• State Resolved
  • Locked Locked
  • Replies 7 replies
  • Subscribers 64 subscribers
  • Views 117 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

LMG5200: Noise on Reference design

Laszlo Techy
Prodigy 90 points
Part Number: LMG5200
Other Parts Discussed in Thread: TIDA-00909, BOOSTXL-3PHGANINV, LAUNCHXL-F280049C, , TIDA-00913, INA240

Tool/software:

Hello,

We have been developing a motor control board using LMG5200s and based on the TI reference design TIDA-00909. The board receives 24V nominally through the main power connector. In addition to powering the LMG5200s, it also provides power for 5V and 3V3 rails (also similarly to the TIDA-00909). These rails power other devices.

We are observing noise issues on all of the power rails that coincide with the SW node transitions. Some of these noise bursts are large enough to disrupt the UART communication between the motor control MCU and other devices, resulting in packet loss whenever the LMG5200s are enabled.

We are investigating what could be done to reduce the noise. Along the way we realized that the reference design is just as noisy as our motor control board. The measurements below were made using BOOSTXL-3PhGaNInv connected to a LAUNCHXL-F280049C. Switching frequency is 10kHz. The motor is commanded to a Home position and is not drawing any significant current (100mA) in a standstill position. This is the minimum noise we observe. When the motors draw more current the noise gets worse.

 Below is a screen capture probing the 3V3 rail using a 1GHz probe (Tektronix TPP1000) and using “paperclip” ground connection to minimize ground loop size. The connection is made between TP1 GND test point and the adjacent 3V3 next to the R50 silkscreen print.

Blue trace is only for triggering purposes (DC coupling, full bandwidth, measuring the SW node of U1 LMG5200). The yellow trace is the 3V3 rail (AC coupling, full bandwidth). The scope capture also shows 3 other saved waveform traces (R1-R2-R3) that illustrate repeatability of the SW node ringdown interference onto the 3V3 rail. (The noise burst 50ns after the initial noise burst shown on the yellow trace only is another SW node triggering.) The magnitude of these bursts (1.4V pk-pk) is unacceptable on the power rail.

The post below discussed slightly related issues, although it was a special application, while ours is just a straight-up motor control, like the reference design.

e2e.ti.com/.../4437367

The recommendations were to

  1. Ensure the measurement technique is tip-and-barrel
  2. Check for PCB layout errors or large inductance loops
  3. Lower the switching frequency

1) Our measurement technique is the closest approximation to tip-and-barrel as is possible to do. 2) The device under test is TI’s reference design and all PCB layout suggestions were followed. And 3) the switching frequency is very low (10kHz).

Yet, it suffers from SW node interference. Do you have any suggestions what we can do to improve noise immunity moving forward?

Best,

Laszlo

  • 0
    TI__Expert 3580 points

    Hi Laszlo,

    thank you for testing the BOOSTXL-3PhGaNInv and your feedback. 

    There are smaller GND bounces (than you report) between the TIDA-00909/TIDA-00913 and the LaunchPad. You can see that on the phase current measurement which is single-ended in the TIDA-00913 design. However these bounces, never caused a fault on the C2000 MCU in my case, please see TIDA-00913 figure 50. The C2000 ADC's measurement error is 2A transient spikes, which translates into a GND noise spike of around 2A/33A*3.3V ~ 0.2V only.    

    How long is the motor cable? Due to the high dV/dt of the GaN-FET, you will likely see phase current oscillations during PWM switching when you use cables longer than around 1m due to transmission line reflections. The frequency depends in the cable length / propagation delay. If this is your use case, could these additional oscillations be an issue? 
    A workaround could then be an LC RF filter (in the report called EMI LRC filter) to reduce the dV/dt slew rate, as shown with TIDA-00909, e.g. figure 9, results with figure 64.
    Note on the filter. I used capacitor placeholders for the inductors L1, as the inductors were not in the Altium library at that time.  

    How do you connect the other MCU to the F280049C through UART? Where does this MCU get 3.3V and GND?

    An option to improve the UART might be to use RS485 to handle common mode noise. Have you tried that?

    Regards,
    Martin

  • 0 in reply to Martin Staebler
    Prodigy 90 points

    Hi Martin,

    I was able to do some additional measurements on the BOOSTXL-3PhGaNInv using tip-and-barrel method between R50 pad and the top of TP1 test point. When making that measurement the noise magnitude is much smaller, now I only see 0.3V pk-pk, as opposed to 1.4V when making the measurement with a short spring clip at the probe tip. So perhaps the noise is not as big as we suspected. 

    For the record, the motor leads used with the evaluation were about a foot long, but that does not seem relevant any more. On our motor control board we do have the EMI filter, and the motor leads are only about 3 inches long.

    On our motor control board we have main power coming in (VBUS) that is directly received by the LMG5200s. This same rail is then stepped down on our board to produce 5V rails and also 3V3 rails to power the MCU. VBUS (and GND) is also passed through a connector to the other board that also steps it down to 5V and 3V3 to power the other MCU on the other end of the UART communications. We have been considering adding a pi-filter between the two boards.

    Using RS-422 is something we have been considering also. We will likely implement it for better noise immunity.

    Best,

    Laszlo

  • 0 in reply to Laszlo Techy
    TI__Expert 3580 points

    Hi Laszlo, 

    thank you for your feedback. It's good to see you measurement fits with what we could see with the C2000 MCU ADC measurements (~0.2V GND bounces).

    RS422 should help you get a more robust board to board communication, ensure you use the corresponding differential line termination.

    Regards,
    Martin  

  • 0 in reply to Martin Staebler
    Prodigy 90 points

    Hi Martin,

    Following up with one more question. In order to further reduce the mentioned ground bounce (even though it appears small), what would be your recommendation? Would increasing the VIN capacitance help? My understanding is that typically ground bounce is associated with an inrush of current at the switching transistor that can locally affect the ground potential. This could maybe be mitigated by added capacitance that provides the charge for this inrush and not affecting grounding elsewhere. But in the case of this LMG5200, it seems that a lot of this noise is though inductive coupling of the high dV/dt switching node. So more capacitance would not necessarily help. Do you have any insight?

    Best,

    Laszlo

  • 0 in reply to Laszlo Techy
    TI__Expert 3580 points

    Hi Laszlo,

    you may experiment increase/decrease the ceramic decoupling caps, which are placed closed to each of the LMG5200, e.g. C4,C5,C6,C7. 
    I have not tried the impact on the transient GND noise when I tested the design though.

    The layout is also important. Ideally we have a star GND connection between analog and power GND and the transient current with the power loop would not impact the analog GND plane. Since we have three half-bridge drivers LMG5200 and each has an internal AGND to PGND connection it was not possible to do an ideal star GND. However as I mentioned already, we did not see an issue with digital CMOS signals like the PWM since the transient GND bounces were in the 0.2V range.

    Regards,
    Martin 

       

  • 0 in reply to Martin Staebler
    Prodigy 90 points

    Thank you, Martin. When you evaluated the ground bounce using the phase current measurement (as opposed to directly measuring voltage on the rails): was that because it is difficult to accurately measure these high-speed transients? As my experience demonstrates, the ground loop size reduction had a huge effect on the measured noise spike voltage magnitude. Would the current measurement chip's output voltage not be susceptible the same way, though? 

  • 0 in reply to Laszlo Techy
    TI__Expert 3580 points

    Hi Laszlo,

    the ADC measurement using the C2000 ADC on the LaunchPad to measure the INA240 analog output over one PWM cycle is a system test and validates the direct impact on potential transient GND noise on the ADC measurement on the Launchpad, which was connected to the TIDA-00909.

    As you could see from the report the analog measurement error was in the range of 0.2V, and should reflect the real system transient GND noise. We can further reduce the impact of PWM switching by using functional isolation between the MCU and the power stage. 

    Regards,
    Martin