The following topic page provides a more detailed explanation for some of the questions: https://www.mathworks.com/help/radar/ug/measurement-accuracy-bias-and-resolution.html
The fusionRadarSensor, like other measurement-level radar model (radarDataGenerator, drivingRadarDataGenerator), uses the object radar cross section (RCS), the range to the object, and the radar loop gain to compute an estimated signal-to-noise (SNR). Generally speaking, a higher SNR would result in better (smaller) noise covariance.
In addition, the radar resolution impacts the radar measurement noise in each of the radar measurement axes: azimuth, elevation, range, and range-rate (Doppler).
Therefore, you can indirectly control the amount of reported noise by specifying the following properties:
Resolutions, or how well the radar resolves between objects in the scene: AzimuthResolution, ElevationResolution, RangeResolution, and RangeRateResolution.
Bias fractions (see the link above): AzimuthBiasFraction, ElevationBiasFraction, RangeBiasFraction, RangeRateBiasFraction.
Radar loop gain: ReferenceRange and ReferenceRCS.
Alternatively, you could simply set the MeasurementNoise value for each detection generated by the radar model. Simply loop through the detections and set the MeasurementNoise property on each detection to the value of your choice.
