You are on page 1of 3

Bandwidth-Waste Avoidance Schemes for Uplink Access in IEEE 802.

16 Broadband Wireless Networks


Zi-Tsan Chou
Networks and Multimedia Institute Institute for Information Industry, Taipei, Taiwan Email: ztchou@nmi.iii.org.tw
frame

Abstract Due to air interference and dynamical queue states in subscriber stations, IEEE 802.16 uplink channel access will unavoidably encounter the idling UL-subframe problem, bandwidth hole problem, and padding waste problem, which are collectively called bandwidth-waste problems. Because of these bandwidthwaste problems, 802.16 uplink goodput will degrade signicantly. To our best knowledge, these problems have not been seriously addressed and solved yet. In this paper, we propose the BWA (bandwidth-waste avoidance) schemes which employ the timeoutbased UL-MAP retransmission scheme to conquer the idling ULsubframe problem and employ the uplink bandwidth reallocation algorithm to conquer the bandwidth-hole and padding-waste problems. Approximate performance analysis does conrm that BWA is much better than 802.16 especially from the viewpoint of bandwidth utilization.

DL subframe broadcast control period Preamble Preamble BS SS1 SS2 SS3 DL-MAP DL-MAP UL-MAP UL-MAP

UL subframe UL-burst period

DL-burst period

DL burst DL burst 1 2

DL burst n

Contention Period

UL burst 1

UL burst 2

UL burst 3

UL burst 1

UL-burst 3 SSTG MDPUs time

Preamble

I. I NTRODUCTION IEEE 802.16 [3], also known as WiMAX, is currently the most promising MAC (medium access control) protocol for high-speed wireless access in metropolitan-scale areas. The PMP (point-to-multipoint) wireless network in 802.16 is composed of a BS (base station) and multiple associated SSs (subscriber stations). In PMP networks, all connections need to go through the BS, where the data directed from SSs to BS is called uplink, while the reverse direction is called downlink. As shown in Fig. 1, in the TDD (time division duplexing) mode, time is divided into xed-sized frames. Besides, each frame is composed of a DL (downlink) subframe and a UL (uplink) subframe. The broadcast control period in the start of the downlink subframe contains synchronization, channel control descriptions, and DL-MAP/UL-MAP for the current downlink/uplink subframe access specication. The UL-MAP message includes at least (1) basic CID (connection ID), which is the unique identifer of the SS scheduled for the particular UL burst and (2) the start time and duration of the corresponding UL burst for that CID. Fig. 1 gives an example of the uplink channel access in 802.16, where the UL-MAP species (SS1 , t1 , x1 ), (SS2 , t2 , x2 ), (SS3 , t3 , x3 ) such that SSi can in turn send its MPDUs (MAC protocol data units) to BS from time ti within its granted transmission duration xi . For a more detailed presentation, please refer to [2], [3]. A. Bandwidth-Waste Problems Undoubtedly, noise is inevitable especially in wireless networks even operating in the licensed band. If DL-MAP is

Fig. 1. An example of IEEE 802.16 uplink channel access. Note that each uplink burst is composed of the preamble, followed by MPDUs and SSTG (SS transceiver turnaround gap).

destroyed due to interference, the entire DL-subframe is still operative since each SS needs to receive all portions of the DL trafc except for the bursts whose modulation and coding (MC) schemes either are not implemented by the SS or are less robust than its current operational MC scheme [2]. However, if the noise source is close to the BS such that UL-MAP is destroyed, then the entire UL-subframe is idle and completely wasted since no SSs know when they can upload MPDUs. Such a phenomenon is called idling UL-subframe problem. The reasons causing this problem are twofold: (1) BS does not know whether UL-MAP is correctly broadcasted to all intended SSs since BS does not receive acknowledgements (ACKs) from them. If all intended SSs simultaneously reply ACKs upon receiving the UL-MAP, then their ACKs collide. If, upon receipt of UL-MAP, each intended SS in turn replies its ACK to the BS, then uplink throughput will signicantly degrades due to large ACK-overheads. (2) Worst, the BS in 802.16 will not get involved in the UL-subframe operations any longer, even if the UL-subframe is completely idle, until the end of the frame [2], [3]. Another bandwidth-waste scenario occurs when the noise source is close to SS so that it cannot correctly receive ULMAP. Take Fig. 1 for example, since SS2 cannot recognize UL-MAP due to interference, the channel time allocated for SS2 (from t2 to t3 ) is completely idle and wasted. We call

frame

such a phenomenon bandwidth hole problem. Refer to Fig. 1, bandwidth-waste scenario can also occur when the granted transmission opportunity xi for SSi is much larger than its current demand di by at least a MAC header. In this case, SSi should transmit padding MPDUs during that unused period from ti + di to ti+1 . In 802.16 [3], the size of a padding MPDU can be up to the maximum of 2041 bytes. Such a phenomenon is called padding waste problem, which may be unavoidable especially during the lifetime of VBR (variable bit rate) ows, (such as rtPS (real-time polling service) ows), because (1) BSs uplink bandwidth allocation algorithm may be imperfectly designed and thus falsely predicts SSs current demand or (2) the SSs demand may suddenly change due to the drop of some real-time data in queue whose deadline is now expired. B. Our Contributions To our best knowledge, the above-mentioned three bandwidth waste problems specically targeted for 802.16 wireless networks have not been addressed and studied before. In this paper, we rst propose the timeout-based UL-MAP retransmission scheme for BS to guarantee that the idling UL-subframe problem will never occur. On the other hand, noise sources may be close to some SSs. In this case, we propose the carrier sensing scheme for SSs and the bandwidth reallocation algorithm for BS such that, when a polled SS cannot recognize the UL-MAP or gives up sending padding MPDUs, then that unused channel time can be shared with residual untransmitted over-demanded SSs. By this way, we can conquer the bandwidth-hole and padding-waste problems. The remainder of this paper is organized as follows. Section II not only describes the BWA schemes in detail, but also provides an approximate goodput comparison between 802.16 and BWA. The nal conclusions are drawn in Section III. II. T HE BWA SCHEMES Since the contention period in UL-subframe is used for SSs in a best-effort manner, the BWA schemes focus the bandwidth utilization only on the UL-burst period. A. Detailed Description We assume that, in the broadcast control period, the BS broadcasts the UL-MAP specifying the basic CID SSi , start transmission time ti , and the granted transmission duration xi for each polled SSi . We call (SS1 , t1 , x1 ), , (SSk , tk , xk ) the polling list.1 After the end of the contention period, the BS in BWA continuously keeps track of the channel activity. If UL-MAP is destroyed due to interference, all polled SSs will not respond. Once the medium idle time exceeds the predened threshold h1 , BS can conclude that either the ULMAP is injured (that is, noise source is close to BS) or SS1 cannot recognize the UL-MAP (that is, noise source is close to SS1 ). Under these circumstances, as shown in Fig. 2(a), the BS will rebroadcast the UL-MAP which contains
1 In 802.16-2004 [2], [3], BS allocates bandwidth only on a per-SS basis. Hence we assume that, on the polling list, SSi = SSj for all 1 i = j k.

broadcast control period

UL subframe contention period UL-burst period

Preamble BS SS1 SS2 SS3 SS4 Preamble BS SS1 SS2 SS3 SS4 Preamble BS SS1 SS2 SS3 SS4

Fig. 2. (a) Since UL-MAP is destroyed due to error, no SSs respond to BS. After the predened medium idle time h1 has elapsed, the BS rebroadcasts the UL-MAP message and thus prevents the idling UL-burst period accident from happening. (b) SS2 correctly receives the UL-MAP during the broadcast control period. However, SS2 detects an interference event from t1 + h1 to t1 + h1 + TULMAP . If SS2 insists on sending its MPDUs from t2 to t3 , then those MPDUs may collide with SS3 s MPDUs sending from t3 to t4 . (c) The unused channel time x2 d2 of SS2 is shared with untransmitted SSs via the uplink bandwidth reallocation algorithm. In this case, SS4 luckily acquires additional bandwidth; namely, x4 > x4 .

the remaining untransmitted SSs on the polling list, that is, (SS2 , t2 , x2 ), , (SSk , tk , xk ) . This way permits the BS to retain control of the uplink, thus stopping the idling ULburst period problem from happening. Note that the value h1 should be greater than the maximum air propagation delay . Besides, the new UL-MAP may not include SS1 since, if we still include SS1 , the QoS requirements of other untransmitted SSs on the polling list may be adversely violated. On the other hand, the unused channel time x1 can be reallocatable to other untransmitted SSs2 on the polling list. k We assume that (1) U = i=1 xi , where U is the length of the UL-burst period, (2) the demanded bandwidth for UGS (unsolicited grant service) ow always remains unchanged, and (3) when uploading data to BS, the SSi with rtPS service ow will piggyback its demanded bandwidth i in
2 For simplicity, we assume that each SS has only one kind of service ow. Thus from now on, SS and service ow are used exchangeably. It is easy to extend our work to the cases that some SSs contain multiple (or different kinds of) service ows. However, such a straightforward, but lengthy, study is out of the scope of this paper.

DL-MAP DL-MAP DL-MAP

UL-MAP UL-MAP UL-MAP

UL-burst 1 ULburst 2 UL-burst 3 UL-burst 4

UL-MAP UL-burst 2 UL-burst 3 UL-burst 4 UL-MAP UL-MAP UL-burst 3 UL-burst 4 UL-MAP

the next frame. Before rebroadcasting the UL-MAP, the BS calculates the sharable channel time S = x1 h1 h2 , where h2 = Tproc + TULMAP + SSTG + , Tproc is the MAC processing delay, and TULMAP is the time for sending the UL-MAP. Then BS rebroadcasts the UL-MAP at time t1 + h1 such that SS2 can upload its UL-burst from time t2 = t1 + h1 + h2 . Specically, the BWAs uplink bandwidth reallocation algorithm works as follows. S1. If S > 0, the BWA employs the weighted-fair sharing scheme [1] to reallocate xi to the untransmitted SSi with rtPS ow according to the following formula. xi = min{i , xi + i }, where i xi i = S (1) xi ) . (2)

the predened threshold, then BS will rebroadcast the ULMAP again and employs the uplink bandwidth reallocation algorithm to deal with that unused channel time ui . By this way, we can successfully conquer the padding waste problem. B. Approximate Analysis For simplicity, we assume that: (1) The noise source is close only to the BS such that the UL-MAP has a probability 0 < p < 1 to be destroyed. (2) There are no noise sources around the SSs. This implies that, all MPDUs sending from SSs to BS are error-free. (3) There are N SSs and each SS has one UGS ow with the same bandwidth need. BS allocates xi = + D for every SSi , where D > h1 + h2 is the time for sending MPUDs and is the sum of SSTG and preamble time. Thus we have U = N (D + ). The uplink goodput G is dened as the fraction of time devoted by SSs to successfully send their MPDUs to BS. Under these assumptions, the uplink goodput of 802.16 is G802.16 = (1p)N D = (1p)D . On the other hand, U +D in BWA, if a polled SSi does not respond to BS due to the damage of UL-MAP, the BS will resend the UL-MAP such that the residual SSs (from SSi+1 to SSN ) can upload their MPDUs if that new UL-MAP is not injured. Hence the uplink goodput of BWA is GBWA = (1 p)D 1 + +D U
N 1

(i >xi )(SSi rtPS) (i

S2. If S 0, let S = S; otherwise S = S SSi rtPS (xi xi ). Then BS evenly distributes S to the untransmitted SSs with nrtPS (non-real-time polling service) or BE (best effort) ows. In other words, xj = xj + S K if SSj {nrtPS, BE}, (3)

where K is the cardinality of the set of the untransmitted nrtPS and BE ows. Similarly, during the UL-burst period, if a polled SS , 2 k, does not respond to the BS, then that unused channel time x will be handled via the above-mentioned bandwidth reallocation algorithm. By this way, we can guarantee that the bandwidth hole problem will never occur. Note that, in BWA, the BS may remove an admitted SS from the polling list if its no-response event has occurred for more than K, say K = 3, consecutive UL-subframes. During the broadcast control period, the BS will broadcast the UL-MAP specifying (SS1 , t1 , x1 ), , (SSk , tk , xk ) . In BWA, if an SS cannot recognize that UL-MAP, then it should keep monitoring the channel activities during the entire ULburst period since it may receive a new UL-MAP again. On the other hand, if SSi recognizes that it was polled by that UL-MAP, it will keep carrier-sensing the channel only from the start of the UL-burst period t1 to the time ti SSTG. Refer to Fig. 1(b), if SSi hears a signal but is unable to detect a meaningful message in the period from t1 to ti SSTG, SSi conjectures that an interference takes place such that the new UL-MAP may be injured. Then SSi should cancel its upload plan unless it receives a new UL-MAP again and never detects an interference event thereafter. In BWA, refer to Fig. 1(c), after transmitting meaningful MPDUs, each polled SSi should stop sending padding bits or padding MPDUs except for padding to reach physical slot (PS) boundary. Since each MPDU carriers the message length information, BS can thus deduce the unused channel time ui of SSi ; namely, ui = xi di , where di is the sum of each MPDU transmission time and is the time for padding to reach PS boundary. Note that the value is obviously less than one PS. In BWA, if ui > h2 + h3 , where h3 is

pi (1 p)(N i)D > G802.16 .


i=1

For example, when p = 0.35, N = 60, and = D/25, we have G802.16 = 62.5%; while GBWA = 95.3%, which is very close to the theoretical limit of the uplink goodput, i.e. 96%. III. C ONCLUSION IEEE 802.16 is currently the most promising MAC standard for wireless metropolitan area networks (WMANs). Due to air interference and dynamical queue states in SSs, the specic bandwidth-waste problems in 802.16 WMANs are denitely inevitable. However, 802.16 does not provide any methods to deal with problems. In this paper, we proposed the BWA schemes, which not only employ the timeout-based UL-MAP retransmission scheme to avoid the idling UL-subframe phenomenon, but also employ the uplink bandwidth reallocation algorithm to conquer the bandwidth-hole and padding-waste problems. Approximate uplink goodput analysis does conrm that BWA is much superior to 802.16 especially in terms of bandwidth utilization. Above all, we believe that WiMAX network access providers may welcome the BWA schemes in that they may bid a very high price for spectrum licenses. R EFERENCES
[1] Zi-Tsan Chou, C.-C. Hsu, and S.-N. Hsu, UPCF: A New Point Coordination Function with QoS and Power Management for Multimedia over Wireless LANs, IEEE/ACM Transactions on Networking, Vol. 14, No. 4, pp. 807820, Auguest 2006. [2] C. Eklund, R. B. Marks, S. Ponnuswamy, K. L. Stanwood, and N.J.M. van Waes. Wireless MAN: Inside the IEEE 802.16 Standard for Wireless Metropolitan Networks. IEEE Press, 2006. [3] IEEE 802.16-2004, IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems. October 2004.

You might also like