-
12V Flybuck Design Posted
11/04/2022 at 18:36 • 0 commentsI revisited this design for the bias supply of a new project I'm working on. I needed a 12V non-isolated output as well as a 12V isolated, and the input needs to work from 30-70V DC. Total power output is 12W with the primary (non-isolated) delivering 900mA and the secondary ('flybuck') about 100mA. AC noise is about 20mV on the primary and 50mV on the secondary.
This layout can be mounted vertically or horizontally and is pictured setup for vertical mounting.
It is slightly smaller than prior layouts and eliminates the bulky electrolytic input cap. It performs well across the entire operating regime with no loading quirks like I had observed with lower input-output voltage margins. Regulation on the primary is ~ 100mV. The secondary's regulation is +/- 300mV and does require about an 8mA load to keep the output from floating up to 20V.
The design workbook & KiCad files are posted in the files section.
-
5V, 5W Version Posted
03/07/2021 at 18:05 • 0 commentsI made a 5V, 5W buck derivative that is smaller than the flybuck design but uses the same input/output filter design for low noise/ripple applications. The inductor core is a Magnetics MPP C055048A2. with 38 turns of 24 AWG. Peak efficiency with a 12V input is ~ 89% at approximately 300mA load. Full load (1A) efficiency is ~ 82%. Lower largely due to DCR losses in the output inductor.
The output voltage will drop ~ 230mV at full load due to these losses. Don't try to move the feedback sense to after the output LC: the phase shift is too great and will result in an unstable control loop (there is no sensible way to compensate with this controller). If tight regulation is required and the application can tolerate a bit more ripple the filter inductor can be removed. Alternatively you could go with an inductor that has a lower DCR but with the associated size/cost penalty.
This design works with a 10-20V input but could easily handle higher input voltages. It includes a bias supply connection which is optional, even up to 20V (the controller uses an internal LDO when the bias supply is left open). Above 20V the bias supply should be used - refer to the data sheet for more information.
-
Main Inductor Characterization
11/14/2020 at 18:53 • 0 commentsIf you build this with the inductor spec in the project you can use this data to validate. The core is a Magnetics MPP #C055050A2, 125u permeability. Do not use any tape between windings: there is no creepage / clearance requirement since this is a non-isolated design and doing so will increase the leakage inductance.
- V1 L=66uH
- Tested @ 20V, 9uS pulse, V2 open
- V1 leakage L=795nH
- Tested @ 5V, 1uS pulse, V2 shorted
- V2 L=458uH
- Tested @ 20V, 25uS pulse, V1 open
- V2 leakage L=5.3uH
- Tested @ 5V, 1uS pulse, V1 shorted
The test jig is a MOSFET driven by a function generator with a large bulk capacitor across the D-S connected to a bench supply. The current ramp is measured with a scope current probe (I use Tek's TCP202). The scope's cursors are used to measure the start/end of the current ramp and the actual rise time (will be longer than the function generator's pulse).
From this the inductance is calculated: L=(V*dt)/di where:
- V=supply voltage
- dt=cursor measured rise time
- di=cursor measured peak current
Parasitic capacitance:
- V1 Fr=1.8MHz, Cp=118pF
- V2Fr=1.56MHz, CP=22pF
This measurements were obtained using a function generator, two passive 1X scope probes, and a 4.7K resistor. One lead of the resistor is connected to a lead of the winding to be measured. The other resistor lead is connected to the function generator's positive lead. The generator's negative lead, along with the scope probe ground leads, are connected to the other lead of the winding. Connect one of the scope probes (say, CH1) to the resistor lead connected to the generator; connect the other probe to the resistors opposite lead (CH2). The other winding should be left open.
Set the scope to trigger on CH1. Program the function generator for a 100kHz, 4V P-P sine wave (no DC offset) and increase in 100kHz increments until the CH2 signal reaches peak amplitude and is in phase with CH1. This is the self resonant frequency (Fr) of the winding. Calculate the parasitic capacitance:
Cp=1/((2*pi*Fr)^2*L)
Where L is the measured inductance of the winding.
- V1 L=66uH
-
Limitations and Problems
11/13/2020 at 21:50 • 0 commentsThe old adage that "you get what you pay for" certainly applies here. A highly integrated solution with simple control schemes poses limitations that aren't easily fixed.
Secondary regulation. The reasons for this have already been discussed and aren't a controller problem per se (any topology will exhibit) but one of coupling between the windings (i.e. leakage inductance). While a well built inductor will minimize this it won't achieve 1% (TI's demonstration boards using commercially built inductors are 10%). Hence, the secondary should be used to power circuits that can tolerate lose regulation.
Conditional stability. This is a result of the complex interaction of V2's low impedance with the control loop and current limit scheme. It is similar to the right half-plane zero found flyback boost converters. The events I've observed to induce this are:
- Vin droops below UVLO and then recovers to some operating value within a few volts of Vin_min (drops that do not go below Vin_min to not induce)
- High capacitive load on V2
Both happen when the secondary is loaded and the primary output is lightly loaded (~>20%) or open. The switching node will go from a stable pattern:
To one of bursts followed by a long off-time:
CH1 is the AC ripple component on the feedback (FB) pin which is necessary for stable operation.
The secondary output voltage will drop to some fraction of its rated output but the primary voltage stays close to its regulated value. If a specific minimum load is applied to the primary, or the secondary load is removed & then reapplied, the controller will resume stable operation. This condition does not occur if both outputs are open or only the primary is loaded (any value). It also does not occur in the opposite direction (sudden increase in Vin).
When the controller encounters these situations it begins by switching with minimum off-time (expected). After a few switching events, current has built-up in the inductor, trips the over-current circuit, and switching terminates. By this time the FB voltage is at or above Vref since V1 is not loaded. Several mS pass (I have measured +20) before the FB voltage drops below Vref and switching resumes. But energy in the secondary is gone so the process repeats. Crucial to this is the lack of synthetic ripple (AC component) on FB pin: it isn't generated due to the short off-times and already high voltage on the FB pin; it doesn't occur during 'cold starts' because the FB voltage is zero.
There is no easy way to fix this. Decreasing R4 to increase the ripple feedback will reduce the V1 load threshold but will not eliminate the instability. A simple hack is to add a diode across the outputs to pull V2 down. A more complex one is to add a synthetic ripple (mabe just voltage sense?) circuit to the secondary and couple to the feedback pin with a resistive summing circuit. Both require a non-isolated secondary.
I decided that the juice isn't worth the squeeze and instead tested the thresholds (Vin drop, minimum V1 load) necessary to avoid this condition (V1 load of ~ 60mA).
Current limit scheme. This is a hazard and likely failure mode when V2 encounters a short or over-current condition with a high input voltage. The specific conditions are: secondary is under load and encounters a short, the primary can be loaded or open, and Vin is high (for my tests 25V).
The controller uses a heuristic scheme to extend the off-time to limit the output current and provide a 'soft' foldback. A key determinant of the off time is the voltage at the feedback pin: the lower it is, the longer the off time. When the secondary shorts, the FB pin voltage remains close to its regulated value (2V) because the V1 output is fine (even when 100% load) and so the off time is short. The next switching cycle begins but is quickly terminated for over-current. The switching cycle looks like this:
The high secondary current (<800mA - well beyond rated 200mA output), combined with the high Fsw sends the controller and rectifer D2 temps through the roof in a matter of seconds. I recorded case temps of 177C for the controller and 184C for D2. Of course this condition does not occur if V1 is shorted (regardless of V2's loading).
I ran a destructive test of this condition to see what would fail first and after 5 minutes, amazingly, neither did. The controller would over-temp, shutdown, and after a few seconds turn back on to repeat the sequence. It was like watching a flashing stop light on the IR camera.
I am sure that one or both would eventually fail. My concerns are fire hazard and that the primary switch fails closed and destroys everything downstream of the supply (a typical failure mode of MOSFETs). A fused crowbar circuit would solve for this.
This is really a crappy control scheme and I'm not sure why TI did this. I've worked with some of their integrated direct-off-line controllers and under these conditions they simply shutdown & reset: they never get out of soft start so there's little current, almost no thermal stress, and certainly no risk of damage to the supply.
Provided V2 doesn't need isolation, the diode across V1/2 idea for instability solves this problem too. Otherwise, a possible solution is to add an external over-current protection circuit; or accept the limitation but do destructive tests to make sure the failure modes are safe & acceptable for the application.
Conclusion. High ratio secondary windings create problems with solutions that can exceed the economic viability of the design. I suspect that this is why in all of TI's documentation and their eval boards the secondaries are 1:1. Use with caution.
-
Build
11/13/2020 at 21:41 • 0 commentsSince one of my design requirements is flexibility, the PCB provides some customization without having to make a new board. That said, you can certainly incorporate this into a project and eliminate it as a disparate component.
Secondary. The secondary can be isolated or tied to the primary ground. If an isolated secondary is desired R7 should be replaced with a 2.2nF MLCC reduce common mode noise.
Controller bias. The controller has an internal linear regulator that is necessary for startup. Beyond startup, it is best to power the controller via one of the outputs if they exceed ~ 9V. As TI points out, high input voltages almost necessitate the use of bias to eliminate the dissipation from the internal regulator. Use a 603 size jumper to connect the appropriate output, and keep in mind that if the secondary is used it must be tied to common ground via a jumper at R7.
Controller mounting. The controller has a thermal pad to improve heat dissipation. The PCB accommodates this and provides a generous copper pour. If you don't have a reflow oven or air pencil apply thermal paste instead. I used this method and the worst case rise was 50C.
Main inductor build. The board is setup to use a toroid core with an OD of 13.5mm. As discussed in the Theory of Operation section, core selection & winding configuration are crucial to minimize leakage inductance. Ignoring core losses, a core with the highest permeability that provides an adequate saturation margin minimizes the number of turns. A high turns ratio complicates the build. If possible use a bifilar winding (1:1 ratios). Otherwise an interleave is best.
For the inductor used in this project I used a simple interleave: winding 1/2 the secondary first, then the primary ('buck' winding), then the remaining secondary. Splitting the primary in half (so P/S/P/S) for four layers might get the leakage inductance a bit lower but it becomes difficult to do manually. The winding details are documented in the design workbook.
Wire size, type, skin effect, etc. is a discussion unto itself but since the currents are relatively small here it is more just a question of wire AWG to minimize DCR balanced with the available winding window. Ideally the secondary would use 28AWG vs. 30 but the increase in leakage inductance isn't worth it: this happens because of the increased air gap between adjacent strands. The window utilization also becomes very high (a machine wound core with 28AWG secondary wouldn't be possible), making the winding sloppy and further degradating regulation.
Installation. I intend to mount the unit as a daughter board using 20AWG wire with nylon spacers (hence, no mounting holes).
-
Theory of Operation
11/13/2020 at 21:40 • 0 commentsTI's datasheet does a good job of explaining the controller architecture so I won't cover that. Additional guidance can be found in application notes AN-1481 & AN-2292. Instead I'll focus on those aspects specific to this implementation.
Main inductor. I wound my own inductor on a Magnetics MPP core because I couldn't find anything off the shelf meeting my spec; if at all possible use a commercially built one (unless you are really good). The secondary winding is not regulated which makes it very sensitive to leakage inductance. In other words, output voltage regulation is dependent on this. This is the reason for the 11% V2 regulation spec. I only got this after three different cores and six winding configurations.
If you do build your own inductor try to keep the turns ratio and overall # turns low. This spec uses a 2.6:1 ratio for the secondary which is a likely limit. A high number of turns / ratio increases leakage inductance. The winding build itself (layers, wire, etc.) also has a huge influence and this is discussed more in the <build> log.
V2 snubber (C18, R8). These were added during evaluation to reduce the leakage inductance ring. This trace shows the unclamped ring.
Although it did not appear to materially increase D2's dissipation I elected to add it for reduced EMI. Trace below is with snubber installed.
At higher input voltages R8 will dissipate ~ 250mW.
LC filters. The input and both outputs include LC filters to reduce ripple and switching spikes. The values for these were obtained empirically by observing the frequency spectrum in the outputs without filters and then calculating the minimum required inductance. A larger value was then selected based on size & saturation margin. Resulting worst-case AC ripple (CH1=V1, CH2=V2):
Input capacitance is sized to minimize input current ripple and for convenience I used the same inductor to filter switching spikes.