Stochastic Resource Allocation for Outage Minimization in Random Access with Correlated Activation

Abstract

A key challenge for random access communications arising in the monitoring of physical phenomena is optimizing the access policy. This is particularly the case when the activity of each sensor is correlated, contrasting with the independence assumption underpinning standard slotted ALOHA schemes. In this paper, we propose a stochastic resource allocation algorithm to reduce outages via maximization of the expected number of sensors that are able to reliably communicate with an access point. Allowing for devices to transmit data over multiple consecutive frames, we show that the proposed algorithm converges with probability one to a locally optimal solution. Moreover, our algorithm significantly outperforms existing methods in terms of the average number of successful transmissions when utilizing successive interference cancellation.

Mitigating User Identification Errors in Resource Optimization for Grant-Free Random Access

“…• We demonstrate that the stochastic resource optimization algorithms recently proposed in [4], [5] can be highly susceptible to user identification errors, leading to suboptimal performance. This is significant as in the absence of user identification errors, these algorithms are optimal in contrast with other heuristic approaches [3] and do not assume perfect knowledge of activity probabilities as in [4], [5].…”

“…• We demonstrate that the stochastic resource optimization algorithms recently proposed in [4], [5] can be highly susceptible to user identification errors, leading to suboptimal performance. This is significant as in the absence of user identification errors, these algorithms are optimal in contrast with other heuristic approaches [3] and do not assume perfect knowledge of activity probabilities as in [4], [5]. • To mitigate the impact of the bias arising from user identification errors, we use importance sampling in a manner analogous to sample selection bias correction [9] and bias reduction in private synthetic data [10].…”

“…In order to avoid high levels of control signaling in massive access systems, it is desirable to provide devices with a timeslot selection policy [2], [3], [4], [5]. This means that when a device has data to transmit, limited control signaling is required in order to determine which resources the device should use.…”

“…In this paper, we consider the impact of imperfect estimation of active devices on resource allocation based on the expected throughput. Our focus is on the problem of designing strategies for allocating a single time slot within a frame to an active device [3], [4], [5], relevant for NB-IoT. In this setting, a key challenge is to account for uncertainties in user activity probabilities, which are ignored in most existing work [2], [3], [4], [5] on resource allocation and coding for random access.…”

“…• To mitigate the impact of the bias arising from user identification errors, we use importance sampling in a manner analogous to sample selection bias correction [9] and bias reduction in private synthetic data [10]. We prove that by weighting the gradient estimates in the algorithms in [4], [5], the resource allocation converges to a stationary point with probability one. • In practice, the weight for the gradient estimates requires the evaluation of the true activity distribution and the imperfect activity distribution induced by erroneous user identification.…”