[Paper Review] Machine Learning Methods for User Positioning With Uplink RSS in Distributed Massive MIMO

K. N. R. Surya Vara Prasad, Ekram Hossain+1 more|arXiv (Cornell University)|Jan 20, 2018

Distributed Sensor Networks and Detection Algorithms3 references3 citations

TL;DR

This paper proposes a Gaussian process regression (GPR) framework for user positioning in distributed massive MIMO systems using uplink received signal strength (RSS), addressing the critical issue of unrealistically small error bars in conventional GPR when test RSS data is noisy. It introduces a numerical approximation GP (NaGP) method that accounts for shadowing noise to produce realistic 2σ error bars, achieving RMSE performance close to the Bayesian Cramer-Rao lower bound (BCRLB).

ABSTRACT

We consider a machine learning approach based on Gaussian process regression (GP) to position users in a distributed massive multiple-input multiple-output (MIMO) system with the uplink received signal strength (RSS) data. We focus on the scenario where noise-free RSS is available for training, but only noisy RSS is available for testing purposes. To estimate the test user locations and their 2σ error-bars, we adopt two state-of-the-art GP methods, namely, the conventional GP (CGP) and the numerical approximation GP (NaGP) methods. We find that the CGP method, which treats the noisy test RSS vectors as noise-free, provides unrealistically small 2σ error-bars on the estimated locations. To alleviate this concern, we derive the true predictive distribution for the test user locations and then employ the NaGP method to numerically approximate it as a Gaussian with the same first and second order moments. We also derive a Bayesian Cramer-Rao lower bound (BCRLB) on the achievable root- mean-squared-error (RMSE) performance of the two GP methods. Simulation studies reveal that: (i) the NaGP method indeed provides realistic 2σ error-bars on the estimated locations, (ii) operation in massive MIMO regime improves the RMSE performance, and (iii) the achieved RMSE performances are very close to the derived BCRLB.

Motivation & Objective

• Address the limitation of conventional Gaussian process regression (CGP) in providing unrealistically small 2σ error bars when test RSS data is corrupted by shadowing noise.
• Develop a method to derive realistic uncertainty estimates (error bars) for user location predictions in RSS-based positioning systems with noisy test data.
• Investigate the performance of machine learning-based user positioning in distributed massive MIMO systems using uplink RSS as the signal fingerprint.
• Derive a Bayesian Cramer-Rao lower bound (BCRLB) to benchmark the achievable root-mean-squared error (RMSE) performance of the proposed methods.
• Demonstrate that the massive MIMO regime enhances positioning accuracy and that accounting for noise in test RSS improves uncertainty estimation.

Proposed method

• Train a Gaussian process model using noise-free RSS data generated from known user locations, path-loss models, and base station antenna positions.
• Apply conventional GP (CGP) for location prediction, treating noisy test RSS as if noise-free, which leads to overly optimistic uncertainty estimates.
• Introduce the numerical approximation GP (NaGP) method to approximate the true predictive distribution of user locations by matching the first and second-order moments of the noisy RSS distribution.
• Use moment-matching to numerically approximate the true predictive distribution as a Gaussian with accurate mean (location estimate) and variance (realistic 2σ error bars).
• Derive the Bayesian Cramer-Rao lower bound (BCRLB) on RMSE using linear algebraic operations on the predictive variance, serving as a performance benchmark.
• Leverage the conditional distribution property of multivariate Gaussians to model the relationship between RSS and user location in the GP framework.

Experimental results

Research questions

• RQ1 Why does conventional GP regression produce unrealistically small 2σ error bars when applied to noisy test RSS data in user positioning?
• RQ2 Can a numerical approximation method be developed to produce realistic uncertainty estimates for user location predictions in RSS-based positioning with noisy test data?
• RQ3 How does the performance of the proposed NaGP method compare to conventional GP in terms of RMSE and uncertainty estimation?
• RQ4 To what extent does operating in the massive MIMO regime improve positioning accuracy compared to conventional MIMO?
• RQ5 How close is the achievable RMSE performance of the proposed methods to the theoretical Bayesian Cramer-Rao lower bound (BCRLB)?

Key findings

• The conventional GP (CGP) method produces unrealistically small 2σ error bars because it ignores the noise in test RSS data, leading to overconfident uncertainty estimates.
• The proposed NaGP method provides realistic 2σ error bars by accounting for the statistical properties of shadowing noise in the test RSS, resulting in more reliable uncertainty quantification.
• Operation in the massive MIMO regime significantly improves RMSE performance compared to conventional MIMO, due to the increased number of antennas and richer signal vector information.
• The RMSE performance of the NaGP method is very close to the derived Bayesian Cramer-Rao lower bound (BCRLB), indicating near-optimal estimation performance.
• The BCRLB can be computed via simple linear algebraic operations on the predictive variance, making it a practical benchmark for evaluating positioning system performance.
• The study reveals that training with noise-free RSS data, while convenient, incurs a performance trade-off in terms of higher RMSE, suggesting a need for improved approximation methods in future work.