Comprehensive Analysis of Cellular Uplink Performance in a Dense Stadium Deployment S.M. Haider Ali Shuvo∗‡ , Hardani Ismu Nabil† , Joshua Roy Palathinkal∗ , Muhammad I. Rochman∗ , and Monisha Ghosh∗ ∗ University of Notre Dame, USA, † Sebelas Maret University, Indonesia.
I. I NTRODUCTION Modern 5G cellular networks provide substantial improvements in downlink performance, but their uplink capacity remains a significant limitation [1], [2]. In the US, most newly deployed 5G bands use time division duplexing (TDD) [3]. In these bands, base stations benefit from high transmit power and larger downlink slot allocations. Conversely, user equipment (UEs) are restricted by limited transmit power and a very small fraction of uplink slots, which limits uplink capacity. At the same time, uplink demand is rapidly increasing due to applications such as live streaming, content creation, and video conferencing [4]. Consequently, commercial 5G cellular networks often fail to meet user uplink expectations, especially during periods of high congestion [5]–[7]. Modern stadiums, in particular, present a major example of “uplink crunch” for Mobile Network Operators (MNOs), with
31-32
6-7
28-29
9-10
27-28
10-11
24-25
13-14
18-19
Abstract—Uplink performance remains a critical limitation in modern 5G networks, where UEs have to balance limited transmission power against propagation challenges. We conducted extensive measurements in the University of Notre Dame’s football stadium, which has a seating capacity of 80,000 spectators, evaluating network behavior under both unloaded (pregame) and severely congested (game day) conditions, with a focus on uplink performance. Analyzing PHY-layer metrics captured via the Rohde & Schwarz QualiPoc, we show that high-frequency TDD bands in the uplink are severely bottlenecked in both the spectral and temporal domains. Despite transmitting near maximum 3GPP power limits, propagation loss inherent to highfrequency bands restricts UEs to low MCS indices and low PRB allocations, even in unloaded networks. This inability to achieve wideband allocation is further compounded by the significantly smaller number of uplink slots compared to downlink slots in TDD frames. Consequently, we observe a severe disparity between uplink and downlink: while high-frequency TDD bands carry the majority of downlink throughput, the network relies heavily on lower-frequency FDD bands for uplink. Additional measurements under favorable propagation conditions around a Verizon COW deployment located in the stadium parking lot also show that this limitation is not solely propagation-driven; rather, the duplexing scheme itself also plays a significant role. Even when TDD bands achieve higher or comparable MCS, FDD bands have a performance edge in the uplink due to the restrictive, downlink-heavy TDD architecture. These findings emphasize the indispensable role of low-frequency FDD spectrum in sustaining uplink capacity, providing insights that will help guide the design of next-generation wireless networks. Index Terms—Uplink performance, 5G wireless Networks, Duplexing scheme, TDD, FDD
19-20
arXiv:2604.04371v1 [cs.NI] 6 Apr 2026
‡ Corresponding Author Email: [email protected]
(a) Target sections
Section 31-32 vomitorium
Approximate measurement locations
(b) Spots per section
Fig. 1. Measurement spots at the stadium seating bowl.
Verizon's Cell on Wheels
Fig. 2. Measurement spots surrounding Verizon’s Cell on Wheels (COW) located in stadium parking lot.
user densities often exceeding 50,000 active users per square kilometer [8]. During games, many users engage in activities such as sharing videos, uploading content, video chatting, and live streaming, which severely stress the uplink [9]. Motivated by this challenge, we conducted extensive measurements at the University of Notre Dame football stadium during game time and in an empty stadium during the pregame period to better understand network performance in such scenarios. In addition, we collected and analyzed data around a Cell on Wheels (COW) deployed by an MNO in a parking lot outside the stadium, an approach operators increasingly use to handle high traffic demand [10]. Through the analysis of these measurements, we make the following contributions: • We present comprehensive real-world measurements comparing 5G performance in the congested stadium bowl and around a COW deployment in the parking lot. • We show that uplink performance in high-frequency
TABLE I C APTURED LTE AND NR PARAMETERS FROM Q UALI P OC . Parameter Latitude, Longitude PCI RSRP/ RSRQ DL/UL ARFCN PDSCH/ PUSCH Throughput DL/UL MCS DL/UL RB PUSCH Tx power
Description UE’s geographic coordinates calculated by the Android API Physical Cell Identifier Signal strength values. For NR, RSRP/RSRQ indicates measurements from the 5G Synchronization Signal (SS) block. Downlink/Uplink Absolute Radio Frequency Channel Number, i.e., center frequency. Throughput values recorded on the Physical Downlink Shared Channel and Physical Uplink Shared Channel, i.e., PHY-layer downlink and uplink throughput. Modulation and Coding Scheme utilized in DL and UL channels. Number of allocated Resource Block per Subframe (LTE) or Slot (NR) in DL and UL. Uplink Transmit power used by the UE.
5G TDD bands are severely limited by a combination of poor signal quality, reflected in low modulation and coding scheme (MCS) values, spectral constraints in which UE power limits large physical resource block (PRB) allocations, and a fundamental temporal bottleneck arising from the lack of available uplink slots in the frame structure. As a result, a larger fraction of uplink traffic is carried over the frequency division duplexing (FDD) anchor in commercial non-standalone (NSA) deployments. • Using measurements collected around the COW, where high-frequency TDD carriers achieve high MCS, we show that even when signal propagation is not a constraint, the uplink still relies heavily on the FDD carrier, indicating that slot allocation disparity is a major bottleneck. We believe these findings provide insights for improving 5G network design and optimizing spectrum management toward better cellular uplink performance. In addition, by analyzing how the interplay of frequency-dependent propagation and duplexing schemes impacts UE transmit power and performance, this work supports the goals of green communication to achieve more energy-efficient network configurations. II. DATA C OLLECTION M ETHODOLOGY Extensive measurements were conducted in the seating bowl of Notre Dame Stadium, which accommodates approximately 80,000 spectators during each game. Data were collected during two football games on Sep. 13 & 20, 2025, denoted as “Game Day”, and, as a baseline, before the season started on Aug. 26 & 28, 2025, denoted as “Pregame”. The data collection was performed using commercial smartphones equipped with Rohde & Schwarz QualiPoc [11], a professional network measurement tool that logs low-level protocol events and key performance indicators (KPIs) by decoding Qualcomm modem signaling messages. A summary of the metrics captured by QualiPoc is provided in Table I. Across all measurement days, we adhered to a standardized data collection protocol at each location: (i) initialize a QualiPoc recording while stationary, (ii) execute two consecutive Ookla Speedtests, and (iii) terminate the
TABLE II S UMMARY OF LTE AND 5G NR C ARRIER C ONFIGURATIONS Operator
Band
Duplex
Verizon
n77 n71 n25 n41 n5
TDD FDD FDD TDD FDD
b13 b5 b66 b2 b66 b2 b12 b66 b2 b30
FDD FDD FDD FDD FDD FDD FDD FDD FDD FDD
T-Mobile AT&T
Verizon
T-Mobile
AT&T
Freq. (MHz) UL DL NR Bands 3700, 3800 677 631 1894 1974 2500, 2600 828 873 LTE Bands 782 751 840 885 1720 2120 1882 1962 1747 2147 1875 1955 709 739 1735 2135 1860 1940 2310 2355
BW (MHz)
# PCI PG† GD†
∆PL (dB)
100 20 20 100, 90 10
4 0 2 17 21
10 1 23 37 23
14.8 0.0 9.0 11.3 1.7
10 10 20 5 15 10 10 10 20 10
23 10 23 23 2 1 28 25 27 24
24 24 29 21 23 23 20 19 24 24
1.3 1.9 8.1 8.9 8.2 8.8 0.40 8.2 8.8 10.7
† PG and GD refer to pregame and game day, respectively. ∆PL denotes
the free space path loss of the corresponding band relative to n71 uplink (677 MHz), calculated as 20 log10 (fU L /677).
recording before relocating. Although Ookla Speedtest captures download, upload, and latency metrics, we omit the latency analysis in this paper due to space constraints. During the two game days, crowd-induced movement restrictions limited measurements to the stairways between seating sections. Consequently, we designated ten target sections and assigned two specific measurement spots per section: one near the vomitorium and another at the upper row. Fig. 1a illustrates the designated stairways, while Fig. 1b shows the measurement spots within a representative section. In contrast, during the pregame phase, the absence of crowd restrictions allowed us to increase the number of spots while keeping them uniformly spaced. Across each measurement day, we recorded approximately 20 to 25 minutes of data per carrier. To accommodate high traffic demand during game time, Verizon deployed a COW in the stadium parking lot. In addition to the bowl measurements, we collected data in the accessible areas surrounding the COW during a game on Oct. 4, 2025. The COW features a three-sector deployment operating with a b66 primary carrier and a n77 secondary carrier. Due to site accessibility, our campaign was limited to two of these sectors, corresponding to Physical Cell Identifiers (PCI) 345 and 347 for the LTE and NR bands. The measurement footprint and the COW location are shown in Fig. 2. Our measurement setup consisted of two Samsung Galaxy S22+ smartphones and one Samsung Galaxy S24+ smartphone operating on the AT&T, T-Mobile, and Verizon networks. All devices used premium, unthrottled data plans to capture peak throughput capabilities. We distributed the SIM card assignments across the devices on different measurement days to eliminate hardware bias. Due to UE hardware limitations, our setup did not support uplink carrier aggregation (CA) between two NR bands. Instead, uplink CA was limited to aggregation between the LTE anchor and a single 5G NR band in NSA mode. In SA networks, such as T-Mobile, the uplink relied entirely on single-carrier allocation without aggregation.
PUSCH Transmit Power: Pregame vs. Game Day Verizon (NSA) T-Mobile (SA)
30
3700/3800 MHz 100 MHz BW
25
782 MHz 10 MHz BW
840 MHz 10 MHz BW
1720 MHz 20 MHz BW
1882 MHz 20 MHz BW
Median 25.0 Median 25.0
1894 MHz 20 MHz BW
25
Median 20.2
10 5 0 5
MCS Index
Tx Power (dBm)
20
15
1894 MHz 20 MHz BW
Median 26.0
Median 23.0
T-Mobile (SA)
1882 MHz 20 MHz BW
2500/2600 MHz 100/90 MHz BW Median 23.0
20
PUSCH MCS: Pregame vs. Game Day Verizon (NSA)
30
782 MHz 10 MHz BW
1720 MHz 20 MHz BW
Pregame Game Day
2500/2600 MHz 100/90 MHz BW
3700/3800 MHz 100 MHz BW
840 MHz 10 MHz BW
15
10
Median 18.0
Median 10.4 Median 12.8
Median 20.3
5 Median 20.6
Pregame Game Day Power Class 3 (23 dBm) Power Class 2 (26 dBm)
0
Median 4.6
Median 9.5
Median 7.3
Median 5.2
Median Median 4.5 3.3
Median 4.9
Median 7.8 Median 3.7
b13 b5 b66 b2 n77 n25 n41 Primary,FDD Primary,FDD Primary,FDD Primary,FDD Secondary,TDD Primary,FDD Primary,TDD
b13 b5 b66 b2 n77 n25 n41 Primary,FDD Primary,FDD Primary,FDD Primary,FDD Secondary,TDD Primary,FDD Primary,TDD
(a) Uplink Transmit Power
(b) Uplink MCS
Fig. 3. Uplink transmit power and MCS distributions across different bands under unloaded (pregame) and loaded (game day) conditions. Note the heavy concentration of transmit power near the 3GPP limit for high-frequency TDD bands (Verizon n77, T-Mobile n41), despite which the Uplink MCS remains severely degraded. Bands displaying only a right-sided violin indicate that the band that was only detected during game day conditions, with no corresponding data points during the pregame measurements.
III. R ESULTS The data collected during the two game days and the pregame period are analyzed to evaluate uplink network performance and quantify the inherent uplink-downlink asymmetry. The findings are presented in the following five subsections. Unless otherwise stated, the ‘Game Day’ label represents the combination of the data from both the first and second games, and all presented results correspond specifically to data collected during active speedtest sessions. A. Deployment Summary Verizon and AT&T maintained 5G Non-Standalone (NSA) connections throughout all measurement periods, while T-Mobile transitioned from exclusively Standalone (SA) mode during pregame to a roughly equal distribution of SA and NSA operation on game day. Only minimal instances of LTE-only operation were recorded for the operators. Table II details the complete list of deployed bands active during game day, where the prefixes “n” and “b” indicate 5G and 4G bands, respectively. During the pregame, we observed limited spectrum utilization: Verizon served only b2 and n77 to our UEs, and T-Mobile utilized only n41. This stark contrast in active bands between the two phases highlights the severe spectrum crunch induced by game day user density. AT&T’s deployment remained similar in both conditions and had no C-band coverage in bowl area. Table II also reports the number of unique per-band PCIs for pregame and game day. We observe a high PCI density on some bands during the game day, in particular TMO n25, n41, b66, and b2. This strongly indicates the use of active Distributed Antenna Systems (DAS) and/or small cells to address extreme capacity demands during the game. On the other hand, we observe a comparatively low number of Verizon
n77 PCIs, which indicates lower number of small cells and/or passive DAS deployment. B. Uplink Power Consumption and Channel Quality We analyze uplink transmit (Tx) power and uplink MCS distributions under empty pregame and packed game day conditions, as illustrated in Fig. 3a and Fig. 3b. For brevity, we focus this analysis on the Verizon and T-Mobile MNOs. Across both time periods, we observe that the high-frequency, wider TDD 5G bands (Verizon n77 and T-Mobile n41) utilize significantly higher transmit power than the lower frequency, narrow FDD bands (b5, b13, b66, b2, and n25). Specifically, T-Mobile’s n41 consistently operates near the 3rd Generation Partnership Project (3GPP) Power Class 3 limit of 23 dBm, while Verizon’s n77 approaches the Power Class 2 limit of 26 dBm [12]. However, despite this higher power consumption, the median attainable MCS at these higher frequency bands remains considerably lower than that of the lower-frequency bands during both pregame and game day measurements. This indicates that the severe propagation loss inherent to highfrequency bands remains limiting even outdoors,thereby restricting the uplink to low MCS near the maximum permissible transmit power limit. C. Uplink-Downlink Asymmetry in High-Frequency Bands To evaluate the disparity between uplink and downlink, we analyze a single representative PCI observed on game day. We compare its MCS and PRB allocation ratio (defined as the ratio of allocated PRBs to the total available PRBs in that band) distributions across representative low- and highfrequency bands for both Verizon (b5 and n77) and T-Mobile (n71 and n41). Fig. 4 presents these distributions via split violin plots, where the left and right halves of each distribution denote the downlink and uplink, respectively. The upper subplot, detailing
MCS Distribution
25
Verizon
Downlink Uplink
T-Mobile
20
Pregame AT&T Allocated PRB / Total PRB
the MCS distribution, illustrates that while the base station successfully maintains a relatively high downlink MCS across all bands, the uplink MCS degrades severely in the higherfrequency bands (n77 and n41) under the same propagation conditions and the same PCI.
80
90%
5
Median 14.6 Median 9.1
Median 8.4
b5,10MHz PCI 367
120
Median 8.9
Median 1.8
n77,100MHz PCI 110
Median 9.3
Median 10.3
n71,20MHz PCI 551
n41,100MHz PCI 46
PRB Allocation
Verizon
Median 1.5
Downlink Uplink
T-Mobile
100 80
0
65% 34%
20
n5 10MHz
Median 40%
Median 25%
b5,10MHz PCI 367
Median 75%
Median 24%
n77,100MHz PCI 110
Median 64%
Median 64%
n71,20MHz PCI 551
Median 93%
Median 28%
n41,100MHz PCI 46
Verizon
48%
71% 50%
29% n5 10MHz
n71 20MHz
n25 20 Hz
43%
31%
Downlink Median Uplink Median n41 n77 100MHz 100MHz
Fig. 5. Median difference between downlink and uplink PRB allocation ratios across pregame (upper) and game day (lower) scenarios. The substantial gap in high-frequency bands between downlink and uplink even in the unloaded network highlights the inability of the UE to access the wideband high-frequency channel due to poor signal propagation and transmit power constraints.
Fig. 4. Comparison of downlink and uplink performance metrics on game day across identical PCIs for T-Mobile and Verizon’s representative low- and high-frequency bands.
Verizon Uplink Carrier Aggregation Pregame vs Game Day Percentage of Uplink Captures
Game Day Pregame
100.0%
100
Similarly, the lower subplot shows a significant reduction in the uplink PRB allocation ratio for the high-frequency bands. These observations highlight a fundamental asymmetry in the link budget: while the base stations utilize their higher transmit power capability to counter high-frequency path loss in the downlink, the power-constrained UE fails to achieve a high MCS despite operating near its maximum permissible transmit power limit, as shown in §III-B. As a result, uplink performance in higher-frequency bands are severely bottlenecked compared to downlink, consistently failing to achieve high MCS or substantial resource block allocations. Fig. 5 illustrates the difference in median PRB allocation ratio between the downlink and uplink across the different 5G frequency bands of the three carriers during pregame and game day. We observe that this allocation gap is substantially larger for the high-frequency bands. Notably, this asymmetry persists even under unloaded pregame conditions, indicating that the uplink limitation is fundamentally power-limited rather than congestion-driven. As established in §III-B, the UE operates near its maximum 3GPP-allowable transmit power while still acheiving only a low MCS. Since these speedtest measurements were obtained in an empty stadium network, where radio resources are theoretically unconstrained, the reduced PRB allocation cannot be attributed to scheduler congestion. Instead, it indicates that the UE lacks sufficient total transmit power to sustain transmission across the full wide bandwidth of high-frequency carriers. To compensate
89%
64% 63%
40 20
T-Mobile
87%
80 60
n41 100MHz
Downlink Median Uplink Median n77 100MHz
Game Day
100
0
20
96%
40
60 40
Verizon
90%
AT&T
5 0
T-Mobile
60
10
Allocated PRB / Total PRB
MCS Index
96%
0
15
Allocated PRB/Total PRB
100
80 60 47.0%
40
29.6%
20 0 b13(P) + n77(S)
12.3%
b5(P) + n77(S)
b2(P) + n77(S)
Carrier Combination
11.1%
b66(P)+n77(S)
Fig. 6. Histogram of Verizon uplink carrier combinations during pregame and game day. The distribution shows that during game day when network is congested and requires high capacity, the utilization of sub-1 GHz LTE anchors (b5, b13) becomes substantially more frequent.
for the severe propagation loss, the available uplink power is concentrated over a small subset of PRBs, thereby maintaining a viable uplink connection at low MCS in the high-frequency bands. D. Prevalence of Lower-Frequency FDD Bands in Uplink Throughput To assess whether higher-frequency TDD bands contribute proportionally to the overall uplink throughput despite their higher transmit power and wider bandwidth, we first examine Verizon’s uplink performance in 5G Non-Standalone (NSA) mode. As depicted in Fig 6, during the congested game day scenario, the network relies heavily on sub-1 GHz carriers (b5,
Uplink throughput contributed by 5G Carrier
Verizon - Downlink and Uplink Throughput per Band (Game Day)
20.0
1.0
Downlink Uplink
17.5 15.0
0.6 0.4 0.2 0.0
0
20
40
Pregame, b2(P)+n77(S) Median: 28.9% Game Day, b5(P)+n77(S) Median: 16.7% 60 80 100
Throughput over 5G/Total throughput
Throughput (Mbps)
Empirical CDF
0.8
12.5 10.0 7.5 5.0 2.5 0.0
Median 0.16
Fig. 7. CDF plots illustrating the percentage of uplink throughput delivered by the Verizon 100 MHz, n77 carrier during pregame and game day.
b13 (10 MHz)
Median 0.85
Median 4.37
b5 (10 MHz)
Median 11.46
Median 0.58
n77 (100 MHz)
Fig. 8. Downlink and uplink throughput distributions across b13, b5, and n77 bands during game day. The high-frequency TDD n77 carrier carries the vast majority of downlink traffic but fails to provide substantial uplink capacity, whereas the lower-frequency sub-1 GHz FDD bands (b5, b13) perform significantly better.
T-Mobile (NSA) - Uplink Throughput(Game Day)
1.0 0.8 Empirical CDF
b13) as primary LTE anchors, accounting for 76.6% of anchor usage, whereas during the pregame, only b2 (1882 MHz) served as the primary LTE anchor. This shift highlights the role of low-frequency FDD bands in sustaining uplink transmission under power-constrained conditions during periods of high user density. Furthermore, our measurements indicate that Verizon’s 5G NSA uplink performance depends heavily on the lower-frequency FDD bands. As shown in Fig. 7, for an uplink CA configuration combining b5 (uplink 840 MHz, 10 MHz BW) and n77 (3700/3800 MHz, 100 MHz BW), the median 5G throughput contribution (5G throughput / total CA throughput) remains below 20% during game day, even though n77 operates with a 6 dB higher median transmit power (Fig. 3a) and ten times the channel bandwidth. A similar trend is observed in the unloaded pregame stadium environment: when aggregating b2 (uplink 1882 MHz, 20 MHz BW) and n77, the median n77 throughput contribution remains below 30%, despite utilizing approximately 14 dB higher median transmit power (Fig. 3a). As discussed in earlier sections, despite this higher transmit power, n77 in both game day and pregame exhibits low MCS values (Fig. 3b), suggesting that poor signal propagation at C-band limits its uplink capacity. A comparison of downlink and uplink throughput between the lower-frequency FDD bands (b5, b13) and the higherfrequency TDD band (n77) during game day further highlights the uplink-downlink asymmetry. As illustrated in Fig. 8, while n77 achieves a high median downlink throughput of around 11.5 Mbps, its median uplink throughput drops to 0.5 Mbps. Conversely, the FDD bands exhibit much stronger uplink throughput relative to their downlink. Specifically, b5 exhibits median downlink and uplink throughputs of 0.9 Mbps and 4.4 Mbps, respectively, while b13 exhibits median values of 0.2 Mbps and 0.9 Mbps. Together, these results reveal a clear directional asymmetry in Verizon’s deployment: the highfrequency TDD n77 band dominates downlink capacity, while uplink throughput depends heavily on the sub-1 GHz FDD bands (b5, b13) during game day. Next, we evaluate T-Mobile’s uplink performance when operating in 5G NSA mode (a discussion of T-Mobile’s 5G
Median 0.90
0.6 0.4 0.2 0.0
0
b2(P)+n71(S) Median: 10.7 Mbps b2(P)+n41(S) Median: 5.7 Mbps 5 10 15 20 25 30 Total Aggregated Uplink Throughput(Mbps)
Fig. 9. Total aggregated uplink throughput CDF during game day for TMobile NSA mode. Notably, the network achieves a higher median throughput when low-frequency n71 FDD band is aggregated as the secondary carrier alongside the b2 anchor, rather than the high-frequency n41 TDD band.
SA mode is provided in §III-E). Unlike Verizon, which relies solely one the high-frequency TDD n77 band for 5G stadium coverage, T-Mobile deploys both a low-frequency FDD 5G carrier, n71 (uplink 677 MHz, 20 MHz BW), and a highfrequency TDD 5G carrier, n41 (2500/2600 MHz, 100/90 MHz BW). This dual deployment allows a direct comparison of aggregated NSA uplink throughput across the two frequency tiers. As illustrated in Fig. 9, although aggregating with n41 enables a higher peak uplink throughput, the carrier combination of b2 with n71 achieves a significantly higher median aggregated uplink throughput (10.7 Mbps) than b2 and n41 (5.7 Mbps). Consistent with the uplink-downlink asymmetry observed in Verizon’s deployment, T-Mobile’s high-frequency carrier exhibits a similar performance gap. When examining perband downlink and uplink performance, we observe that while the high-frequency n41 band delivers a median downlink
50
120 T-Mobile (SA) - Downlink and Uplink Throughput per Band (Game Day)
Verizon COW - Downlink and Uplink Throughput per Band Downlink Uplink
Downlink Uplink
100
30
Throughput (Mbps)
Throughput (Mbps)
40
20 10 0
80 60 40 20
Median 5.69
Median 18.90
b66 (20 MHz)
Median 20.61
Median 5.12
n77 (100 MHz)
0
Median 11.86
n25 (20 MHz)
Median 8.74
Median 78.87
Median 5.99
n41 (100/90 MHz)
Fig. 10. Downlink and uplink performance comparison on the Verizon COW utilizing a primary b66 anchor and secondary n77 band. Despite conducive C-band propagation conditions (for band n77 the median MCS: UL 20/DL 16; median UL Tx power: 19 dBm), it shows a disproportionate throughput distribution. While n77 carries the bulk of the downlink load, the b66 anchor provides a median uplink throughput approximately four times higher.
Fig. 11. Downlink and uplink throughput distributions for T-Mobile’s Standalone (SA) mode, comparing the n25 (FDD) and n41 (TDD) carriers. Despite operating under comparable MCS conditions and the n41 band utilizing 3dB more uplink transmit power, the n25 FDD carrier maintains a distinct performance edge in the uplink, achieving a median throughput of approximately 9 Mbps compared to 6 Mbps for n41.
throughput of 44 Mbps, its median uplink throughput drops sharply to 1.5 Mbps. Conversely, the lower-frequency bands exhibit a much more balanced profile: n71 records median downlink and uplink throughputs of 20 Mbps and 10 Mbps, respectively, while the LTE anchor b2 achieves 3 Mbps and 1 Mbps. Similar to Verizon’s deployment, the T-Mobile data shows the same trend: the high-frequency TDD n41 band dominates downlink capacity, while the low-frequency FDD n71 carrier provides higher uplink throughput. Taken together, the Verizon and T-Mobile results reveal a consistent operational pattern in 5G networks: high-frequency TDD carriers are fundamentally uplink-limited by three compounding constraints: 1) From a propagation perspective, the larger path loss at high frequencies degrades channel quality, yielding low uplink MCS despite operation near the 3GPP transmit power limit, as shown in §III-B. 2) From a spectral perspective, UE transmit power constraints prevent effective utilization of the wideband high-frequency channels, resulting in substantial PRB allocation disparity relative to the downlink. While the base station utilizes a large transmit power budget to maintain high PRB allocation in the downlink, the power-constrained UE is unable to secure a comparable allocation in the uplink, even under unloaded conditions, as shown in §III-C. 3) From a temporal perspective, downlink-heavy TDD slot configurations further restrict uplink transmission opportunities. Consequently, although high-frequency TDD carriers dominate downlink throughput, uplink performance in both T-Mobile and Verizon NSA deployments remains heavily dependent on low-frequency FDD bands.
by low-frequency FDD bands under favourable propagation conditions and high MCS, we analyze measurements collected around the COW deployment during the Oct. 4, 2025, game day for PCIs 345 and 347. Given the close proximity of the measurement locations to the COW and the high likelihood of line-of-sight propagation, both the primary b66 band (uplink 1720 MHz, 20 MHz BW) and the secondary n77 band (3700/3800 MHz, 100 MHz BW) achieve high MCS values in both downlink and uplink while operating substantially below the 3GPP Power Class 2 limit of 26 dBm (the median uplink transmit power is 19 dBm for n77 and 17 dBm for b66). Specifically, the uplink median MCS is approximately 18 for b66 and 20 for n77. However, despite this favorable MCS on n77, Fig. 10 shows that the median n77 uplink throughput remains near 5 Mbps, which is roughly onefourth of the uplink throughput carried by b66. This behavior reverses in the downlink, where n77 carries the vast majority of the load. The 5G contribution ratio further confirms this asymmetry: although n77 carries nearly all downlink traffic (median 97%), its median uplink contribution remains below 10%. Moreover, the median uplink PRB allocation is about 10% lower than the downlink allocation, despite substantial UE transmit power headroom, indicating that uplink resource contention remains significant. The Verizon COW deployment indicates that even under excellent signal propagation and high MCS conditions—where the UE possesses significant transmit power headroom—the network remains heavily reliant on the primary FDD band for uplink capacity. The TDD carrier is firstly structurally limited by its downlink-heavy slot allocation, which reduces temporal uplink allocation. Furthermore, this temporal restriction induces a secondary spectral limitation under load: because the TDD frame provides a low number of uplink slots, user congestion is amplified in the uplink compared to the downlink during high-traffic events, as suggested by their
E. Performance Advantage of FDD in Uplink To assess whether high-frequency 5G TDD carriers can sustain uplink loads comparable to or greater than those carried
median PRB allocation difference of 10%. Consequently, even with conducive propagation condition and Tx power not being a bottleneck, UE fails to achieve an uplink high enough PRB allocation on the wide TDD carrier comparable to its downlink, which combined with the inherently low number of uplink slots, forces the network to depend on the continuous FDD band for uplink capacity. A parallel analysis can be applied to T-Mobile’s 5G SA network. While T-Mobile does not deploy a COW, their SA architecture within the stadium bowl includes both a mid-frequency FDD band n25 (uplink-1894 MHz, 20 MHz BW) and a high-frequency TDD band n41 (2500/2600 MHz, 100/90 MHz BW). Without uplink CA, these carriers are allocated independently. As shown in Fig. 3b, the n25 and n41 bands achieve comparable uplink MCS, with medians of approximately 5 and 4, respectively. In particular, the transmit power allocated to the n25 band is on average 3 dB lower than that of n41 (Fig 3a). Examining the throughput per band (Fig. 11) reveals similar disparity like Verizon. Although n41 delivers far superior downlink performance—carrying roughly eight times the median throughput of n25—the n25 band has performance edge over n41 in the uplink, achieving a median of approximately 9 Mbps compared to n41’s 6 Mbps. This occurs despite n41 utilizing twice the transmit power and both bands having comparable MCS. This reinforces the finding that high-frequency TDD bands are inherently less suitable for uplink traffic compared to FDD bands even when signal propagation and channel conditions are comparable, due to TDD bands asymmetric frame configurations that heavily favor downlink transmission. IV. C ONCLUSION In this paper, we presented a comprehensive measurement study of network performance in a high-density stadium environment, focusing on the fundamental performance contrast between the downlink and uplink. Our analysis reveals that uplink performance on high-frequency wide TDD bands is severely power constrained. Due to extreme propagation losses, UE consistently yields very low MCS and minimal PRB allocations compared to downlink, despite operating near the maximum permissible 3GPP transmit power limits. This leaves UE being unable to use much of the bandwidth even in situation of no network congestion like empty stadium. Analyzing the NSA deployments of both Verizon and TMobile, stark asymmetry emerges in their downlink and uplink behavior. While wide, high-frequency TDD carriers bear the vast majority of the downlink load, the lower-frequency FDD bands are carrier of the bulk of the uplink traffic. Furthermore, we demonstrated that this uplink bottleneck is not strictly a product of poor propagation. By evaluating scenarios with excellent channel conditions (Verizon COW deployment) or with comparable channel conditions between FDD and TDD bands (T-Mobile SA deployment), we pointed the structural limitations of the TDD duplexing scheme. Even when high-frequency TDD bands achieved exceptionally high or comparable MCS values to their FDD counterparts, the
FDD bands maintained a distinct performance edge in the uplink. This highlights that the asymmetric, downlink-heavy frame configuration of TDD bands fundamentally limits uplink capacity, regardless of signal quality. Finally, these findings provide a critical insight regarding the impact of duplexing scheme on power constrained uplink. The TDD strategy relies on granting the UE intermittent, time-multiplexed access to a large bandwidth, with the uplink receiving a small fraction of the temporal slots compared to the heavily favored downlink in most deployments. However, this yield diminishing returns because as our measurements show the power-constrained UE cannot physically access such a wide channel under typical propagation conditions, leaving much of the band inaccessible even when the network is unloaded. In contrast, the FDD strategy allocates a smaller, dedicated chunk of bandwidth that is continuously available for uplink. This constant availability, combined with a narrower channel that is accessible with the UE’s limited power budget, ensures the resource blocks can be better accessed both spectrally and temporally, making FDD architectures more advantageous and efficient for achieving uplink performance in real world scenarios. R EFERENCES [1] D. Xu, A. Zhou, X. Zhang, G. Wang, X. Liu, C. An, Y. Shi, L. Liu, and H. Ma, “Understanding operational 5G: A first measurement study on its coverage, performance and energy consumption,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication on the applications, technologies, architectures, and protocols for computer communication, 2020, pp. 479–494. [2] M. I. Rochman, W. Ye, Z.-L. Zhang, and M. Ghosh, “A comprehensive real-world evaluation of 5G improvements over 4G in low-and midbands,” IEEE Transactions on Cognitive Communications and Networking, 2025. [3] E. Dahlman, S. Parkvall, and J. Skold, 5G NR: The next generation wireless access technology. Academic Press, 2020. [4] Ericsson, “Enhancing Event Experience: Mobility Report,” 2024. [Online]. Available: https://www.ericsson.com/en/reports-and-papers/ mobility-report/articles/enhancing-event-experience [5] I. Khan, M. Ghoshal, J. Angjo, S. Dimce, M. Hussain, P. Parastar, Y. Yu, X. Deng, S. Hawal, S. Huang et al., “How mature is 5G deployment? A cross-sectional, year-long study of 5G uplink performance,” Computer Communications, vol. 237, p. 108153, 2025. [6] M. Ghoshal, Z. J. Kong, Q. Xu, Z. Lu, S. Aggarwal, I. Khan, Y. Li, Y. C. Hu, and D. Koutsonikolas, “An in-depth study of uplink performance of 5G mmWave networks,” in Proceedings of the ACM SIGCOMM Workshop on 5G and Beyond Network Measurements, Modeling, and Use Cases, 2022, pp. 29–35. [7] D. Chmieliauskas and Š. Paulikas, “Evaluation of uplink video streaming qoe in 4g and 5g cellular networks using real-world measurements,” IEEE Access, 2025. [8] Rohde & Schwarz, “Verizon 5G Is the MVP at Levi’s Stadium for Super Bowl LX,” https://www.ookla.com/articles/ verizon-5g-is-the-mvp-at-super-bowl-lx, Feb. 2026. [9] Nokia, “Mass Event Optimization,” Nokia Networks, White Paper, 2020. [Online]. Available: https://www.nokia.com [10] H. Shakhatreh, K. Hayajneh, K. Bani-Hani, A. Sawalmeh, and M. Anan, “Cell on Wheels-Unmanned Aerial Vehicle System for Providing Wireless Coverage in Emergency Situations,” Complexity, vol. 2021, no. 1, p. 8669824, 2021. [11] Rohde & Schwarz, “QualiPoc Android,” Retrieved from https://www.rohde-schwarz.com/us/products/test-and-measurement/ network-data-collection/qualipoc-android 63493-55430.html. [12] 3GPP, “NR; User Equipment (UE) radio transmission and reception; Part 1: Range 1 Standalone,” Technical Specification (TS) 38.101-1, Dec 2022, version 17.8.0. [Online]. Available: https://www.3gpp.org