SiESTA-VR reproducibility dataset, Wireless VR streaming simulation traces

This dataset gathers all data necessary to replicate the figures and tables in the paper "SiEStA-VR, a Simulation Environment for Streaming Applications in Virtual Reality over Wi-Fi 7." Version 3 accompanies the second revision of the manuscript. The simulator is open source: https://github.com/wn-upf/SiEStA-VR. Use branch main. The data in this version were produced by commit 1639365, which is part of main's history; later commits on main change only the post-simulation viewer and the repository's README, and with the viewer off (the default) they write the same output files byte for byte. What changed in version 3 New simulations, 60 seeds. The three Results-section sweeps (Figs. 5-7) were run again on the corrected simulator with 60 seeds per configuration (14,160 runs; version 2 held 30 seeds). Results with one or two users agree with version 2 within seed-to-seed variation; with three or more users they differ, because of two corrections: Access-point queue: the AP removed transmitted downlink packets from its shared queue by packet number alone. Every user's server numbers its packets from zero, so other users' packets with the same numbers were also removed, unsent, and under load a user's video could stop for the rest of the run. The AP now identifies each packet by its source, destination and number. NeSt-VR: a frame delivered in one A-MPDU had zero span and entered NeSt-VR's capacity estimate as zero capacity; such samples are now excluded. The statistics timeout is now checked on every frame, as in ALVR, and its timer starts with the session's first frame. Deterministic runs. Every random draw comes from the run's seed, so the same code, command and seed write the same files byte for byte. The command of every run is included. Per-packet logs not included. The client's per-packet logs (SHARD_RX_stats, 85 GB) are left out to fit the record's size limit. Every quantity the paper derives from them is in the included run summary, and any run's log can be regenerated exactly from its command. Run summary and table scripts. One row per run and user (73,560 rows) with VF-RTT, bitrate and frame loss at deadlines from 30 to 120 ms, and the scripts that compute Tables 7, 8 and A.12 from it. Validation rerun. The single-session validation against the ALVR testbed (Section 6.1, Fig. 3, Table 5) was run again with one spatial stream, the 60 us EHT preamble and an ALVR-like client. Parquet files are compressed with zstd (version 2: snappy); their contents are unchanged. The data of version 2 (30 seeds, previous simulator) remain available as version 2 of this record. Files SiEStA-VR_v3_Results_<sweep>_seeds<a>-<b>.zip (9 files): the simulation runs, one archive per sweep and block of 20 seeds. Each archive unzips into Results_section/Results_rerun_<sweep>/. Sweeps: SLO, static, one 80 MHz link, all ABRs (Fig. 5); MLO, static, one link vs. MLO over two 80 MHz links (Fig. 6); RWK, the same under random-walk mobility (Fig. 7). SiEStA-VR_v3_reproduction.zip: README.md (full description), the run summary (run_summaries/), the paper's tables (tables/), the figure and table scripts (python_scripts/), the 14,160 simulator commands and a script that runs them (sweep_commands/), the list of runs (manifest_runs.csv) and the validation data and scripts (Validation_realtestbed/). README.md: the same description as in the reproduction archive. scripts_and_misc.zip (unchanged from version 2): per-frame encoded video sizes (csv_framesizes/), the MATLAB cross-validation (Validation_matlab/), and the first-round reviewer experiments (MAC PER sweep, VMAF/SSIM/PSNR for four videos, watermark sensitivity). networkTrafficPoisson.m, parall_STR_simulation_toolbox.m (unchanged from version 2): the MATLAB WLAN Toolbox scripts of the MLO check (Appendix A.3). They require the WLAN Toolbox of MATLAB R2025b. Run folders Each run folder name encodes its parameters; sweeps can be selected with regular expressions on it. Field Meaning T Simulated time (s) D Distance to the access point (m) Br Target (or initial) bitrate of the VR stream (Mbps) FPS Frame rate of the VR stream Codec Video codec (HEVC or AV1) GoP Group-of-Pictures size of the encoded stream IR Intra refresh enabled (1) or disabled (0); overrides GoP if enabled VBVframe Per-frame VBV (rate-control buffer) constraint enabled (1) or disabled (0) macPL MAC packet error rate; lost packets are retransmitted aggAMPDU Maximum A-MPDU aggregation size NXR Number of VR users NBG / BGLambda Number of background stations and their traffic arrival rate UL Background traffic direction: 0 (DL), 1 (UL), 2 (DL/UL mixed) S Random seed ABR Adaptive bitrate algorithm: 0 CBR, 1 NeSt-VR, 2 EVeREst (intra), 4 GCC, 5 NADA, 9 ALVR Adaptive SLO80 / MLO80-80 One 80 MHz link / Multi-Link Operation over two 80 MHz links RWALK Random-walk mobility enabled (1) or disabled (0); overrides D if enabled Files per run, one per user <u>: XR_stats_<u>.parquet: one row per frame reported in the client's statistics, as received by the server: VF-RTT (rtt_ms), frame index, size, span, jitter, throughput, requested bitrate and others. BITRATE_stats<u>.parquet: one row per frame the server generated, with its time and bitrate. NESTVR_stats<u>.csv: NeSt-VR's internal state at each decision (NeSt-VR runs only). trace_emu_effects<u>.csv: the emulated wired link between server and AP (a 1 Gb/s cap, no impairments in these sweeps).