One common approach for question answering over speech data is to first transcribe speech using automatic speech recognition (ASR) and then employ text-based retrieval-augmented generation (RAG) on the transcriptions. While this cascaded pipeline has proven effective in many practical settings, ASR errors can propagate to the retrieval and generation steps. To overcome this limitation, we introduce SpeechRAG, a novel framework designed for open-question answering over spoken data. Our proposed approach fine-tunes a pre-trained speech encoder into a speech adapter fed into a frozen large language model (LLM)–based retrieval model. By aligning the embedding spaces of text and speech, our speech retriever directly retrieves audio passages from text-based queries, leveraging the retrieval capacity of the frozen text retriever. Our retrieval experiments on spoken question answering datasets show that direct speech retrieval does not degrade over the text-based baseline, and outperforms the cascaded systems using ASR. For generation, we use a speech language model (SLM) as a generator, conditioned on audio passages rather than transcripts. Without fine-tuning of the SLM, this approach outperforms cascaded text-based models when there is high WER in the transcripts.

Machine-to-machine (M2M) communication, also referred to as Internet of Things (IoT), is a global network of devices such as sensors, actuators, and smart appliances which collect information, and can be controlled and managed in real time over the Internet. Due to their universal coverage, cellular networks and the Internet together offer the most promising foundation for the implementation of M2M communication. With the worldwide deployment of the fourth generation (4G) of cellular networks, the long-term evolution (LTE) and LTE-advanced standards have defined several quality-of-service classes to accommodate the M2M traffic. However, cellular networks are mainly optimized for human-to-human (H2H) communication. The characteristics of M2M traffic are different from the human-generated traffic and consequently create sever problems in both radio access and the core networks (CNs). This survey on M2M communication in LTE/LTE-A explores the issues, solutions, and the remaining challenges to enable and improve M2M communication over cellular networks. We first present an overview of the LTE networks and discuss the issues related to M2M applications on LTE. We investigate the traffic issues of M2M communications and the challenges they impose on both access channel and traffic channel of a radio access network and the congestion problems they create in the CN. We present a comprehensive review of the solutions for these problems which have been proposed in the literature in recent years and discuss the advantages and disadvantages of each method. The remaining challenges are also discussed in detail.

Big Data received from sources such as social media, in-stream monitoring systems, networks, and markets is often mined for discovering patterns, detecting anomalies, and making decisions or predictions. In distributed learning and real-time processing of Big Data, ensemble-based systems in which a fusion center (FC) is used to combine the local decisions of several classifiers, have shown to be superior to single expert systems. However, optimal design of the FC requires knowledge of the accuracy of the individual classifiers which, in many cases, is not available. Moreover, in many applications supervised training of the FC is not feasible since the true labels of the data set are not available. In this paper, we propose an unsupervised joint estimation-detection scheme to estimate the accuracies of the local classifiers as functions of data context and to fuse the local decisions of the classifiers. Numerical results show the dramatic improvement of the proposed method as compared with the state of the art approaches.

We introduce a new tendency measure for a probability mass function (pmf) referred to as “tenor,” and defined in terms of the phase of the first non-zero frequency of the discrete Fourier transform of the pmf. This statistic is in the vicinity of the region of highest probability of the pmf. Unlike mean, tenor is robust against outliers, and unlike mode and median, tenor can be evaluated using only arithmetic operations of addition and multiplication, without the need for comparison operations. We propose a distributed algorithm for computation of tenor in a graph and prove that for large networks represented by Erdos-Renyi graphs, [1] and by Watts-Strogatz graphs (small-world graphs), [2] the distributed algorithm converges. Numerical examples including the distributed computation of the majority vote are presented to demonstrate the operation of the algorithm.

In this paper we consider the problem of cooperative spectrum sensing in cognitive radio networks (CRN) in the presence of misbehaving nodes. We propose a novel approach based on the iterative expectation maximization (EM) algorithm to detect the presence of the primary users, to classify the cognitive radios, and to compute their detection and false alarm probabilities. In contrast to previous work we assume that the FC has no prior information about the radios in the network except that the honest radios are in majority. As shown in the paper this is required for any algorithm to uniquely identify the CRs. Another distinguishing feature is that our approach can classify the radios into more than just two classes of honest and malicious CRs. This applies in cases where the honest CRs have different detection and false alarm probabilities, which may arise when they employ different spectrum sensing techniques or encounter dissimilar channel and noise conditions. Another case is when the CRN includes more than one type of misbehaving CRs. Our numerical results show significant improvements over the widely popular reputation-based classifier (RBC). In particular, with only a few decisions from the CRs, the proposed algorithm can quickly and efficiently classify the CRs whereas the RBC method fails even for networks with a large number of CRs. In all of our numerical results the EM algorithm converged in five or fewer iterations resulting in fast convergence of the proposed method. This makes the proposed method a good candidate for implementation in CRNs. The numerical results are also compared with the Cramer-Rao lower bound and show a close match. Simulation results are also presented to demonstrate the efficacy of the proposed algorithm in the presence of correlated observations among the radios.

In distributed sensing, the statistical model of the data collected by the sensor elements is often unavailable. In addition, these statistics may vary among the sensors and over time, for instance due to: 1) hardware variations; 2) the sensors' geographical locations; 3) different noise statistics; 4) diverse channel conditions between the sensor elements and the fusion center (FC); and 5) the presence of misbehaving sensors sending false data to the FC. In this paper, we consider the problem of centralized binary hypothesis testing in a wireless sensor network consisting of multiple classes of sensors, where the sensors are classified according to the probability density function (PDF) of their received data (at the FC) under each hypothesis. The sensor nodes transmit their observed data to the FC, which must classify the nodes and detect the state of nature. To optimally fuse the data, the FC must also estimate the PDFs of the sensors' observations. We develop a method based on the expectation maximization (EM) algorithm to estimate the PDFs for each sensor class, to classify the sensors, and to detect the underlying hypotheses. The estimation of PDFs is nonparametric in that no prior model is assumed. Simulation results using fewer than three iterations of the EM algorithm demonstrate the efficacy of the proposed method.

The growing demand for wireless services has led to the introduction of new paradigms in spectrum sharing such as the unlicensed ISMand U-NII bands and dynamic or opportunistic spectrum access through cognitive radios. Coexistence of users in these technologies leads to increases in co-channel interference (CCI) which needs to be appropriately mitigated. CCI is often modeled as a white Gaussian noise process and assumed to simply reduce the signal-to-noise (plus interference) ratio. In this paper we consider the effect of CCI by a careful examination of the samples at the output of the matched filter receiver. We show that the timing offset between the interference and the desired signals may result in the correlation of errors in adjacent symbols. We evaluate the bit error rate (BER) resulting from CCI as well as the distribution of the total number of errors in a packet.

Spectrum sensing is a key function of cognitive radios and is used to determine whether a primary user is present in the channel or not. Many approaches have been proposed when both primary user and secondary user employ a single antenna. Recently several techniques have also been proposed assuming that the the secondary user employs multiple antennas. In this paper, we formulate and solve the generalized likelihood ratio test (GLRT) for spectrum sensing when both primary user transmitter and the secondary user receiver are equipped with multiple antennas. We do not assume any prior information about the channel statistics or the primary user's signal structure. Two cases are considered when the secondary user is aware of the energy of the noise and when it is not. The final test statistics derived from GLRT are based on the eigenvalues of the sample covariance matrix. Through analysis we exhibit the role of the eigenvalues in characterizing the signal+noise and noise subspaces in the received data. Simulation results are presented in terms of the receiver operating characteristics and detection probabilities for several cases of interest.

We propose an iterative algorithm based on expectation maximization (EM) to jointly estimate the system parameters and decode the data in a MIMO system using unitary space-time block codes. It is assumed that the receiver is unaware of the channel coefficients and their distribution, the average energy of the received signal, the prior probability of each signal in the constellation, and the noise power. The algorithm works for arbitrary modulation schemes and any channel model including Rayleigh and Rician fading. The complexity of the proposed receiver is computed and it is shown to be significantly lower than previously published methods, as it does not require any matrix inversion or trellis search. The performance of the proposed receiver is evaluated by simulations in terms of symbol error rate (SER) vs. signal-to-noise ratio (SNR) and it is shown that with only a few iterations of the algorithm it achieves a performance close to that of the receiver which knows all the parameters.

Wireless sensor networks are prone to node misbehavior arising from tampering by an adversary (Byzantine attack), or due to other factors such as node failure resulting from hardware or software degradation. In this paper, we consider the problem of decentralized detection in wireless sensor networks in the presence of one or more classes of misbehaving nodes. Binary hypothesis testing is considered where the honest nodes transmit their binary decisions to the fusion center (FC), while the misbehaving nodes transmit fictitious messages. The goal of the FC is to identify the misbehaving nodes and to detect the state of nature. We identify each class of nodes with an operating point (false alarm and detection probabilities) on the receiver operating characteristic (ROC) curve. Maximum likelihood estimation of the nodes' operating points is then formulated and solved using the expectation maximization (EM) algorithm with the nodes' identities as latent variables. The solution from the EM algorithm is then used to classify the nodes and to solve the decentralized hypothesis testing problem. Numerical results compared with those from the reputation-based schemes show a significant improvement in both classification of the nodes and hypothesis testing results. We also discuss an inherent ambiguity in the node classification problem which can be resolved if the honest nodes are in majority.