AWR1642: some questions about SRR demo

Shine

Part Number: AWR1642

Hello Champs,

Customer has some questions about SRR demo.

1. Customer met the same issue as below post. Is it a bug in exp2sp function?
https://e2e.ti.com/support/sensors-group/sensors/f/sensors-forum/911736/awr1642-exp2sp-calculation-result-issue

2. What's the function of quadraticInterpLog2ShortPeakLoc()?

Why the input is cfarDetObjIndexBuf? Because there are 3 coordinate information in the space?

interpOffset = quadraticInterpLog2ShortPeakLoc(obj->cfarDetObjIndexBuf, obj->numDopplerBins, detIdx1, (CFARTHRESHOLD_N_BIT_FRAC + obj->log2numVirtAnt))

3. What is the formula for antilog2 calculation?

Thanks
Regards
Shine

over 2 years ago

0 Kaushik Gowda over 2 years ago

TI__Mastermind 20605 points

Hi Shine,

Please give me until Monday to get back to you on these queries.

Regards,

Kaushik

0 Kaushik Gowda over 2 years ago

TI__Mastermind 20605 points

Hi Shine,

As mentioned in the previous thread, the unit functionality of the exp2sp function has been validated in the test report. Can they confirm if their usage is right?
This function uses a quadratic fit to estimate fractional bin for a more accurate estimate of the peak value. The input is cfarDetObjIndexBuf because we are trying to perform a fit on the list of detected bins only after the 1D CFAR stage.
2^(input/(2^fracBitIn)) {Basically perform a 2^x, in this case input is a fixed-point number}

Regards,

Kaushik

0 Shine over 2 years ago in reply to Kaushik Gowda

TI__Genius 13655 points

Hi Kaushik,

Thank you very much for your great support.

Below is the customer's reply.

1. Below is the snapshot of customer's test.

The test result

2. Is it only using coordinate information to fit? Is there some document for fitting?

3. Would you pleas elaborate the specific information for 2^(input/(2^fracBitIn))? How it is equivalent to 2^x, because all they using is floating point operations.

Thanks.
Regards
Shine

0 Kaushik Gowda over 2 years ago in reply to Shine

TI__Mastermind 20605 points

Hi Shine,

The usage seems to be fine and straightforward. Would it be possible to share this standalone test script with me? I can try to connect with the right team to get some more info on this. Please expect a delay in support concerning this.
It uses the bin number or index for the fit. Unfortunately, there is no document explaining this and customers are requested to go through the source code and understand the algorithm.
Let's assume your fixed number has no fractional bits at all. Then 2^(input/(2^0)) = 2^input where x=input. The function here takes input as unsigned integers and returns a floating-point number based on the evaluation.

Regards,

Kaushik