Base Station Cooperations Under Imperfect Conditions
- 126 Downloads
BS cooperation is dedicated to deal with the inter-cell co-channel interference in 4G/LTE networks. Traditional ways of co-channel interference mitigation either require extra bandwidth for large reuse factor or rely on highly complex signal processing which is not practical for mobile devices. With the development of MIMO technologies and the enhanced BS processing capability, BS cooperation has been proposed to deal with the inter-cell co-channel interference. The basic idea is to treat a group of BSs as a “super-BS” with multiple spatially distributed antennas, also known as network MIMO (Karakayali et al., 2006). By sharing UEs’ CSIs and data, multiple BSs coherently coordinate the transmission and reception; thus other-cell signals are used to assist transmission instead of acting as interference. As a consequence, the network capacity can be greatly enhanced. The idea can be originated to distributed antenna systems (Saleh et al., 1987; Zhang and Andrews, 2008). Based on this technique, people have proposed a novel architecture called distributed wireless communication system (DWCS) (Zhou et al., 2003), where the antennas are distributively deployed and the baseband processors are jointly managed.
Limited capacity of central controller. With the increase on the number of cooperating BSs, a central controller needs to handle the joint precoding for the increased number of UEs. To support high-speed data transmissions, high data processing capacity of the central controller is urgently required.
Synchronization. To realize BS cooperation, symbol-level synchronization among BSs should be guaranteed. However, as the BSs are geographically separated, it is very difficult to achieve high-accuracy synchronization. This difficulty may become even more severe when the number of cooperating BSs increases since the variance of the distances to UEs becomes larger.
Limited CSI acquisition capability. The BS cooperation highly depends on UEs’ CSIs for precoding. Therefore, each BS needs to obtain the CSI of all the UEs, which requires a large amount of resources for channel training and feedback. However, as the feedback channel capacity is limited, the cooperation performance would degrade due to inaccurate channel feedback.
Limited backhaul capacity. The UEs’ data needs to be shared among cooperating BSs, which exponentially increases the burden of backhaul link. BS cooperation cannot be fully realized unless there is sufficient backhaul capacity to support UEs’ data sharing.
To achieve the expected capacity gain of BS cooperation in real systems, the above imperfect conditions should be taken into consideration. In the next section, candidate solutions are listed with a summary of the state-of-the-art research progress.
Dealing with Imperfect Conditions
The imperfect conditions involved in the real systems induce novel design in cooperative transmission. In particular, the research efforts are classified into the following three parts.
Clustering with Limited Cooperation Size
Due to the limited capacity of the central controller as well as the difficulty in synchronization, it is impossible to form a “super-BS” over the whole network. On the other hand, it is also unnecessary as the cooperation gain with large inter-BS distance is marginal due to the pathloss effect. Consequently, grouping BSs into clusters is a practical and efficient solution. The clustering algorithms can be classified into two categories. The first category is static clustering (Huang et al., 2009), in which the BS cooperation cluster is predetermined during network planning stage so that it is easy to implement during network operation stage. However, the problem of static clustering is that the cluster-edge UE is severely degraded due to inter-cluster interference. A modified protocol was proposed in Zhang et al. (2009) by suppressing the interference via inter-cluster cooperation. However, the BS cooperation gain will be sacrificed. Thus, people turn to the second category of clustering, i.e., dynamic clustering according to UEs’ channel conditions. A centralized dynamic clustering algorithm was proposed by Papadogiannis et al. (2008b). It is shown that the dynamic clustering with cluster size 2 can perform the same as the static clustering with size 7. Nevertheless, the centralized algorithm requires a large amount of CSIs as well as high computational capacity of the central controller. An improved algorithm was proposed by Liu and Wang (2009) to reduce the cost on channel acquisition. But still, the limited capacity of the central controller is a performance bottleneck.
To further reduce the computational requirement on the central controller, we proposed a distributed dynamic clustering algorithm (Zhou et al., 2009) where the BSs negotiate with each other distributively to form the clusters. The BSs firstly compute the CSI-related weights to represent the potential gain of joining a certain cluster and then exchange the weights to reach an agreement. It is shown that with a small amount of information exchange among the BSs, the proposed distributed algorithm achieves the same sum-rate performance as the centralized algorithm, but the computational complexity and signaling overhead can be greatly reduced. The dynamic clustering algorithm was further extended to overlapping cluster case (Gong et al., 2011), where a single BS can join multiple clusters simultaneously, and hence, the user fairness can be improved compared with non-overlapping clustering methods.
With clustering technique, the practical constraints including limited capacity of the central controller and synchronization issue can be well solved. Firstly, distributed dynamic clustering reduces the capacity requirement on the central controller. The channel training can be launched locally in each BS, the information exchange can be done by the backhaul link, and the clustering decision is made distributively. Secondly, limited cooperation size reduces the difficulty of synchronization. Simulation results show that the dynamic clustering with sizes 2 to 3 can achieve very high sum data rate. The performance improvement by further increasing the cluster size is marginal. Hence in practice, clustering with sizes 2 to 3 is widely used. In this case, the synchronization is not a difficult task.
Dynamic CSI Acquisition with Limited Feedback
As each BS requires CSIs from all UEs in the same cluster, CSI acquisition should be redesigned for BS cooperation. In TDD mode, by exploiting the uplink-downlink reciprocity, the CSIs of all UEs can be trained via uplink pilot signal. However, the total training slot length is limited. When splitting the training sequence to the UEs, each of them can only be allocated with limited resources and hence has imperfect CSI. In FDD mode, the CSIs can be trained by each UE via broadcasting a training sequence from a BS and then can be fed back to the BS. In this case, each UE needs to feed back the CSIs of multiple cooperating BSs. Due to limited feedback channel capacity, the feedback channel resource allocation should be carefully designed. In the rest of this subsection, we focus on limited feedback in FDD mode.
Limited feedback for a single link or a single cell has been well studied by Love et al. (2008) and the references therein. Different from single-cell case, however, in BS cooperation scenario, there are different pathloss effects between a specific UE and cooperating BSs. Intuitively, the BSs closer to the UE are more valuable for cooperation. As a result, one can ignore CSI from the far BSs and allocate more feedback bits to the near BSs to enhance the overall performance. Papadogiannis et al. (2008a) proposed a threshold-based CSI feedback algorithm to only feed back the CSIs larger than a certain threshold. However, there lack theoretical results on how to determine the threshold. A more dedicated and realistic model is to use quantization and feed back the CSIs by a given number of bits. Quantized CSI feedback was studied by Zhang and Andrews (2010) and Bhagavatula and Heath (2011) for MIMO interference channel, i.e., the CSIs of other UEs are only used for interference suppression. For cooperation, the performance is more sensitive to the CSIs as the precoding matrix strongly depends on the quantized feedback. Based on the above intuition, more bits should be allocated to strong links, while fewer bits should be used on weak links, so that the influence of feedback error is minimized. This idea was realized by Zhou et al. (2011). In this work, a feedback set of a UE is defined as a subset of BSs to cooperatively serve the UE. Each UE optimizes its feedback set distributively according to the expected signal-to-noise-plus-interference ratio (SNIR) which includes the interference caused by the feedback error. The proposed algorithm outperforms the conventional nonadaptive CSI feedback.
The adaptive quantized CSI feedback deals with the limits on feedback capacity. The basic idea is to utilize the feedback bits on the “most valuable” channels. It is possible that no bits are allocated to some BS as the signal strength is too weak. In this case, the practical cluster size shrinks. Hence, considering the feedback capacity constraint, there is a trade-off between the cooperative gain with large cluster size and the interference caused by the feedback error. Again, the cluster size should not be too large in practice.
Feedback Scheduling with Constrained Backhaul
The backhaul link among BSs is responsible for sharing UEs’ data as well as CSI. When the backhaul link capacity is limited, not all the information can be shared. How to allocate the limited backhaul capacity should be carefully studied. The downlink achievable rate region under backhaul capacity constraint was characterized by Simeone et al. (2009). To fully exploit the advantage of BS cooperation, the idea of superposition coding can be introduced to enhance the ability of sharing data. One way is to share the integrated other-cell signal status, i.e., interference, via interference forwarding (Kim, 2008) in the uplink or rate splitting (Maric et al., 2007) in the downlink. The other is to share part of UEs’ data among neighboring BSs for cooperative precoding or interference cancellation (Marsch and Fettweis, 2008).
When jointly considering the CSI feedback constraint, the problem becomes more complicated. Both feedback and backhaul constraints result in imperfect information for cooperation. If the feedback is perfect, the CSI will become imperfect due to limited backhaul. On the other hand, if the backhaul capacity is sufficient, limited feedback still causes interference by imperfect CSI. Therefore, it is necessary to jointly consider both effects. Marsch and Fettweis (2009) jointly considered limited feedback and backhaul constraints and evaluated some detection and coding mechanisms. Based on the analysis, how to adjust CSI feedback scheme under backhaul constraint is further investigated. Intuitively, according to different backhaul capacities, the cooperation mode can switch between joint transmission and interference suppression or partly joint transmission plus partly interference suppression. When the backhaul capacity is sufficient, BSs can fully cooperate to obtain cooperation gain. On the other hand, when the backhaul capacity is insufficient, only the interference information can be exchanged, and interference suppression is adopted.
In summary, the application of BS cooperation in real system encounters imperfect conditions, including limited computational capacity, synchronization issue, limited feedback capacity, and limited backhaul capacity. Candidate solutions have been given with three categories: clustering, feedback bits allocation, and mixed cooperation and interference management policy. These schemes enhance the feasibility of BS cooperation in practice. In the next section, key applications will be given.
The BS cooperation has been standardized in 3GPP (TR36.819, 2012), termed as Coordinated Multipoint (CoMP). The imperfect conditions are considered in the standard. For instance, the cooperation clustering with cluster size 2 is suggested, the feedback policy includes explicit feedback and implicit feedback, and the CoMP categories include joint processing (fully cooperate), coordinated scheduling/beamforming (partially cooperate), and so on. The techniques have been widely applied in 4G/LTE networks.
Recent progress and applications include BS cooperation under energy-efficiency constraint. For instance, in smart grid scenario, BSs need not only cooperate in signal domain but also in energy domain, so that the provided energy is sufficient for cooperation (Xu and Zhang, 2015). In energy-harvesting scenario, where each BS is powered by renewable energy that is intermittent and imbalance, novel cooperation protocol is required to tackle the energy imbalance between cooperating BSs (Gong et al., 2016). The energy constraint can be viewed as another imperfect condition. Hence, the design under energy-efficiency constraint can be considered as an extension on the research of BS cooperation with imperfect conditions.
- Gong J, Zhou S, Niu Z, Geng L, Zheng M (2011) Joint scheduling and dynamic clustering in downlink cellular networks. In: IEEE global telecommunications conference (Globecom), pp 1–5Google Scholar
- Liu J, Wang D (2009) An improved dynamic clustering algorithm for multi-user distributed antenna system. In: International conference wireless communnication signal processing, pp 1–5Google Scholar
- Marsch P, Fettweis G (2008) On base station cooperation schemes for downlink network MIMO under a constrained backhaul. In: IEEE global telecommunications conference (Globecom), pp 1–6Google Scholar
- Marsch P, Fettweis G (2009) On downlink network MIMO under a constrained backhaul and imperfect channel knowledge. In: IEEE global telecommunications conference (Globecom), pp 1–6Google Scholar
- Papadogiannis A, Bang HJ, Gesbert D, Hardouin E (2008a) Downlink overhead reduction for multi-cell cooperative processing enabled wireless networks. In: IEEE 19th international symposium personal, indoor and mobile radio communication (PIMRC), pp 1–5Google Scholar
- Papadogiannis A, Gesbert D, Hardouin E (2008b) A dynamic clustering approach in wireless networks with multi-cell cooperative processing. In: IEEE international conference communication (ICC), pp 4033–4037Google Scholar
- Simeone O, Somekh O, Poor HV, Shamai (Shitz) S (2009) Downlink multicell processing with limited-backhaul capacity. EURASIP J Adv Signal Process 2009(1):840–814Google Scholar
- TR36819 (2012) Coordinated multi-point operation for lte physical layer aspects (release 11). Technical report, 3GPPGoogle Scholar
- Zhou S, Gong J, Niu Z, Jia Y, Yang P (2009) A decentralized framework for dynamic downlink base station cooperation. In: IEEE global telecommunications conference (Globecom), pp 1–6Google Scholar