Due to imperfections introduced in the manufacturing process of the PCB and the SoC calibrations can be applied to compensate for bias in the range and receiver gain and phase introduced from the RF path delays. At a high level, this goal of this procedure is to determine the range bias offset common to all Tx-Rx paths and the gain and phase mismatch of each virtual Tx-Rx pair of an object placed at boresight at a fixed, known distance in the far field. The calibration coefficients generated from this procedure can be applied in post processing to compensate for the relative delay between the Rx paths so that each receiver interprets an object placed at boresight as being at a zero-angle and eliminates any bias in the detected range. The delays are unique to each PCB and sensor and thus each board/sensor pair has a unique set of calibration coefficients. Depending on the placement and orientation of the antennas, the applied correction could be correcting either azimuth or elevation angle. For example, if the receiver antennas are placed horizontally, then it is the azimuth angle that is zero’ed out during this calibration.

Procedure: For this example, assume that the radar sensor has four receiver antennas laid out horizontally so that the zero-angle calibration zero’es out azimuth angle error. The set up used is shown in Figure 2-2. A corner reflector is a preferred target to use for this test as it shows up as a single point, whereas a metal plate can be interpreted as multiple points, which can impact the accuracy of the calibration. The target is kept at a distance large enough so that the difference in azimuth angle from the target to each of the four receivers is negligible. The chirp parameters are chosen appropriately so that it generates an IF of around 1 MHz. The RF sweep is chosen to be within the specified operating range of Radar. One of the transmitters is used for this calibration. ADC output from each receiver chain is simultaneously captured.

First, using a turn-table the elevation beam tilt, if any, is compensated so that the sensor beam is made horizontal. Next the radar sensor is kept at zero-degree azimuth angle from the target. The ADC output is captured and is processed to get the FFT and further the relative phase difference between each Rx path. The range bias coefficient is computed based on the peak position and the known target distance of X. The gain and phase mismatch between each virtual Tx-Rx pair is computed and the coefficients are generated such that each of the virtual channels is forced to zero phase for the object at boresight. The coefficients can then be stored in a LUT and used in angle of arrival computations. Sufficient averaging is done to improve the accuracy. Confirmatory readings are taken at a couple of known angles on either side of zero to confirm that the calibration works.

The mmWave SDK provides a method for generating the calibration coefficients over the command line interface via the Out of Box Demo. For more details, refer to the mmWave SDK User’s Guide. Additionally, the procedure and implementation of the calibration routine in the data-path processing chain can be found in the mmWave SDK install folder at mmwave_sdk_<ver>\packages\ti\datapath\dpc\objectdetection\<chain_type>\docs\doxygen\html\index.html. The user can use the OOB directly to perform the calibration or can port the provided source code into their own custom application.