PVACaponROIPreprocessParams#

Defined in public/src/operator/include/PvaOperatorTypes.h

struct PVACaponROIPreprocessParams#

Parameters for the Capon DoA-on-ROI beamspace-projection preprocessing operator.

Shared host/device GP struct (passed via prog[“GP”].set). Geometry is fixed for a run, so the DFT sizes are precomputed on host and passed here. See OpCaponROIPreprocess.h.

Public Members

int32_t numTx#

transmit channels

int32_t numRx#

receive channels

int32_t numChirps#

Doppler dimension of the range-Doppler cube.

int32_t numAzVirt#

virtual azimuth aperture, usually numTx times the azimuth RX element count

int32_t numElVirt#

virtual elevation aperture, usually the elevation RX element count

int32_t beamAz#

beamspace azimuth bins (coarse-DoA DFT length); independent of numAzVirt

int32_t beamEl#

beamspace elevation bins (coarse-DoA DFT length); independent of numElVirt

int32_t roiAz#

ROI azimuth size in beamspace bins.

int32_t roiEl#

ROI elevation size in beamspace bins (svDim = roiAz*roiEl)

int32_t svDim#

beamspace channels per ROI = roiAz*roiEl (at most 16)

int32_t numAzBins#

steering-grid azimuth directions per ROI (1 to 16)

int32_t numElBins#

steering-grid elevation directions per ROI (1 to 8)

int32_t numDopplerBinsCovMat#

snapshot height H (FBA-expanded Doppler count)

int32_t dopplerNeighborhood#

local Doppler window, >= 2 (effective = 2*floor(N/2))

bool enableFBA#

true: forward-backward averaging (doubles the snapshot count)

int32_t inputQBits#

snapshotsROI fixed-point Q, in [0, 30]

int32_t steeringQBits#

steeringLut / steeringVectorsROI fixed-point Q, in [0, 30]

int32_t maxDet#

max detections (ROIs) per frame: the row count of the per-detection tensors. Submit’s detectionCount tensor says how many of them are processed.

int32_t relDopplerOffset[8]#

Host-precomputed DDM Doppler shift per TX in bins, typically round((ddm[0]-ddm[tx])*numChirps); any value, applied modulo numChirps. Indexed [0, numTx).

float ddmaWeightRe[8]#

DDM decode weight, real part: typically cos(2*pi*ddm[tx]), times any per-TX calibration. Indexed [0, numTx); must be finite.

float ddmaWeightIm[8]#

DDM decode weight, imag part: typically sin(2*pi*ddm[tx]), times any per-TX calibration. Indexed [0, numTx); must be finite.

bool coarseDoAUseFft#

true: run coarse DoA as a batched FFT-64 (16 detections per run) instead of the direct DFT. Requires numElVirt == 1, beamAz == 64, numAzVirt == 64, the generated twiddles, no coarse-DoA grid and the full coarse-DoA field of view. false (default): direct DFT.

int32_t coarseElVirt#

Height of the coarse-DoA position grid, onto which a sparse array is zero-filled to make it separable: snapshots become [numAzVirt*coarseElVirt] with position (x,y) at y*numAzVirt + x, and numElVirt must be 1; 0 (default) = use numElVirt, i.e. no separate grid.

int32_t coarseAzBinLo#

First azimuth bin searched.

int32_t coarseAzBinHi#

Last azimuth bin searched; beamAz-1 is the full range.

int32_t coarseElBinLo#

First elevation bin searched.

int32_t coarseElBinHi#

Last elevation bin searched; beamEl-1 is the full range.

bool useCustomTwiddles#

false (default): Create() generates the DFT tables waz[ka][a] = exp(-j*2*pi*ka*a/beamAz) and wel[ke][e] = exp(-j*2*pi*ke*e/beamEl), and the waz/wel tensors are ignored (may be NULL). true: the caller passes the tables (e.g. calibrated steering vectors); not with coarseDoAUseFft. Table row k is beamspace output bin (k + beam/2) % beam.