Covariance-based sparse estimation with applications to channel sounding and massive random access for IoT applications

Speaker: Giuseppe Caire, TU Berlin

Consider the following problem: we observe multiple noisy samples from a random vector which can be represented as the linear combination of a small number of given dictionary vectors, and we want to identify which elements of the dictionary are "active'', i.e., they are associated to a non-negligible received signal power. This problem may arise in estimating the scattering field as a function of the angle of arrival in an array processing setting, or estimating which users are active in the context of grant-free random access where a very large number of nodes (e.g., sensors) have very sporadic data to transmit. We consider the mathematics of such estimation problem and in particular provide new very efficient algorithms for both scenarios, which go beyond the conventional compressed sensing limits. The key common feature to be leveraged is that the relevant information in this context is contained in the covariance matrix of the received signal. Exploiting the structure of the problem, we can come up with structured covariance estimators that do better than conventional schemes proposed in the literature. Applications to massive MIMO, channel sounding, and massive random access for IoT systems will be discussed.