In the transmitter - based scheme, the transmitter assumes that a certain packet drop could occur in the end - to - end operation. The transmitter sends additional payloads or modifi cations that help the receiver (decoder/destination) to recover the lost packets. The popular options for these payloads are redun- dancy and FEC. Retransmission with the TCP - based method is possible for listening and for the broadcast mode of voice communication, but it is not used in interactive voice conversations.
5.3.1 Retransmission or the TCP - Based Method
Retransmission is a TCP - based method. In principle, retransmission - based methods work for any end - to - end media, signaling, or data packets. A lost
TRANSMITTER- AND RECEIVER-BASED TECHNIQUES 95 packet is identifi ed at the destination, and a request is initiated to the transmit- ter for retransmission of the lost packet. The longest delays of 300 to 400 ms
will happen with inter - regional voice calls. As per calculations given in
RFC0793, RFC2988 [Postel (1981) , Paxson and Allman (2000) ], inter - regional calls create a retransmission time - out of 1.0 to 1.6 seconds. Interactive audio applications have tight latency (delay) bounds. End - to - end delays need to be as low as possible and preferably in the upper limit range of 150 to 250 ms with the exception of intermediate satellite links and hard - to - reach areas. For this reason, interactive voice applications typically do not employ retransmission - based recovery for lost packets. Fax tolerates more end - to - end delay up to 3 seconds. Hence, the TCP - based method is possible for fax transmission, even though other techniques are most popular with fax. In practice, the TCP method is used for VoIP voice and fax signaling packets. Retransmission - based schemes work better when loss rates are relatively low, and they take lower bandwidth than other transmitter - based methods of FEC and redundancy. As loss rates increases, the overhead from retransmission and bandwidth require- ments increases. At higher packet loss, redundancy or FEC becomes more effective for bandwidth utilization.
5.3.2 FEC
Several FEC techniques have been developed to repair losses of data or media during transmission. These schemes rely on sending additional FEC packets in addition to regular Real - time Transport Protocol (RTP) voice packets. At the receiver, a lost packet is recovered from the regular packets and from the FEC packets. FEC techniques are broadly classifi ed as media - independent
FEC, and the media - dependent technique is usually known by the name
“ redundancy ” [Rosenberg and Scholzrinne (1999) ]. In practice, media inde- pendent is the default referenced name for FEC. Media - independent FEC is given in this section, and redundancy is explained in the next section.
FEC schemes were popularly used for bit error recovery. In digital com- munication, many media - independent FEC techniques use block or algebraic codes to produce additional packets in transmission to aid the correction of lost packets. In a generic notation, FEC takes k data packets (with each packet of several bytes) and generates n − k additional FEC check packets for the transmission of n packets over the network. Multiple FEC scheme options are given in RFC2733 that were considered for VoIP voice applications. FEC takes care of bursty or more packet losses based on the employed FEC scheme. In the process of an FEC operation, a certain amount of extra information is sent to the transmitted packets. The transmitter adds error correction payload, and the receiver tries to recover the missing data from extra packets or bytes. This operation adds extra overhead to the transmitted bytes and increases bandwidth.
To make use of FEC benefi ts, the receiver has to hold the received packets
intervals based on the selected FEC scheme. The complexity of FEC varies based on the selected implementation. For RTP payloads, parity - based algo- rithms are popularly used. These parity - based algorithms use the exclusive OR (XOR) operation. Most processors support a 16 - or 32 - bit XOR operation in their general arithmetic. Compared with other voice processing algorithms, FEC processing is insignifi cant because of XOR and packet loss detection logic operations.
Forward error correction is used in many digital communication systems. It is also useful in T.38 [ITU - T - T.38 (2005) ] based fax calls as given in Chapter 16 . A FEC technique also helps to encrypt the transmitted bits that help in sending secured VoIP voice and fax. Some FEC - based techniques are reported as patented for VoIP as a part of a secured VoIP communication.
The example used in RFC2733 [Rosenberg and Schulzrinne (1999) ] is taken here. For the fl ow of packets a, b, c, d, and so on, FEC packets f(a,b) and f(c,d) are generated. The operation f(a,b) is for bitwise XOR logic operation of total payloads a and b. The example given below creates two FEC packets for every four original packets, which makes a total of six - packets of transmission for every four packets. For a, b, c, and d, the packets generated by the sender are given here:
a b c d Original RTP media stream
f a b f c d Addition FEC st
⇒
( , ) ( , ) ⇒ rream
where time progresses to the right. In this example, the error correction scheme introduces 50% of overhead. In this example, if packet b is lost, a and f(a,b) can be used to recover b.
Scheme - 1 of RFC 2733. This scheme creates double the packets and caters to the bursts of two consecutive packet losses. The packets generated by the sender are given here:
a b c d e Original RTP media stream
f a b f b c f c d f d e Ad
⇒
( , ) ( , ) ( , ) ( , ) ⇒ dditional FEC stream
Scheme - 2 of RFC 2733. In this scheme, original packets are not sent. All packets are of FEC. This scheme allows for recovery of all single packet losses and some of the consecutive packet losses but with slightly less overhead than scheme - 1. The packets generated by the sender are given here:
f a b f a c f a b c f c d f c e f c d e only FEC stream no
, , , , , , , ,
( ) ( ) ( ) ( ) ( ) ( ) ⇒
original RTP stream
( )
Scheme - 3 of RFC 2733. This scheme requires the receiver to wait an addi- tional four - packet interval to recover the original media packets. However, it
TRANSMITTER- AND RECEIVER-BASED TECHNIQUES 97 can recover from one, two, or three consecutive packet losses. The packets generated by the sender are given here:
a b c d Original RTP media stream
f a b c f a c d f a b d Add
⇒
( , , ) ( , , ) ( , , ) ⇒ iitional FEC stream
The advantages of the FEC scheme are that its media - independent operation does not depend on the contents of the packet. Recovery is an exact replace- ment for the lost packet and takes relatively less computation. FEC is simple to implement compared with the voice or fax algorithms. The disadvantage of the FEC scheme is increased delay and bandwidth, but the increase in band- width in FEC is relatively less compared with redundancy.
5.3.3 Redundancy
In redundancy - based scheme, packets are sent with a concatenation of pay- loads with the procedures given in RFC2198 [Perkins et al. (1997) ]. Redun- dancy techniques are popularly used in fax pass - through and T.38 - based fax transmission. Fax signals are composed of phase and frequency modulations. In fax transmission, decoder - only based concealment techniques are not used because of the limitations in PLC implementations to recover phase and fre- quency modulations. In redundancy, primary payload is appended with a required number of previous payloads. This is a simple operation of concate- nating the required number of previous basic payloads. Based on the selected media and transport mechanisms, different packet headers are used. RTP - based voice and fax payloads follow implementations as per RFC2198. UDPTL - based fax payloads implement different headers as given in Chapter 16 .
Figure 5.2 shows the basic principle of redundancy. In this example, each frame is considered for a duration of 10 ms. In row 1 of the fi gure, frames marked as 10, 11, 12, 13, and 14, are transmitted for every 10 ms from the transmitter side to the IP network. In row 2, packets with redundancy = 1 are shown. Here the current 10 - ms frame and previous 10 - ms frame are grouped together to send as one packet. In row 3, the previous two frames and current frame are transmitted as one packet for redundancy = 2.
Redundancy = 1 takes care of one packet loss. As an example, if packet with frames of {13, 14} are lost, packets with frames {12, 13} and {14, 15} can make it to recover all the required frames of a lost packet with {13, 14} frames. If
packet drop exceeds more than one packet, redundancy = 2 is required.
Redundancy is advantageous for single packet drops or drops not exceeding the redundancy level.
Redundancy = 2 with two packet drops is shown in the fourth row. As an example, if packets with frames of {12, 13, 14} and {13, 14, 15} are lost, the previous and next packets are used to recover the required frames. To get the
benefi t of this redundancy, the receiver has to wait for extra packet delay. This delay should not vary on a packet basis. Once redundancy is selected, the buffering delay has to hold the required number of packets. In redundancy
= 2, the buffer has to hold two more packets to get the benefi t out of
redundancy.
Bandwidth Considerations for FEC and Redundancy. A redundancy scheme duplicates the payloads in successive packets based on the redundancy level. Redundancy increases the bandwidth in proportion (as approximation) to the redundancy level. To cater for redundancy = 1, bandwidth requirements increase by two times (at least payload). In FEC, one packet loss can be managed with about 50% more bandwidth. In practice, engineers show more favor to redundancy because of simple interpretation. However, FEC gives more benefi ts because of the availability of direct RTP packets along with FEC packets. The receiver that does not support FEC can simply continue the call with main non - FEC packets. FEC also scales better for security features of voice and fax packets.
5.3.4 Interleaving
Interleaving is used in higher throughput data networks. Interleaving disperses the small units of payloads. Hence, this technique is useful for distributing bursty erasures of packets or payloads into short - duration losses. In digital
Figure 5.2. Frames recovery using redundancy. [Payloads in a packet are marked as {10, 11, 12} for simple interpretation. Actual order is {12, 11, 10}.]
subscriber line (DSL), very high - speed DSL (VDSL), wireless, and passive optical networks interleaved techniques are used at the physical layer to protect data from burst errors in physical transmission. The interleaver works by segmenting the payload into units. Several units are combined to form one packet. The units are distributed in several packets. At the destination, it requires several packets to retrieve the units. A loss of a single packet from an interleaved stream results in multiple small gaps in the reconstructed stream, as opposed to the single large gap that would occur in a noninterleaved stream. As an example for voice, a 20 - ms payload of G.711 can be fragmented into four small payloads of 5 ms. The four small 5 - ms frame units can be dis- persed. If some packet is lost, the receiver will see only 5 ms of loss and not the whole 20 ms of consecutive data. A short - duration packet drop is easier to recover for PLC algorithms. In practice, the interleaver and FEC are used in combination. The interleaver disperses the burst errors, and FEC takes care of dispersed packet drops or errors.
The main advantage of interleaving is that it does not increase the band- width requirements when compared with redundancy and FEC. The inter- leaver complexity also changes in different applications. The main disadvantages of the interleaver are interoperability, increase in delay compared with FEC and redundancy, more processing, and extra memory to hold several small units of packets. For voice payloads, redundancy is most popularly used. FEC is used as the next option. The interleaver is not popular on direct voice and fax payloads.