Despite tremendous progress in the field of text-to-video (T2V) synthesis, open-sourced T2V diffusion models struggle to generate longer videos with dynamically varying and evolving content. They tend to synthesize quasi-static videos, ignoring the necessary visual change-over-time implied in the text prompt. Meanwhile, scaling these models to enable longer, more dynamic video synthesis often remains computationally intractable. To tackle this challenge, we introduce the concept of Generative Temporal Nursing (GTN), where we adjust the generative process on the fly during inference to improve control over the temporal dynamics and enable generation of longer videos. We propose a method for GTN, dubbed VSTAR, which consists of two key ingredients: Video Synopsis Prompting (VSP) and Temporal Attention Regularization (TAR), the latter being our core contribution. Based on a systematic analysis, we discover that the temporal units in pretrained T2V models are crucial to control the video dynamics. Upon this finding, we propose a novel regularization technique to refine the temporal attention, enabling training-free longer video synthesis in a single inference pass. For prompts involving visual progression, we leverage LLMs to generate video synopsis - description of key visual states - based on the original prompt to provide better guidance along the temporal axis. We experimentally showcase the superiority of our method in synthesizing longer, visually appealing videos over open-sourced T2V models.

Despite the recent advances in large-scale diffusion models, little progress has been made on the layout-to-image (L2I) synthesis task. Current L2I models either suffer from poor editability via text or weak alignment between the generated image and the input layout. This limits their usability in practice. To mitigate this, we propose to integrate adversarial supervision into the conventional training pipeline of L2I diffusion models (ALDM). Specifically, we employ a segmentation-based discriminator which provides explicit feedback to the diffusion generator on the pixel-level alignment between the denoised image and the input layout. To encourage consistent adherence to the input layout over the sampling steps, we further introduce the multistep unrolling strategy. Instead of looking at a single timestep, we unroll a few steps recursively to imitate the inference process, and ask the discriminator to assess the alignment of denoised images with the layout over a certain time window. Our experiments show that ALDM enables layout faithfulness of the generated images, while allowing broad editability via text prompts. Moreover, we showcase its usefulness for practical applications: by synthesizing target distribution samples via text control, we improve domain generalization of semantic segmentation models by a large margin (~12 mIoU points).

Abstract The generalization with respect to domain shifts, as they frequently appear in applications such as autonomous driving, is one of the remaining big challenges for deep learning models. Therefore, we propose an exemplar-based style synthesis pipeline to improve domain generalization in semantic segmentation. Our method is based on a novel masked noise encoder for StyleGAN2 inversion. The model learns to faithfully reconstruct the image, preserving its semantic layout through noise prediction. Random masking of the estimated noise enables the style mixing capability of our model, i.e. it allows to alter the global appearance without affecting the semantic layout of an image. Using the proposed masked noise encoder to randomize style and content combinations in the training set, i.e., intra-source style augmentation ( $$\textrm{ISSA}$$ ISSA ) effectively increases the diversity of training data and reduces spurious correlation. As a result, we achieve up to $$12.4\%$$ 12.4 % mIoU improvements on driving-scene semantic segmentation under different types of data shifts, i.e., changing geographic locations, adverse weather conditions, and day to night. $$\textrm{ISSA}$$ ISSA is model-agnostic and straightforwardly applicable with CNNs and Transformers. It is also complementary to other domain generalization techniques, e.g., it improves the recent state-of-the-art solution RobustNet by $$3\%$$ 3 % mIoU in Cityscapes to Dark Zürich. In addition, we demonstrate the strong plug-n-play ability of the proposed style synthesis pipeline, which is readily usable for extra-source exemplars e.g., web-crawled images, without any retraining or fine-tuning. Moreover, we study a new use case to indicate neural network’s generalization capability by building a stylized proxy validation set. This application has significant practical sense for selecting models to be deployed in the open-world environment. Our code is available at https://github.com/boschresearch/ISSA .

The generalization with respect to domain shifts, as they frequently appear in applications such as autonomous driving, is one of the remaining big challenges for deep learning models. Therefore, we propose an intra-source style augmentation (ISSA) method to improve domain generalization in semantic segmentation. Our method is based on a novel masked noise encoder for StyleGAN2 inversion. The model learns to faithfully reconstruct the image preserving its semantic layout through noise prediction. Random masking of the estimated noise enables the style mixing capability of our model, i.e. it allows to alter the global appearance without affecting the semantic layout of an image. Using the proposed masked noise encoder to randomize style and content combinations in the training set, ISSA effectively increases the diversity of training data and reduces spurious correlation. As a result, we achieve up to 12.4% mIoU improvements on driving-scene semantic segmentation under different types of data shifts, i.e., changing geographic locations, adverse weather conditions, and day to night. ISSA is model-agnostic and straightforwardly applicable with CNNs and Transformers. It is also complementary to other domain generalization techniques, e.g., it improves the recent state-of-the-art solution RobustNet by 3% mIoU in Cityscapes to Dark Zürich.

Joint synthesis of images and segmentation masks with generative adversarial networks (GANs) is promising to reduce the effort needed for collecting image data with pixel-wise annotations. However, to learn high-fidelity image-mask synthesis, existing GAN approaches first need a pre-training phase requiring large amounts of image data, which limits their utilization in restricted image domains. In this work, we take a step to reduce this limitation, introducing the task of one-shot image-mask synthesis. We aim to generate diverse images and their segmentation masks given only a single labelled example, and assuming, contrary to previous models, no access to any pre-training data. To this end, inspired by the recent architectural developments of single-image GANs, we introduce our OSMIS model which enables the synthesis of segmentation masks that are precisely aligned to the generated images in the one-shot regime. Besides achieving the high fidelity of generated masks, OSMIS outperforms state-of-the-art single-image GAN models in image synthesis quality and diversity. In addition, despite not using any additional data, OSMIS demonstrates an impressive ability to serve as a source of useful data augmentation for one-shot segmentation applications, providing performance gains that are complementary to standard data augmentation techniques. Code is available at https://github.com/boschresearch/one-shot-synthesis.

Abstract Despite their recent successes, generative adversarial networks (GANs) for semantic image synthesis still suffer from poor image quality when trained with only adversarial supervision. Previously, additionally employing the VGG-based perceptual loss has helped to overcome this issue, significantly improving the synthesis quality, but at the same time limited the progress of GAN models for semantic image synthesis. In this work, we propose a novel, simplified GAN model, which needs only adversarial supervision to achieve high quality results. We re-design the discriminator as a semantic segmentation network, directly using the given semantic label maps as the ground truth for training. By providing stronger supervision to the discriminator as well as to the generator through spatially- and semantically-aware discriminator feedback, we are able to synthesize images of higher fidelity and with a better alignment to their input label maps, making the use of the perceptual loss superfluous. Furthermore, we enable high-quality multi-modal image synthesis through global and local sampling of a 3D noise tensor injected into the generator, which allows complete or partial image editing. We show that images synthesized by our model are more diverse and follow the color and texture distributions of real images more closely. We achieve a strong improvement in image synthesis quality over prior state-of-the-art models across the commonly used ADE20K, Cityscapes, and COCO-Stuff datasets using only adversarial supervision. In addition, we investigate semantic image synthesis under severe class imbalance and sparse annotations, which are common aspects in practical applications but were overlooked in prior works. To this end, we evaluate our model on LVIS, a dataset originally introduced for long-tailed object recognition. We thereby demonstrate high performance of our model in the sparse and unbalanced data regimes, achieved by means of the proposed 3D noise and the ability of our discriminator to balance class contributions directly in the loss function. Our code and pretrained models are available at https://github.com/boschresearch/OASIS .

Variational message passing (VMP), belief propagation (BP) and expectation propagation (EP) have found their wide applications in complex statistical signal processing problems. In addition to viewing them as a class of algorithms operating on graphical models, this article unifies them under an optimization framework, namely, Bethe free energy minimization with differently and appropriately imposed constraints. This new perspective in terms of constraint manipulation can offer additional insights on the connection between different message passing algorithms and is valid for a generic statistical model. It also founds a theoretical framework to systematically derive message passing variants. Taking the sparse signal recovery (SSR) problem as an example, a low-complexity EP variant can be obtained by simple constraint reformulation, delivering better estimation performance with lower complexity than the standard EP algorithm. Furthermore, we can resort to the framework for the systematic derivation of hybrid message passing for complex inference tasks. Notably, a hybrid message passing algorithm is exemplarily derived for joint SSR and statistical model learning with near-optimal inference performance and scalable complexity.

Uncertainty estimates help to identify ambiguous, novel, or anomalous inputs, but the reliable quantification of uncertainty has proven to be challenging for modern deep networks. To improve uncertainty estimation, we propose On-Manifold Adversarial Data Augmentation or OMADA, which specifically attempts to generate challenging examples by following an on-manifold adversarial attack path in the latent space of an autoencoder that closely approximates the decision boundaries between classes. On a variety of datasets and for multiple network architectures, OMADA consistently yields more accurate and better calibrated classifiers than baseline models, and outperforms competing approaches such as Mixup, as well as achieving similar performance to (at times better than) postprocessing calibration methods such as temperature scaling. Variants of OMADA can employ different sampling schemes for ambiguous on-manifold examples based on the entropy of their estimated soft labels, which exhibit specific strengths for generalization, calibration of predicted uncertainty, or detection of out-of-distribution inputs.

Training of Generative Adversarial Networks (GANs) is notoriously fragile, requiring to maintain a careful balance between the generator and the discriminator in order to perform well. To mitigate this issue we introduce a new regularization technique - progressive augmentation of GANs (PA-GAN). The key idea is to gradually increase the task difficulty of the discriminator by progressively augmenting its input or feature space, thus enabling continuous learning of the generator. We show that the proposed progressive augmentation preserves the original GAN objective, does not compromise the discriminator's optimality and encourages a healthy competition between the generator and discriminator, leading to the better-performing generator. We experimentally demonstrate the effectiveness of PA-GAN across different architectures and on multiple benchmarks for the image synthesis task, on average achieving 3 point improvement of the FID score.

Driven by 5G requirements, research on alternatives to the popular cyclic-prefix orthogonal frequency division multiplexing (CP-OFDM) waveform recently arose. In particular, non-orthogonal circularly filtered waveforms such as generalized frequency division multiplexing (GFDM) were proposed due to flexibility and robustness. Applying multiple-input multiple-output (MIMO) techniques for future wireless networks are unquestionable and thereby compulsory for any alternative waveform. Despite advancements in accurate MIMO detection algorithms for GFDM, compared with CP-OFDM their complexity still exhibited a higher order of magnitude, impeding an energy-efficient implementation. In this paper, we propose a low-complexity formulation for iterative minimum mean squared error with parallel interference cancellation (MMSE-PIC) detection for non-orthogonal waveforms with localized inter-carrier interference, where we focus on the application to MIMO-GFDM. The proposal achieves complexity similar to CP-OFDM and we evaluate its performance under realistic channel conditions with imperfect channel state information, where we obtain up to 2-dB gain of GFDM compared with OFDM. We confirm our findings by analyzing the measured extrinsic information transfer charts and show that the proposal achieves the performance of optimal maximum likelihood detection. The results point out the MMSE-PIC algorithm as a viable technique for iterative MIMO receiver implementations for non-orthogonal waveforms.

Owing to the advantage of low-complexity, message passing algorithms have been extensively studied for massive multiple-input multiple-output (MIMO) detection. Most of message passing algorithms assume that the channel state information (CSI) is completely known by receiver, which is unrealistic in wireless communication. In this paper, we investigate a robust approximate message passing (RAMP) detection algorithm with imperfect CSI based on minimizing Bethe free energy. For given pilot structure and channel estimation methods, the results of channel estimation should be denoted by a probability density function of CSI rather than its estimation values. Based on such observations, the MIMO detection issue in the presence of channel estimation error is formulated as a Bethe free energy minimization subject to appropriately imposed constraints and the given statistical model of CSI. The Lagrange multiplier theory is employed to identify the stationary points of the constrained Bethe free energy, which give us back the fixed-point equations. This results in an iterative algorithm for detection in massive MIMO systems. Numerical experiments corroborate its superiority in terms of SER performance when the CSI is imperfect.

In this paper, we investigate the uplink multi-user generalized approximate message passing (GAMP) detection for massive MIMO system. As practical channels are spatially correlated, the conventional GAMP performs poorly and its fixed points fall into locally-optimal solutions. In order to analyse the fixed points of GAMP, we regard the detection problem of massive MIMO systems as the Gibbs free energy minimization and derive GAMP by Bethe method to upper-bound Gibbs free energy. To improve the convergence performance of GAMP detection, we propose linear preprocessing with row-orthogonalization for GAMP (RO-GAMP) at the receiver before GAMP detection is executed. Firstly, we derive the structure of linear preprocessing consisted of four design principles: orthogonality of rows of sensing matrices, irrelevance of noise, low-dimension of observation vector and equivalence of Bethe free energy minimization. Secondly, some conditions are presented on preprocessing matrix to satisfy these design principles. Then, we propose two optimal preprocessing matrices for RO-GAMP. When these two matrices are used for massive MIMO OFDM with slow-varying channels, a low-complexity preprocessing method is presented finally. Our numerical results demonstrate the advantage of RO- GAMP over GAMP, in terms of symbol error rate (SER) and convergence rate, for practical massive MIMO channels which exhibits spatial correlation.

Generalized Frequency Division Multiplexing (GFDM) is a flexible non-orthogonal waveform. Due to its flexibility it can be served as a framework to emulate diverse multi-carrier waveforms including orthogonal frequency division multiplexing (OFDM) and single-carrier frequency domain equalization (SC- FDE). Nevertheless, inter-symbol- and inter-carrier- interference may arise in GFDM if the filter roll-off factor is larger than zero. In multiple-input multiple-output (MIMO) scenarios, also inter-antenna- interference further challenges the receiver design. In this paper, we focus on pilot-aided channel estimation for GFDM. In contrast to our prior works, we propose a technique to insert the pilot symbols in a manner such that they are orthogonal to the data symbols in the frequency domain. Based on this design, frequency-domain channel estimation algorithms initially developed for OFDM become straightforwardly applicable. We also examine the impact of such pilot design on the signal properties, including power spectral density (PSD) and peak-to- average-power ratio (PAPR). At the end of the paper, the performance of a MIMO-GFDM system is investigated and compared with the conventional MIMO-OFDM systems.

In cloud-based radio access networks, remote radio units and central baseband units are connected by fronthaul links, which are commonly assumed to be error-free. However, especially for wireless millimeter wave fronthaul links, this might be challenging to achieve, as they face a more unreliable environment than the conventionally used fiber links. In this paper, we hence aim to mitigate the impact of imperfect fronthaul links. For this, we propose the concept of joint radio access and fronthaul reception, which considers to recover the transmitted messages correctly at the centralized baseband unit, rather than to ensure a nearly perfect fronthaul transmission in between. Based on the Bayesian minimum mean square error criterion, we develop a joint access and fronthaul estimation scheme that can be utilized for various signals transported over the fronthaul, including in-phase/quadrature phase (I/Q) samples, soft-bits, synchronization, and reference signals. In addition, we develop an approximated variant of the scheme to reduced complexity, and an iterative extension to further improve the performance. We demonstrate that our scheme can operate under less reliable fronthaul than conventional approaches by numerical simulation for different signals, and show that our method can be implemented in a parallel architecture to achieve a reasonable computational complexity.

As real-time applications typically require short length code words, this paper analyzes the minimum achievable code word error rate of the given finite-length code words, namely, maximum likelihood (ML) decoding error probability. For coding schemes ML decoding is too complex, the key contribution of this paper is to analytically assess the performance gap between ML decoding and a practical decoding scheme. We analyze the combination of turbo codes and bit-interleaved coded-modulation that is a spectrally efficient coding scheme adopted in 3GPP long term evolution (LTE). In single-input single-output (SISO) systems, it was shown in the literature that turbo decoding delivers near-ML decoding performance. In this paper, we extend the analysis to a multi-input multi-output (MIMO) system. In contrast to the SISO case, the turbo principle-based iterative decoding scheme is subject to appreciable performance loss compared with the ML decoding, even in good MIMO channel conditions. For this reason, we further analyze potential reasons and examine possible improvements. By means of simulation, it is shown that convergence to a non-ML code word rather than non-convergence is the key reason for the observed performance loss in good channel conditions.

This paper studies the link level performance of orthogonal frequency division multiplexing (OFDM) and four other advanced waveforms, namely, filtered OFDM (F-OFDM), universal-filtered OFDM (UF-OFDM), filter bank multicarrier (FBMC) and generalized frequency division multiplexing (GFDM). Compared to OFDM, the two filtered variants achieve lower out-of-band (OOB) emissions and can mostly preserve the conventional OFDM-based transceiver design. For the latter two non-orthogonal waveforms, this paper proposes a low complexity implementation of minimum mean square error equalization to jointly tackle the channel and waveform-induced interference. On this basis, the benefits of FBMC and GFDM can be exploited with complexity comparable to the former (quasi-) orthogonal waveforms. The observed benefits include lower peak-to-average power ratio (PAPR) and smaller frame error rate (FER) under challenging doubly dispersive multiple-input multiple-output (MIMO) fading channels. Additionally, linear filtering of FBMC offers an ultra-low OOB emission, while a good compromise in the usage of time and frequency resources can be achieved by circular filtering of GFDM. In the comparison of offset quadrature amplitude modulation (OQAM) versus QAM for non-orthogonal waveforms, OQAM can offer lower PAPR, while smaller FERs can be achieved by QAM in rich multipath fading channels.

New waveform candidates are being investigated for the fifth generation wireless systems. Among the promising candidates, generalized frequency division multiplexing (GFDM) offers the flexibility to address a wide range of requirements (e.g. low latency, coarse synchronization, etc.). Due to the non- orthogonality of GFDM, the transmit signal subjects to inter-symbol and inter-carrier interference. In this paper, the problem of GFDM channel estimation with the aid of reference signals (pilots) is investigated. In GFDM, the receive signal is a combination of pilots, data and the noise part. Hence, when utilizing the conventional estimation techniques, degradation of channel estimation performance due to interference from data symbols further challenges the receiver design for GFDM. We show that if we employ multiple pilots per subcarrier within a single GFDM block, different pilot patterns have significant impact on the resulting interference term and thus, the quality of the channel estimation in GFDM. Such results are then compared with the performance of channel estimation in orthogonal frequency division multiplexing (OFDM) which takes advantage of clear pilot observation.

Upcoming fifth generation (5G) cellular networks will demand more from the physical layer (PHY) than current- generation Orthogonal Frequency Division Multiplexing (OFDM) can deliver. The 5G waveform candidate Universal Filtered OFDM (UF-OFDM) is designed to provide the flexibility required for future applications. However, the introduction of subband filters in UFMC can increase implementation complexity and low-complexity solutions need to be found. State-of-the-art technologies provide an algorithm that performs shorter-length FFTs that can reduce complexity to two to ten times that of OFDM (depending on the allocation sizes), at the cost of only approximating the exact UFMC signal. In this paper we propose a new approximation of the UFMC signal which bases on the similarity of adjacent subcarriers that can be implemented with reduced number of operations. Analysis show that the system can be implemented with only 20% more operations than standard OFDM when accepting some increase in the subband bandwidth. A more accurate solution can be implemented at roughly 3.6 times OFDM complexity. The results can reduce implementation costs for future mobile devices.

The coded modulation (CM) capacity represents the maximum achievable data rate for a CM scheme assuming optimal decoding at the receiver. It is an important analytical tool, providing theoretic limits for near-optimum transceiver design. Next generation wireless communications systems with the use of new technologies, such as massive antennas and nonorthogonal waveforms, tend to be large scale. However, the conventional ways developed for evaluating the CM capacity of small-scale systems expose exponential complexity with respect to the system's input dimension. Therefore, they become infeasible when the input dimension increases by one or two orders of magnitude in large-scale systems of interest. This letter resorts to a lower and upper bound of the CM capacity, allowing for a computationally efficient evaluation with polynomial complexity. In particular, two message passing algorithms, namely expectation propagation (EP) and variational message passing (VMP), are applied to evaluate the bounds. Two applications are examined in the end. The presented results bring valuable information about the system design, allowing one to evaluate the impact of suboptimal implementation in the overall system performance.

Generalized frequency division multiplexing (GFDM) as a nonorthogonal waveform aims at diverse applications in future mobile networks. To evaluate its performance, its capacity limits are of particular importance. Therefore, this paper analyzes its constellation-constrained capacities for cases where the channel state information (CSI) is unknown at the transmitter and perfectly known at the receiver. In frequency selective channels, GFDM may provide advantage over the conventional orthogonal frequency division multiplexing (OFDM) scheme. In order to achieve near-capacity performance, the interaction of data symbols in time and frequency combined with multiple antennas (MIMO) challenges the design of GFDM receivers. This paper, therefore, applies expectation propagation (EP) for systematic receiver design. It is shown that the resulting iterative MIMO-GFDM receiver with affordable complexity can approach optimum decoding performance and outperform MIMO-OFDM in a rich multipath environment. Simulations are also used to illustrate the impact of channel delay spread on the constellation-constrained capacities and on the performance of the novel receiver algorithm.

Within the framework of Monte Carlo simulation, this paper derives a Markov chain Monte Carlo (MCMC) algorithm for efficient detection in multiple-input multiple-output (MIMO) systems using the non-orthogonal multi-carrier waveform termed generalized frequency division multiplexing (GFDM). The proposed MCMC algorithm performs the detection task in frequency domain. Its adopted proposal distribution and Gibbs sampler are tailored under the consideration of complexity and latency for tackling the three-dimensional interference involved in the received signal, i.e., inter-carrier, inter-symbol and inter-antenna interference. By means of simulation, its decoding performance is compared with that achieved by employing the sphere decoding algorithm in a conventional orthogonal frequency division multiplexing (OFDM) based MIMO system. For multi-path fading channels with strong frequency selectivity, the MCMC algorithm proposed for the MIMO-GFDM system can deliver superior performance.

Innovative 5G applications will challenge future cellular systems with new requirements. The OFDM based 4G standard will not be able to address all of them. Generalized frequency division multiplexing is a flexible multicarrier waveform with additional degrees of freedom. This paper presents a strategy towards a flexible FPGA implementation of GFDM, which is reconfigurable at run-time.

For upcoming 5G networks, new challenges are posed on the physical layer, which go beyond increased data rate. Generalized Frequency Division Multiplexing (GFDM) is proposed as a candidate waveform to combat these challenges. However, inherent self-interference between subcarriers of GFDM hinders the application of standard spatial multiplexing (SM) detection algorithms. We present an algorithm that combines maximum likelihood and successive interference cancellation detection techniques that allows to exploit the inherent frequency diversity of GFDM coming from self-interference. Computer simulations reveal that the proposal outperforms OFDM in terms of symbol error rate in fading multipath channels. These findings prove self- interference to be beneficial and that SM can be successfully applied to GFDM.

A low-complexity algorithm for calculation of LMMSE filter coefficients for Generalized Frequency Division Multiplexing (GFDM) in a fading multipath environment is derived. The simplification is based on the block circularity of the involved matrices. The proposal reduces complexity from cubic to squared order. The proposed approach can be generalized to other waveforms with circular pulse shaping.

This paper presents a solution for achieving transmit diversity with generalized frequency division multiplexing (GFDM). Compared to previous works, the proposed solution significantly improves symbol error rate (SER) performance and latency, where both aspects are crucial for future 5G cellular networks. It is shown that widely linear estimation at the receiver side can jointly equalize and demodulate the space-time encoded GFDM signal. Moreover, maximum ratio combining can further increase the SER performance with multiple receive antennas. SER performance is evaluated in Rayleigh fading multipath channels.

When an iterative receiver has more than one iteration loop, scheduling the iterative decoding process is an important issue. To find the optimal scheduling, extrinsic information transfer (EXIT) function was the main tool in the literature. However, it is only very efficient when the codeword is significantly long. In this paper, under the consideration of a packet-based transmission scenario, we propose modeling the scheduling search problem as the foraging problem of ant colony. And then, an ant colony optimization (ACO) algorithm, labeled as max-min ant system (MMAS), is adopted and tailored for our problem. Simulation results show MMAS outperforms the EXIT function-based method when practical length codewords are used.

This paper presents two enhancements for improving the convergence behavior of bit-interleaved coded modulation iterative decoding (BICM-ID) in multiple-input multiple-output (MIMO) systems. As typical tree search based MIMO demapping algorithms for BICM-ID are based on the suboptimal Max-Log-MAP (maximum a-posteriori) criterion, the first enhancement aims at correcting the outputs of such MIMO demappers towards the optimal Log-MAP criterion. For the second enhancement, the deterministic annealing technique is applied to the iterative decoding process. Through simulation results, the efficiency of both enhancements has been demonstrated via the decoding performance and complexity.

For soft-input soft-output sphere decoding (SD), the combination of the Schnorr-Euchner (SE) enumeration and the radius reduction shrink the search space quickly but with considerable computational complexity. In this paper, a low complexity approach is proposed to reduce the tree search space before SD starts. Particularly, such reduction is achieved by 1) tightening the initial radius; 2) restricting the search space to a subset of the symbol lattice. With further enhancement on the tree pruning constraint, simulation results demonstrate significant complexity savings with negligible error rate performance loss.

In this paper, the iterative scheduling issue for multiple-input multiple-output (MIMO) systems with turbo codes is addressed based on the informed asynchronous scheduling, initially used for scheduling low-density parity-check (LDPC) decoding. For the iterative MIMO demapping and turbo decoding, the convergence behavior by using the informed asynchronous scheduling is first demonstrated to be similar to the standard sequential scheduling. As the extrinsic log-likelihood ratio (LLR) calculation at both the MIMO demapper and turbo decoder costs high complexity, an informed message update (IMU) rule is further proposed to select only a subset of extrinsic LLRs for updates within each iteration. By avoiding the re-calculations for the extrinsic LLRs with small changes between successive iterations, significant complexity savings are achieved with negligible error rate performance loss.

We extend the methodology in to flat fading channels. The systematically derived receiver structure corresponds to an iterative solution of the joint maximum likelihood (ML) optimization problem. The heuristically derived BICM with iterative decoding is shown to be the outcome of this joint ML optimization problem.

This paper introduces a new framework of the multiple-input multiple-output (MIMO) detection in iterative receivers. Unlike the conventional methods processing with symbol lattice, we consider the delta symbol lattice, i.e., the difference between two arbitrary points in the symbol lattice. The inherent flexible, symmetric, and sparse properties of the delta lattice enhance the detection in both complexity and performance aspects. Consequently, we propose a delta-list MIMO (DL-MIMO) detection which separately exploits the channel information and the a priori information so that a soft-input soft-output sphere decoder is dispensable. Simulation results demonstrate this hardware-friendly DL-MIMO detection delivers nearly-optimal performance at affordable cost in a practical scenario.

Using the Schnorr-Euchner (SE) order for soft-input sphere decoders is inefficient for implementation, because it requires exhaustive calculation and sorting of partial metrics of all constellation points. Instead, low-complexity methods can be applied by separating the partial metric into channel information and a priori information and solely enumerating based on one of them. With such an orthogonalization, this paper presents an algorithm that effectively combines these two enumerations to deliver an order close to the SE one. Mathematical analyses and simulation results demonstrate that this is the first algorithm allowing for a low-complexity implementation with optimal error rate performance for any number of iterations.