In this paper we study channel estimation for Narrow-band Internet of things (NB-IoT) systems, a low-power radio access technology introduced by the Third Generation Partnership Project (3GPP) as a part of Long Term Evolution (LTE) in Release 13 and recently extended to support Non-Terrestrial Networks (NTN) in Release 17. To improve the quality of channel measurements from pilot signals, we propose a cross-slot smoothing algorithm based on a fixed-lag Kalman smoother (KS). The non-causal and recursive nature of the proposed smoother is suitable for NB-IoT systems, thanks to their delay-tolerant design (due the presence of sub-frame repetitions) and to unitary precoding (which ensures consistency of channel statistics across slots). We propose an efficient implementation of fixed-lag KS, which we call “reduced-redundancy” fixed-lag. Our algorithm eliminates computational redundancies by extending the state vector instead of augmenting it. We tailor our algorithm to practical scenarios, including geostationary (GEO) and low Earth orbit (LEO) satellites. Results show that the proposed algorithm outperforms causal Kalman filter and non-causal moving average or infinite impulse response filters, with comparable computational complexity.

Estimating second-order channel statistics (specifically, frequency-domain covariance matrices) from narrowband reference signals is crucial for channel estimation in 5 G cellular systems. The problem is challenging because of the limited number of available pilot subcarriers. In this work, we reparametrize the problem using a delay-domain characterization. Three algorithms are proposed for estimating the power delay profile (PDP) using least squared error and maximum likelihood estimation frameworks. The algorithms demonstrate a tradeoff between the computational complexity and model insights used to estimate the underlying PDP. The performance of the proposed algorithms is numerically analyzed under different channel conditions and pilot density. The proposed algorithms demonstrate significant gains (up to 3 dB in signal-to-noise ratio), and are robust to noisy delay spread information.

In orthogonal frequency-division multiplexing (OFDM)-based systems, such as 5 G, precise channel estimation (CE) is crucial for reliable connectivity and high data rates. However, in 5G, the sub-band-specific precoding structure of pilot signals, known as demodulation reference signals (DMRS), limits the number of pilot samples per sub-band, which can lead to significant CE errors. This paper provides an in-depth analysis of the influence of CE errors on pilot and data subcarriers in OFDM systems. We rigorously demonstrate that CE errors on data subcarriers can be modeled as independent additive noise, validating assumptions made in previous works but not formally proven. Additionally, we analyze how CE errors on pilot subcarriers introduce bias in noise variance estimation and derive closed-form expressions for this bias under both perfect and imperfect MMSE CE conditions. We propose practical bias removal schemes tailored for residual power-based noise variance estimation. Numerical simulations confirm the accuracy of our theoretical findings and show that the proposed noise variance estimation method significantly reduces the performance gap between narrow-band and wide-band CE.

In wireless communication systems where the received signal model is linear with Gaussian-distributed channel and noise, linear minimum mean square error (LMMSE) channel estimation (CE) achieves the best performance in terms of mean square error (MSE). However, LMMSE CE relies on parameters that may be either unavailable at the receiver (e.g., accurate knowledge of the power delay profile (PDP)) or too complex for practical implementation (e.g., the LMMSE filter size). A suboptimal choice of parameters may severely degrade LMMSE CE performance. Motivated by this observation, we investigate machine learning as a tool for refining and improving CE performance. We show that our proposed low-complexity learning-aided LMMSE CE can overcome the impact of suboptimal parameters and approach the ideal LMMSE performance.

In this paper we investigate the problem of precoder selection assuming compressed channel state information (CSI) feedback over sub-bands. Typically (e.g., in NR Rel-16) the problem is approached by a two-step solution: independent precoder optimization for each sub-band, followed by selection of the compression bases. We show that such conventional approach does not perform well in scenarios of medium/high frequency selectivity. Then, we derive an alternative approach, based on wideband capacity maximization, which provides superior performance under frequency-selective channels, while being compatible with the existing NR Rel-16 framework without any additional signaling. We finally propose an adaptive method in which the receiver dynamically selects the wideband or the sub-band optimization strategy, depending on the instantaneous channel condition. The proposed approach achieves consistent gains (up to 3dB) in all the considered test cases.

Deep learning-based methods have made significant impact and demonstrated superior performance for the classical image and video super-resolution (SR) tasks. Yet, deep learning-based approaches to super-resolve the appearance of 3D objects are still sparse. Due to the nature of rendering 3D models, 2D SR methods applied directly to 3D object texture may not be a good approach. In this paper, we propose a rendering loss derived from the rendering of a 3D model and demonstrate its application to the SR task in the context of 3D texturing. Unlike other literature on the 3D appearance SR, no geometry information of the 3D model is required during network inference. Experimental results demonstrate that incorporating the rendering loss during network training outperforms existing state-of-the-art methods for 3D appearance SR. Furthermore, we provide a new 3D dataset consisting of 97 complete 3D models for further research in this field.

This paper investigates the problem of noise vari-ance estimation in orthogonal frequency domain multiplexing (OFDM)-based systems such as 5G New Radio (NR). Accurate estimation of the noise variance is critical for the receiver performance, especially when applied with linear minimum mean square error (LMMSE) channel estimation (CE). A commonly used method estimates the noise variance from the power of the residual signal at the CE output. In this paper, we prove that such conventional estimator is biased, resulting in underestimation of the noise variance; then, we derive a bias correction method. Simulation results show that the proposed bias correction can significantly improve LMMSE CE performance, achieving up to 1dB gain in terms of block error rate (BLER).

In past few years we have observed an increase in the usage of RGBD sensors in mobile devices. These sensors provide a good estimate of the depth map for the camera frame, which can be used in numerous augmented reality applications. This paper presents a new visual inertial odometry (VIO) system, which uses measurements from a RGBD sensor and an inertial measurement unit (IMU) sensor for estimating the motion state of the mobile device. The resulting system is called the depth-aided VIO (DVIO) system. In this system we add the depth measurement as part of the nonlinear optimization process. Specifically, we propose methods to use the depth measurement using one-dimensional (1D) feature parameterization as well as three-dimensional (3D) feature parameterization. In addition, we propose to utilize the depth measurement for estimating time offset between the unsynchronized IMU and the RGBD sensors. Last but not least, we propose a novel block-based marginalization approach to speed up the marginalization processes and maintain the real-time performance of the overall system. Experimental results validate that the proposed DVIO system outperforms the other state-of-the-art VIO systems in terms of trajectory accuracy as well as processing time.

This paper proposes a novel deep learning architecture for semantic segmentation. The proposed Global and Selective Attention Network (GSANet) features Atrous Spatial Pyramid Pooling (ASPP) with a novel sparsemax global attention and a novel selective attention that deploys a condensation and diffusion mechanism to aggregate the multi-scale contextual information from the extracted deep features. A selective attention decoder is also proposed to process the GSA-ASPP outputs for optimizing the softmax volume. We are the first to benchmark the performance of semantic segmentation networks with the low-complexity feature extraction network (FXN) MobileNetEdge, that is optimized for low latency on edge devices. We show that GSANet can result in more accurate segmentation with MobileNetEdge, as well as with strong FXNs, such as Xception. GSANet improves the state-of-art semantic segmentation accuracy on both the ADE20k and the Cityscapes datasets.

We propose a novel 5D video stabilization solution based on sensor vision fusion. Traditional gyroscope based video stabilization approaches only stabilize the 3D rotation of a camera. They often suffer in scenes with highly dynamic translational movements. In this paper, we model the residual camera motion after 3D rotation based stabilization as residual 2D translation in the image plane, which can be estimated using visual information. The proposed 5D stabilization is achieved by carefully combining sensor based 3D rotation stabilization and vision based residual 2D translation stabilization. The 5D stabilization is prototyped on a smartphone for real time recording. We demonstrate its advantages over two state-of-the-art methods. The 5D stabilization can also be used to solve the challenging foreground object stabilization problem.

Pilot-aided channel estimation has been popular for 4G wideband communication systems. Even though it is convenient, its limitation comes from the fact that the performance is bounded by the density of pilot symbols. Increasing pilot symbols improves channel estimation quality, but it also hurts bandwidth efficiency. In this paper, we advocate data-aided channel estimation for 5G. It is potentially the only solution to improve channel estimation quality without increasing pilot density. In particular, we aim to answer two fundamental questions for data-aided channel estimation: (1) What is the optimal scheme for data-aided channel estimation? (2) How can we make the optimal scheme practical by reducing its complexity? We present how we tackle these problems of data-aided channel estimation and show that the proposed iterative scheme yields significant gain in long term evolution (LTE) signal demodulation.

The mutual information between modulation- constrained input and its output through Gaussian channels has been studied for various purposes such as physical layer abstraction. While the capacity of multiple-input and multiple-output (MIMO) Gaussian channels is well understood and its simple formula exists, this modulation- constrained mutual information for MIMO Gaussian channels has only been computed numerically and its expression for online computation has not been found. To tackle this challenge, this paper proposes several new bounds on the mutual information for MIMO Gaussian channels, which can be used for approximation. The proposed bounds are expressed in the forms that can be computed online. Moreover, these bounds are improved by taking a minimum of upper bounds and a maximum of lower bounds. The proposed bounds are compared with the mutual information obtained through simulation.

Recently, LTE Release 12 and IEEE 802.11ac adopted 256-quadrature amplitude modulation (QAM) as a tool to accommodate high demand of data throughput in wireless systems. Furthermore, currently developing IEEE 802.11ax considers 1024-QAM to further increase the Wi-Fi throughput. In this paper, we propose low complexity schemes in soft detection of high order QAM symbols in multiple input multiple output (MIMO) channels with or w/o prior information (from decoder output). We focus on 256-QAM with two transmitted layer MIMO with the understanding that the scheme can be generalized to higher order modulation and higher rank MIMO channels. We propose an initial candidate reduction (ICR) scheme with 128 initial candidates (refer to the scheme as ICR-128) to reduce the number of Euclidean distance (ED) calculations by half in comparison with soft maximum likelihood (ML) detection and as a result reduce the total hardware size by almost 50%. For initial candidate set selection we rely on linear minimum mean square error (MMSE) detection. The main advantage of the proposed scheme is in initial candidate set selection where we only need to detect I and Q sign of the target layer which simplifies MMSE detection. We further investigate the use of prior information in initial candidate set selection with MMSE soft interference cancellation (MMSE-SIC) which uses prior information in case of iterative detection and decoding (IDD) and/or hybrid automatic re- transmission request (HARQ) re-transmissions. To avoid computational complexity of MMSE-SIC, we propose a low complexity initial detection with prior information.

In Long-Term Evolution (LTE) downlink control channel, a large number of blind decoding attempts are made, while the number of valid codewords is limited. The blind decoding results are then verified using a 16-bit cyclic redundancy check (CRC). However, even with the 16-bit CRC, the false alarm (FA) rate of such blind decoding is inevitably high. This paper investigates the problem of pruning of blind decoding results for reduction of the FA rate. To the best of our knowledge, the approach using a soft correlation metric (SCM) shows the best FA reduction performance among existing schemes. However, following the Bayes principle, we propose novel likelihood-based pruning that provides systematic balancing between the FA rate and the miss (MS) rate. Moreover, the simulation results show that the signal-to-noise ratio (SNR) gain of our proposed scheme is unbounded, with respect to the SCM-based scheme, in the independent and identically distributed (i.i.d.) Rayleigh fading channel. Moreover, the proposed scheme is shown to be less complex than the existing scheme. Finally, it is proved that, as SNR increases, the proposed approach has the decision error probability that approaches the minimum value yielding near-optimal contraction of Voronoi regions for pruning of blind decoding results.

This paper considers the problem of unknown boosting factor estimation for Long Term Evolution (LTE) downlink control channel, whose signal can be boosted or deboosted for power control. The boosting factor needs be estimated and utilized to deploy advanced receiver algorithms. We first show that an iterative algorithm can be used to find the solution for this estimation problem. However, the use of iterative algorithms poses numerous modem implementation challenges. For this reason, we investigate non-iterative estimation of the boosting factor and propose a novel approach based on joint estimation of the boosting factor and modulation symbols. The main idea of the proposed scheme is to utilize various techniques such as bias reduction and approximate dimension reduction and improve the estimation performance of this baseline method. The simulation results show that the proposed non-iterative algorithm can achieve near optimal performance close to that of the maximum likelihood (ML) solution for a wide range of signal-to-noise ratio (SNR).

In cellular communications, user equipment (UE, i.e., mobile device)-side interference cancellation (IC) along with multicell coordinated scheduling can significantly reduce the effect of the downlink intercell interference. To aid UE-side IC, a study item, called network-assisted interference cancellation and suppression (NAICS), has been initiated for Long Term Evolution (LTE) Advanced Release 12. Among NAICS receivers, this paper considers a receiver with interference-aware successive decoding (IASD) capability, which is one of the most advanced UE-side IC techniques. The IASD achievable transmission rate is dependent on the interferer transmission rate. Thus, coordination among multiple cells for rate and UE selection would be necessary for the overall network performance enhancement. In this paper, we consider a single dominant interference model and propose an optimal single-user rate selection algorithm based on the belief-propagation framework. In the multi-user-per-cell case we propose several UE and rate selection algorithms and analyze their performance.

In the presence of co-channel interference in cellular networks, interference mitigation by detecting the desired signal jointly with the interference promises considerable gain over the conventional way of handling the interference as colored Gaussian. Even though such interference-aware detection can improve the performance, it requires some information on the interference. In particular, the modulation format of the interference has to be classified to this end, when it is not signaled by the network explicitly. This paper investigates interference modulation classification methods for interference-aware joint detection. We propose an iterative interference modulation classification algorithm that utilizes the decoded information of the desired signal in order to cancel the desired signal from the received signal. After the cancellation, the remaining signal can be treated as interference plus noise so that we can classify the modulation format of the interference at reduced complexity with small performance loss due to decoding errors.

It has been generally believed that the multiple-input multiple-output channel capacity grows linearly with the size of antenna arrays. In terms of degrees of freedom, linear transmit and receive arrays of length L in a scattering environment of total angular spread \Ω\ asymptotically have \Ω\L degrees of freedom. In this paper, it is claimed that the linear increase in degrees of freedom may not be attained when scattered electromagnetic fields in the underlying scattering environment are statistically correlated. After introducing a model of correlated scattering, which is referred to as the colored scattering model, we derive a capacity upper bound, assuming that the channel is known perfectly at the receiver and in distribution at the transmitter. Unlike the uncorrelated case, the prelog factor of the capacity, i.e., the number of degrees of freedom, in the colored scattering channel is asymptotically limited by \Ω\·min{L, 1/ ΓI} where Γ is a parameter determining the extent of correlation. In other words, for very large arrays in the colored scattering environment, degrees of freedom can get saturated to an intrinsic limit rather than increasing linearly with the array size.

This paper considers the problem of mismatched hypothesis testing, where approximate likelihood functions are used instead of true likelihood functions. Given a hypothesis testing problem, the maximum likelihood (ML) solution is known to be optimal when true likelihood functions are used, but the optimality does not hold anymore if mismatched approximate likelihood functions are employed instead, in order to reduce computational complexity, for instance. In this paper, we investigate the mismatched ML framework using approximate likelihood functions, while the mismatches between the true and the approximate likelihood functions are corrected by additive compensating constants. The probability of error of this mismatched hypothesis testing is analyzed asymptotically, assuming a large number of samples, and the compensating constants that maximize the error exponent are established. The general results on the mismatched hypothesis testing are then utilized in designing and optimizing a digital modulation classifier with low complexity.

This comment corrects a few errors found in the derivation of the maximum likelihood estimate of differential phase in the paper, "A Technique for Orthogonal Frequency Division Multiplexing Frequency Offset Correction." We show that the problem of differential phase estimation can be considered as an estimation problem in the presence of nuisance parameters, which does not satisfy strong ancillarity. The approach in the above paper to solve this problem can be understood as conditioning for elimination of nuisance parameters but without taking proper steps. After making corrections on the proof, it is demonstrated that the estimator in the above paper is inherently suboptimal and thus prior knowledge on the nuisance parameters, if available, can be utilized to further improve the estimation performance.

This paper deals with the problem of classification of digital modulation. In particular, we develop and propose a practical modulation classification scheme based on the likelihood of observations. While ML classification is well known and shows the optimal performance, its computational complexity prevents it from being easily implemented in hardware. On the contrary, our proposed scheme has low computational complexity and near optimal classification performance. Moreover, this scheme is designed to perform in fast fading channels. It is shown that our proposed classifier takes advantage of the channel variation without loosing near optimality.

This paper investigates the problem of reliable tone interference estimation from noisy observations. Performing the DFT of time domain observations is a commonly used tone estimation technique. However, the use of one DFT block of frequency domain samples may not yield reliable estimation of tone interference. To solve this problem, we propose to combine multiple DFT blocks of frequency domain observations which are naturally provided in OFDM systems, improving the reliability of tone interference estimation. To this end, a frequency domain DFT is used to process multiple blocks of observations. It is shown that our estimator can achieve the performance close to the Cramer-Rao bound and the conventional DFT-based estimation with much higher complexity.

We consider beam selection using a fixed beamforming network (FBN) at a base station withMarray antennas. In our setting, a Butler matrix is deployed at the RF stage to formMbeams, and then the best beam is selected for transmission. We introduce some properties of the noncentral chi-square distribution and prove the resulting properties of the beam selection gain verifying that beam selection is superior to antenna selection in Rician channels with anyK-factors. Furthermore, we find asymptotically tight stochastic bounds of the beam selection gain, which yield approximate closed form expressions of the expected selection gain and the ergodic capacity. Beam selection has the order of growth of the ergodic capacity Θ(logM) regardless of user location in contrast to Θ(log(logM)) for antenna selection.

For a multiple-antenna system, we find a simple and accurate expression for the asymptotic distribution of the antenna selection gain when the transmitter selects the transmit antenna with the strongest channel. We use this to estimate the underlying channel capacity distributions and obtain the approximate ergodic capacity. This estimate is compared with upper and lower bounds. This analysis demonstrates that unlike multiple-input/multiple-output (MIMO) systems, the channel for antenna selection systems hardens at a slower rate, and thus a significant multiuser scheduling gain can exist-Theta (1/logm) for channel selection as opposed to Theta (1/radicm) for MIMO, where m is the number of transmit antennas.

We consider a wireless transmission scenario, when a base station is endowed with a fixed beamforming network, where M antennas are employed at the base station to point beams to predetermined azimuthal angles. In our setting, a Butler matrix is deployed at the RF stage to form M beams, and then the best beam is selected for transmission. We derive the distribution of the beam selection gain for this scenario under a Rician channel assumption as a function of both the azimuthal location of the remote unit and the Rician A'-factor. Using some key properties of the noncentral chi-square distribution, we prove that beam selection outperforms antenna selection.

In this paper, we analyze the performance of a communication system that employs Protocol Assisted Switched Diversity (PASD) antennas under Rayleigh fading channel conditions. The PASD system accumulates time-displaced blocks, each with the same information but using different antennas, and then combines their symbols using maximal ratio combining (MRC). In the PASD system, quality of service (QoS) is ensured by comparing the available signal-to-noise ratio (SNR) at the output of the combiner against a preset threshold value. Our analysis shows that this method reduces the number of transmissions significantly as the channel deteriorates as compared with ARQ without memory while it can satisfy the symbol error rate (SER) requirement for most of blocks. It is also shown that deploying more than one antenna is important to stabilize the system when the channel becomes static.

For a multiple antenna system, we compute the asymptotic distribution of antenna selection gain when the transmitter selects the transmit antenna with the strongest channel. We use this to asymptotically estimate the underlying channel capacity distributions, and demonstrate that unlike multiple- input/multiple-output (MIMO) systems,the channel for antenna selection systems hardens at a slower rate, and thus a significant multiuser scheduling gain can exist. Additionally, even without this scheduling gain, it is demonstrated that transmit antenna selection systems outperform open loop MIMO systems at low signal-to-interference-plus-noise ratio (SINR) regimes, particularly for small number of receive antennas. This may have some implications on wireless system design, because most of the users in modern wireless systems have low SINRs.