NRF52833 - High power consumption while running edge impulse model

Hi All,

We have developed a model for continuous motion classifier using a 6-axis IMU interfaced with Nordic’s nrf52833 SoC. The model is working fine, but it consumes a lot of power. Kindly help us reduce the power consumption.

Here are some details of application.

  • The motion is sampled at the rate of 50Hz (i.e the controller reads the data from the IMU at 50Hz)
  • The classifier runs once every 200 ms
  • The time taken to run the classifier is 7ms

Fig 1.

In Fig1, we have not yet started the application. Only BLE is active, the current consumed is 42.38 uA, which I think is okay.

Fig 2.

In Fig 2. 6axis IMU starts to acquire data and BLE is also active. In this case the current consumed is ~400uA. The current consumed by the IMU is ~360 uA which is consistent with what is provided in the IMU datasheet, so there is limited scope to reduce the current consumed.

Fig 3.

In this case, we enabled the classifier and the current consumption jumped to 3.28mA, which is high. How can we reduce this when the classifier is enabled. The time taken by the classifier to run is only 7ms and the classifier runs only once every 200ms.

Kindly let me know, if any specific configuration flags need to be set/unset to improve the current consumption.

Thanks,
RBH

Hi @Ranchpal,

Do you go back to sleep/deep sleep mode between each inference?
Though 7ms is quite low, running every 200ms is still a high duty cycle and will drain the battery quickly.
Also could you zoom in on your last figure to have a smaller resolution, like 10ms?

FYI we have released a blogpost on power consumption a few weeks ago: https://www.edgeimpulse.com/blog/analyze-power-consumption-in-embedded-ml-solutions

Aurelien

Hi Aurelien,

Yes, we are entering sleep mode between each inference.

Yes 200ms is still a high duty cycle, but it does not justify 3.28mA of avg consumption as controller when fully active will consume just around ~6mA (from datasheet).

We have been struggling with this for quite sometime now. Is there a private ticketing mechanism or a paid consultation route available for us to avail from edge impulse so that we can share the code base and sort out this issue quickly.

Thanks,
RBH