26 GHz OFDM and 77 GHz FMCW Radar Dataset for Domain Shift Invariant Blockage Prediction
This dataset is the radar and groundtruth dataset linked to the paper "26 GHz OFDM and 77 GHz FMCW Radar Dataset for Domain Shift Invariant Blockage Prediction". The abstract of this paper is blow. The infor of the other part of the communication OFDM dataset in this paper can be found in the paper that is openly accessible.
This paper presents a novel millimeter wave communication (comms) and radar sensing co-existing dataset. The measurement campaign was performed for blockage prediction with diverse human activities. 26 GHz Orthogonal Frequency Division Multiplexing (OFDM) multi-beam communication testbed and 77 GHz Frequency-Modulated Continuous-Wave (FMCW) multiple input, multiple output (MIMO) radar multi-monostatic set-up were configured. The corresponding bistatic channel state information and multi-monostatic backscattered channels are pre-processed for preliminary domain shift analysis by means of visual pre-processed sample inspection. Domain shift inside a blockage prediction model occurs when measurement circumstances under which model training data was collected significantly differ from the model inference measurement circumstances. Domain shifts cause model performance deterioration in the inference phase. No previous millimeter wave blockage prediction research considers mitigating domain shift in prediction models. We argue that this is caused by no millimeter wave blockage prediction datasets being available with samples collected under a large number of different measurement circumstances. Analysis results indicate presence of different signature presence levels in pre-processed radar backscattered channel samples and different doppler bin energy magnitudes and locations in pre-processed OFDM testbed channel state information samples captured under varying measurement circumstances. Therefore, creating a large enough blockage prediction dataset with samples captured under varying measurement circumstances that induce hard enough domain shifts between model train and inference situations is important to allow model domain shift mitigation research.
- 2023-03-01 first online, published, posted
- MSCA-IF-2020 - Individual Fellowships V.I.P. (Grant agreement ID: 101026885) and Dutch SectorPlan
TU Eindhoven, Department of Mathematics and Computer Science;
Katholieke Universiteit Leuven, Networked Systems
- 9,658,961,632 bytes md5 Ground-truth-videos.zip
- 38,629,265,434 bytes md5 Radar data.zip
- 35,949,246,465 bytes md5 Radar2.zip
- 1,683 bytes md5 Read_Me.txt
- download all files (zip)