Massive machine-type communications protocols have typically been designed under the assumption that coordination between users requires significant communication overhead and is thus impractical. Recent progress in efficient activity detection and collision-free scheduling, however, indicates that the cost of coordination can be much less than the naive scheme for scheduling. This work considers a scenario in which a massive number of devices with sporadic traffic seek to access a massive multiple-input multiple-output (MIMO) base-station (BS) and explores an approach in which device activity detection is followed by a single common feedback broadcast message, which is used both to schedule the active users to different transmission slots and to assign orthogonal pilots to the users for channel estimation. The proposed coordinated communication scheme is compared to two prevalent contention-based schemes: coded pilot access, which is based on the principle of coded slotted ALOHA, and an approximate message passing scheme for joint user activity detection and channel estimation. Numerical results indicate that scheduled massive access provides significant gains in the number of successful transmissions per slot and in sum rate, due to the reduced interference, at only a small cost of feedback.

Consider a massive random access scenario in which a small set of$k$active users out of a large number of$n$potential users need to be scheduled in$b\ge k$slots. What is the minimum common feedback to the users needed to ensure that scheduling is collision-free? Instead of a naive scheme of listing the indices of the$k$active users in the order in which they should transmit, at a cost of$k\log (n)$bits, this paper shows that for the case of$b=k$, the rate of the minimum fixed-length common feedback code scales only as$k \log (e)$bits, plus an additive term that scales in$n$as$\Theta (\log \log (n))$for fixed$k$. If a variable-length code can be used, assuming uniform activity among the users, the minimum average common feedback rate still requires$k \log (e)$bits, but the dependence on$n$can be reduced to$O(1)$. When$b>k$, the number of feedback bits needed for collision-free scheduling can be significantly further reduced. Moreover, a similar scaling on the minimum feedback rate is derived for the case of scheduling$m$users per slot, when$k \le mb$. The problem of constructing a minimum collision-free feedback scheduling code is connected to that of constructing a perfect hashing family, which allows practical feedback scheduling codes to be constructed from perfect hashing algorithms.