• State Answer Suggested
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 62 subscribers
  • Views 1542 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

LM3478 based SEPIC Supply Misbehaves after 3rd PCB spin

NORMAN TRACY90415
Intellectual 400 points
Other Parts Discussed in Thread: LM3478

LM3478 based SEPIC Supply Misbehaves after 3rd PCB spin

Two years ago with Web Bench I designed a SEPIC supply using LM3478 to output 15 VDC @ 0.5A from an input of 3 to 15 VDC. The power supply feeds a 30 watt class-D audio amp IC including 450 uF decoupling capacitance on the chip’s power input pins. The schematics are shown below. Now on our third PCB layout the power supply + audio amp design has worked very well in lab and field testing.

 

SEPIC supply using LM3478 9 VDC in to 15 VDC Out

Supply load is 30 watt class-D audio amp with 450 uF decoupling capacitance

While checking the third revision of the PCB an anomaly was noted. Very large current inrushes on the input and larger than normal supply currents used during operation.

NOSCAP edited.bmp

The top trace of the scope capture shows PVDD the 15 VDC output of the supply come on under MPU control one division in. The middle trace is a clamp on current probe monitoring the circuit’s DC input current, when PVDD switches on 19 A(!) spikes are seen for one major time division. After 100 mS the DC input current drops to ~12 A, note this occurs when signal appears on the bottom trace labeled V@FFR. V@FFR is the amplifier’s output across the load, seriously under sampled at this scope setting it is actually a full scale 10k sine wave. Therefore we see this misbehaving power supply circuit pulling 19 A peaks while being asked to supply 15 V @ ~12mA to an idle power amp then dropping to 12 A peaks 5 A averages when the power amp begins delivering power to its load.

amp_on_0edited.bmp

In the scope capture above time base is 5x faster to look at the time prior to load being applied. I also had my teammate who writes the firmware for this device separate the turn-ons of the power supply and power amp to check the effect of the amp on the supply. Scope triggers from PVDD turning on with its load off. After 100 mS the MPU turns on the power amp as shown on the CH1 yellow trace, based on the frequency of the ~20A spikes on the DC input current probe the power supply circuit takes no notice of the amp turning on. After another 100 mS the MPU begins outputting the 10k sine wave. The power supply circuit does notice the signal turning on and we see the expected frequency increase of the current being drawn from DC input and a welcomed reduction in amperage.

After other circuit checking and debugging efforts I looked at the circuit asking ‘where is all that current going’. The path from Q2 source to ground via the sense resistor looked likely. I tomb stoned the 18 mOhm Rsense_1 and added a wire loop to allow the current probe to be clipped on.

Isense_0 edited.bmp

The scope capture appears to show the excessive current being drawn from DC in is ‘shooting thru’ (to borrow a phrase from H-bridge circuits) Q2 and Rsense_1 to ground.

IsenseZnm edited.bmp

Zoomed capture of current at Q2 and Rsense_1 to ground node while circuit under load delivering signal to amplifier’s output load. Current through Q2 and Rsense_1 is peaking at 10 A averaging 4.68 A.

At this time I welcome all suggestions as to why this circuit has gone bad in its adolescence. Is my theory the excessive current being drawn is ‘shooting thru’ Q2 and Rsense_1 to ground plausible? Or is it just charge flowing back into the input inductor via ground? What has made this good circuit go bad?

LM3478 3 to 15VDC in 15VDC out PS no2 Webench Design and Reports.pdf

  • 0
    TI__Expert 4100 points
    Hi Norman,
    We are looking into this and will get back to you soon.

    In the meantime if you could clarify few questions :
    1. what (if anything) changed from previous board design?
    2. Could you please send schematic for 2nd revision (if there are any changes)?
    3. Can you please try to disconnect power amplifier (including input capacitors) from output of the converter and observe part behvaior? Do you still see excessive inrush current or not?

    From your description part is running in very light load mode (15V, 12mA) immediately after power up. In this mode of operation larger input current is possible, but i would not expect to see ~20A.

    Thank you
  • 0 in reply to Brani Dubocanin
    Intellectual 400 points
    Hello Brani,
    Thank you for the rapid response. For now I suggest setting this aside while I look into something at my end. I believe I have identified a PCB layout issue causing excessive impedance in the nodes feeding power into the LM3478 based supply. Once I have verified or disproved this theory on my bench I will report back to this forum.
  • 0 in reply to NORMAN TRACY90415
    Intellectual 400 points

    October 12th Update

    A variety of possible solutions were investigated. Inadequate current carrying capability in the PCB layout was a promising theory when close inspection revealed places where the high current battery voltage input feed jumped layers via single vias and traversed traces that should be wider for the current carried. Wires added to the single via necked down current paths had no effect. Reverse voltage and switch-on circuits upstream of the misbehaving power supply were checked and tests run bypassing them, again to no effect. The breakthrough came when on the lab bench it was noticed that for about 10 mS after power-on the supply output the desired 15 VDC pulling the expected low current then the current would jump to the multi-amp levels shown above. What happens in the first ten-ish milliseconds and then changes? The microprocessor powers up, loads its code, and configures its GPIO pins. This called attention to the HVPOWEROFF signal connecting LM3478 pin 7 FA/SD to the devices MPU. Could the MPU’s GPIO be loading down the dual purpose FA/SD (Frequency Adjust/Shut Down) pin and affect the LM3478’s behavior? The answer is yes it can and that was the source of the over current with loss of regulation condition.

    Recall that in modern MPUs the GPIO pins can not only be programmed as input or output but also their electrical characteristics are under software control. Also note in this design TI’s advice to use a MOSFET and 10k pull-up resistor to switch the part on then isolate the frequency set resistor from the MPU was not noted or heeded.

    LM3478 datasheet Figure 26 Shutdown Operation in Frequency Adjust Mode

    Compared to the recommended connection the misbehaving circuit (shown in the initial post above) connects RFA directly to ground and connects the MPU’s GPIO pin above RFA directly to FA/SD. This depends on the MPU’s GPIO pin being able to hold FA/SD high to turn off the LM3478 and ‘disappear’ i.e. present very high impedance while the LM3478 is operating so RFA can set the part’s base switching frequency. On the software side with the original enable/disable circuit for the LM3478 the MCU IO pin was driving the LM3478 pin high to disable and low to enable. However by driving the pin low the IO pin was sinking current meant to go through RFA and set the switching frequency.

    After trying a couple of MPU GPIO configurations the software solution was to drive the IO pin high to disable the LM3478 and to drive the pin low then tri-state the pin for normal LM3478 operation. Going immediately from high to tri-state was not satisfactory, apparently first asserting a low draining current off the node so when the MPU pin is tri-stated RFA sees the correct impedance to ground. The following ‘scope capture shows circuit operation after the software fix

     

    This is a vast improvement, the 19 amp current spikes prior to load being applied are gone and when the signal is output to the load (at ~220mS) current average drops to 1.73 amps. Two areas of concern remained the slowed turn-on of the supply and loss of regulation or switch off at 1100 mS.

    When the software fix was tried on a second PCB assembly it was not nearly as effective. The 15 VDC output would collapse after a few hundred mS and drop low enough that output to the load suffered. After other debugging showed critical parts like Csep on the 2nd assembly were OK looking at the FA/SD node with a scope showed the 15 VDC output would collapse when noise appeared on the node. With the supply on FA/SD was very close (@ ~1 VDC) to the 1.3 VDC that TI’s specifies as the turn off threshold and when noise appeared was exceeding the threshold. To gain some noise margin the circuit was reconfigured adding a 10k pull down and moving where the GPIO connects as shown in the following schematic.

    Results with the added 10k resistor and moved GPIO connection are shown in the following two scope captures.

     

    Note improved rise time of the power supply’s output and that it stays on until the output ceases and the MPU turns it off. Also note further reduction in current from 1.73 to 1.23 amps.

    Zoomed in to examine the ~180 mS period when the MPU turns on the supply prior to outputting signal to the power amp that is the supply’s load. After the turn on spikes there is no sign of other spikes. Rather it is good to see that within the resolution of the clip-on current probe no difference in current drawn from the battery is detectable vs. only running the MPU’s 1.8 VDC supply and with 15 VDC supply on and idle.

    Thus closes the strange case of 20 amp current shoot through and the LM3478. The changes detailed above will suffice for now. When I spin the PCB it will include the MOSFET isolated connection between FA/SD and MPU as shown in the datasheet. The takeaways include:

    • It never hurts to reread the datasheet one more time even when the project is rushed.

    • Beware the dual function pin connected to the processer’s I/O pins.

  • 0 in reply to NORMAN TRACY90415
    TI__Intellectual 2225 points
    Hi Norman,
    I'm glad to hear you were able to solve the problem, and thanks for sharing that story with us! I'm sure it will be helpful to someone else down the road.
    Perry