Frame Aggregation Size Determination for IEEE 802.11ac WLAN

Considering Channel Utilization and Transfer Delay

Tomokazu Moriyama, Ryo Yamamoto, Satoshi Ohzahata and Toshihiko Kato

Graduate School of Informatics and Engineering, University of Electro-Communications,

1-5-1, Chofugaoka, Chofu, Tokyo 182-8585, Japan

Keywords: IEEE 802.11ac WLAN, MU-MIMO, Frame Aggregation, Spatial Stream.

Abstract: In order to improve the throughput over WLAN, the IEEE 802.11ac standard, which has been used recently,

introduces the frame aggregation and MU-MIMO (multi-user multiple-input and multiple-output)

mechanisms. The frame aggregation concatenates multiple data frames in one MAC data unit. MU-MIMO

provides SDMA (space division multiple access), which allows multiple STAs (stations) to share space

resources. In an 802.11ac WLAN, MU-MIMO is used in the downlink data transfer in a way that data frames

to multiple STAs are aggregated separately and transmitted at the same time. When traffic loads to individual

STAs are different, however, it is possible that there is a waste in space and time resources. In order to avoid

this waste, several methods to control the frame aggregation size for MU-MIMO are proposed. Those methods

focus mainly on increasing the channel utilization, and so they have a problem that there is a large delay in

transmitting an aggregated data unit. In this paper, we propose a new method to determine the frame

aggregation size considering both channel utilization and delay data frames suffer from in transmission

queues. A performance evaluation result shows that our method provides high channel efficiency with keeping

transmission delay in a relatively small value.

1 INTRODUCTION

One of major interests on WLAN is an improvement

of data transfer throughput. IEEE 802.11ac, the latest

version of WLAN standard, introduced several

mechanisms to increase the throughput of individual

data transfers and that of a WLAN system as a whole.

They include new modulation methods, increased

number of antennas, frame aggregation, and MU-

MIMO. The frame aggregation is originally

introduced in 802.11n (Kim, et al., 2012), and

802.11ac inherits it with expanding the maximum

aggregation size from 65.5 Kbytes to 1 Mbyte (Ong,

et al., 2011). Multiple data frames are aggregated into

a single MAC data unit called A-MPDU (aggregation

MAC protocol data unit).

As for the MIMO technology, 802.11n adopted

only SU-MIMO (single-user MIMO), which is

designed to increase the throughput between one

sender and one receiver (Perez-Neira and Campalans,

2010). On the other hand, 802.11ac has introduced

MU-MIMO, which is a technique to transmit to

multiple receivers at the same time based on SDMA,

in order to increase the overall throughput of a

WLAN system as a whole (Gast, 2013). A separate

data stream in an MU-MIMO communication is

called a spatial stream. It should be noted that

802.11ac supports MU-MIMO only for the downlink

data transfer from an AP (access point) to STAs.

In an actual data transfer, the frame aggregation

and MU-MIMO are used together, and this introduces

a problem that there is a waste (channel idle time) in

some spatial streams when there are variations in

traffic loads from an AP to STAs. More specifically,

the 802.11ac standard defines a procedure that, when

an AP transmits A-MPDUs over multiple spatial

streams using MU-MIMO, it aggregates all data

frames stored in transmission queues for individual

streams. That is, the 802.11ac standard selects the

maximum value among multiple queue lengths as the

frame aggregation size. We call this procedure

mamimum policy in this paper. Although this policy

allows queued data frames to be transmitted

immediately, spatial streams with shorter queue

length will have a wasted time in data transfer.

In order to eliminate this waste in space and time

resources, there are several proposals on how to

determine the frame aggregation size during MU-

Moriyama, T., Yamamoto, R., Ohzahata, S. and Kato, T.

Frame Aggregation Size Determination for IEEE 802.11ac WLAN Considering Channel Utilization and Transfer Delay.

DOI: 10.5220/0006472200890094

In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 6: WINSYS, pages 89-94

ISBN: 978-989-758-261-5

MIMO data transfer. (Nellalta, et al., 2012), (Nomura,

et al., 2014), and (Nomura, et al., 2015) propose

minimum policy, which uses the frame aggregation

size equal to the smallest value among transmission

queue lenghs used by spatial streams which are ready

for MU-MIMO data transfer. (Syed and Trajkovic,

2015) proposes average policy, where the frame

aggregation size is set to the average of transmission

queue lenghs for spatial streams. These policies

improve the channel utilization by decreasing a waste

in space and time resoruces, but the queueing delay

before data frames are transmitted becomes large.

In this paper, we compare three aggregation

policies and clarify the channel utilization and the

delay including both queueing delay and transmission

delay. We also propose a new procedure that

determines a frame aggregation size dynamically

between the minimum queue length and the average

queue length, according to the variations of the queue

lengths among spatial streams. The proposed methods

determines an aggregation size close to the minimum

queue length when the queue length variations are

small, and on the other hand, it determines a size close

to the average queue length when the variations are

large. The rest of paper consists of the following

sections. Section II shows the problem of wasted

space and time resources in MU-MIMO and the

conventional solutions against this problem. Section

III presents the proposed method. Section IV

describes the results of the computer simulation study

and Section V concludes this paper with some

directions for the future work.

2 PROBLEM AND

CONVENTIONAL WORK

2.1 MU-MIMO and Frame

Aggregation

MU-MIMO is a technology adopted by 802.11ac to

improve a WLAN system level throughput. It is based

on the SDMA scheme which transmits directional

radio waves in parallel. In SDMA, an AP can send

data frames to multiple STAs simultaneously. In an

actual environment, STAs sometimes implement one

or a few antennas due to the hardware scale limit,

while APs can be equipped with many antennas. So,

MU-MIMO is an effective way to improve the whole

WLAN system throughput. Currently the 802.11ac

standard regulates that the MU-MIMO downlink data

transfer supports up to eight streams.

The frame aggregation technology is introduced

in 802.11n and is extended in 802.11ac. It is

understood commonly that the frame aggregation

improves the data transfer throughput in MAC layer

(Kim, et al., 2004), (Chosokabe, et al., 2015). There

are two types of frame aggregation; A-MSDU

(aggregation MAC service data unit) and A-MPDU.

In this paper, we focus on A-MPDU, where data

frames (MPDU) including MAC header and FCS

(frame check sequence) are concatenated to form an

A-MPDU. The error detection is performed per

MPDU basis and their reception is reported

independently and inclusively by a single Block Ack

(block acknowledgment) frame.

2.2 Problem of Wasted Space and Time

Resources

As described above, the current 802.11ac standard

tries to aggregate as many MPDUs as possible in an

A-MPDU during MU-MIMO data transfer. This

procedure may bring a problem that there are wasted

time in some spatial streams. Figure 1 shows an

example. In a WLAN in Figure 1(a), AP works as an

Ethernet switching hub and an 802.11ac access point.

Four servers connected to AP via Ethernet are

communicating with four stations, STA1 through

STA4. AP establishes separate spatial streams, s1

through s4. In some moment, the transmission queues

for individual spatial streams contain different

number of MPDUs as shown in this figure. When

those MPDUs come to be transmitted using MU-

MIMO, AP sends all of these frames by aggregating

them into A-MPDUs for individual spatial streams.

The result is given in Figure 1(b). In this case, the

(a) example of MU-MIMO data transfer

(b) status of spatial streams

STA 1

STA 2

STA 3

STA 4

Server 1

Server 2

Server 3

Server 4

Time

Space

Wasted space time

Figure 1: Wasted space time problem during MU-MIMO

data transfer.

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

maximum A-MPDU length (frame aggregation size)

is the length of five MPDUs, which is equal to the

largest queue length among transmission queues just

before A-MPTUs are transmitted (maximum polity).

This is the queue length for spatial stream s1. As for

the other spatial streams, the queue length was not as

large as s1, and so there are some wasted time as

indicated by a shaded part in Figure 1(b). We call this

part a wasted space time. This part will decrease the

channel utilization, and, as a result, degrade the

WLAN system level throughput.

2.3 Conventional Work

In order to avoid this waste, two types of approaches

have been proposed as mentioned above. One is the

minimum polity based approach. The frame

aggregation size will be the smallest queue length

among non-empty queues for spatial streams. In the

example in Figure 1, the frame aggregation size

corresponds to one MPDU length, which is the queue

length for spatial stream s2. Another is the average

policy based approach. The frame aggregation size is

the average queue length of non-empty queues. In

Figure 1, the frame aggregation size will be the length

of three MPDUs, which is the average of five, one,

four and two MPDUs.

It is expected that these policies improve the

channel utilization, because they can reduce wasted

space time. On the other hand, data frames queued in

a transmission queue will suffer from longer delay

until they are actually transmitted. Actually, as for

the MPDU transmission time itself is the same for

three policies, that is, shorter A-MPDUs require only

shorter MPDU transmission time. But, shorter A-

MPDUs will increase the PHY and MAC overheads

introduced in 802.11 WLAN. They include a PLCP

(physical layer convergence protocol) header, RTS

(request to send)/CTS (clear to send) exchanges, and

Block Ack Req/Block Ack exchanges. These

overheads occupy time and space resources and

introduce delay for data frames.

3 PROPOSAL

The minimum policy is the most effective in the

channel usage. However, as the traffic variation

among multiple spatial streams becomes large, the

queueing delay becomes large. On the other hand, the

average policy is expected to decrease the queueing

delay compared with the minimum policy even if the

traffic variation becomes large. However, the channel

usage of the average policy is worse than the

minimum value policy.

We propose a method to control the aggregation

size in response to the traffic variation among spatial

streams. When the traffic variation is small, the frame

aggregation size is set according to the minimum

policy. When the traffic variation is large, the

aggregation size is set according to the average policy.

For this purpose, it is necessary to recognize the traffic

variation by consulting the amount of data in transmis-

sion queues. In our method, the time stamp when a data

frame arrives at the queue is kept with the data itself.

Figure 2 shows a status of an AP establishing

multiple spatial streams with N stations, STA 1

through STA N. A transmission queue is allocated for

each STA, and Figure 2 shows that the queue for STA

i has the longest queue length and that for STA j has

the shortest length. For each data frame, the time

stamp is associated. In the longest queue, they are

max

(1) through T

max

), where I

max

is the number of

data frames in the longest queue. Similarly, the time

stamps in the shortest queue are T

min

(1) through

min

). The total data size in the longest and

shortest queues is D

max

and D

min

, respectively

The proposed method uses the throughput

variation to represent the traffic variation among

spatial streams. Specifically, the throughput for the

longest queue and the shortest queue (S

max

and S

min

respectively) is given by the following equations.













(





)





()

(1)













(





)





()

(2)

The frame aggregation size in the proposed

method, D

prop

, is defined in the following way.

If 



−



≤ __, then





=



+(



−



)×









__

. (3)

STA 1 STA i STA j

STA N

queue for

STA 1

queue for

STA i

queue for

STA j

queue for

STA N

Dmax

Tmax(1)

Tmax

(Imax)

Tmin

(Imin)

Tmin(1)

Dmin

Figure 2: Status of transmission queues in AP.

Frame Aggregation Size Determination for IEEE 802.11ac WLAN Considering Channel Utilization and Transfer Delay

Otherwise,





=



. (4)

Here, D

ave

is the frame aggregation size for the

average polity. It is given by the following equation.









∑









. (5)

When the traffic variation is small, D

prop

is set to

a value close to D

min

in order to increase the channel

utilization. When the traffic variation is large, D

prop

set to a value close to D

ave

in order to decrease the

delay. Note that the frame aggregation size is not

larger than D

ave

4 PERFORMANCE EVALUATION

4.1 Simulation Model

In this section, we show the results of performance

evaluation for three conventional methods and the

proposed method using the Monte Carlo simulation.

Figure 6 shows the simulation model. In the

simulation, each server sends packets to the

corresponding STA through a single AP. AP

aggregates MPDUs and transmits an A-MPDU to an

individual STA using MU-MIMO data transfer. The

traffic load from a server to AP is uniformly random

between 0 and x Mbps. AP has eight antennas and

STA has two antennas. The simulation parameters are

shown in Table 1. With these parameters, the physical

layer data rate per STA is 360 Mbps.

As for the evaluation index for the channel

utilization, we use the wasted space time ratio defined

in the following equation.

 =







(6)

As for the delay, we use the period from the time a

packet arrives at the AP to the time it reaches the

corresponding STA. We call it delay time in the

following subsections.

4.2 Results of Two STA Case

s11, s12

s21, s22

sN1, sN2

Server 1

Server 2

Server N

STA 1

STA 2

STA N

(0, x] Mbps: Uniformly random

Figure 3: Simulation model.

Table 1: Simulation parameters.

Parameter Value

Number of STAs

Modulation

Channel width per STA

Number of spatial stream per STA (N

)

Coding rate (R)

Guard interval (GI)

Packet transmission duration from sever

MPDU size

PHY header transmission duration

RTS transmission duration

CTS transmission duration

SIFS

DISF

Slot time

Cwmin

Bock Ack transmission duration

N (2 or 4)

256-QAM

40 MHz

5/6

0.8 μsec

1 sec

1500 byte

42 μsec

40μsec

28μsec

16μsec

34μsec

9μsec

290μsec

Figures 4 and 5 show the results when there are two

STAs using MU-MIMO with two spatial streams.

The horizontal axes indicate the upper limits of traffic

load (x in Figure 3) of two STAs. The results of the

wasted space time ratio shown in Figure 4 indicate

that, for conventional policies, the smaller the

aggregated size, the better the channel utilization.

The proposed method shows the good characteristics,

which is similar to the minimum policy.

In the result of the delay time shown in Figure 5,

the maximum policy gives the smallest value among

four schemes. The average policy also gives small

delay time, which is comparable with the maximum

policy. On the hand, the minimum policy provides

very large value (hundreds of mili seconds in the

worst case). Although the delay time of the proposed

method is higher than the maximum and average

policies, the value is up to 25 msec and seems to be

tolerable for actual communication.

4.3 Results of Four STA Case

Figure 6 shows the wasted space time ratio when the

number of STAs is four. Each STA uses two spatial

steams with AP. The horizontal axis indicate the

upper limits of traffic load (x in Figure 3) for four

STAs. As described above, the traffic is generated in

a uniform random manner in the area of (0, x] Mbps.

It is clear that maximum policy has the worst

characteristics and the minimum policy provides the

best performance. The average policy is located in the

middle of the maximum and minimum policies. In the

proposed method, when the traffic variation is small,

the improvement is small. However, when the traffic

variation gets large, the performance of the proposed

method becomes closer to that of the minimum policy.

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

Traffic load to

STA1 [Mbps]

Traffic load to

STA2 [Mbps]

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

(a) Maximum policy (b) Minimum policy

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

Figure 4: Results of wasted space time ratio with two STAs.

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

(a) Maximum policy (b) Minimum policy

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

Traffic load limit

to STA1 [Mbps]

Traffic load limit

to STA2 [Mbps]

Figure 5: Results of delay time with two STAs.

Frame Aggregation Size Determination for IEEE 802.11ac WLAN Considering Channel Utilization and Transfer Delay

Figure 7 shows that the delay time from AP to

STAs. The delay time of the maximum policy is the

smallest, and that of the average policy is slightly

larger than the maximum policy. The delay time of

minimum policy is vastly large, and when x exceeds

250 Mbps, the delay time becomes more than 1 sec.

Although the proposed method has larger delay time

than the minimum and average policies, it is less than

one tenth of the delay time of minimum policy over

250 Mbps.

Considering these two performance results, we

confirmed that the proposed method improves the

channel utilization and reduces the delay time.

Figure 6: Wasted space time ratio with four STAs.

Figure 7: Delay time with four STAs.

5 CONCLUSIONS

In this paper, we proposed a method of determining

the frame aggregation size in MU-MIMO data

transfer. Monte Carlo computer simulation showed

that the difference in the aggregation size provides a

trade-off between the channel utilization and the

transfer delay. By appropriately determining the

aggregation size according to the traffic variation for

individual spatial streams, the delay time can be

reduced. The result is that the proposed method

provides 10% of the delay time in the worst case of

the conventional methods, and the channel utilization

of the proposed method is close the best of the

conventional methods. However, these are results in

an early stage. We need to revise our method and

elaborate performance evaluation in more realistic

communication environment.

REFERENCES

Kim, Y., et al., 2012. Adaptive two-level frame aggregation

in IEEE 802.11n wlan. In Proc. IEEE Asia-Pacific

Conference on Communications (APCC 2012).

Ong, E. H., et al., 2011. IEEE 802.11ac: Enhancements for

very high throughput WLANs. In Proc. IEEE Personal

Indoor and Mobile Radio Communications (PIMCR

2011).

Perez-Neira, A. I., Campalans, M. R., 2010. Cross-Layer

Resource Allocation in Wireless Communications:

Techniques and Models form PHY and MAC Layer

Interaction. Elsevier Science.

Gast, M. S., 2013. 802.11ac: A Survival Guide: WI-Fi at

Gigabit and Beyond. O’reilly Media.

Nellalta, B., et al., 2012. On the performance of packet

aggregation in IEEE 802.11ac MU-MIMO WLANs.

IEEE Commun. Lett., vol. 16, no. 10.

Nomura, Y., et al., 2014. High efficient packet aggregation

scheme for multi-rate and voip packet transmissions in

next generation mu-mimo wlans. In Proc. IEEE 2014

International Conference on Advanced Technologies

for Communications (ATC ’14).

Nomura, Y., et al., 2015. Efficient frame aggregation with

frame size adaptation for next generation mu-mimo

wlans. In Proc. 9th International Conference on Next

Generation Mobile Applications, Services and

Technologies.

Syed, A. U., Trajkovic, L., 2015. Improving vht mu-mimo

communications by concatenating long data streams in

consecutive groups. In Proc. IEEE Wireless

Communications and Networking Conference

Workshops (WCNCW 2015).

Lin, Y., Wong, V. W. S., 2006. Frame aggregation and

optimal frame size adaptation for ieee 802.11n wlans.

In Proc. IEEE Global Telecommunications Conference

2006 (GLOBECOM ’06).

Kim, Y., et al., 2004. Throughput enhancement of ieee

802.11 wlan via frame aggregation. In Proc. IEEE

Vehicular Technology Conference 2004 (VTC2004-

Fall).

Chosokabe, Y., et al, 2015. A channel adaptive hybrid

aggregation scheme for next generation wireless lan.

In Proc. IEEE Wireless Communications and

Networking Conference Workshops (WCNCW’ 2015).

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems