buck converters, small voltages and very light or no loads

Thread Starter

ag-123

Joined Apr 28, 2017
276
i'm planning to build a bench power supply, it takes from mains and a transformer brings it down to 22v ac which i'd rectify into DC. for that matter whether it is rms we'd set it aside for the moment.

And to convert that 22v (possibly higher if the specs is after all rms), i'm intending to use buck converters to convert (dc-dc) it down to lower voltages.
Hence, i started doing some spice simulations to anticipate what to expect.
here is the model i'm using for my buck converter simulation

buck22v-3v_sch.png
View attachment buck1.cir
buck22v-3v.png
i'm rather surprised by the long settling times when i wanted 3 volts out of 22 volts, and i'm using a rather low 10 ohm load resistance.
if i change that 10 ohm load to 100 ohm, (e.g. for small mcu, even tens of milliamps is plenty), the settling time stretch out extremely and this same chart is still well above 15 volts at the end of that 3ms.
buck22v-3vb.png
the other thing is as the buck converter start settling from 22v VIN, at 22v it would probably blow semiconductors, mcus e.g. for those 3v ones some of which won't be able to take a voltage any higher than 3.6v

any idea what could be a better way to deal with very small loads (e.g. tens of milliamps ) or even possibly disconnected - no load, and large dropout to small voltages, as well as the high initial voltage ?
 
Last edited:

crutschow

Joined Mar 14, 2008
34,284
You problem is the simple simulation, which has an inherent overshoot due to the L1C1 resonant circuit.
A real converter circuit has feedback and a slow-start circuit so there is no overshoot of the desired voltage when the power is applied.
It does not start with a fixed PWM duty-cycle as your simulation has.

If you want to simulate a real buck converter, download the free LTspice and simulate one of their buck converter circuits, (they have test circuits already designed for simulation).

Below is an example:
You right click on the converter symbol, and then click the top option to open the example circuit.

upload_2019-9-8_12-8-58.png

upload_2019-9-8_12-11-30.png
 
Last edited:

Thread Starter

ag-123

Joined Apr 28, 2017
276
thanks crutschow !

hi bob,
in my simulation, the desired output voltage is Vout = d x Vin = 22 x 0.1363 = 3v
so i used a 13.6% duty cycle. for my switching frequency, i started experimenting with 10khz and increasing it to deal with the ripple voltage.
i think like crutschow mentioned, a real buck converter ic probably won't start right out with a large duty cycle. But if it does that i think it would look like how this simulation works as the switching frequency and duty cycle is fixed from the beginning
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
i found a goof in my buck converter simulation, my transistor/mosfet/switch is ON right from the start. Hence, all that 22v gets sent to Vout right at the beginning.

same circuit, i fixed up the pulse train so that it switch the transistor off at the beginning.
View attachment buck2.cir
buck22v-3vc.png
this time round it looks more manageable as a buck converter running at 20khz, 13.6% duty cycle to achieve 3v load is 10 ohm, this looks very much like a real buck converter.

unfortunately if I use a 100 ohm load (30ma currents).
buck22v-3vd.png
i got a wacky voltage surge that goes up to 9v and rising. same frequency 20khz, 13.6% duty cycle, based on the formulas for steady state
Vout = Vin x D = 22 * .136 = 3v
the voltage surge still happen but less severe even if i reduce the duty cycle down to 2% ! 1us of 50us
i've had to reduce the output capacitor say to values like 10uf or even 1uf (lots of ripples), to hold back that climb
using a small inductor apparently just cause the current and voltage ripples to get much worse, it seemed a moderately large inductor actually helps. it seemed for very light loads, i'd need to stretch out the period and use very small duty cycles
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
another experiment, same buck converter model
100 ohm load, this time i've to use :
duty cycle 1us (pulse) in 200us (period) - 5khz frequency
and reduce output capacitor to 10uf

i'm getting about 0.9v across 100ohm, 9mA !
it is pretty obvious the converter is operating in discontinuous mode
buck22v-3ve.png
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
one more experiment, same buck converter model
1000 ohm load, this time i've to use :
duty cycle 1us (pulse) in 1000us (period) - 1khz frequency !
1us pulse, the other 999us, the load runs on the inductor and capacitor !
and output capacitor 10uf
the converter is operating in discontinuous mode
i'm getting about 1.6v across 1k ohm, 1.6mA, oh wellbuck22v-3vf.png
buck converters are literally formula 1 engines, if the target device is a micropower device o_O
 
Last edited:

Alec_t

Joined Sep 17, 2013
14,280
Doesn't look like discontinuous mode. Each switching pulse is letting some current through to charge the cap. As you're finding, without feedback to the switch the output voltage on the cap is very much dependent on the load.
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
as i'm using ngspice http://ngspice.sourceforge.net/ , i'm not able to plot I(d1), but based on a basic analysis, i(d1) would be zero when the transistor (switch) conducts and it would be equal to i(L1) when the transistor (switch) is off, assuming that its model is after all a perfect diode. but literally i'd wish it model a diode close to real life, it seemed somewhat that way but i've not verified it.

as for v2, if i plot v(2), v(3), i(l1)
buck22v-3v-v2.png
this is for the case of the successful buck simulation a couple posts back. as we expect, when the transistor conducts v2 is at 22v i.e. Vin.
what is less expected is you can see some high frequency ringing between the 22v spikes, they are rather 'irritating'. But i read somewhere, if the converter goes into discontinuous mode, these ringing can literally occur in real buck converters. in this case, i'd think this simulation is pretty real. they do not show up in v(3) as the output capacitor would have removed these high frequency ringing
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
if i plot v(2), v(3), i(L1)
for the case where i used 100 ohm load, 10uf output capacitor
duty cycle 1us (pulse) in 200us (period) - 5khz frequency
buck22v-3v-v2_rl100.png
you can see the rather extreme ringing that shows up, and in particular they show up after the transistor stop conducting (switched off).
in the front part the inductor conducts pretty much a flywheel and supply the load, but a little further on the ringing starts. and inductor current is very much zero this probably matches the scenario of discontinuous operation. what is a little surprising is that v3 (i.e. Vout) is non zero around 1v, which means that currents is after all flowing in the load which is simply a 100 ohm resistor, probably supplied from the capacitor and falling currents from the inductor in combination

the diode and inductor model is likely very good and simulated parasitic capacitances. there is literally no 'transistor' in the circuit, it is a perfect switch. so that (parasitic capacitances) likely contributes to the ringing effects
 
Last edited:

ci139

Joined Jul 11, 2016
1,898
(upload is) LTSpice netlist (regulated - soft start) -- a non-real world , near-ideal circuit
//// -- can't advise for NG-Spice . . . You may try adding 500mΩ or less in series with the fly-back diode . . . also limit your 22V by 2 to 30 Ω (series internal resistance) . . . + set TH other than Zero usually V.th = ( V.ctrl.max - V.ctrl.min ) / 2 ((← the least just because the Spice engine is more likely to detect it at the median))
 

Attachments

Last edited:

Thread Starter

ag-123

Joined Apr 28, 2017
276
placing a resistor in series with the diode did not remove the ringing. however, i think it is ok as these ringing don't show up at v3 (Vout), removed by the output capacitor. just that it would not be appropriate to sample v2 i.e. at the diode, transistor and inductor junction for control purposes.
for a 'simple' case, i.e. fixed load i can set a fixed duty cycle, frequency and run open loop, the only downside is the initial overshoot and so forth.
if i use a buck converter ic, i'd think this would be mostly resolved as i'd think most of them are after all closed loop control buck converters

but i think the *zero (no) load* (or very low load) condition presents a unique condition to the buck (think as well as boost) converters. i've had some lipo power banks where if i draw too little currents, the output simply shut off. i think this reflects what is out there. there are those that lets me use very little mA currents and still run fine and there are those that simply shutdown even when mA currents are still in the low 10s of mA
 
Last edited:

Alec_t

Joined Sep 17, 2013
14,280
Could the spikes be sim artifacts? What are the spice properties of the switching voltage VSW and the switch SW1?
 

ci139

Joined Jul 11, 2016
1,898
sorry the resistor should be equal less 370mV / 3A ≈ 120mΩ (some ringing occurs anyway - because the LC and parasitic inductances and capacitances present -- the overshoot spikes are also expected as the simulator uses quanted time -- they shouldn't be too frequent/excessive however)
 

Thread Starter

ag-123

Joined Apr 28, 2017
276
@Alec_t it is possible too, as i've very much just started playing with ngspice. i'm not too sure what is included in the models but i simply use whatever is generic and available. but if ngspice literally simulated the ringing correctly, then it is a big plus for the spice simulators as you would literally observe them in the real thing
 
Top