This dissertation treats a recently introduced modern random access protocol known as unsourced random access (U-RA). This protocol belongs to the family of grant-free random access protocols which, in contrast to the established grant-based protocols of current mobile communication standards, do not rely on an initial access phase, in which the active devices identify themselves and await the grant of dedicated transmission resources (e.g. time-frequency blocks) from the base station (BS). When messages are short and the number of active devices is large, which is a central specification of upcoming massive machine-type-communication (mMTC) scenarios, such a grant-based procedure is overly wasteful and may increase the delay to an unacceptable level. In a grant-free scenario the active devices transmit their payload right away without awaiting the grant of dedicated resources. A commonly discussed approach is to assign devices unique fixed identification sequences (called pilots in the following). Active devices then transmit their pilot followed directly by their message. The idea of unsourced random access is to go one step further and, ideally, abolish the identification step, such that active devices directly transmit a message from a common predefined message set. Information-theoretically such a behavior is captured by letting each user employ the same codebook, as opposed to the classical information-theoretic treatment of the multiple access problem where each user is assigned an individual codebook. The assumption of a common codebook allows to treat the random access problem in a way that captures the effect of short messages while still taking into account the physical properties of the channel. In this work I build on a previously introduced coding scheme for the U-RA problem on the AWGN channel termed coded compressed sensing (CCS). I introduce a novel decoding algorithm for the CCS scheme based on approximate message passing (AMP) and give an asymptotic analysis. The analysis shows that it is possible, under optimal decoding, to achieve the fundamental communication limit on the AWGN channel even when arbitrary many users communicate without any coordination. Furthermore, I show that the low-complexity AMP algorithm cannot achieve this limit with a uniform power allocation. Instead, an optimized non-uniform power allocation is necessary to improve the performance of the AMP algorithm beyond a certain limit. I use the developed analysis to derive a method to find such power allocation. Numerical evaluation show that the proposed scheme is efficient and the analysis is precise. In the second part of the work I present an extension of the U-RA idea to a block-fading AWGN channel with a massive MIMO receiver. I introduce and study a modified CCS scheme that uses a covariance based decoding algorithm. The analysis shows that the sum-spectral-efficiency of such a U-RA system can grow as O(L/ log L), where L is the length of a coherence block. In the course of the analysis I show that the self Khatri-Rao product of random spherical matrices satisfies the so called restricted isometry property, which is a result of independent interest and find applications in e.g. the analysis of the limits of direction-of-arrival estimation. Remarkably, the CCS scheme shows to work very well even in a high mobility scenario with short coherence block length L ≈ 100 and hundreds of concurrent active users. In such a fast fading scenario pilot based random access would completely fail for this number of users. On the other hand, if the coherence block length is large enough, I show that a more conventional scheme based on randomly chosen pilots and maximum-ratio-combining can efficiently implement the U-RA idea in a massive MU-MIMO setting. Finally, I show that the CCS scheme can be deployed in an distributed setting, where multiple BSs can decode the messages of many active users with minimal corporation between the BSs and no required cell management ("cell-free"). Furthermore, I present a novel iterative scheme that allows to jointly detect the activity and the geographic positions of the active users. The practical performance of all the proposed schemes is evaluated by numerical simulations.
Sparse recovery based grant-free random access for massive machine-type communication
Zulassungsfreier zufälliger Zugriff basierend auf der Wiederherstellung dünnbesetzter Signale für massive maschinenartige Kommunikation
2021
Sonstige
Elektronische Ressource
Englisch