This paper investigates the resource optimization problem for a multi-cell massive multiple-input multiple-output (MIMO) network in which each base station (BS) is equipped with a large number of antennas and each base station (BS) adapts the number of antennas to the daily load profile (DLP). This paper takes into consideration user location distribution (ULD) variation and evaluates its impact on the energy efficiency of load adaptive massive MIMO system. ULD variation is modeled by dividing the cell into two coverage areas with different user densities: boundary focused (BF) and center focused (CF) ULD. All cells are assumed identical in terms of BS configurations, cell loading, and ULD variation and each BS is modeled as an M/G/m/m state dependent queue that can serve a maximum number of users at the peak load. Together with energy efficiency (EE) we analyzed deployment and spectrum efficiency in our adaptive massive MIMO system by evaluating the impact of cell size, available bandwidth, output power level of the BS, and maximum output power of the power amplifier (PA) at different cell loading. We also analyzed average energy consumption on an hourly basis per BS for the model proposed for data traffic in Europe and also the model proposed for business, residential, street, and highway areas.

In the current mobile communication systems, due to the technical limits such as occupied space and implementation complexity in a multi-antenna system, the number of antennas configured on the transmitter/receiver (TX/RX) end is limited. Massive MIMO has been recently proposed as a potential technique that significantly enables the increase the network throughput and energy efficiency [

The amount of mobile data traffic has increased dramatically over the years, which is mainly driven by the massive demand of data-hungry devices such as smart phones, tablets and broadband wireless applications such as multimedia, three-dimensional (3D) video games, e-Health, Car2X communications [_{2}) contribution of the telecommunications sector to the global CO_{2} emission has increased rapidly over the last decade, where mobile operators are among the top energy consumers. Thus, besides spectral efficiency, energy efficiency is crucial, so wireless researchers and engineers need to shift their focus to energy-efficiency oriented design in order to reduce the operation cost for mobile operators and minimize the environmental impact of the wireless domain.

In general, efficient allocation of radio resources (e.g., power, frequency, time, and antennas) plays a crucial role in improving wireless network performance. Taking into consideration different points of view, there are several energy efficiency (EE) metrics to evaluate wireless networks. In particular, authors at [

Resource allocation for EE maximization has been an active area of research in recent years. Design of an optimal resource allocation algorithm can be challenging especially if the underlying resource allocation problems are non-convex [

Furthermore, several works [

In this paper, the aim is to investigate the energy-efficient resource allocation algorithm for a single-cell and multi-cell multi-user massive MIMO system. Our paper examines the effect of two ULD models, namely, boundary focused (BF) and center focused (CF) and is different from [

In order to maximize EE throughout 24-hour operation of the network, it is necessary to optimize M following the daily load profile (DLP) and ULD model. In order to map the user distribution to the DLP, we model each BS as a state dependent M/G/m/m queue [

The above mentioned state dependency is due to the fact that the user rate depends on the number of users that the BS can serve simultaneously. The peak load of the DLP is when BS serves a maximum number of users. Network is dimensioned to handle peak data traffic and it becomes most energy efficient when serving maximum load.

In particular, we divide a cell into two areas and for each area, we consider the user distribution and the optimum number of antennas that maximize EE under the given network load and interference condition for any number of users a cell serves. Generally, a user located at the cell boundary will receive the weakest signals from its BS but the strong interference from other BSs. For this reason, our goal is to determine and compare system performance in these two cell areas.

The specific contributions of the paper are as follows:

We derive two systems for analyzing energy efficiency: adaptive and fixed antenna system, each one contains two ULD models boundary focused (BF) and center focused (CF).

We investigate the total energy efficiency of our considered and fixed antenna system taking into account the number of BS antennas, user load distribution, and cell loading.

We propose a dynamic resource allocation strategy to adapt the number of antennas based on tracking variations of user location distribution and cell loading.

The following section presents description of created models: system, ULD, traffic, and power consumption model. The third section provides a problem formulation and description of the created optimization algorithm. In the fourth part additional scenarios of the basic model are formed to ascertain which system parameter values of the analyzed model result in the best performance during the appliance of the massive MIMO. The results and the discussion of the analyses are also shown in the fourth part. At the end of the paper there is a conclusion.

In this paper, we consider a downlink of a multi-cell massive MIMO system that contains hexagonal cells, each having its own BS and we apply classic wrap-around to avoid edge effects. Each BS transmits a constant output power _{c} (s) coherence time which remain static within a time-frequency coherence block of

We assume that all BSs and UEs are synchronized and operate on time-division duplex (TDD) protocol and that the BS obtains perfect Channel state information (CSI) from the uplink pilots, which is a reasonable assumption for low-mobility scenarios. All BSs employ zero-forcing (ZF) precoding to cancel out intra-cell interference by using the beamforming technique, which is a signal processing technique that cancels out intra-cell interference by beamforming and adapts the power allocation to guarantee the same rate to each user. Numerical results in [

The energy-efficiency (in bits/Joule) of a massive MIMO system is defined in [

In the expression for distance-dependent rate at

Power consumed in a BS depends on the number of active antennas and the number of users served simultaneously. It consists of a fixed part (e.g., control signal, backhaul, direct current-direct current (DC--DC) conversion) and a dynamic part, i.e., radiated transmit power and circuit power which depends on the number of active antennas and number of users served simultaneously.

Based on the models and the practical numbers in [

ULD model formed in this paper defines two coverage areas which are divided with radius

Every area is assigned a set of weighting factors

Network loads vary throughout the day, so in order to capture the daily traffic variation and maximize EE throughout the day, we model each BS as an M/G/m/m state dependent queue [

The steady state probabilities for the random number of users

We allow at most 2% blocking at 100% load, i.e., the probability of serving the maximum allowed number of users,

In this paper, we want to maximize the EE of the system described in Section 2. Our EE problem is divided into smaller ones, and the solution to this problem was found through several steps. In order to achieve better EE in the system, it is necessary to adapt the number of active antennas for any user states. This is done iteratively. First, each BS finds the most energy efficient number of active antennas taking into account the interference from the surrounding BSs and a given number of users. This optimal number of antennas

The constraint

We use the daily load profile proposed for data traffic in Europe however, the mean value can be calculated for another period or part of the day for which we need to have an appropriate load profile for data traffic.

In order to find an optimum number of antennas, we first determine the global optimum point which represents the maximum number of users and antennas within, which we search for the optimal value of the system. Then we formulate the problem to find the optimum cell size, bandwidth, output power level of the BS, and maximum output power of PA. The influence of cell size change on pilot contamination and thus on system performance is analyzed in [

In order to analyze the performance of our established model with dynamic change of antenna number, a reference model was created which uses a previously determined global maximum number of antennas

Since changes in the cell are constantly occurring, it is necessary to adapt radio resources to these changes and establish a dynamic allocation of resources that result in high energy efficiency. Authors in [

Our algorithm aims to find the optimal number of antennas of BS, taking into account the current state in the cell. But the first step of the algorithm is to find the global optimal number of antennas and users

Similar algorithm was proposed in [

The described steps of the proposed strategy are shown in Algorithm 1, and were performed only for one considered cell due to the assumed configuration symmetry in all cells.

The cellular network in our simulation consists of 19 regular hexagonal cells. BS from each cell transmits a constant output power

Cell size is a very important factor in the design of massive MIMO networks that can significantly affect system performance and the amount of capital investment which includes infrastructure costs such as base station equipment, backhaul transmission equipment, site installation, and radio network controller equipment. In this scenario, we analyze the impact of cell size on system performance, especially on EE in the two observed coverage areas: center and boundary. We formed three sub-scenarios for three different cell sizes: sub-scenario with a cell radius of 500, 1000, and 2000 m. Users close to the cell boundaries suffer from strong interference, therefore it is important for these users to determine the optimal cell size.

Our results show that EE and Average UR are significantly higher in the CF area than in BF for each considered cell size. But, for all considered scenarios, compared to a system with a fixed number of antennas, both BF and CF users achieve better results when adaptively selecting the number of antennas. Results for EE for all cell size scenarios are shown in

Our analysis confirms that EE increases with the cell size shrinking, but reducing the cell size will also decrease the array gain. Therefore, the achievable SE decreases when the cell size is too small. The transmit power decreases with the cell size shrinking because the users will be closer to the BS. Therefore, transmit power is much lower than the circuit power, so EE in the network with a small cell size is dominated by the circuit power. Our results are consistent with findings from [

We found that for a cell radius of {500, 1000, 2000} m, the corresponding

Cell size (m) | Cell load (%) | BF | CF | ||||
---|---|---|---|---|---|---|---|

Number of active users in cell | |||||||

30 | 60 | 90 | 30 | 60 | 90 | ||

500 | 10 | 92 | 147 | 198 | 65 | 111 | 156 |

50 | 126 | 187 | 213 | 74 | 121 | 166 | |

100 | 144 | 210 | 213 | 83 | 132 | 178 | |

1000 | 10 | 87 | 150 | 211 | 66 | 120 | 171 |

50 | 95 | 158 | 217 | 68 | 121 | 173 | |

100 | 104 | 166 | 222 | 71 | 124 | 175 | |

2000 | 10 | 110 | 185 | 250 | 90 | 157 | 220 |

50 | 111 | 186 | 250 | 90 | 157 | 220 | |

100 | 113 | 187 | 250 | 90 | 158 | 221 |

We have also calculated weighted average optimal number of antennas

EE gain of our proposed model with adaptive antenna selection has excellent results for both the CF and BF areas and decreases with increasing cell load, but increases with the increase of the cell size. The corresponding EE gain for cell radius of 500 m, and cell load of {10, 50, 100}% for BF is {478.35, 165.44, 92.20}% and for CF is {551.92, 325.70, 201.48}% and is shown on

Also in [

It can be concluded that

Since bandwidth is a limited resource, 5G will require the allocation of additional spectrum for mobile broadband and flexible spectrum management capabilities. A set of new frequency bands will be used (licensed, unlicensed, and shared) to supplement the existing wireless frequencies, enabling larger bandwidth operation and substantially improved capacity.

The use of large blocks of spectrum in higher frequency bands and heterogeneous carrier aggregation (particularly above 6 GHz and up to 100 GHz) in addition to unlicensed spectrum have made wider system bandwidths (up to 3.2 GHz) feasible, leading to higher peak data rates and network capacity [

This section analyzes the impact of the bandwidth values on the performance of the considered system model. In this analysis, the bandwidth values vary from 20 to 100 MHz in increments of 20 MHz. With the increase in bandwidth, there is an increase in EE and Average UR for CF and BF ULD. In all considered bandwidth scenarios, CF ULD achieved better results than BF ULD, however the difference between the results is not drastic, but with increasing bandwidth and network load, it increases in favor of CF ULD. In all considered scenarios, the proposed adaptive system also achieved better results compared to the fixed one. With the increase in bandwidth, the results of the adaptive system are significantly better.

Despite these improvements, with increasing bandwidth

This section analyzes the impact of the different output power levels of the BS (

Values of 10, 20, 30, and 40 W were considered to analyze the impact of the output power level of the BS on EE.

It is very important to analyze how the increase in network load affects the activation of the required number of antennas that ensures a certain output power level of the BS since the increase in the number of active antennas negatively affects the EE. The findings of authors in [

Based on the obtained results, it can be concluded that the best EE for the case of CF ULD is twice as high as the best achievement of EE in BF ULD. The analyzed sub-scenarios for the BF ULD achieve approximate EE values and there is no drastic increase in EE for a particular sub-scenario when the network load increases.

To analyze the impact of the maximum output power of the PA (

For the

Users in the central area for the 1 dB sub-scenario have a fixed activation of 100 antennas regardless of the network load value because for

For the case of ULD BF, the lowest antenna activation is achieved for

Compared to BF ULD, CF ULD achieved better results for Average UR and EE in all sub-scenarios with different

Based on the results, it can be concluded that as the maximum output power of the PA and the output power level of the BS increases, EE decreases for both ULDs.

In addition to the analyzed impact of system parameters on EE for our formed adaptive system model, another important measure of energy consumption and savings that can be achieved during the day by applying the adaptive number of active antennas is average energy consumption per BS expressed in units of kWh:

In this analysis for daily load variation, we use the model proposed for data traffic in Europe [

Based on the previous results, the analyzes in this section were performed for an environment with a cell radius of 500 m, a bandwidth of 20 MHz, the output power level of BS is 20 W and the maximum output power of the PA is 6 dB. Also, this analysis was carried out considering the results of our established two areas of coverage (center and boundary) and two new scenarios with the different assignment of weighting factors, S1 and S2.

Hours | 01–07 | 08–10 | 11–18 | 19–21 | 22–24 |
---|---|---|---|---|---|

During daylight hours (8–21 h) S1 has a higher concentration of users in the central area, while during the night hours (22–7 h) higher weighting factors were assigned to users at the edges of the cell. Scenario S2 has the opposite situation with the assignment of weighting factors. Results for CF and BF ULD in all previous scenarios showed that BF ULD performed worse, which is also the case in scenarios S1 during night hours and S2 during daylight hours for all data traffic models, and it is necessary to activate more antennas. However, the system with adaptive antenna activation in all scenarios can turn off a significant number of antennas. Scenario S1 during daylight hours achieves a maximum shutdown of 77 antennas for the model proposed for data traffic in Europe, while Scenario S2 during night hours has a maximum shutdown of 88 antennas in the case of Business data traffic. Results in [

Compared to the fixed system, the adaptive system achieves significantly lower consumption. In the S1 scenario, a saving of 64% is possible in the Residential model of data traffic, while the maximum saving of average energy consumption in scenario S2 is achieved for the Business data traffic model and is equal to 42%.

In this study, we analyzed the energy efficiency of a load adaptive massive MIMO system with two different ULD models. Our results show that the optimal number of antennas depends primarily on ULD model and secondarily on cell loading. We also determined the impact of cell size, available bandwidth, output power lever of the BS, and dimensioning of a power amplifier (PA) at different cell loading on EE. Our developed resource allocation scheme jointly optimizes the number of BS antennas, user load distribution, and cell loading. BF ULD achieved worse results in the analysis of the impact of system parameters on the EE in the fixed and adaptive model of the system and for different network loads. Strong interference from other BSs and weakest signals from its BS are the reasons for poor results in BF ULD, which is why special attention should be paid to users at the edges of the cell when designing the network and establishing algorithms to solve interference. With increasing bandwidth and cell size, the maximum number of simultaneously served users is reduced, which is why the choice of the optimal value for these system parameters depends on the trade-off between the number of simultaneously served users and the level of desired performance for those users. The adaptive system for CF ULD achieves the best EE gains and for cell radius of 500 m, and cell load of {10, 50, 100} their values are {551.92, 325.70, 201.48}%.

During dimensioning the maximum output power of the PA and the output power level of the BS it was concluded that increases in these parameters lead to a decrease in EE. An increase in network load affects the activation of more antennas to ensure a certain output power level of the BS and that number is significantly higher in BF ULD, even in the best sub-scenario (

Based on the comparison between the performance of fixed and adaptive system model, it was concluded that better results of EE are achieved in the adaptive system for all analyzed scenarios because this proposed system can turn off a significant number of antennas and achieve significantly lower energy consumption. Within the 24 h operation, a savings potential of 64% is achieved in a residential model of data traffic, while saving in other models is in the range of 20–64%.