicar para entender  y aprender casual mente tengo un cargador de baterías que me llego le faltaba una pieza no en contre el diagrama  o schematic
me encontre de tanto buscar por Internet  un archivo en español sobre el diseño con este integrado uc3842
documento en pdf
DISEÑO Y CONSTRUCCIÓN DE UNA FUENTE CONMUTADA pdf
al mirar los cálculos o formulas que a estado haciendo en la pagina 59 donde se esta despejando C1 al repetir la formula como la muestra no logro el resultado no se que pasos uso para hacer ese cálculo o formula para despejar C1 y no se como se saco ese valor  C1 = 22uf
podrían hacerme el favor y me ayudan  como es que se despeja esa formula hasta que se saca el valor de de C1
gracias desde colombia

[technix]

estoy comenzando a estudiar fuentes y me gusta estudiar y practicar para entender  y aprender casual mente tengo un cargador de baterías que me llego le faltaba una pieza no en contre el diagrama  o schematic
me encontre de tanto buscar por Internet  un archivo en español sobre el diseño con este integrado uc3842
documento en pdf
DISEÑO Y CONSTRUCCIÓN DE UNA FUENTE CONMUTADA pdf
al mirar los cálculos o formulas que a estado haciendo en la pagina 59 donde se esta despejando C1 al repetir la formula como la muestra no logro el resultado no se que pasos uso para hacer ese cálculo o formula para despejar C1 y no se como se saco ese valor  C1 = 22uf
podrían hacerme el favor y me ayudan  como es que se despeja esa formula hasta que se saca el valor de de C1
gracias desde colombia

Pérdon. No hablo Español.

Do you mind repost in English? I speak very little Spanish.

[underwol2]

yo tampoco hablo otro idiomas utilizo el traductor de google para poder comprender los temas en los foros de otros idiomas
 

[underwol2]

lo que quiero es como hacer los cálculos con ejemplos para diseño con uc3842 para que no se complique al menos cálculos de solo uc3842

[Cliff Matthews]

lo que quiero es como hacer los cálculos con ejemplos para diseño con uc3842 para que no se complique al menos cálculos de solo uc3842

No lograrás ni limosna aquí si no atenta postear en inglés bro. Hay piles de miembros usando ese traductor añadiendo a la comunidad. No desanimes, dale un chancesito! (pa´ seguir en español es peor..)