AWR2544LOPEVM: Discrepancy in Velocity Measurement

Hi Kaushik,

The maximum velocity for this configuration is 63 m/s (fold of 126m/s), as shown in the 2D FFT velocity boundaries in the previous image. I'm trying to measure a velocity of -11 m/s, but it is being detected as -22 m/s. I've encountered a similar issue in another experiment where a target moving at 15 m/s was detected at 30 m/s.

Could you please repeat the experiment using RTS, for example?

Thanks,

Shlomi

Hi Shlomi,

I currently do not have access to an RTS. Can you confirm if the velocity measurement is halved only at these values or throughout the range of values of all velocities (both directions)?

Regards,

Kaushik

Hi Kaushik, 

I've observed this phenomenon in all my measurements, both within and outside the range of velocities. The measured velocity is doubled, not halved. Could you perform an experiment in front of a moving car?

Thanks, 

Shlomi

Hi Kaushik,

I have conducted two additional experiments using a different radar, the AWR1843AOPEVM, under the same scenario, and the results are correct.

Attached below are screenshots from mmWave Studio of both experiments. You can observe the detections on the range-Doppler map (top-left subplot) and the velocity of the target in the table on the right side of the image.

The two experiments involved a car driving towards the radar and a car passing the radar. In both cases, the detections accurately reflect the car's velocity, unlike the results from the AWR2544LOPEVM.

It is important to note that the configuration for both chips is similar, with the exception of the sampling rate (25 Msps instead of 40 Msps) and the number of samples (128 instead of 256) to maintain the same chirp time. The only difference is the maximum range.

How can we explain the fact that the AWR2544LOPEVM detects double the velocity?

Thanks,

Shlomi

Exp. 1: Car driving towards radar

Exp. 2: Car Passing Radar

Hi Kaushik,

I came across another interesting issue regarding the bin file size.

Based on the first message in this thread, I calculated the expected size of the bin file for the experiment.

Here are my findings:

BinFileSize = [2 bytes per real sample] x [140 chirps] x [4 Rx] x [1 Tx] x [256 ADC samples] x [4000 frames] = 1,146,880,000 bytes.

However, the size of the bin file saved on disk (as shown in the image below) is exactly half of the expected size. Despite this discrepancy, mmWave Studio displays the correct number of frames, which is 4000.

How can we explain this issue?

Hi Shlomi,

Based on your experimental findings, there appears to be a discrepancy somewhere and I need some additional confirmation here:

1. Can you share the complete chirp configuration that you have used for this particular data capture? (Frame, Chirp and Profile Config)
2. Can you confirm that the velocity step that is calculated matches the chirp design?
3. If you try processing the data by yourself and plot the 2D FFT, do you see any shifted peaks or the same results as the Studio's processing?

Regards,

Kaushik

Hi Kaushik,

Both chirp configurations are described in the image below.

You can see two videos of range-Doppler maps from my processing [link], similar to mmWave results.

The video starting with 2544 relates to AWR2544LOPEVM, and the other one relates to AWR1843AOPEVM.

Let's start with the 1843 video. Run it to frame 2480, and you'll see the target detected with a velocity of ~20 km/h, which is correct.

In the 2544 video, run it to frame 960. The velocity is detected near 100 km/h, which is impossible inside a parking garage since the maximum velocity of the car was 40 km/h. Additionally, there are only 2000 frames instead of the 4000 I requested to capture. This issue doesn't occur with the 1843, which has 4000 frames.

The velocity step is correct in both chips at 0.91 m/s, as designed.

If possible, could you record the 2544 in front of a car moving towards the radar and share the results with me?

Thanks,

Shlomi

Hi Kaushik,

I've noticed an issue in mmWave Studio 3.1.3.1. The slider correctly shows the number of frames, but selecting a frame beyond half of the total (e.g., frame 2001 out of 4000) results in a runtime Matlab error. The range-Doppler map freezes and stops processing, which makes me question the firmware's validity.

I'm currently flashing the appimage from this path:
C:/ti/radar_toolbox_2_00_00_06/source/ti/examples/Out_Of_Box_Demo/prebuilt_binaries/out_of_box_2544.appimage

I have two questions:

1. Is this the correct file to flash?

2. Could you please perform a simple recording with the radar in front of a corner (or nothing) and check if you only receive half of the requested frames?

Thanks,

Shlomi

Hi Shlomi,

Thanks for the additional information. I am however not able to access the drive link on my network.

1. Can you let me know the procedure you are following to capture the data?
   1. It appears you are using the 2544 demo and capturing he LVDS data. Is there any difference between OOB demo performance to what you are observing?
   2. Can you try by manually configuring the device?
2. Again, if you try processing the data by yourself and plot the 2D FFT, do you see any shifted peaks or the same results as the Studio's processing?
3. I can try this experiment on my end, but it will take a few days of time to have the setup ready.

Regards,

Kaushik

Hi Kaushik,

1. I'm configuring the device manually through mmWave Studio and then capturing LVDS data. The web demo doesn't support the 2544 device.

2. The link to the drive includes my own data processing. As I mentioned, the number of frames is half of what I configured to capture, which is causing the issue. Can you please verify if the appimage I mentioned in my last message is the correct one to flash?

3. Could you please try it with your own setup and let me know if it works? It's urgent.

Thanks,
Shlomi

Hi Shlomi,

If you're manually configuring the data, then what appimage is flashed on the device should not matter as the device will be booted in SOP2 mode and the full config will happen over the studio interface.

Can you perform an experiment by modifying different parameters of your data captures such as num samples, num chirps/loops, num frames etc. by keeping the rest as 1? This should be able to identify where the 2x difference is coming from.

Regards,

Kaushik

Hi Kaushik,

I agree, when using mmWave studio, the appimage is irrelevant.

Can you elaborate on the experiment you recommend I perform?

As I mentioned earlier, the 2x discrepancy is due to the number of captured frames being half of the requested value. For instance, if I request to capture 8 frames, I only receive 4. Could you please set up your own 2544 device with some chirp configuration and request to capture 8 frames? You should observe in the mmWave studio that you can only toggle between the first 4 frames. This is the simplest and the first thing I suggest checking.

Thanks,

Shlomi

Hi Shlomi,

Thanks for describing further. Can you confirm what happens if you:

• Configure 1 Frame. Is there no capture here or do you get 1 frame?
• Configure 2 Frames. Do you get 1 Frame here or 2 frames?
• You mention the slider toggle in the postproc window is what I am assuming. Can you confirm the same? Also, does this correlate with the actual data size that you have received?

Additionally, can you mention the version of the mmWave Studio you are using for me to reproduce the same?

Regards,

Kaushik

Hi Kaushik,

Before I proceed with any further experiments, I want to ensure that I am loading the correct MSS and BSS bin files. The paths are detailed below. Could you please confirm them?

Additionally, have you tried it yourself? 15 days ago, you mentioned that it would take a few days.

MSS: C:\ti\mmwave_studio_03_01_03_01\rf_eval_firmware\masterss\xwr2xxx_masterss.bin

BSS: C:\ti\mmwave_mcuplus_sdk_04_06_00_01\mmwave_dfp_02_04_13_00\rf_eval\rf_eval_firmware\radarss\xwr25xx_radarss_rprc.bin

Hi Shlomi,

Load the BSS and MSS FW from "C:\ti\mmwave_studio_03_01_03_01\rf_eval_firmware". Can you provide details for the questions asked by Kaushik?

Regards,

Samhitha