axakhan/IOV

---
license: cc-by-4.0
task_categories:
- reinforcement-learning
- tabular-classification
language:
- en
tags:
- network-slicing
- 5G
- IoV
- DRL
- MEC
- URLLC
- eMBB
- mMTC
- vehicular-networks
- 3GPP
- MobFogSim
- SUMO
size_categories:
- 1M<n<10M
---

# IoV-DynSlice-2026

**Dynamic 5G Network Slicing Dataset for Internet of Vehicles — 7.83 Million Rows**

Generated using **MobFogSim** (fog/edge computing simulator) extended with a **3GPP TR 38.901 Urban Macro (UMa) channel model**, driven by a real-world **SUMO vehicular mobility trace** from Islamabad, Pakistan.

Designed for training Deep Reinforcement Learning (DRL) agents (PPO, SAC, TD3) to perform dynamic bandwidth allocation across three 5G network slices: URLLC, eMBB, and mMTC.

---

## Dataset at a Glance

| Property | Value |
|---|---|
| Total rows | 7,834,940 |
| Columns | 42 |
| Seeds | 10 |
| Vehicles per seed | 1,000 |
| Simulation ticks | 786 (seconds) |
| Missing values | 0 |
| File format | CSV |
| License | CC-BY 4.0 |

---

## Simulation Setup

### Infrastructure
| Parameter | Value | Standard |
|---|---|---|
| Base stations (gNBs) | 225 (15×15 grid) | 3GPP TR 38.901 UMa |
| Inter-Site Distance (ISD) | 500 m | 3GPP TR 38.901 Table A.1-2 |
| Coverage area | 56.25 km² (7.5×7.5 km) | Urban core Islamabad |
| Carrier frequency | 3.5 GHz (n78 band) | PTA Pakistan 5G allocation |
| Total spectrum | 100 MHz | 3GPP TS 38.104 FR1 max |
| Tx power | 30 dBm | — |
| BS height | 25 m | UMa standard |
| UE height | 1.5 m | UMa standard |
| Noise figure | 7 dB | — |

### Channel Model
3GPP TR 38.901 Urban Macro (UMa) path loss:

```
PL [dB] = 13.54 + 39.08 × log₁₀(d) + 20 × log₁₀(fc) − 0.6 × (hUT − 1.5)
```

where `d` is distance in metres, `fc` is carrier frequency in GHz, and `hUT` is UE height in metres.

SINR → PLR mapping follows 5-tier lookup table derived from 3GPP link-level curves. Shannon capacity: `BW × log₂(1 + SINR_linear) × (1 − PLR)`.

### Mobility
- **Trace**: SUMO simulation of Islamabad, Pakistan
- **Total SUMO vehicles**: 171,140
- **Selected vehicles per seed**: 1,000 (random selection)
- **Time steps**: 786 seconds
- **Coverage**: 271 km² total SUMO area

### 5G Network Slices
| Slice | SLA Latency | SLA PLR | Base BW | Max BW |
|---|---|---|---|---|
| URLLC | ≤ 1 ms | ≤ 0.1% | 20 MHz | 50 MHz |
| eMBB | ≤ 10 ms | ≤ 5% | 60 MHz | 70 MHz |
| mMTC | ≤ 500 ms | ≤ 10% | 20 MHz | 25 MHz |

---

## Event Types (DPC — Dynamic Priority Controller)

The DPC detects 6 network conditions from real-time traffic signals and adjusts slice priorities and bandwidth allocation accordingly.

| Event | Count | % | Description |
|---|---|---|---|
| VEHICLE_SURGE | 1,909,960 | 24.4% | Rapid increase in vehicle density |
| HANDOVER_STORM | 1,879,990 | 24.0% | High handover rate (>5 per tick) |
| CONGESTION | 1,527,550 | 19.5% | Slow traffic, moderate density |
| EMERGENCY | 1,779,900 | 22.7% | Peak density, URLLC breach |
| NORMAL | 327,570 | 4.2% | Free-flow, low density, no events |
| ACCIDENT | 409,970 | 5.2% | High stopped-vehicle fraction (≥92%) |

### Bandwidth Allocation per Event (MHz)

| Event | URLLC | eMBB | mMTC | Total |
|---|---|---|---|---|
| NORMAL | 20 | 60 | 20 | 100 |
| CONGESTION | 30 | 55 | 15 | 100 |
| VEHICLE_SURGE | 30 | 55 | 15 | 100 |
| HANDOVER_STORM | 35 | 50 | 15 | 100 |
| EMERGENCY | 40 | 45 | 15 | 100 |
| ACCIDENT | 50 | 35 | 15 | 100 |

---

## Column Schema (42 columns)

### Identification
| Column | Type | Description |
|---|---|---|
| `vehicle_id` | int | Unique vehicle identifier |
| `timestamp` | int | SUMO simulation second (0–785) |
| `seed` | int | Random seed used for this run |

### Vehicle State
| Column | Type | Description |
|---|---|---|
| `vehicle_x_m` | float | Vehicle x-coordinate (metres) |
| `vehicle_y_m` | float | Vehicle y-coordinate (metres) |
| `vehicle_speed_mps` | float | Instantaneous speed (m/s) |
| `dist_to_nearest_ap_m` | float | Distance to nearest gNB (metres) |
| `nearest_ap_id` | int | ID of nearest gNB |
| `vehicle_handover` | int | 1 if handover occurred this tick |
| `vehicle_migration` | int | 1 if VM migration occurred |

### Traffic Context
| Column | Type | Description |
|---|---|---|
| `traffic_density_veh_per_km2` | float | Global vehicle density (all SUMO vehicles / 271 km²) |
| `vehicle_count` | int | Total SUMO vehicles active at this tick |
| `event_flag` | int | Numeric encoding of event_type |
| `event_type` | str | NORMAL / CONGESTION / VEHICLE_SURGE / HANDOVER_STORM / EMERGENCY / ACCIDENT |
| `root_cause` | str | DPC root cause classification |
| `congestion_severity` | int | 0=NORMAL, 1=CONGESTION, 2=SURGE/HO-STORM, 3=EMERGENCY/ACCIDENT |

### DRL Reward Signal
| Column | Type | Range | Description |
|---|---|---|---|
| `reward` | float | [−0.44, +0.39] | Composite DRL reward |

Reward formula:
```
r = 0.35×latComp + 0.20×lossComp + 0.15×netComp − 0.10×migPenalty − 0.20×slaPenalty
```

### URLLC Slice Metrics
| Column | Type | Description |
|---|---|---|
| `urllc_latency_ms` | float | URLLC slice latency (ms) |
| `urllc_packet_loss_rate` | float | URLLC PLR |
| `urllc_bandwidth_alloc_mhz` | float | Allocated bandwidth (MHz) |
| `urllc_throughput_mbps` | float | Achieved throughput (Mbps) |
| `urllc_jitter_ms` | float | Latency jitter (ms) |
| `urllc_slice_priority` | int | Current priority level |

### eMBB Slice Metrics
| Column | Type | Description |
|---|---|---|
| `embb_latency_ms` | float | eMBB slice latency (ms) |
| `embb_packet_loss_rate` | float | eMBB PLR |
| `embb_bandwidth_alloc_mhz` | float | Allocated bandwidth (MHz) |
| `embb_throughput_mbps` | float | Achieved throughput (Mbps) |
| `embb_jitter_ms` | float | Latency jitter (ms) |
| `embb_slice_priority` | int | Current priority level |

### mMTC Slice Metrics
| Column | Type | Description |
|---|---|---|
| `mmtc_latency_ms` | float | mMTC slice latency (ms) |
| `mmtc_packet_loss_rate` | float | mMTC PLR |
| `mmtc_bandwidth_alloc_mhz` | float | Allocated bandwidth (MHz) |
| `mmtc_throughput_mbps` | float | Achieved throughput (Mbps) |
| `mmtc_jitter_ms` | float | Latency jitter (ms) |
| `mmtc_slice_priority` | int | Current priority level |

### Network State
| Column | Type | Description |
|---|---|---|
| `network_utilization_percent` | float | Total spectrum utilisation (always 100% — full allocation) |
| `edge_cpu_utilization_percent` | float | Edge server CPU utilisation |
| `global_migration_count` | int | Cumulative VM migrations |
| `global_handover_count` | int | Cumulative handovers |
| `handover_rate_per_tick` | float | Handovers per second at this tick |
| `isci_value` | float | Inter-Slice Contention Index |
| `dynamic_priority_shift_flag` | int | 1 if DPC shifted priorities this tick |
| `sla_violation_count` | int | Number of slices violating SLA (0–3) |
| `packet_delivery_ratio` | float | Network-wide packet delivery ratio |
| `fog_latency_ms` | float | Fog node RTT (ms) from 3GPP physics model |

---

## DRL Environment

### Observation Space (9 features)
```python
obs = [
urllc_latency_ms, # URLLC slice latency
embb_latency_ms, # eMBB slice latency
mmtc_latency_ms, # mMTC slice latency
urllc_bandwidth_alloc_mhz, # current URLLC BW allocation
embb_bandwidth_alloc_mhz, # current eMBB BW allocation
mmtc_bandwidth_alloc_mhz, # current mMTC BW allocation
traffic_density_veh_per_km2,# traffic density
vehicle_speed_mps, # mean vehicle speed
isci_value # inter-slice contention index
]
```

### Action Space (3 continuous values)
Bandwidth allocation in MHz for each slice. Constraint: sum ≤ 100 MHz.
```
action = [urllc_bw, embb_bw, mmtc_bw] ∈ [0, 100]³
```

### Reward
```
range: [−0.44, +0.39]
mean: −0.291 (std: 0.133)
```
| Event | Mean Reward |
|---|---|
| NORMAL | +0.255 |
| VEHICLE_SURGE | −0.251 |
| CONGESTION | −0.282 |
| HANDOVER_STORM | −0.313 |
| ACCIDENT | −0.358 |
| EMERGENCY | −0.403 |

---

## Loading the Dataset

```python
import pandas as pd
from datasets import load_dataset

# Option 1: pandas (recommended for large datasets)
df = pd.read_csv("hf://datasets/axakhan/IOV/combined_dataset_v6.csv")

# Option 2: HuggingFace datasets library
dataset = load_dataset("axakhan/IOV")
df = dataset["train"].to_pandas()

# DRL state columns
STATE_COLS = [
"urllc_latency_ms", "embb_latency_ms", "mmtc_latency_ms",
"urllc_bandwidth_alloc_mhz", "embb_bandwidth_alloc_mhz", "mmtc_bandwidth_alloc_mhz",
"traffic_density_veh_per_km2", "vehicle_speed_mps", "isci_value"
]

ACTION_COLS = [
"urllc_bandwidth_alloc_mhz",
"embb_bandwidth_alloc_mhz",
"mmtc_bandwidth_alloc_mhz"
]

X = df[STATE_COLS].values # observations
a = df[ACTION_COLS].values # actions taken by DPC
r = df["reward"].values # rewards
```

---

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{khan2026iovdynslice,
author = {Khan, Abubakar},
title = {IoV-DynSlice-2026: Dynamic 5G Network Slicing Dataset for Internet of Vehicles},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/axakhan/IOV},
note = {7.83M rows, MobFogSim + 3GPP TR 38.901 UMa + SUMO Islamabad trace,
225 gNBs, ISD=500m, fc=3.5GHz, BW=100MHz, 1000 vehicles × 10 seeds}
}
```

---

## Related Work

- MobFogSim: [github.com/diogomg/MobFogSim](https://github.com/diogomg/MobFogSim)
- 3GPP TR 38.901: Study on channel model for frequencies from 0.5 to 100 GHz
- 3GPP TS 22.261: Service requirements for the 5G system
- SUMO: [sumo.dlr.de](https://sumo.dlr.de)