Part III
Full Duplex in 5G Small
Overview
1 Problem Description and Assumptions
A strategy to enhance the spectral efficiency to increase the system capacity is the usage of MIMO technology with a large number of antennas. However, this strategy brings constraints in terms of space and cost, thus becoming in-feasible. In this part of the dissertation, we analyse the role of FD technology in improving the system capacity of a 5G indoor small cell system [1]. FD allows for simultaneous transmission and reception in the same frequency band, which can theoretically double the system throughput over conven-tional HD systems.
The key limitation in building an operational FD transceiver is the self-interference (SI), i.e., the self-interference generated by the transmitted signal at the receiver end of the same node. Recent results in the SI cancellation tech-nology have shown that∼110 dB of attenuation of the SI are already achiev-able [2], according to certain constraints in bandwidth and transmit power.
Such results indicate that building an operational FD transceiver is already feasible.
FD promises 100% throughput gain over HD transmission mode. How-ever, there are several limitations that may prohibit FD to accomplish that promise. Figure 4.12 shows the three constraints that impact the FD gain.
The first two, namely the SI and the ICI, are depicted in Figure 4.12a, with focus on the green device. The red lines represent interference, whereas the black ones correspond to the desired signals. FD technology requires a high level of attenuation of the transmitted signal to avoid saturating the receiver end of the same node. Even though the recent SI cancellation results indicate that building an operational FD transceiver could be already feasible, the residual SI impacts the performance of a FD node. Furthermore, the increase of ICI caused by FD by doubling the number of interfering streams compared to HD, has a negative effect on the FD performance. The third constraint is
Overview
the traffic profile, shown in Figure 4.12b. Exploiting FD is only possible when there is data in both ends. Realistic networks are represented by bursty and DL heavy traffic, and such traffic is usually not symmetric between DL and UL, being the former much heavier than the latter [3].
(a)Self-interference and inter-cell interference. Comparison between HD and FD in a two cells scenario.
(b)Traffic profile.
Fig. 4.12:Constraints which limit the gain that FD can provide over HD.
Two FD applications are considered, bidirectional FD and BS FD. The for-mer refers to the case where both the AP and the UE are FD capable. In the latter case, only the BS or AP is able to exploit simultaneous transmission and reception in the same frequency band.
The research community has been actively studying FD technology given its potential. Most of the works are focused on SI cancellation techniques, since as explained earlier, FD requires a high level of attenuation of the trans-mitted signal to be operational. The evaluation of FD in realistic scenarios is scarce. Many research works assume full buffer traffic model and isolated cells, which means that two of the main constraints that limit the FD gain are not considered. The main target in these works is the evaluation of the residual SI impact. Then, the throughput obtained with FD is approximately doubled compared to the HD throughput. This work takes a different ap-proach, by assuming ideal SI cancellation and evaluating the potential of FD technology in indoor small cell networks considering the occurrence of ICI and realistic traffic profiles. Furthermore, the main KPI is not only
through-1. Problem Description and Assumptions
put but also packet delay. The following assumptions are considered in this part of the dissertation:
• Ultra-dense indoor small cell network.
A 10×2 grid of small cells is considered, with a size of 10×10 m2 and a wall penetration loss of 5 dB [4]. Each indoor cell contains one AP and 4 UEs, all randomly deployed. The users are affiliated to the AP in the same cell (closed subscriber group).
• Finite buffer traffic model.
As previously described, the traffic asymmetry between UL and DL has an impact on the FD performance. The considered traffic model is the one defined by 3GPP [5]. This model generates payloads of 2 megabytes in average. Both the payloads and their inter-arrival time are extracted according to a negative exponential distribution. Two ratios of the offered load are considered, DL:UL = 1:1 and DL:UL = 6:1, in order to capture the effect of the traffic asymmetry on the FD gain.
• Dynamic TDD system for comparison.
The load fairness basedscheme [6] is the baseline choice for this study, be-cause it shows better performance than the delay fairness basedscheme [6]
when the traffic is asymmetric, while performing well with symmetric traf-fic.
• 4×4 MIMO transceiver with IRC receiver.
The IRC is combined with a taxation-based rank adaptation algorithm that penalizes the use of high transmission ranks to control the network inter-ference level [6].
• Recovery mechanisms and link adaptation.
The considered recovery mechanisms are HARQ, RLC AM and TCP. A fast link adaptation that uses the latest 5 channel measurements to extract the most appropriate MCS and transmission rank is used. The MCS is extracted from a SINR-to-MCS mapping table according to a BLER target of 10%. OLLA is not considered.
• Ideal SI cancellation.
Given the low transmit power (10 dBm) and the short distance among nodes (10×10m2cell size), taking this assumptions becomes appropriate.
This consideration allows for evaluating the upper bound of the gain that FD can provide over a fully dynamic TDD system. Therefore, if the eval-uation shows that such gain is not significant, it may indicates that FD may not be the most suitable technology for interference-limited scenarios.
Finally, the modeling of SI cancellation is out of the scope of this work.
Overview
• Random UE paring and round robin scheduler.
For the BS FD case, the pairing of UEs is done randomly. Therefore, the results presented in this work may slightly improve if a smarter UE pairing algorithm is used. However, the difference in performance would not be significant given the small cell size and hence the marginal UE coupling loss.
2 Main Findings
The analysis of the conditions under which the theoretical FD throughput gain is obtained is provided. Such gain is possible to be achieved under specific conditions, namely ideal ICI, full buffer traffic (100% probability of exploiting FD) and scheduling of same UE in UL and DL to avoid intra-cell in-terference, i.e., UE-to-UE interference. In that specific case, a delay reduction of 50% is demonstrated, thus proving also the potential of FD in improving the latency of the system.
The traffic asymmetry dictates the probability that FD can be exploited, and thus the gain that FD can provide. Then, when the traffic allows to simultaneously transmit and receive, the number of interfering streams are doubled. Consequently, the ICI is increased and the FD performance is nega-tively affected. This situation is described in Figure 4.13 for the bidirectional FD case. In the BS FD scenario there is the impact of the intra-cell interfer-ence, which is also present in an isolated cell and impacts the FD gain.
The performance of the bidirectional FD and the BS FD is different. Con-sequently, the main findings for each FD type are presented separately. In the first place, the BS FD scenario suffers from intra-cell interference, which means that the DL user is highly impacted by the transmission of the user scheduled in UL. In the second place, the schemes used to decide the most appropriate transmission direction are different for bidirectional FD and BS FD. The former exploits FD with a single user every time there is data at both ends. The latter decides the optimal transmission direction of each UE according to theload fairness basedscheme [6] and then decides which users can be scheduled to exploit FD.
Part of the analysis is dedicated to the interaction between TCP and FD.
TCP is a well-known protocol to provide a reliable communication by limit-ing the amount of transmitted data based on TCP acknowledgments [7]. The drawback of this protocol is a reduction in throughput and an increase of the latency. FD shows potential to speed-up the protocol since the acknowledg-ments can be transmitted without delay, by simultaneously transmitting and receiving. Results presented in Paper C show that FD is able to accelerate the TCP protocol and mitigate its drawbacks when the ICI is not the main
2. Main Findings
Fig. 4.13:Flow chart describing how the FD gain behaves in the bidirectional FD case.
limiting factor. However, under strong ICI, the benefits of FD in speeding-up the TCP protocol are hidden by the increased interference, since it limits the amount of transmitted data with the selection of a lower MCS and transmis-sion rank.