Future generations of mobile networks call for concurrent sensing and communication functionalities in the same hardware and/or spectrum. Compared to communication, sensing services often suffer from limited coverage, due to the high path loss of the reflected signal and the increased infrastructure requirements. To provide a more uniform quality of service, distributed multiple input multiple output (D-MIMO) systems deploy a large number of distributed nodes and efficiently control them, making distributed integrated sensing and communications (ISAC) possible. In this paper, we investigate ISAC in D-MIMO through the lens of different design architectures and deployments, revealing both conflicts and synergies. In addition, simulation and demonstration results reveal both opportunities and challenges towards the implementation of ISAC in D-MIMO.
Secure aggregation (SecAgg) is a commonly-used privacy-enhancing mechanism in federated learning, affording the server access only to the aggregate of model updates while safeguarding the confidentiality of individual updates. Despite widespread claims regarding SecAgg's privacy-preserving capabilities, a formal analysis of its privacy is lacking, making such presumptions unjustified. In this paper, we delve into the privacy implications of SecAgg by treating it as a local differential privacy (LDP) mechanism for each local update. We design a simple attack wherein an adversarial server seeks to discern which update vector a client submitted, out of two possible ones, in a single training round of federated learning under SecAgg. By conducting privacy auditing, we assess the success probability of this attack and quantify the LDP guarantees provided by SecAgg. Our numerical results unveil that, contrary to prevailing claims, SecAgg offers weak privacy against membership inference attacks even in a single training round. Indeed, it is difficult to hide a local update by adding other independent local updates when the updates are of high dimension. Our findings underscore the imperative for additional privacy-enhancing mechanisms, such as noise injection, in federated learning.
We present the uplink and downlink of a time-division duplex distributed multiple-input multiple-output (D-MIMO) testbed, based on a 1-bit radio-over-fiber architecture, which is low-cost and scalable. The proposed architecture involves a central unit (CU) that is equipped with 1-bit digital-to-analog and analog-to-digital converters, operating at 10 GS/s. The CU is connected to multiple single-antenna remote radio heads (RRHs) via optical fibers, over which a binary RF waveform is transmitted. In the uplink, a binary RF waveform is generated at the RRHs by a comparator, whose inputs are the received RF signal and a suitably designed dither signal. In the downlink, a binary RF waveform is generated at the CU via bandpass sigma-delta modulation. Our measurement results show that low error-vector magnitude (EVM) can be achieved in both the uplink and the downlink, despite 1-bit sampling at the CU. Specifically, for point-to-point over-cable transmission between a single user equipment (UE) and a CU equipped with a single RRH, we report, for a 10 MBd signal using single-carrier 16QAM modulation, an EVM of 3.3% in the downlink, and of 4.5% in the uplink. We then consider a CU connected to 3 RRHs serving over the air 2 UEs, and show that, after over-the-air reciprocity calibration, a downlink zero-forcing precoder designed on the basis of uplink channel estimates at the CU, achieves an EVM of 6.4% and 10.9% at UE 1 and UE 2, respectively. Finally, we investigate the ability of the proposed architecture to support orthogonal frequency-division multiplexing (OFDM) waveforms, and its robustness against both in-band and out-of-band interference.
A fundamental question in theoretical machine learning is generalization. Over the past decades, the PAC-Bayesian approach has been established as a flexible framework to address the generalization capabilities of machine learning algorithms, and design new ones. Recently, it has garnered increased interest due to its potential applicability for a variety of learning algorithms, including deep neural networks. In parallel, an information-theoretic view of generalization has developed, wherein the relation between generalization and various information measures has been established. This framework is intimately connected to the PAC-Bayesian approach, and a number of results have been independently discovered in both strands. In this monograph, we highlight this strong connection and present a unified treatment of generalization. We present techniques and results that the two perspectives have in common, and discuss the approaches and interpretations that differ. In particular, we demonstrate how many proofs in the area share a modular structure, through which the underlying ideas can be intuited. We pay special attention to the conditional mutual information (CMI) framework; analytical studies of the information complexity of learning algorithms; and the application of the proposed methods to deep learning. This monograph is intended to provide a comprehensive introduction to information-theoretic generalization bounds and their connection to PAC-Bayes, serving as a foundation from which the most recent developments are accessible. It is aimed broadly towards researchers with an interest in generalization and theoretical machine learning.
Recent work has established that the conditional mutual information (CMI) framework of Steinke and Zakynthinou (2020) is expressive enough to capture generalization guarantees in terms of algorithmic stability, VC dimension, and related complexity measures for conventional learning (Harutyunyan et al., 2021, Haghifam et al., 2021). Hence, it provides a unified method for establishing generalization bounds. In meta learning, there has so far been a divide between information-theoretic results and results from classical learning theory. In this work, we take a first step toward bridging this divide. Specifically, we present novel generalization bounds for meta learning in terms of the evaluated CMI (e-CMI). To demonstrate the expressiveness of the e-CMI framework, we apply our bounds to a representation learning setting, with $n$ samples from $\hat n$ tasks parameterized by functions of the form $f_i \circ h$. Here, each $f_i \in \mathcal F$ is a task-specific function, and $h \in \mathcal H$ is the shared representation. For this setup, we show that the e-CMI framework yields a bound that scales as $\sqrt{ \mathcal C(\mathcal H)/(n\hat n) + \mathcal C(\mathcal F)/n} $, where $\mathcal C(\cdot)$ denotes a complexity measure of the hypothesis class. This scaling behavior coincides with the one reported in Tripuraneni et al. (2020) using Gaussian complexity.
We present a new family of information-theoretic generalization bounds, in which the training loss and the population loss are compared through a jointly convex function. This function is upper-bounded in terms of the disintegrated, samplewise, evaluated conditional mutual information (CMI), an information measure that depends on the losses incurred by the selected hypothesis, rather than on the hypothesis itself, as is common in probably approximately correct (PAC)-Bayesian results. We demonstrate the generality of this framework by recovering and extending previously known information-theoretic bounds. Furthermore, using the evaluated CMI, we derive a samplewise, average version of Seeger's PAC-Bayesian bound, where the convex function is the binary KL divergence. In some scenarios, this novel bound results in a tighter characterization of the population loss of deep neural networks than previous bounds. Finally, we derive high-probability versions of some of these average bounds. We demonstrate the unifying nature of the evaluated CMI bounds by using them to recover average and high-probability generalization bounds for multiclass classification with finite Natarajan dimension.
We present a general framework for the characterization of the packet error probability achievable in cell-free Massive multiple-input multiple output (MIMO) architectures deployed to support ultra-reliable low lantecy (URLLC) traffic. The framework is general and encompasses both centralized and distributed cell-free architectures, arbitrary fading channels and channel estimation algorithms at both network and user-equipment (UE) sides, as well as arbitrary combing and precoding schemes. The framework is used to perform numerical experiments that clearly show the superiority of cell-free architectures compared to cellular architectures in supporting URLLC traffic in uplink and downlink. Also, they provide the following novel insights into the optimal design of cell-free architectures for URLLC: i) minimum mean square error (MMSE) spatial processing must be used to achieve the URLLC targets; ii) for a given total number of antennas per coverage area, centralized cell-free solutions involving single-antenna access points (APs) offer the best performance in the uplink, thereby highlighting the importance of reducing the average distance between APs and UEs in the URLLC regime; iii) this observation applies also to the downlink, provided that the APs transmit precoded pilots to allow the UEs to estimate accurately the precoded channel.
We consider a fully digital massive multiple-input multiple-output architecture with low-resolution analog-to-digital/digital-to-analog converters (ADCs/DACs) at the base station (BS) and analyze the performance trade-off between the number of BS antennas, the resolution of the ADCs/DACs, and the bandwidth. Assuming a hardware power consumption constraint, we determine the relationship between these design parameters by using a realistic model for the power consumption of the ADCs/DACs and the radio frequency chains. Considering uplink pilot-aided channel estimation, we build on the Bussgang decomposition to derive tractable expressions for uplink and downlink ergodic achievable sum rates. Numerical results show that the ergodic performance is boosted when many BS antennas with very low resolution (i.e., 2 to 3 bits) are adopted in both the uplink and the downlink.
Low-resolution analog-to-digital converters (ADCs) simplify the design of millimeter-wave (mmWave) massive multi-user multiple-input multiple-output (MU-MIMO) basestations, but increase vulnerability to jamming attacks. As a remedy, we propose HERMIT (short for Hybrid jammER MITigation), a method that combines a hardware-friendly adaptive analog transform with a corresponding digital equalizer: The analog transform removes most of the jammer's energy prior to data conversion; the digital equalizer suppresses jammer residues while detecting the legitimate transmit data. We provide theoretical results that establish the optimal analog transform as a function of the user equipments' and the jammer's channels. Using simulations with mmWave channel models, we demonstrate the superiority of HERMIT compared both to purely digital jammer mitigation as well as to a recent hybrid method that mitigates jammer interference with a nonadaptive analog transform.
All-digital basestation (BS) architectures for millimeter-wave (mmWave) massive multi-user multiple-input multiple-output (MU-MIMO), which equip each radio-frequency chain with dedicated data converters, have advantages in spectral efficiency, flexibility, and baseband-processing simplicity over hybrid analog-digital solutions. For all-digital architectures to be competitive with hybrid solutions in terms of power consumption, novel signal-processing methods and baseband architectures are necessary. In this paper, we demonstrate that adapting the resolution of the analog-to-digital converters (ADCs) and spatial equalizer of an all-digital system to the communication scenario (e.g., the number of users, modulation scheme, and propagation conditions) enables orders-of-magnitude power savings for realistic mmWave channels. For example, for a 256-BS-antenna 16-user system supporting 1 GHz bandwidth, a traditional baseline architecture designed for a 64-user worst-case scenario would consume 23 W in 28 nm CMOS for the ADC array and the spatial equalizer, whereas a resolution-adaptive architecture is able to reduce the power consumption by 6.7x.