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.

AWR1642: Algorithm of TIDEP-0092 reference design and SDK library

Part Number: AWR1642
Other Parts Discussed in Thread: TIDEP-0092

We are using AWR1642 for Vehicle Radar and now simulating  reference design code (TIDEP-0092) SRR part with matlab currently.

Here are some questions about algorithm in reference design and SDK library function

1. Due to multiple received antenna,we have multiple ADC raw data, How to deal with the data with different antenna (data fusion) , if the reference design do some processing between antenna,what method is applied?

2. Does the reference design code do the digital beamforming ? (I have this question due to the sharp antenna pattern but still has the FOV 120 degree)

3. In power detection mechanism,  CFAR has the parameter const1 and const2, how to determine in SRR case ? In most case is given probability false alarm rate and we can calculate the threshold value ?

4. How does peak grouping works in SDK library ? What is the definition of detection matrix ? 

5. In reference design, Fourier beamforming method is applied for DoA estimation, what angle steps and angle interval is applied ? 

6. In reference design, how to calculate frame length in SRR design ? (I tried to calculate by: (effective chirp time + idle time )*number of chirps  +  (effective chirp time + idle time )*number of chirp) fast chirp/slow chirp  (51us +3us)*64+(51us +14.8us)*64 .The results does not match frame length mentioned above.  

7. In SDK library, there is a function called extendedmaxvelocity, I have some question about what SDK user guide mentioned. As follow:

"A simple technique for velocity disambiguation is implemented. It corrects target velocities up to (2*vmax). Enabling this feature results in loss of multiObjBeamForming feature."
Why?
 
Thanks for replies, the answer may help me a lot!
 
Best regards,
Henry Lin
  • Hi Henry

    1. Due to multiple received antenna,we have multiple ADC raw data, How to deal with the data with different antenna (data fusion) , if the reference design do some processing between antenna,what method is applied?

    1a. As in the OOB, the SRR design, performs (at a detected range Doppler bin) an FFT (called the azimuth FFT) across the different Rx channels to estimate the angle. Have you gone through the mmwave training series? [[View:https://training.ti.com/mmwave-training-series]:1230:0] It explains the basic processing flow for FMCW radars.

    2. Does the reference design code do the digital beamforming ? (I have this question due to the sharp antenna pattern but still has the FOV 120 degree

    2a. Yes.

    3. In power detection mechanism,  CFAR has the parameter const1 and const2, how to determine in SRR case ? In most case is given probability false alarm rate and we can calculate the threshold value ?

    3a. The false alarm rate can be estimated from the CFAR threshold which is what we program. If you refer to the user guide (search for cfarCfg), you'll see the definitions of const1, and const2. In general, N = const2, and threshold or alpha = const1 * 6/512.

    4. How does peak grouping works in SDK library ? What is the definition of detection matrix ?

    4a. The detection matrix is simply the log of non-coherent sum of the radar cube across the antenna dimension. Please review the angle estimation, and detection portions of the mmwave training.

    5.  In reference design, Fourier beamforming method is applied for DoA estimation, what angle steps and angle interval is applied ? 

    5a. In addition to a 32-pt FFT, we also do quadratic interpolation.

    6. 6. In reference design, how to calculate frame length in SRR design ? (I tried to calculate by: (effective chirp time + idle time )*number of chirps  +  (effective chirp time + idle time )*number of chirp) fast chirp/slow chirp  (51us +3us)*64+(51us +14.8us)*64 .The results does not match frame length mentioned above.

    6a. Note that the reference design has two subframes (one called SRR and the other called USRR). Also, note that the active frame time (i.e. the time when the radar transmits) is much smaller than the frame repeat time (which is set to 30ms).

    7. In SDK library, there is a function called extendedmaxvelocity, I have some question about what SDK user guide mentioned. As follow:

    "A simple technique for velocity disambiguation is implemented. It corrects target velocities up to (2*vmax). Enabling this feature results in loss of multiObjBeamForming feature."
    Why?
    7a. Please read the documentation for the max-velocity extension method to understand how it works in detail (The documentation is quite good). However, to answer your question, multi-object beam-forming allows one to detect more than one target in a range velocity bin. The extendedmax-velocity allows one to test two different velocity hypothesis on one range-velocity bin (i.e. one hypothesis stating that objects in the range-velocity bin are below the max-unambiguous velocity and one hypothesis that objects in the range-velocity bin are above). If there are more than one object in a range-velocity bin and one is travelling below the max-velocity and one is travelling above the max-velocity, the extended max-velocity method will not work for both objects. Only the stronger object will be correctly disambiguated.
    Regards
    Anil

  • Hi Anil:
    Sorry about my not clearly description, I have using matlab and simulink (phased array tool box) to model the over all DSP processing about reference design (TIDEP-0092)only the SRR part, including envirmonenrtal setting (targets model, FMCW chirp parameters about SRR (fast and slow ), thermal noise, antenna pattern) and DSP processing flow.
    I have take the online training course before, but still have many detail questions about the DSP flow. My simulation model is able to estimate range, velocity and angle.
    Processing flow is as follow
    1. Capture the ADC raw data from 4 receive antennas and doing 2D fft (in chirp and between chirp) with each antenna Individualy. (So I have 4 2D FFT matrix)
    2. Take "abs" operator with FFT matrix then sum (Without information about digital beamforming before)
    3.Two direction CFAR (CA-CFAR and SOCA-CFAR) is applied with the result in 2 .
    4.Peak grouping (local max finder function ) is applied two determine which index do have targets.
    5.Angle estimation is applied with the result (beat frequency index) between antenna to get the targets DoA.
    Above procedures are for fasr chirp, and same processing flow for slow chirps.
    6.Chinese remaineder is applied to extend max velocity.
    After the process flow, I can get the range, velocity and angle estimation results with multiple targets .But there still many detail questions about my model.
    1.What digital beamforming method is applied ?
    2.If doing digital beamforming, how many frames do I need to estimate the parameters? (steering beam in one frame data or between frame data )(Range velocity and angle)
    3.For angle estimation for doing the digital beamforming, it segment the FoV into many intervals, and just doing 32-point in that interval, is that right ? If yes, what are the intervals? (Ex: 0-30 degree 30-60 degree ...)
    4. I have question Frame length in SRR design(7.3ms),how to calculate this parameters with fast and slow chirp configuration ?
    5. Due to Fourier beamforming angle estimation method , I need to steering beam in an interval, how is the angle step and interval (Similar with question 3 )
    6.Is it possible to order ADC capture card DCA-1000 now, if not, when is available?

    Thanks for replies, the answer may help me a lot!

    Best regards,
    Henry Lin
  • Hi Henry,

    I am also having questions. The output of 'digital beamforming' is angle of the target object (or DoA). If you are able to find the angle estimate (i.e. DoA) , why do you need to do 'digital beamforming'? Are you refering to Tx beamforming? If so, we do not do Tx-beamforming in the reference design.

    4. I have question Frame length in SRR design(7.3ms),how to calculate this parameters with fast and slow chirp configuration ?

    Ans: You can find the exact parameters for the SRR configuration in the file \srr\common\srr_config_chirp_design_SRR80.hm where we have the following two lines. Each parameter is 10ns. I have copied the important parameters below

    #define PROFILE_SRR_IDLE_TIME_VAL           (300U)

    #define PROFILE_SRR_RAMP_END_TIME_VAL       (5600U)

    #define CHIRP_SRR_0_IDLE_TIME_VAL             (0U)

    #define CHIRP_SRR_1_IDLE_TIME_VAL             (1180U)

    In other words, each 'fast chirp' duration is 59 us, and each 'slow chirp' duration is (59+11.8). Each type of chirp is repeated 64 times. So, the  total frame time is (59*64) + (70.8*64)  = 8.3ms. Note that the 'ramp-end time' includes the settling time as well, whereas 'effective chirp time' doesn't. The 'effective chirp time' is the time during which we sample the signal. The 'ramp-end-time includes the 'ADC start time' and the 'effective chirp time'.

    We typically start sampling a few microseconds after the ramp has started so that the ramp (and the digital front end) have stabilized.Hence the reason for a non-zero 'ADC start time'.

    The documentation has an error (should be 8.3ms not 7.3 ms). Thank you for finding it.

    Regards

    Anil

  • Hi Anil,

      Thanks for your answers. In SRR design, there is only one active Tx antenna, so it not possible doing beamforming,

    Here are two more questions:

    1.Can you describe more detail about interpolation in DoA estimation.

    2.As your above descriptions,"digital beamforming" mean "fourier beamforming" for angle estimation, is that right ? Or there is another process algorithm about beamforming?

    Thanks a lot.Your answers are very helpful for me!!

    Best regards,

    Henry Lin

  • Hi Henry,

    1. The DoA estimation in 'SRR TI Design' is simply an oversampled FFT. It is performed across received channels for a range-doppler bin that has been detected using CFAR-CA. In the case of the SRR TI design, a 32-pt FFT is performed over 4 Rx antennas.After this step, we compute the magnitude of the FFT output, and then perform quadratic interpolation around the peak.

    The benefit of quadratic interpolation is that the angle estimate accuracy is improved (the accuracy is equivalent to a much larger FFT).

    2. We do 'fourier beam forming'.

    Regards

    Anil