Performance analysis and modeling based on LTE-A field measurements: a city center example

This study analyzes the LTE-A network based on actual field measurements. Measurements were conducted at 80 locations in the Samsun, Turkey, city center, using TEMS Investigation software. RSRP, RSRQ, SINR, CQI, and throughput values were recorded in a stationary position for each measurement while downloading 500 MB of data. The averages of these recorded data were calculated, and detailed analyses were performed. The effects of RSRP, RSRQ, SINR, and CQI on the throughput value in the LTE-A network were examined, and a novel mathematical model was proposed that gives the relationship between them with 88.85% accuracy. It was also observed that the accuracy of the proposed model could be increased by 4% with GRNN.


Introduction
The Long-Term Evolution (LTE) technology has advanced with the 3rd Generation Partnership Project (3GPP) Release-10 LTE-Advanced (LTE-A) technology [1], ushering in a new generation of communication systems that can optimize and autonomously utilize frequency channels [2].
LTE-A technology has continued to evolve with Release-11 [3], and Release-12 [4] was designed to meet users' higher data rate expectations with mobile applications and increase mobile traffic. In Release-11, the frequency band is used in an unordered manner using the non-contiguous intra-band scheme to enhance spectral efficiency. In Release-12, unordered usage coincides with a non-contiguous inter-band system in various frequency bands. Increasing the bandwidth with Release-12 means that high frequencies and signal-tointerference plus noise ratio (SINR) values can be obtained for small cells. Release-12 can achieve up to 750 Mbps downlink (DL) and 150 Mbps uplink (UL) throughput with a bandwidth of up to 100 MHz, using the multiple input multiple output orthogonal frequency division multiplexing (MIMO-OFDM) technology and 256 QAM modulations. LTE-A is a homogeneous network that employs advanced base stations (eNodeBs, eNBs) to cover the entire cell and uses the same transmission power level, modulation technique, and antenna configurations to provide service to user units [5]. The downlink quality between the eNodeB and user equipment (UE) in LTE-A systems is determined by wireless channel parameters. In the LTE-A system, the status of the wireless channel is expressed by the channel quality indicator (CQI). The CQI, determined by the user unit, is usually represented with the SINR value calculated using the reference signal received power (RSRP) and reference signal received quality (RSRQ). There is a direct relationship between the CQI value and throughput (THRP). There are many studies on the performance of LTE and LTE-A systems in the literature. There are many studies on the performance of LTE and LTE-A systems in the literature. In [6], the performance of the LTE cellular system based on field test measurements was evaluated. In [7], RSRP, RSRQ, and reference SINR at various points in space which is covered by an LTE macro base station operating at 2100 MHz were measured. In [8], researchers examined the performance of the automatic CQI reporting algorithm to determine the best approach. In [9], they mapped the downlink SNR to CQI in the Flat Rayleigh channel with fast fading across different transmission modes. In [10], they investigated the feasibility of 256 QAM using SINR geometry and proposed two methods for designing CQI tables to support 256 QAM transmission. In [11], researchers modeled the download throughput of LTE networks. In [12], they designed and implemented a multi-layered performance monitoring system to investigate LTE network performance from the subscriber perspective. In [13], researchers analyzed the performance of a real LTE network through drive tests. In [14], they analyzed practical measurement results from a live LTE network in Australia to evaluate the effects of SNR on throughput and possible relationships among SINR, RSRP, RSSI, and RSRQ. In [15], they analyzed a real-world dense LTE network to study CQI reports. In [16], researchers proposed maximum throughput scheduling for LTE based on CQI feedback. In [17], they proposed a CQI calculation scheme for LTE/LTE-A systems. In [18], various propagation models with path loss for LTE data measurement in urban areas were compared. In contrast to studies in the literature, this study conducted LTE-A network field measurements at 80 locations in the city center of Samsun, Turkey, using the TEMS Investigation mobile network testing software. The RSRP, RSRQ, SINR, CQI, and THRP values were recorded for each measurement location. Detailed analyses were performed to investigate the relationships between RSRP, RSRQ, SINR, and CQI and to evaluate their effects on THRP.

LTE-A technology
The LTE-A system includes several advances, such as increased spectrum efficiency, carrier aggregation (CA), and the utilization of multiple antennas to deliver faster data rates than the LTE version [19]. Users on an LTE-A network can simultaneously use more than one band and combine different bandwidths to increase the efficiency of the CA technique. CA allows for higher data rates in LTE-A because it can provide a significantly broader transmission bandwidth by combining component carriers (CC) in the same or separate frequency bands [20]. Thus, by combining the 1.4/3/5/10/15/20 MHz bandwidths (maximum of five bands), a 100 MHz bandwidth can be used. CA may be efficiently utilized using MIMO techniques that can be increased up to 8 × 8 in the downlink. This method improves spectral efficiency and increases the user's THRP value. The main factors affecting the data rate in an LTE-A network are the distance between transceivers of the line of sight (LOS), signal and noise power, bandwidth, and the number of users. The four essential radio resource management metrics in the LTE-A network are CQI, RSRP, RSRQ, and the received signal strength indicator (RSSI) picked up by the carrier. The wireless channel directly impacts system performance in the LTE-A network, and all users estimate channel state information and transmit it to the eNodeB. CQI is crucial in LTE-A for the eNodeB to decide the corresponding modulation and encoding scheme. CQI contains information regarding the wireless channel's quality, and each UE in the LTE-A network measures the SINR value, scales it to the CQI value, and reports this value to the eNodeB. 3GPP defines the RSRP, RSRQ, and RSSI measurements, whereas UE providers determine SINR. The adaptive modulation and coding (AMC) system of the LTE-A network adjusts modulation techniques based on SINR and CQI values during communication, and the throughput varies correspondingly. SINR determines CQI, which defines the THRP that the UE can support under real-world radio conditions by adopting the modulation scheme. The higher the SINR for a measurement range, the higher the CQI, and thus the higher the THRP value. RSRP and RSRQ are fundamental signal level and quality measures for LTE-A networks. The RSRP theoretically ranges from − 140 dBm to − 44 dBm in 1 dBm intervals. When the UE is as close to the eNodeB as possible, the signal quality is good [21], and the RSRP value is high; when the UE is further away, the signal quality is poor RSRP value is low. RSRP levels for usable signals are typically about − 70 dBm near the eNodeB and approximately − 120 dBm at cell edges. When RSRP is insufficient to find a solid intercellular migration or cell reselection choice, RSRQ measurement gives additional information. RSRQ may accept values ranging from − 3 to 19.5 dB in 0.5 dB intervals. Low RSRQ values indicate no interference, whereas high RSRQ values indicate a high level of interference. Table 1 shows the evaluations of connection quality in the LTE-A network based on the RSRP, RSRQ, and SINR values [22]. Higher efficiency is obtained if the SINR is excellent for a measurement range; on average, RSRP and SINR are related, and the smaller the difference between RSSI and RSRP, the better the RSRQ.

LTE-A field measurements
This study utilized TEMS Investigation [23] to gather data in the city center of Samsun and examine the impact of CQI, SINR, RSRP, and RSRQ channel parameters on the THRP value. Figure 1 provides   shown in Figure 2. Table 2 provides statistical evaluations for RSRP, RSRQ, SINR, CQI, and THRP. As the [25,26] studies considered mean values in their system evaluations, this study also evaluates the average values. The correlation coefficients (CC) between RSRP, RSRQ, SINR, CQI, and THRP were also analyzed and are shown in Table 3. The table shows the highest correlation between CQI and THRP, with a value of 0.8418. The second-highest correlation is between SINR and CQI, with a value of 0.8393.

Measurement results and analysis
Regression analyses were performed on the data to determine the mathematical formulation between THRP and RSRP, RSRQ, SINR, and CQI. For clarity, RSRP is denoted by X 1 , RSRQ by X 2 , SINR by X 3 , CQI by X 4, and THRP by Y. The relationship between X 1 and Y can be expressed using simple regression analysis, as seen in Eq. (1).
The normalized root mean squares error (NRMSE) approach is used to test its accuracy between the estimated where Y n is the real, Ŷ n is the predicted value, n is the index number, and N represents the total number of data.
While the NRMSE between the estimated Y using Eq. (3) and the real Y value is 0.1964, the NRMSE are 0.1686 and 0.1565 respectively for Eqs. (4) and (5). Multiple linear regression analysis was used to estimate the Y value using X 1 , X 2 , X 3 , and X 4 to improve estimation  The NRMSE between the estimated and the measured Y value using Eq. (6) is 0.1155. Utilizing four-channel variables and multiple regression analysis improves prediction accuracy by around 2.5%. Figure 4 shows the variation in measured and estimated THRP levels based on measurement location.
To find a better relationship between THRP and RSRP, RSRQ, SINR, and CQI, the Generalized regression neural network (GRNN) method was applied to data using MATLAB software (MATLAB ® 2020b). GRNN has a fast computational speed and good nonlinear approximation performance [27]. In the simulations, 70% of the data (56) were randomly used for training, and 30% were used for testing (24). This process was repeated 1000 times, and the minimum NRMSE was selected. Table 4 lists the parameters used in the simulation. In this study, only the result of the 0.85 value of the spread of the radial basis function is given because it gives the lowest NRMSE in simulations. As a result of the simulation, the lowest NRMSE value was calculated as 0.0191 for training and 0.0714 for testing. The regression performances of the GRNN method for training, testing, and all data are shown in Fig. 5a-c, respectively.

Conclusion
In this study, real-time throughput analyses were performed on the LTE-A network at 80 locations in the city center of Samsun. A 500-MB file was uploaded to Dropbox for each measurement, and RSRP, RSRQ, SINR, CQI, and THRP values were recorded. The relationship between THRP and RSRP, RSRQ, SINR, and CQI was analyzed and mathematically modeled. The accuracy of the mathematical model between RSRP and THRP is 81.8%, while the accuracy for RSRQ, SINR, and CQI is 80.3%, 83.1%, and 84.3%, respectively. The THRP model using RSRP, RSRQ, SINR, and CQI has an accuracy of 88.85%. To increase the model's accuracy, the GRNN structure was used, and THRP was estimated with an accuracy of up to 92.86%. The proposed models have only been verified for the scenarios we measured. Model accuracies can be increased by using more data from different environments and other ANN structures.
Data Availability Data will be made available on request.