2
\$\begingroup\$

I've got a buck-boost power supply design based on the LM5118 by TI. A schematic of the device is shown below:

SMPS Schematic

(Link to TI Workbench)

EDIT Many of the comments mention the layout, hence I've included a screenshot of the top and bottom side of the board here. I've also included the schematic so that the component references match up.

Top side: Top side PCB

Bottom side: Bottom side PCB

Schematic: Schematic 01

Schematic 02

NOTE: The schematic does use slightly different parts than what the TI Workbench suggests. These were calculated by following the process in the datasheet and using the Excel-based calculator TI mention in the datasheet. When this original setup didn't work, the components were replaced to match the ones suggested by the Workbench, but that still hasn't solved the issue.

What I've found is that with a load of 0.5A, between 10-16V input, the device emits a loud whining noise, which appears to be coming from the inductor. At the same time, the output voltage has a 1.5V pk-pk ripple at about 5kHz. Once the voltage is raised above approximately 16V, the noise stops and the power supply appears to be functioning correctly, with a clean 13.5V output.

I've scoped the gate signals for the transistors and noticed that when the power supply is generating that noise, the signals to the gate appear to be discontinuous. The controller will drive the gate at the switching frequency (approx 300kHz) for a period of time then there will be a gap, where the gate is just turned off for a period of time, then the switching resumes. I haven't measured the frequency of this overall behaviour - at a guess, I bet it's not going be too far off the 5kHz.

I had another power supply designed using the same chip that was initially designed for 28V output and worked perfectly fine. The feedback resistors were then modified to provide a 13.5V output and it continued to work fine across the entire voltage range. The only difference between that design and this one were the transistors (SQM120N10-3M8-GE3) and the diodes (MBRS3201T3G).

Any thoughts on what I could look for to help troubleshoot this?

\$\endgroup\$
  • 1
    \$\begingroup\$ try playing around with the loop compensation, the type 2 RC network connected to the COMP pin. Some simple math should help you calculate the right values for the RC compensation network. Section 10.2.2.17 in the datasheet for the LM5118 has more info, the RC feedback components are marked in the [schematic] (imgur.com/XiNsRwL) \$\endgroup\$ – user7994 Sep 17 '15 at 21:47
  • 2
    \$\begingroup\$ Your issue is occurring when transitioning from boost to buck mode, and this strongly suggests loop compensation (including layout) or perhaps output to input coupling via PCB tracks. I will note that ringing occurs at high duty cycles (at least for buck mode), and there is a mode dependency on duty cycle (data sheet figure 10) and that rather reinforces my suspicion. \$\endgroup\$ – Peter Smith Sep 18 '15 at 8:26
  • 1
    \$\begingroup\$ On the transition region from buck to buck-boost - figure 16 shows the transition starting at about 2.5V above V(out), which would fit with your observed 16V clean operation. \$\endgroup\$ – Peter Smith Sep 18 '15 at 8:58
  • 1
    \$\begingroup\$ It might be worth you looking at TI's reference PCB design for this part and comparing how they have done it: ti.com/lit/ug/snva334b/snva334b.pdf \$\endgroup\$ – Andy aka Sep 18 '15 at 11:30
  • 1
    \$\begingroup\$ Having seen apparently benign layouts cause the weirdest problems in SMPS circuits - lets just say I am always suspicious of this set of hidden schematic elements. There are two things interesting in the layout: 1. There appears to be a solid ground plane - if this is solid between pin 1 and pin 20, you could easily have output switch node noise coupling right back into the input which can easily destabilise the circuit. 2. There is a track underneath L1, where you can expect severe noise (shielded inductors aren't that shielded). Look at the Vin pin on a scope. \$\endgroup\$ – Peter Smith Sep 18 '15 at 12:03
2
\$\begingroup\$

Looks like everyone who said it was the loop compensation was right! But I've led everyone on a bit of a wild goose chase - the problem was a mistake on the schematic that must have occurred when it was being modified from one iteration to the next: R6 is meant to connect to COMP, but is connected to FB! So, it was indeed the loop compensation not working properly!

\$\endgroup\$
1
\$\begingroup\$

I suggest looking here: http://www.smpstech.com/problems.htm, the guy seems quite knowledgeable and the site is full of interesting SMPS-related reading.

\$\endgroup\$
0
\$\begingroup\$

Sorry, i am too lazy to read all comments.

Still, i have some thoughts for the subject. DC/DC often have notable ripple of frequency like 0.1 or 0.01 of the switching frequency. That means, that the control circuit does not work very well. First thing you could try is playing with all those caps of compensation network. Also adding a small cap in parallel with top feedback resistor may be really helpful.

Now, in your layout i can't see the MOSFETs. Either i just miss them on my cell phone screen, or they are far from the driver. In that case you are going to experience issues with EMI caused by high gate and drain currents. Be careful with that too.

\$\endgroup\$
  • \$\begingroup\$ Thanks. The transistors are there: look to the left and right of L1 \$\endgroup\$ – Amr Bekhit Sep 18 '15 at 11:15
  • \$\begingroup\$ So they are q1,q3, not m1,m2? \$\endgroup\$ – Gregory Kornblum Sep 18 '15 at 11:18
  • \$\begingroup\$ Yes, that's them. \$\endgroup\$ – Amr Bekhit Sep 18 '15 at 11:20
  • \$\begingroup\$ Just look at all the way gate current must pass to q1. It's probably like 2A with rise time of maybe 20nsec. Pretty awful. \$\endgroup\$ – Gregory Kornblum Sep 18 '15 at 11:21
  • \$\begingroup\$ No no, it's Q2 and Q3. \$\endgroup\$ – Amr Bekhit Sep 18 '15 at 11:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.