Searcharxiv⌕ Search

arXiv subjects

Esa Ollila

Publications and source records attributed to Esa Ollila.

At least 19 recordsLinked to original sources

Performance Analysis of Quantization-Aware Neural Channel Estimation with Low-Precision Inference

Neural channel estimators can improve radio performance of 5G receivers, but real-time deployment in the radio access network (RAN) imposes strict latency and power budgets. Prior studies of quantized neural channel estimators assess this cost through mean-squared error or analytic operation counts. In this paper, we train a 1D convolutional neural network (CNN) channel estimator for the 5G NR physical uplink shared channel (PUSCH) with INT8 and FP8 quantization-aware training (QAT), and we evaluate its radio performance together with its latency, power, and energy on an NVIDIA H100 GPU. On TDL-A and TDL-C channels at user equipment (UE) speeds up to 35 m/s, both 8-bit models stay within 0.025 dB of FP32 at 10% block error rate (BLER). At batch size 128, INT8 reduces the latency by 16%, the incremental GPU power by 33%, and the energy per query by 26%, compared with 6%, 26%, and 21% for FP8. INT8 is therefore the preferred format.

eess.SP↗

Zeroth-Order Riemannian Optimization on Fixed-Rank Manifolds for LLM Fine-Tuning

Zeroth-order (ZO) optimization estimates descent directions from function values alone, which removes the activation storage required by backpropagation and makes it attractive for memory-efficient fine-tuning of large language models (LLMs). Existing low-rank ZO methods either parameterize the additive update through Euclidean factors with a non-unique representation, or restrict each finite-difference direction to be low-rank. In this work, we propose to constrain each additive update block to a fixed-rank matrix manifold while keeping the pretrained weights frozen, so the learned update remains rank-constrained throughout fine-tuning. On this geometry, we build a Riemannian ZO gradient estimator that projects a random matrix onto the tangent space, normalizes it to correct the scale mismatch induced by projection, and further propose Zeroth-order Riemannian Gradient Descent (ZO-RGD) and its smoothed variant, ZO-RGD-S. We show that the ambient finite difference matches its retracted counterpart, the normalized estimator recovers the Riemannian gradient in expectation, and ZO-RGD converges under trajectory-regularity conditions. Experiments on LLM fine-tuning tasks show that the proposed algorithms are competitive with strong forward-only baselines across model families and tasks, with gains in several settings.

cs.LG↗

Covariance Learning for Direction-of-Arrival Estimation with Sparse Linear Arrays

Covariance learning (CL) estimates source powers and sensor-noise variance by minimizing the Type-II (marginal) negative log-likelihood of the array measurements. We develop both grid-based and gridless CL direction-of-arrival (DOA) estimators applicable to arbitrary sparse linear array (SLA) geometries. The grid-based DOA estimator minimizes the Type-II objective on an angular grid via successive convex approximation (SCA) and estimates the noise variance by a scalar bisection search. The gridless DOA estimator then refines each grid-based DOA over the continuum using a Newton method. Simulations on a six-sensor nested array, including an underdetermined case with nine sources, show that the proposed DOA estimators outperform state-of-the-art methods, despite jointly estimating the sensor noise variance.

eess.SP↗

Efficient SCA Algorithms for Sparse Bayesian Learning

Type-II sparse Bayesian learning (SBL) for the multiple measurement vector model recovers a sparse signal power vector by minimizing the marginal negative log-likelihood via the expectation-maximization (EM) algorithm, referred to as M-SBL. We revisit the covariance learning (CL) problem and develop three parallel first-order algorithms within the successive convex approximation (SCA) framework. CL-SCA minimizes per-coordinate surrogates formed by partially linearizing the log-determinant term, while CL-FP uses a damped projected fixed-point iteration derived from stationarity conditions. Both methods have closed-form updates. We prove that the fixed points of both update maps are exactly the KKT points of the CL problem, that each update direction satisfies a quantitative descent inequality, that the damped iterates achieve sufficient decrease with an $\mathcal{O}(T^{-1/2})$ best-iterate stationarity rate, and we derive a backtracking rule that retains this rate without step-size tuning. If damping factors are bounded away from zero, the residual vanishes along the whole sequence and every limit point is a KKT point. Local analysis motivates CL-Hyb, which combines CL-SCA and CL-FP. Experiments on device activity detection in massive random access show that CL-Hyb-bt matches the best CL solvers' accuracy at a fraction of the runtime and is over an order of magnitude faster than EM.

eess.SP↗

Computationally Efficient Neural Receiver via Axial Self-Attention

Deep learning-based neural receivers offer promising physical-layer solutions for next-generation wireless systems. We propose an axial self-attention transformer neural receiver that achieves state-of-the-art Block Error Rate (BLER) performance with significantly improved computational efficiency during inference and large-scale training. By factorizing attention operations along temporal and spectral axes, the proposed architecture reduces computational complexity from $O((TF)^2)$ to $O(T^2F+TF^2)$, yielding substantially fewer floating-point operations and attention matrix multiplications per transformer block. Experimental validation under 3GPP Clustered Delay Line (CDL) channels demonstrates consistent performance gains across varying mobility scenarios. Under non-line-of-sight conditions, our proposed axial neural receiver outperforms global self-attention and convolutional neural receiver baselines at 10% BLER and 1% BLER respectively, with reduced computational complexity.

eess.SP↗

A Majorization-Minimization Framework for Activity Detection in Mixed Near-and-Far-Field Random Access

Grant-free random access in massive machine-type communications requires detecting a small active set from length-L uplink pilots received at an M-antenna base station. Classical covariance-based detectors are largely built on the far-field (FF) model, where the M antenna-domain observations are treated as independent snapshots, and inference reduces to an L $\times$ L covariance description. This model becomes inadequate in mixed near-field (NF) and FF access, where NF devices induce device-specific structured spatial covariances and the aggregate observation no longer fits the FF snapshot structure. In this paper, we develop a unified covariance-aware Rician framework for activity detection that treats FF and NF devices by a single likelihood model. Within this framework, we propose a majorization-minimization projected gradient descent (MM-PGD) detector for the resulting relaxed likelihood. For scalable exact likelihood evaluation, we further derive a Kronecker-Woodbury implementation that exploits the mixed NF/FF covariance structure, and avoids factorizing the full LM $\times$ LM covariance matrix. Numerical results show that MM-PGD is on par with the strongest coordinate-wise baseline in the all-FF limit, while its advantage becomes more pronounced as the fraction of NF users increases under the tested regimes.

eess.SP↗

Floating-Point Microformat Quantization and Pruning for Efficient MU-MIMO Neural Receivers

Neural receivers outperform conventional 5G NR processing chains, but their compute and memory demands hinder real-time deployment. For a standard-compliant multi-user MIMO neural receiver, the 4-bit number format, not merely the bit width, determines whether compression preserves the gain over classical receivers. We apply weight and activation quantization-aware training (QAT) and, separately, 50% magnitude pruning, comparing INT8/INT4 and FP8 (E4M3)/FP4 (E2M1) weights with INT8 post-ReLU activations. Trained on 3GPP UMi channels and evaluated on TDL-B and TDL-C, 8-bit weight-activation models remain within 0.05 dB of FP32 at 10% and 1% block error rate (BLER). At 4 bits, uniform INT4 loses 3.3-3.7 dB and falls below LS-LMMSE, whereas FP4 more than halves this loss (1.3-1.4 dB) and still outperforms it by about 0.5 dB, even after pruning. FP4's denser near-zero grid matches the trained weight distribution, and FP4 avoids the residual-path over-pruning seen with INT4. An analytic cost model projects 66x fewer bit-operations and 8.8x less weight storage for pruned 4-bit-weight inference.

eess.SP↗

Covariance-Aware MM-PGD for Mixed Near-/Far-Field Activity Detection

Grant-free activity detection with mixed near-field (NF) and far-field (FF) devices is an important problem that can be addressed via covariance-based detectors. The difficulty is that NF users induce device-specific structured spatial covariances, whereas FF users are well approximated by isotropic covariances. Under a unified Rician model, we first formulate activity detection as a relaxed maximum-likelihood problem in the full LM-dimensional vectorized observation space. We then develop a covariance-aware majorization-minimization projected gradient descent (MM-PGD) detector. It updates the full activity vector jointly and avoids the per-coordinate high-rank subproblems that arise in the NF regime. Numerical results over SNR, antenna-count, and NF-ratio sweeps show that MM-PGD achieves up to 20x lower miss-detection probability than the strongest coordinate-wise baseline. The advantage is most pronounced at high NF ratios, while in the all-FF case MM-PGD performs on par with the strongest coordinate-wise baseline.

eess.SP↗

Joint Activity Detection and Channel Estimation for Massive Random Access Using SBL and SCA

In massive machine-type communication (mMTC) applications, a key challenge is joint device activity detection and channel estimation (JADCE) under grant-free random access, as a massive number of devices with sporadic traffic seek to connect to the base station. We address JADCE for massive random access using a covariance learning-based sparse Bayesian learning (SBL) approach. Specifically, we first use the successive convex approximation (SCA) framework to partially linearize the scaled negative log-likelihood function (LLF) of the data, then minimize it to estimate the sparse vector of devices' signal powers. After identifying active devices from these power estimates, empirical Bayesian estimation is used to obtain channel estimates. Simulation results demonstrate the efficiency and performance superiority of the proposed CL-SCA method compared to other existing methods.

eess.SP↗

Nuisance parameters and elliptically symmetric distributions: a geometric approach to parametric and semiparametric efficiency

Elliptically symmetric distributions are a classic example of a semiparametric model where the location vector and the scatter matrix (or a parameterization of them) are the two finite-dimensional parameters of interest, while the density generator represents an \textit{infinite-dimensional nuisance} term. This basic representation of the elliptic model can be made more accurate, rich, and flexible by considering additional \textit{finite-dimensional nuisance} parameters. Our aim is therefore to investigate the deep and counter-intuitive links between statistical efficiency in estimating the parameters of interest in the presence of both finite and infinite-dimensional nuisance parameters. Previous seminal works have addressed this problem by leveraging a general result: if the statistical model has a specific group invariance, then the projection operator onto the semiparametric nuisance tangent space can be asymptotically expressed as a conditional expectation with respect to the maximal invariant sub-$σ$ algebra. In this article, we show that, for the statistical model of elliptical distributions, the projection operator can be explicitly computed without relying on the above-mentioned asymptotic approximation. This allows us to obtain original results also for the case in which the location vector and the scatter matrix are parameterized by a finite-dimensional vector that can be partitioned in two sub-vectors: one containing the parameters of interest and the other containing the nuisance parameters. As an example, we illustrate how the obtained results can be applied to the well-known \virg{low-rank} parameterization. Furthermore, while the theoretical analysis will be developed for Real Elliptically Symmetric (RES) distributions, we show how to extend our results to the case of Circular and Non-Circular Complex Elliptically Symmetric (C-CES and NC-CES) distributions.

math.ST↗

Finer Parameter Steps for Low-Rank PEFT: A Controlled Study with CP Tensor Adapters

Low-rank adapters are usually compared by sweeping a small set of ranks, but the rank also fixes the resolution of the parameter budget. For a $2048{\times}2048$ OPT attention projection, increasing LoRA by one rank stores $4096$ trainable scalars, leaving large gaps between feasible low-budget adapter sizes. This paper asks whether a tensorized adapter with finer capacity increments changes the observed accuracy--budget trade-off. We instantiate this question with fixed-component canonical polyadic (CP) tensor adapters. Under a $32{\times}64{\times}32{\times}64$ tensorization, one normalized CP component stores $193$ trainable scalars per projection, about $21$ times smaller than one LoRA rank step. We compare CP adapters and LoRA on OPT-1.3B across SST-2, RTE, and BoolQ under matched target modules, training protocol, data caps, and seed schedules. CP trains stably and fills the gaps between LoRA ranks, but the effect is task-dependent: SST-2 reaches an early low-budget plateau, BoolQ benefits from additional CP components before saturating slightly below LoRA, and RTE remains LoRA-favored. Finer parameter steps are therefore useful for diagnosing PEFT budget sensitivity, but they do not by themselves guarantee a better accuracy--budget curve.

cs.LG↗

Frequency Range 3 for ISAC in 6G: Potentials and Challenges

Spanning 7-24 GHz, frequency range 3 (FR3), is a key enabler for next-generation wireless networks by bridging the coverage of sub-6 GHz and the capacity of millimeter-wave bands. Its unique propagation characteristics, such as extended near-field regions and spatially nonstationary fading, enable new transmission strategies. This article explores the potential of FR3 for integrated sensing and communication (ISAC), which unifies wireless communication and environmental sensing. We show that FR3's bandwidth and multiple-input multiple-output (MIMO) capabilities enable high-resolution sensing, multi-target tracking, and fast data transmission. We emphasize the importance of ultra-massive MIMO with extremely large aperture arrays (ELAAs) and the need for unified near-field and far-field channel models to support efficient ISAC. Finally, we outline challenges and future research directions for ELAA-based ISAC in 6G FR3.

cs.IT↗

A Tutorial on AI-Empowered Integrated Sensing and Communications

Integrating sensing and communication (ISAC) can help overcome the challenges of limited spectrum and expensive hardware, leading to improved energy and cost efficiency. While full cooperation between sensing and communication can result in significant performance gains, achieving optimal performance requires efficient designs of unified waveforms and beamformers for joint sensing and communication. Sophisticated statistical signal processing and multi-objective optimization techniques are necessary to balance the competing design requirements of joint sensing and communication tasks. As model-based approaches can be suboptimal or too complex, deep learning offers a powerful data-driven alternative, especially when optimal algorithms are unknown or impractical for real-time use. Unified waveform and beamformer design problems for ISAC fall into this category, where fundamental design trade-offs exist between sensing and communication performance metrics, and the underlying models may be inadequate or incomplete. This tutorial paper explores the application of artificial intelligence (AI) to enhance efficiency or reduce complexity in ISAC designs. We emphasize the integration benefits through AI-driven ISAC designs, prioritizing the development of unified waveforms, constellations, and beamforming strategies for both sensing and communication. To illustrate the practical potential of AI-driven ISAC, we present three case studies on waveform, beamforming, and constellation design, demonstrating how unsupervised learning and neural network-based optimization can effectively balance performance, complexity, and implementation constraints.

cs.IT↗

Efficient Quantization-Aware Neural Receivers: Beyond Post-Training Quantization

As wireless communication systems advance toward Sixth Generation (6G) Radio Access Networks (RAN), Deep Learning (DL)-based neural receivers are emerging as transformative solutions for Physical Layer (PHY) processing, delivering superior Block Error Rate (BLER) performance compared to traditional model-based approaches. Practical deployment on resource-constrained hardware, however, requires efficient quantization to reduce latency, energy, and memory without sacrificing reliability. In this paper, we extend Post-Training Quantization (PTQ) by focusing on Quantization-Aware Training (QAT), which incorporates low-precision simulation during training for robustness at ultra-low bitwidths. In particular, we develop a QAT methodology for a neural receiver architecture and benchmark it against a PTQ approach across diverse 3GPP Clustered Delay Line (CDL) channel profiles under both Line-of-Sight (LoS) and Non-LoS (NLoS) conditions, with user velocities up to 40 m/s. Results show that 4-bit and 8-bit QAT models achieve BLERs comparable to FP32 models at a 10% target BLER. Moreover, QAT models succeed in NLoS scenarios where PTQ models fail to reach the 10% BLER target, while also yielding an 8x compression. These results with respect to full-precision demonstrate that QAT is a key enabler of low-complexity and latency-constrained inference at the PHY layer, facilitating real-time processing in 6G edge devices.

eess.SP↗

Anisotropic Tensor Deconvolution of Hyperspectral Images

Hyperspectral image (HSI) deconvolution is a challenging ill-posed inverse problem, made difficult by the data's high dimensionality.We propose a parameter-parsimonious framework based on a low-rank Canonical Polyadic Decomposition (CPD) of the entire latent HSI $\mathbf{\mathcal{X}} \in \mathbb{R}^{P\times Q \times N}$.This approach recasts the problem from recovering a large-scale image with $PQN$ variables to estimating the CPD factors with $(P+Q+N)R$ variables.This model also enables a structure-aware, anisotropic Total Variation (TV) regularization applied only to the spatial factors, preserving the smooth spectral signatures.An efficient algorithm based on the Proximal Alternating Linearized Minimization (PALM) framework is developed to solve the resulting non-convex optimization problem.Experiments confirm the model's efficiency, showing a numerous parameter reduction of over two orders of magnitude and a compelling trade-off between model compactness and reconstruction accuracy.

eess.IV↗

Robust Activity Detection for Massive Random Access

Massive machine-type communications (mMTC) are fundamental to the Internet of Things (IoT) framework in future wireless networks, involving the connection of a vast number of devices with sporadic transmission patterns. Traditional device activity detection (AD) methods are typically developed for Gaussian noise, but their performance may deteriorate when these conditions are not met, particularly in the presence of heavy-tailed impulsive noise. In this paper, we propose robust statistical techniques for AD that do not rely on the Gaussian assumption and replace the Gaussian loss function with robust loss functions that can effectively mitigate the impact of heavy-tailed noise and outliers. First, we prove that the coordinate-wise (conditional) objective function is geodesically convex and derive a fixed-point (FP) algorithm for minimizing it, along with convergence guarantees. Building on the FP algorithm, we propose two robust algorithms for solving the full (unconditional) objective function: a coordinate-wise optimization algorithm (RCWO) and a greedy covariance learning-based matching pursuit algorithm (RCL-MP). Numerical experiments demonstrate that the proposed methods significantly outperform existing algorithms in scenarios with non-Gaussian noise, achieving higher detection accuracy and robustness.

eess.SP↗

Generalized Nonnegative Structured Kruskal Tensor Regression

This paper introduces Generalized Nonnegative Structured Kruskal Tensor Regression (NS-KTR), a novel tensor regression framework that enhances interpretability and performance through mode-specific hybrid regularization and nonnegativity constraints. Our approach accommodates both linear and logistic regression formulations for diverse response variables while addressing the structural heterogeneity inherent in multidimensional tensor data. We integrate fused LASSO, total variation, and ridge regularizers, each tailored to specific tensor modes, and develop an efficient alternating direction method of multipliers (ADMM) based algorithm for parameter estimation. Comprehensive experiments on synthetic signals and real hyperspectral datasets demonstrate that NS-KTR consistently outperforms conventional tensor regression methods. The framework's ability to preserve distinct structural characteristics across tensor dimensions while ensuring physical interpretability makes it especially suitable for applications in signal processing and hyperspectral image analysis.

eess.SP↗

Efficient Deep Neural Receiver with Post-Training Quantization

Deep learning has recently garnered significant interest in wireless communications due to its superior performance compared to traditional model-based algorithms. Deep convolutional neural networks (CNNs) have demonstrated notable improvements in block error rate (BLER) under various channel models and mobility scenarios. However, the high computational complexity and resource demands of deep CNNs pose challenges for deployment in resource-constrained edge systems. The 3rd Generation Partnership Project (3GPP) Release 20 highlights the pivotal role of artificial intelligence (AI) integration in enabling advanced radio-access networks for 6G systems. The hard real-time processing demands of 5G and 6G require efficient techniques such as post-training quantization (PTQ), quantization-aware training (QAT), pruning, and hybrid approaches to meet latency requirements. In this paper, we focus on PTQ to reduce model complexity by lowering the bit-width of weights, thereby enhancing computational efficiency. Our analysis employs symmetric uniform quantization, applying both per-tensor and per-channel PTQ to a neural receiver achieving performance comparable to full-precision models. Specifically, 8-bit per-channel quantization maintains BLER performance with minimal degradation, while 4-bit quantization shows great promise but requires further optimization to achieve target BLER levels. These results highlight the potential of ultra-low bitwidth PTQ for efficient neural receiver deployment in 6G systems.

eess.SP↗