Periodic Reporting for period 4 - LOLITA (Information Theory for Low-Latency Wireless Communications)
Periodo di rendicontazione: 2021-09-01 al 2023-02-28
The majority of wireless connections in the next generations of wireless systems will most likely be originated by autonomous machines and devices rather than by the human-operated mobile terminals for which traditional broadband services are intended. It is thus expected that enhanced mobile-broadband services will be complemented by new services centered on machine-type communications (MTC). An important emerging area among MTC systems is that of low-latency communications, which targets systems that require reliable real-time communication with stringent requirements on latency and reliability.
The design of low-latency wireless communication systems is a great challenge, since it requires a fundamentally different design approach than the one used in current high-rate systems. Indeed, current systems exchange packets of several thousand bits. For such packet lengths, there are error-correcting codes that can correct transmission errors with high probability at rates close to the channel capacity. Consequently, the design of current systems is supported by the extensive information-theoretical knowledge we have about wireless communications. In contrast, low-latency systems exchange packets of only several hundred bits, so the rate of the error-correcting code must be significantly below the capacity to achieve the desired reliability. Consequently, for such systems capacity is not a relevant performance measure, and design guidelines that are based on its behavior will be misleading. Currently, we are lacking the theoretical understanding of low-latency wireless communication systems that would be crucial to design them optimally. LOLITA addresses this problem by establishing the theoretical framework required to describe the fundamental tradeoffs in low-latency wireless communications.
In particular, by applying and advancing methods from Finite Blocklength Information Theory, LOLITA has developed:
1) Rigorous closed-form approximations of the maximum coding rate of wireless communication channels.
2) Characterizations of the fundamental limits of massive random-access systems.
3) Rigorous and efficiently to compute approximations of state-of-the-art bounds on the maximum coding rate of wireless communication channels.
On the one hand, the obtained results provide accurate performance benchmarks for system designers of next-generation wireless communication systems. On the other hand, they provide invaluable insights on the optimal design of such systems.
The design of low-latency wireless communication systems is a great challenge, since it requires a fundamentally different design approach than the one used in current high-rate systems. Indeed, current systems exchange packets of several thousand bits. For such packet lengths, there are error-correcting codes that can correct transmission errors with high probability at rates close to the channel capacity. Consequently, the design of current systems is supported by the extensive information-theoretical knowledge we have about wireless communications. In contrast, low-latency systems exchange packets of only several hundred bits, so the rate of the error-correcting code must be significantly below the capacity to achieve the desired reliability. Consequently, for such systems capacity is not a relevant performance measure, and design guidelines that are based on its behavior will be misleading. Currently, we are lacking the theoretical understanding of low-latency wireless communication systems that would be crucial to design them optimally. LOLITA addresses this problem by establishing the theoretical framework required to describe the fundamental tradeoffs in low-latency wireless communications.
In particular, by applying and advancing methods from Finite Blocklength Information Theory, LOLITA has developed:
1) Rigorous closed-form approximations of the maximum coding rate of wireless communication channels.
2) Characterizations of the fundamental limits of massive random-access systems.
3) Rigorous and efficiently to compute approximations of state-of-the-art bounds on the maximum coding rate of wireless communication channels.
On the one hand, the obtained results provide accurate performance benchmarks for system designers of next-generation wireless communication systems. On the other hand, they provide invaluable insights on the optimal design of such systems.
In LOLITA, we have focused on the following three research directions:
1) Rigorous closed-form approximations of the maximum coding rate of multiple-input multiple-output (MIMO) wireless channels: We have derived high-SNR normal approximations that become accurate as the packet length and the signal-to-noise ratio (SNR) tend to infinity. The approximations are available in closed form, so they enable mathematical analyses of the maximum coding rate.
2) Characterizations of the fundamental limits of massive random-access systems: In massive random-access systems, the number of users is of the same order as the packet length, and devices access the system at random. We have obtained scaling laws of the capacity of such systems in terms of the order of growth of the number of users. We have further analyzed how interference burstiness stemming from the sporadic access of devices can benefit the transmission of packets. Inter alia, our results demonstrate the need for non-orthogonal access schemes (such as NOMA) of massive random-access systems.
3) Rigorous and efficiently-to-compute approximations of state-of-the-art bounds on the maximum coding rate of wireless communication channels: We have developed saddlepoint approximations of state-of-the-art bounds on the maximum coding rate of single-antenna wireless channels. These bounds need to be evaluated numerically, but they are indistinguishable from the bounds and can be computed at a negligible computation cost. A generalization of these bounds to MIMO channels is underway.
The approximations in 1) and 3) complement each other and provide an extensive characterization of the performance of low-latency wireless communication systems. The characterizations of the fundamental limits of massive random-access systems in 2) address the limitations of interference in wireless networks, which is probably one of the major challenges for the design of next-generation wireless systems.
Our results have been presented at the most important conferences in information theory, most notably the IEEE International Symposium on Information Theory. They further have been published in the most important journals in the field, such as the IEEE Transactions on Information Theory and the IEEE Transactions on Wireless Communications. In addition, in the last month of LOLITA, we organized the Information Theory and Tapas workshop where we presented the results obtained in LOLITA to international experts in information theory.
1) Rigorous closed-form approximations of the maximum coding rate of multiple-input multiple-output (MIMO) wireless channels: We have derived high-SNR normal approximations that become accurate as the packet length and the signal-to-noise ratio (SNR) tend to infinity. The approximations are available in closed form, so they enable mathematical analyses of the maximum coding rate.
2) Characterizations of the fundamental limits of massive random-access systems: In massive random-access systems, the number of users is of the same order as the packet length, and devices access the system at random. We have obtained scaling laws of the capacity of such systems in terms of the order of growth of the number of users. We have further analyzed how interference burstiness stemming from the sporadic access of devices can benefit the transmission of packets. Inter alia, our results demonstrate the need for non-orthogonal access schemes (such as NOMA) of massive random-access systems.
3) Rigorous and efficiently-to-compute approximations of state-of-the-art bounds on the maximum coding rate of wireless communication channels: We have developed saddlepoint approximations of state-of-the-art bounds on the maximum coding rate of single-antenna wireless channels. These bounds need to be evaluated numerically, but they are indistinguishable from the bounds and can be computed at a negligible computation cost. A generalization of these bounds to MIMO channels is underway.
The approximations in 1) and 3) complement each other and provide an extensive characterization of the performance of low-latency wireless communication systems. The characterizations of the fundamental limits of massive random-access systems in 2) address the limitations of interference in wireless networks, which is probably one of the major challenges for the design of next-generation wireless systems.
Our results have been presented at the most important conferences in information theory, most notably the IEEE International Symposium on Information Theory. They further have been published in the most important journals in the field, such as the IEEE Transactions on Information Theory and the IEEE Transactions on Wireless Communications. In addition, in the last month of LOLITA, we organized the Information Theory and Tapas workshop where we presented the results obtained in LOLITA to international experts in information theory.
LOLITA has advanced the state of the art in information theory and wireless communications in various directions. Normal approximations of the maximum coding appeared before in the information theory literature and were used in the wireless communications literature as a proxy for the maximum coding rate of wireless communication systems. However, the majority of these works has focused on simple channel models that provide only limited insights on the behavior of practical communication systems. In this project, we have developed normal approximations that are tailored to wireless communication channels. As such, it provides a more accurate benchmark for low-latency wireless communication systems.
The state-of-the-art bounds on the maximum coding rate that are available in the information theory literature are remarkably tight. However, they must be evaluated numerically which, for practical channel models, can be very costly or in some cases even infeasible. We have developed saddlepoint approximations of these bounds that are indistinguishable from the bounds while having a negligible computational cost. They thus allow for a characterization of the maximum coding rate of low-latency wireless communication systems in regimes that have not been studied before in the literature, but that are of high practical importance. We have further demonstrated that the saddlepoint approximations recover the arguably most important approximations available in the information theory literature: the normal approximation and error exponents. Normal approximations and error exponents apply to different regimes of packet lengths and error probabilities and are therefore usually considered separately. Our saddlepoint approximations thus provide a unifying characterization of the two regimes. For a comparison of these approximations for a single-antenna system with packet length 168 and signal-to-noise ratio 6dB, see the attached figure.
A common assumption in the analysis of massive random-access systems is that the number of active users is proportional to the packet length. We have relaxed this assumption and instead studied the performance of such systems as a function of the order of growth of the number of users. This allows us to better understand how the behavior of the number of users affects the performance of massive random-access systems. It further allows us to assess the robustness of the assumptions made in the literature.
The state-of-the-art bounds on the maximum coding rate that are available in the information theory literature are remarkably tight. However, they must be evaluated numerically which, for practical channel models, can be very costly or in some cases even infeasible. We have developed saddlepoint approximations of these bounds that are indistinguishable from the bounds while having a negligible computational cost. They thus allow for a characterization of the maximum coding rate of low-latency wireless communication systems in regimes that have not been studied before in the literature, but that are of high practical importance. We have further demonstrated that the saddlepoint approximations recover the arguably most important approximations available in the information theory literature: the normal approximation and error exponents. Normal approximations and error exponents apply to different regimes of packet lengths and error probabilities and are therefore usually considered separately. Our saddlepoint approximations thus provide a unifying characterization of the two regimes. For a comparison of these approximations for a single-antenna system with packet length 168 and signal-to-noise ratio 6dB, see the attached figure.
A common assumption in the analysis of massive random-access systems is that the number of active users is proportional to the packet length. We have relaxed this assumption and instead studied the performance of such systems as a function of the order of growth of the number of users. This allows us to better understand how the behavior of the number of users affects the performance of massive random-access systems. It further allows us to assess the robustness of the assumptions made in the literature.