Blind Adaptive Beamformer Based on Orthogonal Projections for GNSS

The quality of the ranging data provided by a global navigation satellite systems (GNSS) receiver largely depends on the synchronization error, that is, on the accuracy of the propagation time-delay estimation of the line-of-sight (LOS) signal. In case the LOS signal is corrupted by several superimposed delayed replicas (reflective, diffractive, or refractive multipath) and/or additional radio frequency interference (RFI), the estimation of the propagation time-delay and thus the position can be severely degraded using state-of-the-art GNSS receivers. Multi-antenna GNSS receivers enable application of array processing for effective multipath and interference mitigation. Especially, beamforming (spatial filtering) approaches have been studied intensely for GNSS in the past years due to a balanced trade-off between performance and complexity. Usually these beamforming approaches require knowledge of the spatial signature (spatial reference) of the desired signal and thus require detailed knowledge of the direction-of-arrival (DOA) of the LOS signal and/or non-LOS signals, the antenna response, the array geometry, and other hardware biases. Even if the antenna array response can be approximately determined, either by empirical measurements (array calibration) or by making certain assumptions (e.g. identical sensor elements in known locations), the true antenna array response can be significantly different due to for example changes in antenna location, temperature, calibration inaccuracy and the surrounding environment. Thus, robust beamforming algorithms were developed in order to cope with errors in the array response model to be applied to derive the spatial filter. In this work we propose a blind adaptive beamforming approach based on orthogonal projections for GNSS, for which knowledge about the array response and spatial reference for the LOS signal are not required. The proposed approach is capable of adaptively mitigating RFI and multipath components based on orthogonal projections. In order to derive the needed projectors adaptively two eigendecompositions of the estimate of the spatial covariance matrix before (pre-correlation) and after (post-correlation) despreading are performed. Based on these eigendecompositions appropriate estimations of subspaces are achieved in order to derive projectors onto the interference free and multipath free subspaces, respectively. At pre-correlation stage the covariance matrix estimation can be evaluated over a short time interval in order to realize good performance in case of a jammer with a high time-frequency dynamic (e.g. Chirp-like jammer). For the implementation within a real-time receiver, dedicated building-blocks are used. Computation of the covariance matrix and the projection into the interference free subspace is performed by hardware-macros at sampling-rate. In contrast, the eigendecomposition is executed on a processor achieving projector update-rates in the kHz-range. Implementation issues addressing quantization losses related to wordlength configurations for both hardware building-blocks are discussed. Once interfering signals are removed from the input signals, wordlengths can be reduced in order to minimize implementation costs for the subsequent despreading or correlation. Wordlength reduction is realized using a digital automatic gain control (AGC). At post-correlation stage all available degrees of freedom are used for multipath mitigation, noise reduction and further cancellation of residual interferences. After despreading, projection into the multipath free subspace becomes an individual process for each channel of the receiver. Considering the computational load on a navigation processor, this is a very challenging task since covariance matrices and eigendecompositions have to be computed individually for each channel. A cost-analysis in terms of processing cycles on an embedded processor for the covariance matrix computation and eigendecomposition is provided. In addition, the relation between the covariance observation time and multipath mitigation performance are analyzed for selected scenarios. Simulation results show that the proposed blind adaptive beamforming approach based on orthogonal projections achieves effective interference and multipath mitigation capabilities compared to state-of-the-art non-blind beamforming algorithms. The overall complexity required by the blind beamformer is discussed and a feasible hardware implementation is derived. The accuracy and numerical stability of estimation of the spatial covariance matrix before and after despreading are shown for both block interval and recursive estimation methods. Providing costs in terms of computational requirements and navigation performance related to a specific implementation a trade-off between estimation robustness, time-frequency characterization and mitigation capability is derived. Based on this analysis a complete multi-antenna GNSS receiver architecture is proposed taking into account hardware complexity and navigation performance. A software bit accurate representation of the receiver hardware platform is used for performance evaluation. As the proposed blind approach does not require precise a priori information about the DOAs of the LOS (spatial reference) or non-LOS signals and about the antenna array response, robustness with respect to errors in the antenna array response model and additional hardware biases can be achieved without further increase of complexity.