Part IV
Autonomous
Overview
1 Problem Description and Assumptions
Part III of the thesis concluded that FD technology may not be the most suit-able strategy for increasing the capacity of indoor small cell networks. This last part of the dissemination explores scenarios where FD can result in sig-nificant gains by studying its potential in providing fast discovery for D2D communication. D2D has been positioned as a strong candidate for future 5G systems, given its potential to offload the infrastructure and to provide support for the URLLC use case.
To initiate a D2D communication, it is previously required that devices discover their peers. This procedure is known as discovery phase, and it can be performed with the involvement of the infrastructure or autonomously by the devices. With the former option, the latency and the control overhead are increased, since devices need to exchange information with the base sta-tion in order to set up the communicasta-tion. However, having the base stasta-tion controlling the network discovery procedure may bring benefits in terms of interference management. When devices work autonomously, the exchange of messages is performed directly among them, thus reducing the overhead and the latency. According to the requirements for next generation access technologies, the control latency should not exceed 10 milliseconds [1]. For this reason, this work focuses on the second strategy, since the most effective solution to provide fast discovery is to avoid the involvement of the infras-tructure.
To be able to set up the communication, a device should be aware whether it has been discovered by its peers. Consequently, a feedback mechanism should be introduced. Ideally, the considered procedure should not increase the control overhead, to avoid impacting the discovery time. This last part of the thesis introduces a piggybacking mechanism for the feedback. The acknowledgment is embedded in the discovery signal and therefore it does
Overview
not require a dedicated transmission.
Conventional HD systems may not be sufficient to meet the 10 millisec-onds latency target. The main problem of the HD transmission mode is that a device cannot listen and transmit at the same time, thus posing a trade-off between the time spent transmitting and the time dedicated to listening. To avoid the constraint that HD poses on the transmission probability, FD tech-nology is considered. This transmission mode allows a device to be continu-ously listening to transmissions from neighbors while being able to transmit its own discovery message.
Commonly, the transmission of the discovery message is based on a pre-defined periodicity or a fixed transmission probability. However, this con-figuration does not allow to control neither the collisions nor the idle slots.
Providing dynamism in terms of transmission probability would be then the most appropriate solution. However, there are two constraints. A high trans-mission probability would cause a large number of collisions, thus increasing the discovery time. On the other hand, if such probability is too low, then a large number of idle slots would be generated and the discovery time would also increase. Note that the transmission probability should be related to the number of neighbors: a large set of peer devices would lead to a low transmission probability and vice versa. Nevertheless, in a real network, the information on the number of neighbors is not available. Hence, an algorithm to estimate the number of neighboring devices is required.
The main assumptions considered in this part of the dissertation are listed below:
• Nodes are synchronized in time and frequency.
The discovery procedure is performed autonomously by the devices, i.e., the cellular infrastructure is not involved in that phase, but still provides time and frequency synchronization.
• Dedicated spectrum for the D2D discovery.
No interference from the cellular network or the D2D devices exchanging data is considered. Furthermore, a pool of frequency resources is available.
The maximum pool size considered in this work is 4. During reception, the devices can listen to all the resources simultaneously.
• Devices are randomly deployed in a certain area at the same time.
Aclusteris defined as the set of neighbors within the coverage range of a device, plus the own device. A simplified channel model which considers only the path loss is assumed. Then, the coverage range is determined taken into account the transmit power, the path loss and the noise power.
When every device is able to reach all the other devices in the network, all the devices’ clusters coincide. This case refers to the single cluster network.
2. Main Findings
The opposite case is a multi-cluster network. An example is shown in Figure 5.18.
(a)Example of a single cluster network. (b)Example of a multi-cluster network.
Fig. 5.18:Types of scenario analyzed in this part of the dissertation.
• 4×4 MIMO technology with interference cancellation receivers.
Ideal interference cancellation and transmission rank one are used in this work. The interfering streams that are not cancelled are treated as noise.
For the sake of comparison, a receiver which treats all interfering streams as noise is also considered.
• Ideal SI cancellation.
All devices are assumed to be FD capable with ideal SI cancellation. This assumption is in line with the previous part of the thesis.
• The time slot duration is 0.25 milliseconds.
In line with the 5G concept presented in [2], since D2D communication is foreseen to be part of it.
• The discovery message and piggybacked feedback is contained in a sin-gle time-frequency resource.
The quantification of the overhead and the design of the discovery message is left for future work. The assumption is that the information needed for the discovery phase, e.g. the device identifier, and the acknowledgments to one or more neighbors can be mapped in a single time-frequency resource block. The discovery message is assumed to be transmitted with a fixed MCS and transmission rank one.
2 Main Findings
The proposed system design is evaluated in the single and the multi-cluster scenarios. A basic receiver that treats interference as noise and one which is able to ideally cancel the three strongest interfering streams, according to
Overview
the transmission rank one assumption, are used. In addition, the number of deployed devices and the size of the frequency resources pool is varied.
The single cluster results presented in Paper D show that the discovery time follows a ’U’ shape curve as a function of the transmission probability. In the left side of the curve, corresponding to low transmission probabilities, a large number of idle slots is generated, thus wasting resources. On the other hand, the right side of the curve represents the high transmission probability region where collisions occur often. The minimum of the curve corresponds to the optimal transmission probability leading to the minimum discovery time, which depends on the network conditions and the scenario.
The target of varying the number of deployed devices and the size of the frequency resources pool is to have different representations of the sys-tem congestion. This parameter is defined as the number of network devices divided by the size of the frequency resources pool. Therefore, fixing the number of frequency resources and increasing the number of network de-vices increases the system congestion. On the other hand, if the number of devices is fixed, increasing the frequency resources leads to a reduction of the system congestion. Based on the system congestion parameter, the optimal transmission direction and the minimum discovery time are extracted, indi-cating that providing dynamism in terms of transmission probability could bring benefits to the system in reducing the discovery time.
An important finding of Paper D is that, in order to fulfill the strict latency target for next generation systems of 10 milliseconds, one of the requirements is to equip devices with MIMO technology and interference cancellation re-ceivers. Under this assumption, FD technology is able to further reduce the discovery time and meet the latency target. Otherwise, the gain that FD can provide is rather limited.
The multi-cluster scenario is more challenging than the single cluster in terms of interference management. In the single cluster scenario, the num-ber of interfering streams is the same for all the receiving devices. On the contrary, in the multi-cluster scenario, devices could perceive different num-ber of interfering streams. Consequently, devices need to carefully select the most appropriate transmission probability to optimize the discovery time but to avoid harming the neighboring devices. It should be highlighted that the conclusion on the advanced receivers extracted from the single cluster sce-nario is still valid in the multi-cluster network.
Given the potential of FD technology to reduce the discovery time, un-der the assumption of using interference cancellation receivers, two solutions are proposed. Firstly, an algorithm to estimate the number of neighbors and select the most appropriate transmission direction. Secondly, a signaling ex-change mechanism to reduce the network interference. The results presented in Paper E demonstrate the potential of the proposed solutions in reducing the network interference at the expenses of a minor increase in the overhead,
3. Included articles
leading to a reduction in the discovery time. In addition, with the proposed scheme, a larger system congestion can be handled while meeting the 10 milliseconds latency target.
3 Included articles
The main body of this last part of the thesis is composed by 2 articles.