Learning Based Reconfigurable Sub-nyquist Sampling Framework for Ultra-wideband Angular Sensing

Reconfigurable and Intelligent Ultra-Wideband Angular Sensing: Prototype Design and Validation

“…Himani Joshi and Sumit J. Darak are with Electronics and Communications Department, IIIT-Delhi, India-110020 (e-mail: {himanij,sumit}@iiitd.ac.in), Mohammad Alaee-Kerahroodi and Bhavani Shankar Mysore Rama Rao are with Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg, Luxembourg-1359 (e-mail: {mohammad.alaee,Bhavani.Shankar}@uni.lu) of active transmissions in the wideband spectrum and hence leads to sensing failure (or characterization failure) whenever the number of active transmissions exceeds the number of ADCs. To overcome this drawback, we recently proposed reconfigurable SNS [9], [10] which unlike the traditional SNS methods [3]- [8], can dynamically digitize the non-contiguous bands of the wideband spectrum for a given number of ADCs and its architecture is based on non-contiguous SNS [14]. Reconfigurable SNS allows skipping of any number of bands in the wideband spectrum, and this is especially required when some part of the spectrum may not be useful due to high traffic, security constraints and high propagation loss.…”

“…Reconfigurable SNS allows skipping of any number of bands in the wideband spectrum, and this is especially required when some part of the spectrum may not be useful due to high traffic, security constraints and high propagation loss. This enables the digitization of the spectrum significantly wider than that of the conventional SNS thereby leading to higher spectrum transmission opportunities (and hence, high network throughput) [9], [10]. Reconfigurable SNS demands additional intelligence to dynamically learn the spectrum statistics and select appropriate bands of the spectrum for digitization which can be accomplished using learning algorithms [9], [10].…”

“…This enables the digitization of the spectrum significantly wider than that of the conventional SNS thereby leading to higher spectrum transmission opportunities (and hence, high network throughput) [9], [10]. Reconfigurable SNS demands additional intelligence to dynamically learn the spectrum statistics and select appropriate bands of the spectrum for digitization which can be accomplished using learning algorithms [9], [10].…”

“…We explore multi-play multi-armed bandit based learning algorithm to embed intelligence needed for dynamic selection of the noncontiguous spectrum in reconfigurable SNS. The proposed approach offers significant savings in receiver complexity over [3], [5], [8], [9] and feasible hardware architecture than [3], [7], [8], [10]. 3) Compared to previous works, the proposed spectrum characterization (frequency band status and DoA estimation) approach does not need prior knowledge of received signal distribution and spectrum statistics.…”

“…Such prototype and subsequent experimental performance analysis of various UWAS approaches offer important insights which may not be possible in conventional simulation-based experiments. Other than prototype design, the proposed work also offers new contributions at the algorithm level compared to our previous works in [9], [10]. The proposed approach exploits sparse-array, unlike the traditional uniform linear array (ULA) in [9] and offers a higher number of DoA estimations than [9].…”