Predicting Lifespan of Ground-to-Air Multipath Components in mmWave UAV Channels
Wahab Khawaja 1 , Rune H. Jacobsen 1 , Sajid Hussain 2 , and Ismail Guvenc 3 · 2025
Read Predicting Lifespan of Ground-to-Air Multipath Components in mmWave UAV Channels by Wahab Khawaja 1 , Rune H. Jacobsen 1 , Sajid Hussain 2 , and Ismail Guvenc 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-09044.txt.
Introduction
Radio signals bouncing between a ground station and a drone are born and die as the drone moves, breaking the usual steady-channel assumptions. Wahab Khawaja and co-authors set out to predict how long each reflected path will live. They turn each path's trace across receiver positions into a time series, grouped into path bins.
A semi-Markov model captures how bins evolve. Prediction methods range from weighted sums and classic classifiers to deep neural networks. The networks win clearly, forecasting each path's lifespan, including its death, with high accuracy. The 2025 work helps engineers design links for drones that respect how fast the channel rewrites itself.
Catalog record
| CREATOR | Wahab Khawaja 1 , Rune H. Jacobsen 1 , Sajid Hussain 2 , and Ismail Guvenc 3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-09044
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Predicting Lifespan of Ground-to-Air Multipath Components in mmWave UAV Channels Wahab Khawaja 1 , Rune H. Jacobsen 1 , Sajid Hussain 2 , and Ismail Guvenc 3 Abstract In mobile ground-to-air (GA) propagation channels, the birth and death of multipath components (MPCs) are frequently observed, and the wide-sense stationary uncorrelated scattering (WSSUS) assumption does not always hold. Several methods exist for tracking the birth and death of MPCs, however, to the best of knowledge of authors, there is no existing literature that addresses the prediction of the lifespan of the MPCs in non-WSSUS GA propagation channels. In this work, we consider the GA channel as non-WSSUS and individual MPCs across receiver positions are represented as time series based on the Euclidean distance between channel parameters of the MPCs. These time series representations, referred to as path bins, are analyzed using a semi-Markov chain model. The channel parameter variations and dependencies between path bins are used to predict the lifespan of path bins using weighted sum method, machine learning classifiers, and deep neural networks. For comparison, the birth and death of path bins are also modeled using a Poisson distribution and a Markov chain. Simulation results demonstrate that deep neural networks offer highly accurate predictions for the lifespan (including death) of MPC path bins in the considered GA propagation scenario. I Introduction Ground-to-air (GA) communications using unmanned aerial vehicles (UAVs) have secured significant attention in the recent decade [ 1 , 2 ] . GA communications using UAVs are envisioned for 5 G and beyond [ 3 , 4 ] . The integration of 5G terrestrial networks with non-terrestrial networks (NTN) including UAVs is highlighted in the latest release of the 3GPP [ 5 ] . The 5G and beyond communication networks are expected to dominantly use millimeter wave (mmWave) frequencies. In the mmWave mobile GA propagation channel, there are sparse multipath components (MPCs) that experience frequent births and deaths [ 6 , 7 , 8 ] . Accurate prediction of lifespan of MPCs in GA channels is essential for maintaining reliable links and optimizing resources in NTN. This enables better handovers and minimizes link failures under dynamic conditions. There are different methods available in the literature for tracking the lifespan of MPCs including their birth and death [ 9 , 10 , 11 ] . In [ 9 ] , nonlinearities due to the birth and death of MPCs in a vehicular propagation channel were detected and tracked using a modified Kalman filter. In [ 10 ] , the analysis of the birth and death of MPCs in a time-variant channel was carried out using spatial point process. Assumption of wide sense stationary channel transfer function using point process perspective helped in the analysis. In [ 11 ] , a statistical model for time-varying UAV air-to-ground (AG) channel was presented. The stochastic model provided characterization of the parameters of the birth and life of MPCs. Furthermore, it was shown that regular patterns can be used to describe the evolution of MPCs. Birth and death of MPCs was also analyzed using AG propagation channel measurements in [ 12 ] . However, to the best of knowledge of authors, no existing study has focused on predicting the lifespan of MPCs in mobile GA propagation channels, motivating this research to bridge that gap. This work highlights the need to address the non-WSSUS nature of GA channels in vehicular communications [ 13 ] , where UAV motion induces non-WSS, and static scatterers create correlated MPCs. (a) (b) (c) Fig. 1 : The GA propagation scenario showing life span, and birth and death instances of MPCs along the UAV trajectory. The contributions of this work are as follows: We analyze a mobile GA directional propagation channel with scatterers positioned on one side shown in Fig. 1 . First, the tracking of the birth and death of MPCs is obtained by organizing the MPCs as respective time series over receiver (RX) positions. The time series representation of a MPC called the path bin is obtained using the Euclidean distance among the channel parameters of MPCs at different RX positions [ 8 ] highlighted in Fig. 1 . The lifespan of path bins defined as the number of RX positions covered until their death is predicted using predictive modeling based on the analysis of the characteristics of the path bins and the correlation among the path bins. Weighted sum (WS) models, machine learning (ML), and deep neural networks (DNN) are used to predict the lifespan of path bins. The birth and death of path bins are also modeled as independent processes using Poisson distribution and Markov chain. The prediction performance of different methods is then compared. Geometric ray-based method is used for simulations. The outline of the paper is as follows, Section II provides the time series representation of MPCs as path bins over the RX positions, Section III provides different methods for the prediction of lifespan of MPC path bins, the simulation setup and results are discussed in Section IV , and Section V concludes the paper. II Time Series Representation of MPCs We consider our GA propagation channel shown in Fig. 1 to be a time-varying channel with correlated scattering. We model our channel \boldsymbol{H(t,n)} as follows: \displaystyle\boldsymbol{H(t,n)} \displaystyle=\sum_{m=1}^{M_{n}}\alpha_{m,n}(t)\delta\big(t-\tau_{m,n}\big)\delta\big(\boldsymbol{\theta^{(\rm T)}}-\theta_{m,n}^{(\rm T)}\big) \displaystyle\times\delta\big(\boldsymbol{\phi^{(\rm T)}}-\phi_{m,n}^{(\rm T)}\big)\delta\big(\boldsymbol{\theta^{(\rm R)}}-\theta_{m,n}^{(\rm R)}\big)\delta\big(\boldsymbol{\phi^{(\rm R)}}-\phi_{m,n}^{(\rm R)}\big) \displaystyle\times\exp{\bigg(\frac{-j2\pi\Delta d_{m,n}}{\lambda}\bigg)}~, (1) where H(t,n) is the channel impulse response at time instance t at n^{\rm th} RX position, \alpha_{m,n} and \tau_{m,n} are the complex amplitude and delay of the m^{\rm th} MPC at n^{\rm th} RX position, respectively, \boldsymbol{\theta^{(\rm T)}} and \boldsymbol{\phi^{(\rm T)}} are the elevation and azimuth angle vectors at the transmitter (TX), respectively, \boldsymbol{\theta^{(\rm R)}} , \boldsymbol{\phi^{(\rm R)}} are the elevation and azimuth angle vectors, respectively, at the RX, \theta_{m,n}^{(\rm T)} , \phi_{m,n}^{(\rm T)} are the elevation and the azimuth angles of the m^{\rm th} MPC corresponding to n^{\rm th} RX position at the TX, respectively, \theta_{m,n}^{(\rm R)} , \phi_{m,n}^{(\rm R)} , are the elevation and the azimuth angles of the m^{\rm th} MPC corresponding to n^{\rm th} RX position at the RX, respectively, n=1,2,\cdots,N and M_{n} are the total number of RX positions and MPCs at n^{\rm th} RX, respectively, and \Delta d_{m,n}=d_{m,n}-d_{0} , where d_{m,n} is the distance of the m^{\rm th} MPCs at n^{\rm th} RX position, d_{0} is the reference distance. The phase of the MPCs \big(\frac{2\pi\Delta d_{m,n}}{\lambda}\big) at different RX positions are assumed to be uniformly distributed between [0 - 2 \pi ). The MPCs observed at the RX exhibit strong correlations because of dependencies in channel parameters across RX positions along the straight-line trajectory. The predictable variation in MPC parameters, caused by consistent changes in distance to scatterers, supports effective prediction of MPC lifespan through Euclidean distance and correlation-based methods. The correlation among the channel parameters of MPCs at n^{\rm th} and k^{\rm th} RX positions can be represented by the matrix \boldsymbol{R} defined as \boldsymbol{R}=\begin{bmatrix}{\textbf{c}}_{1,1}&{\textbf{c}}_{1,2}&{\textbf{c}}_{1,3}&\cdots&{\textbf{c}}_{1,M_{k}}\\ {\textbf{c}}_{2,1}&{\textbf{c}}_{2,2}&{\textbf{c}}_{2,3}&\cdots&{\textbf{c}}_{2,M_{k}}\\ \vdots&\vdots&\vdots&\cdots&\vdots\\ {\textbf{c}}_{M_{n},1}&{\textbf{c}}_{M_{n},2}&{\textbf{c}}_{M_{n},3}&\cdots&{\textbf{c}}_{M_{n},M_{k}}\end{bmatrix}~, where {\textbf{c}}_{m_{n},m_{k}}~{\rm for}~n\neq k is the correlation vector representing correlation among the channel parameters of the MPCs, m_{n} and m_{k} at the n^{\rm th} and k^{\rm th} RX positions, respectively. This vector is given by \displaystyle{\textbf{c}}_{m_{n},m_{k}}= \displaystyle\big[c_{m_{n},m_{k}}^{(\alpha)},~c_{m_{n},m_{k}}^{(\tau)},~c_{m_{n},m_{k}}^{(\theta^{(\rm T)})},~c_{m_{n},m_{k}}^{(\phi^{(\rm T)})},~c_{m_{n},m_{k}}^{(\theta^{(\rm R)})},~c_{m_{n},m_{k}}^{(\phi^{(\rm R)})}\big], (2) where c_{m_{n},m_{k}}^{(\alpha)},~c_{m_{n},m_{k}}^{(\tau)},~c_{m_{n},m_{k}}^{(\theta^{(\rm T)})},~c_{m_{n},m_{k}}^{(\phi^{(\rm T)})},~c_{m_{n},m_{k}}^{(\theta^{(\rm R)})},~c_{m_{n},m_{k}}^{(\phi^{(\rm R)})} are