Do I need per-phase current sense?

[mck1117]

I'm working on a project to build a fan controller for high power automotive cooling fans, some of which can pull upwards of 60 amps.  So because it's fun, I want to build a multiphase buck regulator.  The STM32F334 has the magical HRTIM timer that can generate up to 5 phases of synchronous buck signals (with deadtime and easy phase shedding!) in hardware, so that's the plan.

The planned architecture is a simplified case of a synchronous multi phase buck regulator: high and low side MOSFETs, LC output filter on each phase.  Since the output accuracy and stability requirements are essentially none, I don't need huge loop bandwidth, and will probably actually run open loop.  Three phases for now, maybe more later.

So here's the question: do I actually need independent inductor current sense on each phase? I'm planning to use a hall effect (Allegro) current sensor on the output, primarily for overcurrent (aka short circuit) detection, in addition to deciding how many phases should be running at any particular time.  I assume that I'll get reasonable current balance between the phases just because they're all the same (modulo input/output resistance from traces), and I can estimate whether I'll be in CCM/DCM by the input/output voltages, and output current.

Is there any other reason that I need sensing on every phase?

[uer166]

The only safe way to go is a current-mode control scheme, which necessitates individual current sense per inductor. Moreover you need separate current loop control instances as well. Peak, average current, whatever. Then there is an outer voltage control loop that sets the setpoint for all the individual current loops.

If you run open loop, you will have great current overshoot whenever there is a load transient, probably damaging the FETs. Do you need it to be syncrhonous? Having a sync buck can also require bidirectional current sensing as well, since the topology can boost the output voltage back into the input.

[mck1117]

If you run open loop, you will have great current overshoot whenever there is a load transient, probably damaging the FETs.

Is this true even if the load is almost purely a series LC? Remember I'm driving a fan (600W big brushed DC motor), not building a general purpose buck.  In case of a true output short, I'm happy to just trip a fault and turn the regulator off, which a single output current sensor provides.

Do you need it to be syncrhonous? Having a sync buck can also require bidirectional current sensing as well, since the topology can boost the output voltage back into the input.

It doesn't necessarily have to be synchronous, other than power dissipation.  600 watts output power is a lot.

[uer166]

Is this true even if the load is almost purely a series LC?

It is true especially if the load is LC. You have an LC load permanently attached anyway (the inductor+output capacitor). Try a simulation of a PWM source that starts up into an LC circuit and you'll see what I mean.

[mck1117]

Oops, that was a typo.  A brushed DC motor is almost a pure LR, not an LC.

[Siwastaja]

If you plan to use PWM control and use same duty cycle for all phases, you may be surprised how poorly the phases share current. The higher the efficiency of the system, the worse the problem.

Unless you want to push your luck, bare minimum, you'll need at least a crude per-phase overcurrent comparator to prevent inductor saturation which increases the dI/dt rapidly. And once you have this comparator, why not use it for the control loop as well, to share current properly? So make it accurate enough, say 5-10%.

Active measurement enforcing current sharing is good to have because then you don't need to massively oversize all phases to handle worst-case currents.

[jbb]

Yes, Siwastaja is right - per phase current sense and control is the way to go. In continuous conduction mode, a few nanoseconds can make a few milivolts difference to the average DC voltage and if you’ve got very low resistances (because you want low conduction losses) then quite a bit of current can flow. If one inductor current ‘walks’ up enough to reach saturation then you’ll have a bad time.

There is one special case; discontinuous conduction mode. In this mode the inductor currents drop to zero each cycle, so a small timing difference will produce only moderate current mismatch. However, discontinuous mode means high current ripple, high core losses, and special consideration for how to drive your synchronous rectifiers...

[Siwastaja]

Yes you can mentally model the PWM controller stage as a voltage source with series resistance of the total system loss (MOSFET loss, trace/wiring loss, motor resistance loss). If these losses are very low, then regulating current requires very careful adjustment of voltage; doing that in open-loop, even slight variation in voltage causes large variation in current (dI = dV/R, R-->0 ==> dI->inf).

[mck1117]

Yes, Siwastaja is right - per phase current sense and control is the way to go. In continuous conduction mode, a few nanoseconds can make a few milivolts difference to the average DC voltage and if you’ve got very low resistances (because you want low conduction losses) then quite a bit of current can flow. If one inductor current ‘walks’ up enough to reach saturation then you’ll have a bad time.

Ah hah - that's the insight I needed.  I added a correct-order-of-magnitude amount of resistance between the phases (input and output, only 300uOhm) in the SPICE model to simulate a connector being on one side of the PCB, nearer to one phase, and there's pretty significant imbalance, on the order of 20-30%.

[Yansi]

What's the point of all this? If the fan is a standard brushed DC, use a simple PWM control, straight to the fan motor. (for 60A, use synchronous diode).  Otherwise, any LC multiphase convertery is pure money and efficiency loss.

[Siwastaja]

Indeed, for controlling DC motor, a DC motor controller is the best choice, which is a half-bridge (synchronous or non-synchronous) without LC filter. Requirement for current sensing is the same, though.

DC motors provide a lot of inductance for you; so much you would never use such large amount if you designed an equivalent DC/DC supply. But the inductance is there for you: you can as well use it! This large inductance means, f_sw will be significantly lower, and ripple current will be lower as well, than what you would design for a DC/DC power converter. Some 10kHz usually works well from the ripple current viewpoint; you may want to go just beyond audible range, though.

End result will be single-phase, though.

Any PWM generation unit then works for this purpose, no need to use HRTIM especially. It's a plus if the overcurrent comparator signal can be connected to the timer directly, STM32 Advanced Control Timers have such break input that can be routed from the internal analog comparator, but software interrupt can be acceptable too if you do this with a simple 8-bit AVR/PIC for example.

[Alti]

For the fun of building multiphase that is ok.
But multiphase are built for low ripple and fast step response.
This is NOT what DC fans require.