Radio Access Network (RAN) disaggregation is emerging as a key trend in beyond 5G, as it offers new opportunities for more flexible deployments and intelligent network management. A relevant problem in disaggregated RAN is the functional split selection, which dynamically decides which baseband (BB) functions of a base station are kept close to the radio units and which ones are centralized. In this context, this paper firstly presents an architectural framework for supporting this concept relying on the O-RAN architecture. Then, the paper analyzes how the functional split can be optimized to adapt to the different load conditions while minimizing energy costs.
In the ever-evolving field of artificial neural networks and learning systems, complex-valued neural networks (CVNNs) have become a cornerstone, achieving exceptional performance in image processing and telecommunications. More precisely, in digital communication systems, CVNNs have been delivering significant results in tasks like equalization, channel estimation, beamforming, and decoding. Among the CVNN architectures, the complex-valued radial basis function neural network (C-RBF) stands out, especially when operating in noisy environments such as 5G multiple-input multiple-output (MIMO) systems. In such a context, this paper extends the classical shallow C-RBF to deep architectures, increasing its flexibility for a wider range of applications. Also, based on the parameter selection of the phase transmittance radial basis function (PT-RBF) neural network, we propose an initialization scheme for the deep C-RBF. Via rigorous simulations conforming to 3GPP TS 38 standards for digital communications, our method not only outperforms conventional initialization strategies like random, K-means, and constellation-based methods but it also seems to be the only approach to achieve successful convergence for deep C-RBF architectures. These findings pave the way to more robust and efficient neural network deployments in complex-valued digital communication systems.
The large number of individually controlled antennas in large-scale MIMO systems makes the complexity of uplink receive combining a serious challenge. This is why simple methods, such as maximum ratio combining, are often considered, even if lower spectral efficiency is attained. In this paper, we investigate the performance of a reduced-complexity combining scheme that exploits second-order statistics of the channel. Compared to maximum ratio combining, substantial improvements are achieved, especially when the number of active terminals increases. This is attained with limited computational cost and remarkable robustness to the imperfect channel statistics knowledge.
The low-resolution aware linear minimum mean-squared error (LRA-LMMSE) channel estimator for low-resolution MIMO receivers was shown to achieve lower mean-squared error (MSE) with the use of a comparator network, which is composed of several simple comparators with binary outputs. In this study, we improve the analysis for the MSE of a comparator network aided MIMO system employing 1-bit quantization by deriving a bound for the error when using an approximated version for the auto-correlation of the quantized signal. By simulations, we compare the obtained analytical expression with the state-of-the-art approximation, showing that the proposed analytical expression is closer to the actual MSE obtained numerically. We also simulate independent comparator networks, which appear to yield a lower bound on the MSE for specific configurations.
In this paper, constrained optimization methods with a sparsity-aware Newton-type direction update are applied to adaptive filtering. Different versions of the proposed algorithms, which include a sparsity-aware RLS and a CG algorithm, present lower computational complexity than traditional algorithms based on LMS-Newton, since the proposed algorithms require the use of a smaller Hessian estimate. The proposed algorithms are tested in acoustic echo cancellation and time-varying channel identification. Some of them present a computational reduction of 25% of that of the traditional LMS-Newton in the best case scenario. Most of the proposed algorithms present faster convergence than the benchmark algorithms in several scenarios.
