Augmented Outdoor RSS Dataset for Single- and Two-Transmitter Localization

Augmented Outdoor RSS Dataset for Single- and Two-Transmitter Localization

Overview

This record contains a physics-informed augmented derivative of the outdoor RSS localization dataset A Dataset of Outdoor RSS Measurements for Localization. The original dataset was collected on the University of Utah / POWDER campus using FRS/GMRS-band radios at 462.7 MHz and includes outdoor RSS measurements for 0-TX, 1-TX, and 2-TX localization scenarios.

The present release extends that dataset with synthetic but physically grounded samples generated using PhARMNet: Physics-informed Augmentation and RF Modeling Network. PhARMNet combines terrain-aware propagation features derived from DSM/building maps and TIREM-based modeling with neural RSS prediction to generate additional transmitter-receiver measurements in regions and transmitter configurations that were not exhaustively covered in field collection.

This augmented dataset is designed to support:

• wireless transmitter localization
• single- and multi-transmitter learning
• out-of-distribution generalization studies
• data augmentation for sparse outdoor RSS datasets
• physics-informed wireless propagation modeling

Relationship to the Original Dataset

The original dataset provides real-world outdoor RSS measurements collected from the POWDER testbed and includes:

This augmented release does not replace the original measured dataset. Instead, it provides additional synthetic samples that are compatible with the original JSON schema and intended to be used alongside the original data.

What This Record Adds

This release contains two augmented subsets:

These samples were generated from the original POWDER-FRS measurements using PhARMNet-based propagation models.

Augmentation Methodology

The augmentation pipeline is based on PhARMNet, which combines real measurements with terrain-aware and physics-based propagation features extracted from DSM/building data and TIREM. In the PhARMNet framework, each transmitter-receiver pair is represented using a set of 14 terrain-aware and propagation-aware features, including line-of-sight / non-line-of-sight indicators, diffraction-related quantities, elevation angles, obstacle counts, knife-edge effects, and shadowing geometry.

Single-transmitter augmentation

For the 1-TX case, grouped receiver-specific measurements from the original single-transmitter dataset were used to train receiver-type-aware PhARMNet RSS predictors. These trained models were then evaluated over precomputed transmitter coordinate libraries in order to synthesize new RSS values for additional transmitter locations.

Two-transmitter augmentation

For the 2-TX case, a two-input variant of the PhARMNet propagation model was used. New transmitter-pair configurations were formed from available single-transmitter locations, and trained models were used to predict RSS measurements for these synthetic transmitter pairs. This makes it possible to enlarge the 2-TX dataset substantially beyond the small number of directly measured two-transmitter samples available in the original record.

Why augmentation is needed

The original outdoor dataset is valuable but sparse, especially in the multi-transmitter setting, where the number of measured 2-TX samples is limited. The PhARMNet paper specifically motivates augmentation as a way to improve spatial coverage and support better localization performance in sparse and out-of-distribution regions.

File Format

The augmented files follow the same general JSON organization as the original dataset family.

Each top-level entry is indexed by a timestamp-like key and contains:

Sample structure

{
  "2022-04-25 14:11:02": {
    "rx_data": [
      [-75.14, 40.76, -111.85, "receiver-name"]
    ],
    "tx_coords": [
      [40.767, -111.846]
    ],
    "metadata": [
      {"power": 1, "transport": "augmented", "radio": "TXA"}
    ]
  }
}

For the augmented 2-TX case, tx_coords contains two transmitter coordinates and metadata contains two corresponding transmitter entries.

Important Notes

1. Synthetic values: RSS values in this record are model-generated, not directly measured in the field.
2. Schema compatibility: The JSON structure is intentionally designed to remain compatible with the original dataset format.
3. Timestamp fields: Timestamp keys are synthetic unique identifiers created for schema consistency; they should not be interpreted as original collection times.
4. Coordinate format: Final exported coordinates are given in GPS latitude/longitude, consistent with the original dataset representation.
5. RSS units: Final exported RSS values are provided in dB, after converting model outputs back from normalized training-space values.

Intended Use

This dataset may be useful for:

• training and evaluation of RSS-based localization models
• studying sparse-data and out-of-distribution localization
• comparing measured and synthetic propagation data
• testing single- vs. multi-transmitter localization pipelines
• physics-informed augmentation for wireless networking tasks

Because this is an augmented derivative dataset, users are encouraged to use it together with the original measured data and to clearly distinguish between measured and synthetic samples in downstream experiments.

Provenance and References

This augmented dataset is derived from:

1. Original dataset:
   Frost Mitchell, Aniqua Baset, Sneha Kumar Kasera, and Aditya Bhaskara.
   A Dataset of Outdoor RSS Measurements for Localization.
   Zenodo. DOI: 10.5281/zenodo.7259895
2. Augmentation methodology:
   Md Mumtahin Habib Ullah Mazumder, Frost Mitchell, Aditya Bhaskara, Sneha Kumar Kasera, and Neal Patwari.
   Bridging Data Gaps: Enhancing Wireless Localization with Physics-Informed Data Augmentation.
   Proceedings of the ACM on Networking, 2025. DOI: 10.1145/3768995

Suggested Citation for This Record

If you use this augmented dataset, please cite:

• this Zenodo record,
• the original POWDER-FRS dataset, and
• the PhARMNet paper.