Elza Erkip

In frequency division duplexing systems, downlink massive multiple-input multiple-output (MIMO) precoding algorithms rely on accurate channel state information (CSI) feedback from users. This paper investigates the tradeoff between the CSI feedback overhead and the resulting user performance in terms of achievable sum rate. Our approach consists of determining the precoding directly from the user feedback. We employ a deep learning-based design for an end-to-end precoding-oriented feedback architecture, including learned pilots, user compressors for finite-rate feedback, and base station processing to determine precoding vectors. We propose a novel loss function that maximizes the sum of achievable rates while minimizing the CSI feedback overhead. We consider both single- and multi-cell multi-user MIMO systems, analyzing the impact of intra- and inter-cell interference on the CSI feedback strategy design, as well as robustness. Simulation results demonstrate that our approach outperforms previous precoding-oriented methods and offers greater efficiency than conventional methods that separate CSI compression and precoding.

Efficient and low-latency wireless connectivity between the base station (BS) and a sparse set of sporadically active devices from a massive number of devices is crucial for emerging massive machine-type communications (mMTC). This paper addresses the challenge of identifying active devices while meeting stringent access delay and reliability constraints in mMTC environments. A novel multi-stage active device identification framework is proposed where we aim to refine a partial estimate of the active device set using feedback and hypothesis testing across multiple stages eventually leading to an exact recovery of active devices after the final stage of processing. In our proposed approach, active devices independently transmit binary preambles during each stage, leveraging feedback signals from the BS, whereas the BS employs a non-coherent binary energy detection. The minimum user identification cost associated with our multi-stage non-coherent active device identification framework with feedback, in terms of the required number of channel-uses, is quantified using information-theoretic techniques in the asymptotic regime of total number of devices ℓ when the number of active devices k scales as k = Θ(1). Practical implementations of our multi-stage active device identification schemes, leveraging Belief Propagation (BP) techniques, are also presented and evaluated. Simulation results show that our multi-stage BP strategies exhibit superior performance over single-stage strategies, even when considering overhead costs associated with feedback and hypothesis testing.

The relay channel, consisting of a source-destination pair along with a relay, is a fundamental component of cooperative communications. While the capacity of a general relay channel remains unknown, various relaying strategies, including compress-and-forward (CF), have been proposed. In CF, the relay forwards a quantized version of its received signal to the destination. Given the correlated signals at the relay and destination, distributed compression techniques, such as Wyner-Ziv coding, can be harnessed to utilize the relay-to-destination link more efficiently. Leveraging recent advances in neural network-based distributed compression, we revisit the relay channel problem and integrate a learned task-aware Wyner-Ziv compressor into a primitive relay channel with a finite-capacity out-of-band relay-to-destination link. The resulting neural CF scheme demonstrates that our compressor recovers binning of the quantized indices at the relay, mimicking the optimal asymptotic CF strategy, although no structure exploiting the knowledge of source statistics was imposed into the design. The proposed neural CF, employing finite order modulation, operates closely to the rate achievable in a primitive relay channel with a Gaussian codebook. We showcase the advantages of exploiting the correlated destination signal for relay compression through various neural CF architectures that involve end-to-end training of the compressor and the demodulator components. Our learned task-oriented compressors provide the first proof-of-concept work toward interpretable and practical neural CF relaying schemes.

Database de-anonymization typically involves matching an anonymized database with correlated publicly available data. Existing research focuses either on practical aspects without requiring knowledge of the data distribution yet provides limited guarantees, or on theoretical aspects assuming known distributions. This paper aims to bridge these two approaches, offering theoretical guarantees for database de-anonymization under synchronization errors and obfuscation without prior knowledge of data distribution. Using a modified replica detection algorithm and a new seeded deletion detection algorithm, we establish sufficient conditions on the database growth rate for successful matching, demonstrating a double-logarithmic seed size relative to row size is sufficient for detecting deletions in the database. Importantly, our findings indicate that these sufficient de-anonymization conditions are tight and are the same as in the distribution-aware setting, avoiding asymptotic performance loss due to unknown distributions. Finally, we evaluate the performance of our proposed algorithms through simulations, confirming their effectiveness in more practical, non-asymptotic, scenarios.

Running deep visual analytics models for real-time applications is challenging for mobile devices. Offloading the computation to edge server can mitigate computation bottleneck at the mobile device, but may decrease the analytics performance due to the necessity of compressing the image data. We consider a “split computing” system to offload a part of the deep learning model's computation and introduce a novel learned feature compression approach with lightweight computation. We demonstrate the effectiveness of the split computing pipeline in performing computation offloading for the problems of object detection and image classification. Compared to compressing the raw images at the mobile, and running the analytics model on the decompressed images at the server, the proposed feature-compression approach can achieve significantly higher analytics performance at the same bit rate, while reducing the complexity at the mobile. We further propose a scalable feature compression approach, which facilitates adaptation to network bandwidth dynamics, while having comparable performance to the non-scalable approach.

We consider lossy compression of an information source when the decoder has lossless access to a correlated one. This setup, also known as the Wyner-Ziv problem, is a special case of distributed source coding. To this day, practical approaches for the Wyner-Ziv problem have neither been fully developed nor heavily investigated. We propose a data-driven method based on machine learning that leverages the universal function approximation capability of artificial neural networks. We find that our neural network-based compression scheme, based on variational vector quantization, recovers some principles of the optimum theoretical solution of the Wyner-Ziv setup, such as binning in the source space as well as optimal combination of the quantization index and side information, for exemplary sources. These behaviors emerge although no structure exploiting knowledge of the source distributions was imposed. Binning is a widely used tool in information theoretic proofs and methods, and to our knowledge, this is the first time it has been explicitly observed to emerge from data-driven learning.

The re- identification or de-anonymization of users from anonymized data through matching with publicly available correlated user data has raised privacy concerns, leading to the complementary measure of obfuscation in addition to anonymization. Recent research provides a fundamental understanding of the conditions under which privacy attacks, in the form of database matching, are successful in the presence of obfuscation. Motivated by synchronization errors stemming from the sampling of time-indexed databases, this paper presents a unified framework considering both obfuscation and synchronization errors and investigates the matching of databases under noisy entry repetitions. By investigating different structures for the repetition pattern, replica detection and seeded deletion detection algorithms are devised and sufficient and necessary conditions for successful matching are derived. Finally, the impacts of some variations of the underlying assumptions, such as the adversarial deletion model, seedless database matching, and zero-rate regime, on the results are discussed. Overall, our results provide insights into the privacy-preserving publication of anonymized and obfuscated time-indexed data as well as the closely related problem of the capacity of synchronization channels.

We consider lossy compression of an information source when the decoder has lossless access to a correlated one. This setup, also known as the Wyner-Ziv problem, is a special case of distributed source coding. To this day, practical approaches for the Wyner-Ziv problem have neither been fully developed nor heavily investigated. We propose a data-driven method based on machine learning that leverages the universal function approximation capability of artificial neural networks. We find that our neural network-based compression scheme, based on variational vector quantization, recovers some principles of the optimum theoretical solution of the Wyner-Ziv setup, such as binning in the source space as well as optimal combination of the quantization index and side information, for exemplary sources. These behaviors emerge although no structure exploiting knowledge of the source distributions was imposed. Binning is a widely used tool in information theoretic proofs and methods, and to our knowledge, this is the first time it has been explicitly observed to emerge from data-driven learning.

Massive Machine-Type Communications (mMTC) is a key service category in the current generation of wireless networks featuring an extremely high density of energy and resource-limited devices with sparse and sporadic activity patterns. In order to enable random access in such mMTC networks, base station needs to identify the active devices while operating within stringent access delay constraints. In this paper, an energy efficient active device identification protocol is proposed in which active devices transmit On-Off Keying (OOK) modulated preambles jointly and base station employs non-coherent energy detection avoiding channel estimation overheads. The minimum number of channel-uses required by the active user identification protocol is characterized in the asymptotic regime of total number of devices ℓ when the number of active devices k scales as k = Θ(1) along with an achievability scheme relying on the equivalence of activity detection to a group testing problem. Several practical schemes based on Belief Propagation (BP) and Combinatorial Orthogonal Matching Pursuit (COMP) are also proposed. Simulation results show that BP strategies outperform COMP significantly and can operate close to the theoretical achievability bounds. In a partial-recovery setting where few misdetections are allowed, BP continues to perform well.