Multivariate Performance Evaluation of CPRI based BBU Feed-In for Indoor Active DAS

Multivariate Performance Evaluation of CPRI based BBU Feed-In for Indoor Active DAS

Syed Muhammad Ali Moazzam Pasha

Mobile Services Provider, KSA

ABSTRACT

Cellular deployments in multilayered dense sharing environ- ments can be challenging from power and infrastructure foot- print requirements. One of the most pressing issues in Hybrid Indoor deployments is the realization of a sustainable and share- able Distributed Antenna System (DAS) with an optimized wattage, thereby limiting the carbon footprint and also en- abling a commercially viable infrastructure viz a viz improved spectral eciency and higher payload delivery per watt and acreage. A direct CPRI interface with the processing units in a typical topology circumvents the need for Radio Transceivers before the head end, which helps in alleviating high power consumption and space. This however requires benchmarking performance, most notably the eects of Cascaded Error Vec- tor Magnitude (EVM) and Uplink (UL) Noise since these are some of the most fundamental Physical layer determinants for payload delivery and spectral eciency. This paper intends to evaluate the eects of these two based on some readily available performance KPIs and employ statistical techniques to view the impact for direct CPRI interface at a higher level of abstraction.

1. INTRODUCTION

   The rapid growth of indoor mobile trac and high-capacity venues such as airports, hospitals, stadiums, and commercial complexes have signicantly increased the demand for high- performance Indoor Building Systems (IBS). The distributed antenna system (DAS) has become the preferred solution for providing seamless indoor coverage and higher capacity; how- ever, the large number of active network elements required in conventional active and hybrid DAS deployments has also resulted in increased energy consumption, operational ex- penditure (OPEX) and infrastructure complexity. As mobile operators continue to pursue sustainable networks, reducing the energy footprint of indoor radio access infrastructure has become an important research and operational objective. But most importantly, with increased complexity, the number of cascaded nodes and active components has led to increased dis- tortions and noise dened by Error Vector Magnitude (EVM) and Noise Rise Values (NRV). Considerable research has fo- cused on improving the energy eciency of base stations, but this paper investigates the impact of Direct BBU Feed-In as a means of improving system spectral eciency and radio link quality due to improved EVM and UL noise. While measuring the EVM is costly and requires specialized instruments, proxy performance indicators can help in benchmarking and this pa- per provides an Machine Learning based approach to ascertain whether direct BBU feed-in solutions circumvents these issues in link quality.

   Fig. 1. Power Consumption distribution in different types of Base Stations.

   1. Cellular Energy Consumption

      As illustrated in Fig. 1 , the Power Amplier (PA) constitutes the largest single contributor to base station energy consumption, accounting for approximately 47% of the power consumption in micro base stations and 36% in pico base stations. Although active and hybrid Indoor Building System (IBS) architectures are not identical to standalone micro or pico base stations, their radio front-end architectures share many common functional blocks, including RF transceiver chains, power amplication and baseband processing. Consequently, the power consump- tion breakdown of micro and pico base stations provides a useful approximation of the relative contribution of the PA stage in active indoor radio equipment [1].

   2. Hybrid DAS Eective Power Requirement

      Unlike outdoor cellular deployments, where high RF transmit power is required to overcome large propagation losses and provide wide-area coverage, hybrid Active DAS primarily functions as a signal distribution network. Radio Transmit power required at the DAS interface remains signicantly lower than that of a conventional outdoor base station because the radio signal is converted into low power optical signal with input power requirement usually around 0 dBm. Consequently, the use of a high-power amplication stage represents a substantial architectural overhead for many IBS deployments.

   3. Direct BBU Feed-In

   This observation provides the motivation for the proposed Di- rect BBU Feed-In architecture, in which the distributed antenna network is driven directly from the centralized Baseband Unit (BBU), thus eliminating the intermediate PA stage traditionally employed within the Active DAS architecture. Figures 2 and 3 depict the two architectures showing that the complete Power Ampliers- Radio stage is removed in the direct BBU feed-in architecture. The CPRI specication [2] provides a framework

   Fig. 2. Conventional RF Feed-In DAS Architecture.

   EVM has a direct impact on receiver Signal-to-Interference- plus-Noise Ratio (SINR). Although EVM is fundamentally a modulation quality metric rather than a power ratio, the distor- tion represented by EVM behaves as an additional interference component that reduces the spectral eciency of the system. Under the assumption that distortion sources are uncorrelated with the desired signal, the relationship can be approximated by

   Fig. 3. CPRI Based BBU Feed-In DAS Architecture.

   for this implementation but details of actual implementation are subject to original equipment manufacturers (OEMs) interfaces and are beyond the scope of this paper. Nevertheless the pro- posed architecture has the potential to substantially reduce the overall energy consumption of IBS deployments. But beyond energy savings, the removal of an active amplication stage reduces the introduction of nonlinear distortion [5], oering the potential for improved uplink Noise and Error Vector Mag- nitude (EVM). These advantages make Direct BBU Feed-In an attractive architecture for next-generation energy-ecient indoor wireless networks, particularly as operators seek to reduce the operational expenditure while also evaluating the complete RF chain rather than an individual PA in isolation [4].

2. RADIO LINK IMPAIRMENTS : EFFECTS OF

   CASCADED ACTIVE NODES

   1. Cascaded EVM

      Error Vector Magnitude (EVM) is a fundamental performance metric that quanties the modulation accuracy of a wireless transmitter or receiver by measuring the deviation between the received modulation symbols and their ideal constella- tion points. In practical radio systems, EVM represents the cumulative eect of RF impairments introduced throughout

      SINRe 1 , (1)

      EVM2

      where EVM is expressed as a linear quantity rather than a percentage. Consequently, increasing EVM reduces the eective SINR available for demodulation, even when the received signal strength remains unchanged.

      The reduction in eective SINR directly impacts the achiev- able spectral eciency of LTE and NR systems. According to Shannons capacity relationship,

      = log2 (1 + SINR), (2)

      where represents the achievable spectral eciency in bits/s/Hz. A decrease in SINR due to higher EVM there- fore reduces the maximum achievable throughput for User Equipment (UE). In practical LTE systems, this manifests as the scheduler selecting a lower Modulation and Coding Scheme (MCS), transitioning, for example, from 256-QAM to 64-QAM or from 64-QAM to 16-QAM in order to satisfy Block Error Rate (BLER) requiremens. The reduction in modulation order and coding rate subsequently decreases user throughput and cell capacity.

      This relationship is particularly signicant for Active Dis- tributed Antenna Systems (DAS), where the transmitted signal is cascaded through multiple active RF stages before reaching the distributed antennas. Each additional amplier, frequency conversion stage, or active remote unit contributes incremental distortion and noise, producing what may be termed cascaded EVM. Although no universal analytical expression exists for cascaded EVM, an empirical alternative could be the RSS of EVM for all stages.

      EVMtotal = /EVM2 + EVM2 + EVM2 + ·· ·. (3)

      the signal chain, including power amplier (PA) non-linearity,

      1 2 3

      phase noise, IQ imbalance, local oscillator instability, quantiza- tion noise, thermal noise, and frequency-dependent distortion. Since modern cellular systems employ cascaded RF architec- tures comprising multiple active and passive components, the overall EVM reects the aggregate distortion introduced by each stage.

      Recent work [4] demonstrated that the EVM generated by a non-linear power amplier cannot be considered independently from the surrounding RF network. Their system-level analysis showed that when a non-linear PA is cascaded with a generic linear network, such as impedance matching circuits, lters, couplers, or distribution networks, the resulting EVM depends on the combined response of the entire RF chain rather than the PA alone. Their experimental wideband load-pull mea- surements further demonstrated that impedance mismatch can signicantly increase the measured EVM over wide modula- tion bandwidths, highlighting the importance of evaluating the complete transmission chain rather than individual components in isolation.

      For cellular systems such as LTE and NR, degradation in

      The cumulative impairment increases the eective distortion observed at the receiver, reducing SINR and ultimately limiting the achievable throughput.

   2. Cascaded Uplink Noise

   The Uplink (UL) noise oor of a wireless communication sys- tem represents the minimum level of unwanted electrical noise present at the receiver input that limits the detection of weak signals. In practical cellular radio systems, the receiver noise oor is not determined by a single component but rather by the cumulative contribution of all active and passive elements within the receive chain. Low Noise Ampliers (LNAs), du- plexers, lters, mixers, variable gain ampliers (VGAs), cables, couplers, repeaters and distributed antenna equipment each introduce additional thermal noise and attenuation, collectively increasing the overall receiver noise gure (NF). This cumula- tive degradation is commonly referred to as the cascaded noise gure or cascaded noise oor, and is fundamentally described by the Friis cascade equation [6].

   For a receiver comprising cascaded stages, the overall noise

   factor is expressed as

   2 1

   3 1

   1

   TABLE 1. Signicance of Performance Indicators and Expected Impact of Direct BBU Feed-In

   total = 1 +

   1 + 12 + ··· + [l 1 , (4)

   KPI Layer Expected Impact Primary RF Mech-

   =1

   anism

   where and denote the noise factor and gain of stage , respectively. The Friis equation demonstrates that the rst few stages of the receiver chain dominate the overall system noise gure, emphasizing the importance of low-noise front-

   DL/UL SINR Physical Reduced cascaded EVM and receiver noise

   Average CQI Physical Improved signal

   end design. Conversely, passive losses occurring before the rst amplication stage directly increase the receiver noise

   DL/UL BLER Link

   quality

   Improved demodu-

   lation and decoding

   gure because attenuation degrades both signal power and signal-to-noise ratio before amplication.

   DL/UL Packet Loss

   Link Fewer HARQ re- transmissions and

   In LTE and 5G cellular systems, an increase in cascaded noise oor directly reduces the eective receiver Signal-to-

   DL Spectral E- ciency

   MAC/PHY

   decoding failures

   Higher MCS selec-

   tion

   Interference-plus-Noise Ratio (SINR). As the cascaded noise oor increases, weaker user signals become increasingly dif-

   DL/UL Through- put

   User Higher spectral ef- ciency and lower

   cult to distinguish from the background noise, particularly for users located near the cell edge or within high-loss indoor environments. The reduction in eective SINR forces the scheduler to select more robust but lower-order Modulation and Coding Schemes (MCS), reducing the achievable spectral eciency and ultimately decreasing user throughput.

   Average UE Headroom

   Uplink

   retransmissions

   Improved uplink

   sensitivity and lower required UE transmit power

   This relationship is well established in LTE receiver design. For example, [7] demonstrates cascaded noise gure analysis for an LTE receiver front-end consisting of duplexers, LNAs, mixers, lters and variable gain ampliers. The analysis shows how each stage contributes to the cumulative receiver noise gure, which directly determines receiver sensitivity and carrier-to-noise ratio. The same methodology is subsequently used to calculate the minimum detectable LTE signal level, illustrating the direct relationship between cascaded noise gure, receiver sensitivity and communication performance.

   The impact of cascaded noise is particularly relevant in Indoor Building Systems (IBS) employing Active or Hybrid Distributed Antenna Systems (DAS). According to Friis cas- caded noise equation, the overall receiver noise gure depends not only on the noise gure of each stage but also on the gain of the preceding stage. Passive insertion losses introduced by splitters, couplers, duplexers, and attenuators directly increase the eective system noise gure because they attenuate the re- ceived signal before subsequent amplication. By eliminating the high-power RF transmission stage and feeding the Active DAS directly from the BBU, RF attenuation and associated cascade losses are reduced and thus improving receiver sensi- tivity and uplink SINR. Conversely, the cumulative eect of higher receiver noise oor reduces the uplink SINR and results in increased uplink Block Error Rate (BLER), lower Channel Quality Indicator (CQI), reduced spectral eciency and lower user throughput. Removing one of the primary sources of RF non-linearity, the PA stage, reduces the accumulated EVM while simultaneously lowering thermal noise generation.

3. BENCHMARKING FRAMEWORK

   To quantify the impact of the proposed Direct BBU Feed- In architecture on indoor cellular network performance, a rudimentary performance evaluation framework was developed using operational Key Performance Indicators (KPIs) collected from commercial Indoor Building System (IBS) deployments. The study compares indoor sites employing the proposed Direct BBU Feed-In architecture against conventional hybrid Active DAS deployments utilizing the legacy BBURF feed-in architecture. Since the proposed architecture fundamentally

   modies the RF signal chain by eliminating the intermediate power amplier stage, the evaluation focuses on network-level performance indicators that directly reect improvements in signal quality and user experience.

   A. Layer 1 Metrics Overview

   Physical layer or Layer 1 metrics such as Downlink Spectral Eciency and Uplink (UL) Signal-to-Interference-plus-Noise Ratio (SINR), Average Channel Quality Indicator (CQI), UE Power Headroom, in addition to other performance indicators such as actual Cell Average Throughput and Average CQI, are used to quantify improvements in radio signal quality resulting from reductions in cascaded Error Vector Magnitude (EVM) and noise oor. Link-layer performance is also evaluated using Downlink and Uplink Block Error Rate (BLER). These metrcs characterize the reliability of the radio link and provide an indication of the quality of Radio Link

   Collectively, these KPIs enable the proposed architecture to be evaluated from multiple complementary perspectives.

4. STATISTICAL METHODOLOGY AND DATASET

   Statistical analysis focused on rst establishing the ground truth about supposed improvements in the performance of Direct BBU feed-in systems in comparison to the legacy RF Feed-in systems. Then the approach is to classify the two solutions in relation to the performance and visualize how distinct the two systems are. For this study a total of 52 cells were selected with 28 cells with a working direct BBU feed in solution to the DAS. All cells had comparable trac and connected users, two factors which can inuence the KPIs. The data was based on an hourly resolution with 9563 independent observation for the 52 cells with each data point consisting of 18 performance measurements such as UL/DL Cell throughput, DL Spectral Eciency, UL SINR, UL/DL BLER etc.

   A. Linear Mixed-Eect Model

   To establish a statistical baseline prior to the multivariate anal- ysis, a Linear Mixed-Eects Model (LMM) was employed to quantify the inuence of the deployment architecture on the selected downlink and uplink KPIs. The model accounts for

   TABLE 2. Mixed-Eects Model Results

   KPI Architecture Users

   	Eect			Eect
   DL Spectral Eciency	+0.6113	3.88 × 104		+0.0496	< 1016
   DL Throughput	+5.4606	2.06 × 106		+0.4165	< 1016
   DL BLER	+0.0335	0.0380		-0.00110	0.300
   Average CQI	-0.1261	0.704		+0.00776	5.85 × 104
   UL SINR (PUSCH)	+4.863	6.79 × 1015		+0.1423	< 1015
   UL Throughput	+2.0355	0.0246		+0.0208	0.726
   UL BLER	+0.1228	0.0192		+0.00020	0.955
   Average UE Headroom	-2.1286	0.0949		+0.0237	0.0206

   the repeated hourly observations collected from each cell by treating cell identity as a random eect, while deployment architecture (0 = legacy BBU-RF Feed-In, 1 = Direct BBU Feed-In) and connected users were modeled as xed eects. Consequently, the estimated architecture coecient represents the average performance dierence between the two deploy- ment architectures after accounting for variations in trac load and cell-specic characteristics.

   = 0 + 1 + 2 + + , (5)

   Fig. 4. First 3 Principal Components explain 90 Percent Variance

   and classication of the architectural dierences.

   A. Multivariate Performance Analysis

   To complement the mixed-eects analysis, a multivariate work- ow was used to investigate whether the selected KPIs collec- tively formed distinct operating clusters associated with the

   where

   denotes the deployment architecture,

   represents

   two indoor deployment architectures. Prior to dimensionality reduction, multivariate outliers were identied and removed

   the Average Connected Users, is the random intercept

   associated with cell , and is the residual error.

   B. Univariate Performance Results

   The preliminary mixed-eects analysis indicates that Direct BBU Feed-In is primarily associated with improvements in Radio performance metrics. After adjusting for dierences in trac load, statistically signicant gains are observed in DL Spectral Eciency (+0.61 bps/Hz), DL Throughput (+5.46 Mbps), UL SINR (+4.86 dB), and UL Throughput (+2.04 Mbps). These improvements persist after accounting for con- nected Users, suggesting that the observed performance gains cannot be attributed solely to dierences in network loading.

   Conversely, Average CQI and Average UE Headroom do not exhibit statistically signicant dependence on deployment architecture, while both DL and UL BLER require further investigation since the estimated direction of the architecture coecient does not fully align with the improvements observed in throughput and SINR. These results suggest that the principal impact of Direct BBU Feed-In is reected in downlink and uplink payload delivery capability while secondary metrics such as BLER, CQI and UE headroom not exhibiting any statistically signicant dependency. These results are compiled in Table 2

5. MULTIVARIATE PERFORMANCE ANALYSIS

   While the mixed-eects model provides statistical evidence of the independent inuence of the deployment architecture on individual KPIs, it evaluates each performance indicator separately. Cellular network performance is inherently mul- tidimensional, with capacity, radio quality, reliability, and trac load interacting simultaneously. So the next stage of the analysis extends beyond individual KPI comparisons and em- ploys multivariate techniques to investigate the combined eect on these indicators. This enables the identication of latent cavrying dimensions and performance patterns associated with Direct BBU Feed-In and conventional BBU-RF Feed-In deploy- ments. This provided a more comprehensive understanding

   using robust statistical criteria. A total of 324 observations (3.23%) were classied as outliers and excluded from further analysis. The relatively small proportion of discarded samples indicates that the dataset was largely representative of normal network operation while eliminating observations that could disproportionately inuence the latent feature extraction.

   B. Principal Component Analysis

   Following outlier removal, Principal Component Analysis (PCA) was applied to the standardized KPI dataset compris- ing downlink spectral eciency, downlink throughput, uplink throughput, uplink SINR, average CQI, and UE headroom. PCA was employed to transform the correlated KPI space into a reduced set of orthogonal latent variables while preserving the majority of the information contained in the original mea- surements. The rst three principal components accounted for the overwhelming proportion of the total variance and were therefore selected for subsequent visualization and clustering. The resulting three-dimensional latent space revealed two par- tially overlapping but visibly shifted groups, suggesting that the two deployment architectures exhibit dierent multivari- ate operating characteristics despite sharing common trac conditions and propagation environments.

   C. Gaussian Mixture Model

   In Fig. 5, the multivariate KPI observations exhibited overlap- ping and approximately bimodal behavior. Thus a Gaussian Mixture Model (GMM) was selected to identify the underly- ing performance regimes. Unlike hard-partitioning methods such as K-means, GMM allows observations to have prob- abilistic classication in multiple clusters and can therefore accommodate the observed overlap between the two regimes. The bimodal structure further suggested that a two-component GMM was an appropriate model for characterizing the under- lying performance populations.

   To objectively identify these operating regimes without us- ing any prior knowledge of the deployment architecture, an unsupervised Gaussian Mixture Model (GMM) was trained

   Fig. 5. Original Clusters for RF and BBU Feed-In

   Fig. 6. AIC-BIC criterion for clusters count

   using the rst three principal components. The otimal number of Gaussian components was determined using the Akaike In- formation Criterion (AIC) and Bayesian Information Criterion (BIC), both of which exhibited an elbow at two components. Fig. 6, indicating that a two-cluster solution provided the best balance between model complexity and goodness-of-t. This validates the original two architecture explainability derived from the mixed eects linear model.

   The resulting GMM partitioned the observations into two predominant multivariate operating regimes as shown in Fig

   7. Although the clusters exhibited partial overlap, reecting the common operating environment of indoor LTE networks, a clear association with the deployment architecture was ob- served. Approximately 84% of the legacy RF Feed-In obser- vations were assigned to the corresponding operating regime, whereas 63% of the Direct BBU Feed-In observations were assigned to the second regime. The higher classication rate for RF Feed-In suggests that its performance characteristics are comparatively homogeneous, while the lower classication rate for Direct BBU Feed-In indicates greater variability in operating conditions. This behaviour is expected, as Direct BBU Feed-In sites encompass a broader range of trac loads and deployment scenarios, resulting in a wider spread in the latent KPI space.

   Overall, the multivariate analysis provides independent evi- dence that the two deployment architectures occupy system- atically dierent regions of the KPI feature space. Unlike the mixed-eects models, which quantify the impact of de-

   Fig. 7. Clustering based on GMM

   Fig. 8. Higher MCS Values for BBU Feed-In

   ployment architecture on individual KPIs after controlling for trac load, the PCA-GMM framework demonstrates that these KPI improvements collectively manifest as distinct multivari- ate operating regimes. Consequently, the clustering results reinforce the statistical ndings of the mixed-eects models and indicate that the performance gains associated with Direct BBU Feed-In are not conned to isolated KPIs but represent a broader shift in overall network operating behaviour.

6. MCS DISTRIBUTION ANALYSIS

   The MCS distribution analysis provides a more granular view of the link-adaptation behavior of the two architectures. Across the 32 downlink MCS bins and 29 uplink MCS bins, the DL distributions show only subtle dierences between RF Feed-In and Direct BBU Feed-In, with no clear observable shift in the distribution of pegged values. In contrast, the UL distribution exhibits a more distinct dierence, Fig 8, with Direct BBU Feed-In showing a higher proportion of observations in the higher MCS bins. This shift is consistent with the higher UL SINR observed in the mixed-eects analysis and suggests that the improved uplink radio conditions associated with Direct BBU Feed-In may enable the scheduler to sustain more ecient modulation and coding levels.

7. INFERENCE

   1. RF Mechanism

      The observed performance dierences can be interpreted in terms of the RF architectures. Conventional RF Feed-In

      requires high-power RF transmission followed by attenuation and distribution through multiple RF components, introducing additional insertion losses. Direct BBU Feed-In reduces the dependence on this high-power RF chain and associated attenuation, providing a plausible explanation for the observed improvement in UL SINR and DL spectral eciency, as validated by mixed eects model. Since cascaded EVM and UL noise were not directly measured in this study, these mechanisms should be regarded as engineering interpretations rather than directly measured eects.

   2. Link Adaptation

      The improvement in radio conditions associated with Direct BBU Feed-In is expected to inuence LTE link adaptation through the selection of MCS levels. The observed increase in UL SINR and DL spectral eciency suggests more favorable radio conditions for transmission. Analysis of the complete MCS distribution, rather than only the dominant MCS 29 bin, therefore provides a means of determining whether these improvements translate into greater utilization of higher MCS levels. The observed eect was more pronounced in UL where BBU Feed-In showed better performance than legacy RF Feed-In solution.

   3. Network Performance

      The mixed-eects results demonstrate that Direct BBU Feed-In is associated with improvements across several network-level KPIs after accounting for trac load and cell-specic variabil- ity. In particular, Direct BBU Feed-In increased DL spectral eciency by 0.611 bps/Hz, DL throughput by 5.46 Mbps, UL SINR by 4.86 dB, and UL throughput by 2.04 Mbps. The PCA-GMM analysis further showed two partially overlapping performance regimes, with 84% of RF Feed-In and 63% of Direct BBU Feed-In observations associated with their respec- tive dominant regimes, providing complementary evidence of architecture-dependent performance characteristics.

   4. Engineering Implications

   The results indicate that Direct BBU Feed-In can provide both network-performance and architectural benets by reducing reliance on high-power RF amplication and associated RF dis- tribution components. The measured improvements in SINR, spectral eciency, and throughput suggest that simplifying the RF chain does not compromise network performance and may provide measurable gains. In addition, removing unnec- essary active RF stages has the potential to reduce equipment power consumption, thermal load, and space requirements. Direct power, EVM, and cascaded-noise measurements are nevertheless required to quantify these benets independently.

8. LIMITATIONS AND FUTURE WORK

   While the eects of variability due to trac load and concurrent users have been factored into the mixed-eects model, Indoor DAS deployments can have dierent response to factors like network density, trac proles, clutter, proximity to outdoor Macro sites, ingress of outdoor Macro RF power etc. This can directly have an impact on performance which can diminish the eect of gains from architecture types. Ideally, benchmarking should be done on the same site in tandem but this can be challenging from deployment and prohibitive from commercial point of view. While results in a lab environment can be isolated from the external factors, system response to trac and load gradient is possible only with live trac.

9. CONCLUSION

   Multivariate analysis presented in this study provides a frame- work for evaluating Direct BBU Feed-In against legacy BBU- RF Feed-In architectures using operational KPIs. The mixed- eects results indicate statistically signicant improvements in DL spectral eciency, DL throughput, UL SINR, and UL throughput after accounting for trac load and cell-specic variability. Analysis based on GMM further indicates that the two architectures occupy distinctly dierent, although partially overlapping, multivariate performance regions. MCS distribu- tion analysis provides a further mechanism for investigating the radio-quality improvements in relation to LTE link adapta- tion. Direct measurements of power consumption, EVM, and cascaded noise remain necessary to independently quantify the underlying RF and energy mechanisms.

10. ACKNOWLEDGEMENT

Direct CPRI feed-In solutions are deployed by dierent DAS manufacturers and exact implementation varies for every ven- dors commercial product. However, benchmarking framework has been developed independently based on performance met- rics developed by cellular services provider for a Tier-1 ICT OEM supplier. The author has used AI-assisted tools for language editing and document structuring.

REFERENCES

1. A. Gati et al., Key technologies to accel- erate the ICT Green evolutionAn opera- tors pint of view, arXiv:1903.09627, 2019, doi: 10.48550/arXiv.1903.09627.

2. CPRI Cooperation, CPRI Specication V6.0: Common Public Radio Interface (CPRI); Interface Specication, Aug. 30, 2013. [Online]. Available: http://www.cpri. info

3. Y. Josse, B. Fracasso, G. Castignani, and N. Montavont, Energy-ecient deployment of Distributed Antenna Sys- tems with Radio-over-Fiber link, Reference details to be completed from the veried source.

4. G. P. Gibiino, A. M. Angelotti, A. Santarelli, and

   P. A. Traverso, Error vector magnitude measurement for power ampliers under wideband load impedance mismatch: System-level analysis and VNA-based im- plementation, Measurement, vol. 187, Art. no. 110254, 2022, doi: 10.1016/j.measurement.2021.110254.

5. E. Uthirajoo, H. Ramiah, J. Kanesan, and A. W. Reza, Wideband LTE Power Amplier with Integrated Novel Analog Pre-Distorter Linearizer for Mobile Wireless Communications, PLOS ONE, vol. 9, no. 7, Art. no. e101862, 2014, doi: 10.1371/jour- nal.pone.0101862.

6. H. T. Friis, Noise Figures of Radio Receivers, Proceed- ings of the IRE, vol. 32, no. 7, pp. 419422, July 1944, doi: 10.1109/JRPROC.1944.232049.

7. T. J. Rouphael, Wireless Receiver Architectures and De- sign: Antennas, RF, Synthesizers, Mixed Signal, and Dig- ital Signal Processing. Oxford, U.K.: Newnes/Elsevier, 2014.