DSP-Based Calibration and Linearity Testing in Automotive FMCW Radar Receivers with Low-Quality Test Signals

Radar systems have found their way into the automotive industry for decades, enabling vehicles to sense their surroundings and utilize the gathered information for safety features, such as automatic emergency braking. Today, modern vehicles are equipped with a variety of advanced driver assistant systems (ADASs), including blind spot warning, forward collision warning, and adaptive cruise control. Radar has emerged as one of the key sensing technologies because of its all-weather capabilities and robustness against adverse lighting conditions. Specifically, frequency-modulated continuous wave (FMCW) radar has become the most popular choice for automotive applications due to its simple and cost-effective hardware. Since ADASs directly influence vehicle dynamics based on data provided by radar sensors, these sensors must meet stringent safety requirements. Ensuring compliance with these requirements involves extensive testing after production. However, the sensors’ functionality must also be guaranteed throughout their entire operational lifetimes, which necessitates the development of on-chip monitoring concepts. These concepts continuously track specific sensor parameters to ensure reliable performance. Typical monitoring tasks include measuring the gains and phases of the sensors’ transmit and receive paths. However, as sensors become increasingly sensitive and capable of higher resolution, the complexity of the monitoring tasks has also grown. One emerging sensor characteristic to be monitored is the linearity of the radar receivers. In general, linearity testing requires a highly precise test signal source, which is rarely available on-chip for monitoring purposes.
This thesis explores novel concepts for on-chip linearity monitoring in automotive FMCW radar receivers, with a focus on utilizing a low-quality on-chip test signal generator (TSG). Two distinct concepts are developed to address this challenge.
The first concept, referred to as the artificial slow-time (AST) linearity test, addresses the on-chip testing of so-called intercept points. To achieve this, the AST linearity test exploits the high repeatability of an on-chip TSG, which ensures that the same low-quality test signal can be generated multiple times. By applying simple analog modifications to the test signal before coupling it into the receive path, the AST linearity test enables the determination of the receiver’s intercept points through digital signal processing, despite the low quality of the test signal. This approach not only overcomes the limitations of low-quality test signal sources but also provides a cost-effective solution for on-chip linearity monitoring. In this thesis, the theoretical framework of the AST linearity test is validated through both simulations and experimental measurements.
The second concept, referred to as homogeneity enforced calibration (HEC), focuses on the system identification and calibration of one specific component in the radar receiver, namely the analog-to-digital converter (ADC). Similar to the AST linearity test, HEC is tailored to be used with a low-quality on-chip TSG. This is achieved by exploiting the fact that a nonideal ADC violates the homogeneity condition, which is a basic property of linear systems. By inserting the same low-quality test signal into the ADC twice, where, at the second time it is analogously scaled prior to the analog-to-digital conversion, an error signal can be defined. This work shows that the error signal can be employed in cost functions for minimization, enabling the identification of the ADC’s calibration parameters. By relying solely on the ADC’s output signals, this method eliminates the dependence on external high-precision test equipment. This makes the approach highly suitable for integration into automotive radar systems, where compactness, cost-efficiency, and reliability are critical. Two novel filters based on HEC are derived: the HEC Wiener filter and the HEC stochastic gradient descent (SGD) approach. Furthermore, a bilinear extension of HEC is introduced, which eliminates its key implementation challenge, i.e., a precisely known analog scaling factor. For the bilinear extension, two additional filters are developed: the bilinear homogeneity enforced calibration (BL-HEC) Wiener filter and the BL-HEC SGD approach. The resulting BL-HEC not only relaxes the requirements on the TSG’s nonlinearity but also eliminates the need for a precisely known scaling factor, significantly enhancing the practical applicability of HEC. BL-HEC is verified through simulations and measurements conducted on state-of-the-art automotive radar sensors. Ultimately, HEC is explored as a general solution for nonlinear system equalization, highlighting its versatility and demonstrating its applicability to a wide range of systems beyond ADC calibration. In addition, this thesis introduces two filters for broader applications: the bilinear homogeneity enforced least squares filter and the bilinear homogeneity enforced recursive least squares filter.