AWR1642BOOST: algorithm、antenna pattern、reference design code

Hi Henry,

Answers are inline

1. I have found the antenna pattern angle vs. gain figure plot in reference document, but do not find the gain vs. frequency response plot, Is there any available document ?

We do not have any plots showing 'antenna gain vs RF frequency'. What I can share is S11 as a function of frequency. Would that be useful?

2. In reference design code USRR mode, it implemented virtual array in time domain, so the equivalent element number of receive antenna is 8, so the estimation of DoA size is 8 or using 4、4 (independent estimation twice then joint considering)

All 8 'virtual rx antenna' samples are used for the estimation of the DoA.

3. In clustering phase, does the process data (input of clustering) has arranged from small to larger in velocity domain? I have studied the reference code and I think it has different result with the arrangement of velocity.

Could you provide more information on this? (It could be a bug). In practice, the detected object lists are arranged in increasing positive velocity, because of the way we do detection. We process the zero-velocity bin first, and then the next velocity-bin and so on.

4.In tracking phase, we need to estimate the covariance of the measurement and the reference code use to CRLB to estimate the var of the parameters, but I still do not understand how to estimate the noise in range、angle、velocity domain?

Noise (for range and velocity) is simply one of the outputs of the CFAR algorithm. When we perform the CFAR-algorithm, we check whether a given 'Cell-under-test' is larger than it's 'neighborhood noise-floor' by a given threshold. For each detected object,  the noise-floor is used to compute the SNR of that object.

For angle, we use the FFT output to compute an SNR, Given a peak, we find locations of nulls corresponding to the same peak if that peak was due to a pure sinusoid - we use the energy in those nulls to compute the 'noise floor'. Let me know if you have more questions.

5. There are two mode in reference code (USRR/SRR) and can close or open optionally in matlab GUI interface, if I close the USRR mode, What is this mean? It will fill the gap with SRR chirp or just empty ?

The GUI doesn't change anything on the device. It only makes the visualization simpler. So if you disable USRR, the chip will still generate the USRR chirps, and perform the USRR processing, but the GUI will not display the output.

6. Does the reference design do the peak grouping after CFAR? If yes, how its work? Does it find the local max in 3*3 (in 2D -FFT detection block)?

Peak-grouping is used after CFAR in the SRR mode. We don't look for local max in a 3x3 array, instead, we look for local-max first in a 1x3 array (for doppler) and then in a (3x1) array for range)

7. Is the CFAR input define as the is 2*log2(abs(fft))?

You can read the documentation corresponding to the 'log computation function' mmwavelib_log2Abs32 (which is used in secondDimFFTandLog2Computation), but essentially, it would be round(log2(abs(x[i]))*2^8).

8. When using chinese remainder theory to extend the max velocity, how to pair the result for fast chirp and slow chirp ? with power similarity or position similarity? If slow chirp detection result has 4 objects and fast chirp has 3 objects, the result of the output is 3 or 4?

The reference design ignores a lot of corner case such as the ones you described. What it does is detect objects in the fast chirp using CFAR and then match those detected objects with objects in the 'slow chirp' using velocity similarity (given a particular hypothesis), and power similarity. You can see the implementation in the function disambiguateVel().

Thanks for answering!

Best regards,

Henry Lin