Real-time measurements and performance analysis of closed-loop MIMO service for mobile operators

: As fifth generation (5G) networks are starting to become commercial, user expectations in terms of new services become high as well. This signifies that mobile communications service providers need to build robust 5G new services as quickly and cost-eﬀiciently as possible. Many new technologies rely on closed-loop (CL) and multiple input multiple output (MIMO) technologies due to emerging cooperation between nodes in next generation networks. In this paper, we first compare different multiantenna transmission modes namely: transmit diversity, open-loop (OL), and CL MIMO spatial multiplexing strategies to provide mobile network operator (MNO) services in terms of their characteristics, ,limitations and benefits. Later we investigate how launching a large-scale CL MIMO deployment strategy can affect the various key performance indicators (KPIs) of the existing services provided by Mobile Network Operators (MNOs) in real-operational network infrastructure in Turkey. Our practical experimental results indicate that, compared to OL MIMO system, CL MIMO can achieve large performance on a practical setup, where up to 3% improvement in cell average throughput, 9% in user throughput, 6% in spectrum eﬀiciency, and 9% in channel quality indicator (CQI) and modulation coding scheme (MCS) are obtained, while reduction by 25% and 17% on sum delay and initial block error rates (IBLER) are observed.


Introduction
The Fifth Generation (5G) networks are expected to provide low latency, massive connectivity, and high throughput to diverse set of applications and devices. One of the important features of 5G networks will be to support various subscriber services [1]. Techniques such as Multiple-input multiple-output (MIMO) provide significant increase on the effectiveness of the mobile transmission that reduces the radio interference across the cell. Thus, better user experience for MNOs services can be achieved as a result. Millimeter-wave (mmWave) based 5G service can be affected by the surrounding environment, such as building shadowing, reflection, rain attenuation, etc. This delay and noise sensitivity would make providing feedback on Closed-Loop (CL) systems more critical in terms of their coverage enhancement abilities. The primary requirement to enable CL systems is the capability of end-user devices to support the system. Mobile communication systems allow the usage of channel information to implement CL techniques between user equipment (UE) and Base Station (BS).
The information in this channel can be utilized to increase the system's effectiveness without a requirement * Correspondence: engin.zeydan@cttc.cat This work is licensed under a Creative Commons Attribution 4.0 International License. of complex receiver architecture. Moreover, lossless and reliable data transmission with CL systems is also important for some next generation services that are mission-critical and cannot tolerate a single packet loss.
MIMO systems contain multiple antennas to improve the connectivity of UEs. To run MIMO algorithms effectively, a significant coordination support is required between BSs and UEs. In order to utilize all the capabilities of MIMO technology, two types of mechanisms are defined in long term evolution (LTE)-advanced and 5G. These are Open-Loop (OL) and CL MIMO schemes. OL MIMO has been used in the literature in combination with different schemes such as Downlink (DL) multi-user Sparse Code Multiple Access (SCMA) (MU-SCMA) or coordinated multi-point (CoMP) transmission for Ultra Dense Network (UDN) [2]. In time and frequency response of the spatial channel, there are variations that affect the performance of the channel. CL MIMO method leverage the channel feedback information to track these variations accurately [3]. CL MIMO systems are under the focus of industry for 5G communications. 5G New Radio (NR) data transmission is expected to utilize CL MIMO dynamic pre-coding scheme as one way to improve the spectral efficiency [4]. 5G Stand Alone (SA) [5] beam management technique requires CL MIMO system to determine the best beam from the UE point of view. CL MIMO solution operates in a dynamic manner, since the mobility of the UE changes in the coverage area. For this reason, it is essential to investigate the performance gains of CL MIMO solution in pre-5G deployments.

Related work and motivation
In the literature, many studies on the enhancements and the performance gains for CL and OL based MIMO systems exist. CL and OL based single user MIMO solutions have been studied in combination with Non-Orthogonal Multiple Access (NOMA) in [6]. Closed loop spatial multiplexing (CLSM) scheme is envisioned to be utilized in various 5G applications including UDNs, massive MIMO, and mmWave (and/or terahertz)-based BSs [7]. In [8], a general expression for the probability density function is proposed to derive the system Bit Error Rate (BER) and ergodic capacity in CL for MIMO systems with performance results where it has been shown that ideal CL MIMO provides a 2 dB theoretical performance link gain corresponding to 20% higher spectrum efficiency over OL MIMO. The article in [9] examined the effect of the transmit antenna correlation on the CL throughput of a 2 × 2 CL MIMO system. Potential learning schemes to achieve and exceed performance of existing MIMO schemes and their performance evaluations using both OL and CL operations in MIMO systems are presented in [10]. The paper in [11] provides a detailed performance comparison between CL and OL MIMO schemes based on system level simulation results and show 2 dB link gain (corresponding to 20% spectral efficiency) theoretical improvements in ideal conditions and 1 dB (corresponding to 10% capacity gain in fully loaded network scenario) when practical constraints are considered. In [12], the impact of different cellular network deployments, antenna configurations, and transmission schemes on achievable performance are examined through theoretical and simulation studies for multiantenna heterogeneous networks (HetNets).
System-level simulations are conducted in [13] with the specified aspects mostly affect the capacity and spectrum efficiency of LTE network which also includes CL MIMO. CL MIMO is shown to provide valuable signal-tointerference-plus-noise ratio (SINR) gains compared to OL MIMO via simulations even under Channel Status Information (CSI) feedback error scenarios in [14]. However both of these approaches ( [13,14]) are based on system level simulations that lack experimentation with real-world operational environments inside a MNO infrastructure. There are also system-level studies for MIMO in the literature. A system level analysis on user-centric scheduling for a flexible 5G radio design with using CL MIMO is presented in [15]. OL and CL training systems are proposed in [16] as a framework that is using successive channel prediction/estimation at the user for Frequency Division Duplexing (FDD). Relationship between beam-forming and MIMO techniques is investigated in [17]. The authors in [18] design a precoding matrix using wider beamwidths in both CL and OL MIMO transmission schemes to enhance the throughput values for 5G new radio (NR) systems. An iterative algorithm based on an alternating minimization procedure using interference alignment for MIMO is proposed in [19]. A method that dynamically configures the transmission parameters for multiple MIMO streams is proposed in [20]. CL MIMO system is extended for 5G networks in patent [21] so that each receiver is adapted to transmit at least one type of feedback information selected from a antenna selection group. Another CL MIMO system patent in [22] is proposing variations on the feedback information, so the payload may be configured to 6 bits including a Precoding Matrix Index (PMI) or 4 bits representing a PMI and 2 bits representing a differential SINR. A detailed comparison of CL and OL MIMO in LTE is given in [11]; however, the analysis results are based on system level simulations. The patent in [23] is proposing a method for dynamic switching between CL and OL MIMO for multistream processing. The authors in [24] investigate the experimental performances of vehicular throughput of different transmission modes (TMs) (including transmit diversity, OL, and CL spatial multiplexing transmission) in different carrier frequencies and MIMO configurations modes. The authors report that TM-2 mode is shown to be utilized mostly due to its robustness against mobility and poor channel conditions during field measurements of a residential district in İstanbul, Turkey. The paper in [25]  The rest of the paper is organized as follows: Section 3 presents the system model and concepts related to different MIMO deployment strategies as well as comparisons of OL and CL MIMO strategies to provide MNO services. The experimental results are presented in Section 4 as well as provides discussions on main takeaways issues that need to be considered. Finally, in Section 5, we provide the conclusions and future work of the paper.

System model and concepts
We assume that there are K l UEs with N transmit and receive antennas in each cell l ∈ L with where L = {1, 2, . . . , L} is the set of cells and L is the total number of cells in the considered geographic region. The channel between UE-k in cell l is denoted by h l l,k which is a N × 1 vector. In OL MIMO transmission scheme (e.g. in LTE TM-3 mode), the received DL signal y l,k ∈ C at UE-k in cell l is given by , where BS l transmits the signal x l = ∑ K l i=1 s l,i and receiver noise n l,k = N C (0, σ 2 ). The received SINR at user , where P k,l is the transmitted power from BS l ∈ L to UE-k. In CL MIMO transmission scheme (e.g. in LTE TM-4 mode), the received DL signal y l,k ∈ C at UE-k in cell j is given by , where BS l transmits the signal , where w j m,i is the pre-coding vector to cancel the interference at cell l for UE-k. The throughput of a cellular network in a given area which is measured in bps/km 2 can be be calculated as , where B is the bandwidth, D is the average cell density, and SE is the per-cell Spectral Efficiency (SE) which represents the amount of information transferred per second over a unit bandwidth [27]. In practical BS deployments, when BS receives the SINR values, it first maps it into CQI and later into spectral efficiency using 3GPP specification tables (see Table 7.2.3-1 in [28]). Finally it loops through MCS indexes to find the best Transport Block Size (TBS)-MCS pair that can approximate the obtained spectral efficiency and maps an MCS index into a TBS (see Table 7.1.7.1-1 in [28]) during one Transmission Time Interval (TTI).
3GPP specification has also worked on defining propagation models for urban channels. For urban areas which of interest in our experiments, 3GPP has defined macro-cell propagation model [29]: , where R is the base station-UE separation in kilometres, f is the carrier frequency in MHz, Dhb is the base station antenna height in metres, measured from the average rooftop level. The UE transmit power P l,k from , where P max is the UE maximum transmit power, R min is the minimum power reduction ratio to prevent UEs with good channels to transmit at very low power level, CL is the path coupling loss defined as max{path loss-G T x -G Rx , MCL}, where path loss is propagation loss plus shadowfading, G T x is the transmitter antenna gain in the direction of the receiver, G Rx is the receiver antenna gain in the direction of the transmitter, MCL is minimum coupling loss (selected to be 70 dB in macro cell urban areas), 0 < γ < 1 is the balancing factor for UEs with bad channel and UEs with good channel, and CL x−ile is the x-percentile CL value. [29].  In UEs with a RI value of 1, the same data streams flow over different antennas. In this case, the throughput is expected to be less, while the data loss due to radio conditions is also expected to be less. TM3 and TM4 MIMO provide higher peak throughput when using rank 2 that allows two code words on two antennas with spatial multiplexing.  CSI which is an indicator how good or bad the channel is at a specific time. Based on these reports, eNodeB determines the transport block sizes to send the data, which in turn can be directly converted into throughput. PMI is used to combine the two signals transmitted by the two antennas of eNodeB. The PMI value is measured and reported by UE. This UE based PMI determination method is more suitable for UEs than the one that is Both CL and OL MIMO spatial multiplexing schemes support rank 1 and rank 2 transmissions. In case where rank 1 is used, both schemes become similar to transmit diversity. In rank 2 transmission, codewords are transmitted on multiple antennas by using spatial multiplexing. Under certain conditions (e.g at high SINR), spatial multiplexing can enhance the transmission data rate. During practical operation, first initial random access (RA) transmit diversity is used towards UEs. After the reception of Radio Resource Control (RRC) connection setup message by UEs, OL or CL MIMO with spatial multiplexing is used. The transmission scheme and correspondingly the CSI content is adjusted using RRC signaling.

Experimental results
We performed experiments for monitoring and comparing CL and OL MIMO between 02 August 2018 and Partnership Project (3GPP). All the experiments were done in urban areas which corresponds to urban macro model in [29]. The system scenario is set for the FDD system parameters and FDD coexistence scenario as specified in [29]. The selected experiment models and the experiment configurations of carriers for transmitters are configured as defined in [32].        Finally, Figure 7 shows the overall KPI improvements after CL MIMO is activated in different cities of Turkey. We can observe obtained gains in terms of DL user and cell throughput, spectrum efficiency, CQI values, and modulation usability increase in 64 QAM and 16 QAM in different locations of Turkey. The highest increase has been on the usage of 64 QAM, whereas the lowest percentage gains are on cell DL throughput values. On the other hand, QAM usage by BSs has been reduced dramatically by around 40% in all cities. Figure 7 shows that depending on the CQI increments, the highest user DL throughput was experienced in Antalya. The reason for this situation can be explained by the fact that the experimental feature test was Main observations and takeaways: CL MIMO is expected to be utilized in combination with different evolving technologies in 5G networks. For this reason, it is essential to observe its potential benefits in comparison with OL MIMO scheme. Note that CL MIMO is a mechanism used to continuously adapt the transmitted signal to suit the channel characteristics, whereas, in OL MIMO, the communications channel does not utilize explicit information regarding the propagation channel. Hence, the success of CL MIMO scheme depends on the quality of channel estimation as the UE measures the channel to send reports back to BS. On the other hand, potential communication system impairments such as transmit & receive phase differences don't affect the performance of OL MIMO scheme. During the observation period of our practical experimental setup, overall improvements have been observed in major LTE KPIs including CQI, 64 QAM usage in DL, UE & cell throughput, spectral efficiency, packet delay, and IBLER values after switching MIMO scheme from OL to CL spatial multiplexing in the considered major cities of Turkey. CL MIMO is generally more efficient than OL MIMO in low mobility conditions where user's radio conditions don't change rapidly. This signifies that majority of the UEs inside the considered sites are in low mobility conditions. However, in locations where high mobility exists (e.g. in highways), it may be more suitable to proceed with OL MIMO scheme. Moreover, in specific locations where high amount of traffic is available, communication overhead of CL MIMO can deteriorate the obtained gains in terms of the considered KPIs. Therefore, an adaptive approach where appropriate switching capability between OL and CL MIMO transmission scheme can be necessary depending on the hour of the day (e.g. on peak or low hour traffic) or the characteristics of the locations (e.g. residential, business, shopping area, airport, etc.). On the other hand, the improvements in throughput values were also not too high (around 3% to 9%). This signifies low rank usage choice of BSs in CL MIMO scheme. Moreover, uncalibrated antennas can also cause for the differences in codeword-qualities. Theoretically, CL MIMO has been shown to provide up to 2dB theoretical link gain corresponding to up to 20% theoretical spectrum efficiency gain compared to OL MIMO schemes as demonstrated in [8] in ideal conditions. On the other hand, our experimental results have indicated that up to 6% for spectrum efficiency gain has been be achieved with CL MIMO scheme under real operation scenarios. This can be due to nonideal environments where many environmental factors, such as density of the cell, mobility, spatial and temporal patterns of user traffic, BS and UE antenna configurations, etc. as well as impairments related to differences in received signal's amplitude, timing/frequency offsets, and phase noise in CL MIMO at the BS site that have degraded the performance gains.
Note that OL transmit diversity can yield better performances at low SINR regimes, whereas OL & CL spatial multiplexing MIMO schemes can work better in high SINR regimes (as also demonstrated with system level simulations in [11]. Therefore, depending on the optimal SINR point, using transmit diversity or spatial multiplexing schemes need to be decided based on the KPI measurements. This can be beneficial for cities such as İzmir, which is observed to have the lowest MCS index and CQI values. Note also that potential 5G services that can be provided by mmWave (e.g. at 28 Ghz) spectrum can be affected by the surrounding environment, such as building shadowing, reflection, rain attenuation, etc. [33]. However, our frequency ranges of eNodeBs used throughout the experiments were on the order of 800 Mhz and 1800 Mhz. Therefore, the effect of such environmental factors, such as atmospheric conditions are not having a big impact on the performance of CL MIMO systems.

Conclusions
In this paper, we have run real-world experiments on OL and CL MIMO strategies. We have observed different KPIs that are of interest to MNOs. First, we have described the characteristics, limitations, and benefits of both MIMO transmission schemes. Then, we have run practical experiments to demonstrate the performance gain of CL MIMO in major and the most crowded cities of Turkey. Based on the obtained experimental results, CL MIMO is shown to outperform 3% OL MIMO for cell average throughput, 9% for user throughput, 6% for spectrum efficiency, 9% in CQI and MCS as well as 25% and 17% reductions on sum delay and IBLER. Therefore, CL MIMO is demonstrated to be a strong practical candidate technique for future wireless networks via a practical experimental set-up. A possible future study area can focus on the reducing the computational burden at the UE introduced by the CQI, RI, and PMI reporting via alternative channel estimation models such as channel transfer functions that are characterizing end-to-end performance. Additionally, an adaptive switch scheme between different MIMO modes, which can leverage the strengths of OL transmit diversity, OL & CL spatial multiplexing MIMO depending on radio environment, and observing their impact on large-scale nationwide deployments are of interest.