Delay Analysis for Wireless Fading Channels with Finite Blocklength Channel Coding

Abstract

Upcoming low-latency machine-to-machine (M2M) applications are currently attracting a significant amount of interest from the wireless networking research community. The design challenge with respect to such future applications is to allow wireless networks to operate extremely reliably at very short deadlines for rather small packets. To date, it is unclear how to design wireless networks efficiently for such novel requirements. One reason is that existing performance models for wireless networks often assume that the rate of the channel code is equal to the Shannon capacity. However, this model does not hold anymore when the packet size and thus blocklength of the channel code is small. Although it is known [1] that finite blocklength has a major impact on the physical layer performance, we lack higher-layer performance models which account in particular for the queueing effects under the finite blocklength regime.A recently developed methodology [2] provides probabilistic higher-layer delay bounds for fading channels when assuming transmission at the Shannon capacity limit. Based on this novel approach, we develop service process characterizations for fading channels with finite blocklength channel coding, leading to novel probabilistic delay bounds that can give a fundamental insight into the capabilities and limitations of wireless networks when facing low-latency M2M applications. In particular, we show that the Shannon capacity model significantly overestimates the delay performance for such applications, which would lead to insufficient resource allocations. Finally, based on our (validated) analytical model, we study various important parameter trade-offs highlighting the sensitivity of the delay distribution under the finite blocklength regime.

Deep Learning for Hybrid 5G Services in Mobile Edge Computing Systems: Learn From a Digital Twin

“…and the approximation V u k ≈ 1 is applied, which is accurate when the receive SNR is higher than 5 dB [28,29]. To satisfy the maximal transmit power constraint, we can obtain the minimal g th,u k by substituting (24) into P t,u k = P max,u , i.e.,…”

Learning to Optimize with Unsupervised Learning: Training Deep Neural Networks for URLLC

“…1) Achievable Rate in Finite Blocklength Regime: In URLLC, the blocklength of channel coding is short due to the short transmission duration, and hence the impact of decoding errors on reliability cannot be ignored. Shannon's capacity formula cannot be employed to characterize the probability of decoding errors [11]. The achievable rate in finite blocklength regime is required.…”

“…Although the achievable rate is in closed-form, it is still too complicated to obtain graceful results. As shown in [11], if the signal-to-noise ratio (SNR) α k g k Pmax N0W k ≥ 5 dB, V k ≈ 1 is accurate. Since high SNR is required to ensure ultra-high reliability and ultra-low latency, such approximation is reasonable.…”

Unsupervised Deep Learning for Ultra-Reliable and Low-Latency Communications

“…In URLLC, the blocklength of channel coding is short due to the short transmission duration, and hence the impact of decoding errors on reliability cannot be ignored. Since Shannon's capacity formula cannot be employed to characterize the probability of decoding errors [16], we consider the achievable rate in finite blocklength regime. In quasi-static flat fading channels, when channel state information is available at the transmitter and receiver, the achievable rate of the kth user (in packets/frame) can be accurately approximated by [12],…”

“…Although the achievable rate is in closed-form, it is still too complicated to obtain graceful results. As shown in [16], if the signal-to-noise ratio (SNR) α k g k P k N0W k ≥ 5 dB, V k ≈ 1 is accurate. Since high SNR is required to ensure ultra-high reliability and ultra-low latency, such approximation is reasonable.…”