Overview

To design and evaluate the technologies of the future, we model and simulate standard-complient environmets. With Deep RT, we aim to fascilitate setting a scientific methodology to evaluate and banchmark different algorithms proposed for deep learning in wireless communications.

The project is part of the work at CDL of the TU Wien. The aim of CDL is to enhance the dependability (reliability and timeliness) of wireless communications even at high mobility, such as to support applications that require beyond best-effort services, e.g., road-safety applications, augmented and virtual reality, UAV control and communication. Our main focus thereby is on 5G and beyond mobile communication technologies.

Read our work to better understand the application of the published datasets and modeled scenarios.

Below, we share channel traces as well as carefully modeled scenarios that match the real-life environment setup. More specifically, Deep RT has four main components:

1. Scenarios,
2. Ray-traces,
3. Channel state information, and
4. Deep learning models.

Scenarios

Scenario models represent the environment that channel traces are obtained. The initial scenario model is obtained from the OpenStreetMaps. Other files include additional objects modeled and altered geometries to simulate certain environment settings. For example, in order to simulate a time-varying environment, we model an environment in which the scattering objects change over the time. Each file has the complete environment that is assumed to be stationary during that time snap. A collection of $T$ files correspond to the realization of the environment over $T$ time snapshots.

{{% alert note %}} Testing alert note {{% /alert %}}

Ray-traces

Scenario files can be used to generate your own dataset in any ray-tracing tool that you have available. We provide ray-traces too. Our traces are obtained using the state-of-the-art and commercial ray-tracers commonly used from the research community. These tools are from Matlab and Remcom. For the convinience, we provide both versions. You can also use the ray-traces with our system- or link-level simulators available at VTC.

Channel state information

After collecting all the path parameter values from the ray-traces, we obtain final CSI from a multi-path channel model. We provide such information as ready-to-use and benchmark with our work where we have used such a system model and dataset. For each scenario, we provide the corresponding CSI dataset.

Deep learning models

To understand how we utilize the datasets and deep learning, we provide some examples. Advanced algorithms proposed by our group members will be published soon.

Some notations and definitions

• $R$: number of transmitters.
• $N_{r}$: total number of antennas.
• $M$: total number of RRHs or BSs.
• $\mathcal{S}$: Set of scattering objects with $\left| \mathcal{S} \right| = S$.
• $T$: number of time-snapshots.
• $\mathbf{H}_{r}^{t}$: estimated channel matrix of a user $r$ at $t \in \lbrace 1, \ldots, T \rbrace$.
• $N_{c}’$: pilot subcarriers in an OFDM system.
• $\widehat{\mathbf{h}}_{k}^{t}$: estimated channel vector for $k$-th subcarrier at $t$.