2.1  Introduction

The previous chapter presented the fundamental principles required for understanding the digitization and transfer of audio within the context of this book. This chapter presents some of the background theory that is necessary in order to provide the framework for understanding different kinds of communication interface, and so support the choice of technology to be applied to a particular circumstance.

The main subject areas covered in this chapter are the 4-layer and 7-layer reference models and quality of service (QoS). The reference models are useful tools in understanding network technologies and the issues that they present. Both models are taught to network professionals today, and have a bearing on how network technologies have evolved. Quality of service can be interpreted as a buzz-phrase around commercial organizations and so should be treated with caution; it is a difficult phrase to define, as it can mean different things to different people, depending on the circumstance. The section on QoS puts the term in the context of computer networks, where it is a useful tool, since it puts a name to the service expected of the network, which would otherwise be difficult to describe.

The main advantage of understanding these concepts is to clarify thought, and present information that has been ratified by academic thinking and business markets. By presenting this information, it is hoped that the reader need not spend too much time conceptualizing the problems or defining the requirements of a particular subject area.

These concepts may be thought of in a similar way to project management tools. For every project manager, there is a right and wrong way to manage a project; however, it is arguable that there is no right or wrong way to manage a project, there is only the success or failure of that project to consider. There are several practices that have been proven over time to reduce the likelihood of failure, and the popularity and success of these practices means that project managers do not have to spend too much time worrying about how to best approach the organization of a complex task. Arguably, common sense, background knowledge, and experience are all important, but for the newcomer it is quicker and more useful to cross reference against practices that have already been developed and used successfully.

2.2  Bandwidth and Information Availability

The term bandwidth is used by electrical engineers to refer to the frequency range of an analogue signal. In the network industry, the term is sometimes used to refer to the capacity or the data rate of a link. Note that because of sampling, quantization, and compression, the bit rate needed for a given bandwidth analogue signal is potentially many times less than might be implied.

The principal endeavour of the computer networking industry is the availability of bandwidth, but no matter how high the figures achieved, there is never enough bandwidth. Amdahl’s law states: one I/O bit per instruction, which roughly translates into the increasing data transport rate required as CPU capacity increases. According to the statement, the amount of data that needs to be transported by an interface is directly proportional to the rate at which the CPU can perform instructions. The amount of information created by any multimedia application (such as those which generate or manipulate audio, video or use some kind of feedback and interaction) is extremely large and sensitive to QoS, as seen in Chapter 1.

Most studios using digital audio are forced to use a combination of open-standard and proprietary techniques in order to transfer data from its source to the recording media, since one vendor may not have products that fit all the steps in the audio production chain. In such cases, interconnections, media filters, and translation devices of various types ensure translation and synchronization between the various devices in use. This method of transfer is difficult to manage and can be extremely complex, especially over long distances when all the component elements are considered.

Audio is intrinsically related to time since its perception by the ear occupies a particular time span. Digital audio data cannot usefully be heard without the correct decoding mechanism, and even then an accurate reproduction would only result if the same understanding of time is used during the playback and recording processes. Otherwise, the playback will be altered when compared to the original recording.

When audio is played back (or streamed) over networks, the realtime nature of audio becomes a problem, particularly for applications written to support real-time audio streaming over the Internet, such as Internet radio and telephony.

2.3  Data Interchange in Computer Networks

The most appropriate way to present the issues surrounding the transfer of digital audio data over networks is to present the problems of different classes of data as they apply to professional audio use. Armed with this knowledge, it is then appropriate to discuss the capabilities of various technologies available to the studio. Most of the technologies have specific limitations in terms of the amount of data that a studio would like to be able to transfer, but others have limitations because of the role they are designed to fulfil. A more informed understanding reveals that many of the technologies have common features and mechanisms, and can be made very useful to the multimedia industry.

2.3.1  Characterization

Within information technology, a network is defined as a series of points (nodes or stations) interconnected by communication paths (www.whatis.com, 1999). Networks can be connected together to form internetworks and can also contain subnetworks.

Using such a broad definition, most normal audio facilities have several different networks already in use, although they may not be thought of as such. For example, connecting several devices together for control via MIDI creates a network, as does connecting devices together to exchange digital audio information or synchronization data.

Data communications networks can be classified in different ways. For example, a network can be characterized by its geography or by its topology. The most common topologies are bus, ring, and star configurations. MIDI is an example of a bus topology, which is formed by a single, open ended network length, whose ends do not meet. A ring configuration is formed naturally enough, by connecting the two ends of the cable together and unless the technology is designed to work under both conditions, a bus should not be turned into a ring and a ring should not be turned into a bus.

For instance, if the two ends of a MIDI network were connected together, MIDI messages might traverse the whole ring and start again, leading to all sorts of problems with timing errors and repeated messages.

Figure 2.1 shows the classic diagrammatic representation of bus, ring and star topologies that will be adopted throughout this book. Star topologies are more commonly used since the advent of hubs, which turn bus and ring topologies into a star. This is explained in more detail later.

Networks can also be characterized in terms of spatial distance (geography) as LANs, MANs and WANs, as we have seen. In audio terms, a MIDI network is usually contained within one or two rooms, and would be considered a local area network. LANs can get quite large, containing hundreds of stations, and spanning an entire campus of several buildings, or can be quite small, limited to two or more devices within a department or studio.

Networks can also be characterized by the transmission protocol being used to carry the data and this might be likened to determining whether a transport network is designed to carry cars and trucks, or railway carriages. Physical network cabling carries both types of traffic at the same time. The two different traffic types cannot see each other, although the flow of one may effect the other. This coexistence of protocols on the same cable is known as creating a logical network. A number of techniques are available for combining logical networks in this way, such as surrounding one type of data with the protocols for another, in the same way that cars may be transported on a rail network by placing them onto a carriage designed for the purpose.

Other characterizations or classifications include synchronous or asynchronous, the nature of the connections and the types of physical cable in use.

2.3.2  Packet Switching Networks

Packet switching networks proliferate a type of communication capable of dealing with multiple destinations, multiple devices attached to the same cable, and unanticipated connections. Although resilient to change, a delay is incurred because the amount of data being transmitted cannot be anticipated in advance. Packet switching networks are more efficient if some amount of delay is acceptable and for multiple distributed connections.

History

The US Department of Defense (DoD) initiated the early work on network technology in the early 1960s. The project to investigate the ability to maintain command and control over missiles and bombers after a nuclear attack was named as the Advanced Research Projects Agency NETwork (ARPANET) and focused on the distribution of data (Kristula, 1997). The research produced a document that described several ways to accomplish this and the final proposal was a packet switched network, described thus:

The Internet

The ARPANET project involved both military and academic institutions and eventually the network developed into the public domain and became the Internet.

Information on the Internet architecture and related protocols are published in requests for comments (RFCs) that are available electronically via the Internet or in hard copy from the Defense Data Network DN Network Information Center (Room EJ291, SRI International, 333 Ravenswood Avenue, Menlo Park, CA 94025, USA).

Character

Packet switching networks are known as connectionless, since data travels with address information attached directly to it, in order to find its way through the network. This is explained more graphically in Figure 2.2. Furthermore, packet switching networks are often multiple access cables, with more than one device attached to the same cable.

2.3.3  Circuit Switching Networks

A circuit switched network is a type of communication where a dedicated channel (or circuit) is established for the duration of a transmission. The most ubiquitous circuit switching network is the telephone infrastructure, which links together wire segments to create a single unbroken line for each telephone call.

History

The history of circuit switching networks belongs within the telecommunications industry. Circuit switching networks developed to speed up the time taken to connect calls and are ideal for communications that require signals to be transmitted in real time, such as voice conversations. The popular picture of rows of girls connecting calls is an original circuit switched network.

With the advent of silicon switching, circuits can be made and broken far more quickly. The signalling bandwidth of the complete circuit can be used for the transmission of digital data. More detailed discussion on appropriate technologies can be found in later chapters.

The infrastructure was developed to transmit data over longer distances, and the appropriate acronym POTS stands for the Plain Old Telephone System. Appropriate progress has seen circuit switching technologies develop into a more ubiquitous technology, available for use more generally.

Character

Circuit switching networks are sometimes called connection-oriented networks, meaning that a connection appears to havebeen made directly between the two devices at each end of the circuit (Figure 2.3) and should not be confused with connection orientation mentioned further up the 7-layer model. Packet switching networks, whilst being connectionless, are made to be connection oriented by using a higher-level protocol. Transmission Control Protocol, for example, makes Internet Protocol networks connection oriented.

2.3.4  Point to Point Connections

Point to point connections might not present themselves comfortably as a network type, but the use is generally to connect devices together for the purposes of communication or transfer of data.

A point to point connection would normally be associated with two devices, one connected to another, such as for copying a digital audio tape from one machine connected directly to another, as shown in Figure 2.4. The connection is enclosed, and no other devices can see or address the network.

Although most common audio standards utilize a point to point topology, connections can be more complex over other physical network types by creating a logical connection between devices. Transmission Control Protocol for instance makes connectionless networks connection oriented, whereas Fibre Channel takes a different approach to multiple access.

2.4 US Department of Defense 4-Layer  Model

Among the many significant publications from the ARPANET project was the Department of Defense 4-layer architecture (4-layer model), illustrated in Figure 2.5. It proved difficult to design protocols adhering to this model and it was several years before the transport control protocol/Internet protocol (TCP/IP) was finalized and deployed.

In the meantime, the International Standards Organization worked on the 7-layer model, which identifies the steps that need to be addressed for general communication purposes.

Although this book uses the 7-layer model as a reference, the 4-layer model is also important and is explained briefly under the following four headings.

2.4.1  Network Access Layer

The network access layer is responsible for delivering data over the physical cable and can be considered to be attached to the actual physical medium in use, such as copper or fibre optic cable.

The network access layer defines the resistance, voltages, and access mechanisms in use. Definitions exist for the restrictions in cable lengths, appropriate cable types, number of connections and so on. This layer defines the restrictions in terms of the network or single cable segment and describes how differences in voltage (or light intensity for fibre optics etc.) are turned into recognizable data.

2.4.2  The Internet Layer

The Internet layer is responsible for managing the delivery of data across a series of different physical networks that might interconnect a source and destination machine. Routing protocols are associated with this layer, such as the Internet Protocol or IP, the Internet’s fundamental protocol. This is the most significant layer when considering connecting networks together to create an internetwork, since it describes how data are sent over longer and more complex routes than can be defined by the physical layer.

2.4.3  The Host to Host Layer

The host to host layer deals with problems once the data have made their way from one machine to another, such as handshaking between computers. This is a common technique even in audio machines whereby connections are negotiated in much the same way as DAT machines, modems or fax machines might do with frequency or bit rate.

In more detail, handshaking is the exchange of information between two devices in order to establish the protocol to use for communication (Figure 2.6). Since the devices at each end of the line may have different capabilities, a handshaking procedure is used in order to determine the highest transmission speed that both can use. In DAT machines, handshaking is used in order to determine the sample frequency and word length of the incoming signal for instance, much like establishing the correct shaped key in order to open the door of further communication.

2.4.4  The Process Layer

The process layer contains protocols that implement user-level functions, such as mail delivery (SMTP – simple mail transfer protocol), file transfer (FTP) and remote login (TelNet). Each of these provides a different type of service over the network.

Further details of the 4-layer model are incorporated in descriptions of the 7-layer model in the next section.

2.5  The ISO 7-Layer Open Systems Interconnection Reference Model

Taking the lead from the 4-layer model, the International Standards Organization (ISO) undertook a redefinition of the model in order to address what was seen as the shortfalls of the original. The result, which never became a ratified standard, was the ISO OSI reference model (see Figure 2.7), which is taught to network professionals today and known hereafter as the 7-layer model. Protocols adhering strictly to the OSI model are notoriously difficult to understand and are indeed bloated and inefficient (Entry for OSI Seven Layer Model, 1999), but its most significant contribution is the underlying philosophy of networking as represented by its layered model.

The model describes, in seven layers, all of the problems that need to be solved in order to allow multiple computers to talk to each other on a network.

It is useful to understand the model and what it presents, since the descriptions of the processes contained within the seven layers describe all the steps necessary to place data on a cable for receipt by a known recipient. The modularity of the 7-layer model allows it to be a useful tool wherever it is necessary to place data on a cable (or other transport medium). It is useful to think of digital audio transfer in these terms, as the exercise illustrates what steps have been implemented and therefore what functionality is missing from the candidate solutions and architectures.

Although the 7-layer model is considered to be an improvement over the older DoD 4-layer model, one of the layers was subsequently split into two further sub-layers.

The 7-layer model’s basic functions match those of the 4-layer model, although the work of the ISO/OSI committee described the functions in more detail, leading to the increase in the numbers of layers. It is easy to get bogged down in the theory behind the 7-layer model, and it is worth taking a step back to look at the work as a whole. By doing so, it is easier to appreciate that it describes every stage in the translation of electrical signals into comprehensive and error-free data communications between digital devices. As with the 4-layer model, the OSI model is designed to be modular. That is, any mechanism or set of rules designed to fulfil the requirements of a layer can be interchanged with any suitable mechanism for that particular layer, without affecting the functional mechanisms applied to the adjacent layers. For instance, two mechanisms that fulfil the requirements of the physical and data link layers are token passing and CSMA/CD (or Ethernet and its cousins). Each has very different rules of engagement, and may even utilize different cable types or electrical parameters. However, this should not effect the ability to run upper layer protocols, such as TCP/IP upon the physical medium. This would give rise to common descriptions such as ‘TCP/IP over Ethernet’ which describes several functional layers of the model.

As with the 4-layer model, it remains difficult to implement the 7-layer model with exact partitions at the layer boundaries, and a good deal of energy is expended by the network industry in trying to get new network technologies to talk to established ones. The boundaries between the layers would be better illustrated with dotted adjoining lines, since functionality often affects adjacent layers.

The table shown in Figure 2.8 shows the functions and methods associated with each layer. To understand the model, imagine that the network cable is attached to the bottom of the physical layer, as suggested in Figure 2.7.

2.5.1  The Physical Layer

The physical layer provides the services associated with physically connecting to the cable and creating the link. These include the encoding of data into voltages, or whatever binary states the medium requires (such as frequency in fibre optic cable, microwave, and infrared linkages) and the encoding of the data using mechanisms such as non-return to zero. This layer is responsible for physically transferring messages between nodes. The physical connection may take several forms, and the path of a message through an internetwork may use several different types of physical layer mechanism before reaching its destination.

In simple terms, the physical layer can be thought of as attaching to the cable (or other medium) and being responsible for identifying data upon it. The physical layer is also responsible for configuring the link. This means that various other parameters are defined here, such as the maximum number of stations on a segment, the voltage and impedance used on the medium, the maximum length of cable and so on.

The physical layer provides the data link layer with a method of moving bits between two machines. It is recognized as error prone and unreliable because of the conditions, such as geographical distance and environmental changes including radio frequency interference, through which the physical layer carries information. As a result, other layers ensure data integrity by using error checking and correction mechanisms.

2.5.2  The Data Link Layer

The data link layer provides the error checking functionality that makes the unreliable physical layer reliable. In addition, the data link layer also provides flow control such that messages of a given size may be reliably transmitted.

This layer has subsequently been split into two sub-layers. The sub-layers are called the media access sub-layer (MAC) and the logical link control sub-layer (LLC). The MAC sub-layer is responsible for those functions that are associated with the physical layer directly below it, such as the error control that is required in order to make the physical layer reliable. This will normally include some mechanism whereby a packet or frame, as it is called at this layer, can be resent if it is lost or damaged, as well as determining whether the cable is available to accept transmission.

The LLC layer on the other hand concentrates on taking data from the layers above and marshalling them downwards, and so has more to do with controlling the flow of the data, and splitting it into correctly sized messages (packets, frames or datagrams) for transmission.

2.5.3  The Network Layer

The network layer is an important layer for connecting networks together to form internetworks, since it is responsible for routing messages between nodes. When a message reaches an intermediate node in the network, the network layer implementation at that node will be able to determine which route that message should take on the next stage of its journey to the destination device. This may entail determining which link, of several connected to the node, should carry the message onwards. In order to operate efficiently, the network layer uses its own information gathering systems called routing protocols, and several of these are in use, such as those covered in the next chapter.

Although network layer protocols tend to exist between nodes, rather than at the sending and receiving stations as shown in Figure 2.9, they do effect the original data, by altering addressing information, which allows other nodes to determine the destination of the data. In this way, devices operating specifically at the network layer are able to determine the correct route or path that the packet needs to take in order to find its destination.

Each device in the network layer may strip the original addressing information from the frame and add its addressing information, or it may append more addressing information to the frame, making it longer. The disadvantage of appending addresses is obvious – frames get longer. Some devices operating at the network layer may have a maximum frame size, and so frames that are too long may get split into the correct size until eventually, the original data form only a very small part of the information that arrives at the destination. It is very important to retain this information if the returned data are to navigate their way back to the sender. On the return journey, each network layer node strips the new information away until the data are in the intended form at the time it reaches the original sender.

Network layer devices have their own sets of protocols and rules, which are used to communicate the status of particular links to one another in order that the fastest routes for data to travel may be discovered before sending the data onward.

Taken together, the first three layers are referred to as the subnet layers. To provide full connectivity, all subnet layers must be implemented in each node of the network

2.5.4  The Transport Layer

Layers 4 through 7 have less to do with the mechanisms for moving information around, since at this point the data have been created and moved from one node to another. These layers are collectively called the end-to-end layers because their services are required only in the end nodes and not in the intermediate nodes. The transport layer manages end-node to end-node communication, delivering the units of data of whatever size, from one device across the network to the receiving device. While the data link layer ensures that a message will not be damaged, the subnet layers may not necessarily guarantee that all messages will be delivered, or if they are delivered, in what order. The transport layer is therefore called upon to handle flow control, retransmission and message sequencing.

The transport layer provides long-range resilience to the data. Since data can arrive in the wrong order, the transport layer will reorder the data before passing to the upper layers.

Examples of this mechanism appear in transport control protocol (TCP), where each packet of data is numbered so that the transport layer can restore the sequence of packets at the receiving node.

The transport layer also inspects the number of each packet, to ensure that each one has turned up. In the event of a packet being lost, the transport layer requests that the packet is resent.

2.5.5  The Session Layer

The session layer is responsible for establishing, managing, synchronizing and terminating network sessions between devices. This allows the host computer to assign resources to conversations between itself and the sending station.

The session layer mechanisms take the form of handshaking, and negotiating the format and rules for the conversation to take place. Once this has been set up, there is little activity at this layer until the conversation is disconnected, whereupon the session layer will handle the termination of the session in a controlled fashion, so as to ensure that the resources become available for further conversations.

This layer is especially important in the Internet, where popular Internet sites may receive millions of requests per hour, all needing to be sorted and allocated resources in the most efficient way.

Each frame of data contains the resource allocation identifier, so that the incoming data can be immediately identified and routed to the correct resource within the host device.

2.5.6  The Presentation Layer

The presentation layer takes care of translating data to and from standardized formats, as may be negotiated by the users. This may include character set translation, or any of several forms of data conversion, for example between two sets of floating point numbers. While layer 6 transformations often serve to present data to the receiving application in a convenient and mutually understandable format, they are also carried out for purposes directly related to the communication, such as data encryption, compression and so on.

2.5.7  The Application Layer

By the time the data reach the application layer, they are now in the form that the receiving program can understand, and have been stripped of all the data that may have been added to it for the purposes of the network layer (and all the other layers). Furthermore, any compression or encryption has been removed and the data are available for direct processing.

The application layer forms an interface to the user program and is often understood to encompass an application program that performs a network specific user service, such as file transfer or mail delivery.

Digital audio data could be thought of as sitting within this layer, since it only needs to be decoded by the D/A conversion process to be understood. More strictly, the D/A process itself would be one of the services offered by the application layer.

2.6  Quality of Service

It is the intention of this book to offer best-practice information on the implementation of networks for audio purposes, only where appropriate. This information is referenced where necessary but conflicting information can often be found. Best practice is too often dependent upon the restrictions that are imposed upon any particular project. For instance, it may be desirable to install the best possible network technology within an establishment, but this is impossible because of the budget assigned to the project. On the other hand, it may be desirable to install the cheapest or most commonly used network technology, but this is useless for the purpose for which the network is intended.

The latest advances in communication technologies have made it possible to develop new services, particularly multimedia services such as audio and video retrieval and videoconferencing. The wide deployment that the Internet has recently experienced has raised interest in providing such services over networks in general. However, the main problem that these services present is that their requirements are completely different from those of traditional data communication. To transmit audio or video flows it is essential to preserve their time dependencies, but most current networks do not guarantee the data delivery within any given amount of time.

2.6.1  What is QoS?

A great number of parameters have to be understood by both the vendor and consumer before the correct network design can be realized. Other than cost, installation times, network management facilities, failure recovery, and traffic throughput, there is another set of parameters that make up the quality of service (QoS) that is demanded from the network. As mentioned previously, different types of data have different delivery requirements, and these help to determine the QoS.

The term ‘quality of service’ is more accurately expanded to, ‘the quality with which a service is offered’, and can be applied to any service, from car washing to network provision. To understand the term in its appropriate spirit, the definition found in Webster’s Dictionar. under the entry for quality assurance reads in part:

The service in question is the ability of a network to transfer data to agreed requirements. A suitable network will be one whose QoS is at least as good as those required by the data type.

2.6.2  Defining the Parameters of QoS in Digital Audio Transfer

Defining the quality of service has always been an important part of specifying a network, since it would be bad practice to install a network that struggled with e-mail, when it was installed to deliver real-time audio, although the requirement was not always associated with the phrase. There is no technology that can deliver unlimited network bandwidth, on demand for all the possible types of service. Since this is the case, it is necessary to understand what sort of network will fulfil the requirements of a particular environment. A full analysis of the requirements will lead the network designer to choose the correct technology, and should have a significant impact on the eventual design of the network. It is the job of the network designer to get enough information about the requirements in order to ascertain which technology and design fulfils the requirements.

To illustrate this by example, a small studio whose customers require soundtracks for multimedia training and games is running a database library of reference sounds, short audio clips, speech, and music. Although there is a need for recording audio onto hard disk, the studio manager agrees that the recording activity and processing of files in a multi-track environment can occur at the DAW located in the studio area as shown in Figure 2.10. The server containing the database is located in a purpose-built machine room elsewhere in the facility. The network is required to deliver audio files to the workstation when a project is opened from within the DAW software. At the point of opening the project, the software requests the associated project files from the server and several hundred megabytes of data may be sent over the network from the server to the workstation. Once these have been received, the files are loaded onto the workstation and the network is no longer required.

With planned and tested configuration, the workstation could be disconnected from the network and the engineer could continue normal operation. This offers some resilience in design, and ensures that a network failure would not prevent the audio engineers from performing their tasks or cause them to lose any information in the middle of a production session. The studio manager is keen on implementing a networked solution so that the DAW files are consolidated in one place for safely backing up, and can be made available to other studios within the facility. In this example, the studio manager is also keen to see the system put in place, so that the disk space can be monitored, and studio charges calculated by factoring the engineers’ time together with the disk space used.

The eventual installation must not compromise the engineers’ productivity but should enhance it. The files should therefore be delivered in a timely fashion so that the engineer is not kept waiting too long, each time a project is loaded, as this time is unproductive and soon mounts up. In order to fulfil the parameters laid out by the studio manager, a packet switching solution may be considered, since although the time factor has been specifically mentioned, it is not as important to the ‘file-loading’ nature of the description as the cost.

In the next example, illustrated in Figure 2.11, the studio manager of a large post-production complex has specified the requirements.

The complex has a centralized machine room and racks full of signal processors located in each studio. The studio manager would like to take the rack-based digital signal processors (DSP) out of the studios, placing them in the secure and environmentally controlled machine room; all the signal processors can be controlled remotely. The idea in this example is that the DSP is bunched together as a single resource and is available to any one of the studios in the complex and so can be treated as a commodity. The only limitation that the customer is prepared to accept is the limitation of the power of the DSP resource, so the network must not become the bottleneck. The studio manager is concerned that equipment is underutilized and wants to find a way of dishing out DSP on demand to any of the studios that need it, so that the usage can be measured and resources planned and justified. Once again it is important that the productivity of the engineers is not compromised in any way, meaning that the audio must travel the network from the studio, enter the processing rack and return to the studio for monitoring without any significant delay. In order to achieve this, the network designer has identified that a streaming solution is required, but that the network must be segmented in such a way that the best performance can be administered. The answer in this particular case is to ensure that the delay is incurred only after the recording has been completed and therefore to record audio onto a local workstation. Once the recording has been made, a slight delay, in the order of less than 5 ms, may be acceptable only during playback. In this way, the engineer decides on the DSP algorithms (for instance equalization, reverberation, compression, and so on) and sends the control information out to the DSP resource rack, which configures the DSP resource. The resulting audio experience can be compared with the original unaffected recording.

This scenario reflects the normal studio modus operandi, except that the audio never leaves the digital domain. The only difference is that the cable topology allows processing equipment to be located further away making it more flexible in terms of operation. The solution presented is only possible because the network designer determined that audio was initially recorded locally and so some time–displacement delay could be incurred during the production phase.

Even so, the network capacity needs to be significantly higher than the first example, and the possibility of delivery failure of any data was not an option once the audio stream had begun transmission.

The two examples presented here highlight the differences in requirements between two studios. In the first, a file store-and-forward solution will suffice, whilst the second specification requires a streaming solution. These two requirements would very rarely lead to the installation of the same network technology, and it is these differences that are expressed in the definition of the quality of service of each of the networks. The QoS is determined by the functional requirements that the network is expected to fulfil, and by the budget available for the project.

There are many parameters to bear in mind when designing a network, such as the physical location and type of the cables, any security, failure recovery and redundancy that may be required. Even the details of the storage throughput and the service record of third party providers (in the case of managed services such as those provided by telecommunication providers) might need to be considered.

2.6.3  Determining the QoS

Before looking at the different types of data networks, it is necessary to understand the demands and solutions that any proposed network is expected to undertake. Because unlimited bandwidth is not available to unlimited users, it is necessary to determine the exact nature of the services that the network is intended to support. As much relevant detail as possible should be acquired before attempting the design.

It would be a relatively simple task to determine the QoS necessary from any mechanism employed to transfer audio in the examples above. The first example could be satisfied with a simple LAN technology, such as fast Ethernet (100MB/s) or Token Ring (16MB/s or 32MB/s). The second would benefit from a more rigorous design, perhaps utilizing a technology such as asynchronous transfer mode (ATM) or Fibre Channel.

The second example is perhaps the most interesting from the point of view of most audio facilities, and part of the QoS can be determined using a fixed calculation. The studio has the requirement to move a number of streams of audio data around, and is using hi-resolution audio within the facility. The amount of data can therefore be calculated as

From the above calculation, it can be said that there must be a data rate of around 2.4 Mbits/s guaranteed for each stream on the network. Furthermore, since the requirements of the professional audio industry are being discussed, it is critical that the integrity of the data remains intact, and so delivery must be perfect every time.

Perhaps the most difficult part of determining the strategy from the point of view of the network designer is to determine the number of audio streams that might be in use at any one time on the network. The number of streams in the audio industry varies, and the acceptable number increases regularly.

Audio has not yet made a wholesale switch to using network technologies. This may be because there are relatively few uses to which limited bandwidth can be put in audio terms, because of the amount of data that digital audio create and the limitations that audio transmission places on the network in terms of time. This is compounded by the nature of public networks such as the Internet, which provide packet-based delivery, unrelated to time in the way that audio playback demands. However, this is not to say that audio has been put to all the possible uses for the current bandwidth limitations.

2.6.4  Identification of Usefulness

Some of the uses for conveniently available audio are listed below. This is not meant to be a complete list, and includes common uses for networks in the audio industry that are already in use, those in development and some which are intended purely to spark the imagination.

In terms of QoS, it is the availability of the audio as a time-displaced service that stands out as common to most of the items listed above, and it is this which separates the requirements of the audio industry from such distribution networks as broadcast and retail.

2.6.5  Broadcast

In traditional broadcast networks, such as those employed by television and radio, once the broadcast of a particular item has begun, any delay that occurs once the transmission has started will spoil the experience of the recipient. If a consumer misses a broadcast, then the experience will not be repeatable in any way other than employing some method of time displacement.

Within the audio industry, it is the availability and speed of delivery that make the digital format useful, or not. For example, if an audio engineer in a particular studio required the availability of an item that was only available once, as in the public broadcast example, then this method of distribution is completely useless (unless some other time-displacement method was used to capture the broadcast, as mentioned). If the same engineer waits for a CD to be delivered through a retail distribution network, then this is also less than ideal, because of the excessive period of time spent waiting for the material, during which the engineer is redundant.

2.6.6  Categories of QoS

By narrowing the analysis to professional use only, the extremes can be excluded and it is possible to categorize the remaining requirements into four categories of quality of service:

User Oriented

User oriented transmission offers subjective image and audio streaming quality, but is capable of file transmission as per the first of the studio examples used above. File transmission has a low time dependency, since the files are being delivered from one place to another before being manipulated. However, the requirement for reliable transmission remains high, so that any file remains intact following delivery. Although streaming is possible, quality is not guaranteed over time and any applications attempting to utilize streaming must take care to account for delivery failure of individual packets, and structure the stream to an inconsistent QoS.

In general, a single file is requested from a device attached to the network, and this is delivered to the client computer, where it is loaded into memory for processing by software control. Since time is not a significant factor in this situation, it is possible to increase the reliability of an otherwise unreliable mechanism by retransmitting any lost information, provided the file is reconstructed afterwards, in order to satisfy the requirements.

Consumer Broadcast

Although consumer broadcast in its traditional sense will not be covered specifically, this general heading also covers on-demand consumption, and one-to-many transmissions.

Internet radio broadcast is an example worth looking at, where a consumer may be searching for a service that covers a particular event, such as a sporting event not covered by traditional broadcast media.

The consumer carries out a search for an Internet service providing coverage of an event. The Internet service provider (ISP) will provide a commentary on the event, which is turned into a digital stream. The event is not ‘broadcast’ in the traditional sense, since the information will only be sent to computers which have specifically requested the stream. A slight delay of a few seconds between the transmission and receipt will not affect the enjoyment by the consumer, and may be an anticipated part of the service. In fact such a delay is common even in television broadcasts, where an event may be taking place on the other side of the world, and a propagation delay is incurred because of the distances involved.

For the purposes of the consumer, this is still considered ‘live’, although for live musicians attempting to play together over the link, the delay makes the task impossible to achieve. This was demonstrated during the famous Live Aid broadcast in 1986 where such an attempt resulted in an impossible time-keeping task for the performers.

In the case of Internet radio’s coverage of sporting events, the transmission may be the only coverage of the event, and so the consumer will be more inclined to accept poorer quality reception, involving dropouts and compressed audio. This is not to say that it is acceptable to provide information that is so degraded as to impair important information about the event. In other words, there is a point when the service becomes so degraded that the consumer will not accept the service. Therefore, the clarity and continuity of the final output has a level that can be defined. Work in this area is where Internet broadcast software manufacturers do much of their research.

Another common definition of on-demand QoS that is often associated with delivery of video is that the consumer can choose when to watch a particular program. An example of this, which is discussed in the book by Bill Gates (CEO of Microsoft Corporation at the time) entitled The Road Ahea. is that a popular film is made available by a service provider for a period of time, say a month. During this time, the viewer can ask to see the film at a time that is convenient to him or her. At the time requested, the film is broadcast to that viewer, and any other viewers who have also requested the film at that time. This scenario is some way from being reality for most people, although early tests have been carried out and the service is being operated in a few controlled locations (Lucent Technologies, 2000).

Format and Synchronization Oriented

Format and synchronization oriented QoS includes video resolution, frame rate, storage format, compression schemes and the skew between the beginning of audio and video sequences.

Commonly associated with studio processing this service level is the first of the uses for audio transfer that increases the QoS demands towards specifically designed networks. In this category, the QoS can suffer from a small delay such as that defined in the second option, consumer broadcast, above, but there must be no deterioration in the quality of the audio that is received. Applications for this would be in almost any typical studio where an audio signal is being engineered through processors and analysers before being submitted as a final product. Specifically, this may involve streams of audio being sent to processors such as mixers, equalization, effect units, and audio compressors.

In order to retain the highest possible quality of the audio, there must be no dropouts in the stream. The most obvious example of this is when streaming to a tape mechanism, when a dropout in the stream could have an audible effect on the final product. Currently, the most common form of transfer mechanisms installed to meet these requirements are from the professional audio industry, and include AES/EBU and its consumer version S/PDIF covered in Chapter 4.

Performance Oriented

Performance orientation accounts for factors such as end-to-end delay and bit rate. For video streams, delay variations in excess of 500 μs are considered annoying with variations in excess of 650 μs being intolerable.

A multimedia-enabled network must deliver a continuous stream of data that arrives at its destination at a fixed rate, even if the network becomes heavily loaded with multiple users and other data streams.

This category requires the most rigorous QoS, since a low latency time is required, as well as high quality audio. The environment in which the transfer is used might include live concerts, where performance audio is generated at the stage, from where it is transferred to the mixing platform and to the D/A process performed at the amplification stage. From there, the final analogue signal is sent out through the loud speakers or to the monitor speakers located near the performers.

2.6.7  Clarifying QoS

Clarification is now needed, for the last few pages have talked almost exclusively about the term ‘quality of service’ as if it were a product of networks and audio alone.

From the perspective of the client, the quality of any IT services can be thought of in fairly simple terms. Even though a network is made up of a number of complex parts, each of which may be supplied by a different vendor, the client sees the ability to transfer audio as a single service, not a collection of independent services. The consumer is not concerned with how the service is provided, or what components make the service possible. This is true even when looking at a smaller system, such as the simple transfer of audio between one device and another over a fibre, such as in the common case of making a recording to a digital audio tape (DAT) machine. If the transfer does not work, then the chances are that a component is failing, or is not compatible with other components in the chain. No matter which, the service of audio transfer is failing.

The concerns of the client can be classified into broad headings and questions relating to any installation should be aimed at determining these expectations for:

Availability

Availability focuses upon the question of whether the client can use the services when the client wishes to use them. This depends upon when the service is required. In the example of the administrative offices of a studio complex, this may be from 9 am until 5 pm from Monday to Friday. On the other hand, the studio rooms themselves may be in operation from 10 am until midnight except at weekends when a 24-hour service is required.

For a video-on-demand service provider, a number of choices may be presented. The provider may choose to offer each movie at specific times only, say, twice a night on weekdays, or five times at the weekend. In the end, the service provider permutates the provision of the service to manageable times, whilst considering the consumers’ demands, in order to offer a successful service. Most IT managers are aware of the scheduled hours of operation for the service they are operating, and upgrades and other planned maintenance will be scheduled for other times.

Performance

For IT personnel, performance of a network will generally be measured in terms of packets per second, transactions, response times, or other actual measurements that are available from the equipment within the service. Understandably, the customer is not concerned with such jargon, and describes things in rather less specific language such as ‘Does it function at an acceptable speed?’ It is the job of the designer or IT manager to turn broad descriptions of requirement into measurements and evidence that the service is performing as expected. The question of speed may be the question of response time in an online transaction system. In another case it might be the time that it takes to move a copy of a file from one office to another, or the time required to load an application to a desktop system from a server.

In an audio system, performance stands out in any definition of QoS as one of the main problems because of the unaccustomed quantity of data and the time dependency that audio intrinsically requires.

Accuracy

As with performance, accuracy is also a significant issue for the transfer of audio. It is enough to say that the demands of the different QoS definitions result in different requirements for accuracy. For instance, nothing less than 100% accuracy will do for the transfer of audio in performance orientation, whilst some compromises may be negotiated when defined as a consumer broadcast for special interest groups over the Internet as described earlier.

A simple example of accuracy would be whether e-mail is delivered to the correct recipient or not. Similarly, in the case of applying transactions to a database, it is essential that the change be applied to the proper version. It can be seen that if services do not accurately perform their functions, high availability and high performance are worthless.

Cost

Although performance and accuracy are directly related to QoS, the cost of the service cannot be ignored. A well-known saying in Information Technology is ‘Fast, cheap, good – you can have two out of three’ and this can be adapted to the delivery of almost any service.

It is common practice for network vendors to present a number of solutions to an invitation to tender (ITT) for the supply of a service and let the customer choose the most suitable. Three separate proposals are common, with the best offering being the most expensive, but able to offer all of the functionality and speed that the customer has asked for, perhaps with a few extra bells and whistles that the vendor hopes will spark the imagination of the customer. The second will typically match the customer’s requirements more precisely and will be the middle offering in terms of cost. The third offering will usually be the bare bones system that barely fulfils the basic customer demands, and will be the one with the lowest cost.

2.6.8  Managing Expectations

It is possible with most services to provide whatever it is the customer asks for. However, in order to do so may mean an unreasonable cost.

To illustrate this, imagine a ruler marked out in centimetres starting from 0 as shown in Figure 2.12. From one end to the other end is a fixed value of say 20 cm. A mark is placed halfway down the ruler, marking out two equal lengths of 10 cm. For the next stage, another mark is added exactly halfway between the first mark and 0. Then another mark is made exactly halfway between the second mark and 0. This exercise can be continued indefinitely, with a mark being placed halfway between the last mark and the end of the ruler each time. Although the measurement between the marks decreases by 50% each time, each consecutive mark will be exactly halfway between the last mark and zero. If there were a cost or effort associated with making each mark, then the cost for the project would increase by the same amount for each mark that is made, for increasingly diminished returns, and the objective would never be reached. Using this analogy then, a perfect service that is always available with excellent performance and 100% accuracy is simply not possible.

Notes and Further Reading

Entry for OSI Seven Layer Model (1999) Connected: An Internet Encyclopeadia. http://www.freesoft.org/CIE/index.htm.

Kristula, Dave (1997) The History of the Internet. http://www.davesite.com/webstation/net-history.shtml. Various other sources.

Lucent Technologies, Murray Hill, New Jersey, USA (2000) http://www.lucent.com/press/0398/980326.bla.html. Various other sources.

www.whatis.com. Entry by Harbeck, Reg (1999) What is a Network? WhatIs.com, 55 West Chestnut Street, Kingston, NY 12401, USA.