Large-Scale and Hierachical Bayesian Inference for Future Mobile Communication Networks

• Applicant: Professor Dr.-Ing. Gerhard P. Fettweis, since 2/2019
• Subject Area: Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering
• Term: from 2018 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 392016367

• Final Report Year: 2022

Final Report Abstract

The project LHBiCOM advocated the use of Bayesian statistical inference methods to solve large dimensional signal processing problems in future communications systems with enormous traffic demands, massive connections, and/or non-Gaussian/non-linear signal distortions. Future communications systems bring great challenges to signal processing engineering from tractability and fidelity perspectives. The classic linear minimum mean square error (LMMSE) estimator is optimal for jointly Gaussian estimation problems. The number of observations needs to be at least as large as the number of unknowns. In contrast, future communications systems involve high dimensional non-Gaussian and/or non-linear estimation problems due to new features like massive connections and coarse quantization. As the involved maximum likelihood or maximum a-posteriori problems are in general either analytically intractable or computationally expensive, signal processing solutions with acceptable complexity must be developed. Targeting at high fidelity solutions with tractable complexity, within this project, we first studied variational Bayesian inference. A framework providing various message passing based solutions is developed by making different approximations of the underlying distributions. By minimizing the Kullback-Leibler (KL) divergence between these alternate distributions and the true likelihood or a-posteriori functions, this framework shows that various message passing algorithms (including belief propagation (BP), expectation propagation (EP), variational message passing (VMP), approximate message passing, and generalized approximate message passing) can be derived from the same objective function, i.e., by minimization of the KL divergence. Their different updating rules result from differently formulated constraints, e.g., marginalization consistency for BP, moment matching for EP, ignorance of local dependencies for VMP. Noticing this, we treat constraints as the degree of freedom in algorithm design, and the framework becomes a handy tool for developing novel message passing algorithms, e.g., for estimating latent variables of different statistical properties, for developing edge-dependent hybrid messages, and for fitting the statistical models in largescale non-linear and/or non-Gaussian systems. Furthermore, the application of different signal processing methods, especially message passing based ones, has been studied for large-scale MIMO systems with sparse unknown variables, non-linear quantization errors, and imperfect channel state information (CSI). In particular, we have conducted in-depth research on the following five large-scale MIMO systems with either non-Gaussian likelihood functions, statistical CSI, sparse angular-domain channels, or sparse user activity patterns: 1) sparse signal recovery and sparse Bayesian learning in massive machine-type communications; 2) novel antenna array topologies with robust line-of-sight spatial multiplexing gain under strong quantization distortions including its channel equalization; 3) angular domain channel estimation for massive MIMO- OFDM communications; 4) joint channel estimation and data detection for grant-free massive machine-type communications with imperfect CSI; 5) beam alignment and channel estimation in fast time-varying communications like vehicle-to-everything (V2X) communications. The derived low-complexity algorithms and novel system designs show significant performance advantages while being computationally tractable.

Publications

• “Fast beam alignment through simultaneous beam steering and power spectrum estimation using a frequency scanning array,” in International ITG Workshop on Smart Antennas, Hamburg, Germany, Feb. 2020
  C. Jans, X. Song, W. Rave, and G. Fettweis
  
• “Frequency-selective analog beam probing for millimeter wave communication systems,” in Proc. IEEE Wireless Commun. and Netw. Conf., June 2020, pp. 1–6
  C. Jans, X. Song, W. Rave, and G. Fettweis
  
• “Network massive MIMO transmission over millimeter-wave and terahertz bands: Mobility enhancement and blockage mitigation,” IEEE J. Sel. Areas Commun., vol. 38, no. 12, pp. 2946–2960, 2020
  L. You, X. Chen, X. Song, F. Jiang, W. Wang, X. Gao, and G. Fettweis
  
• “Pilot reuse for vehicle-to-vehicle underlay massive MIMO transmission,” IEEE Trans. Veh. Technol., vol. 69, no. 5, pp. 5693–5697, 2020
  L. You, M. Xiao, X. Song, Y. Liu, W. Wang, X. Gao, and G. Fettweis
  
• “Deterministic pilot design and channel estimation for downlink massive MIMO-OTFS systems in presence of the fractional Doppler,” IEEE Trans. Wireless Commun., vol. 20, no. 11, pp. 7151– 7165, 2021
  D. Shi, W. Wang, L. You, X. Song, Y. Hong, X. Gao, and G. Fettweis
  
• “On one-bit lineof-sight MIMO communications at flexible communications distances,” IEEE Wireless Commun. Lett., vol. 10, no. 1, pp. 116–120, 2021
  X. Song, L. Landau, W. Wang, L. You, X. Gao, and G. P. Fettweis
  
• “Sparse channel estimation via hierarchical hybrid message passing for massive MIMO-OFDM systems,” IEEE Trans. Wireless Commun., vol. 20, no. 11, pp. 7118–7134, 2021
  X. Liu, W. Wang, X. Song, X. Gao, and G. Fettweis
  
• “Unifying message passing algorithms under the framework of constrained Bethe free energy minimization,” IEEE Trans. Wireless Commun., vol. 20, no. 7, pp. 4144–4158, 2021
  D. Zhang, X. Song, W. Wang, G. Fettweis, and X. Gao
  
• “On robust millimeter wave line-of-sight MIMO communications with few-bit ADCs,” IEEE Trans. Wireless Commun., 2022
  X. Song, S. Ma, P. Neuhaus, W. Wang, X. Gao, and G. Fettweis