Figure 2.1 A General View of Multiplexing
Introduction
Why use multiplexing?
Efficient use of transmission
links
Frequency Division
Multiplexing, FDM
Time Division Multiplexing,
TDM
Bandwidth Considerations
Switching
Port buffering
Switched Networks
Circuit Switched Networks
Packet Switched Data
Networks PSDN
Virtual Circuit (connection oriented PSDN)
Datagram (connectionless PSDN)
Appendix
The object of this section of the notes is to introduce some of the ideas relating to multiplexing and switching that will be fundamental to understanding the material presented in this course. It is likely that you have covered some of these ideas in other units but none-the-less it will prove instructive to review these principles before embarking upon this program of instruction.
Technology and materials science have increased the amount of information that can be carried over a single communications channel immensely since the days of the telephone network that once used twisted pairs throughout the entire network, dedicating one voice channel to each twisted pair. We now have fibre optic cables that are capable of carrying nearly over a million voice grade channels simultaneously. In order to utilise this greater capacity we need some methods to share this great capacity among the users of the network end-points to ensure that the lines are used as efficiently as possible. This sharing is known as multiplexing. By multiplexing our signals we can place many single independent signals on one communications channel. Thus many low data rate signals (low bandwidth)are able to share the high bandwidth of one single transmission channel.
There are three main types of multiplexing in general use.
To efficiently use a transmission medium when the charges are being
levied per unit time, it is fundamental to understand the technique of
multiplexing.
Figure 2.1 A General View of Multiplexing
The above link is able to carry n separate channels of data. The multiplexer (MUX) combines data from n input lines and transmits them over a higher capacity data link. The demultiplexer (DEMUX) accepts the multiplexed data stream, separates the data according to the channel, then delivers the data to the appropriate output lines. This allows two or more devices to share the same transmission line. If the devices in an organisation can efficiently share the communication link then this can drop the cost of the facility by allowing the resource to be spread over many users.
Figure 2.2 The Operation of Frequency Division Multiplexing
This technique is suitable for analogue signalling. FDM allows users to share the line in a parallel sense i.e. several signals are carried simultaneously. If the useful bandwidth of the transmission medium exceeds the required bandwidth of the signals to be transmitted then FDM may be used. The bandwidth of the transmission medium is divided into channels, which can be used simultaneously for data transmission. Each channel has an individual bandwidth centred about the carrier frequency and the channels are arranged so as not to interfere with each other, the channels being separated by guard bands. The signals are modulated onto different carrier frequencies. This technique is simple but inefficient in terms of bandwidth.
In fig 2-2, the incoming analogue signals A, B and C are each passed to an individual modulator which modulates the signal onto a carrier frequency that has been chosen to lie within the bandwidth of the transmission medium. The drawback of this method is that the FDM MUX only accepts analogue signals as its input so any digital signals would require conversion to analogue before multiplexing and this can be a source of distortion and noise. At the receiving end there is a FDM DEMUX which filters the three incoming signals from the transmission medium and separates them once more into the separate analogue signals A, B and C. Being analogue it is possible to amplify the signals but this will amplify any noise too.
Time Division Multiplexing, TDM
This technique is used for multiplexing digital signals onto a digital transmission line. Regardless of the nature of the signals upon entry to the transmission system (digital or analogue) they must be transmitted to their respective destinations in digital format. Analogue speech signals are sampled, quantised and encoded into digital form. It is not economic to send just one encoded voice channel at 64 kbit/sec over a single transmission channel so the method of multiplexing is to send the signals separated in time rather than frequency or phase, hence the name ~TDM.
Figure 2.3 The Operation of Time Division Multiplexing
In practice, TDM frames last 125 ?s giving a total of 8000 frames per
second. Each frame in the multiplexed stream has a framing bit or byte
(depending upon the implementation scheme) to mark the start of each frame.
This is to enable the receiving end to decode the multiplexed stream. The
number of slots per frame varies with location and data rate. There are
two major systems in use:
In both schemes, a byte is inserted per slot and each slot carries
one channel of 64 kbit/sec. For an American T1 channel capable of carrying
1.544 Mbit/second, one 125 ?s frame has 24 time slots plus one framing
bit. Each T1 time slot carries one 64 kbit/sec voice grade channel.
The European standard allows 32 time slots per 125 ?s frame using slots
0 and 16 for the framing byte and signalling byte respectively. This leaves
30 free slots for data channels. The total capacity of an E1 channel is
2.048 Mbit/sec although the useful capacity is 1.92 Mbit/sec.
Figure 2.4 North American Example of TDM on T1 Channel
Figure 2.5 European Example of TDM channel on E1 Channel
Over the years there have been developed several TDM hierarchies, the
most common of which are depicted below in table 2.1.
Table 2.1 Comparison between US and European
Digital Hierarchies
The hierarchies depicted above create a system where for instance six T3 channels can be multiplexed onto one T4 channel and four E4 channels can be multiplexed onto one E5 channel. If we wish to obtain a voice channel from an E5 channel it must first be taken down to E4 then E2 before we can recover the signal.
Provided that the transmitting and receiving multiplexers remain in synchronisation, it is a relatively simple job to recover the relevant input channels at the receiving end.
The one disadvantage of digital systems is the extra bandwidth that is required to carry the digital signal when compared to its analogue counterpart.
Example
A standard T1 channel transmits 24 voice channels designed to carry
4 kHz each. The digital transmission rate on the T1 channel is 1.544 Mbit/sec
which requires a bandwidth of approximately 772 kHz (Nyquist). It can be
seen that 24 x 4 kHz is only 96 kHz. Thus only 96 kHz would be required
to carry the 24 analogue channels. This is approximately 8 times less than
the bandwidth required to carry the 24 channels digitally (772 kHz / 96
= 8.04). The extra bandwidth is traded off against the considerably lower
signal to noise ratios.
One problem posed by TDM is the wasted bandwidth when the incoming
digital channels are carrying bursty data traffic rather than continuous
speech channels. This results in wasted slots (wasted bandwidth) on the
Tn (or En) channel. This can be seen in fig 2.6 below.
Figure 2.6 Wasted Bandwidth when deploying Time Division Multiplexing in bursty traffic
With TDM, each input channel is assigned a recurring time slot and if there is no signal present to be carried by the transmission line then empty slots will be carried on the line but this represents wasted bandwidth. Studies of the characteristics of today's digital traffic show the bursty nature of transmissions and thus if we are dealing with a large percentage of bursty digital traffic, we will be wasting our channel capacity. The solution here is to use Statistical Time Division Multiplexing known as STATMUX or SMUX. With STATMUX, the recurring slots are filled as they are needed. This means that more channels can be multiplexed onto the transmission channel than the sum of their theoretical maximum capacities. Fewer slots are wasted and this produces a better channel usage efficiency. There is a requirement however to provide each input channel with some form of identification so that it can be identified at the receiving end. This is one source of overhead in the scheme.
Figure 2.7 Statistical Multiplexing of
bursty traffic
Buffering (temporary storage space) will be required within the SMUX to cope with the bursty nature of the traffic arriving on the input channels.
[Asynchronous Multiplexing is an extension of statistical multiplexing where the tributary signals are not related in time. They may have differing data rates too. This is a key innovation and is used in ATM. Substantial buffering is needed in the multiplexer. This will be covered later in the course.]
Switches have both input and output ports and can be used to transfer
data from an input port through to an output port. Any input port can be
switched to any output port (provided it is free) and many connections
may be made simultaneously. There are several different types of switches,
the differences being within the switch fabric (construction method/ type).
Types include space division, time division and hybrid. A non-blocking
switch can connect all inputs simultaneously whereas a blocking switch
may not be able to connect all inputs as needed. When purchasing these
switches it is necessary to enquire whether the switch is blocking or non-blocking
and also what the capacity of the switch fabric is. This is measured in
Gbit/sec or Tbit/sec.
Figure 2.8 The Anatomy of a Layer 2 Switch
In fig 2.8, an incoming frame arrives at an input port where it is buffered. The frame must wait in the input buffer until the connection and control processor can switch the input through to the required output. Once it has been switched through to the output port, the frame is buffered until it can be released onto the output medium. In the case of CSMA/ CD the frame must wait until the segment is free before the frame can be transmitted.
Port buffering is required to provide non-blocking performance and also
to minimise the number of dropped frames or cells. There are two main types
here:
Figure 2.9 Comparison of Switch Buffering
Schemes
Until recently, LAN technologies have relied on the transmissions being frame broadcast. Thus station to station transmission communication is achieved by frames transmitted on one shared medium. This is fairly easy to implement but inefficient in terms of the overall communication rate for the network as a whole. The stations are capable of transmitting at a higher rate than the sharing technique (CSMA/ CD)for the medium will allow. Frame broadcast also does not scale beyond LAN size.
MANs and WANs have always relied on two types of switched transmissions, circuit switched and packet switched. These two types are important in ATM.
In a circuit switched network, a physical connection must be established
between the source and the destination. This type of connection provides
a guaranteed (fixed) latency for the connection. It provides a channel
that is dedicated via the network's switches and cables and is usually
full duplex.
Figure 2.10 A Circuit Switched Data
Network
A slot on each transmission channel is freed from end to end for the exclusive use of each user, providing the full duplex link. There will be a setup delay as the circuit is established from end to end, during which a complex signalling conversation takes place between the source, network nodes and the destination. This setup usually takes of the order of 10s to 1000s of milliseconds as the network tries to find a path from source to destination. If the network traffic is high then it may not be possible to route the circuit at all, thus the call is blocked.
Assuming that the state of the network is such that the call may be made then as soon as the connection has been made data may begin to flow. The network will perform no processing on the data whatsoever (c.f. datagram), the routing work having already been done. The type of data that is transmitted is immaterial to the network so it may be framed or continuous. The decreased error rate of modern networks means that the network need not perform error control either.
Older analogue networks displayed a high level of error with a constant background noise level that required error detection schemes to ensure the integrity of received data. Today's digital networks display random single-bit errors and occasional longer errors of a few milliseconds that occur about once per month. The error detection and correction functions are carried out in the receiving entities’ protocol stacks. After the transmission is finished, further signalling is required to release the circuit. This is known as teardown.
The delay experienced as an individual bit traverses the circuit switched network is very small and constant and this is a good method of transferring isochronous traffic.
Packet switched data networks require that data is sent as a series
of packets that conform to the network's protocol. If we use transmission
control protocol (TCP) which is connection oriented we can ensure that
the recipient is present before starting transmission of data.
The data we want to transmit is firstly created and stored within the
computer. The data will be segmented into appropriate sized chunks (segments)
that the PSDN can accept. Each segment is encapsulated into a packet by
adding a header and a trailer.
Figure 2.11 Segmentation and Transmission
of Data in a PSDN
At the destination, the reverse process takes place with the segments
firstly being buffered then the segments are presented to the user.
Figure 2.12 A Packet Switched Data
Network
The interface to the PSDN is via a packet assembler and disassembler and is often referred to as a PAD. In fig 2.12, each node has a buffer or FIFO (first in first out) stack to store packets whilst waiting for a chance to be transmitted along to the next node towards the destination. PSDN switches are more complex than circuit switches. They have internal buffering to store packets while waiting to be forwarded to the next hop. Protocol processing and error control (CRC-16) are often extra tasks to be performed by the switching nodes. Within the PSDN the packets are transmitted at full speed with independent data rates for each hop. The packets are allowed all of the bandwidth for some of the time.
Packet flow control and switch congestion are major issues for PSDNs. If the lines are congested and the buffers overflow then packets will be lost. The flow of information across such a network is not blocked but the delay experienced by an individual packet will vary with the traffic flow within the network. It is possible to place increased priorities upon each packet and this may be used with isochronous traffic to carry voice and video data.
Before packets are sent, the user requests connection to a source, i.e. a source-destination path. This is called a virtual circuit (VC). Each virtual circuit is assigned an identifier number. The VC routing information is stored at each switch in routing tables. Because of the size of the networks it is not possible to assign a fixed VC number across the network. This would require knowledge of every single virtual circuit that is currently in use at the moment the call is set up. Thus the VC numbers are local to each hop across the network and are unlikely to be using the same number on adjacent hops. There is a setup delay while the circuit is routed and the routing tables are updated. The source tags its packets with the assigned VC identifier number and the intermediate switches route the packets according to the routing table. All the packets follow the same route and arrive at the destination in sequence.
In a datagram network there is no circuit setup delay. Each packet is sent as soon as it is ready. Each packet is routed independently, the chosen routes dynamically varying according to traffic flows at various parts of the network. Each packet must be given a sequence number because packets that travel by a shorter route may overtake a previously transmitted packet that has had to take a longer route to the destination. Thus the order that packets arrive at the destination is not guaranteed, but the numbering ensures that when the packets have finally arrived they may be rearranged in their original sequence.
Nyquist. Harry Nyquist in 1933 made the discovery that if an analogue signal is sampled at such a rate that is equal to or greater than twice the highest frequency component of the analogue signal, the samples will contain all of the components necessary to firstly digitise the signal and then to recover the signal once more in analogue form without loss of signal integrity. It is known as the Nyquist criterion and is described by the equation:
fs >= 2 BW
where fs is the sampling frequency and BW is the bandwidth of the input signal.
For telephone speech signals (the basis of the original conversion from analogue to digital transmission systems) the voice channel bandwidth was set at 4 000 Hz because the majority of the speech energy is contained within a BW of 4 kHz. Using the Nyquist criterion it can be seen that the sampling rate is:
fs >= 2 x 4 kHz, giving a minimum sampling rate of 8000 samples per
second.