Multiagent distributed clustering scheme is proposed herein to process data which are collected by dispersed sensors that are not under centralized control. Two methods based on distributed dual averaging (DDA) algorithm are proposed, which are able to incorporate network structure and do not require exchange of centroid estimates, which makes it appealing for security conscious applications. The first method provides the framework for distributed clustering using the DDA algorithm with predefined regularization parameter. The second method, called Adaptive DDA (ADDA), relaxes the condition concerninga prioriknowledge about the centroids, assumed in the first method, without losing clustering performance. This is achieved by properly regularizing the problem where a data-driven approach is used to determine the regularization parameter. The proposed methods are further extended via the proposed Bin method to scenario where processing agents store unbalanced amount of data with non-IID class distribution. Experiments are conducted on both real-life and synthetic data. Numerical results show the efficacy of the proposed approaches compared to state-of-art centralized algorithm and other distributed approaches.

This work considers unsupervised learning tasks being implemented within the federated learning framework to satisfy stringent requirements for low-latency and privacy of the emerging applications. The proposed algorithm is based on Dual Averaging (DA), where the gradients of each agent are aggregated at a central node. While having its advantages in terms of distributed computation, the accuracy of federated learning training reduces significantly when the data is nonuniformly distributed across devices. Therefore, this work proposes two weight computation algorithms, with one using a fixed size bin and the other with self-organizing maps (SOM) that solves the underlying dimensionality problem inherent in the first method. Simulation results are also provided to show that the proposed algorithms' performance is comparable to the scenario in which all data is uploaded and processed in the centralized cloud.

This work considers a design of network in which multiple transmission points (TPs) cooperatively serve users by jointly precoding shared data. Considered problem formulation jointly designs the beamformers and performs TP-UE link selection, which aims in improving overall system rate. Proposed distributed Augmented ADMM algorithm features parallelization among TPs, which has practical importance for computational load distribution and reducing signaling overhead in backhaul. This approach is different from others in literature because it solves a design problem that involves a coupling constraint which no existing algorithm is able to solve. Simulation results are also provided to show that the proposed distributed algorithm performance outperforms previously proposed distributed consensus optimization method and is comparable to its centralized counterpart.

Performance of object classification using 3D automotive radar relies on accurate data association and multitarget tracking' which are greatly affected by data bias and proximity of objects to each other. A regularized fuzzy c-means (RFCM) algorithm is proposed herein to resolve the data association uncertainty problem that has shown to outperform the conventional FCM algorithm. The proposed method exploits results from the companion tracker to increase performance robustness. Simulation results using simulated and field data have proven the efficacy of the proposed method.

Performance of MIMO precoder for ultradense HetNets can be hindered by a lack of accurate channel state information. A novel two stage precoding scheme, known as Orthogonalized Sparsity Enhanced Mismatch Model-LQ (OSEMM-LQ) precoding, is proposed herein. Simulation results have shown it can achieve higher receive signal power and BER performance compared to the recently proposed SEMM and conventional norm ball mismatch model (NBMM) precoding schemes.

Performance of precoder-based spatial intercell interference cancelation in heterogeneous networks is often hampered due to lack of accurate channel state information. Performance can be augmented by modifying the design of the precoder to incorporate the channel estimate mismatch by using deterministic and probabilistic mismatch models. Previously proposed models either have been deemed too conservative (deterministic) or are prone to error due to inaccuracy in the probability distribution function and corresponding parameters (stochastic). A new deterministic mismatch model is proposed herein in an attempt to alleviate these problems. Different from all previously proposed deterministic models, the proposed model, called sparsity enhanced mismatch model (SEMM), exploits the inherent sparse characteristics of MIMO interference channels. The SEMM has two variants, i.e., SEMM (angular) and SEMM (eigenmode). The SEMM incorporates a basis expansion model to bring forth the inherent sparsity, which exists in MIMO interference channels. In the context of precoder design for heterogeneous network, it is analytically shown, and by simulation, the proposed mismatch models enable the aggressor-transmitter (A-Tx) to allocate more transmission power to the sparse elements of the interfering link so that performance in the communicating link is enhanced compared with conventional norm ball mismatch model.

Transmission over doubly-selective fading (DSF) interference channel often relies on the use of robust precoder due to a lack of accurate channel state information, with performance often depending on the conservativeness of the mismatch model. Previously proposed mismatch models either have been deemed too conservative (deterministic models) or are prone to error due to inaccuracy in the probability density function (pdf) and corresponding parameters (stochastic models). A deterministic mismatch model called Sparsity Enhanced Mismatch Model - Reverse discrete prolate spheroidal sequence, or SEMMR, is proposed herein in an attempt to alleviate this problem. Different from all previously deterministic models, the proposed model exploits the inherent sparse characteristics of DSF interference channels which lead to a two-stage robust transceiver design that outperforms precoding only strategy incorporating conventional norm ball mismatch model (NBMM). The inherent sparsity in the channel is brought forth by modeling the channel using a basis expansion model (BEM) where discrete prolate spheroidal sequence (DPSS) is used as a basis. Analytical and simulation results are provided to validate the performance gains of the SEMMR transceiver over the NBMM precoder.

A robust superimposed training sequence design is proposed for spatially correlated MIMO channel estimation. The proposed scheme does not require accurate knowledge about the spatial correlation matrix and it is shown to outperform robust correlated MIMO channel estimators such as relaxed MMSE (RMMSE) and least-squares-RMMSE (LSRMMSE). Since the training sequence is overlaid into the data stream, spectral efficiency of the system is higher than those that use time-multiplex pilots. A solution for the sequence can be obtained easily by using an iterative algorithm which is guaranteed to converge as long as the training sequence matrix is initialized to have full rank.

In this paper, we proposed a dynamic ordered subcarrier selection algorithm (DOSSA) for OFDM based video transmission system. The proposed scheme is shown to achieve lower bit error rate (BER) than the previously proposed OSSA by first selecting a fraction of the subcarriers with highest channel gain. The content information is then exploited in order to extend the OSSA to achieve unequal error protection (UEP) for packets of different importance. Simulation results show that system that utilizes the proposed scheme can achieve higher PSNR, especially at low SNR, compared to those that use the equal error protection (EEP) OSSA.

Herein proposed is a new frequency domain equalizer (FEQ) to suppress channel induced interference such as ICI and ISI, co-channel interference, and overlaid systems interference. Unlike earlier schemes, this proposed algorithm requires no temporal oversampling nor the use of more than one receive antenna. All the above is achieved by exploiting null subcarriers (a.k.a. virtual/unused/unmodulated subcarriers) inherent in standard multicarrier systems, and by a generalized sidelobe cancellation (GSC) like scheme. This proposed method can offer superior bit error rate over earlier methods as well as simpler computations over another GSC-like scheme.

A higher-order statistics based minimal transmit redundancy space-time FIR precoder-blind equalizer is proposed. For most block based transmission systems such as OFDM, a long guard period, which is longer than or equal to the channel order, is required to avoid inter-block interference. This guard period is a type of redundancy which consumes valuable bandwidth. The blind equalizer that we have proposed can blindly equalize the received precoded signal without the long guard period requirement. Simulation results have shown that the BER performance gain increases as more received data are used to design the equalizer which allows accurate estimation of the cumulant matrices. The proposed algorithm was also shown to be robust toward channel order overestimation.

Time-domain equalizers (TEQs) have been applied extensively to shorten the channel impulse response, thus enhancing the transmission efficiency of multitone and multicarrier systems with cyclic prefix. Recent developments on TEQs have mainly been focused on minimizing the additive noise power, minimizing the ISI power or maximizing the total throughput. This paper takes a different approach by minimizing the detection error or bit error rate (BER) using the Chernoff bound, under a fixed ISI power of the equalized channel, simulation results showed that this approach achieves better performance in terms of the bit error rate when compared to the aforementioned approaches and is robust to channel noise and channel length. Both numerical and analytical solutions have been derived.

In statistical speech recognition, misclassification often occurs when there is a mismatch between the incoming signal and the acoustics model inside the recognizer. In order to combat this problem, techniques such as Cepstral Mean Subtraction, Vocal Tract Normalization, adaptation and pronunciation model can be used. In this paper, we proposed a new approach based on transformation technique where the output distribution function in the HMM model, a Gaussian probability density function, could be transformed to match the estimated distribution of the incoming signal by using a memoryless invertible nonlinearity function. Since the new density still has a Gaussian form, the function could be completely characterized by using the Expectation Maximization (EM) algorithm. 1.