Description This dataset provides 800 realistic ray-tracing-based channel realizations and URLLC QoS metrics, generated using DeepMIMO v2 [1]. It is designed to support research on predicting the minimal number of active RIS (Reconfigurable Intelligent Surface) reflective elements (N) that satisfy stringent 6G URLLC requirements in single-user scenarios: End-to-end latency ≤ 1 ms (propagation delay + 0.3 ms processing delay) Reliability ≥ 99.9999% (physical-layer SER ≤ 10⁻⁶) Achievable data rate≥ 300 Mbps For each user position and RIS configuration, the dataset includes: Complex channel coefficients: h_{TU}: direct BS-to-UE channel (real & imaginary parts) h_{TR}: BS-to-RIS channel h_{RU}: RIS-to-UE channel User position coordinates (x, y, z) in meters Optimal beam index (beam_id) from UPA codebook search Optimal RIS phase-shift vector Φ (real & imaginary parts) Optimal BS beamforming vector w (real & imaginary parts) Minimal sufficient N, achievable data rate, latency, reliability, best linear SNR Feasibility link & optimal_N (the smallest N that meets all constraints, or 0 if infeasible) The objective is to jointly optimize the BS beamforming vector w, the RIS phase-shift matrix Φ, and the number of active RIS elements N, to minimize hardware complexity while meeting 6G URLLC QoS requirements. Generation Details DeepMIMO version: v2 Scenarios & Parameters: RIS Size (N) DeepMIMO Scenario Carrier Frequency Transmit Power 64 O1_3p5 3.5 GHz 10 kW 256 O1_28 28 GHz 100 kW 512 O1_28 28 GHz 100 kW 1024 O1_60 60 GHz 100 kW Bandwidths (B): 500 MHz, 200 MHz, 100 MHz, 50 MHz (applied across all RIS sizes) Users per scenario: 50 user requests (different positions) Total combinations: 4 RIS sizes × 4 bandwidths × 50 users = 800 scenario–user combinations Codebook generation: 3D UPA codebook with oversampling factor of 2 [2], steering angles uniformly quantized over [0, π), phases constructed via exponential terms and Kronecker products. Reliability calculation: Physical-layer reliability ρ = 1 − Q(√(SINR / 2)) (approximation for Gray-coded 64-QAM SER) Latency model: end-to-end latency = propagation delay + 0.3 ms fixed processing delay Other key parameters: LoS-dominant paths, antenna spacing 0.5λ, noise PSD −174 dBm/Hz, MRT beamforming at BS, unit-modulus constraint on RIS phases. Files dataset_nn_single_user_labeled_2.csv → Channel coefficients, positions, beam_id, Φ, w, feasibility link, optimal_N dataset_nn_single_user_urllc.csv → QoS metrics (achievable data rate, latency, reliability, best_linear_snr) per user and RIS size References [1] A. Alkhateeb, "DeepMIMO: A generic deep learning dataset for millimeter wave and massive MIMO applications," arXiv, 2021.[2] A. Taha, Y. Zhang, F. B. Mismar and A. Alkhateeb, "Deep Reinforcement Learning for Intelligent Reflecting Surfaces: Towards Standalone Operation," in 2020 IEEE 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Atlanta, GA, USA, 2020.