3.10 Appendix A
4.2.1 Centralized Protocol for Implementing the System
In this chapter, we adopt a centralized approach to solve the maximum video quality prob- lem in D2D networks for the following advantages [8]. First, it satisfies and executes reliably the service provider policy of maximizing the mean video quality over all devices. Second, it can optimally solve the maximum video quality problem using the high computational capa- bilities of the central station. Third, the centralized implementation requires low processing capabilities at the devices. Finally, it is adaptive to the mobility of devices. On the other hand, a fully distributed approach can be adopted to perform the decision making process at the devices and the base station separately. However, such distributed approach suffers from high processing requirements at the devices, high sensitivity to devices mobility and high security risks [8]. Therefore, we now describe the possible implementation processes of the centralized IDNC system, where a central station forms the SCM Y and the FSM F, and coordinates the global decision making process in each time slot.
Coverage Zone
The devices exchange Hello messages among themselves in order to determine their coverage zones (i.e., neighbouring devices). Each device broadcasts one bit Hello message. Other
O(M −1) neighboring devices generate one bit response message. Consequently, a device discovers its coverage zone using M bits. The coverage zones of allM devices in the network can be discovered using M2 bits. With respect to the stringent deadline for delivering N
video packets, the locations of the devices in a network are assumed to be static. However, the devices’ locations can change from a set of packets delivery to another set, in which case the coverage zones are determined again using M2 bits. Therefore, forN video packets delivery, the communication overhead is M2 bits.
Note that this chapter considers an idealistic Hello message of 1 bit. In practice, a transmitted packet includes several overhead bits such as control bits, an ID of the source, a tag for the control packet and redundancy bits for the integrity of data. Since the size of the Hello message has no impact on the network coding solution, we consider a simple 1 bit message. In fact, the Hello messages, regardless of whether network coding is used or not, are needed to be exchanged to form a D2D communication network and finding the optimal size for Hello messages belongs to the study of D2D network design and analysis.
Packet Reception Probability
In this chapter, the network coding is performed at the network layer. With an efficient channel coding performed at the physical layer, an abstraction of channel model at the net- work layer is often considered, where a transmitted packet is either received or lost with an average erasure probability. This channel erasure probability is a slowly changing param- eter in the network and can be estimated based on the test (or the past) packet reception performance over the channel. Such test messages can be repeated a number of times in order to make a more accurate estimation of the erasure probability. Once the packet recep- tion probabilities connecting a device to other devices are estimated, the device sends this information to the central station. A channel erasure probability can be represented using
⌈log2100⌉bits, where 100 is the maximum erasure probability in percentage. Since each of M devices sends M−1 channels’ information connecting this device to other M−1 devices, the overall communication overhead is M2⌈log
2100⌉ = 7M2 bits. Using this information,
the central station forms the SCM Y. Note that the frequency of updating the channel erasure probabilities depends on the available network resources and the speed at which the fading gain of the channel changes. For example, the fading gain of the channels fluctuate for mobile devices and, consequently, the frequent update on the channel status is required to reduce the channel estimation error.
FSM Update
Each device sends a positive/negative acknowledgement to the central station indicating a received/lost packet. Note that a device needs to use one bit to acknowledge a received packet. Since there are M devices in the network, the overall communication overhead from feedback is M bits per time slot. With the feedback reception, the central station updates the FSM Fin each time slot.
Centralized Decision
In each time slot, the central station selects a set of transmitting devices and their packet combinations using an IDNC algorithm. It then informs the transmitting devices separately about the packet combinations while using the indices of individual packets. In fact, a packet combination can be formed by XORingO(N) individual packets. The central station sends a
bitmap ofN bits to each transmitting device, where the entries with 1’s are the indices of the source packets that are XORed together. In a partially connected D2D network, there can be at most M
2 transmitting devices since a device cannot receive and transmit simultaneously.
The overall communication overhead to inform at most M
2 transmitting devices about their
packet combinations is O(MN) bits, which is negligible compared to the typical size of a packet in wireless networks.