Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Resource Allocation and Impact Analysis of the System Parameters on Performance in Multi-Cell Massive MIMO Networks

Resource Allocation and Impact Analysis of the System Parameters on Performance in Multi-Cell... B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 Submitted: August 29, 2022 Original scientific paper Accepted: November 5, 2022 RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS 1,2 1 Samira Mujkić , Suad Kasapović Abstract: This paper investigates the energy-efficient resource allocation algorithm for a massive multiple input multiple output (MIMO) system, in which each base station adapts the number of antennas to the daily load profile. Our paper examines the effect of two user location distribution (ULD) models, on the energy-efficiency (EE) of load adaptive masive MIMO system. We propose a resource allocation strategy to adapt the number of antennas based on tracking variations of ULD and cell loading maximizing the EE. We also evaluate impact of cell size, available bandwidth and output power level of the BS on EE at different cell loading. Keywords: energy efficiency, massive MIMO, traffic load, user location distribution INTRODUCTION selects the optimal set of users from the much larger pool of potential users, for a given set of beamforming weights, th The new technology of the 5 generation (5G) standard in polynomial time. Furthermore, several works [3] the that will most significantly increase energy efficiency is a authors analyse how the three main design parameters; technology called massive MIMO. Massive MIMO will also namely the number K of active user equipment (UEs), the increase network bandwidth and thus meet the current M number of antennas at the base station (BS), and the demand for higher data rates. transmit power must be chosen to improve EE. With excessive power consumption in wireless A low-complexity resource allocation scheme has been communications networks, beside spectral efficiency (SE), recently investigated in the literature. In particular, authors EE has become another significant metric for evaluating in [4] developed an algorithm that jointly optimizes the the performances of wireless communications systems. number of antennas, user selection and power allocation The energy consumption of a base station (BS) with no in a multi-user massive MIMO system, while the work load is at least half of the maximum energy it consumes in [5] optimizes user scheduling, power allocation and at the peak. Therefore, it is very important that 5G beamforming in MIMO networks implementing user- networks are able to adapt their power consumption with centric clustering. temporal changes of load [1]. In order to improve the energy efficiency of the 5G base station it is necessary In [6] the authors investigated the impact of power to develop joint optimization schemes and algorithms to amplifiers (PA) dimensioning on the energy efficiency of save computational and transmission power in the main load adaptive massive MIMO system. A common algorithm unit of the base station and radio frequency (RF) chains for antenna selection and resource block (RB) allocation together. Management of radio resources requires the was created in [7] together with power optimization establishment of strategies and algorithms that will use algorithm. Resource allocation method presented in [8] limited resources in the most efficient way. Due to the optimally select data and pilot powers along with the constant changes in the cells, it is necessary to establish training duration and maximize the spectral efficiency for a dynamic allocation of resources where the radio the three deferent receivers. The paper [9] also analysed resources will follow the changes and be adaptable to the energy efficiency of a system with a large number of them. Resource allocation established in this way should antennas where the transmission antennas were selected result in high energy efficiency. and the authors proposed two algorithms for the selection of transmitting antennas based on the application of In [2] the authors developed a joint beamforming and user sequential and binary search algorithms. scheduling algorithm based on fractional programming and the Hungarian algorithm. The Hungarian algorithm In [10] the authors show that during low traffic demand, e.g., late night, it is optimal to turn off a fraction of the Faculty of Electrical Engineering, University of Tuzla, Bosnia and Herzegovina Correspondence email: samira.mujkic@untz.ba © 2022 Author(s). This is an open access article licensed under the Creative Commons Attribution License 4.0. (http://creativecommons.org/licenses/by/4.0/). 1 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS antennas while minimizing total power consumption if the The number of users served by each BS is denoted by K , power consumption in the transceiver circuits is taken into while the number of used antennas is denoted by M . In account along with the RF amplifiers. network with massive MIMO the ratio of these two units must be M >K . Users have devices with one antenna. i i This paper investigates the energy-efficient resource User devices with multiple antennas are not considered allocation algorithm for a massive MIMO system, in which due to the computational complexity. each base station has a large number of antennas and it adapts the number of antennas to the daily load profile. Two The architecture of analysed massive MIMO system is user location distribution (ULD) models, central focused shown in Figure 1. (CF) ULD and boundary focused (BF) ULD, are defined in this paper. The formed ULD model defines two coverage areas with different user densities. ULD analysed in this paper is similar to models in [11] but in our paper we have also analysed the influence of certain system parameters on EE for different ULD model at different cell load. In particular, we aim at bringing new insights on how the number M of antennas at the BS, the number K of active UEs, and other system parameters should be chosen in order to achieve maximal EE. It was assumed that cells Figure 1: The architecture of massive MIMO system are identical in terms of all configuration and aspects. Each BS is modelled as an M/G/m/m state dependent BS transmits a constant output power P , which is queue [12] in order to more easily calculate the distributed equally among PAs and each PA is connected probability of a certain state of the user in the network. to one antenna element. The Rayleigh channel, which This dependence is a consequence of that the user rate remains static within a time-frequency coherence block directly depends on the number of users served by the of U=B T symbols, is considered, where B is coherence c c c base station simultaneously. bandwidth, while T represent coherence time. To the best of our knowledge, there is no existing work The large-scale fading is assumed to be the same for all the that provides mechanism to cope with the daily load antennas, because the distance between any UE and the variation and maintain high EE throughout the day in a BS is much larger than the distance among the antennas. massive MIMO network. This paper aims to fill this gap in the literature where we consider efficiency taking into We assume that all BSs and UEs employ a time-division account daily load variation, the number of base station duplex (TDD) protocol and zero-forcing (ZF) precoding and antennas and user load distribution. We make the that BSs have perfect channel state information (CSI). following contributions: One of the basic definitions of massive MIMO energy - We derive a new downlink EE equation by considering efficiency (in bits/Joule) represents EE as a ratio of spectral- beside M, K, also the effect of daily load variations for efficiency (sum-rate in Mbit/s) and total consumed power the massive MIMO system. (in Joule/s) [13]: max - We derive adaptive antenna system that adapts the R(kM , ) ∑ i k=1 number of active BS antennas to daily load variations. EE(K , M )= (1) ii total p KM , ( ) i ii The remaining of this work is organized as follows: in The user rate at UE in the cell i is expressed as: Section 1 we present the created model for system,    power consumption, distribution of user location and α K QQ max 12 RK ( )= 1− log 1+ (2) UE i   2  traffic. The second section describes EE maximization TB Q + Q  cc  34  problem. Simulation results for the created system sub- scenarios are described in Section 3. At the end of the Where necessary overhead for channel acquisition in equal to paper there is a conclusion.  α K max 1−  TB  cc 1. ANALYSED MODELS α is the pilot reuse factor, K is the maximum number of max served users and T B is time-frequency coherence block c c 1.1. System model during which channel remain static. Q =p(M /K ) represents 2 i i mean transmit power per user and Q =(M -K ) is the effective 2 i i This paper presents a massive MIMO system that array gain. Average noise power that interferes with the considers only the downlink of a multi-cell system. The signals coming from the serving base station is: cells are hexagonal and each has its own BS. Bδ Q = d UE ( ) i Z 2 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 whereas Total baseband power when the BS serves K users simultaneously with M antennas can be expressed as:  d UE J ( )  jZ Q = pM  4 ∑ j j=1 d (UE )   i Z P KM ,, = P K+= P KM ( ) ( ) ( ) BB ii C 01 i C ii is average inter-cell interference power. The path loss from K BK max the serving BS at origin cell i to UE is equal to d (UE ) while R k P + P + P + ++ P ( )( ) z i z ∑ i COD DEC syn BS k=1 3TL d (UE ) is path loss from interfering BS. (6) C BS j z 2 2   BM K 1 3BM K ii i i 1.2. Power consumption model 2++   L TL BS  C  BS The power consumption (PC) is the sum of the PA when the average output power is p and power consumed max R k ( ) by different analogue components and digital signal where is the total distance-dependent rate ∑ i k=1 processing. When the transmitter is equipped with a calculated for one base station according to (2), whereas large number of antennas circuit power consumption L is BS computational efficiency and B is the bandwidth. BS i dominates the system power consumption. We propose a new refined circuit power consumption model for massive 1.3. User Location Distribution model MIMO systems: Two coverage areas, central and boundary, are defined in total P K , M Pp+=P M Pp+ P K+ ( ) ( ) ( ) ( ) i i i PA C i PA C 0 i this paper. They are divided with radius r , but such that (3) P (KM , )+ P it holds r >r and r <r . r represents the radius of C1 i i FIX D min D max min the circle within which are not distributed users due to the requirement that the minimum distance of the user where P (p) is the total input power of a traditional power PA from BS is 35 m (r =35 m) and r represents the radius amplifier (TPA) and can be approximated as [14], [15]: min max of the circumscribed circle of hexagon. Figure 2 shows P p ≈⋅ pP ( ) (4) geometry of coverage areas. PA max,PA where η is the maximum TPA efficiency and P is max,PA maximum output power. Prominent research directions to minimize power amplifier losses include [16]: - dimensioning the power amplifier, - use of low peak-to-average-power-ratio (PAPR) techniques and - PA-aware design. P (K ) is a part of circuit power and is dependent on the C0 i number of simultaneously served users K while P (K ,M ) i C1 i i depends also of the number of base station antennas M . For circuit power consumption we use the model proposed Figure 2: Geometry of coverage areas in [17]: Central area users are distributed within a circle defined P = P +P +P +P +P (5) by radius r except area defined by radius r . Due to C FIX CE TC C/D LP D min the above in the area of the central area we have the where P is fixed power with constant quantity for difference between the radii r and r . Border area users FIX D min control signalling, site cooling, load independent power of are distributed in the band outside the circle with radius r baseband processing and backhaul infrastructure. P is and inside the hexagon with the radius of the circumscribed CE power of the channel estimation process. P accounts for circle r . A set of weighting factors γ={γ ,γ } is assigned TC max c b the power consumption of the transceiver chains and is to each area and it is a function of time, so that it can given by P =MP +P +KP where P is the power per take different values during the day. The sum of weighting TC TC syn UE BS RF chain at the BS required to run the circuit components, factors at any time instant is always unity i.e., ∑γ=1. As P is power consumed by the local oscillator (LO) and P the cell is split up into two coverage areas, we have a set syn UE is power required by all circuit components of each single- of radii r={r ,r } and in our main simulation scenario we D max antenna UE. P is power consumed for linear processing have that r={300,500} than a set of weighting factors for LP operations at the base station and P is power of the BF ULD are γ ={0.3,0.7} and a set of weighting factors C/D BF channel coding P and decoding P . for CF ULD are γ ={0.8,0.2}. COD DEC CF = S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS 1.4. Traffic model the ability to adapt the number of active antennas for any user states, taking into account the interference from the We consider the daily load profile proposed for data traffic surrounding BSs. In this system BS discovers the most in Europe [18] and model the load at each BS as a state- energy efficient number of active antennas M for a given opt dependent M/G/m/m queue [11] where M is distribution number of users by maximizing EE: of the inter-arrival time of users, G the distribution of max R kM , ( ) the service time, m the number of servers, and m is the ∑ i k=1 M = arg max : opt total capacity of BS. State dependency results from the fact p KM , ( ) i ii (8) that the user rate depends on the number of users served PAPR constraint: M ≥ ∗10 Mk ≥+1 i i by the BS simultaneously. The M/G/m/m queue dictates max,PA that for exponential user arrival with rate λ and general distribution of service rate. Number of servers is equal to The first constraint is due to keep PAPR of the transmitted a maximum number of users. signal at 8 decibel (dB) and second constraint is requirement of ZF precoding. The change in the number of users at BS will result in the change of the service rate for every user. Service rate for n In addition to the optimal value of the number of antennas users that are served simultaneously will be μ(n), but if there for a given number of users, it is very important to is new user served by BS the service rate will change to determine a weighted-average value of M during time opt μ(n+1). Also, when BS stops with serving one user there is instant t is: a change in service rate and it becomes μ(n-1). t max  M π kM∗ k ( ) ( ) (9) avg i opt k=1  The steady state probabilities for the random number of users n in the BS i are modelled as: n In order to analyse the performance of our proposed   λET [ ]  1 adaptive system model, we also created a system  ππ (n)= (0) ii  nn !µ µ n−1 µµ 21 ( ) ( ) ( ) ( ) model that has a fixed number of antennas active all the  (7) time without regard to changes in network. This model    λET [ ] −1 mK =  1 max 1  is called a fixed or reference system model. π 0 = ( )   i ∑ j=0 µ nn µ −1 µµ 21 j! ( ) ( ) ( ) ( )     First action of the proposed algorithm is to determine the global optimal number of antennas and users (M , K ). where π (0) is the probability when no user in cell i, E[T] glopt glopt i 1 This global optimum of antennas is used to initialize the is the expected service time when there is only one user number of antennas at all BS in fixed system and also in BS, μ(n)=(R (n))/(R (1)) is the service rate and it is ratio i i the initial value of the number of antennas in adaptive of the data rate of n users in BS, to data rate of one system. We do not have the closed form expression user. For different user states the data rates are calculated for the M (k), but since M and k are integer values, according to (2). opt the optimum number of antennas can be found by exhaustive search over each user state. According to In a queuing system with no buffer space, the blocking (8) the vector of antennas that maximizes EE was found probability is probability of having K number of users. max and then according to (9) the weighted average optimal As in the works [6], [19], [1] and [11], the probability of number of antennas, M was calculated. Number of simultaneously serving K number of users is 0.02. avg max antennas at a certain number of users and load M is iterate iteratively updated with M , until it converges. Average number of users in other time intervals is avg calculated using λ which is determined by π (K )=0.02. max i max Similar algorithm was proposed in [11] where authors Corresponding average number of users at time instant t have analysed the same conditions but ULD model is in which cell load is x is determined by λ =λ (x /100). t t max t different. Proposed strategy, which contains previously described 2. PROBLEM FORMULATION steps is shown in Algorithm 1. Due to the assumed configuration symmetry in all cells, our proposed The problem to be solved is the maximization of the EE strategy was performed only for one considered cell. of the massive MIMO system described in Section 1. In Optimization algorithm with the same approach has order to simplify the problem solving, the assumption been already proposed in [20]. that the multi-cell environment is symmetric in terms of all configuration and aspects was used. Based on these assumptions, the number of antennas for all cells will be the same. In order to achieve better EE in the massive MIMO system, this paper propose system model that has = B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 3.1. Influence of cell size on performance of massive MIMO system In this section we analyse how deferent cell size influence the performance of the considered system model in the central and boundary coverage areas. Influence of cell size on EE and average user rate (UR) was analysed for cell sizes of 500 m, 1000 m and 2000 m. In comparison to BF ULD, better results of EE and average UR are achieved for CF ULD in each considered cell size, but both ULD models, achieved better results for adaptive system. Higher energy efficiency is achieved for scenarios with a smaller cell size. However, when the cell is too small, the antenna gain and spectral efficiency will be reduced, which is why it is important to determine the size of the cell. The results for the average UR for BF ULD and all considered cell sizes are shown in Figure 3, and for CF ULD in Figure 4. 3. RESULTS AND DISCUSSION The Matlab software was used to test the performance of the massive MIMO system, which in simulation environment consists of 19 regular hexagonal cells. Monte-Carlo insertion method was used to place 10 000 test points in each cell. Test points are not distributed in the circle with cell radius of 35 m, which is determined by the minimum distance from the base station. In order to avoid border effects and achieve a situation where all base stations receive interference from all directions equally, a wrap-around topology was used. Figure 3: Average UR for BF ULD at different cell size. More precisely, for each combination of user device and base station, eight alternative base station locations are considered and it is determined which one has the shortest distance to the user device. A bandwidth of 20 MHz was used for communication with a total receiver noise power of −96 dBm. The coherent bandwidth is 180 kHz, and the coherent time is 10 ms, which results in a coherent block of 1800 samples. The channel model uses a Rayleigh fading channel, and the path attenuation factor is 3.76, which determines the loss of transmission of the base station signal to the user device. The base station operates on a carrier frequency of 2 GHz. In order to find optimal size of the cell, bandwidth and base station output power level additional sub-scenarios have been created in which influence of certain system Figure 4: Average UR for CF ULD at different cell size. parameter on EE was analysed. 5 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS The best results of the average UR are achieved for a 3.2. Influence of bandwidth on performance of massive fixed system, CF ULD and a cell size of 500 meters. With MIMO system the increase of the network load, the average UR in the adaptive system decreases and for network loads of 10- Allocation of additional spectrum in 5G, which will bring 30% there is a significant decrease in average UR, while wider bandwidths is necessary. The maximum channel for network loads of 30-100% the average UR has almost bandwidth defined for Frequency range 1 (< 6 GHz) is the same values 100 MHz, while the minimum channel bandwidth defined for Frequency range 2 (24–54 GHz) is 50 MHz and the The fixed antenna system has approximate values for maximum is 400 MHz. all network load values, which is expected because this model of antenna system always has the same This section analyses the influence of the bandwidth number of activated antennas, regardless of the network values on the EE and average UR. The values for which load and the number of users. In order to compare the the impact is analysed are values from 20 MHz to 100 adaptive and fixed system results, EE gain was used, MHz with a step of 20 MHz. which represents a simple ratio of these two quantities, i.e. (EE -EE )/EE . EE gain of adaptive antenna Figure 6 shows the results for EE of network load for BF adaptive fixed fixed system has excellent results for both, the CF and BF ULD at different bandwidth values (20 MHz, 60 MHz and ULD, and increases with the increase of the cell size, but 100 MHz), while Figure 7 shows the results for CF ULD. it decreases with increasing cell load. The corresponding EE and average UR are increased when bandwidth is EE gain for cell radius of 500 m, and cell load of {10, increased, for both CF and BF ULD, but CF ULD achieved 50, 100} % for BF is {204.96, 39.56, 0}% and for CF is better results for all analysed bandwidth values. The {244.32, 125.88, 49.51}%. The achieved values of EE difference between the achieved results of CF and BF gain for different cell radii are shown in Figure 5. ULD is not significant, but when bandwidth and network load are increased, this difference is also increased in benefit of CF ULD. Figure 5: EE gain for different cell size and ULD model Figure 6: EE for BF ULD at different bandwidth values. The conclusion for this part of the analyses in which the influence of cell size was considered is the following: - The maximum number of active users K , and EE max increases with decreasing cell size, - M and the difference in EE gain between the max fixed and adaptive antenna system decreases with decrease cell size. - As cell size and network load increase, average UR in the adaptive antenna system decrease. - Based on the results of all sub-scenarios, the best network performance was achieved when R =500, max which represents the optimal cell size. Figure 7: EE for CF ULD at different bandwidth values. 6 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 In analysed scenarios with different bandwidth, the proposed adaptive system achieved better results than the fixed one. The results of the adaptive system are significantly better, when the bandwidth in increased. The difference between the achieved results of EE for fixed and adaptive system in BF ULD is smaller and with increasing network load EE achieves approximate values in both considered models. When the bandwidth is increased from 20 MHz to 60 MHz, the EE value is doubled for BF ULD and network loads of 20-100%. Also, with CF ULD the same level of EE improvement was achieved in the mentioned two scenarios with different bandwidth but for network loads of 30-100%. Further increase in bandwidth, i.e. up to 100 MHz still results in a significant increase in EE but in a smaller ratio. In the case of CF ULD and a bandwidth of 100 MHz, an EE Figure 9: Average UR for CF ULD at different bandwidth values. of slightly less than 50 [Mbit/Joule] was achieved, which is three times higher than in the case of the bandwidth With BF ULD, the difference between the average UR of the of 20 MHz used in the baseline scenario. Also, BF ULDs fixed and adaptive system model decreases with increasing achieve a threefold increase in EE when the bandwidth is network load, so that at a network load of 90%, these two 100 MHz compared to the scenario with a bandwidth of models have almost the same values. On the other hand, 20 MHz. Fixed antenna system, in both ULD models at the with CF ULD, for almost all network load values, there is aforementioned network load range values, also achieved an equal difference between the average speed per user three times the EE for the 100 MHz bandwidth compared of a fixed and adaptive system model. For CF ULD there is to the sub-scenario having a 20 MHz bandwidth. the biggest difference between the results of the fixed and adaptive system model. This difference is smaller at lower Figure 8 and Figure 9 show the results of the achieved network loads in scenarios with higher bandwidth. average UR for BF and CF ULD, respectively. The obtained results show that in this scenario, higher average UR were Due to the abovementioned, the highest EE gains were achieved in the fixed system model and in the case of CF achieved in the case of CF ULD, which are significantly higher ULD. For both ULD models, the highest average UR was for network loads up to 40% in the baseline scenario (BW achieved with a bandwidth of 100 MHz, which is more = 20 MHz), after which approximate values were achieved. than 4 times higher than the baseline scenario with 20 The results for EE gain and loss of UR are shown in Figure MHz, i.e. the increase in bandwidth increases and the 10 and Figure 11. Despite these improvements, maximum average speed per user. number of users and antennas are significantly reduced with increasing bandwidth so that a scenario with the bandwidth Figure 8: Average UR for BF ULD at different bandwidth values. Figure 10: EE gain at different network load, bandwidth and ULD model 7 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS Figure 11: User-rate loss at different network load, bandwidth and ULD model Figure 12: Optimal number of active antennas of 100 Hz have M =96 and K =43. Therefore, choosing max max the appropriate bandwidth will depend on the compromise reached between the number of users served simultaneously and the desired level of performance for those users. 3.3. Influence of the output power level of the BS on performance of massive MIMO system For the analysis of the influence of the output power level of the BS on the performance of the system under consideration, 4 scenarios were created. The values for which the impact is analysed are values from 10 W to 40 W with a step of 10 W. The output power level of the BS is directly related to the number of active antennas, as well as the EE, which is why it is very important to analyse how the increase in network load affects the activation of the antenna at the base station. BF ULD requires significantly higher activation of antennas Figure 13: EE at different values of output power level of the BS with increasing network load than CF ULD. In both cases of and ULD models. ULD, the highest activation of antennas was for P =40 W. CF ULD achieved the best results when P =20 W and in this case it is necessary to activate an average of 60 antennas 4. CONCLUSION at maximum network load. The best results for BF ULD are achieved, when P =10 W and in this scenario it is necessary Massive MIMO technology is considered crucial for the to activate on average 3 times more antennas at maximum further development of wireless networks, and due to its network load i.e., 185 antennas. characteristics and capabilities, it is the basic technology for achieving the requirements set for the 5G standard Figure 12 shows the results of the optimal number of active and the future 6G standard. In this paper, we investigate BS antennas in the case of CF ULD and both system models resource allocation of a load adaptive massive MIMO (adaptive and fixed). system with different ULD models. Our developed resource allocation scheme is different from the existing The results have the same trend of behaviour, i.e. the largest ones due to the fact that it jointly optimizes the user load M is for P =40 W, and the smallest for P =10 W W for distribution, number of BS antennas and cell loading. opt c c both system models. Increasing the number of users in the network linearly increases the number of optimal antennas, We also determined the impact of cell size, available but in the case of higher network load (100%) a higher M bandwidth and output power lever of the BS at different opt is recorded, compared to the network load of 20%. The cell loading on EE and average UR. When the number of representation of EE for all analysed scenarios with different service antennas in massive MIMO systems is determined, P values is shown in Figure 13. other system parameters can be adjusted (radiation power, number of users, etc.) to achieve higher spectral 8 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 efficiency at the cost of reduced energy efficiency, and [5] H. A. Ammar, R. Adve, S. Shahbazpanahi, G. vice versa. Boudreau and K. V. Srinivas, “Downlink Resource Allocation in Multiuser Cell-Free MIMO Networks With The results of the analysis confirm that EE increases User-Centric Clustering,” in IEEE Transactions on with decreasing cell size, however, decreasing cell size Wireless Communications, vol. 21, no. 3, pp. 1482- also affects the reduction of antenna array gain. Spectral 1497, 2022, doi: 10.1109/TWC.2021.3104456 efficiency is reduced when the cell size is too small. Also, [6] M. M. Aftab Hossain, R. Jäntti and C. Cavdar, as cell size decreases, transmission power decreases as “Dimensioning of PA for massive MIMO system users will be closer to the base station. In the scenario with load adaptive number of antennas,” 2014 with a bandwidth of 100 MHz, the achieved EE is three IEEE Globecom Workshops (GC Wkshps), Austin, times higher than in the baseline scenario (20 MHz). TX, USA, 2014, pp. 1102-1108, doi: 10.1109/ GLOCOMW.2014.7063580 Also, the results of the proposed adaptive system model [7] M. Ataeeshojai, R. C. Elliott, W. Krzymien, C. are compared with the results of the fixed system under Tellambura and J. Melzer, “Energy-Efficient the same conditions in the network, all with the aim of Resource Allocation in Single-RF Load-Modulated determining the improvement of the system’s performance Massive MIMO HetNets,” IEEE Open Journal of the when the number of BS antennas is adjusted to the Communications Society, pp. 1-1, 2020 network condition. The results showed that the adaptive [8] M. Miriestahbanati, “Resource Allocation in system model achieved better results for EE and average Collocated Massive MIMO for 5G and Beyond”, UR in all scenarios. For all-network loads and for both MS thesis, Concordia University, Montreal, Quebec, system models, adaptive and fixed system, worse results Canada, 2021 for EE and average UR were obtained for the BF ULD case, [9] H. Li, L. Song and M. Debbah, “Energy Efficiency which is a consequence of interference from neighbouring of Large-Scale Multiple Antenna Systems with base stations. Transmit Antenna Selection,” IEEE Transactions on Communications, vol. 62, no. 2, pp. 638-647, 2014 The paper assumes that the CSIs for the receiver are [10] H. V. Cheng, D. Persson, E. Björnson and E. G. perfectly known. However, this assumption may be Larsson, “Massive MIMO at Night: On the Operation unrealistic. An interesting and important problem is to of Massive MIMO in Low Traffic Scenarios,” in Proc. analyse the behaviour of the proposed system model IEEE ICC, London, UK, June 2015 when the CSI data are unreliable. Also, it is very important [11] M. A. Abuibaid and S. A. Çolak, “Energy-efficient to deal with the pilot contamination. Future research massive MIMO system: Exploiting user location should investigate the appropriate design of frequency distribution variation,” AEU – International Journal reuse factors during training and power control to reduce of Electronics and Communications, vol. 72, pp. pilot contamination effect. 17–25, 2017, doi: https://doi.org/10.1016%2Fj. aeue.2016.11.013 [12] J. Y. Cheah and J. M. Smith, “Generalized M/G/C/C state dependent queueing models and pedestrian REFERENCES traffic flows,” Queueing Syst. vol. 15, pp. 365-386, March 1994 [1] M. M. A. Hossain, C. Cavdar, E. Bjornson and R. [13] R. M. Asif, J. Arshad, M. Shakir, S. M. Noman and Jantti, “Energy-Efficient Load-Adaptive Massive A. U. Rehman, “Energy Efficiency Augmentation in MIMO,” 2015 IEEE Globecom Workshops (GC Massive MIMO Systems through Linear Precoding Wkshps), San Diego, CA, USA, 2015, pp. 1-6, doi: Schemes and Power Consumption Modeling,” 10.1109/GLOCOMW.2015.741418 Wireless Communications and Mobile Computing, [2] A. A. Khan, R. S. Adve and W. Yu, “Optimizing 2020, doi: 10.1155/2020/8839088 Downlink Resource Allocation in Multiuser MIMO [14] M. M. A. Hossain and R. Jantti, “Impact of efficient Networks via Fractional Programming and the power amplifiers in wireless access,” 2011 IEEE Hungarian Algorithm,” in IEEE Transactions on Online Conference on Green Communications, Wireless Communications, vol. 19, no. 8, pp. 5162- (Online Conference) Piscataway, NJ, USA, 2011, pp. 5175, 2020, doi: 10.1109/TWC.2020.2990176 36-40, doi: 10.1109/GreenCom.2011.6082504 [3] E. Björnson, L. Sanguinetti, J. Hoydis and M. Debbah, [15] D. Persson, T. Eriksson and E.G. Larsson, “Amplifier- “Optimal Design of Energy-Efficient Multi-User Aware MultipleInput Multiple-Output Power MIMO Systems: Is Massive MIMO the Answer?,” in Allocation,” in IEEE Commun. Letters, vol. 17, IEEE Transactions on Wireless Communications, no. 6, pp. 1112-1115, June 2013, doi: 10.1109/ vol. 14, no. 6, pp. 3059-3075, 2015 doi: 10.1109/ LCOMM.2013.043013.130050 TWC.2015.2400437 [16] N. R. Koppisetti, “Massive MIMO for 5G Wireless [4] A. Salh, L. Audah, Q. Abdullah, N. Shahida, S. A. Hamzah, Networks, An Energy Efficiency Perspective”, et. al, “Energy-Efficient Low-Complexity Algorithm in Vancouver: MS thesis, University of British Columbia, 5G Massive MIMO Systems,” Computers, Materials & Continua, vol. 67, no. 2, pp. 3189-3214, 2021 9 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS [17] E. Bjornson, L. Sanguinetti, J. Hoydis and M. BIOGRAPHY Debbah, “Designing Multi-User MIMO for Energy Efficiency: When is Massive MIMO the Answer?,” in Samira Mujkić received the B.S., M.S. and Ph.D. Proc. of IEEE WCNC, Istanbul, Turkey, April 2014 degrees in electrical engineering from the University of [18] G. Auer et. al., “D2.3: Energy efficiency analysis of Tuzla, Tuzla, Bosnia and Herzegovina, in 2011, 2014 the reference systems, areas of improvements and and 2022, respectively. From 2012 to 2014, she was target breakdown,” INFSO-ICT-247733 EARTH, ver. a Teaching Assistant and from 2014 to 2017 she was 2.0, (2012), http://www.ict-earth.eu/. Accessed on: a Lecturer with the College of Computer Science and Jun. 13, 2022 Business Communications eMPIRICA, Brcko, Bosnia and [19] M. M. A. Hossain, C. Cavdar, E. Björnson and R. Herzegovina. She currently works at the Government of Jäntti, “Energy Efficiency of Massive MIMO: Coping the Brcko District of Bosnia and Herzegovina. Her main with Daily Load Variation,” 2015, https://arxiv.org/ research interests include radio resource allocation, abs/1512.01998. Accessed on: Jun. 26, 2022 wireless communications, MIMO and massive MIMO [20] S. Mujkic, S. Kasapovic and M. Abuibaid, “Energy- systems. efficient resource optimization for massive mimo networks considering network load,” Computers, Suad Kasapović received the B.S. degree in Electrical Materials & Continua, vol. 71, no.1, pp. 871–888, Engineering from the University of Tuzla, Tuzla, Bosnia 2022 and Herzegovina, in 1996. He then received the M.Sc. and Ph.D. degrees in fields Telecommunications and Informatics at the Faculty of electrical engineering and computing, from the University of Zagreb, Croatia in 2002 and 2007, respectively. He is currently an Associate Professor with the Department of Telecommunications, University of Tuzla. His general research interests include signal processing and wireless communications, with emphasis on MIMO communication systems, practical issues in 5G systems, and wireless sensor network, and etc. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png B&H Electrical Engineering de Gruyter

Resource Allocation and Impact Analysis of the System Parameters on Performance in Multi-Cell Massive MIMO Networks

Mujkić, Samira; Kasapović, Suad
B&H Electrical Engineering , Volume 16 (2): 10 – Dec 1, 2022
Download PDF
10 pages

Loading next page...
 
/lp/de-gruyter/resource-allocation-and-impact-analysis-of-the-system-parameters-on-1L0xwEiEPn

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
© 2022 Samira Mujkić et al., published by Sciendo
eISSN
2566-3151
DOI
10.2478/bhee-2022-0012
Publisher site
See Article on Publisher Site

Abstract

B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 Submitted: August 29, 2022 Original scientific paper Accepted: November 5, 2022 RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS 1,2 1 Samira Mujkić , Suad Kasapović Abstract: This paper investigates the energy-efficient resource allocation algorithm for a massive multiple input multiple output (MIMO) system, in which each base station adapts the number of antennas to the daily load profile. Our paper examines the effect of two user location distribution (ULD) models, on the energy-efficiency (EE) of load adaptive masive MIMO system. We propose a resource allocation strategy to adapt the number of antennas based on tracking variations of ULD and cell loading maximizing the EE. We also evaluate impact of cell size, available bandwidth and output power level of the BS on EE at different cell loading. Keywords: energy efficiency, massive MIMO, traffic load, user location distribution INTRODUCTION selects the optimal set of users from the much larger pool of potential users, for a given set of beamforming weights, th The new technology of the 5 generation (5G) standard in polynomial time. Furthermore, several works [3] the that will most significantly increase energy efficiency is a authors analyse how the three main design parameters; technology called massive MIMO. Massive MIMO will also namely the number K of active user equipment (UEs), the increase network bandwidth and thus meet the current M number of antennas at the base station (BS), and the demand for higher data rates. transmit power must be chosen to improve EE. With excessive power consumption in wireless A low-complexity resource allocation scheme has been communications networks, beside spectral efficiency (SE), recently investigated in the literature. In particular, authors EE has become another significant metric for evaluating in [4] developed an algorithm that jointly optimizes the the performances of wireless communications systems. number of antennas, user selection and power allocation The energy consumption of a base station (BS) with no in a multi-user massive MIMO system, while the work load is at least half of the maximum energy it consumes in [5] optimizes user scheduling, power allocation and at the peak. Therefore, it is very important that 5G beamforming in MIMO networks implementing user- networks are able to adapt their power consumption with centric clustering. temporal changes of load [1]. In order to improve the energy efficiency of the 5G base station it is necessary In [6] the authors investigated the impact of power to develop joint optimization schemes and algorithms to amplifiers (PA) dimensioning on the energy efficiency of save computational and transmission power in the main load adaptive massive MIMO system. A common algorithm unit of the base station and radio frequency (RF) chains for antenna selection and resource block (RB) allocation together. Management of radio resources requires the was created in [7] together with power optimization establishment of strategies and algorithms that will use algorithm. Resource allocation method presented in [8] limited resources in the most efficient way. Due to the optimally select data and pilot powers along with the constant changes in the cells, it is necessary to establish training duration and maximize the spectral efficiency for a dynamic allocation of resources where the radio the three deferent receivers. The paper [9] also analysed resources will follow the changes and be adaptable to the energy efficiency of a system with a large number of them. Resource allocation established in this way should antennas where the transmission antennas were selected result in high energy efficiency. and the authors proposed two algorithms for the selection of transmitting antennas based on the application of In [2] the authors developed a joint beamforming and user sequential and binary search algorithms. scheduling algorithm based on fractional programming and the Hungarian algorithm. The Hungarian algorithm In [10] the authors show that during low traffic demand, e.g., late night, it is optimal to turn off a fraction of the Faculty of Electrical Engineering, University of Tuzla, Bosnia and Herzegovina Correspondence email: samira.mujkic@untz.ba © 2022 Author(s). This is an open access article licensed under the Creative Commons Attribution License 4.0. (http://creativecommons.org/licenses/by/4.0/). 1 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS antennas while minimizing total power consumption if the The number of users served by each BS is denoted by K , power consumption in the transceiver circuits is taken into while the number of used antennas is denoted by M . In account along with the RF amplifiers. network with massive MIMO the ratio of these two units must be M >K . Users have devices with one antenna. i i This paper investigates the energy-efficient resource User devices with multiple antennas are not considered allocation algorithm for a massive MIMO system, in which due to the computational complexity. each base station has a large number of antennas and it adapts the number of antennas to the daily load profile. Two The architecture of analysed massive MIMO system is user location distribution (ULD) models, central focused shown in Figure 1. (CF) ULD and boundary focused (BF) ULD, are defined in this paper. The formed ULD model defines two coverage areas with different user densities. ULD analysed in this paper is similar to models in [11] but in our paper we have also analysed the influence of certain system parameters on EE for different ULD model at different cell load. In particular, we aim at bringing new insights on how the number M of antennas at the BS, the number K of active UEs, and other system parameters should be chosen in order to achieve maximal EE. It was assumed that cells Figure 1: The architecture of massive MIMO system are identical in terms of all configuration and aspects. Each BS is modelled as an M/G/m/m state dependent BS transmits a constant output power P , which is queue [12] in order to more easily calculate the distributed equally among PAs and each PA is connected probability of a certain state of the user in the network. to one antenna element. The Rayleigh channel, which This dependence is a consequence of that the user rate remains static within a time-frequency coherence block directly depends on the number of users served by the of U=B T symbols, is considered, where B is coherence c c c base station simultaneously. bandwidth, while T represent coherence time. To the best of our knowledge, there is no existing work The large-scale fading is assumed to be the same for all the that provides mechanism to cope with the daily load antennas, because the distance between any UE and the variation and maintain high EE throughout the day in a BS is much larger than the distance among the antennas. massive MIMO network. This paper aims to fill this gap in the literature where we consider efficiency taking into We assume that all BSs and UEs employ a time-division account daily load variation, the number of base station duplex (TDD) protocol and zero-forcing (ZF) precoding and antennas and user load distribution. We make the that BSs have perfect channel state information (CSI). following contributions: One of the basic definitions of massive MIMO energy - We derive a new downlink EE equation by considering efficiency (in bits/Joule) represents EE as a ratio of spectral- beside M, K, also the effect of daily load variations for efficiency (sum-rate in Mbit/s) and total consumed power the massive MIMO system. (in Joule/s) [13]: max - We derive adaptive antenna system that adapts the R(kM , ) ∑ i k=1 number of active BS antennas to daily load variations. EE(K , M )= (1) ii total p KM , ( ) i ii The remaining of this work is organized as follows: in The user rate at UE in the cell i is expressed as: Section 1 we present the created model for system,    power consumption, distribution of user location and α K QQ max 12 RK ( )= 1− log 1+ (2) UE i   2  traffic. The second section describes EE maximization TB Q + Q  cc  34  problem. Simulation results for the created system sub- scenarios are described in Section 3. At the end of the Where necessary overhead for channel acquisition in equal to paper there is a conclusion.  α K max 1−  TB  cc 1. ANALYSED MODELS α is the pilot reuse factor, K is the maximum number of max served users and T B is time-frequency coherence block c c 1.1. System model during which channel remain static. Q =p(M /K ) represents 2 i i mean transmit power per user and Q =(M -K ) is the effective 2 i i This paper presents a massive MIMO system that array gain. Average noise power that interferes with the considers only the downlink of a multi-cell system. The signals coming from the serving base station is: cells are hexagonal and each has its own BS. Bδ Q = d UE ( ) i Z 2 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 whereas Total baseband power when the BS serves K users simultaneously with M antennas can be expressed as:  d UE J ( )  jZ Q = pM  4 ∑ j j=1 d (UE )   i Z P KM ,, = P K+= P KM ( ) ( ) ( ) BB ii C 01 i C ii is average inter-cell interference power. The path loss from K BK max the serving BS at origin cell i to UE is equal to d (UE ) while R k P + P + P + ++ P ( )( ) z i z ∑ i COD DEC syn BS k=1 3TL d (UE ) is path loss from interfering BS. (6) C BS j z 2 2   BM K 1 3BM K ii i i 1.2. Power consumption model 2++   L TL BS  C  BS The power consumption (PC) is the sum of the PA when the average output power is p and power consumed max R k ( ) by different analogue components and digital signal where is the total distance-dependent rate ∑ i k=1 processing. When the transmitter is equipped with a calculated for one base station according to (2), whereas large number of antennas circuit power consumption L is BS computational efficiency and B is the bandwidth. BS i dominates the system power consumption. We propose a new refined circuit power consumption model for massive 1.3. User Location Distribution model MIMO systems: Two coverage areas, central and boundary, are defined in total P K , M Pp+=P M Pp+ P K+ ( ) ( ) ( ) ( ) i i i PA C i PA C 0 i this paper. They are divided with radius r , but such that (3) P (KM , )+ P it holds r >r and r <r . r represents the radius of C1 i i FIX D min D max min the circle within which are not distributed users due to the requirement that the minimum distance of the user where P (p) is the total input power of a traditional power PA from BS is 35 m (r =35 m) and r represents the radius amplifier (TPA) and can be approximated as [14], [15]: min max of the circumscribed circle of hexagon. Figure 2 shows P p ≈⋅ pP ( ) (4) geometry of coverage areas. PA max,PA where η is the maximum TPA efficiency and P is max,PA maximum output power. Prominent research directions to minimize power amplifier losses include [16]: - dimensioning the power amplifier, - use of low peak-to-average-power-ratio (PAPR) techniques and - PA-aware design. P (K ) is a part of circuit power and is dependent on the C0 i number of simultaneously served users K while P (K ,M ) i C1 i i depends also of the number of base station antennas M . For circuit power consumption we use the model proposed Figure 2: Geometry of coverage areas in [17]: Central area users are distributed within a circle defined P = P +P +P +P +P (5) by radius r except area defined by radius r . Due to C FIX CE TC C/D LP D min the above in the area of the central area we have the where P is fixed power with constant quantity for difference between the radii r and r . Border area users FIX D min control signalling, site cooling, load independent power of are distributed in the band outside the circle with radius r baseband processing and backhaul infrastructure. P is and inside the hexagon with the radius of the circumscribed CE power of the channel estimation process. P accounts for circle r . A set of weighting factors γ={γ ,γ } is assigned TC max c b the power consumption of the transceiver chains and is to each area and it is a function of time, so that it can given by P =MP +P +KP where P is the power per take different values during the day. The sum of weighting TC TC syn UE BS RF chain at the BS required to run the circuit components, factors at any time instant is always unity i.e., ∑γ=1. As P is power consumed by the local oscillator (LO) and P the cell is split up into two coverage areas, we have a set syn UE is power required by all circuit components of each single- of radii r={r ,r } and in our main simulation scenario we D max antenna UE. P is power consumed for linear processing have that r={300,500} than a set of weighting factors for LP operations at the base station and P is power of the BF ULD are γ ={0.3,0.7} and a set of weighting factors C/D BF channel coding P and decoding P . for CF ULD are γ ={0.8,0.2}. COD DEC CF = S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS 1.4. Traffic model the ability to adapt the number of active antennas for any user states, taking into account the interference from the We consider the daily load profile proposed for data traffic surrounding BSs. In this system BS discovers the most in Europe [18] and model the load at each BS as a state- energy efficient number of active antennas M for a given opt dependent M/G/m/m queue [11] where M is distribution number of users by maximizing EE: of the inter-arrival time of users, G the distribution of max R kM , ( ) the service time, m the number of servers, and m is the ∑ i k=1 M = arg max : opt total capacity of BS. State dependency results from the fact p KM , ( ) i ii (8) that the user rate depends on the number of users served PAPR constraint: M ≥ ∗10 Mk ≥+1 i i by the BS simultaneously. The M/G/m/m queue dictates max,PA that for exponential user arrival with rate λ and general distribution of service rate. Number of servers is equal to The first constraint is due to keep PAPR of the transmitted a maximum number of users. signal at 8 decibel (dB) and second constraint is requirement of ZF precoding. The change in the number of users at BS will result in the change of the service rate for every user. Service rate for n In addition to the optimal value of the number of antennas users that are served simultaneously will be μ(n), but if there for a given number of users, it is very important to is new user served by BS the service rate will change to determine a weighted-average value of M during time opt μ(n+1). Also, when BS stops with serving one user there is instant t is: a change in service rate and it becomes μ(n-1). t max  M π kM∗ k ( ) ( ) (9) avg i opt k=1  The steady state probabilities for the random number of users n in the BS i are modelled as: n In order to analyse the performance of our proposed   λET [ ]  1 adaptive system model, we also created a system  ππ (n)= (0) ii  nn !µ µ n−1 µµ 21 ( ) ( ) ( ) ( ) model that has a fixed number of antennas active all the  (7) time without regard to changes in network. This model    λET [ ] −1 mK =  1 max 1  is called a fixed or reference system model. π 0 = ( )   i ∑ j=0 µ nn µ −1 µµ 21 j! ( ) ( ) ( ) ( )     First action of the proposed algorithm is to determine the global optimal number of antennas and users (M , K ). where π (0) is the probability when no user in cell i, E[T] glopt glopt i 1 This global optimum of antennas is used to initialize the is the expected service time when there is only one user number of antennas at all BS in fixed system and also in BS, μ(n)=(R (n))/(R (1)) is the service rate and it is ratio i i the initial value of the number of antennas in adaptive of the data rate of n users in BS, to data rate of one system. We do not have the closed form expression user. For different user states the data rates are calculated for the M (k), but since M and k are integer values, according to (2). opt the optimum number of antennas can be found by exhaustive search over each user state. According to In a queuing system with no buffer space, the blocking (8) the vector of antennas that maximizes EE was found probability is probability of having K number of users. max and then according to (9) the weighted average optimal As in the works [6], [19], [1] and [11], the probability of number of antennas, M was calculated. Number of simultaneously serving K number of users is 0.02. avg max antennas at a certain number of users and load M is iterate iteratively updated with M , until it converges. Average number of users in other time intervals is avg calculated using λ which is determined by π (K )=0.02. max i max Similar algorithm was proposed in [11] where authors Corresponding average number of users at time instant t have analysed the same conditions but ULD model is in which cell load is x is determined by λ =λ (x /100). t t max t different. Proposed strategy, which contains previously described 2. PROBLEM FORMULATION steps is shown in Algorithm 1. Due to the assumed configuration symmetry in all cells, our proposed The problem to be solved is the maximization of the EE strategy was performed only for one considered cell. of the massive MIMO system described in Section 1. In Optimization algorithm with the same approach has order to simplify the problem solving, the assumption been already proposed in [20]. that the multi-cell environment is symmetric in terms of all configuration and aspects was used. Based on these assumptions, the number of antennas for all cells will be the same. In order to achieve better EE in the massive MIMO system, this paper propose system model that has = B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 3.1. Influence of cell size on performance of massive MIMO system In this section we analyse how deferent cell size influence the performance of the considered system model in the central and boundary coverage areas. Influence of cell size on EE and average user rate (UR) was analysed for cell sizes of 500 m, 1000 m and 2000 m. In comparison to BF ULD, better results of EE and average UR are achieved for CF ULD in each considered cell size, but both ULD models, achieved better results for adaptive system. Higher energy efficiency is achieved for scenarios with a smaller cell size. However, when the cell is too small, the antenna gain and spectral efficiency will be reduced, which is why it is important to determine the size of the cell. The results for the average UR for BF ULD and all considered cell sizes are shown in Figure 3, and for CF ULD in Figure 4. 3. RESULTS AND DISCUSSION The Matlab software was used to test the performance of the massive MIMO system, which in simulation environment consists of 19 regular hexagonal cells. Monte-Carlo insertion method was used to place 10 000 test points in each cell. Test points are not distributed in the circle with cell radius of 35 m, which is determined by the minimum distance from the base station. In order to avoid border effects and achieve a situation where all base stations receive interference from all directions equally, a wrap-around topology was used. Figure 3: Average UR for BF ULD at different cell size. More precisely, for each combination of user device and base station, eight alternative base station locations are considered and it is determined which one has the shortest distance to the user device. A bandwidth of 20 MHz was used for communication with a total receiver noise power of −96 dBm. The coherent bandwidth is 180 kHz, and the coherent time is 10 ms, which results in a coherent block of 1800 samples. The channel model uses a Rayleigh fading channel, and the path attenuation factor is 3.76, which determines the loss of transmission of the base station signal to the user device. The base station operates on a carrier frequency of 2 GHz. In order to find optimal size of the cell, bandwidth and base station output power level additional sub-scenarios have been created in which influence of certain system Figure 4: Average UR for CF ULD at different cell size. parameter on EE was analysed. 5 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS The best results of the average UR are achieved for a 3.2. Influence of bandwidth on performance of massive fixed system, CF ULD and a cell size of 500 meters. With MIMO system the increase of the network load, the average UR in the adaptive system decreases and for network loads of 10- Allocation of additional spectrum in 5G, which will bring 30% there is a significant decrease in average UR, while wider bandwidths is necessary. The maximum channel for network loads of 30-100% the average UR has almost bandwidth defined for Frequency range 1 (< 6 GHz) is the same values 100 MHz, while the minimum channel bandwidth defined for Frequency range 2 (24–54 GHz) is 50 MHz and the The fixed antenna system has approximate values for maximum is 400 MHz. all network load values, which is expected because this model of antenna system always has the same This section analyses the influence of the bandwidth number of activated antennas, regardless of the network values on the EE and average UR. The values for which load and the number of users. In order to compare the the impact is analysed are values from 20 MHz to 100 adaptive and fixed system results, EE gain was used, MHz with a step of 20 MHz. which represents a simple ratio of these two quantities, i.e. (EE -EE )/EE . EE gain of adaptive antenna Figure 6 shows the results for EE of network load for BF adaptive fixed fixed system has excellent results for both, the CF and BF ULD at different bandwidth values (20 MHz, 60 MHz and ULD, and increases with the increase of the cell size, but 100 MHz), while Figure 7 shows the results for CF ULD. it decreases with increasing cell load. The corresponding EE and average UR are increased when bandwidth is EE gain for cell radius of 500 m, and cell load of {10, increased, for both CF and BF ULD, but CF ULD achieved 50, 100} % for BF is {204.96, 39.56, 0}% and for CF is better results for all analysed bandwidth values. The {244.32, 125.88, 49.51}%. The achieved values of EE difference between the achieved results of CF and BF gain for different cell radii are shown in Figure 5. ULD is not significant, but when bandwidth and network load are increased, this difference is also increased in benefit of CF ULD. Figure 5: EE gain for different cell size and ULD model Figure 6: EE for BF ULD at different bandwidth values. The conclusion for this part of the analyses in which the influence of cell size was considered is the following: - The maximum number of active users K , and EE max increases with decreasing cell size, - M and the difference in EE gain between the max fixed and adaptive antenna system decreases with decrease cell size. - As cell size and network load increase, average UR in the adaptive antenna system decrease. - Based on the results of all sub-scenarios, the best network performance was achieved when R =500, max which represents the optimal cell size. Figure 7: EE for CF ULD at different bandwidth values. 6 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 In analysed scenarios with different bandwidth, the proposed adaptive system achieved better results than the fixed one. The results of the adaptive system are significantly better, when the bandwidth in increased. The difference between the achieved results of EE for fixed and adaptive system in BF ULD is smaller and with increasing network load EE achieves approximate values in both considered models. When the bandwidth is increased from 20 MHz to 60 MHz, the EE value is doubled for BF ULD and network loads of 20-100%. Also, with CF ULD the same level of EE improvement was achieved in the mentioned two scenarios with different bandwidth but for network loads of 30-100%. Further increase in bandwidth, i.e. up to 100 MHz still results in a significant increase in EE but in a smaller ratio. In the case of CF ULD and a bandwidth of 100 MHz, an EE Figure 9: Average UR for CF ULD at different bandwidth values. of slightly less than 50 [Mbit/Joule] was achieved, which is three times higher than in the case of the bandwidth With BF ULD, the difference between the average UR of the of 20 MHz used in the baseline scenario. Also, BF ULDs fixed and adaptive system model decreases with increasing achieve a threefold increase in EE when the bandwidth is network load, so that at a network load of 90%, these two 100 MHz compared to the scenario with a bandwidth of models have almost the same values. On the other hand, 20 MHz. Fixed antenna system, in both ULD models at the with CF ULD, for almost all network load values, there is aforementioned network load range values, also achieved an equal difference between the average speed per user three times the EE for the 100 MHz bandwidth compared of a fixed and adaptive system model. For CF ULD there is to the sub-scenario having a 20 MHz bandwidth. the biggest difference between the results of the fixed and adaptive system model. This difference is smaller at lower Figure 8 and Figure 9 show the results of the achieved network loads in scenarios with higher bandwidth. average UR for BF and CF ULD, respectively. The obtained results show that in this scenario, higher average UR were Due to the abovementioned, the highest EE gains were achieved in the fixed system model and in the case of CF achieved in the case of CF ULD, which are significantly higher ULD. For both ULD models, the highest average UR was for network loads up to 40% in the baseline scenario (BW achieved with a bandwidth of 100 MHz, which is more = 20 MHz), after which approximate values were achieved. than 4 times higher than the baseline scenario with 20 The results for EE gain and loss of UR are shown in Figure MHz, i.e. the increase in bandwidth increases and the 10 and Figure 11. Despite these improvements, maximum average speed per user. number of users and antennas are significantly reduced with increasing bandwidth so that a scenario with the bandwidth Figure 8: Average UR for BF ULD at different bandwidth values. Figure 10: EE gain at different network load, bandwidth and ULD model 7 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS Figure 11: User-rate loss at different network load, bandwidth and ULD model Figure 12: Optimal number of active antennas of 100 Hz have M =96 and K =43. Therefore, choosing max max the appropriate bandwidth will depend on the compromise reached between the number of users served simultaneously and the desired level of performance for those users. 3.3. Influence of the output power level of the BS on performance of massive MIMO system For the analysis of the influence of the output power level of the BS on the performance of the system under consideration, 4 scenarios were created. The values for which the impact is analysed are values from 10 W to 40 W with a step of 10 W. The output power level of the BS is directly related to the number of active antennas, as well as the EE, which is why it is very important to analyse how the increase in network load affects the activation of the antenna at the base station. BF ULD requires significantly higher activation of antennas Figure 13: EE at different values of output power level of the BS with increasing network load than CF ULD. In both cases of and ULD models. ULD, the highest activation of antennas was for P =40 W. CF ULD achieved the best results when P =20 W and in this case it is necessary to activate an average of 60 antennas 4. CONCLUSION at maximum network load. The best results for BF ULD are achieved, when P =10 W and in this scenario it is necessary Massive MIMO technology is considered crucial for the to activate on average 3 times more antennas at maximum further development of wireless networks, and due to its network load i.e., 185 antennas. characteristics and capabilities, it is the basic technology for achieving the requirements set for the 5G standard Figure 12 shows the results of the optimal number of active and the future 6G standard. In this paper, we investigate BS antennas in the case of CF ULD and both system models resource allocation of a load adaptive massive MIMO (adaptive and fixed). system with different ULD models. Our developed resource allocation scheme is different from the existing The results have the same trend of behaviour, i.e. the largest ones due to the fact that it jointly optimizes the user load M is for P =40 W, and the smallest for P =10 W W for distribution, number of BS antennas and cell loading. opt c c both system models. Increasing the number of users in the network linearly increases the number of optimal antennas, We also determined the impact of cell size, available but in the case of higher network load (100%) a higher M bandwidth and output power lever of the BS at different opt is recorded, compared to the network load of 20%. The cell loading on EE and average UR. When the number of representation of EE for all analysed scenarios with different service antennas in massive MIMO systems is determined, P values is shown in Figure 13. other system parameters can be adjusted (radiation power, number of users, etc.) to achieve higher spectral 8 B&H Electrical E g n e i g, Volume 16, Issue 2, 2022:1-10 n i e r n ISSN:2566-3143, eISSN:2566-3151, DOI: 10.2478/bhee-2022-0012 efficiency at the cost of reduced energy efficiency, and [5] H. A. Ammar, R. Adve, S. Shahbazpanahi, G. vice versa. Boudreau and K. V. Srinivas, “Downlink Resource Allocation in Multiuser Cell-Free MIMO Networks With The results of the analysis confirm that EE increases User-Centric Clustering,” in IEEE Transactions on with decreasing cell size, however, decreasing cell size Wireless Communications, vol. 21, no. 3, pp. 1482- also affects the reduction of antenna array gain. Spectral 1497, 2022, doi: 10.1109/TWC.2021.3104456 efficiency is reduced when the cell size is too small. Also, [6] M. M. Aftab Hossain, R. Jäntti and C. Cavdar, as cell size decreases, transmission power decreases as “Dimensioning of PA for massive MIMO system users will be closer to the base station. In the scenario with load adaptive number of antennas,” 2014 with a bandwidth of 100 MHz, the achieved EE is three IEEE Globecom Workshops (GC Wkshps), Austin, times higher than in the baseline scenario (20 MHz). TX, USA, 2014, pp. 1102-1108, doi: 10.1109/ GLOCOMW.2014.7063580 Also, the results of the proposed adaptive system model [7] M. Ataeeshojai, R. C. Elliott, W. Krzymien, C. are compared with the results of the fixed system under Tellambura and J. Melzer, “Energy-Efficient the same conditions in the network, all with the aim of Resource Allocation in Single-RF Load-Modulated determining the improvement of the system’s performance Massive MIMO HetNets,” IEEE Open Journal of the when the number of BS antennas is adjusted to the Communications Society, pp. 1-1, 2020 network condition. The results showed that the adaptive [8] M. Miriestahbanati, “Resource Allocation in system model achieved better results for EE and average Collocated Massive MIMO for 5G and Beyond”, UR in all scenarios. For all-network loads and for both MS thesis, Concordia University, Montreal, Quebec, system models, adaptive and fixed system, worse results Canada, 2021 for EE and average UR were obtained for the BF ULD case, [9] H. Li, L. Song and M. Debbah, “Energy Efficiency which is a consequence of interference from neighbouring of Large-Scale Multiple Antenna Systems with base stations. Transmit Antenna Selection,” IEEE Transactions on Communications, vol. 62, no. 2, pp. 638-647, 2014 The paper assumes that the CSIs for the receiver are [10] H. V. Cheng, D. Persson, E. Björnson and E. G. perfectly known. However, this assumption may be Larsson, “Massive MIMO at Night: On the Operation unrealistic. An interesting and important problem is to of Massive MIMO in Low Traffic Scenarios,” in Proc. analyse the behaviour of the proposed system model IEEE ICC, London, UK, June 2015 when the CSI data are unreliable. Also, it is very important [11] M. A. Abuibaid and S. A. Çolak, “Energy-efficient to deal with the pilot contamination. Future research massive MIMO system: Exploiting user location should investigate the appropriate design of frequency distribution variation,” AEU – International Journal reuse factors during training and power control to reduce of Electronics and Communications, vol. 72, pp. pilot contamination effect. 17–25, 2017, doi: https://doi.org/10.1016%2Fj. aeue.2016.11.013 [12] J. Y. Cheah and J. M. Smith, “Generalized M/G/C/C state dependent queueing models and pedestrian REFERENCES traffic flows,” Queueing Syst. vol. 15, pp. 365-386, March 1994 [1] M. M. A. Hossain, C. Cavdar, E. Bjornson and R. [13] R. M. Asif, J. Arshad, M. Shakir, S. M. Noman and Jantti, “Energy-Efficient Load-Adaptive Massive A. U. Rehman, “Energy Efficiency Augmentation in MIMO,” 2015 IEEE Globecom Workshops (GC Massive MIMO Systems through Linear Precoding Wkshps), San Diego, CA, USA, 2015, pp. 1-6, doi: Schemes and Power Consumption Modeling,” 10.1109/GLOCOMW.2015.741418 Wireless Communications and Mobile Computing, [2] A. A. Khan, R. S. Adve and W. Yu, “Optimizing 2020, doi: 10.1155/2020/8839088 Downlink Resource Allocation in Multiuser MIMO [14] M. M. A. Hossain and R. Jantti, “Impact of efficient Networks via Fractional Programming and the power amplifiers in wireless access,” 2011 IEEE Hungarian Algorithm,” in IEEE Transactions on Online Conference on Green Communications, Wireless Communications, vol. 19, no. 8, pp. 5162- (Online Conference) Piscataway, NJ, USA, 2011, pp. 5175, 2020, doi: 10.1109/TWC.2020.2990176 36-40, doi: 10.1109/GreenCom.2011.6082504 [3] E. Björnson, L. Sanguinetti, J. Hoydis and M. Debbah, [15] D. Persson, T. Eriksson and E.G. Larsson, “Amplifier- “Optimal Design of Energy-Efficient Multi-User Aware MultipleInput Multiple-Output Power MIMO Systems: Is Massive MIMO the Answer?,” in Allocation,” in IEEE Commun. Letters, vol. 17, IEEE Transactions on Wireless Communications, no. 6, pp. 1112-1115, June 2013, doi: 10.1109/ vol. 14, no. 6, pp. 3059-3075, 2015 doi: 10.1109/ LCOMM.2013.043013.130050 TWC.2015.2400437 [16] N. R. Koppisetti, “Massive MIMO for 5G Wireless [4] A. Salh, L. Audah, Q. Abdullah, N. Shahida, S. A. Hamzah, Networks, An Energy Efficiency Perspective”, et. al, “Energy-Efficient Low-Complexity Algorithm in Vancouver: MS thesis, University of British Columbia, 5G Massive MIMO Systems,” Computers, Materials & Continua, vol. 67, no. 2, pp. 3189-3214, 2021 9 S. Mujkić, S. Kasapović: RESOURCE ALLOCATION AND IMPACT ANALYSIS OF THE SYSTEM PARAMETERS ON PERFORMANCE IN MULTI-CELL MASSIVE MIMO NETWORKS [17] E. Bjornson, L. Sanguinetti, J. Hoydis and M. BIOGRAPHY Debbah, “Designing Multi-User MIMO for Energy Efficiency: When is Massive MIMO the Answer?,” in Samira Mujkić received the B.S., M.S. and Ph.D. Proc. of IEEE WCNC, Istanbul, Turkey, April 2014 degrees in electrical engineering from the University of [18] G. Auer et. al., “D2.3: Energy efficiency analysis of Tuzla, Tuzla, Bosnia and Herzegovina, in 2011, 2014 the reference systems, areas of improvements and and 2022, respectively. From 2012 to 2014, she was target breakdown,” INFSO-ICT-247733 EARTH, ver. a Teaching Assistant and from 2014 to 2017 she was 2.0, (2012), http://www.ict-earth.eu/. Accessed on: a Lecturer with the College of Computer Science and Jun. 13, 2022 Business Communications eMPIRICA, Brcko, Bosnia and [19] M. M. A. Hossain, C. Cavdar, E. Björnson and R. Herzegovina. She currently works at the Government of Jäntti, “Energy Efficiency of Massive MIMO: Coping the Brcko District of Bosnia and Herzegovina. Her main with Daily Load Variation,” 2015, https://arxiv.org/ research interests include radio resource allocation, abs/1512.01998. Accessed on: Jun. 26, 2022 wireless communications, MIMO and massive MIMO [20] S. Mujkic, S. Kasapovic and M. Abuibaid, “Energy- systems. efficient resource optimization for massive mimo networks considering network load,” Computers, Suad Kasapović received the B.S. degree in Electrical Materials & Continua, vol. 71, no.1, pp. 871–888, Engineering from the University of Tuzla, Tuzla, Bosnia 2022 and Herzegovina, in 1996. He then received the M.Sc. and Ph.D. degrees in fields Telecommunications and Informatics at the Faculty of electrical engineering and computing, from the University of Zagreb, Croatia in 2002 and 2007, respectively. He is currently an Associate Professor with the Department of Telecommunications, University of Tuzla. His general research interests include signal processing and wireless communications, with emphasis on MIMO communication systems, practical issues in 5G systems, and wireless sensor network, and etc.

Journal

B&H Electrical Engineering – de Gruyter

Published: Dec 1, 2022

Keywords: energy efficiency; massive MIMO; traffic load; user location distribution

There are no references for this article.