Automotive Systems Laboratory, Inc.

Craig S Jacobs

Michael Edward Farmer

Shan Cong

A neural network radar processor ( ) comprises a multilayer perceptron neural network ( ) comprising an input layer ( ), a second layer ( ), and at least a third layer ( ), wherein each layer has a plurality of nodes ( ), and respective subsets of nodes ( ) of the second ( ) and third ( ) layers are interconnected so as to form mutually exclusive subnetworks ( ). In-phase and quadrature phase time series from a sampled down-converted FMCW radar signal ( ) are applied to the input layer, and the neural network ( ) is trained so that the nodes of the output layer ( ) are responsive to targets in corresponding range cells, and different subnetworks ( ) are responsive to respectively different non-overlapping sets of target ranges. The neural network is trained with signals that are germane to an FMCW radar, including a wide range of target scenarios as well as leakage signals, DC bias signals, and background clutter signals.

Neural network radar processor

A leakage calibration and removal system and method estimates the complex in-phase and quadrature phase (I/Q) components of a leakage signal for each beam location in the sampled down-converted radar signal in a radar system (10). In a digital embodiment, the stored leakage calibration signal (264) is subtracted (206) from the sampled radar signal, and the resultant signal is processed (208, 210, 212) to detect targets. A leakage calibration process (250) is activated if a leakage signal test (214) indicates a problem for a sufficient number of consecutive scans (216), wherein for each beam location, M consecutive I/Q waveforms are averaged (252), known targets are removed (254, 256, 258), and the resulting signal is scaled (262) and stored (264) as a new leakage calibration signal if the variance is within acceptable limits (262). In a hybrid embodiment, the stored leakage signal (364) is converted to analog form (366), subtracted (301) from the analog down-converted radar signal (300), and scaled by a variable gain (303) before the complex I/Q components are sampled therefrom (302, 304). A leakage calibration process (350) is activated every Nth scan (316), wherein for each beam location, M chirp waveforms are averaged (352) and checked for variance (354). If the variance is within acceptable limits, the leakage signal is updated by a Kalman Filter (356), stored as the new leakage calibration signal (364), and the Kalman gain matrices are updated (358).

Growing community of inventors

Farmington Hills, MI, United States of America

Craig S Jacobs