Evaluating system capacity - Facilitating Internet of Things on the Edge

For the performance assessment, we used the following indicators: (i) the fraction of

time when at least one node is disconnected from the mesh network, (ii) the mean

number of disconnected nodes, and (iii) the mean throughput per-node of the mesh

network.

In the considered scenario, it is assumed that devices of all users support multi-

connectivity[2]. Consequently, a capable node instantly switches to another con-

nection in cases where the current link is unavailable. In this work, the number of

simultaneously supported links is referred to as the degree of multi-connectivity (M ).

Depending on the instant positions of users on the floor, the number of simultane-

ous links can vary in the range from 0 toM .

The performance evaluation of the considered system was performed in a system-

level simulator. This simulator is based on the Stage source code[29, 74]. The simu-

lation of the mesh network was performed over the 3D model of a building floorplan

(Fig. 5.1). The 3D model allows for a determination of LoS conditions between two

arbitrary points in the environment. In the initial stage of simulation, we executed

a coordinate simulation of the users’ mobility and blockage dynamics. The simula-

tion process utilized a discrete timeline, where at every iteration, the LoS condition

was checked for all the users of the mesh. The data about users coordinates in every

iteration, as well as LoS conditions, were recorded to the database.

In the next step, we worked exclusively with the database for evaluating the target

metrics. It starts with estimating path loss among all the users. If the signal level is

lower than a preset threshold, the simulator considers that there is no direct link

Figure 5.1 Visualization of the simulation

between these users. Contrary, if a direct connection between two users exists, the

simulator estimates the throughput utilizing the Shannon formula. In the third step,

the simulator estimates overheads introduced by medium access control and routing

procedures. These results are in a network-level topology graph, where every link is

associated with throughput.

During the simulation experiments, each set of input parameters (referred to as a

simulation round), run for 1200 seconds of the simulation time, with a time step of

0.25 seconds. It is assumed that all of the involved processes (arrival, service, block-

age) are stationary; the steady state always exists in the system. The starting point

of the steady-state period has been defined from the exponentially weighted mov-

ing average (EWMA) statistics (weight parameter 0.05) and follows the procedure

described in[52]. Table 5.1 summarizes simulation parameters used in our experi-

ments.

Our analysis starts with connectivity and throughput of time-dependent behav-

ior analysis. For this purpose, we used traces of these metrics. An example of such

traces is demonstrated in Fig. 5.2. Particularly, the connectivity trace illustrated in

Fig. 5.2a, allows us to conclude that disconnection intervals are comparably short

but frequent. Such behavior can be explained by short-lived outage events caused by

user mobility and dynamic blockage. An example of throughput variations among

users is shown in Fig. 5.2b. As it can be observed, the throughput significantly de-

viates. For some users, the deviations are rather smooth, primarily caused by users’

mobility. For other users, the throughput trace is characterized by sharp peaks; these

Table 5.1 Simulation parameters.

Parameter

Value

Carrier frequency,

f

28 GHz

Antenna array

16× 16 el. (planar array)

Channel model

3GPP InH

Transmission power

1 W

Receiver sensitivity

-91 dBm

Mean expected area covered by blockers,

p

0.15 Number of users in the mesh network

{8, 10, 12, 14, 16}

Velocity of users

1 m/s

Mobility of users

RDM model

Degree of multi-connectivity

{2, 3, 4, ∞}

(a)Connectivity trace

(b)Throughput trace

Figure 5.2 Example of connectivity and throughput traces

are associated with the blockages (building geometry and dynamic blockers).

Fig. 5.3 the time fraction when at least one node is disconnected from the mesh,

as a function of the number of nodes and multi-connectivity degree (M ). This metric

characterizes an integral measure of connectivity in the network. It demonstrated

fundamentally different behavior for different degrees of multi-connectivity. Partic-

ularly, for multi-connectivityM

= 2,3, the metric increases as the number of users

ity that at least one user finds itself in unfavorable position becomes higher with

greater number of users in the network, and the number of available simultaneous

connections may be insufficient to eliminate this. However, if the degree of multi-

connectivity is high enough, the effects of connection diversity enhances the over-

all connectivity in the mesh. Therefore, we may conclude that multi-connectivity

in mmW meshes significantly improves connectivity for scenarios with dynamic

blockages. However, the number of simultaneously supported connections should

be rather high, which may cause notable control overheads.

Fig. 5.4 shows the mean number of disconnected users depending on the degree

of multi-connectivity and the total number of users in the network. ForM= 2,3 the

network is not able to scale appropriately as the number of disconnected users grows.

Further increasingM , allows the mean number of disconnected users to dip below

1. and the mean number of disconnected users decreases with a greater number of

users in the mesh network.

Fig. 5.5 indicates the mean throughput experienced by a user in the mesh, de-

pending on the number of nodes and a number of simultaneously supported links.

Figure 5.4 Mean number of disconnected users

As it can be noticed, the throughput behavior is qualitatively similar for all the val-

ues ofM . In a case when there are no restrictions for the number of simultaneously

supported links, the mean throughput experienced by the user is 3−5 times higher if

compared to theM= 2. It is worth noting that while increasing the degree of multi-

connectivity, the throughput approaches the higher bound relatively slow, e.g., the

mean throughput atM= 4 is only a half of that for M = ∞. Such a behavior is differ

from the one reported for outdoor scenarios in, e.g.,[26, 28], where both capacity