Channel Sounding with nRF54L15 – Poor Accuracy Compared to Nordic's Demo

Hi Nordic team,

I'm currently working on a real-time distance measurement project using channel sounding with the nRF54L15 DKs. I’ve been using the channel_sounding_ras_initiator and channel_sounding_ras_reflector samples from your official GitHub repository, and I’ve managed to get the system running successfully.

However, the performance I’m getting in terms of distance accuracy and responsiveness is noticeably worse than what’s shown in your YouTube demo (https://www.youtube.com/watch?v=dbQKqXN80ms). In your video, the distance tracking appears smooth and accurate, even when the devices are in motion.

In my case, the measurements are much more noisy, and there’s a significant delay (sometimes several seconds) between physical movement and updated distance readings. I’ve only made minor modifications to the sample code—such as increasing buffer sizes and applying a median filter to reduce noise.

My questions are:

  1. Is the code used in the YouTube demo exactly the same as the channel_sounding_ras_initiator and channel_sounding_ras_reflector samples, or was it modified for the demo?

  2. Are there specific hardware requirements or environmental conditions (e.g., antenna placement, orientation, shielding) that are crucial for achieving the best performance?

  3. Is any calibration or parameter tuning needed beyond what is included in the sample projects?

I’d really appreciate any tips or insights on how to get results closer to what you demonstrate in the video.

Thanks in advance,

Ibai

Parents
  • I have a similar impression when running the same samples without modifications in the latest nrf sdk version (also tried stepping back to 3.0 exactly) with two nrf54l15dk. The IFFT value is much better than the PBR and RTT value, and it seem fairly good at times when the initiatior and reflector are oriented towards each other and nothing else is obstructing them. But just some small adjustment can make the value jump by several meters at times or when I walk around in a room it's off by several meters. I suspect the demo was in quite good conditions, but I can't say for sure.

    Based on my reading of appnotes from other manufacturers is that you need more antenna paths for good accuracy. Two antennas on either side seems to be a sweet spot, then you get four antenna paths in total. As far as I understand, Nordic supports multiple antennas in software, but has no hardware for it (or any hardware reference designs or recommendations).

    Also, the algorithm that calculates the distance from the raw values play a role. I don't know how big compared to the antenna setup at this point, but there are companies focusing on just developing proprietary algorithms for this. Nordic's efforts towards this appears to be a work in progress with some significant improvements made in 3.0.

    Someone from Nordic, please correct me if I'm wrong.

Reply
  • I have a similar impression when running the same samples without modifications in the latest nrf sdk version (also tried stepping back to 3.0 exactly) with two nrf54l15dk. The IFFT value is much better than the PBR and RTT value, and it seem fairly good at times when the initiatior and reflector are oriented towards each other and nothing else is obstructing them. But just some small adjustment can make the value jump by several meters at times or when I walk around in a room it's off by several meters. I suspect the demo was in quite good conditions, but I can't say for sure.

    Based on my reading of appnotes from other manufacturers is that you need more antenna paths for good accuracy. Two antennas on either side seems to be a sweet spot, then you get four antenna paths in total. As far as I understand, Nordic supports multiple antennas in software, but has no hardware for it (or any hardware reference designs or recommendations).

    Also, the algorithm that calculates the distance from the raw values play a role. I don't know how big compared to the antenna setup at this point, but there are companies focusing on just developing proprietary algorithms for this. Nordic's efforts towards this appears to be a work in progress with some significant improvements made in 3.0.

    Someone from Nordic, please correct me if I'm wrong.

Children
No Data
Related