Post Go back to editing

LT3476 limit cycle

In my experiment with LT3476 for LED dimming, I saw a phenomenon that seems like a limit cycle. I need some help to explain this. For my circuit employing the LT3476, I used 21K for Rt resistor, that means the switching frequency is 1[MHz], and used 20[KHz] PWM dimming signal for PWM input pin, and set reference current for 1.5[A] with Vadj and shunt resistor. I did not use shutdown feature. Power supply is 24[Vdc]. And this is important point as I think, 20K resistor and 220[pF] capacitor connected in series to Vc pin. I did suspect that this controller gain is too large, so some non-linearity in the LT3476 might cause the limit cycle. The limit cycle itself was seen at the zoomed waveform of switching node voltage, and it was observed during LTspice simulation and real experiment. When I used 10K resistor and 220[pF] capacitor, the limit cycle disappeared,

One more question : I used an inductor that is about half an inch distant from the LT3476, value 10[uH] for switching LED current. And I know this inductor is not directional, that is it does not matter whether I connected the inductor in one direction or connected in reverse direction. The data sheet of inductor also ensures this point. But the limit cycle mentioned above seems that it does receive some effect from the orientation of inductor on the PCB. That means, with one orientation of inductor, limit cycle was seen, but with other orientation limit cycle is not shown. Is there any explanation on this observation?

When designing a controller for LT3476, we may need a small signal AC model of the LT3476. Would you provide such a model?



Second Post
[edited by: dbsuh at 2:40 AM (GMT 0) on 20 Dec 2018]
  • Let's start with a single channel.  What is the voltage of your LED string?  Is your circuit a buck-mode topology or some other topology like a boost?  Did you use one of the example application circuits from the datasheet or the demonstration circuit DC976A as a starting point for your design?  If so, which starting point circuit did you use?  Demonstration circuit DC976A is usually the fastest way to get hardware up and running.

    Your 20kHz pwm dimming frequency to dim the LED brightness seems fast.  Your eye will not see the LED turning on and off if the pwm dimming frequency is greater than around 100Hz.  A 20kHz pwm dimming frequency will have less dynamic range than a 100Hz pwm dimming frequency.

    Normally the orientation of the inductor does not matter except for very subtle effects like noise generation.  If I connect a wire to my inductors to observe inductor current, I always put the wire on the side of the inductor that is at a DC potential, so the inductor terminal that is connected to the  SW node stays low and close to the pcb and does not radiate noise excessively.  In reality, inductors usually work even if both terminals are held off the pcb by an inch or so of wire.  

    The VC pin components determine the stability of the LT3476 regulation loops.  Normally you test stability of a current regulator by putting a small step command on the LT3476 VADJ pin and observing the LED current.  If the step command on the VADJ pin does not cause the LED current to ring, then generally the loop is stable.  You can test your circuit for that behavior using LTspice - I doubt if you need a small signal model to compensate the LT3476 loop.

  • Thanks a lot for your answer.

    My LED is about 3.4[V] drop, 1.5[A] rating. There are 4 LEDs in series in one string. Other 3 channels are of similar composition. So, of course, my circuit is buck, not a buck-boost or anything else. I saw DC976A circuit and used it as a starting point of my design. My application is not for human eyes, but for industrial camera in material inspection. That is why PWM dimming frequency was determined to be 20[kHz]. As you know, for camera application, the light intensity must be high, and at the same time on-time must be short enough to prevent heat problems. And to add some explanation, lighting is done on burst mode, that is, not a continuous on mode. Of course, dynamic range will be restricted, but PWM dimming pulse is generated by a 120[MHz] clock, so dimming duty is controlled by that clock.

    For the limit cycle I saw, it was seen at the simulation, too. This is the important point, as I think.

    For the inductor problem, I suspect that distance between inductor and LT3476 is too short. Will you tell me the distance between inductor and LT3476 in DC976A?

    I will simulate the step response on LTspice as you described.

    Sincerely yours,

  • I simulated the step response with LTspice. The step was 0.15[A] equivalent. But before the step and after the step, the limit cycle seen at the SW node voltage zoomed was all the same. As before, 20k resistor and 220[pF] capacitor in series to the Vc pin.

  • Try to upload your LTspice file and I will try to see the problem that you are seeing.  Use conditions that will allow me to see the problem when I run the simulation that you upload.  If you cannot upload the file, perhaps you can go through Technical Support and Technical Support will forward your simulation file to me. 

  • I can see from your LTspice file that the behavior of SW is erratic.  Your LED current is too high.  Worst case SW Current Limit for the LT3476 is 1.5A, so your peak inductor current must remain below 1.5A.  That means your maximum DC LED current is probably about 1.2A.  I show the relevant excerpt from the LT3476 EC table below:

    LT3476 DS EC table excerpt_2.jpg

    Your compensation RC also is not correct.  I used a 1nF from the VC pin to GND in my simulation and the output current transient response looks stable.  I show my simulation circuit and the results of my transient test in the image below.  I will also try to upload my simulation file.

    sim results_1.png

    LT3476 _20190109 Andy_2.zip