Principal Research Engineer
Huawei Munich Research Center

Email: firstName.lastName (at) live.com
Bio , CV , LinkedIn , Google Scholar
Research interests
  • Wireless communications
  • Vehicular communications (V2X)
  • Resource allocation
  • Channel modeling
  • Reinforcement learning

Current Research Projects


Channel modeling


The work on channel modeling is predominantly focused on analysing the peculiarities of V2V channels, in particular how the V2V channel is impacted by the presence of surrounding vehicles. Since measurements showed considerably different channel characteristics in terms of both large and small scale parameters, we classify the V2V links into three types: line of sight (LOS), non-LOS due to vehicles (NLOSv), and non-LOS due to static objects (e.g., buildings, foliage). Over the years, the work on this topic ranged from city-wide characterization of received power through efficient deterministic modeling, to stochastic analysis of state transitions, to path loss, shadow fading, and fast fading characterization of LOS, NLOSv, and NLOS channels for geometry-based stochastic channel modeling.

I am also working on the prediction of channel characteristics through data analysis. In particular, we have looked into whether V2V measurement data containing additional features (e.g., distance, Tx-Rx locations, type of environment, time of day, obstruction type) can be used to estimate the different aspects of radio channels (including path loss, fading, and interference) in a non-parametric fashion.

Vehicle blockage measurments GEMV^2 Comparison with Log distance path loss model Vehicle blockage measurments

Albeit in a simplified form, the following aspects of my work on channel models found its way into 3GPP (TR 37.885) and ETSI (TR 103-257-1) standard:

Resource allocation


Resource allocation is a difficult problem even when the network is static and the connections between the scheduler and users are highly reliable. In case of a higly mobile vehicular networks with varying propagation conditions, the problem becomes particularly challenging. Adding to the mix patchy network coverage exacerbates the problem further. State of the art typically employs a combination of heuristics and local sensing to allocate the resources and typically trades off reliability and efficiency to achieve.

We investigate whether a centralized learning scheduler can be taught to efficiently pre-assign the resources to vehicles for-of-coverage V2V communication. Specifically, we employ a reinforcement learning (RL) scheduler that interacts with the vehicular environment and learns to predict possible resource collisions and half-duplex limitations so that it can efficiently (pre-)allocate the resources to vehicles. Our initial results (described in IEEE VTC Spring '18, IEEE VNC '18, and VTC Fall '19) are encouraging, showing that the employed Vehicular Reinforcement Learning Scheduler (VRLS) can learn efficient strategies for assigning resources, even in case of uncertainty caused by varying radio conditions, vehicular traffic density, and vehicle speeds.

Vehicular Reinforcement Learning Scheduler Vehicular Reinforcement Learning Scheduler


Code

In addition to standardized models, some of the research I worked on found its way into freely available code:

Measurement data


Students

I have been fortunate to work with some extraordinary students.

Current students


Past students


Ph.D. students

Undergrad and M.S. students at CMU

Older projects


Some of my older work can be found on this link.