TR2018-060

Reduced-Dimension Symbol Detection in Random Access Channel


Abstract:

In a growing number of IoT applications, an access point communicates with IoT users whose number prevents assignment of orthogonal channel resources to them, and renders network management and control impractical. Consequently, symbols transmitted from active users collide, which necessitates their separation on the receiver side. Given that an IoT user transmits with low probability, the sparsity in the user activity domain has been exploited in a number of previous works, where the symbol separation problem is formulated and solved as a sparse recovery problem. However, an excessive computational complexity remains a challenge in such approaches, where the number of filters in the equivalent receiver filter bank is equal to the overall number of users. Consequently, reduced-dimension processors facilitating low complexity symbol separation are sought. We propose here two novel reduced-dimension processors. In addition, a scheme, which aims to remove an impractical requirement that the receiver knows channels from all users is proposed and studied in conjunction with the reduced-dimension processors. Furthermore, pre-whitened reduced-dimension processors and a variety of approaches for the design of dimensionality reduction transformation matrix are considered. Finally, the results are validated with simulations. As such, the tests show that the proposed processors considerably outperform the benchmark, and also that the pre-whitened processors outperform their nonprewhitened counterparts.