Illustration

The near-far effect is now illustrated with different traffic types and transport pro-tocols. For improving the downlink, two solutions are proposed, namely the relay based and the window based solution. For the uplink, packet sizes of the terminals can be adapted to their physical mode.

Downlink Traffic

A simple scenario is considered with two terminals in the cell (see figure 2.35).

Terminal 1is assumed to be close to the AP, i.e., it can transmit and receive packets most of the time at 11 Mbps. In our simulations, terminal 1 is at 5 m. After 5 s of simulation, terminal 2 is introduced at d m (35 or 45 m) from the AP with the same type of traffic. 35 m is in the 2 Mbps area and 45 m is in the 1 Mbps area.

The AP is sending packets of 1024 bytes at a constant bit rate over UDP to the two terminals. The input is so high, that it has always something to send. The buffer of the AP is assumed to be infinite and serves the packets according to First In First Out policy.

Simulation results show a very fair behavior of the IEEE 802.11 DCF MAC protocol, since both terminals get the same throughput irrespective of their distance to the AP, i.e., irrespective of the physical mode they use for data transmission (figure 2.36).

But, this fairness leads to a very bad situation for terminal 1 that experiences a

AP terminal 1 5m

terminal 2 d ∈ { 35m; 45m }

Figure 2.35: Scenario with two terminals to illustrate the near-far effect on the downlink.

Figure 2.36: Terminals and aggregate throughput vs. simulation time, UDP down-link traffic, d∈ {35m; 45m}.

significant performance degradation. When terminal 2 is at 35 or 45 m, terminal 1 sees a loss of respectively 57% and 86% of its throughput. The aggregate throughput drops too. This is not the case when terminal 2 is at 5 m or even at 25 m (figure 2.37)

Indeed, as explained in [110] and for equal size packets, a 1 Mbps terminal will occupy the channel approximately 11 times more than a 11 Mbps terminal to transmit a packet. Its data rate is smaller, but its channel occupancy is higher. This phenomenon leads to an equal throughput for both terminals. Besides, since most of the time the channel is used by the low bit rate terminal, the aggregate throughput is also reduced.

0 5 10 15 20 25 30 35 40 45 50

Figure 2.37: Terminals and aggregate throughput vs. simulation time, UDP down-link traffic, d∈ {5m; 25m}.

A field experimentation in an office building illustrates also the near-far effect. As shown in figure 2.38, six measurements have been done for six positions of terminal 2 thanks to a radio sniffer. The UDP throughput and the physical mode distribution is given on figure 2.39 when terminal 2 is alone in the considered cell and 1500 byte packets are sent from the AP. We see a predominance of the 2 Mbps data rate.

Figure 2.40 shows the UDP throughput of two terminals when terminal 1 is close to the AP and terminal 2 moves along the six measurement points. The throughput of terminal 1 alone in the cell is also given. Until 64 m, we have an illustration of the

149 m

Figure 2.38: Field experimentation in an office building.

52 54 56 58 60 62 64 66 68 70 72 0.2

0.4 0.6 0.8

Distance AP / terminal 2 [m]

UDP Throughput [Mbps]

Distance AP / terminal 2[m]

Fraction

11 Mbps 5.5 Mbps 2 Mbps 1 Mbps

Figure 2.39: UDP throughput and physical mode distribution vs. the distance AP/terminal 2, field experimentation.

near-far effect: The aggregate throughput is reduced and for terminal 2 positions between 60 and 64 m, both the terminals have a similar throughput. After 64 m, terminal 2 experiences a lot of disconnections and terminal 1 takes advantage of this to increase its performance.

Note that in the downlink case, it is inefficient to adapt the packet size to the physical mode, e.g., allowing only long packets for terminal 1 and short ones for terminal 2. For the same input load for both users, the number of short packets would indeed exceed the number of long ones in the AP buffer and so on the channel (figure 2.41). Terminal 2 packets would in this case still occupy the channel most of the time.

It is however possible to adopt a scheduling policy at the AP allocating more time for the transmissions of terminal 1.

It can be expected that the impact of near-far effect will smooth out as the number of terminals increases. The degradation of performance depends on the proportion of terminals in each physical mode areas. Let us consider four terminals at 5 m, and a single terminal atd m (45 m and 5 m) as shown in figure 2.42. In the first simulation (figure 2.43), the significant degradation of the aggregate throughput is again observed (from 3460 Kbps to 1530 Kbps, i.e., −56%). This result has to be

52 54 56 58 60 62 64 66 68 70 72 0

0.5 1 1.5 2 2.5 3 3.5 4 4.5

Distance AP / terminal 2 [m]

UDP Throughput [Mbps]

terminal 1 terminal 2 Aggregate terminal 1 alone

Figure 2.40: Terminals and aggregate UDP throughput vs. the distance AP/terminal 2, field experimentation.

compared with that of figure 2.36. In this case, the aggregate throughput doesn’t exceed 1 Mbps. The difference is explained by the proportions of terminals in the two physical modes areas: 50% are in the 11 Mbps zone in figure 2.36, while they are 80% in figure 2.43. Terminals 1 to 4 see their throughput dropping from 864 Kbps to 306 Kbps. For d = 5 m it drops to 693 Kbps but the aggregate throughput is constant.

We now turn to a scenario with TCP and two terminals as in figure 2.35. Sim-ulations are done with an advertized window of 64 segments of 1024 bytes and the near-far effect is still observable (figure 2.44).

It can be noted that at 45 m the throughput is less stable. This is due to a higher PER of terminal 2 in this particular scenario. If a packet is lost, the TCP congestion window for terminal 2 is reduced. Terminal 1 takes advantage to receive more data. Consequently, the aggregate throughput increases as well.

We now propose two solutions for the near-far effect on the downlink. The first one is to have a relay node in the direction of terminal 2. The second one is based on an adaptive advertized window of TCP.

Queue

Figure 2.41: Variable packet sizes on the downlink in case of near-far effect.

AP 5m

terminals 1 to 4 terminal 5 d ∈ { 35m; 45m }

Figure 2.42: Scenario with five terminals to illustrate the near-far effect on the downlink.