The 5G Throughput Gap
Marketing materials promise multi-gigabit 5G speeds, but real-world throughput often falls far short. Ericsson's 2025 Mobility Report shows that the global median 5G download speed is 186 Mbps -- roughly 10% of the theoretical peak for a 100 MHz TDD carrier with 4x4 MIMO. T-Mobile US, operating the largest mid-band 5G network in the United States, reports a median of 245 Mbps on their n41 (2.5 GHz) deployment, while peak-hour speeds in dense urban areas can drop to 80--120 Mbps.
Understanding why throughput degrades requires analyzing every link in the chain: radio conditions (SINR, MIMO rank), scheduling parameters (MCS, PRB allocation), transport/backhaul capacity, and core-network policies (session AMBR, flow QoS). This article provides a systematic diagnostic framework with worked calculations at each stage.
The 5G NR Throughput Formula
The theoretical peak throughput for 5G NR downlink is defined in TS 38.306 clause 4.1.2:
Throughput = Sigma(v_layers x Q_m x f x R_max x NRB x 12 / Ts_mu) x (1 - OH)
Simplified for a single component carrier:
| Parameter | Symbol | Description | Typical Value (n41 TDD) |
|---|---|---|---|
| MIMO layers | v | Number of spatial streams | 2--4 (depends on UE, rank) |
| Modulation order | Q_m | Bits per symbol | 6 (64-QAM), 8 (256-QAM) |
| Coding rate | R | Max coding rate | 0.926 (948/1024) |
| Bandwidth (PRBs) | NRB | Number of resource blocks | 273 (100 MHz, SCS 30 kHz) |
| Subcarriers per PRB | -- | Fixed at 12 | 12 |
| OFDM symbols per slot | -- | 14 per slot (normal CP) | 14 |
| Slot duration | Ts | 0.5 ms for 30 kHz SCS | 0.5 ms |
| TDD DL ratio | -- | DDDSU pattern DL fraction | ~70% (typical slot config) |
| Overhead | OH | DMRS, CSI-RS, CORESET, etc. | 14--18% |
Worked Example 1 -- Theoretical vs Realistic Throughput
Configuration: n41, 100 MHz, SCS 30 kHz, TDD (DDDSU), 4x4 MIMO, 256-QAM Theoretical peak (TS 38.306):- Layers: 4
- Q_m: 8 (256-QAM)
- R_max: 0.926
- PRBs: 273
- Subcarriers: 12
- Symbols per slot: 14
- Slots per second: 2000 (SCS 30 kHz)
- Raw: 4 x 8 x 0.926 x 273 x 12 x 14 x 2000 = 2.73 Gbps (full duplex)
- TDD DL ratio (~70%): 2.73 x 0.70 = 1.91 Gbps
- Overhead (16%): 1.91 x 0.84 = 1.60 Gbps theoretical peak DL
- SINR: 15 dB -> MCS 20 (64-QAM, R=0.67) instead of MCS 27 (256-QAM, R=0.926)
- Effective bits/symbol: 6 x 0.67 = 4.02 vs theoretical 8 x 0.926 = 7.41
- Reduction factor: 4.02/7.41 = 54.3% of peak modulation efficiency
- MIMO rank: 2 (average) instead of 4
- Rank reduction: 2/4 = 50% of peak spatial multiplexing
- PRB utilization: 60% (shared with other UEs)
- PRB reduction: 60%
- Realistic DL: 1.60 x 0.543 x 0.50 x 0.60 = 260 Mbps
This aligns closely with T-Mobile's reported median of 245 Mbps -- the throughput gap is explained by real-world SINR, MIMO rank, and cell loading.
Root Cause Analysis Framework
Root Cause Categories and Impact
| Category | Root Cause | Typical Throughput Impact | Diagnosis Tool |
|---|---|---|---|
| RF Conditions | Low SINR (< 5 dB) | 50--80% reduction | Drive test (SINR heatmap) |
| RF Conditions | Low MIMO rank (rank 1 instead of 4) | 50--75% reduction | UE diagnostic mode, RAN counters |
| RF Conditions | High interference (neighbor cell) | 30--60% reduction | PCI confusion analysis, RSRP delta |
| Scheduling | Low MCS due to BLER target | 20--50% reduction | MAC-layer KPIs, BLER counters |
| Scheduling | Low PRB utilization (congestion) | Proportional to 1/users | Cell PRB utilization counters |
| Scheduling | Suboptimal scheduler weights | 10--30% reduction | Vendor OSS scheduler config |
| Transport | Backhaul bottleneck | Caps at transport capacity | Backhaul utilization monitoring |
| Core/Policy | Session AMBR throttling | Caps at AMBR value | PCAP/subscriber config check |
| Core/Policy | QoS flow shaping at UPF | Variable | UPF QER inspection |
| Device | UE capability limitation | 20--50% reduction | UE capability enquiry (RRC) |
Diagnosing SINR and MIMO Rank Issues
SINR to MCS Mapping
The gNB selects the MCS (Modulation and Coding Scheme) based on the UE's CQI report, which reflects the SINR. The mapping is defined in TS 38.214 Tables 5.1.3.1-1 to 5.1.3.1-3.
| SINR Range (dB) | CQI | MCS Index | Modulation | Approx. Code Rate | Spectral Efficiency (bps/Hz) |
|---|---|---|---|---|---|
| < -3 | 1 | 0--2 | QPSK | 0.08--0.19 | 0.15--0.38 |
| -3 to 2 | 2--4 | 2--6 | QPSK | 0.19--0.44 | 0.38--0.88 |
| 2 to 8 | 5--8 | 7--14 | 16-QAM | 0.37--0.60 | 1.48--2.41 |
| 8 to 15 | 9--12 | 15--22 | 64-QAM | 0.46--0.77 | 2.73--4.62 |
| 15 to 25 | 13--15 | 23--27 | 256-QAM | 0.68--0.93 | 5.42--7.41 |
A UE at the cell edge with SINR = 3 dB operates at approximately 1.0 bps/Hz, while a UE near the cell center at SINR = 20 dB achieves 6.0 bps/Hz -- a 6x difference from the same cell.
MIMO Rank Distribution
The MIMO rank (number of spatial layers) depends on channel conditions, antenna correlation, and UE capability. Typical distributions from live networks:
| Environment | Rank 1 | Rank 2 | Rank 3 | Rank 4 | Avg Rank | Source |
|---|---|---|---|---|---|---|
| Dense urban (n41) | 15% | 45% | 25% | 15% | 2.4 | T-Mobile US 2025 |
| Suburban (n41) | 25% | 40% | 20% | 15% | 2.25 | T-Mobile US 2025 |
| Indoor (n78) | 20% | 50% | 20% | 10% | 2.2 | Deutsche Telekom 2025 |
| Rural (n77) | 35% | 45% | 15% | 5% | 1.9 | Vodafone UK 2024 |
| mmWave (n260) | 10% | 30% | 30% | 30% | 2.8 | Verizon 2025 |
The dominant factor limiting rank is antenna correlation in LoS conditions and SNR in NLoS. Rural sites with dominant LoS paths often show high rank-1 percentages even with good SINR.
Worked Example 2 -- Impact of MIMO Rank Optimization
A suburban n41 cell shows the following pre-optimization rank distribution:
- Rank 1: 35%, Rank 2: 45%, Rank 3: 15%, Rank 4: 5%
- Average rank: 0.35x1 + 0.45x2 + 0.15x3 + 0.05x4 = 1.90
After antenna tilt optimization and cross-polarization calibration:
- Rank 1: 20%, Rank 2: 40%, Rank 3: 25%, Rank 4: 15%
- Average rank: 0.20x1 + 0.40x2 + 0.25x3 + 0.15x4 = 2.35
Throughput improvement from rank alone: 2.35 / 1.90 = 23.7% increase
For a cell averaging 200 Mbps DL user throughput, this translates to an improvement of approximately 47 Mbps. SK Telecom reported a similar 20--25% throughput gain from MIMO rank optimization on their 3.5 GHz (n78) macro sites through antenna panel re-alignment and beamforming weight updates.
PRB Utilization and Cell Congestion
PRB (Physical Resource Block) utilization directly reflects cell loading. When the scheduler has fewer PRBs available per user, individual throughput drops proportionally.
| PRB Utilization | Cell Load State | Expected Per-User Impact | Action |
|---|---|---|---|
| 0--30% | Light load | Minimal -- users get full allocation | Monitor, no action needed |
| 30--60% | Moderate | 20--40% reduction from peak | Normal operation, watch trends |
| 60--80% | High | 50--70% reduction from peak | Consider capacity additions |
| 80--95% | Congested | 70--90% reduction, QoE degradation | Urgent: split cell, add carrier, offload |
| > 95% | Overloaded | Severe degradation, RRC failures | Critical: emergency capacity action |
Backhaul and Transport Bottlenecks
Even with excellent radio conditions, throughput is capped by the backhaul:
| Backhaul Type | Typical Capacity | Sufficient For | Bottleneck Risk |
|---|---|---|---|
| Fiber (10GE) | 10 Gbps | Full 5G capacity (multiple carriers) | Low |
| Fiber (1GE) | 1 Gbps | Single carrier, moderate load | Medium in dense urban |
| Microwave (E-band) | 2--10 Gbps | Good for most sites | Medium (rain fade) |
| Microwave (traditional) | 200--500 Mbps | Insufficient for 5G | High |
| Satellite (LEO) | 100--300 Mbps | Emergency / rural only | Very high |
A common hidden bottleneck is aggregation node congestion -- even if each cell site has 10GE fiber, the aggregation router may be oversubscribed. Vodafone Germany identified that 12% of their 5G throughput complaints in 2024 were caused by aggregation switch oversubscription during peak hours, resolved by upgrading to 100GE aggregation rings.
Core Network Throttling
The 5G Core can limit throughput through:
- Session AMBR: Configured per subscriber in the UDM, enforced at the UPF. A subscriber with a 100 Mbps plan will be capped regardless of radio conditions.
- Per-flow MBR: Individual QoS flow limits.
- APN-AMBR / DNN-AMBR: Aggregate limit across all PDU sessions on a DNN.
- FUP (Fair Usage Policy): After consuming a data threshold (e.g., 50 GB), the operator reduces the session AMBR.
| Throttling Point | Mechanism | Where Enforced | How to Diagnose |
|---|---|---|---|
| Session AMBR | QER in UPF | UPF (N4 from SMF) | Check subscriber profile in UDR, PFCP QER |
| Per-flow MBR | QER per QFI | UPF | PFCP session dump |
| DNN-AMBR | Aggregate QER | UPF | UDR DNN subscription data |
| FUP throttle | AMBR update via PCF | SMF modifies UPF QER | PCF policy trace, check data usage |
| TCP optimization | TCP proxy or split | UPF or middlebox | Packet capture, check TCP window |
Systematic Diagnosis Checklist
| Check | Tool | What to Look For | Fix |
|---|---|---|---|
| 1. UE capability | RRC UE Capability Enquiry | Max layers, BW, 256-QAM support | Upgrade device |
| 2. SINR | Drive test / MDT | Median SINR < 10 dB | Tilt/power optimization, interference mgmt |
| 3. MIMO rank | RAN counters (RI distribution) | Avg rank < 2.0 | Antenna calibration, beamforming update |
| 4. MCS distribution | MAC-layer KPIs | Avg MCS < 15 | Improve SINR, adjust BLER target |
| 5. PRB utilization | Cell-level counters | > 70% at peak | Add carrier, split cell, offload to small cell |
| 6. DL BLER | MAC HARQ stats | > 10% initial BLER | Outer-loop link adaptation tuning |
| 7. Backhaul utilization | Transport NMS | > 80% sustained | Upgrade backhaul capacity |
| 8. Session AMBR | Subscriber config / PCAP | AMBR lower than radio capacity | Adjust plan, check FUP status |
| 9. TCP performance | Packet capture | Small window, high RTT | TCP optimization, reduce RTT path |
| 10. Carrier aggregation | RRC config, CA activation | CA not activated or limited | Enable CA bands, check UE support |
Carrier Aggregation Impact
Carrier aggregation (CA) combines multiple NR carriers to multiply throughput. The impact is substantial:
| CA Configuration | Bands | Aggregate BW | Expected Peak DL | Operator Example |
|---|---|---|---|---|
| Single carrier | n41 (100 MHz) | 100 MHz | 1.6 Gbps | T-Mobile US baseline |
| 2CC CA | n41 + n25 (100+20 MHz) | 120 MHz | 1.9 Gbps | T-Mobile US urban |
| 3CC CA | n41 + n25 + n71 (100+20+10 MHz) | 130 MHz | 2.1 Gbps | T-Mobile US layered |
| NR-DC (SA+mmWave) | n41 + n260 (100+400 MHz) | 500 MHz | 4.5 Gbps | T-Mobile US hotspot |
| LTE-NR DC (EN-DC) | B66 + n41 (20+100 MHz) | 120 MHz | 1.8 Gbps | T-Mobile US NSA |
T-Mobile US reported that enabling 3CC NR-CA on n41+n25+n71 improved median urban throughput from 245 Mbps to 340 Mbps -- a 39% improvement. SK Telecom achieved over 3 Gbps median in selected Seoul districts using NR-DC with n78 (3.5 GHz, 100 MHz) and n258 (28 GHz, 400 MHz).
Optimization Priorities by Impact
Based on field data across multiple operators, the following ranking reflects typical throughput impact from each optimization action:
| Priority | Optimization | Typical Improvement | Effort | ROI |
|---|---|---|---|---|
| 1 | Enable carrier aggregation | 30--50% | Medium (SW + config) | Very High |
| 2 | MIMO rank optimization | 15--25% | Low (antenna tuning) | High |
| 3 | Interference management (PCI, tilt) | 10--20% | Medium (planning) | High |
| 4 | Backhaul upgrade (where bottlenecked) | Variable (removes cap) | High (CAPEX) | Medium-High |
| 5 | Scheduler tuning (BLER target, fairness) | 5--15% | Low (parameter change) | High |
| 6 | Small cell densification | 30--50% in target area | Very High (CAPEX) | Medium |
| 7 | Session AMBR / FUP policy review | Removes artificial cap | Low | High |
| 8 | TCP optimization at UPF | 5--15% | Medium (SW feature) | Medium |
Key Takeaway: Low 5G throughput is rarely caused by a single factor. The typical user experiences only 10--15% of theoretical peak due to the compounding effects of real-world SINR (reducing MCS), limited MIMO rank, cell loading (PRB sharing), and TDD duty cycle. A systematic diagnosis following the checklist above -- from UE capability through radio conditions, scheduling, transport, and core policy -- will identify the dominant bottleneck. Carrier aggregation and MIMO rank optimization typically deliver the highest return on optimization effort, while backhaul upgrades and cell densification address fundamental capacity constraints.