AWR1642: High Accuracy Demo; Phase Estimation Correction Algorithm

Alex Shirley

Part Number: AWR1642

TI Team,

There is a section of code in the AWR1642 High Accuracy Demo, specifically in the RADARDEMO_highAccuRangeProc_rangeEst function in RADARDEMO_highAccuRangeProc_priv.c file which was left in the source, but was commented out. I'm trying to understand the underlying algorithm behind this, because when I uncomment the code, the accuracy drops off significantly, and I'm not sure if that's a side effect of what is going on, or if there is an error in the source. Is there an associated white paper for this algorithm or what?

if (highAccuRangeHandle->enablePhaseEst) {
            float phaseCoarseEst1, phaseCoarseEst2, phaseInitial;
            double PI = 3.14159265358979323846f;
            double real, imag, denom, initReal, initImag, dtemp1, dtemp2, dtemp;
            float phaseEst, totalPhase = 0.f, phaseCorrection, rangePhaseCorrection;
            __float2_t demodSig, corrSig;

            inputPtr = (__float2_t *)highAccuRangeHandle->inputSig;
            phaseCoarseEst1 = 2 * (float)PI * highAccuRangeHandle->fc
                    * (divsp(2 * (*estRange), (float)3e8) + highAccuRangeHandle->adcStartTimeConst);
            phaseCoarseEst2 = (float)PI * highAccuRangeHandle->chirpSlope
                    * (divsp(2 * (*estRange), (float)3e8) + highAccuRangeHandle->adcStartTimeConst)
                    * (divsp(2 * (*estRange), (float)3e8) + highAccuRangeHandle->adcStartTimeConst);
            phaseInitial = phaseCoarseEst1 - phaseCoarseEst2;

            denom = divdp(1.0, (double)highAccuRangeHandle->fft1DSize);
#if 0
            dtemp1 = cos((double)phaseInitial);
            dtemp2 = sin(-(double)phaseInitial);
            initReal = cos(2.0 * PI * highAccuRangeHandle->chirpRampTime * (double) freqFineEst * denom);
            initImag = sin(-2.0 * PI * highAccuRangeHandle->chirpRampTime * (double) freqFineEst * denom);
#else
            dtemp1 = cosdp_i((double)phaseInitial);
            dtemp2 = sindp_i(-(double)phaseInitial);
            initReal = cosdp_i(2.0 * PI * highAccuRangeHandle->chirpRampTime * (double)freqFineEst * denom);
            initImag = sindp_i(-2.0 * PI * highAccuRangeHandle->chirpRampTime * (double)freqFineEst * denom);
#endif

            //sample @ t = 0;
            corrSig = _ftof2((float)dtemp1, (float)dtemp2);
            demodSig = _complex_mpysp(_amem8_f2(inputPtr++), corrSig);
            RADARDEMO_atan((cplxf_t *)&demodSig, &phaseEst);
            totalPhase += phaseEst;

            if ((highAccuRangeHandle->enableFilter == 0) && (highAccuRangeHandle->enableLinearFit == 0)) {
                //sample @ t = 1;
                dtemp = dtemp1 * initReal - dtemp2 * initImag;
                imag = dtemp1 * initImag + dtemp2 * initReal;
                real = dtemp;
                corrSig = _ftof2((float)real, (float)imag);
                demodSig = _complex_mpysp(_amem8_f2(inputPtr++), corrSig);
                RADARDEMO_atan((cplxf_t *)&demodSig, &phaseEst);
                totalPhase += phaseEst;

                for (j = 2; j < (int32_t)highAccuRangeHandle->fft1DSize; j++) {
                    dtemp = real * initReal - imag * initImag;
                    imag = real * initImag + imag * initReal;
                    real = dtemp;
                    corrSig = _ftof2((float)real, (float)imag);
                    demodSig = _complex_mpysp(_amem8_f2(inputPtr++), corrSig);
                    RADARDEMO_atan((cplxf_t *)&demodSig, &phaseEst);
                    totalPhase += phaseEst;
                }
                phaseCorrection = totalPhase * (float)denom;
                rangePhaseCorrection = divsp((phaseCorrection * (float)3e8),
                                             (4.f * (float)PI * highAccuRangeHandle->fc));
                *estRange += rangePhaseCorrection;
            } else {
                //Not implemented yet
            }
        }

over 8 years ago

0 Cesar over 8 years ago

TI__Guru**** 151884 points

Hi

As you will notice in the code, two different functions are used to compute the sine and cosine values. The commented code uses sin(), cos() where the code that is being used uses cosdp_i() and sindp_i()

This is not related to the algorithm, it is related to the precision. The functions cosdp_i and sindp_i are probably double precision.

Thank you
Cesar

0 Alex Shirley over 8 years ago in reply to Cesar

Intellectual 390 points

That's not the issue, the entire posted section of code is the commented section from the RADARDEMO_highAccuRangeProc_rangeEst function - the entire function is so long, I thought everyone would get lost in the rest of the code.

0 Michelle Liu over 8 years ago

TI__Genius 11415 points

The code was initially used for phase based range refinement. It shows good result in simulation but in reality, it's not good for performance evaluation because of the lack of impairments modeling. Therefore the code is commented out and not intended for customer to use.

Thanks and Regards,

Michelle

0 Alex Shirley over 8 years ago in reply to Michelle Liu

Intellectual 390 points

Do you have a recommendation for an impairments modeling textbook?

0 Radio-Joe over 8 years ago in reply to Alex Shirley

TI__Mastermind 37500 points

Section 4.5 of the Precision FMCW Short Range Radar for Industrial Applications, discusses Phase topics for FMCW, but not modeling.
Regards,
Joe Quintal

Sensors

Sensors forum

AWR1642: High Accuracy Demo; Phase Estimation Correction Algorithm