Vermogensallocatie en (de)activering van antennes en symbolen voor energiereductie in sub-6 GHz basisstations

The increase in the energy consumption of fifth-generation (5G) base stations (BSs) is a growing concern in terms of environmental and economic sustainability. A reason for this increase is that wireless networks are often designed to meet the peak requirements, which have been constantly rising in each new network generation. In this thesis, we adopt an alternative approach and aim at operating the BSs such that their energy use is minimized for a fixed target quality-of-service (QoS). Specifically, we consider multi-antenna BSs using orthogonal frequency-division multiplexing (OFDM) over multiple symbols, and analyze the allocation of transmit power at the BS, number of utilized BS antennas, and number of utilized transmit symbols by the BS. The optimized resource allocation adaptively changes with the target QoS. We model the energy consumption of the BS as due to the power amplifiers (PAs), analog and digital processing modules, and supply and cooling units. The consumption of each component is averaged over different states in a time frame, which are working mode and data transmission or reference signaling, micro-sleep, idle, and sleep modes. Moreover, we model the consumption of subcomponents (e.g., the digital-to-analog converters (DACs) as part of the analog processing). Consumed power is linearly proportional to consumed energy in our framework. The power and communications models are applied to prospective BSs operating in the upper mid-band (7-24 GHz) with thousands of antenna elements and large bandwidths. The results show that hybrid analog-digital beamforming achieves the best performance in bits per Joule. The focus is then placed on the allocation of transmit power and (de)activation of antennas in a BS operating in a sub-6 GHz. The problem is first formulated as the design of digital precoders that minimize the PAs consumption under QoS constraints. We derived closed-form expressions for the precoders that depend on fixed-point equations on the transmit powers of the antennas. In single-carrier systems, the proposed solution utilizes only a few antennas, while the baseline approach utilizes all the antennas to maximize the array gain. When many subcarriers are utilized, no antennas are deactivated from the PA consumption perspective. However, we show that under the considered assumptions on the channel, we can incorporate the consumption of the whole BS and minimize it subject to QoS and transmit power constraints. The proposed solution activates a subset of the antennas (dependently on the number of users) with uniform allocated power. The number of activated antennas is retrieved in closed form and depends on the long-term channel statistics. We then extended the precoding design problem for PA-power minimization to distributed cooperating BSs, in the case where many subcarriers are utilized. We find that the optimized strategy generally activates only a subset of the BSs. In these activated BSs, all the antennas are activated. To reduce the complexity of the initial solution, we propose an allocation based on random matrix theory (RMT), and show that it provides a reliable approximation of the optimal allocation. The RMT-based solution brings benefits in terms of consumption and complexity of the fronthaul (FH) links, used to connect the BSs with the central processing unit. The last two chapters incorporate the (de)activation of transmit (OFDM) symbols. By considering a system with many subcarriers and a fixed precoder, we considered the problem of BS-consumption minimization subject to QoS and transmit power constraints. The transmit power is expressed in terms of the number of activated antennas and time slots using the QoS constraints and the assumptions on the channel. We then proved the convexity of the resulting two-dimensional objective function (relaxed as continuous) and a sufficient condition for the convexity of the optimization domain. The final proposed solution is retrieved by using a low-complexity algorithm and computing a rounding on the continuous variables. In the evaluations, the power model's parameters are adopted from a model in the literature validated on operator measurements of sub-6 GHz BSs, and the large-scale fading parameters are derived from data reported by the same sub-6 GHz BSs. The energy performance is evaluated for different BS configurations, varying the target QoS, and considering enabled or disabled power-saving modes in the PA and analog processing. We find that minimizing the number of activated antennas ("rush-to-mute") is often quasi-optimal in terms of energy consumption, and serving the maximum number of users is beneficial in terms of energy efficiency.