In this work we consider the front-end receivers for flat fading MIMO transmission, whose essential feature is the calculation of the log-likelihood ratios (LLRs) for the transmitted bits. When the number of transmit antennas and the modulation size grow, the exact calculation of the LLRs is most often unfeasible due to the necessity to enumerate the symbols affecting the output. In this work, to avoid the burden of explicit enumeration, instead of calculating the likelihoods exactly, we explore the possibilities offered by the saddlepoint approximation of the likelihood functions. The resulting fixed-complexity approach significantly outperforms the well known linear-filter based minimum mean square error detection.

In this work, we consider the front-end receivers for flat fading MIMO transmission, whose essential feature is the calculation of the log-likelihood ratios (LLRs) for the transmitted bits. When the number of transmit antennas and the modulation size grow, the exact calculation of the LLRs becomes unfeasible due to the necessity to enumerate the symbols affecting the output. In this work, to avoid the burden of explicit enumeration, instead of calculating the likelihoods exactly, we explore the possibilities offered by the saddlepoint approximation of the likelihood functions. The resulting fixed-complexity approach significantly outperforms the well known linear-filter based minimum mean square error detection. A complexity analysis identifies the main computational bottleneck of the proposed detection algorithm.

Based on the Expectation Propagation framework, we derive a MIMO detector that consists of an interference cancellation (IC) step, followed by linear minimum mean square error (LMMSE) filtering. The distinguishing feature of our algorithm, compared to conventional IC-LMMSE detectors for iterative MIMO receivers, is the way how we compute the symbol statistics that are used in the IC and LMMSE steps. In an earlier publication, it has been shown empirically that full posterior feedback from the decoder to the IC-LMMSE detector can outperform the usual choice of extrinsic feedback, which seems to contradict the Turbo principle. The systematic derivation in the present paper provides an explanation of this phenomenon, and results in a further improved algorithm that outperforms the conventional IC-LMMSE detector with either extrinsic or posterior feedback. While we consider only multi-antenna systems, we note that the same principle can be applied to similar problems in signal processing, such as Turbo equalization for frequency-selective channels, or multi-user detection in CDMA systems.

We derive a Markov chain Monte Carlo algorithm for data detection in multi-antenna systems. The proposed algorithm is based on the Rao-Blackwellization principle, and is parameterized by a value eta that can be interpreted as a system temperature. We demonstrate by simulations that systems with increased temperature (η>;1) significantly outperform the straightforward choice of η=1. Due to its log-domain formulation, our algorithm combines near-optimal performance with low implementation complexity.

Recently, MIMO detectors which are based on Markov Chain Monte Carlo (MCMC) simulation techniques have been proposed as alternatives to, e.g., the well-known sphere detector. In this paper, we present a systematic analysis of the performance of MCMC detectors. We study the impact of several parameters such as the list size and the number of independently running chains. As a performance criterion, our analysis is based on the mutual information over an equivalent modulation channel, rather than on coded bit error rates, because this metric is independent of the outer channel code and provides valuable insights over the whole SNR range of interest. Furthermore, we show that combining the MCMC detector with a hard-output sphere detector removes the error floor at high SNR, which is a well-known problem of the MCMC principle.

The problem of data detection in MIMO systems is usually studied under the assumption of perfect channel knowledge at the receiver. In practice, however, only a (potentially very poor) estimate of the channel is available. It has been shown that detection algorithms which take the unreliability of the channel estimate into account can significantly outperform their mismatched counterparts (i.e., detectors which use the estimate in lieu of the true channel matrix). In this paper, we extend the Markov Chain Monte Carlo (MCMC) MIMO detector to the case of imperfect channel knowledge. Due to the low complexity of the MCMC algorithm, this technique is also applicable in systems where brute force approaches are computationally infeasible.

A well-known problem of OFDM signals is their high peak-to-average power ratio, which can cause the clipping of peaks in the transmitted signal due to the limited linear range of high power amplifiers. Operating the amplifier solely in the linear region, i.e., with a large output back-off, results in an undistorted transmitted signal, but at the expense of a low SNR at the receiver. Driving the amplifier into saturation, on the other hand, maximizes the received SNR, but causes severe nonlinear signal distortions. In this paper we study the resulting tradeoff. We show both theoretically and by simulations that, despite the distortions, it is optimal to operate the amplifier close to saturation. Furthermore, we analytically derive a modification of the Bussgang noise cancellation algorithm which was recently shown in the literature to improve the EXIT characteristic of the algorithm.

We examine the application of beamforming in a single frequency network (SFN) for the provision of multicast services. A single frequency network is characterized by the simultaneous transmission of the same signal from multiple base stations. Since each user receives the signal from several base stations, the beamforming weights for many different base stations need to be jointly optimized for the transmission in order to give optimal results. The determination of the beamforming weights that maximize the minimum SNR is a nonconvex problem. A constraint relaxation however yields a convex program, which can be solved with fast algorithms at only a slight loss of performance. Furthermore, this convex formulation of the problem gives solutions that perform significantly better than omnidirectional transmission.

The high peak-to-average power ratio is a major drawback of OFDM systems. Many PAPR reduction techniques have been proposed in the literature, among them a method that uses a subset of tones that do not carry any data, but are modulated such that the PAPR of the resulting time domain signal is minimized. Another problem of OFDM systems is the high out-of-band power caused by the sidelobes of the modulated tones. The OBP can be reduced by modulating reserved tones at the edges of the occupied spectrum so that the sidelobes of the data carriers are reduced. In this paper, we propose to consider both optimization problems jointly. This way, the amount of PAPR and OBP reduction can be significantly enhanced in comparison to a system that performs two separate optimization steps. Furthermore, the joint reduction algorithm offers more flexibility, because the relative weighting of the two optimization criteria can easily be changed, resulting in a smooth trade-off curve.