This paper proposes unprecedented Expectation Propagation (EP)-based receivers for Circular Faster-Than-Nyquist (CFTN) signaling. This concept yields a Minimum Mean-Square-Error equalizer for InterSymbol Interference (ISI) processing combined with a block which we call Constellation Matcher in charge of the symbol estimates realignment with the constellation. From this framework, we explore different scheduling strategies leading to iterative EP-based receivers with our without decision feedback. Then, by extending the family of the EP process to a subset of non-circular Gaussian distributions results in a Widely Linear (WL) equalization. This new WL-EP receiver best fits the CFTN model, allowing enhanced performance at the cost of a slightly increased complexity. Also, we propose to restrict the Gaussian family to circular Gaussian distributions with identity covariance matrices up to a scaling factor. Combined with CFTN, this particular family allows a low-complexity Frequency-Domain processing of the equalization without requiring any cyclic prefix. The proposed EP-based receivers for CFTN are then evaluated for different spectral efficiencies and computational complexities. Our simulations show that they completely handle the ISI up to 5 bits/s/Hz and double the spectral efficiency compared to Nyquist signaling with almost no performance loss.

This paper proposes a new Widely Linear Minimum Mean Square Error with Decision Feedback Equalizer (WL-MMSE-DFE) based on the Expectation Propagation (EP) formalism for Faster-Than-Nyquist (FTN) communications. The system model is first derived using a matrix formalism avoiding any aliasing on the Nyquist filter. Then, after recalling the principles of EP, we consider a WL exponential family for estimating the transmitted symbols. This leads us to an iterative receiver composed of a new WL-MMSE-DFE equalizer dedicated to FTN which exchanges messages with a WL constellation matcher. Moreover, we develop a WL-EP-MMSE-DFE algorithm that outperforms existing solutions at the price of a minor complexity increase. Finally, the proposed receiver allows FTN to achieve its Matched Filter Bound while doubling the spectral efficiency of Nyquist signaling up to 3 bits/s/Hz.

This paper proposes a new, simple and highly adaptive filter design for distortion-based peak to average power ratio (PAPR) reduction algorithms. The proposed method is based on a prototype filter design and is capable to efficiently address multiband signals. Recent practical implementations recommend the use of the equiripple (ER) filter. Numerical results show that for a given complexity and for a given PAPR reduction, the proposed method outperforms the performance of the ER filter in terms of power spectral density and error vector magnitude (EVM). Finally, the proposed method is extended to address the potential heterogeneity of the EVM specifications, even within the same band. We show that our approach provides an additional gain in terms of PAPR reduction capability.

Direct-conversion architectures suffer from the mismatch between the in-phase (I) and the quadrature-phase (Q) branches, commonly called I/Q imbalance. Even low I/Q imbalances imply poor performance of orthogonal frequency division multiplexing (OFDM) systems. In this paper, we propose a new algorithm that uses both training and data symbols in a decision-directed fashion to jointly estimate the channel and compensate for high receiver I/Q imbalance. Simulation results show that our method can compensate for high I/Q imbalance values and also estimate a frequency selective channel.