5G NSA Network Slicing

5G NSA Network Slicing

Muhammad Arif Saeed (1), Mohammed Babar Ahmed (2), Noaman Taufiq (3)

Saudi Telecom Company

Abstract – This paper presents the implementation and field trial results of 5G Non-Standalone (NSA) Network Slicing using Nokias Slice-Aware Scheduler in a live commercial 5G network. The trial was conducted across two clusters site ZXXXX and ZYYYY using SPID-based resource dif- ferentiation to separate Smartphone (Non-SPID) and Router (SPID-16) user groups. The scheduler leverages configurable minimum and maximum Physical Resource Block (PRB) quo- tas per resource group to dynamically allocate shared radio resources every 100 ms. Trial results demonstrate consistent improvement in smartphone downlink MAC throughput, with gains of up to 30% in targeted sectors, alongside controlled router throughput within acceptable bounds. Spectral effi- ciency, PRB utilization per slice, and Ookla SpeedTest mea- surements further validate the effectiveness of the approach. Key parameter tuning insights, cluster-specific observations, and recommendations for city-wide deployment are discussed.

Index TermsNSA slicing, 5G NR, slice-aware scheduler, SPID, PRB quota, network resource management, Nokia gNB, QoS differentiation.

1. INTRODUCTION

   The rapid proliferation of heterogeneous device types in 5G Non-Standalone (NSA) networksmost notably smart- phones and Fixed Wireless Access (FWA) routershas introduced new complexities in radio resource manage- ment. Although these devices coexist on the same physical radio layer and contend for the same pool of Physi- cal Resource Blocks (PRBs), their traffic characteristics and Quality of Service (QoS) expectations differ sub- stantially. Smartphones typically generate latency-sensitive, session-based traffic requiring responsive scheduling and consistent throughput, whereas FWA routers produce sus- tained, high-volume data flows that can dominate available radio resources. In the absence of differentiated scheduling mechanisms, router-driven bulk traffic can monopolize PRBs, resulting in degraded performance and diminished user experience for smartphone subscribers.

   To address this imbalance, NSA Network Slicing has emerged as a practical and deployable solution. Nokias Slice-Aware Schedulerimplemented through fea- ture sets CB008132, CB009084, CB008828, CB008746,

   and CB008709enables operators to define Network Re- source Groups (NRGs) with configurable minimum and maximum PRB allocations. This capability allows traffic from distinct device categories to be isolated into logical slices with guaranteed resource quotas, ensuring fair coex- istence and improved service quality across heterogeneous user populations.

   This paper presents results from a live field trial con- ducted across two commercial clusters in Saudi Arabia: ZXXXX and ZYYYY clusters. The study evaluates the impact of NSA slicing on per-slice throughput, PRB uti-

   lization, and end-user experience, with performance vali- dated using Ookla SpeedTest measurements. The findings provide empirical evidence on how slice-aware scheduling can enhance smartphone performance in mixed-device en- vironments while maintaining efficient utilization of radio resources.

2. Background and Related Work

   Network slicing is a core concept in 5G, defined in 3GPP Release 15 and onwards, enabling multiple virtual networks over a shared physical infrastructure. In standalone (SA) architectures, slicing is typically implemented via S-NSSAI identifiers end-to-end from the core to the RAN. In NSA deployments, where the 5G NR radio is anchored to an LTE core (EPC), full SA-grade slicing is not directly applicable. Nokias NSA Slicing approach provides an equivalent RAN-side capability using Subscriber Profile IDs (SPIDs) to identify and differentiate device groups.

   Prior work on RAN slicing has demonstrated that static quota-based approaches can provide isolation but at the cost of spectral efficiency [6]. Dynamic schedulers that share a common resource pool and compute per-group target shares based on instantaneous demand offer a better balance. The Nokia Slice-Aware Scheduler implements such a dynamic common resource pool model, computing target shares every 100 ms.

3. SYSTEM ARCHITECTURE AND FEATURE

   DESCRIPTION

   The NSA Slicing feature is applicable to FR1 (sub- 6 GHz) deployments. The architecture involves the follow- ing object hierarchy at the gNB level:

   • NRBTS: Hosts activation flags and SPID group defi- nitions (NRSPIDGRP).

   • NRRESOURCEGROUPPROFILE: Defines multi-

     ple resource groups per profile, each associated with a SPID group or S-NSSAI identifier.

   • NRRESOURCEGROUP: Carries the scheduling pa- rameters (minQuota, maxQuota) and resource type identifier (resourceDN).

   • NRCELL: References the applicable DL and UL resource group profiles via nrResourceGroupProfileDN and nrResourceGroupProfileUlDN.

   The scheduler divides available PRBs into two logical pools: a dedicated pool (protected minimum quota per group) and a common resource pool (the remainder shared proportionally). For a given cell with three resource groups each having minQuota of 10%, 20%, and 30%, and a common pool of 40%, the target share for each group is computed as:

   ‌TargetShare(RGi) = minQuotai+ maxQuota mi inQuota i ×

   The Cell-level assignment of NRCELL.nrResourceGroupProfileDN and nrResourceGroupProfileUlDN is determined

   CommonPool (1)

   k (maxQuotakminQuotak ) by the maximum UE count observed by NSA per sector.

   Co-sector NRCELLs across all frequency bands (n28, n40, n77, n78) are configured with the same profile to

   This dynamic computation runs every 100 ms, ensuring the allocation adapts to instantaneous cell load conditions. For the example above, target shares are 25%, 33.3%, and 41.7% for RG#1, RG#2, and RG#3, respectively.

4. Parameter Configuration

   Table I presents the key NRBTS-level parameters acti- vated for the trial. SPID group 1 (Smart Phone / Non-SPID) and SPID group 2 (Router / SPID-16) are defined as the two resource groups.

   TABLE I

   NRBTS and SPID Group Parameter Configuration

   ensure consistent scheduling behavior.

5. Trial Results and Analysis

   1. ZXXXX Cluster

      The ZXXXX cluster was used as the initial trial site to validate the slice scheduler behavior. The trial was conducted in three phases: a hybrid phase combining SPID steering with slicing, followed by two progressive slicing phases with tighter router quotas (30/70 then 20/50 min/max for routers).

      Key observations from the ZXXXX trial include: (i)

      1. Smartphone DL streaming throughput showed a clear and consistent upward trend across all three phases.

         Object

         NRBTS

         Parameter

         actSpidBasedSlicingNSA

         Value

         true

      2. Router DL throughput was progressively limited as max quota was reduced, demonstrating effective slice

         NRBTS actSliceAwareScheduler true

         NRBTS actSliceAwareSchedulerUlAndEnh true NRBTS actHighSliceWeightFactor true NRSPIDGRP-1 userLabel / spid Smart Phone / 1

         NRSPIDGRP-2 userLabel / spid Router / 16

         Table II defines three resource group profiles designed for different cell loading conditions. Profile-0 tarets high- user cells (NSA UEs 40) where router resource al-

         isolation.

      3. PRB utilization for SPID-16 (Router) was hard- capped at the configured max quota level, confirming correct scheduler enforcement.

      4. n40 showed elevated Non-SPID user counts due to n78 coverage gaps, creating contention that was effectively managed by the slice quotas.

         location is more tightly capped. Profile-1 is applied to low-to-medium load cells (UE <40), with relaxed router maxQuota and a higher smartphone minQuota to im- prove user experience. Profile-2 provides a custom baseline for cells requiring case-by-case tuning.

         TABLE II

         Resource Group Profile Quota Configuration and Target Shares

         Profile Resource min% max% Target%

         Group

   2. ZYYYY Cluster N-KPI Analysis

      The ZYYYY cluster was the primary production valida- tion site. P(S)Cell Load Balancing was activated alongside NSA Slicing to rebalance users across frequency layers prior to applying slice quotas. Table III summarizes the key N-KPI changes observed post-implementation.

      ‌TABLE III

      ZYYYY CLUSTER N-KPI SUMMARY (PRE VS. POST)

      Profile- 0

      (High UE

      40)

      NRRG- 1

      (Smart- phone) NRRG- 2

      50 100 68.75

      20 50 31.25

      Throughput

      KPI			PRE	POST
      DL	MAC	UE			+5.2%

      DL Traffic Volume +1.2% NSA Avg Users (n78) 45.9%

      NSA Avg Users (n40) +18.2%

      (Router) Avg DL MAC SDU 104.7 Mbps 184.2 Mbps +75.9%

      1 1 (Low/Med (Smart-				PDCCH CCE Starva-	0.06%	0.00%	98.2%
      UE phone) <40) NRRG-	20	80	36.36	PDSCH Block Rate	0.39%	0.02%	95.5%
      (Router)				DL CA Reconfig At-			+320%
      Profile- NRRG- 2 1 (Custom) (Smart-	50	100	61.54	tempts DL CA Reconfig SR	99.61%	99.88%	+0.27%

      Profile-

      NRRG-

      2

      60 100 63.64 UE Thr (n78)

      tion Rate (Fmt 3)

      phone) NRRG- 2

      (Router)

      20 100 38.46

      Per-band DL throughput analysis showed that smart- phones consistently outperformed routers as expected. For n77/n78 cells with NSA UE count 40, updating the router

      maxQuota to 60% and smartphone minQuota to 60% (Profile-1) further improved both smartphone and router throughput simultaneously, with no degradation observed for any user group. For n40 and n28 bands, the more conservative max quota of 50% for routers was maintained without throughput degradation.

      PRB utilization per slice confirmed proper quota en- forcement: NRG-2 (Router) was capped at its configured maxQuota, while the mean PRB utilization tracked the computed target load values from (III). AEQD collocated sites showed elevated n40 user counts as a deliberate outcome of the P(S)Cell load balancing strategy, with acceptable throughput impacts.

   3. ZYYYY Cluster Pre/Post Analysis

      The ZYYYY cluster was activated after the baseline. The deployment sequence included: (1) Activating NSA Slicing,

      (2) Optimizing low-user n78/n77 [7] cells in the 60/100, 20/60 Profile-1 configuration, (3) Changing a test configuration, and (4) Activating the exhaustive scheduler as a complementary optimization. Throughput analysis showed improvement in smartphone DL and UL throughput across all frequency bands (n28, n40, n77, n78) [3]. Router throughput remained within acceptable bounds despite the tighter quota enforcement.

      Post period analysis confirmed that overall smartphone throughput continued to improve after the peak traffic period, with Ookla DL and UL speedsshowing an upward trend. The PRB utilization tracking followed themax/min quota settings precisely, validating the schedulers deter- ministic enforcement behavior.

   4. Ookla SpeedTest S-KPI Validation

   Ookla SpeedTest measurements providedindependent end-user validation of network improvements. Across both ZXXXX and ZYYYY clusters, 5G DL and UL speeds showed clearimprovement trends across POST-1, POST-2, and POST-3 measurement periods compared to PRE. Modern chipset performance was maintained throughout, confirming that the slicing configuration did not introduce regression for premium devices. These results support the proposal for a broader city-wide cluster deployment.

6. DISCUSSION

   The trial results confirm that NSA Slicing with the Slice- Aware Scheduler is an effective mechanism for differenti- ating service quality between smartphone and router user groups in live 5G NSA networks. Several practical insights emerged:

   • Profile selection based on cell UE count (threshold: 40 UEs) provides a simple and effective adaptive policy. High-user cells benefit from tighter router caps, while low-user cells can afford more generous router quotas.

   • The dynamic common resource pool computation ensures high spectral efficiency by avoiding static partitioning. Unused quota from one group is available to the other within the max quota constraints.

   • AEQD collocated sites require careful P(S)Cell load balancing configuration to prevent unintended n40 concentration. The ccLoadThreshold parameter on n28 was a critical tuning lever.

   • The event high-traffic period demonstrated scheduler robustness under peak load conditions, with PRB quota enforcement remaining precise.

   • Future enhancements including 26R1 (CB012238 QoS and UE fairness) and 25R3 (CB012810 low-latency slice scheduling) are expected to further improve differentiation quality for latency-sensitive applications.

7. CONCLUSION

This paper has presented a comprehensive field trial of NSA Network Slicing in a commercial 5G network [7]. The slice-aware scheduler successfully enforced PRB quota constraints per resource group while maintaining high spectral efficiency through a dynamic common pool model. Smartphone users experienced consistent throughput im- provements of 576% across bands and clusters, while router users were managed within their quota bounds with-out service disruption. Ookla SpeedTest® measurements independently validated the improvements at the end-user experience level. The findings support the recommendation for city-wide deployment of NSA Slicing with per-profile tuning, and highlight the importance of iterative quota optimization as traffic patterns evolve.

REFERENCES

1. 3GPP TS 38.300, NR; NR and NG-RAN Overall Description, Release 16, 2020.

2. ‌Nokia, Slice Aware Scheduler Feature Description, Nokia Net- works, 2023. (CB008132, CB009084, CB008828, CB008746, CB008709).

3. Mohammed Yahiya Pasha Gulam 2L-4H Beams Antenna, IJERT Volume 15, Issue 05 , May 2026

4. X. Zhang et al., Dynamic Resource Slicing for 5G NR Networks: A Survey, IEEE Communications Surveys & Tutorials, vol. 22, no. 4, 2020.

5. Mohammed Yahiya Pasha Gulam Performance Analysis of 5G N71 Depoyment in Live Network Environments, IJERT Volume 15,

   ‌Issue 04 , April 2026

6. ‌O. Simeone et al., Slice Isolation and Efficiency Tradeoffs in Radio Access Network Slicing, IEEE Transactions on Wireless Communications, 2018.

7. Mohammed Yahiya Pasha Gulam Performance Evaluation of N40 Integration Using AZNA in 5G Networks, IJERT Volume 15, Issue 04 , April 2026