| dc.description.abstract | Massive machine-type communications (mMTC) is a 5G and beyond use case, where the network is expected to serve millions of devices per square kilometre. Typical mMTC devices include smart energy meters, pressure sensors, temperature indicators, smart factory equipment, etc. These devices sporadically transmit short packets, i.e., they transmit a short burst of data once in a while and then largely remain inactive. In order to serve mMTC scenarios, we need to use grant-free random access (GFRA) protocols since they have the advantage of a low control and signalling overhead as well as non-orthogonal use of the channel. GFRA for mMTC is a relatively new research topic and has received immense interest in the recent past. In this thesis, we analyze several practical aspects of irregular repetition slotted aloha (IRSA), which is a GFRA protocol for mMTC.
IRSA is a distributed GFRA protocol where users transmit multiple replicas of their packets in randomly selected resource blocks within a frame to a base station (BS). The BS recovers the packets using successive interference cancellation (SIC). Existing studies have analyzed IRSA with idealized assumptions, i.e., neglecting fading, path-loss, channel estimation errors, pilot contamination, multi-cell interference, etc. These non-idealities can greatly reduce the performance of the system and must be accounted for in the design and analysis of any mMTC system.
In this thesis, we first analyze channel estimation in IRSA, exploiting the sparsity structure of IRSA transmissions, when non-orthogonal pilots are employed across users to facilitate channel estimation at the BS. Allowing for the use of non-orthogonal pilots is important, as the length of orthogonal pilots scales linearly with the total number of devices, leading to prohibitive overhead as the number of devices increases. Next, we present a novel analysis of the throughput of IRSA under practical channel estimation errors, and with the use of multiple antennas at the BS. Finally, we theoretically characterize the asymptotic throughput of IRSA using a density evolution based analysis. Simulation results underline the importance of accounting for channel estimation errors in analyzing IRSA, which can even lead to 70% loss in performance in severely interference-limited regimes. We also provide novel insights on the effect of parameters such as pilot length, SNR, number of antennas at the BS, etc, on the system throughput.
Next, we develop a novel Bayesian user activity detection (UAD) algorithm for IRSA, which exploits both the sparsity in user activity as well as the underlying structure of IRSA transmissions. We then derive the Cramer-Rao bound (CRB) on the mean squared error in channel estimation. We empirically show that the channel estimates obtained as a by-product of the proposed UAD algorithm achieves the CRB. Then, we analyze the signal to interference plus noise ratio achieved by the users, accounting for UAD, channel estimation errors, and pilot contamination. Finally, we illustrate the impact of these non-idealities on the throughput of IRSA via Monte Carlo simulations. For example, in a system with 1500 users and 10% of the users being active per frame, a pilot length of as low as 20 symbols is sufficient for accurate user activity detection. In contrast, using classical compressed sensing approaches for UAD would require a pilot length of about 346 symbols. Our results reveal crucial insights into dependence of UAD errors and throughput on parameters such as the length of the pilot sequence, the number of antennas at the BS, the number of users, and the SNR.
Then, we develop an enhanced version of IRSA that can be operated at the peak performance even at high system loads. IRSA can be used to serve a large number of users in mMTC while achieving a near-zero packet loss rate (PLR). However, in overloaded mMTC scenarios, the system is interference-limited, and the PLR is close to one. We develop a variant of IRSA in the interference limited-regime, namely Censored-IRSA (C-IRSA), in which users with poor channel states self-censor, i.e., they refrain from transmitting their packets. This censoring depends on a censor threshold that can be varied depending on the number of users in the system. Firstly, we empirically and theoretically analyze the performance of C-IRSA. Next, we derive the optimal choice of the censor threshold via a semi-analytic approach and a PLR-optimal algorithmic approach. This choice of the threshold maximizes the throughput while achieving zero PLR among uncensored users. Through extensive numerical simulations, we show that C-IRSA operates at full system throughput at high system loads compared to vanilla IRSA which has near-zero throughput.
After this, we analyze IRSA in the multi-cell (MC) and cell-free (CF) setups, accounting for pilot contamination, channel estimation errors, and multi-user interference. Via extensive simulations, we illustrate that, in practical settings, MC IRSA can have a drastic loss of throughput, up to 70%, compared to SC IRSA. Further, MC IRSA requires a significantly higher training length, in order to support the same user density and achieve the same throughput: for example, MC IRSA may need about 4−5x compared to SC IRSA. We provide insights into the effect of system parameters such as number of antennas, pilot length, and SNR on the throughput of MC IRSA and CF IRSA. With the proposed CF architectures, we show that we can achieve more than 14x improvement in the throughput of CF IRSA compared to a massive MIMO SC setup. We also study the densification trends in MC IRSA, where we observe an inverse behaviour in the throughput compared to CF IRSA.
Finally, we optimize the repetition distributions in IRSA with the throughput and the energy efficiency objectives. Via extensive numerical simulations, we study the effect of various system parameters such as the maximum repetition factor, the average repetition factor, the number of antennas, and the pilot length, on the repetition distributions, the inflection load, and the peak energy efficiency. Compared to the best existing distributions, we show that our optimized distributions can achieve up to 58% increase in the inflection load and up to 49% increase in the peak energy efficiency.
Overall, this thesis analyzes and designs the IRSA protocol under several practical non-idealities. The developed algorithms vastly outperform state-of-the-art and can efficiently serve mMTC applications. | en_US |
