Computer Networks 4th Ed Andrew S. Tanenbaum [Electronic resources] نسخه متنی

4.6 Bluetooth

In 1994, the L. M. Ericsson company became interested in connecting its mobile phones to other devices (e.g., PDAs) without cables. Together with four other companies (IBM, Intel, Nokia, and Toshiba), it formed a SIG (Special Interest Group, i.e., consortium) to develop a wireless standard for interconnecting computing and communication devices and accessories using short-range, low-power, inexpensive wireless radios. The project was named Bluetooth, after Harald Blaatand (Bluetooth) II (940-981), a Viking king who unified (i.e., conquered) Denmark and Norway, also without cables.

Although the original idea was just to get rid of the cables between devices, it soon began to expand in scope and encroach on the area of wireless LANs. While this move makes the standard more useful, it also creates some competition for mindshare with 802.11. To make matters worse, the two systems also interfere with each other electrically. It is also worth noting that Hewlett-Packard introduced an infrared network for connecting computer peripherals without wires some years ago, but it never really caught on in a big way.

Undaunted by all this, in July 1999 the Bluetooth SIG issued a 1500-page specification of V1.0. Shortly thereafter, the IEEE standards group looking at wireless personal area networks, 802.15, adopted the Bluetooth document as a basis and began hacking on it. While it might seem strange to standardize something that already had a very detailed specification and no incompatible implementations that needed to be harmonized, history shows that having an open standard managed by a neutral body such as the IEEE often promotes the use of a technology. To be a bit more precise, it should be noted that the Bluetooth specification is for a complete system, from the physical layer to the application layer. The IEEE 802.15 committee is standardizing only the physical and data link layers; the rest of the protocol stack falls outside its charter.

Even though IEEE approved the first PAN standard, 802.15.1, in 2002, the Bluetooth SIG is still active busy with improvements. Although the Bluetooth SIG and IEEE versions are not identical, it is hoped that they will soon converge to a single standard.

4.6.1 Bluetooth Architecture

Let us start our study of the Bluetooth system with a quick overview of what it contains and what it is intended to do. The basic unit of a Bluetooth system is a piconet, which consists of a master node and up to seven active slave nodes within a distance of 10 meters. Multiple piconets can exist in the same (large) room and can even be connected via a bridge node, as shown in Fig. 4-35. An interconnected collection of piconets is called a scatternet.

Figure 4-35. Two piconets can be connected to form a scatternet.

In addition to the seven active slave nodes in a piconet, there can be up to 255 parked nodes in the net. These are devices that the master has switched to a low-power state to reduce the drain on their batteries. In parked state, a device cannot do anything except respond to an activation or beacon signal from the master. There are also two intermediate power states, hold and sniff, but these will not concern us here.

The reason for the master/slave design is that the designers intended to facilitate the implementation of complete Bluetooth chips for under $5. The consequence of this decision is that the slaves are fairly dumb, basically just doing whatever the master tells them to do. At its heart, a piconet is a centralized TDM system, with the master controlling the clock and determining which device gets to communicate in which time slot. All communication is between the master and a slave; direct slave-slave communication is not possible.

4.6.2 Bluetooth Applications

Most network protocols just provide channels between communicating entities and let applications designers figure out what they want to use them for. For example, 802.11 does not specify whether users should use their notebook computers for reading e-mail, surfing the Web, or something else. In contrast, the Bluetooth V1.1 specification names 13 specific applications to be supported and provides different protocol stacks for each one. Unfortunately, this approach leads to a very large amount of complexity, which we will omit here. The 13 applications, which are called profiles, are listed in Fig. 4-36. By looking at them briefly now, we may see more clearly what the Bluetooth SIG is trying to accomplish.

Figure 4-36. The Bluetooth profiles.

The generic access profile is not really an application, but rather the basis upon which the real applications are built. Its main job is to provide a way to establish and maintain secure links (channels) between the master and the slaves. Also relatively generic is the service discovery profile, which is used by devices to discover what services other devices have to offer. All Bluetooth devices are expected to implement these two profiles. The remaining ones are optional.

The serial port profile is a transport protocol that most of the remaining profiles use. It emulates a serial line and is especially useful for legacy applications that expect a serial line.

The generic object exchange profile defines a client-server relationship for moving data around. Clients initiate operations, but a slave can be either a client or a server. Like the serial port profile, it is a building block for other profiles.

The next group of three profiles is for networking. The LAN access profile allows a Bluetooth device to connect to a fixed network. This profile is a direct competitor to 802.11. The dial-up networking profile was the original motivation for the whole project. It allows a notebook computer to connect to a mobile phone containing a built-in modem without wires. The fax profile is similar to dial-up networking, except that it allows wireless fax machines to send and receive faxes using mobile phones without a wire between the two.

The next three profiles are for telephony. The cordless telephony profile provides a way to connect the handset of a cordless telephone to the base station. Currently, most cordless telephones cannot also be used as mobile phones, but in the future, cordless and mobile phones may merge. The intercom profile allows two telephones to connect as walkie-talkies. Finally, the headset profile provides hands-free voice communication between the headset and its base station, for example, for hands-free telephony while driving a car.

The remaining three profiles are for actually exchanging objects between two wireless devices. These could be business cards, pictures, or data files. The synchronization profile, in particular, is intended for loading data into a PDA or notebook computer when it leaves home and collecting data from it when it returns.

Was it really necessary to spell out all these applications in detail and provide different protocol stacks for each one? Probably not, but there were a number of different working groups that devised different parts of the standard, and each one just focused on its specific problem and generated its own profile. Think of this as Conway's law in action. (In the April 1968 issue of Datamation magazine, Melvin Conway observed that if you assign n people to write a compiler, you will get an n-pass compiler, or more generally, the software structure mirrors the structure of the group that produced it.) It would probably have been possible to get away with two protocol stacks instead of 13, one for file transfer and one for streaming real-time communication.

4.6.3 The Bluetooth Protocol Stack

The Bluetooth standard has many protocols grouped loosely into layers. The layer structure does not follow the OSI model, the TCP/IP model, the 802 model, or any other known model. However, IEEE is working on modifying Bluetooth to shoehorn it into the 802 model better. The basic Bluetooth protocol architecture as modified by the 802 committee is shown in Fig. 4-37.

Figure 4-37. The 802.15 version of the Bluetooth protocol architecture.

The bottom layer is the physical radio layer, which corresponds fairly well to the physical layer in the OSI and 802 models. It deals with radio transmission and modulation. Many of the concerns here have to do with the goal of making the system inexpensive so that it can become a mass market item.

The baseband layer is somewhat analogous to the MAC sublayer but also includes elements of the physical layer. It deals with how the master controls time slots and how these slots are grouped into frames.

Next comes a layer with a group of somewhat related protocols. The link manager handles the establishment of logical channels between devices, including power management, authentication, and quality of service. The logical link control adaptation protocol (often called L2CAP) shields the upper layers from the details of transmission. It is analogous to the standard 802 LLC sublayer, but technically different from it. As the names suggest, the audio and control protocols deal with audio and control, respectively. The applications can get at them directly, without having to go through the L2CAP protocol.

The next layer up is the middleware layer, which contains a mix of different protocols. The 802 LLC was inserted here by IEEE for compatibility with its other 802 networks. The RFcomm, telephony, and service discovery protocols are native. RFcomm (Radio Frequency communication) is the protocol that emulates the standard serial port found on PCs for connecting the keyboard, mouse, and modem, among other devices. It has been designed to allow legacy devices to use it easily. The telephony protocol is a real-time protocol used for the three speech-oriented profiles. It also manages call setup and termination. Finally, the service discovery protocol is used to locate services within the network.

The top layer is where the applications and profiles are located. They make use of the protocols in lower layers to get their work done. Each application has its own dedicated subset of the protocols. Specific devices, such as a headset, usually contain only those protocols needed by that application and no others.

In the following sections we will examine the three lowest layers of the Bluetooth protocol stack since these roughly correspond to the physical and MAC sublayers.

4.6.4 The Bluetooth Radio Layer

The radio layer moves the bits from master to slave, or vice versa. It is a low-power system with a range of 10 meters operating in the 2.4-GHz ISM band. The band is divided into 79 channels of 1 MHz each. Modulation is frequency shift keying, with 1 bit per Hz giving a gross data rate of 1 Mbps, but much of this spectrum is consumed by overhead. To allocate the channels fairly, frequency hopping spread spectrum is used with 1600 hops/sec and a dwell time of 625 µsec. All the nodes in a piconet hop simultaneously, with the master dictating the hop sequence.

Because both 802.11 and Bluetooth operate in the 2.4-GHz ISM band on the same 79 channels, they interfere with each other. Since Bluetooth hops far faster than 802.11, it is far more likely that a Bluetooth device will ruin 802.11 transmissions than the other way around. Since 802.11 and 802.15 are both IEEE standards, IEEE is looking for a solution to this problem, but it is not so easy to find since both systems use the ISM band for the same reason: no license is required there. The 802.11a standard uses the other (5 GHz) ISM band, but it has a much shorter range than 802.11b (due to the physics of radio waves), so using 802.11a is not a perfect solution for all cases. Some companies have solved the problem by banning Bluetooth altogether. A market-based solution is for the network with more power (politically and economically, not electrically) to demand that the weaker party modify its standard to stop interfering with it. Some thoughts on this matter are given in (Lansford et al., 2001).

4.6.5 The Bluetooth Baseband Layer

The baseband layer is the closest thing Bluetooth has to a MAC sublayer. It turns the raw bit stream into frames and defines some key formats. In the simplest form, the master in each piconet defines a series of 625 µsec time slots, with the master's transmissions starting in the even slots and the slaves' transmissions starting in the odd ones. This is traditional time division multiplexing, with the master getting half the slots and the slaves sharing the other half. Frames can be 1, 3, or 5 slots long.

The frequency hopping timing allows a settling time of 250260 µsec per hop to allow the radio circuits to become stable. Faster settling is possible, but only at higher cost. For a single-slot frame, after settling, 366 of the 625 bits are left over. Of these, 126 are for an access code and the header, leaving 240 bits for data. When five slots are strung together, only one settling period is needed and a slightly shorter settling period is used, so of the 5 x 625 = 3125 bits in five time slots, 2781 are available to the baseband layer. Thus, longer frames are much more efficient than single-slot frames.

Each frame is transmitted over a logical channel, called a link, between the master and a slave. Two kinds of links exist. The first is the ACL (Asynchronous Connection-Less) link, which is used for packet-switched data available at irregular intervals. These data come from the L2CAP layer on the sending side and are delivered to the L2CAP layer on the receiving side. ACL traffic is delivered on a best-efforts basis. No guarantees are given. Frames can be lost and may have to be retransmitted. A slave may have only one ACL link to its master.

The other is the SCO (Synchronous Connection Oriented) link, for real-time data, such as telephone connections. This type of channel is allocated a fixed slot in each direction. Due to the time-critical nature of SCO links, frames sent over them are never retransmitted. Instead, forward error correction can be used to provide high reliability. A slave may have up to three SCO links with its master. Each SCO link can transmit one 64,000 bps PCM audio channel.

4.6.6 The Bluetooth L2CAP Layer

The L2CAP layer has three major functions. First, it accepts packets of up to 64 KB from the upper layers and breaks them into frames for transmission. At the far end, the frames are reassembled into packets again.

Second, it handles the multiplexing and demultiplexing of multiple packet sources. When a packet has been reassembled, the L2CAP layer determines which upper-layer protocol to hand it to, for example, RFcomm or telephony.

Third, L2CAP handles the quality of service requirements, both when links are established and during normal operation. Also negotiated at setup time is the maximum payload size allowed, to prevent a large-packet device from drowning a small-packet device. This feature is needed because not all devices can handle the 64-KB maximum packet.

4.6.7 The Bluetooth Frame Structure

There are several frame formats, the most important of which is shown in Fig. 4-38. It begins with an access code that usually identifies the master so that slaves within radio range of two masters can tell which traffic is for them. Next comes a 54-bit header containing typical MAC sublayer fields. Then comes the data field, of up to 2744 bits (for a five-slot transmission). For a single time slot, the format is the same except that the data field is 240 bits.

Figure 4-38. A typical Bluetooth data frame.

Let us take a quick look at the header. The Address field identifies which of the eight active devices the frame is intended for. The Type field identifies the frame type (ACL, SCO, poll, or null), the type of error correction used in the data field, and how many slots long the frame is. The Flow bit is asserted by a slave when its buffer is full and cannot receive any more data. This is a primitive form of flow control. The Acknowledgement bit is used to piggyback an ACK onto a frame. The Sequence bit is used to number the frames to detect retransmissions. The protocol is stop-and-wait, so 1 bit is enough. Then comes the 8-bit header Checksum. The entire 18-bit header is repeated three times to form the 54-bit header shown in Fig. 4-38. On the receiving side, a simple circuit examines all three copies of each bit. If all three are the same, the bit is accepted. If not, the majority opinion wins. Thus, 54 bits of transmission capacity are used to send 10 bits of header. The reason is that to reliably send data in a noisy environment using cheap, low-powered (2.5 mW) devices with little computing capacity, a great deal of redundancy is needed.

Various formats are used for the data field for ACL frames. The SCO frames are simpler though: the data field is always 240 bits. Three variants are defined, permitting 80, 160, or 240 bits of actual payload, with the rest being used for error correction. In the most reliable version (80-bit payload), the contents are just repeated three times, the same as the header.

Since the slave may use only the odd slots, it gets 800 slots/sec, just as the master does. With an 80-bit payload, the channel capacity from the slave is 64,000 bps and the channel capacity from the master is also 64,000 bps, exactly enough for a single full-duplex PCM voice channel (which is why a hop rate of 1600 hops/sec was chosen). These numbers mean that a full-duplex voice channel with 64,000 bps in each direction using the most reliable format completely saturates the piconet despite a raw bandwidth of 1 Mbps. For the least reliable variant (240 bits/slot with no redundancy at this level), three full-duplex voice channels can be supported at once, which is why a maximum of three SCO links is permitted per slave.

There is much more to be said about Bluetooth, but no more space to say it here. For more information, see (Bhagwat, 2001; Bisdikian, 2001; Bray and Sturman, 2002; Haartsen, 2000; Johansson et al., 2001; Miller and Bisdikian, 2001; and Sairam et al., 2002).