Geo2ComMap: Deep Learning-Based MIMO Throughput Prediction Using Geographic Data

TL;DR

This paper tackles the computational burden of predicting MIMO-OFDM throughput over large areas by introducing Geo2ComMap, which maps geographic data to full-area throughput maps using sparse measurements. The core is the U-Net-TP model, a 5-channel input network that integrates Building Maps, isotropic PG maps, and sparse RI/CQI/TP samples to predict complete Tput maps, with improvements from a special sampling strategy to reduce high-error regions. Key findings show that incorporating RI/CQI significantly improves accuracy, and advanced architectures like AG U-Net-TP and MO U-Net-TP achieve median errors as low as 27.35 Mbps (BM1) and 38.62 Mbps (BM2) with low inference times, while still producing reliable RI and P_dir estimates. The approach enables fast, area-wide wireless performance estimation, supporting planning and resource management in evolving networks, and the authors provide public code and datasets for reproducibility.

Abstract

Accurate communication performance prediction is crucial for wireless applications such as network deployment and resource management. Unlike conventional systems with a single transmit and receive antenna, throughput (Tput) estimation in antenna array-based multiple-output multiple-input (MIMO) systems is computationally intensive, i.e., requiring analysis of channel matrices, rank conditions, and spatial channel quality. These calculations impose significant computational and time burdens. This paper introduces Geo2ComMap, a deep learning-based framework that leverages geographic databases to efficiently estimate multiple communication metrics across an entire area in MIMO systems using only sparse measurements. To mitigate extreme prediction errors, we propose a sparse sampling strategy. Extensive evaluations demonstrate that Geo2ComMap accurately predicts full-area communication metrics, achieving a median absolute error of 27.35 Mbps for Tput values ranging from 0 to 1900 Mbps.

Figures (8)

• Figure 1: Architecture of Geo2ComMap, which is a 5-input channel U-Net designed to achieve efficient and precise Tput prediction.
• Figure 2: Building Maps used in the experiments. (a) BM1, characterized by a prominent open space. (b) BM2, featuring densely clustered buildings.
• Figure 3: Absolute error comparison of different input configurations in testing BM1 and testing BM2. The left box plot includes outliers (represented as black dots beyond the whiskers), while the right box plot excludes them. Each box represents the IQR of the absolute error, with the red line inside indicating the median value. The whiskers extend to the furthest data points within the range defined as Q1$-1.5\cdot$IQR to Q3$+1.5\cdot$IQR, where Q1 and Q3 denote the first and third quartiles, respectively. Horizontal lines outside the boxes represent the outlier boundaries.
• Figure 4: RMSE values for different combinations of sparse points used in training and testing. (a) RMSE in BM1. (b) RMSE in BM2.
• Figure 5: Absolute error comparison for different numbers of sampled sparse points, where $n$ denotes the number of sparse points.