Effect of Variable-Length Frames TOP
Although an Ethernet LAN limits the information field length to 1500 bytes, a Token Ring LAN operating at 16 Mbps can have an information field that can be up to approximately 18 Kbytes in length. If a LAN station obtains access to the serial interface of the router or FRAD between frames transporting digitized speech, the delay will depend on the operating rate of the serial connection to the network and the length of the frame. Concerning the former, there are significant delay differences between accessing a frame relay network at 56/64 Kbps and a Tl operating rate of 1.544 Mbps. For example, at 64 Kbps, one 8-bit byte carried in the information field of a frame relay frame requires 125 x 10"6 seconds (8/64/000) for transmission. Thus, a 1500-byte information field in a frame relay frame that results from the transportation of data in the maximum-length Ethernet information field to the FRAD or router would require 0.1875 seconds (125 x 10~6 x 1500) to be transmitted to the network, without considering the overhead associated with control fields. This means that, excluding frame overhead and processing time associated with the FRAD or router forming a frame for placement onto the access line connected to the frame relay network, a lengthy Ethernet LAN packet could delay a frame transporting voice by almost 0.2 second.|
Frame length handling |
Frame prioritization |
|
CIR selection |
Frame loss handling |
|
Echo cancellation |
Frame delay handling |
|
Silence suppression |
Voice-compression method |
|
Frame overhead |
Telephony signaling |
|
Multiplexing technique |
Service-level agreements |
Considering End-to-End Delay TOP
It should be noted that the previously described delay represents only the delay associated with accessing the frame relay network. The actual end-to-end delay can be significantly higher, since slight delays are introduced at each node in the frame relay network, and additional delay may result if a frame transporting data arrives at a network node prior to a frame transporting digitized voice. The latter situation can easily occur if your organization uses a frame relay network to interconnect a number of LANs, with PVCs set up to provide an any-LAN-to-any-LAN communications capability. In this situation, there is a higher probability that a frame transporting data will arrive between a sequence r frames transporting digitized speech at a network node for delivery to a FRAD or router via an access line from the network to a subscriber's location. In fact, a station on a LAN at the originating site could be transmitting to a different destination than the destination of the frame carrying digitized voice, while a station on a LAN at a third location could be transmitting a sequence of frames carrying data to the destination location of the digitized-voice frame. If any frame carrying data arrives before the frame carrying digitized voice, then the delay in effect doubles. Thus, the use of 64-Kbps access lines could result in an end-to-end delay of 0.3750 seconds, a gap sufficient in duration to result in the distortion of reconstructed speech.Worst-Case Delay TOP
Although a delay of 0.375 seconds is significant, that delay assumes a worst-case situation when frames transporting digitized speech are delayed by frames transporting data in an information field 1500 bytes in length. Now let's assume that the LAN used is a Token Ring network operating at 16 Mbps. As previously discussed, this network can transport frames with an information field of up to 18/000 bytes in length. Since the maximum length of the information field of a frame relay frame is 8192 characters, this means that the FRAD or router would fragment the contents of a lengthy Token Ring information field that exceeds 8192 bytes into two or more frame relay frames, with the first frame always having a maximum-length information field of 8192 bytes. Again, returning to an access line operating rate of 64 Kbps, this means that the worst-case delay resulting from a frame transporting data will be 125 x 10~6 seconds/byte x 8192 bytes, or 1.024 seconds, clearly a most unsuitable situation. Once again, it becomes possible for a long frame transporting data from a different location to arrive at the destination FRAD slightly ahead of the frame transporting digitized voice. This action doubles the 1.024-second delay between frames transporting voice to 2.048 seconds, making a bad situation intolerable.Varying the Access Line Operating Rate TOP
Now let's assume our organization installs a Tl circuit as the local access line to a frame relay network operator. Although this transmission facility operates at 1.544 Mbps, the operating rate includes an 8-Kbps sequence of framing bits that cannot be used for the transmission of the contents of frame relay frames. This means that the data transmission capacity of the access line will be 1.536 Mbps, or 24 times the capacity of a 64-Kbps circuit. This also means that the latency or delay resulting from a frame transporting data being processed by a router or FRAD just prior to a frame transporting digitized voice will adversely affect the digitized voice frame by 1 /24th the time indicated by our prior computations for the same situation, resulting from the use of a 64-Kbps access line. Table 9-2 summarizes the one-way delays resulting from 1500- and 8192-byte frames carrying data being processed by a FRAD or router prior to a frame transporting digitized voice via access lines operating at data rates from 56 Kbps to 1.544 Mbps, including fractional Tl operating rates of 128 Kbps, 256 Kbps, 384 Kbps, 512 Kbps, and 768 Kbps.
FrameLength |
||
Access Line |
||
Operating Rate |
1500 Bytes |
8192 Bytes |
56 Kbps |
0.21428571 |
1.17028571 |
64 Kbps |
0.18750000 |
1.02400000 |
128 Kbps |
0.09375000 |
0.51200000 |
256 Kbps |
0.04687500 |
0.25600000 |
384 Kbps |
0.0312500 |
0.17066667 |
512 Kbps |
0.02343750 |
0.12800000 |
768 Kbps |
0.01562500 |
0.08533333 |
1.544 Mbps |
0.00781250 |
0.04266667 |
Data Frame Delay Times (in Seconds) as a Function of Frame Length and the Access Line Operating Rate |
In examining the entries in the table above your first instinct to solve the delay problem caused by a variable-length information field may be to significantly increase the operating rate of the access line to the frame relay network. Although this would undoubtedly reduce the delay time caused by frames that transport data arriving at a FRAD or router prior to frames that transport digitized voice/ it is also important to note that very rarely does a frame transporting data arrive at a FRAD or router as a single entity. That is/ if a workstation initiates a file transfer there is a high probability that a flow or sequence of LAN frames will arrive at the FRAD or router/ and in fact may fill its buffer memory unless there is a mechanism that subdivides buffer memory into independent queues and uses a priority mechanism to service data placed in each queue. Even then/ separate queues and a priority scheme may not be sufficient, because at certain access line operating rates/ a lengthy frame transporting data can adversely delay a frame transporting digitized voice. Thus, a better solution to this problem is both fragmentation and prioritization of data.
Fragmentation provides a mechanism to subdivide relatively long frames into a series of shorter, less-delay-creating frames. Through a fragmentation process, an even flow of minimized-length frames will be created, such that frames transporting digitized voice do not have to wait too long behind frames transporting data. This concept is illustrated below it shows how a long data frame is subdivided into a sequence 01 shorter-length frames by a FRAD or router. In this example, it was assumed that the FRAD or router was programmed to interleave frames transporting voice and data.
However,
in many operational situations, fragmentation by
itself or with simple interleaving is not sufficient. What is then needed is a frame-prioritization
scheme that works in conjunction with frame fragmentation.
One important factor
you must recognize is the effect of fragmentation. Although the fragmentation of
packets transporting voice is a necessity to minimize delays to packets
transporting digitized voice, the fragmentation
process increases network overhead. Simply put,
prior to implementing a voice-over-frame relay application it is extremely
important to determine the potential effect of fragmentation on your data
transportation.
Frame Prioritization TOP
Prioritization is a technique in which frames are processed based on predefined criteria. When a FRAD or router is servicing both data and digitized voice, it logically makes sense to prioritize delay-sensitive traffic, such as digitized voice, ahead of non-delay-sensitive traffic or less-delay-sensitive traffic.Memory Partitioning and Priority Queues TOP
Prioritization must be used with fragmentation to be effective. Many equipment vendors will partition the memory of their FRAD or router into priority queues. Frames containing data that exceed a predefined length are first fragmented. Then, all frames transporting data, including fragmented and nonfragmented frames, are placed into a low-priority queue. Similarly, frames transporting digitized speech that exceed a predefined length are also fragmented. However, unlike frames transporting data, frames carrying digitized voice are placed into a high-priority queue. The primary reason frames transporting digitized voice are fragmented is that the sampling source produces a lengthy frame, and fragmentation will produce a more regular flow of voice information. It also reduces the potential that the loss or delay of a packet as it flows through the network will adversely affect the reconstruction of the voice signal at its destination. The method used to prioritize traffic will obviously provide a preference to voice However, depending on the type of data presented to the FRAD or router, there may be certain types of frames transporting different types of data that also require prioritization For example, the transmission of SNA traffic can result in session timeouts if frames carrying such data are adversely delayed Thus, prioritization techniques must consider the type of data being transmitted as well as the fact that frames transporting digitized speech should receive a higher priority than most frames transporting dataCIR Selection TOP
We can obtain an appreciation for the role of the CIR upon voice transport by examining its relationship to the line access rate, the committed burst size (Be) and the excess burst size (Be) As a refresher. Be represents the amount of data m bits that a frame relay network agrees to transfer under normal network conditions during a predefined time interval (Tc.) Most frame relay network operators set Tc at 1 second, resulting m the following relationship
The above picture
illustrates the relationship between the previously mentioned frame relay
metrics. In examining the picture note that
only frames within the CIR are guaranteed
delivery. In comparison, when your transmission
rate exceeds the CIR your frames will either have their discard eligible (DE)
bit set or may actually be discarded when frames exceed the sum of the committed
burst size and the excess burst size. For example,
assume your organization has a contract with a frame relay service provider for
a CIR of 128 Kbps. However, you use a Tl
line access rate of 1.544 Mbps that permits bursts well above the CIR. As your FRAD
or router bursts above the 128 Kbps CIR each frame that exceeds the 128 Kbps
rate will have its DE bit set. Because there is no way to tell whether a DE bit
setting represents the discard region or the discard eligible region after a
frame flows through the first network switch, that
switch is the only switch that will first discard frames arriving at a rate
greater than the sum of Be +
Be prior to discarding frames in the discard eligible region. Thereafter,
any switch under congestion in the network will drop frames with their DE bit
set regardless of the frame rate into the network.
While an occasional
dropped frame will not adversely affect a voice conversation, it is important to
remember that unlike the transport of data where higher layers in the protocol
stack compensate for frame loss via retransmission this is not possible when
real-time voice is transported. This means that the transport of data where
subscribers commonly burst above the CIR knowing that higher layers in the
protocol stack compensate for frame loss can result in poor results when voice
is transported. To obtain reliable delivery sufficient for good voice quality,
you have to have a high enough CIR to cover voice usage.
Another way to achieve a similar result is to consider the service-level
agreement (SLA) guarantees with respect to all
frames and DE marked frames.
Frame Loss Handling TOP
When data is transmitted over a frame relay network, the loss of a frame due to congestion or error results in higher layers at the end points performing a retransmission of the lost frame. Although the retransmission process is commonly used to correct a previously lost frame, it introduces a delay that is not suitable for handling digitized speech. That is, if a frame transporting digitized speech is for some reason dropped by the network, the retransmission of the frame results in an arrival delay that distorts the reconstruction of the digitized voice signal. Recognizing this fact, equipment vendors handle frame loss with respect to the transportation of digitized speech in one of two ways. Some vendors simply generate a period of silence, while other vendors use the contents of a sequence of previously arrived frames to generate speech for the missing interval. This is accomplished by means of an interpolation technique based on the contents of the last or a few previously arrived frames. For both techniques, vendor equipment will not retransmit lost frames, as their arrival at the destination would result in a sufficient degree of delay that would render their use impractical. It is important to note that the second technique actually represents a combination of techniques. A short period of time can transpire prior to a frame either being marked as missing or delayed and an interpolation of the prior frame being used. That period or gap is commonly compensated for by the generation of noise however, because the gap is a very short period of time that is commonly unnoticeable to the human ear, some vendors may elect to use a period of silence. Thus, interpolation may be combined either with noise or silence.Echo Cancellation TOP
Echo is a phenomenon resulting from the connection of two-wire subscriber access lines to the four-wire infrastructure used to form the long-distance trunks that interconnect telephone company central offices to one another. The actual connection of two-wire to four-wire circuits is performed by a device referred to as a hybrid. When energy is transmitted across the hybrid, a portion is reflected back, and this reflection is called an echo. The echo can be an annoyance to the speaker, especially when the reflection of his or her voice is delayed sufficiently to become highly noticeable. There are actually two types of echoes in voice conversations: near-end and far-end. The near-end echo results from the reflection of energy at the hybrid in the caller's central serving office. The far-end echo is caused by the hybrid located in the central office serving the called party. From the perspective of degree of disturbance, the far-end echo travels a much longer distance and would therefore usually arrive at the talker's location well after he or she spoke, causing an echo similar to what you might encounter in a canyon. To suppress the effect of echoes, communications carriers use echo suppressors. An echo suppressor is an electronic device inserted into a four-wire circuit to function as a blocking mechanism with respect to reflected energy. Frame relay networks do not use echo-suppression equipment in their networks. This is because those networks were constructed to support the transfer of digital data. However, since the annoyance factor of an echo is a function of its delay, the transmission of digitized speech over a frame relay network can result in disturbing echoes when two sites separated by a sufficient distance are interconnected via a frame relay network. Since the network operator does not use echo-suppression equipment, this becomes the responsibility of equipment vendors. Although most equipment vendors who market products for the voice-over-frame relay market include an echo-suppression or echo-cancellation capability, not all do. However, prior to using this feature as an evaluation discriminator, it is important to note that only when the average round-trip propagation time exceeds 25 to 50 ms is the use of echo suppression recommended. Otherwise, the use of a frame relay network to connect a few locations in close proximity to one another may not be necessary.As frames are routed through a network, they will encounter a variety of nonuniform processing delays. These delays, which depend on such factors as the activity level of a switch, the length of a frame processed ahead of another frame, and the processing power of the switch, result in a uniform data stream by the time it is transmitted into a frame relay network being received at its destination with random delays between frames. This is illustrated in the picture below, which shows how precise time intervals between a sequence of frames presented to a frame relay network could be altered during their flow through the network, resulting in random delays between received frames.
Jitter and Jitter Compensation TOP
The delay between received frames is commonly referred to as jitter and can result in awkward-sounding regenerated speech. Recognizing this problem, many voice-over-frame relay equipment vendors incorporate a buffer area in their products. Buffering the frames that transport digitized speech removes the variable delays, in effect facilitating the process by which voice is reconstructed to prevent annoying periods of random delays. The buffer area used to compensate for jitter is referred to as a jitter buffer. Most jitter buffers can be selected to a value between 0 (no buffering) to 255 ms of hold time or delay. The setting of the jitter buffer governs the amount of hold time. That is, a 50-ms jitter buffer setting means up to 50 ms of speech can be held and extracted according to the RTF time-stamp to provide a smoothed output. It is important to note that the jitter buffer hold time adds to the overall end-to-end delay. Thus, under most situations you more than likely will set the jitter buffer to a relatively low value. For example, if the end-to-end delay is 120 ms on average, with a low of 90 ms and a high of 130 ms/ the variability is 40 ms. In this example you would set the jitter buffer to 40 ms or perhaps even beyond since you would still be below a total of 250 ms/ which represents the point where a voice conversation begins to resemble a half-duplex transmission.Silence Suppression TOP
Although many frame relay network providers do not bill for the number of bytes transmitted, from a practical standpoint it makes no sense to transmit digitized voice samples that contain only a period of silence. Frames containing a period of silence add to network traffic and can adversely affect the transmission and delivery of other frames carrying digitized voice samples from the same or different callers. Recognizing this problem, some equipment vendors include a silence-suppression capability in their equipment. The effective implementation of silence suppression requires the use of a time tag for each frame transporting a digitized speech sample. This time tag enables the receiver to reconstruct a person's voice, including gaps and pauses between words and sentences, which results in the ability to maintain the natural quality of a person's speech. Unfortunately, the methods used to suppress periods of silence currently vary between some vendor products. However, the adoption of the Frame Relay Forum Voice over Frame Relay Implementation AgreementVoice-Compression Method TOP
Although many technical issues can affect the quality of reconstructed voice, the most important long-term factor is the method of voice compression used. As previously discussed in Chapter 5, voice-compression methods can be generally grouped into three distinct categories: waveform coding, vocoding, and hybrid coding, with the latter representing a combination of the first two categories. Although such waveform-encoding methods as PCM and ADPCM result in very high quality reconstructed speech, they also require a large amount of bandwidth to transmit a digitized conversation, with PCM consuming 64 Kbps and ADPCM requiring 32 Kbps to provide a toll-quality conversation. Waveform coding provides such a high level of reconstructed voice that it serves as a benchmark for users to compare the quality of other compression methods, but it does not maximize the utilization of bandwidth. In fact, on certain access lines, such as a 64-Kbps connection to a frame relay network, the use of PCM or ADPCM would either preclude the use of the connection for data or consume all or half of the bandwidth of the access line—neither of which is a pleasant situation. Thus, a more practical employment of voice on a frame relay network requires the use of low-bit-rate compression algorithms, either resulting from vocoding or hybrid coding techniques.Low-Bit-Rate Coding TOP
Currently, equipment vendors offer products that support a variety of voice-compression algorithms, with many vendors using algorithms in the Code Excited Linear Prediction (CELP) family, whose employment can result in voice digitization rates from a high of approximately 16 Kbps down to 2.4 Kbps. Table 9-3 lists three examples of low-bit-rate standardized voice-compression algorithms commonly supported by frame relay equipment vendors.|
ITU Standard |
Compression Method |
|
|
G.723.1 |
ACELP |
5.3/6.3 Kbps |
|
G.728 |
LD-CELP |
16 Kbps |
|
G.729 |
CS-ACELP |
8 Kbps |
|
Low-Bit-Rate Standards-Based Voice-Compression Methods |
Frame Overhead TOP
One of the most overlooked facts associated with the transmission of voice over a data network is the overhead associated with transporting voice. While many publications indicate that an active conversation occurring over a frame relay network may only require 4 Kbps, that is only part of the story. To obtain an appreciation for the full story let's begin at the beginning and examine how many trade publications compute the 4 Kbps bandwidth consumption and, unfortunately, stop there. If we assume a speech coder operating at 8 Kbps is used, we would start our bandwidth consumption computation at 8 Kbps. Next, we must consider the overhead associated with the header and trailer of the frame relay and RTF timing and sequencing information with respect to the actual data transported. Because small intervals of speech are transported in a frame the actual overhead can be approximately 25 percent or 2 Kbps. Thus/ the gross frame relay bandwidth becomes 8 Kbps + 2 Kbps, or 10 Kbps. If we assume silence suppression is supported/ approximately 60 percent of a conversation's period consists of silence. This is because most conversations are half-duplex, which results in a 50 percent period of silence. After adding time for thinking, pausing for air, and perhaps scratching behind an ear, we can easily add an additional 10 percent period of silence. Thus/ if we subtract 60 percent from 10 Kbps, we obtain a net bandwidth of 4 Kbps. Table 9-4 summarizes the previously discussed computations. While the previously described computations are true, they only tell part of the story. If your organization is transmitting data frames typically longer than 64 bytes they will be fragmented. To illustrate the effect this has upon the efficiency of transmitting data, let's assume a frame transporting 1500 bytes requires fragmentation. Without fragmentation and assuming a 2-byte control field the overhead is 6 bytes to transport 1500 bytes of information. If fragmentation becomes necessary due to the need to minimize delay the 1500 bytes will be transported through the use of 24 frames. The first 23 frames would each transport 64 data bytes, while the 24th frame would transport 28 data bytes. Because 24 frames are now required overhead increases to 6 bytes/frame x 24 frames, or 144 bytes. Thus overhead increased from 6/1500 or .04 percent to 9.6 percent for this example. This is the other part of the story most publications ignore and which in reality, needs careful consideration.Function |
Bandwidth |
8 Kbps Codec |
8 Kbps |
Frame overhead |
2 Kbps |
Gross bandwidth |
10 Kbps |
Less 60% silence suppression -6 Kbps |
-6 Kbps |
Net bandwidth |
4 Kbps |
|
Computing Voice over Frame Relay Net Bandwidth Requirements |
Telephony Signaling TOP
The effective transmission of voice over a frame relay network requires the ability to alert the destination to an incoming call and to inform the originator of the progress of the call. Such activities are associated with telephony signaling, and a FRAD or voice-compliant router must be able to transfer appropriate telephony signaling. In doing so, it is important to note that the meaning of a telephone signal can be represented by a transition as well as by time between pulses and the sequence of pulses. This makes the appropriate coding and transfer of telephony signaling much more challenging than the simple passage of data signals that are represented by a transition from a high to low pulse or a low to high pulse. Since PBXs may use different signaling methods, it is important to ensure compatibility between the signaling method supported by the FRAD or router and the signaling method used by the PBX. In their quest to enhance the performance of voice over frame relay and to develop a mechanism that can differentiate their products from those of others, vendors offering voice-capable FRADs have incorporated different bandwidth-optimization methods into their products. Two such techniques that are incompatible with one another are logical link multiplexing and subchannel multiplexing. Logical link multiplexing (LLM) enables frames transporting voice and data to share the same PVC. This technique is suitable for the situation in which your organization uses a voice server on a LAN to digitize voice. It requires both digitized voice-encoded frames as well as frames transporting LAN data to be carried via a common frame relay connection to the LAN. By using logical link multiplexing, you can more than likely reduce your organization's cost to use a frame relay network, since most network operators include a cost component based on the number of PVCs used. A second multiplexing technique used by some FRAD manufacturers is subchannel multiplexing (SM). Under subchannel multiplexing, portions of multiple voice conversations are combined within one frame. By transmitting samples of several voice conversations within one frame the overhead of the frame in comparison to its payload is reduced. This can be an especially important consideration when transmitting multiple digitized voice conversations over a low-speed frame relay access circuit, such as a 56- or 64-Kbps access line. TOP