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.

LM5046 MOSFET drive outputs unstable under low-load conditions.

Other Parts Discussed in Thread: LM5046, LM5045

We are using an LM5046 to provide isolated +/-82V dc rails to power an IRS2092-based 100V output full-bridge PA/Voice alarm class-D amplifier.  The amplifier stage works well in providing a direct drive 100V rms audio output up to 220W.

Our problem is with the LM5046 isolated full-bridge dc-dc stage. We have two amplifier channels on our PCB, each channel having two isolated dc-dc stages (a high power +/-82V stage, and a low-power flyback +/-12V stage).  We have no problem with the 12V low power flyback stage.  The input power is from 21V to 30V dc.

We have no problems with one of the high power dc-dc stages on our 4-layer pcb, but the high-power (250W) dc-dc stage on our second channel does not control the phase-shift of the full-bridge stage correctly at low power levels, causing an asymmetric drive to the primary of our transformer, resulting in incorrect operation and failed MOSFETs.

Given that the first channel works OK, and that the second channel only has slight layout differences compared to the first on our PCB, I suspected that we were getting noise on the board affecting the switching.  Note that each channel has its own dc input connectors, so I can power up just one of the channels at a time.

To investigate this I removed our dc-dc transformer on this channel, and replaced it with a 2.7k resistor, so that I could monitor the differential waveform at the bridge output with a 'scope, with only low currents flowing on the PCB tracks and ground plane.  To produce an artificial feedback signal I connected a 47k potentiometer across the transistor of the NFB opto-coupler so that I could emulate the NFB signal.

Looking at the effective PWM waveform across the load resistor I saw that at full power the two driving square waves are operating correctly.  As I reduced the potentiometer resistance value one of the drive waveforms shifted its phase to produce the expected PWM waveform across the load resistor.  However, once the differential signal was showing the equivalent of approx. 30% PWM width, ie the transferred power was getting lower, the phase-shifted  drive signal suddenly changed from the correct 50% duty cycle one to a duty cycle of approx. 70%.  This is harmless into a resistor load, but is catastrophic into a transformer load as the primary is now being driven asymmetrically, resulting in saturation of the ungapped ETD39 ferrite core.

As the other channel on the PCB works correctly, (and I have tried another LM5046 on the faulty channel to prove it isn't the chip) I can only assume that we have some sort of noise coupling problem.  Can you give me any idea of what might cause this.  It does not seem to be the over-current trip, as the resistor replacing the transformer primary is only taking a very small current.  I have found that the 33k resistor on the RT pin is very sensitive, as if I put the tip of a 'scope probe on the RT net the switching frequency increases considerably. This net trace on the PCB is less than 10mm long.

I have attached a .PDF of the dc-dc converter circuitry. (Please let me know if this isn't attached so that I can email it to you directly)

Regards,

Chris Morriss.

LM5046 dc-dc stage.pdf

  • Having just read what I have written, I hope you are not confused between my temporary load resistor, and the potentiometer I attached to emulate the NFB signal. (I don't see how I can edit my original post)

    ORIGINAL POST NOW EDITED TO REMOVE AMBIGUITY (I HOPE).

  • Is a TINA SPICE model available for the LM5046? I tried using the PSPICE model you have, but TINA throws up a syntax error when tried to bring the .LIB file into TINA.
  • Hi Chris,

    Unfortunately we don't have the TINA model for LM5046, and I don't think .LIB can be imported to TINA.

    For the output unstable issue, one possibility is the loop phase margin not enough, to prevent the a circuit from oscillation, I suggest measure the gain and phase with a network analyzer and finally adjust the compensation components, ensure the phase margin of at least 45 degrees and gain margin of -12 dB, 60 degrees of phase margin is better. TI has several design notes and application papers on the subject of compensation such as www.ti.com/.../slva662.pdf.

    Best Regards
    Oliver
  • No, it's not a gain/phase issue around the feedback loop, and is most probably due to unwanted coupling from the HS1 node to some other sensitive pin on the chip. The problem is that I can't determine what pin is picking up the noise!

    I took out the transformer, and used a 2.2k resistor load to replace the transformer primary.  I then ran the converter open-loop, but used a variable resistor across the opto-coupler output transistor to emulate the effect of the NFB.  The problem with asymmetric drive on the HS1/LO1 outputs is still there when the converter is only providing a low output power. The drive signals on the HO2/LO2 outputs remains at the correct 50% duty cycle at all times.

    I may try the use of the conventional PWM output LM5045 (pin compatible) to replace the LM5046 and see if this behaves properly.

    As the converter outputs a high voltage (82V), I am using conventional high-speed diodes as the rectifier, and not using the synchronous rectification capability of the chip.  As far as I know, this should not affect the operation of the drive signals to the mains power switching full-bridge MOSFETs though. (The SSSR pin is connected to 0V to disable the sync rectifier operation).

  • After much thought and cutting and patching of fine PCB tracks I have found the solution.
    The four gate drive lines to the bridge MOSFETs from the LM5046 where picking up noise from the outputs of the bridge, as the bridge output tracks were running close to the gate drive tracks. Separating these signals cured the problem. It appears that the noise is internally getting into the chip on the gate drive lines and screwing up the gate drive logic circuitry.

    An analogous effect also occurs on the LM5045, which is not surprising as most of the internal circuitry of the two chips is the same.