Buck converter fine tuning

Discussion in 'The Projects Forum' started by johsey, Feb 25, 2011.

  1. johsey

    Thread Starter New Member

    Sep 29, 2010

    I need some help on buck converter fine tuning. Following are the operating mode and the simulation schematic/results.

    Operating mode
    VIN = 24V
    Varying RLOAD (PTC): 3.5 to 11Ω

    VOUT is set to 19.6V during 4s then switched to 6V for 1s and switched again to 19.6V and so on (cycle).

    The schematic of the buck converter based on the LT3845AL controller:
    refer to attached file "LT3845A_based_buck_converter.jpg"

    The VOUT and inductor current plots:
    refer to attached file "SIM-LT3845A_based_buck_converter.jpg"

    At the beginning, VOUT is set to 19.6V. After 1ms, the controller is turned off during 4ms. Then, VOUT is set to 6V during 1ms and is switched to 19.6V again. (this scenario is for simulation only)

    It seems that the soft-start function doesn't act as I would expect: it should reduce the inductor current spike and the output voltage overshoot by increasing the start time of the controller, shouldn't it?

    My questions are:

    1. How to efficiently reduce the output voltage overshoot?
    2. How to efficiently reduce the inductor current "spike" at each transients?
    3. What is the real impact of this inductor current "spike" on the entire circuit (short- and long-term)?
    4. What can be improved in general?
    Thank you.
  2. SgtWookie


    Jul 17, 2007
    You do not have a source of current for the regulators' Vcc. Since your test involves a wide range of output voltages, you won't be able to derive Vcc from the regulators' output. Read the section in the datasheet titled "Power requirements" beginning the 2nd column on page 10.

    "Soft-Start" information begins on page 13.

    Did you start by trying the supplied test model for the regulator? It exhibits very little overshoot.
  3. SgtWookie


    Jul 17, 2007
    I'm not quite sure why you're changing the voltage divider to change the output voltage. Didn't you really want to change the output current due to the PTC load instead?

    If so, it would be a lot easier to use a current source as a load, and have the current source pulse at regular intervals.

    At any rate, if Css is pulled to ground momentarily, the soft-start feature will be re-initialized.

    See the attached; it's the Linear Technology supplied test fixture with Rload changed to a current source set up as a load to approximate a 3.5 to 11 Ohm load with a 12v output. There are some minor overshoots/undershoots when the load changes, but regulation is pretty good.
    Last edited: Feb 26, 2011
  4. johsey

    Thread Starter New Member

    Sep 29, 2010
    SgtWookie, thank you for your answers.

    Mistake on my side: I've misinterpreted this section...

    No. I wasn't aware of it.

    In our application, the output voltage can be changed to different values in order to do a temperature regulation: the PTC is used as a heater. Thus the need for a voltage divider with different resistances values for the feedback.

    I will try with the supplied test model.

    Thank you.
  5. SgtWookie


    Jul 17, 2007
    re: regulator Vcc
    That might be a bit nit-picky on my part. The regulator supplies its' own Vcc from Vin during start-up, but to keep power dissipation down during operation, you'd normally want to source Vcc from the secondary side. However, since your application apparently involves varying the output over a wide range, that's not going to work for you.

    re: test model
    Many Linear Technology devices have test models supplied. Drop a component into the schematic, right-click on the device, and then left-click on the "Open this macromodel's test fixture" button. Keep in mind that you'll want to save it under a different filename before making any mods to it, or you might lose the mods when LTSpice downloads updates from the Web. Besides, you normally would want to keep the original test fixture as-is.

    I'm starting to understand the problem. Were you planning on actually using a set of fixed resistors to switch in and out, or more like an NTC on the low side of the voltage divider to provide temperature regulation?

    Rather than having lots of switches and fixed resistors in your model, you might consider using a current source between R5 and R6 in the test fixture. That way you can control the rise/fall time to the FB input rather than have it be a sudden change in value.

    Your simulation is reasonably close to the test model, except for the higher switch frequency, different inductor, lack of external Vcc source, and different MOSFETs. The big questions are, what's the rate of change for your real PTC, and the rate of change you'll have for the feedback input via changing resistance values?
  6. johsey

    Thread Starter New Member

    Sep 29, 2010
    In my case, yes, it won't be possible. I have to supply VCC with an external source: I think, a voltage divider on the main VIN (in order to be within specifications on VCC) should do the work, I don't want to add a voltage regulator for this purpose only.

    Good to know. I haven't been working with LT parts until now.

    To be more clear, our current system is based on a PTC resistor which acts as a liquid heater and as the temperature sensor too (not very clean but we have to live with this).

    One 5s cycle is divided as follows:

    • 4s of liquid heating by applying a "high" voltage (can be between 12 and 19.6V) to the PTC
    • 1s during which the power supply is turned off (in order to stabilize the PTC at the liquid temperature) and then turned on again with a "low" voltage (6V) applied to the PTC in order to measure its resistance and thus to calculate the temperature.
    The next cycle "heating" voltage is chosen depending on the measured temperature (proportional regulation).

    A multiplexer is used in order to select the feedback resistance and then to set the output voltage.

    Rate of change for the real PTC: hard to define because it depends on the system temperature and the applied voltage (I have to test it)
    Rate of change for the feedback input: 2 times / cycle (5s), with the lowest time between two feedback changes equal to ~50ms (at the end of the cycle).

    My actual problem are the current surges in the inductor when switching from the "low" voltage to the "high" voltage (as shown in the simulation in my first post).
  7. SgtWookie


    Jul 17, 2007
    This is very strange; you select a very efficient synchronous buck regulator, and then you add later that you'll be attempting to supply Vin with a voltage divider? That's not going to work well at all. If your Vin will/could exceed the regulator's maximum ratings (65v in this case), you need to look at a different regulator.

    What is your actual Vin range, without using a voltage divider?

    Well, here you're writing 1s, 4s, etc. but in the simulation, your timing is in milliseconds.

    I don't know what your thermal coupling will be for your PTC to the substance that you are heating, but 1mS seems like a very short amount of time. OTOH, running this simulation for several seconds would take quite a while, and result in VERY large output files.

    As I mentioned (or implied) previously, pulling Css low will re-initialize the soft-start feature, minimizing overshoot.

    I'm wondering why you chose 6v as the "test" voltage? Seems to me that you could leave the voltage the same for both heating and testing. You could measure the current via the existing current sense resistor (8m Ohms) and measure the Vout, then determine the PTC resistance via R=E/I.

    Rather than changing the regulators' output voltage, you could leave that constant, and simply vary the regulators' ON time vs OFF time.
  8. johsey

    Thread Starter New Member

    Sep 29, 2010
    What I was talking about is the Vcc back driving voltage from an external voltage source. Vin is 24V±5% which seems to be out of Vcc range, for this reason I need to back drive Vcc pin with a lower voltage -> voltage divider from Vin to a lower value within the Vcc specifications?

    The Vin pin will be supplied with 24V (without voltage divider).

    What I wanted in a first step was to observe the current surges in case of fast transients due to a feedback voltage change and OTOH, the stabilization delay of the output voltage/current after a soft-start.

    And as you mentioned, running test for several seconds takes a while. I am sure there are ways to improve the simulation speed but I'm not an experienced user of LTSpice. :)

    I tried to restart the soft-start feature by pulling the shutdown pin low, generating a SHDN UVLO event which, as explained in the datasheet, should discharge Css. I tested it, but it seems that Css continues to charge even if SHDN is pulled low.

    Pulling Css low by using a transistor seems to work fine.

    A "high" voltage would increase the PTC self-heating during the measurement step, thus increasing the measurement error. By using a lower voltage, the PTC self-heating is reduced (it lets the PTC resistance decrease and stabilize near the liquid temperature).

    6V was used in previous working designs for this circuit. Since we have to "update" this design for backward compatibility, I started with this value. I have to test this circuit with other values in order to find the best trade-off between PTC self-heating and measurement voltage ripple.

    I would prefer using an additional current sense resistor just after Cout in order to measure the average current instead of wide changing inductor current.

    Would you do that by applying a PWM signal on the low-side transistor on the Css pin? Since an ON/OFF signal will be seen by the load, how would you measure the average voltage/current?

    Thank you for your answers.