Calibration of Large Coherent MIMO Radar Arrays: Channel Imbalances and 3D Antenna Positions

One of the main challenges for next generation automotive radars is the improvement of angular resolution to a sub-degree level. In this context, wide aperture automotive radars of 1m length or more and resolution close to 0.1 degrees in azimuth and 0.5 degrees in elevation could be beneficial. To enable coherent processing of arrays with such large aperture, prior (i.e offline) and online calibration are necessary: channel imbalances (gains and phases) and three dimensional coordinates of transmit and receive elements need to be determined. We propose a calibration strategy based on alternating steps between the two subtasks of i) channel imbalance estimation with 'known' array positions, by applying a singular value decomposition to the resulting tensor calculus problem; and ii) antenna position estimation with 'known' channel imbalances, by numerically maximizing the Bayesian posterior probability; in both cases operating on range/Doppler snapshots of disjoint targets (with potentially unknown locations). Simulation studies based on the parameters of a MIMO 8x6 linear sparse array show promising results as long as the initial position errors do not exceed half a wavelength (2 mm), beyond which we observe strong effects of ambiguity. Experimental results with real measurements show that after calibration in laboratory conditions, our MIMO 8x6 demonstrator with 50cm aperture is able to resolve two targets at the same range with angular separation at least as close as 0.4 degrees.