RT Dissertation/Thesis
T1 Energy-efficient digital and hybrid precoding algorithms for interference-limited cellular networks
A1 Urquiza Villalonga, David Alejandro
AB The fifth-generation (5G) of cellular communications demands high values of data rates, energyefficiency (EE), reliability, and multi-connectivity. In this context, ultra-dense networks(UDNs) are envisioned as a promising solution to meet the requirements of 5G and beyond5G (B5G) communications. UDNs are characterized by the deployment of a large numberof small-cells (SCs) in the coverage area of the traditional macrocell. Each SC is composedof a low-power small base station (SBS) that provides short-range communication to the userequipment (UE). Network densification increases the spectral efficiency (SE), EE, spectrumallocation, and coverage. However, their performance is limited by the increase of inter-userinterference caused by near SBSs that share the same time and frequency resources. Severalemerging technologies have been proposed to enable the development of UDNs. The use ofmultiple-input multiple-output (MIMO) systems combined with precoding techniques has beenproposed to increase the SE and EE. Interference alignment (IA) arises as a precoding strategyto manage interference while increasing the achievable degrees of freedom (DoF). However,the IA algorithms available in the literature have several limitations. In this thesis, novel digitaland hybrid IA algorithms have been proposed to solve the main open problems of the existingsolutions and hence to further improve the performance of UDNs.First, a two-tier network deployment is studied. Based on the cognitive radio (CR) paradigm,it is assumed that the SBS applies spectrum sensing techniques to detect spectrum holes inthe frequency bands of the macrocell. The unoccupied bands are used by the SBS to transmitinformation to its corresponding UE. The more challenging aspect of the proposed model isthe assumption that the SBS is equipped with an energy harvesting (EH) device that collectsenergy from the environment. Unlike previous strategies that are based on a constant sensingpower value, novel power allocation policies are proposed in this thesis to maximize detectionperformance under energy constraints. Assuming that energy consumption only depends onthe test statistic to detect the presence of signals, a closed-form expression is derived to relateenergy consumption with the number of processed samples and the technology-dependentparameters of the processing unit. Since the obtained closed-form equation is concave withrespect to the processing power, dynamic power allocation policies are proposed to maximizethe probability of detection in offline and online scenarios based on convex optimization theory, dynamic programming (DP), and heuristic solutions with lower complexity (Constant Powerand Greedy policies). The performance of proposed policies is validated by simulations forcommon detection techniques such as matched filter (MF), quadrature matched filter (QMF),and energy detector (ED). The proposed CR-EH policies enable the continuous operation ofenergy self-sufficient communications nodes guaranteeing also a high level of SE.Next chapters are addressed to the interference cancellation on a dense deployment of SCs.Clustered-IA has been widely studied to manage the interference in these dense scenarios.Nonetheless, the setups considered in previous works relied on oversimplified channel modelsand/or enforced single-stream transmission. In this thesis, a novel cluster design based on sizerestricted𝑘-means algorithm is proposed to divide the SCs into different clusters consideringboth path loss and shadowing effects. Therefore, a more realistic solution than those availablein the current literature is provided. Unlike previous works, this clustering method can alsocater to spatial multiplexing scenarios. Also, several design parameters such as the numberof transmit antennas, multiplexed data streams, and deployed SBSs are analyzed in order toidentify trade-offs between performance and complexity. The relationship between density ofnetwork elements per area unit and performance is also investigated. Moreover, it is shown thatthe SE degradation due to the inter-cluster interference in UDNs points to the need of designingan interference management algorithm that accounts for both, intra-cluster and inter-clusterinterference.Reported IA techniques neglect the nonlinear distortion (NLD) induced by power amplifiers(PAs). Thus, their performance is severely degraded in highly power-efficient transmissions operatingclose to the saturation point. Distortion-aware precoding techniques have been studiedto reduce NLD in single base station scenarios either single-user or multi-user. Nevertheless, itsextension to UDNs with multiple SBSs and UEs is not straightforward. In this thesis, a novelIA algorithm, named NLD-IA, is proposed to alternatively design the precoding and combiningvectors to cancel both interference and nonlinearities, so that signal-to-interference-plus-noiseratio (SINR) is maximized in UDNs. Precoding vectors are obtained via a non-convex optimizationproblem that models the NLD correlation. An analytical solution based on the Newtonmethod over the complex field is developed to solve this problem with low computationalcomplexity. Simulation results show that the proposed NLD-IA significantly reduces interferenceand NLD. Thus, it significantly outperforms previous IA algorithms for a wide range ofsignal-to-noise ratio (SNR), network densification, and saturation level. Another problem of network densification is that spectral resources are being saturated. Anattractive solution is to extend the communication services to the millimeter-wave (mmWave)frequencies that provide a huge available spectrum to support broadband applications. Nevertheless,the implementation of fully digital (FD) precoding strategies becomes prohibitivein massive MIMO (mMIMO) systems working at mmWave due to their hardware complexity.Therefore, hybrid digital-analog beamforming (HB) is considered to reduce the numberof radio frequency (RF) chains. When the analog stage is implemented with low-resolutionphase shifters (PSs), the SE is severely affected. In this thesis, a novel HB method to achievea near-optimal FD performance for mMIMO systems with the minimum required number ofRF chains is developed. The reduction of RF chains arises as a key procedure to obtain an EEdesign. The proposed HB scheme is based on a quantization-aware matrix composition (QMC)to systematically compensate for the residual quantization errors. Furthermore, a quantizationstrategy, called sum Euclidean quantization (SEQ), is introduced to improve the accuracy ofthe analog stage. Numerical results reveal that, unlike previously reported HB methods, a nearoptimalSE can be achieved with our proposal while minimizing the number of RF chains andthus, the hardware cost and power consumption.On the other hand, although IA algorithms have been widely studied following an FD approach,there are only a few works on HB-IA. Therefore, a new HB-IA is also proposed in this thesis forUDNs based on an iterative partial construction (IPC) procedure with the novelty of awarenessof PSs quantization. The IPC algorithm relies on algebraic properties to iteratively compute theanalog and digital matrices, via a partial submatrix composition, such that the Euclidean distanceto the FD-IA solution is reduced. An additional digital block, based on an IA algorithm,is also applied to cancel the residual interference. Numerical results reveal that the proposedHB IPC-IA significantly outperforms reported works achieving a near FD-IA performance.Since MIMO precoding is a key enabling technology for 5G/B5G cellular networks, the 3rdGeneration Partnership Project (3GPP) has standardized precoding techniques for single-userand multi-user MIMO systems to achieve a balanced trade-off between performance, complexity,and signal overhead. The implementation of these precoding strategies is covered in thisthesis providing a comprehensive guide. The performance of the 5G New Radio (NR) precoderis compared with theoretical precoding techniques such as singular value decomposition(SVD) and block-diagonalization (BD) to quantify the margin of improvement of the standardizedmethods. Several configurations of antenna arrays, number of antenna ports, and parallel data streams are considered for simulations. Moreover, the effect of channel estimation errorson the system performance is analyzed. For a realistic framework, the SE values are obtainedfor a practical deployment based on a clustered delay line (CDL) channel model. These resultsprovide valuable insights for system designers about the implementation and performance ofthe 5G-NR precoding matrices.Finally, a practical implementation of IA algorithms on a hardware platform using universalsoftware radio peripherals (USRPs) is addressed. A synchronization stage, based on theSchmidl and Cox method, is implemented. In contrast to previous works, a hardware testbed,able to model heterogeneous SINR networks, is proposed. The role of both closed and openloop for channel estimation is evaluated. Then, the impact of channel state information (CSI)updating on the SE and bit error rate (BER) is also analyzed. All the results are based on realmeasurements providing a clear understanding of the benefits and limitations of IA techniquesin interference-limited cellular networks.
YR 2024
FD 2024-01-23
LK https://hdl.handle.net/10016/43924
UL https://hdl.handle.net/10016/43924
LA eng
NO Mención Internacional en el título de doctor
DS e-Archivo
RD 17 jul. 2024