72

	(norm.)	1
		0.8
	thrp.	0.6
		0.2
	saturation
		0.4
		0

5. Evaluation of IEEE 802.11e with the IEEE 802.11a Physical Layer

		BPSK1/2 (6 Mbit/s)
		16QAM1/2 (24 Mbit/s)
		64QAM3/4 (54 Mbit/s)
thrp. increases with increasing
number of backoff entities
		with address 4, w/o WEP encrypt.
10	20	40	60
	number of backoff entities

saturation thrp. (norm.)

1

0.8

0.6

0.4

0.2

0

BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s) 64QAM3/4 (54 Mbit/s)

thrp. increases with increasing number of backoff entities

with address 4, w/o WEP encrypt.

10	20	40	60
	number of backoff entities

(a) 48 byte frame body size.

	1	with address 4, w/o WEP encrypt.
(norm.)
	0.8
thrp.	0.6
	0.2
saturation
	0.4
		BPSK1/2 (6 Mbit/s)
		16QAM1/2 (24 Mbit/s)
		64QAM3/4 (54 Mbit/s)

0

10	20	40	60
	number of backoff entities

saturation thrp. (norm.)

1

0.8

0.6

0.4

0.2

0

(b) 512 byte frame body size.

with address 4, w/o WEP encrypt.

BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s) 64QAM3/4 (54 Mbit/s)

10	20	40	60
	number of backoff entities

(c) 1514 byte frame body size.

(d) 2304 byte frame body size.

lines: modified Bianchi approx.; markers: WARP2 simulation results

Figure 5.8: Normalized saturation throughput for different PHY modes, and a varying number of backoff entities, for lower priority (AC=”lower”). EDCF parameters as defined in Table 5.2. RTS/CTS are used. The respective legacy saturation throughput is illustrated in Appendix D.4, Figure D.4, p. 238.

5.1.2.2.2Higher Priority AC Saturation Throughput

Figure 5.9 and Figure 5.10 illustrate the resulting saturation throughput for the same scenarios as discussed in the last section, but now for the higher priority AC. As before, results for scenarios without the usage of RTS/CTS, and results with the use of RTS/CTS are given in the two figures.

When comparing the results with the lower and legacy priorities, it can be observed that the higher priority AC shows a larger saturation throughput for a smaller number of backoff entities. However, with an increased number of backoff entities, the saturation throughput for the higher priority AC decreases considerably, because of the higher collision probability.

It worth noting that for this AC with its high probability of collisions, simulation results and analytical approximation do not always show the same saturation throughput.

5.1HCF Contention-based Channel Access

1 BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s)

(norm.)	0.8		64QAM3/4 (54 Mbit/s)
thrp.	0.6
	0.2
saturation
	0.4
			with address 4, w/o WEP encrypt.
	0
	10	20	40	60
		number of backoff entities

saturation thrp. (norm.)

1

0.8

0.6

0.4

0.2

0

73

BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s) 64QAM3/4 (54 Mbit/s)

with address 4, w/o WEP encrypt.

deviation with higher collision probability

10	20	40	60
	number of backoff entities

(a) 48 byte frame body size.

(b) 512 byte frame body size.

saturation thrp. (norm.)

1

0.8

0.6

0.4

0.2

0

BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s) 64QAM3/4 (54 Mbit/s)

deviation with higher collision probability

with address 4, w/o WEP encrypt.

10	20	40	60
	number of backoff entities

(c) 1514 byte frame body size.

saturation thrp. (norm.)

1

0.8

0.6

0.4

0.2

0

BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s) 64QAM3/4 (54 Mbit/s)

with address 4, w/o WEP encrypt.

10	20	40	60
	number of backoff entities

(d) 2304 byte frame body size.

lines: modified Bianchi approx.; markers: WARP2 simulation results

Figure 5.9: Normalized saturation throughput for PHY modes, and a varying number of backoff entities, for higher priority (AC=”higher”). EDCF parameters as defined in Table 5.2. RTS/CTS are not used. The equivalent legacy saturation throughput is illustrated in Appendix D.4, Figure D.3, p. 237.

For a larger number of backoff entities, simulation results and analytical approximation deviate from each other and show not the same results with the accuracy as observed for the lower and legacy priority AC.

This effect results mainly from the definition of the collision duration in Bianchi’s legacy 802.11 model. In the analytical approximation, the collision time Tcoll is defined without considering the ACK timeout; see Section D.3, p. 234.

This is a valid assumption only for the backoff entities that are not transmitting the colliding frames, but not accurate enough for backoff entities that are transmitting colliding frames. The higher the collision probability, the less accurate is the approximation.

However, it can be observed from Figure 5.9 and Figure 5.10 that the analytical approximation and the simulation results show at least qualitatively the same saturation throughput.

74	5. Evaluation of IEEE 802.11e with the IEEE 802.11a Physical Layer

1 BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s)

(norm.)	0.8		64QAM3/4 (54 Mbit/s)
thrp.	0.6
	0.2
saturation
	0.4
			with address 4, w/o WEP encrypt.
	0
	10	20	40	60
		number of backoff entities

(a) 48 byte frame body size.

		1	with address 4, w/o WEP encrypt.
	(norm.)
		0.8
	thrp.	0.6
		0.2	BPSK1/2 (6 Mbit/s)
	saturation
		0.4
			16QAM1/2 (24 Mbit/s)
			64QAM3/4 (54 Mbit/s)

0

10	20	40	60
	number of backoff entities

(c) 1514 byte frame body size.

1 BPSK1/2 (6 Mbit/s)

16QAM1/2 (24 Mbit/s)

(norm.)	0.8			64QAM3/4 (54 Mbit/s)
thrp.	0.6	deviation with higher
saturation	0.2	collision probability
	0.4
		with address 4, w/o WEP encrypt.
	0	10	20	40	60
			number of backoff entities
		(b) 512 byte frame body size.
(norm.)	1
	0.8
thrp.	0.6
	0.2	BPSK1/2 (6 Mbit/s)
saturation
	0.4	with address 4, w/o WEP encrypt.
		16QAM1/2 (24 Mbit/s)		deviation with higher
				collision probability
		64QAM3/4 (54 Mbit/s)
	0
		10	20	40	60
			number of backoff entities

(d) 2304 byte frame body size.

lines: modified Bianchi approx.; markers: WARP2 simulation results

Figure 5.10: Normalized saturation throughput for different PHY modes, and a varying number of backoff entities, for higher priority (AC=”higher”). EDCF parameters as defined in Table 5.2. RTS/CTS are used. The equivalent legacy saturation throughput is illustrated in Appendix D.4, Figure D.4, p. 238.

5.1.3QoS Support with the HCF Contention-based Channel Access

Focus of interest in this EDCF performance evaluation is to develop a method to control the channel access priorities between the different ACs, when backoff entities are operating in parallel and according to different ACs. This is referred to as share of capacity per AC and more relevant for the QoS analysis than the saturation throughput in isolated operation. The saturation throughput in isolated operation, where all contending backoff entities operate with the same EDCF parameter set, i.e., within the same AC, are discussed in the previous section. In contrast, in this section the achievable throughput per AC (share of capacity per AC) and the mutual influences between the ACs are investigated for the shared operation. In shared operation, backoff entities operate with different EDCF parameter sets, according to the different ACs.

5.1 HCF Contention-based Channel Access

75

5.1.3.1Share of Capacity per Access Category

A method to quantify the mutual influences between the ACs is developed in the following, based on the expected idle time duration when operating in saturation.

Recapturing the model described in Appendix D including all modifications described earlier in this chapter, the probability that the channel is busy in a generic slot time is given by

P	=1 −P	=1 − 1 −τ [AC ]	)	N [AC ] ,
CCAbusy	CCAidle	(
equivalent to what is	defined for	legacy 802.11	in Equation (D.6), p. 234.
N [AC ]is the number of backoff entities of a particular AC and τ [AC ]					is the
probability that a backoff entity of this AC transmits at a generic slot time.

The number of consecutive idle slots in this model depends on the expected duration of the idle backoff phase, i.e., the expected contention window length, given in slots, until the first backoff entity attempts its resource allocation by

initiating a transmission. This can be expressed by PCCAbusy and PCCAidle . For example, the probability that one single generic slot is idle between two busy

generic slots is given by PCCAbusy PCCAidle . However, the probability that n consecutive generic slots are idle and thus form a longer idle period between

two busy generic slots is P

P

n

. Therefore, the expected contention

window length,

CCAbusy

CCAidle

E CW [AC ] ,can be calculated to

E CW [

AC ]

=

slot

E = P

1

P

1 + 2

P

2

+3 P

3

+...

CCAbusy

CCAidle

CCAidle

CCAidle

E

= P

1 + 2

P

1

+3

P

2

+...

PCCAidle

CCAbusy

CCAidle

CCAidle

E

− E = P

1 + P

1 +

2 P

2 +...

PCCAidle

CCAbusy

CCAidle

CCAidle

1

= PCCAbusy 1 −P

CCAidle

E = PCCAbusy

PCCAidle

.

(1 −P

)2

CCAidle

As a result, with Equation (D.6), the expected number of consecutive idle slots, i.e., the expected size of the contention window is written as

76

5. Evaluation of

IEEE 802.11e with the IEEE 802.11a Physical Layer

E[CW [ AC ]]

=

(1

−P

)

PCCAidle

=

PCCAidle

slot

CCAidle

(1 −P

)2

1 −PCCAidle

(

[

])

CCAidle

−

τ

N

[AC ]

=

1

AC

.

1 −

1

−τ

[

AC

])

N [AC ]

(

Again, the parameter τ is the probability that a backoff entity is transmitting at a generic slot, and N[ AC ] defines the number of backoff entities.

It is well known that σ -persistent CSMA (Kleinrock and Tobagi, 1975) results in the following expected duration of the idle phase, when N backoff entities operate at the same time:

E[idle]=		(1 −σ )N	aSlotTime .
		−(1 −σ )N
1

This is confirmed by Calì et al. (2000b). Three known types of CSMA are often distinguished.

•In persistent CSMA, MSDUs arriving while the channel is busy have to wait for the channel to become idle again (backlogging), before being delivered immediately. This leads to a relatively high probability of collisions in scenarios with many backoff entities.

•In nonpersistent CSMA, in contrast, MSDUs arriving during a busy channel are also backlogged, but when the channel becomes idle again, MSDUs are delivered with a certain probability at this particular time. Bertsekas and Gallager (1992) refer to this type of CSMA to CSMA slotted ALOHA. If they are not delivered at this time, then a delivery attempt will start at the next slot with the same probability.

•Finally, in σ -persistent CSMA, MSDUs that collided before and are waiting for retransmission while the channel is busy, and MSDUs arriving while the channel is busy are delivered with different probabilities, once the channel becomes idle again.

In Calì et al. (2000a, 2000b), the 802.11 DCF is approximated by σ -persistent CSMA. The difference between the 802.11 (E)DCF and the σ -persistent CSMA lies in the selection process of the backoff interval, i.e., the size of the contention window. Whereas 802.11 EDCF adopted a binary exponential backoff, the size of the contention window in σ -persistent CSMA is calculated from a geometric

5.1 HCF Contention-based Channel Access	77
distribution6 with the parameter σ	(Calì et al., 2000b). However,

Calì et al. (2000a, 2000b) show that in the case the system of contending backoff entities is in saturation, the throughput results of the σ -persistent CSMA approximate the achievable throughput of the 802.11 EDCF, if the average backoff intervals, i.e., the expected size of the contention window, E[CW [ AC ]], of the two different CSMA types are equal. For this reason, the mutual influences between the different ACs in the 802.11 EDCF are in this thesis evaluated based on the assumption that the saturation throughput of the EDCF can be approximated with the saturation throughput of σ -persistent CSMA. In the following, the mutual influences of σ -persistent CSMA backoff entities with different σ -parameters are analyzed instead of the mutual influences of backoff entities that operate with the binary exponential backoff CSMA. The results of this analysis will be compared to 802.11 EDCF simulation results with binary exponential backoff.

For each

individual

AC,

the

expected

size

of the

contention

window, i.e.,

E CW [AC ] , is used to parameterize σ -persistent CSMA per AC

!

1

1 −τ [AC ]

)

N [AC ]

E CW [AC ] =

=

(

= E CW [AC ] . (5.4)

σ [AC ]

1 − 1 −τ

[

AC

])

N [AC ]

(

	σ-persistent CSMA with N	Bianchi approximation with N
	contending backoff entities per AC
		contending backoff entities per AC

The left side of this equation shows the expected value for the contention window from the geometric distribution in σ -persistent CSMA, and the right side shows the expected value for the contention window as calculated from Bianchi’s model. This assumption is used in the following, to calculate the mutual influences between the different ACs. Figure 5.11 and Figure 5.12 show the access probabilities of the three ACs that are parameterized according to Table 5.2, for 3 and 8 backoff entities per AC, respectively. The shape of the geometric distribution is clearly visible. As expected, with a higher number of backoff entities, the expected size of the contention window decreases, as a result, the probability of access at earlier slots increases.

6 The probability	density	function of the geometric		distribution is defined as
p (x ) =1 −(1 −σ ) x	x ≥1 , and		p (x ) = 0 else, with the parameter σ (0..1) . The expected value
is directly obtainable from σ		as	E[ X ] =1/σ . The operator	rounds towards minus infinity,

hence, the distribution is discrete and memoryless. The cumulative probability function is defined as P (x ) = σ (1 −σ )x −1 with x =1, 2, 3... , and P (x ) = 0 else; (Görg, 1997).