Estimation of Structured Wireless Channels
To exploit the full potential of future massive MIMO or millimeter wave communication systems, it is crucial to have accurate channel estimates, i.e., to have accurate measurements of the characteristics of the propagation scenario. Only with reliable channel state information (CSI), the base station can form sharp beams to the desired users in order to avoid inter-user interference and to benefit from the array gain of a large antenna array.
Algorithms for channel estimation can be improved by exploiting side information about the channel, such as statistical information (e.g., covariance matrices) or the assumption that the channel is a superposition of a small number of signal paths. In addition, in scenarios where such mathematical models are not accurate enough to reflect the actual propagation environment, further side information about the channel structure could be extracted from data collected in a measurement campaign. To this end, machine learning techniques like the autoencoder may be used. Autoencoders have been successfully applied to the problem of finding structured representations of data in the context of image processing. For this reason, we consider them as a promising tool in future studies.
Our research activities include investigating how channel models can be exploited to design neural-network-based channel estimation algorithms. Another topic is channel estimation in frequency division duplex (FDD) systems, where the downlink channel cannot be easily deduced from the uplink channel estimates due to a lack of reciprocity between the channel realizations. Further interests lie in designing CSI feedback in MIMO systems.