Multiplexing VoIP Packets over Wireless Mesh Networks: A Survey

Abstract

Wireless mesh networks (WMNs) have been increasingly applied in private and public networks during the last decade. In a different context, voice over IP (VoIP) has emerged as a new technology for making voice calls around the world over IP networks and is replacing traditional telecommunication systems. The popularity of the two technologies motivated the deployment of VoIP over WMNs. However, VoIP over WMNs suffers from inefficient bandwidth utilization because of two reasons: i) attaching 40-byte RTP/UDP/IP header to a small VoIP payload (e.g., 10 bytes) and ii) 841 μs delay overhead of each packet in WMNs. Among several solutions, VoIP packet multiplexing is the most prominent one. This technique combines several VoIP packets in one header. In this study, we will survey all the VoIP multiplexing methods over WMNs. This study provides a clear understanding of the VoIP bandwidth utilization problem over WMNs, discusses the general approaches in which packet multiplexing methods could be performed, provides a detailed study of present multiplexing techniques, shows the aspects that hinder the VoIP multiplexing methods, discusses the factors affected by VoIP multiplexing schemes, shows the merits and demerits of different multiplexing approaches, provides guidelines for designing a new improved multiplexing technique, and provides directions for future research. This study contributes by providing guidance for designing a suitable and robust method to multiplex VoIP packets over WMNs.

1. Introduction

Voice over IP (VoIP) has grown tremendously in the last few years. According to [1], the amount of VoIP traffic over the Internet in 2015 is around 156 petabytes per month. In 2013, 214 billion minutes of VoIP calls were run on the Skype application alone [2]. The main driver behind this tremendous adoption of VoIP is that VoIP calls are low cost and are sometimes free. A second driver is the ability to integrate other services with voice such as voicemail to email transcription, text chatting, video, and …etc. Another vital driver is that VoIP efficiently improves bandwidth utilization when compared to the traditional PSTN (public switched telephone network) system. Whereas, the 64-Kb/s channel in a PSTN can accommodate many VoIP calls, which require less than 10 Kb/s [3], [4].

Wireless local area network (WLAN), specifically IEEE 802.11 standard, has been increasingly applied in various networks, from business networks to home networks [5]. To install an IEEE 802.11 network, every access point (AP) needs to be connected to another network through a wired link. This wired connection imposes extra installation complexity and high installation costs, especially in environments that need several APs. To handle this issue, IEEE 802.11s [a.k.a wireless mesh networks (WMNs)] emerged as an extension to IEEE 802.11. Instead of using several wired APs as in an IEEE 802.11 network, WMNs require only one AP to connect to a wired link; the wireless mesh routers are used to extend the coverage area [6], [7], [8]. Seamless installation, scalability, turnaround device failure, and low troubleshooting costs have promoted the wide utilization of WMNs, thereby replacing the typical IEEE 802.11 standard [9]. With this distribution, the deployment of VoIP over WMN is an attractive solution that has gained considerable attention in all sectors [10].

However, VoIP over WMNs is facing two main problems. First, the QoS of voice is degrading over WMN due to packet loss, which is the number of lost packets before they reached the destination, delay, which is the time consumed to transmit the packet between a caller and a callee, and jitter, which is the variation of delay. The increasing of these three factors (packet loss, delay and jitter) in WMNs is due to i) the high susceptible to channel interference, ii) the limitation of the avilable bandwidth, iii) the increasing in channel contention because of increasing the typical transmission-control protocol (TCP) traffic, such as Web and e-mail traffic, iv) the access delay which is the time between arriving the VoIP packet to the AP until it is either successfully transmitted over the WMN or dropped because it has waits long to be transmitted, and v) using distributed coordination function (DCF) channel, which does not consider the sensitivity of VoIP to delay, since point coordination function (PCF) cannot be used [8], [11], [12], [13]. Apart from QoS, a key problem is the inefficient bandwidth utilization resulting from large header overhead (as discussed in section 2.2), thereby limiting the WMN transmission capacity (number of simultaneous calls) [14], [15], [16]. Many efforts have been made to solve this issue. One of these efforts is VoIP packet multiplexing, in which multiple VoIP packets are combined in one header [15], [17], [18]. This work discusses the present packet multiplexing methods of VoIP packets over WMNs. The work provides a clear understanding of the inefficient bandwidth utilization of VoIP over WMN and the impact of packet multiplexing on improve bandwidth utilization and how it reflects to VoIP QoS. In addition, this work provides some general directions for designing a robust multiplexing technique. To the best of our knowledge, this survey is the first one that focuses on VoIP multiplexing methods over WMNs.

The rest of this paper is organized as follows: Section 2 briefly describes WMNs, VoIP, and multiplexing VoIP packets. Section 3 analyzes the present multiplexing methods of VoIP packets over WMNs. Section 4 discusses the present multiplexing methods in general. Section 5 provides guidelines for designing a multiplexing method and introduces suggestions for future research. Finally, Section 6 elaborates the conclusions.

 

2. Background

This section highlights the topics related to this work in order to provide a better understanding to the reader. These topics include WMNs, VoIP over WMNs, and VoIP packet multiplexing.

2.1 WMNs

In the last decade, WMN has become a new multi-hop wireless standard that extended from the IEEE 802.11 standard. In WMN, the two IEEE 802.11 devices, namely, AP and client, are turned into mesh APs (MAPs) and mesh clients, respectively. In addition, WMN introduces a new device called mesh points (MPs). A mesh client can be any end device with a wireless network interface card, such as laptops and mobile phones. MAP is a device connected to the wired network in one side and broadcasts the data wirelessly on the other side. An MP can be viewed as a WMN router and is used along with MAP to extend the range of the wireless network [19], [20]. Fig. 1 shows the general architecture of WMN.

Fig. 1.Architecture of WMN

With regard to the WMN frame format, the frame header of the original IEEE 802.11 standard was extended to include new control fields for mesh technology. These new fields are embedded in the IEEE 802.11 frame body, as shown in Fig. 2. The total size of the mesh header is 40 bytes for IEEE 802.11 plus mesh control fields of up to 24 bytes, thereby resulting in 64 bytes of mesh header. The header overhead is considered to be large especially when compared with small packet applications, such as VoIP as we will see in the next subsection [20], [21].

Fig. 2.WMN frame format

2.2 VoIP over WMNs

VoIP is a technology that allows voice calls through an IP network. VoIP quality is highly sensitive to delay and packet loss (high delay and packet loss result in poor VoIP quality). To travel over an IP network, the voice is captured and digitized by hardware or software called a codec. The codec then compresses and converts the digitized data to small size frames (voice packet payload). Each frame is produced in a specific time period. Enlarging the frame size will increase the time period. Such enlargement will increase the end-to-end delay, thereby resulting in poor VoIP quality. Accordingly, the frame is limited to specific small sizes. The frame size and time period vary depending on the type of codec. Table 1 shows some of the voice codecs. For example, the G.729 codec produces a 20-byte packet payload (2 frames) in a 20 ms duration. The produced frame is then attached to a 40-byte RTP/UDP/IP header and passed down to the lower layer [data link layer (DLL)], which is attached to another header depending on the technology [18], [22], [23], [24].

Table 1.Voice codecs

In the case of WMN technology, in addition to the 40-byte RTP/UDP/IP header, a header of up to 64-bytes (as shown in the previous section) is attached to the voice frame, thereby resulting in a 104-byte header. If an 11 Mb/s network is considered, then the transmission time of 104 bytes voice packet is 104*8/11= 76 μs. Each packet should wait 50 μs distributed interframe space (DIFS) and typically 310 μs backoff (BO) time to make sure that the channel is idle before transmitting the packet. In addition, each packet is followed by 10 μs short inter-frame space (SIFS) and 11 μs acknowledgment (ACK) to process a received frame and ACK the frame, respectively. Further, 192 μs physical layer header [144 μs preamble and 48 μs physical layer convergence protocol (PLCP)] is attached to each packet and ACK. Accordingly, the physical layer header of the data, ACK, physical layer header of ACK, backoff (BO) time, SIFS, and DIFS add around 765 μs overhead to the 76 μs mesh header, thereby resulting in around 841 μs overhead of each packet. Meanwhile, the transmission time of a 20-byte G.729 payload in the 11 Mb/s network is only 20*8/11=14 μs. Therefore, the voice data consume less than 2% of the transmission time [20], [21], [25], [26], [27]. Accordingly, a large header overhead is added to each packet, thereby wasting bandwidth resources. Fig. 3 shows the overhead of sending one VoIP packet assuming a 20-byte G.729. VoIP packet multiplexing is one of the techniques for alleviating the header overhead problem [15], [17], [18]. This technique will be discussed in the next subsection.

Fig. 3.Time to send one VoIP packet

2.3 Packet Multiplexing

The basic philosophy of packet multiplexing is to combine several VoIP packets in one header. The multiplexing entity at the sender side checks the packet destination and then combines the packets that travel in the same route in one header, thereby producing one multiplexed packet (mux-pkt). Upon receiving the mux-pkt, the de-multiplexing entity at the receiver side separates and restores the combined packets. Fig. 4 depicts the general packet multiplexing/de-multiplexing process. Packet multiplexing is one of the main techniques for handling the large overhead problem caused by carrying a VoIP packet over WMNs. Fig. 5 demonstrates the multiplexing of three packets into one packet. The top part of the figure (Fig. 5) shows the time needed to transmit three packets separately. As discussed in section 2.2, 841 μs overhead is required to transmit each packet, with a total of 2523 μs (841 * 3) to transmit the three packets. The bottom part of the figure (Fig. 5) shows the time needed to transmit the mux-pkt that resulting from multiplexing the three packets. The mux-pkt requires only 841 μs, instead of 841 μs for each of the three separate packets. Therefore, a considerable header overhead reduction can be achieved using packet multiplexing methods. Clearly, multiplexing more packets in a mux-pkt (larger mux-pkt size) achieves better overhead reduction and provides higher bandwidth utilization. Packet multiplexing can occur at the DLL, network, transport, or even in the application layer. The high layer causes high bandwidth utilization because the header overhead of the multiplexed packets will be reduced [8], [15], [28].

Fig. 4.Multiplexing/de-multiplexing process

Fig. 5.Reducing header overhead by packet multiplexing

 

3. VoIP Multiplexing Methods Over WMNs

This section discusses the current multiplexing methods of WMNs. First, the general multiplexing approaches over WMNs are introduced. Then, the multiplexing methods are divided into two groups: adaptive and non-adaptive multiplexing methods. Finally, a brief summary of all the multiplexing methods is provided.

3.1 Multiplexing Approaches

Different general approaches of packet multiplexing methods over WMNs are available. This subsection discusses these approaches, these are: hop-to-hop, end-to-end, adaptive and non-adaptive multiplexing.

3.1.1 Hop-to-hop multiplexing

This approach performs packet multiplexing/de-multiplexing and enforces multiplexing delay at all hops throughout the route between source and destination. The end-to-end delay increases because of the multiplexing/de-multiplexing process at each hop. However, hop-to-hop approach achieves high bandwidth utilization because the possibility of multiplexing more packets increases with the number of hops. In addition, the route between the two hops limits the maximum size of mux-pkt. In particular, a high-quality route carries large mux-pkt, and a low-quality route carries small mux-pkt, thereby improving the bandwidth utilization of multiplexing methods [14], [29].

3.1.2 End-to-end multiplexing

This approach performs packet multiplexing at the source hop and packet de-multiplexing at the destination hop only. Therefore, the end-to-end delay that results from multiplexing/de-multiplexing process is less than that in the hop-to-hop approach. However, this approach achieves lower bandwidth utilization than the previous approach does because packet multiplexing occurs at the source hop only. In addition, the mux-pkt size is limited by the weakest link; hence, the bandwidth utilization of multiplexing methods is degraded [14], [29].

3.1.3 Adaptive multiplexing

In this approach the mux-pkt size is controlled and changed adaptively based on the link quality parameters, including; load, end-to-end delay, bit error rate, and congestion. Whereas, in poor quality links large mux-pkt is more susceptibility to noise, interference, distortion or bit synchronization errors, which increases the bit error rate. Therefore, increases the packet size increases its possibility to get damaged and, thus, dropping. On the other hand, in high quality links large mux-pkt reduces the header overhead and improves bandwidth utilization. Accordingly, this approach bounded the mux-pkt size by link quality parameters [7], [10], [15].

3.1.4 Non-adaptive multiplexing

In this approach, mux-pkt size is bounded by certain thresholds, such as time period, number of packets, and size. The actual link quality status is ignored, and, thus, the mux-pkt size might not suit the link quality; hence, the performance of the multiplexing method might be degraded [7], [15]. In all approaches, the mux-pkt size should be less than the maximum transmission unit (MTU) size, which is 2300 bytes in WMNs [30].

3.2 Present Multiplexing Methods Over WMNs

This section introduces the present multiplexing methods over WMNs. The methods are divided into two main groups: adaptive and non-adaptive multiplexing methods. As mentioned above, the main difference between the adaptive multiplexing and non-adaptive multiplexing is that the mux-pkt size in the adaptive multiplexing is changing adaptively based on the link quality, while in the non-adaptive multiplexing is bounded by certain thresholds. However, the general multiplexing process of both adaptive multiplexing and non-adaptive multiplexing is similar. Fig. 6 shows the general process of a multiplexing method. In most papers, other contributions in addition to packet multiplexing were provided, such as header compression, mobility, and routing. In this work, we will mention these contributions but will not discuss them. Additional details on these contributions can be found in the original papers, given that this work highlights and discusses only packet multiplexing methods.

Fig. 6.The general multiplexing process

3.2.1 Non-adaptive Multiplexing Methods

The studies in [31], [32] proposed a hop-by-hop multiplexing mechanism that works at the DLL layer. Three configurable parameters are used to control the packet size and the multiplexing process: SIZEmin, which is the minimum size of the mux-pkt; SIZEmax, which is the maximum size of the mux-pkt; and MAXdelay, which is the maximum waiting time to multiplex the packets. The multiplexing process works as follows: If available packet size > SIZEmin, then the packets are multiplexed while mux-pkt size < SIZEmax. If waiting time > MAXdelay, then the packets are multiplexed while mux-pkt size < SIZEmax. Accordingly, the mux-pkt is produced by attaching a 20-byte IP header to the mux-pkts. At least two packets should be available to accomplish multiplexing. Finally, the mux-pkt is transmitted to its destination. The de-multiplexer uses the information in the IP header to find if this packet is mux-pkt and to extract the multiplexed packets within the mux-pkt. As recommended by these works, SIZEmin and MAXdelay should trade off bandwidth utilization and delay. For example, increasing MAXdelay to multiplex more packets increases the end-to-end delay. By contrast, decreasing MAXdelay to reduce the end-to-end delay decreases the mux-pkt size, thereby reducing the bandwidth utilization. The result for the tested cases showed that the bandwidth utilization was improved to more than four times that in the traditional situation (without multiplexing). The QoS (delay, jitter, and packet loss) is acceptable when the number of connections is around 80 with “no multiplexing” scenario and 354 with “multiplexing” scenario. If the number of connections exceeds these connections, then the QoS is degraded. In terms of mean opinion score (MOS), the “no multiplexing” scenario shows better MOS because this scenario imposes less delay than the “multiplexing” scenario does.

In [33], Ganguly, Samrat et al. proposed a hop-by-hop multiplexing method that multiplexes multiple packets from different flows that travel in the same route in one mux-pkt. First, the network is monitored to determine the hop(s) at which the flows converge (multiple flows pass through the same hop simultaneously). If the interference of that hop is within a certain level, then the hop is chosen as the multiplexer. Otherwise, another hop with multiple converged flows and acceptable interference is chosen as the multiplexer. Afterwards, the network is monitored to find the hop where the flows are diverging. This hop is chosen as the de-multiplexer. After determining the multiplexer and de-multiplexer, the packets are combined in a single mux-pkt at the DLL layer up to a certain size. Then, the aggregated packet is transmitted through a route determined based on the shortest-hop routing method. Upon reaching the de-multiplexer, packets are de-multiplexed and further transmitted to their destination. In addition to multiplexing, this invention compresses the 40-byte RTP/UDP/IP header to 26 bytes, thereby improving the bandwidth utilization.

Kekre, H. B. et al. [26] combined packet multiplexing and multicasting (M-M). The primary idea of M-M method is to compress the RTP/UDP/IP header of the VoIP packet and attach a mini-header with a unique ID to distinguish the multiplexed packets at the de-multiplexer. Then, several packets are multiplexed in one large packet at the network layer. Finally, the mux-pkt is multicast by a single transmission to all stations in a multicast group. The multiplexer multiplexes the packets for certain fixed period (T). The value of T should be chosen to compromise between bandwidth efficiency and end-to-end delay.

The authors mentioned that the packets can be unintentionally captured and read by a certain node by multicasting the packets to all nodes. To solve this issue, the authors proposed the encryption of the voice packets. Although encrypted, the packet is still more susceptible to being hacked than in the other methods without multicasting. In addition, the encryption/decryption process adds extra delay and degrades the overall quality. Multicasting the packets to all nodes can also cause large processing overhead on the nodes with a data not intended to them.

Khyati Marwah et al. [34] improved the e-model of the International Telecommunication Union (ITU) to measure the effect of VoIP packet multiplexing on VoIP quality over WMNs. They extended the rating factor (R-factor) equation by including the multiplexing delay Ad parameter. The proposed e-model was implemented over the end-to-end multiplexing method, and several packets were multiplexed at the MAC layer. In the evaluation test, the number of multiplexed packets in the mux-pkt increased gradually. Results showed that increasing the number of multiplexed packets increases the multiplexing delay, thereby increasing the end-to-end delay and influencing the R-factor. The results also showed that, if the number of multiplexed packets increases, then the MOS decreases. Increasing delay and decreasing the MOS have negative effects on the QoS of VoIP. Accordingly, the packet multiplexing is acceptable up to a certain threshold to avoid the low QoS of VoIP. However, the proposed e-model extension is implemented only in fixed mux-pkt size. Therefore, the authors fail to demonstrate packet multiplexing by including adaptive packet multiplexing. In addition, the effect of other factors on VoIP QoS, such as packet loss and jitter, were not explored.

In summary, this subsection discusses the non-adaptive multiplexing methods. The work in [34] only measures the effect of packet multiplexing on VoIP quality over WMNs. In [26], although the work proposes to bound the mux-pkt size with T time period, no experiments are performed to evaluate the proposed method. In addition, multicasting the packets to all stations causes unneccesary overhead on the network and reduces its perfomance. The works in [31] and [32] bound the mux-pkt size with three parameters: SIZEmin, SIZEmax, and MAXdelay. Accordingly, the mux-pkt size can be selected properly. However, these works perform the multiplexing/de-multiplexing process at each hop; hence, the end-to-end delay increases, thereby degrading the VoIP quality. Notably, the bandwidth utilization improves. The work in [33] proposes a noticeable method that performs the multiplexing/de-multiplexing process only at specific hops where packets converge/diverge. As a result, the delay that results from multiplexing/de-multiplexing process at each hop is reduced, and high bandwidth utilization is maintained. In addition, the multiplexing/de-multiplexing hops are chosen to be within a specific interference level to ensure an improved multiplexing/de-multiplexing process. Therefore, this method can be selected as the best among the non-adaptive multiplexing methods. However, the non-adaptive multiplexing methods usually use certain fixed values to determine the mux-pkt size, thereby degrading the performance of the multiplexing method. In other words, the aforementioned network conditions (interference and load) are changing continuously, especially in wireless networks. Therefore, when the network conditions are good (low interference and low load) the mux-pkt size may increase, thus improving bandwidth utilization. By contrast, when the network conditions are poor (high interference and high load) the mux-pkt size may decrease, thereby reducing the delay caused by the multiplexing process time and the packet loss induced by the large mux-pkt size. Consequently, the VoIP conversation quality is improved. Therefore, the mux-pkt size in a packet multiplexing method should change adaptively according to the network conditions. Such feature is incorporated in the adaptive multiplexing methods that will be discussed in the following subsection [7], [10], [15].

3.2.2 Adaptive Multiplexing Methods

Niculescu, Dragos et al. [35] proposed a multiplexing method that multiplexes VoIP packets at the network layer. The multiplexing occurs at every hop from source to destination unless it reaches the MTU. However, this method is not solely a hop-to-hop multiplexing method because it does not force a multiplexing delay at each hop throughout the path. This method forces a variable multiplexing delay (based on the link status) on the ingress hop only and uses the natural queuing delay to multiplex at the intermediate hops. Therefore, similar to end-to-end multiplexing methods, the proposed method reduces the end-to-end delay that results from hop-by-hop multiplexing. Moreover, similar to hop-by-hop multiplexing methods, this method improves bandwidth utilization by performing multiplexing at each hop. The detailed multiplexing algorithm is shown in Fig. 7. In addition to packet multiplexing, a header compression and routing algorithm for selecting the best route was proposed in this work. The result for the tested cases showed that the bandwidth utilization improved slightly compared with that in the traditional situation. The method also does not impose any additional delay because it uses “the wait for the MAC availability” for packet aggregation.

Fig. 7.Multiplexing algorithm

Another method [36] that combines the features of hop-by-hop and end-to-end multiplexing methods was proposed by Kim, Kyungtae, and Sangjin Hong. Their proposed method forces delay only at the ingress hop and takes advantage of the queuing delay at the intermediate hops. The multiplexing method is as follows: First, the packets from the same flow are multiplexed together at the ingress hop (or edge multiplexer) in a mux-pkt, within a variable period of time for each flow. Then, the resultant mux-pkt are multiplexed together at the DLL layer at the edge multiplexer. Afterwards, the backhaul multiplexer, which works between hops only, multiplexes the resultant packets in one large packet. If the size of this packet reaches a specific size, then it is transfered directly to the destination to save resources; otherwise, it is transfered to the next hop for further multiplexing. A header compression algorithm was also proposed in this work. However, the proposed header compression is different from that in [35]. The result for the tested cases showed that the bandwidth utilization improved by nearly four times that in the traditional situation. Furthermore, the method does not impose any additional delay owing to its use of “the wait for the MAC availability” for packet aggregation.

Kim, Kyungtae et al. [38] proposed to multiplex multiple VoIP frames at the DLL layer. The mux-pkt size changes adaptively based on two thresholds: time and size thresholds. An algorithm was proposed to show where the delay should occur and the duration of the delay in each hop. Therefore, a better result can be obtained by this approach than when the typical hop-to-hop approach is used. If one of the two thresholds is reached, then the mux-pkt is transmitted to its destination. The detailed multiplexing algorithm is shown in Fig. 8. The result for the tested cases showed that the bandwidth utilization improved by more than four times that in the traditional situation. The QoS (jitter and packet loss) improved slightly compared with that in the traditional method.

Fig. 8.Multiplexing algorithm

The same multiplexing method in [37] was exploited by Ganguly, Samrat et al. in [38] with minor modification. Specifically, multiple VoIP packets were multiplexed at the network layer instead of the DLL layer. In addition to packet multiplexing, a header compression technique was proposed, which increases the number of running calls and improves bandwidth utilization. This work provides two other contributions unrelated to bandwidth utilization. First, their proposed method maintains the VoIP quality by sending the packets through the route with less delay and packet loss. Second, a method that supports mobility is used, in which the call is kept stable (without disruption) when moving to other APs while the call is ongoing. In [9], the same multiplexing method and header compression were combined together and proposed as a patent. The result for the tested cases showed that the bandwidth utilization improved by two to three times that in the traditional situation. The QoS (jitter and packet loss) is improved compared with that in the traditional method.

The work in [39] implemented a hop-to-hop multiplexing method that adaptively multiplexes the packets at the DLL layer based on the link status. The optimal mux-pkt size is calculated at each hop using a specific equation of signal-to-noise ratio (SNR). Four parameters are used to control the mux-pkt size and the multiplexing process: SIZEmin, which is the minimum size of the mux-pkt; SIZEmax, which is the maximum size of the mux-pkt and is determined as a function of SNR; MAXdelay, which is the maximum waiting time to multiplex the packets; and SIZEfactor, which compromises between packet loss caused by frame errors and congestion. SIZEmax is determined as a function of SNR. If available packet size > SIZEmin, then the packets are multiplexed while mux-pkt size < SIZEmax * SIZEfactor. If waiting time > MAXdelay, then the packets are multiplexed while mux-pkt size < SIZEmax. After accomplishing multiplexing, a 20-byte IP header is attached to each mux-pkt. At least two packets should be available to perform the multiplexing process. The result for the tested cases showed that the bandwidth utilization was improved by more than three times that in the traditional situation. In addition, the QoS (delay, jitter, and packet loss) is acceptable when the number of connections is around 40 with “no aggregation” scenario and 120 with “aggregation” scenario. If the number of connections exceeds these connections or traffic is low, then the QoS is degraded.

The works in [40] and [41] proposed a method that combines packet multiplexing and packet differentiation. Upon receiving the packets, the hop differentiates them using the differentiated service code point (DSCP) field of the IP. These packets are then distributed into four queues; one of the queues is for VoIP traffic. Each queue uses a packet multiplexing method to save bandwidth. For the VoIP, the work proposes an adaptive multiplexing method that works at the DLL layer. This method imposes a multiplexing delay on the ingress hop only and uses the natural queuing delay for multiplexing at the intermediate hops. Therefore, this method reduces the end-to-end delay that results from hop-by-hop methods and improves bandwidth utilization by performing multiplexing at each hop. To calculate the optimal mux-pkt size, the measurable routing metrics and the number of stations in range were used in a specific formula. The mux-pkt is produced and transmitted when multiplexing period is expired or optimal size is reached. The result for the tested cases showed that the bandwidth utilization was improved by around four times that in the traditional situation. The method imposes less delay when the number of flows is small and higher delay when the number of flows increases. However, the packet loss is less than that in the traditional method.

In [42], the author proposed a novel multiplexing method that multiplexes the VoIP packets from the same call in one mux-pkt. Typically, VoIP packets from the same call are multiplexed by elongating the voice payload at the source of the call, thereby increasing the packetization delay. Unlike the typical methods, this work proposed to multiplex the VoIP packets from the same call at each router from source to destination with zero multiplexing delay. The resulted mux-pkt shares the same UDP and IP header. The multiplexing is performed based on the congestion level of the router. The basic philosophy of this method is that the current bandwidth is sufficient if no congestion is detected. Therefore, performing multiplexing is unnecessary. By contrast, if congestion is detected and the packets are waiting in the queue, then the method uses this waiting time to multiplex the packets from the same call without any additional multiplexing delay. Clearly, the mux-pkt size is adaptive based on the congestion level in the router. For example, if light congestion is detected on the router, then the number of multiplexed packets is low; if high congestion is detected on the router, then the number of multiplexed packets is high. The result showed that the number of calls was double those in the traditional situation, without any additional delay.

Hasegawa, Jun et al. [43] proposed a unique method that combines both packet multiplexing and network coding. Their proposed technique utilizes some network features to improve the network transmission capacity. The proposed method is called bidirectional packet multiplexing and coding (BiPAC). BiPAC performs multiplexing and network coding in each hop and inserts a mini-header between the DLL and IP headers to restore multiplexed packets. Among many network coding techniques, XOR-based network coding was used by BiPAC because this coding is efficient in WMNs. Encoding increases packet overhead because the packet IDs of all XORed packets must be added as a header for network coding. To reduce header overhead, BiPAC multiplexes the packets and then performs the XOR operation on the multiplexed packets. The result for the tested cases showed that the bandwidth utilization was improved by nearly four and a half times that in the traditional situation. The method imposes high delay when the number of flows is small and nearly the same delay when the number of flows increases. However, the packet loss is less than that in the traditional method.

The work in [10] proposed a hop-to-hop multiplexing method that multiplexes several packets at the network layer. The method was implemented over DCF channel because it is more common than PCF channel. The author argued that the proposed multiplexing method was implemented to run over IP level routing, and thus, the method can be implemented to run over DLL layer routing. A mini-header separates the packets within the mux-pkt. The receiver uses the information in the mini-header to extract the packets from the mux-pkt. The size of mux-pkt changes adaptively based on the link status. The signal-to-noise and interference ratio (SNIR) for measuring connection quality was used in a certain function to calculate the optimal mux-pkt size. In [30], [44], and [45], similar multiplexing method was adopted and implemented under different WMN scenarios. The result for the tested cases showed that the bandwidth utilization improved by more than three times that in the traditional situation. The QoS (delay, jitter, and packet loss) is improved compared with that in the traditional method.

The work in [46] combined RObust header compression (ROHC) and packet multiplexing of VoIP packets. A series of testing and evaluation found that ROHC and packet multiplexing highly improved bandwidth utilization and decreased end-to-end delay. However, the bandwidth utilization was degraded because of the processing time of ROHC and packet multiplexing. To reduce the processing time, this work proposed a specific-purpose processor model instead of a general-purpose processor. The proposed model handles ROHC and packet multiplexing. Result showed that the performance improved when using the proposed processor model than when the typical processor is used.

In summary, this subsection discusses the adaptive multiplexing methods. The adaptive multiplexing methods can be divided into two groups. The first group is composed of hop-to-hop methods, including the works in [10], [30], [39], [42], [43], [44], and [45]. The second group comprised hybrid (hop-to-hop/end-to-end) methods, including the works in [36], [37], [38], [40], and [41].

In the first group (hop-to-hop adaptive multiplexing methods), the works in [10], [30], [45], and [46] use the SNIR in a specific equation to evaluate the network performance and thus determine the mux-pkt size. The works in [42] and [43] use size and delay in a specific equation to determine the mux-pkt size. The work in [39] uses SNR and delay in a specific equation to determine the mux-pkt size. In [39], two quality measures (SNR and delay) are used to evaluate the link status, thereby providing a more suitable mux-pkt size as evidenced by the result. Such a result is obtained because the work in [39] improves the bandwidth utilization and keeps acceptable VoIP quality. However, in general, the hop-to-hop adaptive group provides suitable mux-pkt size to a high extent because the methods in this group use the network conditions to calculate the size. In addition, this group achieves high bandwidth utilization because the possibility of multiplexing more packets increases with the number of hops. However, multiplexing packets on every hop in the route imposes additional delay and thus degrades the voice quality. Accordingly, a packet multiplexing method should improve bandwidth utilization and avoid imposing additional delay, as in the second group (hybrid adaptive multiplexing methods) summarized below [37], [38], [40].

In the second group (hybrid adaptive multiplexing methods), the work in [30] uses MTU to determine the mux-pkt size. The work in [36] uses size to determine the mux-pkt size. The works in [37], [38], and [9] use size and delay to determine the mux-pkt size. The works in [40] and [41] use routing metrics and number of stations in specific equation to determine the mux-pkt size. In [35] and [36], the works depend on the MTU and size, respectively, to determine the mux-pkt size. These works also ignored the link status. Therefore, the resulting mux-pkt size might be unsuitable for the network, thereby potentially reducing the efficiency of bandwidth utilization and degrading the voice quality, as discussed earlier. In [40] and [41], the works may also reduce the efficiency of bandwidth utilization and degrade voice quality because they use routing metrics and number of stations to determine the mux-pkt size. These methods also did not consider the link status. The works in [37], [38], and [9] use delay in addition to size and thus resulting in more suitable mux-pkt size. Accordingly, voice quality and bandwidth utilization are better. In addition, they maintain the VoIP quality by sending the packets though the route with less delay and packet loss. However, contrary to the work in [32], which performs the multiplexing at the MAC layer, the work in [38] performs multiplexing at the network layer. The latter work has been patented in [9]. Therefore, more efficient header compression, better bandwidth utilization, and improved voice quality are obtained. Accordingly, the work in [38] and its patent [9] may be considered the best available VoIP multiplexing methods over WMNs. However, in general, the hybrid adaptive group provides a suitable mux-pkt size to a high extent because they use the network conditions to calculate the size. In addition, this group forces a variable multiplexing delay (based on the link status) on the ingress hop only and uses the natural queuing delay for multiplexing at the intermediate hops. Therefore, similar to end-to-end multiplexing methods, these methods reduce the end-to-end delay that results from the hop-by-hop multiplexing. Moreover, similar to hop-by-hop multiplexing methods, they improve bandwidth utilization by performing multiplexing at each hop. By contrast, this group shows bandwidth utilization that is close to that gained by the hop-by-hop method group. Therefore, using hybrid adaptive multiplexing methods is more favorable for WMNs [37], [38], [40].

This section introduces the present methods of multiplexing VoIP packet over WMNs. The multiplexing methods were divided into adaptive and non-adaptive methods. In non-adaptive methods, the mux-pkt size was determined based on specific thresholds: delay and size. In adaptive methods, the mux-pkt size was determined based on the link quality parameters: delay, size, SNR, SNIR, routing metrics, and number of stations. Table 2 summarizes the adaptive and non-adaptive methods.

Table 2.Summary of adaptive and non-adaptive multiplexing methods

 

4. Discussion

4.1 Effect of packet multiplexing on bandwidth utilization

The main purpose of the VoIP packet multiplexing methods is to improve network bandwidth utilization. Multiplexing methods achieve this aim in several aspects. First, on the one hand, the typical VoIP packet payload size between 10 and 30 bytes depends on the codec. On the other hand, up to 104-byte header is added to each payload. Accordingly, the header overhead, which is the relative ratio between the header size and the packet size, is between around 77.5% to 92%. Fig. 9 demonstrates the header overhead with different payload sizes. When multiplexing several packets in one header, the header overhead decreases depending on the number of multiplexed packets in the mux-pkt, as shown in Fig. 10. Second, as discussed in Section 2.2, 841 μs delay is induced by each small packet. Therefore, when multiplexing several small packets in one large packet, the 841 μs delay is shortened in this large packet, thereby increasing the number of transmitted packets. Third, the capacity (number of concurrent calls) of the link will increase because of two reasons: i) the reduced header overhead, which saves the bandwidth and allows the link to accommodate more calls; and ii) the shortened 841 μs delay of each small packet to one large packets, which reduces the link busy time and allows more calls to be accommodated in the link [15], [47], [48], [49]. All the aforementioned factors (header overhead, delay of each packet, and capacity) reflect the bandwidth utilization. On the basis of these factors, multiplexing methods highly improve bandwidth utilization. Fig. 11 shows the capacity of the method in [38], because this method has been selected by the author as one of the best methods. As we can see, the capacity with packet multiplexing is greater than the capacity without multiplexing. However, the capacity decreases when the number of hops increases.

Fig. 9.Header overhead ratio without multiplexing

Fig. 10.Header overhead ratio with multiplexing

Fig. 11.Number of calls with and without multiplexing

4.2 Effect of packet multiplexing on VoIP quality

VoIP technology must provide high conversation quality to ensure user satisfaction. Several factors affect the quality of VoIP. The first factor is delay. According to the ITU, the acceptable end-to-end delay should be within 150 μs. In addition to the natural delay (packetization, transmission, processing, and queuing delay), the multiplexing methods increase the end-to-end delay because of the imposed multiplexing period. Furthermore, the delay increases on the basis of the processing time that results from multiplexing/de-multiplexing operation. However, the multiplexing methods compensate for this extra delay by decreasing the queuing delay through processing and forwarding one large packet instead of many small packets. In addition, the 841 μs delay is shortened to one large packet instead of several small packets. Multiplexing methods might increase or decrease the delay. Therefore, a mechanism to estimate the amount of time to hold the packet for aggregation and compensate the delay is needed [15], [16], [50], [51], [52]. The second factor is jitter. A high jitter corresponds to a low voice quality. The main causes of jitter are queuing time, contention, and the use of different routes. Multiplexing methods insignificantly affect the jitter in that they improve it slightly. The possible reasons are as follows: i) packet multiplexing decreases the number of packets and thus reduces the contention; ii) packets need to be delayed for some time to accomplish multiplexing; and iii) fewer routes will be traversed because one large packet will be sent instead of multiple small packets [7] , [37], [38]. The third factor is packet loss. For VoIP, long delayed packets are considered lost. The acceptable packet loss ratio depends on the codec and packet size. Different codecs use different packet loss concealment (PLC) algorithms to fill the lost packet with approximate packet. However, less than 4% packet loss is generally acceptable. Multiplexing methods decrease packet loss by improving buffer utilization. This phenomenon is driven by replacing the small packet header with one header of large packet; thus, the buffer takes in more packets. However, a larger packet is more susceptible to damage in high error rate links. Accordingly, a larger packet is more likely to be lost, thereby increasing packet loss rate. In addition, a lost large packet is considered lost bursty traffic. Thus, PLC algorithms are less effective [8], [10], [50], [53]. Therefore, the number of packets to be aggregated into the mux-pkt is an important factor and requires a suitable mechanism to be calculated. R-score is a scale to measure voice quality from 1 to 100. The result in [38], as it has been elected by the author as one of the best methods, shows decreasing in packet loss, nearly same delay and jitter in comparison ot traditional method.

4.3 Handicaps of packet multiplexing

Several issues limit VoIP packet multiplexing over WMNs. First, packet multiplexing degrades the QoS of VoIP. As discussed above, the delay, jitter, and packet loss will increase if the multiplexing method is unsuitable [7], [8], [18], [37], [38], [54]. Second, bandwidth utilization is ineffective in the case of few calls and multi-path routing because fewer packets are available for multiplexing [52], [55]. Third, multiplexing multiple packets from different streams requires the same QoS to all streams, thereby eliminating the traffic prioritization feature [52], [55]. Finally, multiplexing/de-multiplexing operation increases the processing overhead on the hops [46]. However, all these obstacles can be alleviated by choosing suitable criteria for designing a multiplexing method, as discussed in Section 5. Designing a suitable multiplexing method that improves bandwidth utilization is possible but remains a challenge for researchers. Table 3 summarizes the elements affected by packet multiplexing.

Table 3.Elements affected by packet multiplexing

 

5. Guidelines and Potential Research

The main goal of VoIP packet multiplexing is to maximize the link capacity while enhancing or at least maintaining the same VoIP QoS. Several issues should be considered when designing a multiplexing method to achieve this goal. First, an important issue is to choose between adaptive method and non-adaptive method. The mux-pkt size is the main important factor to consider when designing a multiplexing method [7]. As discussed earlier, the mux-pkt size in non-adaptive multiplexing methods is bounded by certain thresholds and does not consider the link status. This design might cause improper mux-pkt size. The improper mux-pkt size causes many problems. For example, i) a large packet size increases packet loss in high error rate links, as shown in the discussion earlier; and ii) in stable links, a small mux-pkt size degrades the bandwidth utilization efficiency. By contrast, adaptive multiplexing methods change the mux-pkt size adaptively based on the link status. Therefore, these methods provide proper mux-pkt size and thus result in better bandwidth utilization and fewer packets compared with those in non-adaptive multiplexing methods [7], [8], [10], [15]. Accordingly, the adaptive multiplexing methods are preferred over non-adaptive multiplexing ones. Second, in the adaptive multiplexing methods, mux-pkt size should be calculated based on a robust equation that considers suitable quality parameters. Third, the multiplexing layer should be selected properly. The multiplexing layer highly affects the bandwidth utilization. In case of application layer, multiple voice frames combined in one packet at the sender side reduce the header overhead effectively. This reduction leads to a significant increase in the packetization delay. Moreover, the mux-pkt will be sensitive to bursty packet loss. In case of network layer, efficient header compression might be applied along with the packet multiplexing, thereby highly improving bandwidth utilization. In addition, multiplexing can be adjusted to suit the network situation. In case of DLL (MAC) layer, multiplexing can be adjusted to suit the network situation and obtain a greater header overhead reduction than in the network layer. However, the bandwidth utilization that results from combining packet multiplexing and header compression at the network layer is higher than that obtained by reducing header overhead at the DLL. Accordingly, multiplexing at the network layer achieves better bandwidth utilization because packet multiplexing and header compression are combined. The multiplexing can be adjusted to suit the network situation, which makes this method the suitable choice. However, its protocol fields need to be changed during the encapsulation process. In addition, WMN MAPs usually operate at the DLL layer and does not work at higher layers [7], [51], [56]. Fourth, the hop-to-hop methods achieve better bandwidth utilization than end-to-end methods do. By contrast, end-to-end methods impose less delay than hop-to-hop methods do. Therefore, combining the two techniques obtains better results. The combination can be achieved by forcing multiplexing delay only on the ingress hop and using the natural queuing delay for multiplexing at the intermediate hops. Specifically, the additional delay is imposed only at the ingress/egress node as in the end-to-end methods, and multiplexing is conducted at each hop to achieve bandwidth utilization close to that in hop-to-hop methods [14], [29], [35]. A robust and efficient multiplexing method can be designed when these issues are considered. However, other trends are not investigated in this work. This limitation may be reason that the performance of the multiplexing methods appears more improved over the other methods. These trends can be addressed in future research. Some of these trends are as follows:

 

6. Conclusion

The prominent characteristics of making voice calls over an IP network have encouraged users and service providers to migrate their traditional telecommunication systems to VoIP telecommunication systems. WMNs have been widely applied in all sectors, including homes, schools, universities, and companies. Thus, combining VoIP and WMN is an inevitable issue in all sectors. Inefficient bandwidth utilization is one of the serious problems that hinder the propagation of VoIP over WMNs. Packet multiplexing combines several VoIP packets in one header and is one of the main methods for handling bandwidth utilization problems. In this work, we surveyed the entire current packet multiplexing methods of VoIP packets over WMNs. This study provided a detailed investigation of the present multiplexing methods over WMNs. This study also provided a clear understanding of the bandwidth utilization problem, guidelines for designing a new multiplexing method, and directions for future research. We plan to extend our study by including all the VoIP packet multiplexing methods over wired and wireless networks (e.g., 802.11 standard).