• Nenhum resultado encontrado

The Distributed Channel State Information Setting . 41

No documento Paul de Kerret (páginas 68-71)

2.2 The Challenges of Obtaining CSIT

2.2.2 The Distributed Channel State Information Setting . 41

been focused on the problem of TX cooperation with imperfect CSIT. How- ever, it is usually assumed that the channel estimate, although imperfect, is the same at all the TXs involved in the joint processing. This means that either the precoding is done in a central node or that the precoding is distributed across the TXs with the channel estimate beingperfectly shared between the TXs. This CSIT scenario will be called hereafter thecentralized CSIT case. This assumption comes partly from the legacy of previous works where all the transmit antennas were colocated so that this assumption was

justified, and partly because it makes the model simple and intuitive.

However, this assumption is likely to be breached in many scenarios where the TXs are not colocated. Indeed, precoding in a centralized node can be considered only in some scenarios and requires a centralized architec- ture which does not scale well with the number of cooperating TXs. With distributed precoding, the CSI acquisition is inherently acquired at each TX through a different feedback channel. Two scenarios are actually considered for the acquisition of the CSIT in wireless networks and both scenarios are illustrated in Figure 2.1.

The first one consists in direct broadcast of the local estimates from each RX to all the listening TXs. This scenario is interesting as it does not require any CSIT sharing through the backhaul network. It is however not possible in the current 3GPP LTE-A standards [74].

The alternative is an over-the-air feedback from the UE to the home base station alone, followed by an exchange of the local estimates over the backhaul, as it is currently advocated by 3GPP LTE-A standards [74]. Shar- ing the channel estimates without delay and without quantization requires expensive fiber-based backhaul links or dedicated wireless links which will not be available everywhere or will be too costly. In many settings, the CSI sharing will therefore not be possible without further quantization loss and without a certain delay due both to scheduling and to protocol latency.

Either case, the channel estimates available at the various TXs will not be exactly the same. This becomes particularly clear as the number of cooperating TXs increases. Indeed, in both cases, the practical difficulties of acquiring the CSI at the TXs in a timely manner become more challenging as the number of TXs increases. In addition, the amount of CSI to provide to each TX increases very quickly with the size of the network: every TX has to receive the whole multi-user channel matrix which is of size K×K for a setting withK TX/RX pairs and only single-antenna nodes.

This means that a form of CSI discrepancy between the channel esti- mates at the different TXs will arise in many scenarios. In order to capture multiple-antenna precoding scenarios whereby different TXs obtain an im- perfect and imperfectly shared estimate of the overall multi-user channel, we introduce the framework of distributed CSIT. In this CSIT configura- tion, TX j is assumed to have received its own estimate of the multi-user channel, denoted by H(j), before the transmission occurs. It then designs its transmit parameters solely based on this channel estimate and without additional communications with the other TXs.

In spite of its practical relevance, very few works have considered this CSIT configuration, although the analysis of distributed cooperation is gain-

ing in momentum. In [75], a robust algorithm was designed for joint pre- coding across two TXs having distributed CSIT. However, the algorithm provided is computationally demanding and does not provide any insight.

In [76], the capacity of the two-user interference channel (IC) is studied when each TX knows perfectly a subset of the channel coefficients which are fixed.

This work discusses whether it is possible to transmit reliably without know- ing some channel coefficients. It does not study however the transmission with imperfect and imperfectly shared CSIT. Recently, distributed schedul- ing with local state information has also been investigated in [82] and a heuristic algorithm to efficiently exploit the local information available has been developed. Posterior to our publications, some works have pursued analyzing the impact of distributed CSIT over both JP across the TXs [77]

and IA [78, 79].

Several fundamental questions follow from this distributed CSIT config- uration and have not yet been answered. Although these questions prove to be difficult and to a large extent remain open, we shed some light on this problem in this thesis. Specifically, we show that significant gains can be realized from the consideration of the distributed CSIT framework.

In the first part of this thesis, we study the design of precoders based on distributed CSIT. Considering first JP, we evaluate the performance of conventional precoders designed for the case of centralized CSIT when con- fronted to distributed CSIT. We show the extremely deleterious impact of the CSIT discrepancies over the performance and we discuss the design of robust precoders. We also study how distributed CSIT impact the CSI feed- back requirements. Finally, we discuss the impact of distributed CSIT over the IA algorithms.

In the second part of this thesis, we consider the other face of the prob- lem and we assume this time the transmission scheme to be fixed and we study the spatial allocation of feedback resources. We study what are the requirements for an efficient TX cooperation in terms of feedback and back- haul architecture. In particular, we study how complete and accurate should the estimate H(j) be for each j given some performance requirements. By optimizing directly the spatial allocation of the CSIT, we go closer towards providing each TX solely with the information which it really needs for an ef- ficient transmission. We show how this approach leads to strong reductions of the CSIT requirements at virtually no cost, and is therefore promising to make TX cooperation more practical under the constraints of realistic networks.

2.3 Contributions and Publications

2.3.1 Contributions Presented in this Thesis

No documento Paul de Kerret (páginas 68-71)