LT3757A 5V to 170V Flyback Design Inquiries

Hallo Soohan,

soohan said:

The input caps don't do anything after startup in terms of smoothing out the current spikes. I'm pretty sure this is a modelling issue where I haven't taken some parasitics into account, but I'm not sure how to fix this.

your have an ideal voltage source in parallel to your input caps. As the ideal source keeps the capacitor voltage exactly constant, there is no current flowing in or out your input caps. Adding some internal resistance to your source (as it would be in a real voltage source) will change this.

soohan said:

so I found that all of these issues improved after increasing the inductance values in my coupled inductor. By changing them to 10uH/1000uH, the oscillations went away and the current spikes decreased to about 5A.

You have choosen the switching frequency to be 100 kHz (by setting Rt to 140kOhm). So you have one single switching pulse every 10µs, and every single pulse needs a certain energy (and therefor a certain current) to provide the needed output power.The energy stored in your inductor is 0,5*L*I^2. You want an output power of 170V*40mA=6,8W, you need 68µJ per 10µs pulse. And 68µJ in an 1µH inductor means, that the current has to go up to almost 12A.

In simulation your inductor is not saturated, as your simulation model for the inductor does not know anything about saturation. In real life it would depend on your real inductor if it would stand these high current pulses without saturation.

You can change this by

- either increasing the inductance (you found this already).

- or increasing the switchting frequency (setting Rt to a lower value). You have choosen the lowest frequency specified for the LT3757. If you reduce Rt and increase the switching frequency, you get more pulses per time, and every single pulse requires less energy (less current) for the needed output power.

Besides that two more hints:

- I would recommend to use a "real diode" model for simulation instead ot the intrinsic model of D1

- you have build a low pass filter (R4, C4) in the current sense circuit. Depending on the switching frequency you finally choose this will mean, that you limit the average switching current, not the peak current. And this may be a bad idea, since the peak current may already be destructive while the average current ist still acceptable.

best regards

Achim

Achim,

Thanks a ton for your feedback, it was very helpful for understanding what was going on!

I've introduced the following changes to the design:

• Added 50mOhm internal resistance to 5V source
  • After increasing the value of the input caps, they're taking the brunt of the input current spikes. I'll have to figure out if the inrush current for charging the caps will be an issue though.
• Increased the switching frequency to 400kHz (with coupled inductor values of 10uH/1000uH)
  • Between these two changes, I was able to remove the LPF on the sense pin altogether.

However, when I try to add a real diode model to replace the intrinsic model, I run into some issues.

• The datasheet shows a GSD2004S for the HV reference design. 
• There's another project online that has designed a very similar flyback converter. On their design, they used a RFN1LAM6STR (simulations below).

I have simulated both in place of the intrinsic model, but for each one, I see the following changes:

• Huge increase on the current draw from the input (spikes of almost 30A!).

• Huge increase of the current being drawn from the output capacitor (intrinsic model had peak current at around 300mA):

• Huge increase of the voltage on the sense pin

The output voltage is able to reach the programmed 170V (I'm assuming the blanking period on the sense pin ignores those voltage spikes above 100mV)

Adding the LPF back to the sense pin did not resolve the issue. I'm pretty sure the compensation network/pin isn't involved here.

I don't understand how a change of the diode on the output could cause this. Any explanation would be appreciated!

Draft1.1.asc

Hello Soohan,

soohan said:

I don't understand how a change of the diode on the output could cause this.

The realistic diode models include the effect of reverse recovery time.

When the transistor is off, the diode is forward biased and delivers current to the output. When the transistor is just starting to conduct, the diode gets backward biased and should block the current. But this does not happen immediately, the diode is still conducting in backward direction for some ns - the reverse recovery time. This is the short current pulse you observe and which was not modelled by the simple diode model.

best regards

Achim

Achim,

That makes sense, thanks for the explanation. My (hopefully) last question on this topic is: do I need to compensate for these spikes due to the reverse recovery time? With the diode I selected, they only last for around 25-50ns.

Is this something I need to clamp, or simply part of the switching losses/reduction in efficiency?

EDIT: Also, when I change the coupling factor of the inductors from 1 to 0.985 (to account for max leakage inductance), the current spikes on the primary side seem to disappear! Is the leakage inductance acting as a filter in this case?

Achim,

That makes sense, thanks for the explanation. My (hopefully) last question on this topic is: do I need to compensate for these spikes due to the reverse recovery time? With the diode I selected, they only last for around 25-50ns.

Is this something I need to clamp, or simply part of the switching losses/reduction in efficiency?

EDIT: Also, when I change the coupling factor of the inductors from 1 to 0.985 (to account for max leakage inductance), the current spikes on the primary side seem to disappear! Is the leakage inductance acting as a filter in this case?

Hello Soohan,

I don't think, there are good methods to compensate this current peak - it is indeed part of the switching losses. One may just look for better (faster) diodes, but your choice of RFN1LAM6S is already quite good. (it's promoted as Super Fast Recovery Diode). For lower voltages a schottky diode might still be better, but for your application the choosen diode is good.

Regarding the changes you observe with a more realistic leakage inductance: I'm not really sure if one can describe this with a simple, straight forward explanation.

best regards

Achim