1 Handover Management based on the Number of Retries for VoIP on WLANs Shigeru Kashihara Yuji Oie Department of Computer Science and Electronics, Kyushu Institute of Technology Kawazu 68-4, Iizuka, Japan Telephone: Fax: Abstract Ubiquitous networks enabling access by mobile nodes (MNs) at anytime and at any location will consist of several different WLANs, so that it is very likely that the MN will traverse several hotspots during communication. Thus, a single communication can experience several handovers between WLANs, during which the quality of the communication may be degraded due to packet loss. In the present study, we attempt to develop an effective handover management scheme for VoIP communication in which the quality is sensitive to packet loss. For this purpose, we propose a handover management scheme that employs a transport protocol that supports multiple connections for VoIP communication and that makes handover decisions based upon a new criterion, i.e., the number of retries experienced by a data frame. The proposed handover management scheme minimizes packet loss, as well as redundant traffic due to parallel transfer of identical packets, by predicting the occurrence of packet loss in advance based on the number of retries and properly selecting a single-path or multi-path by examining the communication quality on the wireless link. I. INTRODUCTION Wireless local area networks (WLANs) based on the IEEE standard  have gained popularity due to their low cost, simplicity of installation and high data rates. WLANs are being set up in public spaces, called hotspots, such as airports and coffee shops. In the near future, the these hotspots, which will be administrated by different organizations, will become more numerous, until eventually overlapping to provide continuous coverage for a wide area, and will serve as ubiquitous networks through which mobile nodes (MNs) can continuously access the Internet from any location. In ubiquitous networks, MNs can traverse hotspots while maintaining Voice over IP (VoIP) communication. Thus, a single VoIP communication can experience several handovers, which can lead to packet loss, resulting in the deterioration of VoIP communication quality. Therefore, preventing packet losses during handover is crucial for high-quality VoIP communication. In order to support mobility of MNs, many studies have investigated Mobile IP (MIP) . Since MIP supports transparency above the IP layer, the MN can move across different IP networks without interrupting communication. However, during handover between WLANs with different IP addresses, the MN is subject to a period during which packets cannot be sent or received due to both link switching delay and IP protocol operations (layer 2 and 3 handover process). Packet losses due to such disruptions degrade the quality of VoIP communication. In order to avoid packet losses due to handover, the MN should not only reduce the layer 2 and 3 handover processing period, but also move from a WLAN having a bad wireless link condition to another WLAN before packet loss occurs. Although signal strength is one type of information that indicates the condition of a wireless link, properly estimating the occurrence of packet loss from signal strength in advance is difficult at the MN, because signal strength fluctuates abruptly with the distance from the access point (AP) and any interfering objects located between the MN and the AP. The focus of the present study is handover management that, unlike MIP, is based on end-to-end processing only, without any additional network entities, that will work very well even during handover between WLANs that are administrated by different organizations. In this context, we propose handover management that employs a transport protocol that supports multiple connections for VoIP communication and makes handover decisions based on a new criterion, i.e., the number of retries of a data frame. The proposed handover management scheme minimizes packet loss, as well as redundant traffic due to parallel transfer of identical packets, by predicting the occurrence of packet loss in advance based on the number of retries and properly selecting a single-path or multi-path by examining the communication quality on wireless links. Through simulation experiments, the proposed scheme is shown to maintain the required communication quality for VoIP during handover. II. TWO TYPES OF PACKET LOSS DURING HANDOVER Two types of packet loss can occur during handover: (1) packet loss by intermittence due to layer 2 and 3 handover process, and (2) packet loss due to the inherent characteristics of WLANs. These types of packet loss are described below. Handover technologies for traversing different IP networks without interrupting communications are being investigated in a number of studies. In particular, several handover technologies using MIP have been proposed. However, MNs using MIP experience a period during which they are unable to send or receive packets due to both link switching delay and IP protocol operations. The handover process using MIP is as follows: (1) Channel scan to search for the next AP, (2)
2 Association with a new AP, (3) Acquisition of IP address in a new WLAN, and (4) Binding update to the Home Agent (HA) and the Corresponding de (CN). These handover processes can be divided into layer 2 handover (1,2) and layer 3 handover (3,4). In an empirical study , A. Mishra et al. reported that the layer 2 handover period is 5-4 ms. In addition, considering acquisition of IP address from DHCP (3 ms ) and binding update (one-way delay) over layer 3 handover, it is clear that the disruption period caused by layer 2 and 3 handover processes severely impacts the communication quality of VoIP. Although FMIPv6  and HMIPv6  have been proposed to reduce this handover processing period, it is difficult to deploy these methods in WLANs that are administrated by different organizations. In WLANs, wireless resources are shared among several MNs. Thus, packet loss can inherently occur depending upon the condition of the wireless resource. Namely, packet loss can be caused by frame errors due to the degradation of signal strength or by contention among MNs sharing the resource. III. HANDOVER MANAGEMENT BASED ON NUMBER OF RETRIES Packet losses severely impact communication quality, because the sender never retransmits the same data even if packet losses are occurring under VoIP communication. In order to maintain the communication quality of VoIP during handover, it is necessary to prevent the two types of packet loss described in Section II. In this section, we propose a handover management scheme by which to avoid packet losses. The proposed scheme is an attempt to support end-to-end mobility using multihoming without requiring any hierarchical structure or other additional network entities, as are required by MIP. When an MN acquires a new IP address, the MN must notify the CN of the IP address. Although mechanisms by which to inform the CN of the addition or deletion of IP addresses, such as mobile Stream Control Protocol (msctp) , were considered, they are not discussed herein. A. Reduction of Handover Processing Period When an MN moves between WLANs administrated by different organizations, a disruption period due to the layer 2 and 3 handover process occurs. In order to reduce packet losses due to this disruption period, we proposed a multihoming MN that is equipped with two or more WLAN interfaces. In order to simplify the explanation of the proposed method, the MN is assumed to have two WLAN interfaces (IF1 and IF2). First, the MN communicates with the CN through IF1. While communicating through IF1, MN searches for another AP using IF2. When a new AP is discovered, the MN connects to the new AP by IF2. At this time, the MN is connected with two different WLANs. The layer 2 and 3 handover processing periods are reduced by being connected to two or more WLANs before executing handover. B. A vel Criterion by which to Estimate the Occurrence of Packet Losses In order to move between WLANs while maintaining the VoIP communication quality, the MN should select the better WLAN according to the condition of the wireless link. Although signal strength is one type of information that indicates the condition of a wireless link, it is difficult for the MN to properly perceive the condition of a wireless link based on the variation of signal strength, because signal strength fluctuates abruptly with the distance from the AP and any intervening objects located between the MN and the AP. Therefore, we considered the number of retries of a data frame as a novel criterion by which to detect whether the condition of a wireless link is becoming better or worse. When data frames or ACK frames are lost, the sender retransmits the same data frames until the number of retries reaches a predetermined retry limit. In the IEEE82.11 specification , two types of retry limits are provided: the longframe retry limit and the short-frame retry limit. These two retry limits depend on the size of the data frame. When a data frame is longer than the RTS Threshold (default: 2347 bytes), the long-frame retry limit for the data frame is four. On the other hand, when a data frame is shorter than the RTS Threshold, the short-frame retry limit for the data frame is seven. In the present paper, the short-frame retry limit is employed because the VoIP data size is small. When an ACK frame for the sent data frame cannot be received, even after the sender transmits the same data frame seven times (original data frame and six retransmitted data frames), then the data frame is treated as a lost packet on the sender. Thus, the number of retries, allows the MN to detect whether the condition of the wireless link being used is becoming better or worse and may enable the MN to determine when the handover process should be started before packet loss actually occurs. We investigate through simulation experiments using the Network Simulator Version 2 (Ver. 2.27)  how the distance between the MN and the AP affects both the number of retries of a data frame between the MN and the AP and packet losses. Figure 1 depicts the simulation model. The WLAN based on IEEE 82.11b  is constructed by the infrastructure mode. The MN is assumed to communicate with the CN using VoIP on a WLAN of only 11 Mb/s. The sender sends packets of 2 bytes to the receiver at 2 ms intervals using the G.711 codec, i.e., the consumed bandwidth for one direction is 8 kb/s. Figure 2 illustrates the ratio of packets experiencing data frame retransmission and lost packets among all packets as a function of the distance between the MN and the AP. Retry indicates packets without any data frame retransmission, and Retry 1 - Retry 6 indicate packets that experienced data frame retransmissions from once to six times, respectively. In addition, Packet Loss indicates packets that are unrecovered within the short-frame retry limit. In Figure 2, the packet loss ratio increases sharply from approximately 17 m. On the other hand, the ratio of packets that experienced data
3 CN 1 Mb/s 3 ms AP (82.11b) VoIP (G.711) packet size:2 bytes interval: 2 ms 1 Mb/s 5 ms Transport Layer Handover Manager (HM) IF1 s state IF2 s state information information Ret_IF1 Ret_IF2 SC_IF1 SPT SC_IF2 IP Layer LLC LLC MAC MAC MN PHY PHY WLAN IF1 WLAN IF2 Fig. 1. Simulation model Fig. 3. Cross-layer architecture Ratio [%] Retry Retry 1 Retry 2 Retry 3 Retry 4 Retry 5 Retry 6 Packet Loss Distance [m] Fig. 2. Ratio of packets experiencing data frame retransmission and lost packets to the distance between the MN and the AP frame retransmission increases gradually before packet losses occur, and, at 17 m, approximately 4% of packets experience retransmission. Therefore, the occurrence of packet loss due to the degradation of the condition of a wireless link, that is, the degradation of signal strength, can be detected beforehand based on the number of retries. Data frame retransmissions occur due to not only the degradation of signal strength, but also the collision of data frames sent by two or more MNs. In Reference , VoIP communications on which 2 byte voice data packets are sent at 2 ms intervals are shown to be able to accommodate approximately 1 calls on an 11 Mb/s WLAN. Similar results were obtained through our simulation experiments. In addition, from our simulation results, we obtain the result that approximately 98% of packets do not experience data frame retransmission within 1 calls for VoIP communication. Therefore, the number of retries can be employed as a novel criterion based on which to predict the occurrence of packet loss due to the degradation of signal strength when there is adequate bandwidth to accommodate VoIP communication. C. Cross-Layer In the proposed scheme, a Handover Manager (HM) on the transport layer controls the handover process according to the condition of a wireless link. The cross-layer approach  is employed so that the MAC layer informs the HM of the deterioration of a wireless link. Figure 3 shows the architecture of the proposed scheme. Suppose that an MN has two WLAN interfaces. The MAC layer on each interface informs the HM of the number of retries whenever the ACK frame is received or the number of retries reaches a predetermined retry limit. The HM records the number of retries from MAC layer in Retry Counters (Ret IF1, Ret IF2) on the HM. D. Handover Manager When the condition of a wireless link is degraded, i.e., the number of retries increases, the sender sends two identical data packets to the receiver over two WLANs (multi-path transmission) in order to prevent packet loss. In the case of switching to another WLAN without multi-path transmission, the VoIP communication quality might be degraded due to not knowing the condition of the next wireless link in advance. Therefore, the HM should investigate the condition of both wireless links using multi-path transmission when the number of retries is growing. Then, when the HM discovers a WLAN that has a wireless link of good condition, the transmission returns to single-path transmission through this WLAN. Next, we explain the operation of the HM during handover. An MN with two WLAN interfaces (IF1, IF2) is assumed to be communicating with the CN through IF1, and IF2 is assumed not to be connected to any AP. When the MN discovers a WLAN other than that connected by IF1, the MN connects with the AP by IF2. At this time, both interfaces can be communicating with the CN at anytime. Figure 4 illustrates the process of switching to multi-path transmission. The MAC layer on IF1 informs the HM of the number of retries whenever an ACK frame is received or the number of retries reaches the retry limit. After recording the number of retries in Ret IF1, the HM compares Ret IF1 with the Multi-Path Threshold (), which is the threshold for switching to multi-path
4 Single-Path Send a packet MAC Layer informs HM of the number of retries HM records the number of retries in Ret_IF1 Ret_IF1 Multi-Path WLAN IF1 Multi-Path Send a packet (IF1) SC_IF1 Send a packet (IF2) is reset MAC Layer informs HM MAC Layer informs HM to of the number of retries of the number of retries HM records the number of retries in Ret_IF1 SC_IF2 is reset to HM records the number of retries in Ret_IF2 Ret_IF1== Ret_IF2== SC_IF1++ SC_IF1 SPT SC_IF2++ SC_IF2 SPT Single-Path Single-Path WLAN IF1 WLAN IF2 Fig. 4. Single-path transmission to multi-path transmission Fig. 5. Multi-path transmission to single-path transmission transmission. In the case that Ret IF1 exceeds, the HM detects the deterioration of the condition of the wireless link and switches to multi-path transmission in order to prevent packet loss. In multi-path transmission, since the MN sends two identical data packets to the CN through both WLANs, the network load doubles. Therefore, an operation through which to return to single-path transmission is needed. Figure 5 illustrates the operation for returning to single-path transmission. Here, we focus on only the operation of IF2, because both interfaces have the same operation in multi-path transmission. When switching to multi-path transmission, the number of retries that a packet experiences is used as a switching criterion. However, since the number of retries is basically random, it is not appropriate that the stability of a wireless link is detected from the number of retries for only one packet. Therefore, in order to measure the stability of a wireless link, we provide a Stability Counter (SC) for each interface (SC IF1, SC IF2) on the HM and the Single-Path Threshold (SPT), which is the threshold for returning to single-path transmission. The SC is operated only in multi-path transmission. When the number of retries from the MAC layer on IF2 is zero, the HM increases SC IF2 by one; otherwise the HM resets SC IF2 to zero. When SC IF2 exceeds the SPT, the HM judges that the WLAN used by IF2 is stable and returns to single-path transmission through IF2. That is, the HM prevents packet losses while properly switching between single-path and multipath transmission during handover. IV. SIMULATION EXPERIMENTS In this section, the performance of the proposed scheme is evaluated through simulation experiments. The primary concern is how often packet loss occurs and how much traffic is added due to parallel transfer during handover. Furthermore, the and SPT employed by the HM will be tuned in order to attain low network load while meeting the requirement VoIP (G.711) packet size:2 bytes interval: 2 ms 1 ms MN(2) WLAN(A) Fig. 6. CN(1) AP1 (82.11b) MN(1) 4 ms 3 ms MN(1) 3 m 4km/h CN(1) 5 ms AP2 (82.11b) 1 ms WLAN(B) Simulation model (MN(1) moves from WLAN(A) to WLAN(B)) for packet loss. We implement the proposed scheme in the Network Simulator Version 2 (Ver. 2.27). A. Simulation Model Figure 6 illustrates the simulation model. In our simulation, 1 MNs first execute VoIP communication using the CNs in WLAN(A), after which MN(1) with two WLAN interfaces moves from WLAN(A) to WLAN(B) at a walking speed of 4 km/h. Both WLANs are assumed to be administrated by different organizations. The one-way delay to the CN from each WLAN differs: that from WLAN(A) is set to 35 ms and that from WLAN(B) is set to 1 ms. te the hidden terminal problem is not considered. B. Simulation Results The influences of the and the SPT on packet loss and network load are examined in detail. Table 1 shows the communication quality required by VoIP . The and SPT should be properly adjusted so as to satisfy the communication quality requirement.
5 TABLE I REQUIRED COMMUNICATION QUALITY FOR VOIP Quality Good Average Poor Round trip delay < 15 ms 15-4 ms > 4 ms Jitter < 2 ms 2-5 ms > 5 ms Lost packet < 1% 1-3 % > 3% Occurence of Packet Loss SPT=1 SPT=2 SPT=3 SPT=4 SPT=5 Lost Packet Rate [%] SPT=1 SPT=2.5 SPT=3 SPT=4 SPT= Fig. 7. Average lost packet rate during handover 1) Lost Packet: Figure 7 shows the average lost packet rate during handover. From Table 1, in order to maintain the required communication quality for VoIP, the lost packet rate should be maintained at or below 3%. When the is three or four, the requirement on lost packet rate is satisfied. As shown in the figure, the achieved lost packet rate is less than 2%. On the other hand, the lost packet rate is larger for other values and exceeds 3% for some values. The reason why the lost packet rate is increased by the is as follows. Figure 8 shows how frequently packet loss occurs during handover. One occurrence of packet loss is considered to be either one lost packet or consecutive lost packets. When the is equal to one, the HM can frequently switch to milti-path trasnsmission in response to only one retry, and can often switch back to single-path transmission soon beacuse the HM is too sensitive to a retry. In the simulation model, the delay to the CN from WLAN(B) is smaller than that from WLAN(A), and there is no contention in WLAN(B). Therefore, a packet sent from WLAN(B) immediately after the start of multi-path transmission can arrive at the CN earlier than packets sent from WLAN(A) just before the start of multipath transmission. The packets sent from WLAN(A) arriving at the destination after the packet sent from WLAN(B) are regarded as lost packets. For the above reason, packet loss can occur frequently when the is equal to one, as shown in Figure 8. On the other hand, consider the case in which the is set to seven. A wireless link of WLAN(A) is shared by 1 MNs and an AP. Thus, frames are very likely to experience Fig. 8. Average number of occurrences of packet loss a waiting time for sending frames due to contention. In addition, even if the wireless link condition is becoming worse and a data frame suffers retries, the MN will continue to try to transmit the same data frame until its transmission fails seven times. When the number of retries reaches seven, several frames will be queued in the buffer on IF1. After that, the HM switches to multi-path transmission, causing bursty lost packets, because a packet sent through WLAN(B) arrives at the CN earlier than packets queued on IF1. Therefore, when the is small, the number of packet loss occurrences increases, although a large number of consecutive packets are not lost. On the other hand, when the is large, the number of packet loss occurrences decreases, although a large number of consecutive packets can be lost. Bursty lost packets, which adversely affect VoIP communication quality more severely than random lost packets, are very likely to occur in WLANs . Therefore, in order to maintain VoIP communication quality, consecutive lost packets should also be limited. The distribution of the number of consecutive lost packets is shown in Figure 9 for s of three and four. The number of consecutive lost packets is relatively small for the MTP of three, compared with that for the of four. The number of consecutive lost packets is not so sensitive to the SPT, except for the SPT of one, for which the number of consecutive lost packets is relatively large. Therefore, an of three and an SPT of two or more is recommended. 2) Network Load: In the proposed method, the network load due to the multi-path transmission should be decreased as much as possible. Figure 1 illustrates the network load during handover. The network load is relatively small when the is large and the SPT is small. When the is set to three and the SPT is set to two, the load of the network is 1.4 times (8 kb/s * 1.4 = 8.32 kb/s), an increase of only.4%. This is an acceptable value for the network load in order to prevent packet loss for VoIP communication during handover.
6 Cumulative Distribution Network Load =3 SPT=1 =3 SPT=2 =3 SPT=3.4 =3 SPT=4 =3 SPT=5 =4 SPT=1 =4 SPT=2.2 =4 SPT=3 =4 SPT=4 =4 SPT= Fig Consecutive Lost Packets Distribution of consecutive lost packets during handover Fig. 1. SPT=1 SPT=2 SPT=3 SPT=4 SPT=5 Network load during handover V. CONCLUSIONS The focus of this paper has been to develop an effective handover management scheme for VoIP communication in which the quality is sensitive to packet loss. For this purpose, we have proposed a handover management scheme that employs a transport protocol that supports multiple connections for VoIP communication and makes handover decisions based on the number of retries of a data frame, thus reducing packet losses during handover. An MN having two or more WLAN interfaces prevents packet loss during handover by estimating in advance the possibility that packet loss will occur based on the number of retries and properly choosing multi-path or single-path transmission. Simulation results show that an MN can move between WLANs with the low lost packet rate required for VoIP. In addition, the increase of the network load due to multi-path transmission should be limited. The obtained simulation results have shown that the additional network load can be reduced by appropriately adjusting the and SPT while satisfying the required lost packet rate. With an of three and an SPT of two, the network load is increased by only.4%. In other words, in order to carry VoIP traffic at 8 kb/s, a bandwidth of 8 kb/s * 1.4 = 8.32 kb/s will be consumed. Finally, the obtained results show that the proposed scheme can achieve the required communication quality for VoIP, even during handover, while limiting the amount of redundant traffic due to multi-path transmission to an acceptable level. ACKNOWLEDGMENT This study was supported in part by the Japan Society for the Promotion of Science, by a Grant-in-Aid for Scientific Research (A)(1525), and by the Ministry of Internal Affairs and Communications (MIC) of Japan. REFERENCES  IEEE 82.11, 1999 Edition, Available at  C. Perkins, ed., IP Mobility Support for IPv4, IETF RFC3344, August 22.  D. Johnson, et al., IP Mobility Support in IPv6, IETF RFC3775, June 24.  A. Mishra, et al., An Empirical Analysis of the IEEE MAC Layer Handoff Process, ACM SIGCOMM Computer Communication Review, Vol.33, Issue 2, April 23.  A. Dutta, et al., Application Layer Mobility Management Scheme for Wireless Internet, Proc. of IEEE 3G Wireless 21, San francisco, 21.  R. Koodli, Fast Handovers for Mobile IPv6, IETF Internet Draft, draftietf-mobileip-fast-mipv6-8.txt, October 23.  H. Soliman, et al., Hierarchical Mobile IPv6 mobility management (HMIPv6), IETF Internet Draft, draft-ietf-mipshop-hmipv6-2.txt, June 24.  S. J. Koh, et al., Mobile SCTP for Transport Layer Mobility, IETF Internet Draft, draft-sjkoh-sctp-mobility-4.txt, June 24.  R. Stewart, et al., Stream Control Protocol (SCTP) Dynamic Address Reconfiguration, IETF Internet Draft, draft-ietf-tsvwgaddip-sctp-9.txt, June 24.  The Network Simulator: ns-2,  IEEE 82.11b-1999, Available at  K. Medepalli, et al., Voice Capacity of IEEE 82.11b, 82.11a and 82.11g Wireless LANs, Proc. of Globecom 24, Dallas, December 24.  S. Shakkottai, et al., Cross-Layer Design for Wireless Networks, IEEE Communications Magazine, Volume 41,. 1, pp.74-8, October 23.  A. Schmitter, et al., Analysis of network conformity with voice over IP specifications, Irish Systems and Signals Conference, Limerick, Ireland, pp , July 23.  A. D. Clark, Modeling the Effects of Burst Packet Loss and Recency on Subjective Voice Quality, Proc. of IP Telephony Workshop 21, pp , April 21.