We propose a generalized convolutional neural network (CNN) architecture that first decomposes the input signal into subbands by an adaptive filter bank structure, and then uses convolutional layers to extract features from each subband independently. Fully connected (FC) layers finally combine the extracted features to perform classification. The proposed architecture restrains each of the subband CNNs from learning using the entire input signal spectrum, resulting in structural regularization. Our proposed CNN architecture is fully compatible with the end-to-end learning mechanism of typical CNN architectures and learns the subband decomposition from the input dataset. We show that the proposed CNN architecture has attractive properties, such as robustness to input and weight-and-bias quantization noise, compared to regular full-band CNN architectures. Importantly, the proposed architecture significantly reduces computational costs, while maintaining state-of-the-art classification accuracy. Experiments on image classification tasks using the MNIST, CIFAR-10/100, Caltech-101, and ImageNet-2012 datasets show that the proposed architecture allows accuracy surpassing state-of-the-art results. On the ImageNet-2012 dataset, we achieved top-5 and top-1 validation set accuracy of 86.91% and 69.73%, respectively. Notably, the proposed architecture offers over 90% reduction in computation cost in the inference path and approximately 75% reduction in back-propagation (per iteration) with just a single-layer subband decomposition. With a two-layer subband decomposition, the computational gains are even more significant with comparable accuracy results to the single-layer decomposition.

The use of cellular networks for massive machine-type communications (mMTC) is an appealing solution due to the wide availability of cellular infrastructure.Estimating the number of devices (network load) is vital for efficient allocation of the available resources, especially for managing the random access channel (RACH) of the network. This paper considers a two-priority RACH and proposes two network load estimators: a maximum likelihood (ML) estimator and a reduced complexity (RCML) variant. The estimators are based on a novel model of the random access behavior of the devices coupled with a flexible analytical framework to calculate the involved probabilities. Monte Carlo simulations demonstrate the accuracy of the proposed estimators for different network configurations. Results depict increased estimation accuracy using non-uniform preamble selection probabilities compared to the common uniform probabilities at no extra computational cost.

We consider a narrowband multi-user MIMO reconfigurable intelligent surface (RIS)-assisted wireless communication system and use tensor signal modelling techniques to jointly estimate all communication channels including the RIS-assisted link and the direct-path link between the access point and user equipment. We model the received signal as a third-order tensor comprising two additive CANDECOMP/PARAFAC (CP) decomposition terms corresponding to the direct-path and the RIS-assisted links. Based on this model we propose an enhanced iterative alternating least squares (E-ALS) algorithm to simultaneously estimate both the direct-path and RIS channels, and we derive the corresponding Cramér-Rao Bounds (CRB). Numerical results show that compared to recent previous works which estimate the direct-path and RIS links during separate training stages, the E-ALS method provides a more accurate estimate by efficiently using all pilots transmitted throughout the full training duration without turning the RIS OFF. For a sufficient number of transmitted pilots, the E-ALS method’s accuracy comes close to the CRB for the RIS channels and attains the CRB for the direct-path channel.

Device-to-device (D2D) communication is an enabling technology for fog computing by allowing the sharing of computation resources between mobile devices. However, temperature variations in the device CPUs affect the computation resources available for task offloading, which unpredictably alters the processing time and energy consumption. In this paper, we address the problem of resource allocation with respect to task partitioning, computation resources and transmit power in a D2D-aided fog computing scenario, aiming to minimize the expected total energy consumption under probabilistic constraints on the processing time. Since the formulated problem is non-convex, we propose two sub-optimal solution methods. The first method is based on difference of convex (DC) programming, which we combine with chance-constraint programming to handle the probabilistic time limitations. Considering that DC programming is dependent on a good initial point, we propose a second method that relies on only convex programming, which eliminates the dependence on user-defined initialization. Simulation results demonstrate that the latter method outperforms the former in terms of energy efficiency and run-time.

We propose a convolutional neural network (CNN) architecture for image classification based on subband decomposition of the image using wavelets. The proposed architecture decomposes the input image spectra into multiple critically sampled subbands, extracts features using a single CNN per sub-band, and finally, performs classification by combining the extracted features using a fully connected layer. Processing each of the subbands by an individual CNN, thereby limiting the learning scope of each CNN to a single subband, imposes a form of structural regularization. This provides better generalization capability as seen by the presented results. The proposed architecture achieves best-in-class performance in terms of total multiply-add-accumulator operations and nearly best-in-class performance in terms of total parameters required, yet it maintains competitive classification performance. We also show the proposed architecture is more robust than the regular full-band CNN to noise caused by weight-and-bias quantization and input quantization.

Synchronization is a challenging problem especially for distributed systems, such as out-of-coverage D2D networks, as no common reference time is available. In such cases, devices use distributed synchronization algorithms, however, accurately determining when to stop the synchronization process is as challenging as achieving synchronization since they do not have the synchronization status of other devices in the network. From energy efficiency and performance perspective, the synchronization process should be stopped at all devices at the same time. In addition, to counteract the effect of propagation delays during synchronization, timing-advance (TA) clocks should be employed. This could be achieved in the synchronization process, however, after the devices are synchronized, there is no central mechanism to instruct them on TA clocks for transmitting or receiving data packets. In this paper, we propose a synchronization algorithm which, in an energy-efficient manner, allows devices to (i) acquire the synchronization status of others and terminate the synchronization process as soon as all devices in the network are synchronized, (ii) allow the synchronized devices to properly advance/regress their clocks prior to data communication by tracking their relative timing. We numerically demonstrate that the maximum synchronization error over multipath channels is around 0.6μs and it can be maintained during data communication.

This paper addresses the design of multiuser MIMO amplify-and-forward relaying within a cloud radio access network (C-RAN) from an energy-efficiency perspective. The aim is to jointly select remote radio heads and optimize their transceiver in order to assist the communication between multiple source-destination pairs. We formulate the design problem as an interference leakage minimization subject to per-relay power constraints along with linear signal preserving constraints at the destinations. To obtain an energy efficient relaying solution, the objective function is penalized with a regularization term which promotes group-sparsity among the resultant relaying weights. A low-complexity iterative algorithm based on the alternating direction method of multipliers (ADMM) is then proposed to solve the regularized problem. Simulation results demonstrate the explicit benefits of the proposed algorithm, which results in notably lower power consumption and computational complexity than conventional relaying design methods.

Despite the importance of distributed learning, few fully distributed support vector machines exist. In this paper, not only do we provide a fully distributed nonlinear SVM; we propose the first distributed constrained-form SVM. In the fully distributed context, a dataset is distributed among networked agents that cannot divulge their data, let alone centralize the data, and can only communicate with their neighbors in the network. Our strategy is based on two algorithms: the Douglas-Rachford algorithm and the projection-gradient method. We validate our approach by demonstrating through simulations that it can train a classifier that agrees closely with the centralized solution.

Time synchronization is a key aspect of device-to-device (D2D) schemes, particularly in decentralized networks where no reference time is available. Distributed phase-locked loops (DPLL) is a synchronization algorithm well suited for decentralized situations. In this work, we study DPLL in the context of 5G networks, where we include in our analysis several practical aspects of D2D communication, such as propagation delays, multipath propagation, and the use of single-carrier frequency division multiple access (SC-FDMA). We propose practical methods to compensate for their effects, and introduce new performance metrics to evaluate the merits of the synchronization algorithm. Through simulations at the physical layer, which capture the effects of analog-digital conversions, we demonstrate that time synchronization in a decentralized setting is possible under the constraints specified by the 3GPP for D2D applications.

Deductive logic and its variants enjoy the common property of monotonicity. For tasks such as inductive reasoning and belief revision, this was eventually deemed a serious flaw, prompting attempts to construct non-monotonic versions of logic. With the introduction of the idea of probabilistic reasoning to AI, particularly with the advent of Bayesian networks (BNs), the aforementioned monotonicity was no longer an issue: Probability is inherently non-monotonic. In this work, we introduce the notion of relevance effect which bears on exploiting BNs to generate realizations of relevant variables to be used for potentially improving the performance of a learning model on a supervised classification task. We explore the potential of using the relevance effect in the context of Deep Belief Networks (DBNs) with a focus on relational domains. We show that although the idea is at odds with the non-monotonicity of probabilistic reasoning, we attain an improvement in learning performance in different simulations on both synthetic and real-world scenarios. The observation that adopting this notion has improved the performance of a powerful model like DBNs hints to its potential to be practiced so as to enhance the performance of supervised learning methods in general. We furthermore highlight the connections as well as the implications of our work to the psychology literature.

We present a new Kernel Recursive Least Squares (KRLS) algorithm that is able to efficiently track time-varying systems. In order to alleviate the detrimental effect of a large dictionary size on the algorithm's tracking ability, we decouple the equality between dictionary size and weight vector size, an equality that has been encountered in all previous KRLS algorithms. In the proposed method, the maximum size of the weight vector is fixed and is independent from the dictionary size. We introduce the Kernel Subspace Pursuit algorithm which we use to choose a subset of the dictionary that tracks best the most recent received data samples. The selected dictionary elements are then used in the KRLS iterations. We show through simulations that our algorithm outperforms existing KRLS algorithms in tracking time-varying systems.

Orthogonal Matching Pursuit (OMP) can denoise a signal by greedily approximating a least-squares (LS) estimate as a linear combination of elements (atoms) of a dictionary. OMP iteratively decomposes a signal through deterministic atom selections at each iteration step. Recently proposed randomized OMP algorithms employ random atom selections instead and have the potential to further improve denoising. Typically, the best approximation from these algorithms can be obtained only within a narrow range of iterations. In this paper, we propose a novel multi-stage randomized OMP (MS-ROMP) denoising approach that performs successive ROMP runs, each denoising the obtained estimate from the previous one. We show through simulations that, under certain conditions, this can significantly improve denoising performance by producing a good approximation after any number of iterations beyond the sparsity level.

We consider the problem of channel estimation for OFDM-based amplify-and-forward (AF) two-way relay networks (TWRNs). While previous works have adopted a pilot-based approach, we propose a semi-blind approach that exploits both the transmitted pilots as well as the received data samples to improve the estimation performance. Our proposed semi-blind estimator is based on the Gaussian maximum likelihood (GML) criterion which treats that data symbols as Gaussian-distributed nuisance parameters. The GML estimates are obtained using an iterative quasi-Newton method. To assist in the estimation of the individual channels, we adopt a superimposed training strategy at the relay. We design the pilot vectors of the terminals and the relay to optimize the estimation performance. Furthermore, we derive the semi-blind and pilot-based Cramer-Rao bounds (CRBs) to use as performance benchmarks. Finally, we use simulation studies to show that the proposed method provides substantial improvements in estimation accuracy over the conventional pilot-based estimation and that it approaches the semi-blind CRB as SNR increases. These improvements are possible using only a limited number of OFDM data blocks, which demonstrates the practicality of the semi-blind approach.

In this paper, we derive the Cramer-Rao bound (CRB) for semiblind channel estimation in amplify-and-forward two-way relay networks employing square QAM, assuming flat-fading channel conditions. The derived bound is exact as it is based on the true likelihood function that takes into account the statistics of the transmitted data symbols. Using the new bound, we show that exploiting even a limited number of transmitted data symbols in addition to the pilot symbols leads to substantial estimation accuracy improvements over conventional pilot-based estimation. We also propose a semiblind expectation-maximization-based estimation algorithm that performs very close to the exact CRB at an affordable computational cost. The superior accuracy of the semiblind approach makes it possible to significantly reduce the training overhead for channel estimation, thus offering a higher throughput and a better tradeoff between accuracy and spectral efficiency. We also derive the modified CRB, which approximates the exact CRB at high SNR for low modulation orders.

We present a joint estimator of the time of arrival (TOA) and angle of arrival (AOA) for impulse radio ultrawideband (UWB) systems in which an antenna array is employed at the receiver. The proposed method consists of two steps: 1) preliminary estimation of the TOA and the average power delay profile (APDP) using energy-based threshold crossing and log-domain least-squares fitting, respectively; and 2) joint TOA refinement and AOA estimation by local 2-D maximization of a log-likelihood function (LLF) that employs the preliminary estimates from the first step. The derivation of the LLF relies on an original formulation in which the superposition of images from secondary paths is modeled as a Gaussian random process, whose second-order statistical properties are characterized by a wideband space-time correlation function. In addition to the APDP, this function incorporates a special gating mechanism to represent the onset of the secondary paths, thereby leading to a novel form of the LLF. Closed-form expressions for the Cramer-Rao bound on the variance of the TOA and AOA estimators are also derived, which formally take into account pulse overlap through this gating mechanism. In simulation experiments based on multipath UWB channel models featuring both diffuse and directional image fields, our approach exhibits superior performance to that of a competing scheme from the recent literature.

We consider the problem of channel estimation for OFDM-based amplify-and-forward (AF) two-way relay networks (TWRNs). Unlike previous works which were based on a fully pilot-based approach, we propose a semi-blind approach that exploits both the transmitted pilots as well as the received data samples to provide an enhanced estimation performance. Superimposed training is adopted at the relay to assist in the estimation of the individual channels. We base our semi-blind estimator on the maximum-likelihood (ML) criterion and employ an iterative low-complexity Quasi-Newton method to obtain the ML semi-blind channel estimates. As a performance benchmark we derive the semi-blind Cramer-Rao bound (CRB). Using simulation studies, we show that the proposed approach provides a substantial improvement in estimation accuracy over the conventional pilot-based approach.

A novel joint TOA and AOA estimator is proposed for impulse radio Ultra-Wideband (IR-UWB) systems, in which a uniform linear array of antennas is employed at the receiver. The proposed method consists of two steps: (1) coarse estimation of the TOA and the average power delay profile; (2) joint TOA refinement and AOA estimation by maximization of a novel log likelihood function (LLF) using the coarse estimates from the first step. The derivation of the LLF is based on an original approach in which the pulse image from the primary path is modeled as a deterministic component while the superposition of the images from the secondary paths is modeled as a Gaussian random process. In addition, a special gating mechanism is used to characterize the secondary paths, thereby leading to a previously unknown form of the LLF in step (2). According to simulation experiments based on standard UWB channel models, our approach exhibits superior performance to that of a competing scheme from the recent literature.

Sub-Nyquist maximum likelihood (ML)-based time of arrival (TOA) estimation methods for ultra-wideband (UWB) signals normally assume a priori knowledge of the UWB channel in the form of the average power delay profile (APDP). In practice however, and despite its importance, the APDP is not always available. To address this issue, we develop in this paper a joint estimator of TOA and APDP. Knowing that the APDP of a UWB channel usually consists of several clusters, each with specific exponential decay rate, a parametric APDP model of this type is employed. The parameters of this model are estimated via a least-squares fitting approach; then the estimated APDP is used to form a likelihood function and obtain a ML estimator of the TOA. Simulations show that the TOA estimated jointly in this way achieves a good accuracy in practical scenarios. The proposed APDP estimate can also help to boost the performance of previously reported TOA estimators that assume a priori APDP knowledge, although the proposed ML scheme generally offers superior performance.

We consider the problem of channel estimation for amplify-and-forward two-way relays assuming channel reciprocity and M-PSK modulation. In an earlier work, a partially-blind maximum-likelihood estimator was derived by treating the data as deterministic unknowns. We prove that this estimator approaches the true channel with high probability at high signal-to-noise ratio (SNR) but is not consistent. We then propose an alternative estimator which is consistent and has similarly favorable high SNR performance. We also derive the Cramer-Rao bound on the variance of unbiased estimators.

We consider the problem of channel estimation for amplify-and-forward (AF) two-way relay networks (TWRNs). The majority of works on this problem develop pilot-based algorithms that allocate significant resources for training. We will show in this work that such overhead is not necessary when the terminals employ M-ary phase-shift keying (M-PSK). Using the constant-modulus nature of the transmitted symbols, we develop a relaxed blind maximum-likelihood (ML) channel estimator. We study the performance of the ML estimator in the high SNR and large sample-size scenarios, demonstrating that it performs well in both cases. As a benchmark, we also present and analyze an intuitive low-complexity estimator based on sample-averaging. Simulation studies are used to compare the mean-squared error performance of the two algorithms.

We analyze the mean-squared error (MSE) performance of widely linear (WL) and conventional subspace-based channel estimation for single-input multiple-output (SIMO) flat-fading channels employing binary phase-shift-keying (BPSK) modulation when the covariance matrix is estimated using a finite number of samples. The conventional estimator suffers from a phase ambiguity that reduces to a sign ambiguity for the WL estimator. We derive closed-form expressions for the MSE of the two estimators under four different scenarios which vary in the amount and accuracy of the information available for ambiguity resolution. Our work demonstrates that the relative performance of WL and conventional subspace-based estimators is strongly related to the accuracy of ambiguity resolution and shows that the less information available about the actual channel for ambiguity resolution, or the lower the accuracy of this information, the more favorable the WL estimator becomes.

We develop a novel, microwave beamforming algorithm based on the minimum mean-square error (MMSE) criterion. Relative to perturbation algorithms, the proposed method requires only a fraction of weight changes to estimate the beamformer weights, thereby making it more suitable for high-data-rate applications. Numerical simulations reveal that the proposed algorithm offers promising performance and is less susceptible to performance degradation than the perturbation algorithm when used with slow-switching and/or finite-resolution phase shifters and amplitude-control devices.

The problem of blind channel estimation for downlink space-frequency block coded multi-carrier code division multiple access (SFBC MC-CDMA) schemes is considered. For these schemes, the authors first develop a system model for complex modulated signals, which reduces the multichannel estimation problem to a single-input single-output problem. Then, they present an intuitive subspace-based channel estimation method along with the corresponding necessary and sufficient conditions under which the channel estimate is unique (within a complex scalar). Their studies highlight two interesting properties of SFBC MC-CDMA systems: (i) there is no antenna order ambiguity (also known as permutation ambiguity) even though only one spreading code is assigned to each user; (ii) channel identifiability is guaranteed, regardless of the channel zeros location. They also establish the unbiasedness of the channel estimates and derive closed-form expressions for the mean-square-error of the estimates as well as the corresponding Cramer-Rao bound (CRB). In the derivation of the CRB, they suggest a novel approach which assumes the knowledge of only the spreading code of desired user. This approach results in a tighter bound than the CRB derived based on the knowledge of all users' signatures.

Blind channel estimation algorithms for the downlink of space-frequency block coded multi-carrier code division multiple access (SFBC MC-CDMA) schemes are presented. We first formulate two channel estimation methods based on the second-order statistics (SOS) of the frequency-domain received signal, namely: (i) a minimum variance (MV) criterion, and (ii) a subspace-based approach. Then, we highlight the unique structure of the input covariance matrix. The matrix structure allows us to propose modifications of the presented algorithms that are computationally efficient and offer enhanced performance in practical cases where only an estimate of the received signal covariance matrix is available. Moreover, we address the issue of channel identifiability by investigating the necessary and sufficient conditions under which the channel estimates are unique (within a complex scalar).

We propose a method to approximate the bit-error-rate (BER) performance of antenna-array-based CDMA systems employing sample-matrix-inversion (SMI) minimum-variance-distortionless-response (MVDR) space-time (ST) linear receivers. In the proposed method, a form of the Gaussian Q(ldr) function is approximated by a finite series of exponentially decaying cosines that are optimized using a recently proposed relative error measure. The proposed method provides accurate closed form approximations to the BER performance for a wide range of practical scenarios with very low complexity.

A comprehensive performance analysis of the minimum variance (MV) channel estimator for multicarrier code-division multiple access (MC-CDMA) systems is presented. In particular, we provide closed form expressions for the asymptotic bias and the mean-square-error (MSE) of the estimates as well as the corresponding Cramer-Rao bound (CRB). The derived formula for the bias due to the additive noise characterizes accurately the behavior of the actual MV estimator in the case of heavy system loading or small processing gain. In addition, in the derivation of the CRB, we suggest a novel approach which assumes the knowledge of only the spreading code of the desired user. This approach results in a tighter bound than the CRB derived based on the knowledge of all users' signatures.

We consider the problem of joint estimation of direction-of-arrival (DoA), propagation delay, and complex channel gain for antenna-array-based DS/CDMA communication systems over frequency selective multipath channels. We propose a MUSIC-type estimation algorithm which utilizes the spatial smoothing preprocessing technique. The proposed algorithm essentially breaks the multipath-induced coherency within the received signals and recovers the full signal subspace spanned by all dominant signal paths of all users. This allows the use of MUSIC-type DoA and delay estimators for the individual paths of the user of interest. Based on the angle and timing information, we then estimate the multipath fading coefficients. Simulation results illustrate the effectiveness of this approach.

We propose a method to approximate the bit-error-rate (BER) performance of CDMA systems employing sample-matrix-inversion (SMI) maximum-signal-to-interference-plus-noise-ratio (MSINR) and SMI minimum-variance-distortionless-response (MVDR) linear receivers. In the proposed method, two forms of the Gaussian Q-function are approximated by finite series of exponentials and exponentially decaying cosines that are optimized using a relative error measure. The proposed method provides highly accurate closed form approximations to the BER performance for a wide range of practical scenarios with very low complexity

Minimum variance distortionless response (MVDR)-type blind channel estimation algorithms for the downlink of space-frequency block coded multi-carrier code division multiple access (SFBC-MC-CDMA) schemes are developed and analyzed. The performance of the channel estimator under sufficiently large data record size and small noise assumptions is studied by deriving analytical expressions for the bias and the variance of the estimator. Analysis of the mean-squared error (MSE) of the proposed estimator shows that at sufficiently high SNRs the squared norm of the bias can be considered negligible compared to the variance of the estimator. Furthermore, we establish a bound on the MSE of the biased estimates which is specifically derived for downlink transmissions where the knowledge of only the spreading code of the desired user is available. Finally, simulation examples illustrate the performance of the MVDR channel estimator and the accuracy of our analytical results

We investigate the problem of blind channel estimation for the downlink of a variable spreading gain (VSG) multi-rate DS/CDMA (MR CDMA) system which uses the orthogonal space-time block code (STBC) proposed by Alamouti. In contrast to traditional symbol-level STBC CDMA systems, we consider a chip-level ST block coding system for which we derive a simple subspace based blind channel estimation method. For this algorithm we investigate the necessary and sufficient conditions for the channel to be uniquely identifiable up to a complex multiplicative constant. Furthermore, we analyze the behavior of the channel estimation algorithm as a function of the data samples utilized, establishing its unbiasedness and providing a closed form expression for the mean-square-error (MSE) as well as the Cramer-Rao bound for the channel estimation performance

We consider the problem of identifying the location of a mobile client in an IEEE 802.11b wireless LAN. We propose an estimation algorithm that utilizes received signal strength (RSS) measurements provided by most 802.11b physical layer implementations. In contrast to traditional radio map-based methods that require a burdensome map building phase, the proposed algorithm requires no prior training by employing a probabilistic RSS model that is derived from indoor signal propagation models and the blueprint of the building in which the location estimation system is being deployed. Experimental studies demonstrate the accuracy of the proposed method.

A combined space-frequency (SF) block coding and multi-carrier code division multiple access (MC-CDMA) scheme for downlink transmissions is presented, in which SF block coding is performed at the chip level, i.e. after code spreading. For this scheme, we develop linear single-user joint SF decoding and detection algorithms that exhibit low complexity, low decoding delay and enhanced performance for frequency-selective fading channels compared to their symbol-level counterparts. In addition, it is demonstrated that by exploiting the signal structure imposed by the chip-level SF block coding, blind channel estimation (with only a scalar ambiguity) is feasible, without assigning multiple signatures to each user. Finally, simulation examples demonstrate the performance of the proposed algorithms.

Sensor networks have exciting potential applications in agriculture and medicine, where after the application of treatment, it is beneficial not merely to track the response but to assess the causal impact of the treatment reception. We describe a distributed algorithm for the evaluation of the average causal effect of treatment reception upon response. Our procedure applies the expectation-maximization algorithm across a graphical model of the system, using local message-passing techniques. The key collaborative step in the algorithm is simple message aggregation and averaging, which we perform over a tree network topology. Finally, for completeness purposes, we describe a simple framework for the construction and maintenance of the tree topology that provides a robust mechanism for executing the algorithm using spread-spectrum or ultra-wideband communication.

The presence of the desired signal during estimation of the minimum mean-square error (MMSE)/minimum-variance distortionless-response (MVDR) and auxiliary-vector (AV) filters under limited data support leads to significant signal-to-interference-plus-noise ratio (SINR) performance degradation. We quantify this observation in the context of direct-sequence code-division multiple-access (DS-CDMA) communications by deriving close approximations for the mean-square filter estimation error, the probability density function of the output SINR, and the probability density function of the symbol-error rate (SER) of the sample matrix inversion (SMI) receiver evaluated using both a desired-signal-"present" and desired-signal-"absent" input covariance matrix. To avoid such performance degradation, we propose a DS-CDMA receiver that utilizes a simple pilot-assisted algorithm that estimates and then subtracts the desired signal component from the received signal prior to filter estimation. Then, to accommodate decision-directed operation, we develop two recursive algorithms for the on-line estimation of the AV and MMSE/MVDR filter and we study their convergence properties. Finally, simulation studies illustrate the SER performance of the overall receiver structures.

We investigate the data-record size requirements of sample-matrix-inversion-based minimum-variance-distortionless response and maximum-signal-to-interference-plus-noise-ratio adaptive algorithms to meet a given performance objective in joint space-time signal-detection problems for direct-sequence code-division multiple-access systems. We derive closed-form expressions that provide the data-record size that is necessary to achieve a given performance confidence level in a neighborhood of the optimal performance point, as well as expressions that identify the performance level that can be reached for a given data-record size. This is done by using close approximations of the involved probability density functions. The practical significance of the derived expressions lies in the fact that the expressions are functions of the number of antenna elements, the number of multipaths, and the system spreading gain only, while they depend neither on the ideal input covariance matrix, which is not known in most realistic applications, nor on the exact ideal performance value.

We consider a multilayer perceptron neural network (NN) receiver architecture for the recovery of the information bits of a direct-sequence code-division-multiple-access (DS-CDMA) user. We develop a fast converging adaptive training algorithm that minimizes the bit-error rate (BER) at the output of the receiver. The adaptive algorithm has three key features: i) it incorporates the BER, i.e., the ultimate performance evaluation measure, directly into the learning process, ii) it utilizes constraints that are derived from the properties of the optimum single-user decision boundary for additive white Gaussian noise (AWGN) multiple-access channels, and iii) it embeds importance sampling (IS) principles directly into the receiver optimization process. Simulation studies illustrate the BER performance of the proposed scheme.

For pt.I see ibid., vol.51, p.983-94 (2003). We investigate the coarse synchronization performance of blind adaptive linear self-synchronized receivers for asynchronous direct-sequence code-division multiple-access communications under finite data record adaptation. Based on transformation noise modeling techniques, three alternative methods are developed, leading to analytical expressions that approximate the probability of coarse synchronization error of matched-filter-type and minimum-variance distortionless-response-type receivers. The expressions are explicit functions of the data record size and the filter order and reveal the effect of short data-record sample matrix-inversion implementations on the coarse synchronization performance. Besides their theoretical value, the derived expressions provide simple, highly-accurate alternatives to computationally demanding performance evaluation through simulations. The effect of the data record size on the probability of coarse synchronization error is further quantified through the use of a receiver synchronization resolution metric. Numerical and simulation studies examine the accuracy of the theoretical developments and show that the derived expressions approximate closely the actual coarse synchronization performance.

Blind adaptive linear receivers are considered for the demodulation of direct-sequence code-division multiple-access signals in asynchronous transmissions. The proposed structures are self-synchronized in the sense that adaptive synchronization and demodulation are viewed and treated as an integrated receiver operation. Two computationally efficient combined synchronization/demodulation schemes are proposed, developed, and analyzed. The first scheme is based on the principles of minimum-variance distortionless-response processing, while the second scheme follows the principles of auxiliary-vector filtering and exhibits enhanced performance in short data-record scenarios. In both cases, the resulting receiver is a linear structure of order exactly equal to the system processing gain. Simulation studies included in this paper demonstrate the coarse synchronization as well as the bit-error rate performance of the proposed strategies.

We investigate the relative output signal-to-interference-plus-noise ratio (SINR) performance of two linear direct-sequence code-division multiple-access multiuser detectors: the full decorrelator and the partial decorrelator. We derive necessary and sufficient conditions on the system parameters under which the partial decorrelator outperforms the full decorrelator in the output SINR sense. As a side study, we consider a blind implementation of the full decorrelator that is based on eigendecomposition of the interference-plus-noise autocovariance matrix and can be easily modified to provide a partial decorrelator. Simulation studies illustrate the relative SINR and bit-error rate performance of the full and partial decorrelator under perfectly known and sample-average-estimated input statistics.

We investigate the data-record-size requirements of the minimum-variance-distortionless-response beamformer to meet a given performance objective in signal detection and direction-of-arrival estimation problems. For signal detection problems we consider the output-energy performance measure while for direction-of-arrival estimation problems we adopt a spectrum-based measure defined as the ratio between the estimated and the ideal spectrum. In both cases, closed form expressions are derived that provide the data record size that is necessary to achieve a given performance confidence level in a neighborhood of the optimal performance point. This is done by utilizing close approximations of the involved probability density functions and Markoff-type inequalities. The practical significance of the derived expressions lies in the fact that the expressions are functions of the number of antenna elements only, while they are independent of the ideal input covariance matrix which is not known in most realistic applications.

We investigate the coarse synchronization performance of matched-filter-type (MF) and minimum-variance-distortion less-response-type (MF) near self-synchronized receivers for asynchronous direct-sequence code-division-multiple-access communications under finite data record adaptation. Analytic expressions are derived that approximate closely the probability of coarse synchronization error and provide low-cost highly-accurate alternatives to the computationally demanding performance evaluation through simulations. The expressions are explicit functions of the data record size N and the filter order p and reveal the effect of short-data-record sample-matrix-inversion (SMI) implementations on the coarse synchronization performance.

We consider blind adaptive linear receivers for the demodulation of DS/CDMA signals in asynchronous transmissions. The proposed structures are self-synchronized in the sense that adaptive synchronization and demodulation are viewed and treated as an integrated receiver operation. Two computationally efficient combined synchronization/demodulation schemes are proposed, developed and analyzed. The first scheme is based on the principles of minimum-variance-distortionless-response (MVDR) processing, while the second scheme follows the principles of auxiliary-vector filtering and exhibits enhanced performance in short data record scenarios. The coarse synchronization performance of combined synchronization/demodulation receivers under finite data record adaptation is also investigated. Analytic expressions are derived that approximate closely the probability of coarse synchronization error of the conventional correlator and the MVDR type combined synchronization/demodulation scheme and provide low cost highly accurate alternatives to the computationally demanding performance evaluation through simulations.

We consider the problem of robust detection of a spread-spectrum (SS) signal in the presence of unknown correlated SS interference and additive non-Gaussian noise. The proposed general SS receiver structure is comprised by a vector of adaptive chip-based nonlinearities followed by an adaptive linear tap-weight filter and combines the relative merits of both nonlinear and linear signal processing. The novel characteristics of our approach are as follows. First, the nonlinear receiver front-end adapts itself to the unknown prevailing noise environment providing robust performance for a wide range of underlying noise distributions. Second, the adaptive linear tap-weight filter that follows the nonlinearly processed chip samples results in a receiver that is proven to be effective in combating SS interference as well. To determine the receiver parameters, we propose, develop, and study three adaptive schemes under a joint mean-square error (MSE), or a joint bit-error-rate (BER), or a joint MSE-BER optimization criterion. As a side result, we derive the optimum decision fusion filter for receivers that utilize hard-limiting (sign) chip nonlinearities. Numerical and simulation results demonstrate the performance of the proposed schemes and offer comparisons with the conventional matched-filter (MF), the decorrelator, the conventional minimum-variance-distortionless-response (MVDR) filter, and the sign-majority vote receiver.

Receiver architectures in the form of a linear filter front-end followed by a hard-limiting decision maker are considered for DS-CDMA communication systems. Based on stochastic approximation concepts a recursive algorithm is developed for the adaptive optimization of the linear filter front-end in the minimum BER sense. The recursive form is decision driven and distribution free. For additive white Gaussian noise (AWGN) channels, theoretical analysis of the BER surface of linear filter receivers identifies the subset of the linear filter space where the optimal receiver lies and offers a formal proof of guaranteed global optimization with probability one for the two-user case. To the extent that the output of a linear DS-CDMA filter can be approximated by a Gaussian random variable, a minimum-mean-square-error optimized linear filter approximates the minimum BER solution. Numerical and simulation results indicate that for realistic AWGN DS-CDMA systems with reasonably low signature cross-correlations the linear minimum BER filter and the MMSE filter exhibit approximately the same performance. The linear minimum BER receiver is superior, however, when either the signature cross-correlation is high or the background noise is non-Gaussian.