19.6.1 Medium Access Protocols
Medium access control (MAC) for wireless ad hoc networks is currently a very active research topic. The characteristics of the network, the diverse physical-layer technologies available, and the range of services envisioned render a difficult task the design of an algorithm to discipline the access to the shared medium that results efficient, fair, power consumption sensitive, and delay bound. A number of issues distinguish wireless MAC protocols from those used in wireline networks (Chandra, 2002), as quoted next.
Half-duplex operation. Due to self-interference (i.e., the energy from the transmitter that leaks into the receiver), it is difficult to construct terminals able to receive while transmitting. Therefore collision detection while sending data is not possible and Ethernet-like protocols cannot be used. Since collisions cannot be detected, wireless MAC protocols use collision avoidance mechanisms to minimize the probability of collision. Time varying channel. In multipath fading channels, the received signal is the sum of time-shifted and attenuated copies of the transmitted signal. With the change of the channel characteristics as well as in the relative position of terminals, the signal envelope varies as a function of time. The signal experiences fading that may be severe. The nodes establishing a wireless link need to sense the channel so as to assess the communication link conditions.
Burst channel errors. Wireless channels experience higher bit error rate than wireline transmissions.
Besides, errors occur in bursts as the signal fades, resulting in high probability of packet loss. Therefore, an acknowledgement mechanism must be implemented so that the packet retransmission may be possible in case of packet loss.
Location-dependent carrier sensing. Because the signal strength decays with distance according to a power law, only nodes within a specific range are able to communicate. This gives rise to the hidden and exposed terminals and the capture effect, as described next.
Hidden terminal. Refer to Fig. 19.51 where the relative positions of terminals A, B, and C are shown. B is within range of both A and C but A and C are out of range of each other. If terminal A is transmitting to B and terminal C wishes to transmit to B, it incorrectly senses that the channel is free because it is out of range of A, the current transmitter. If C starts transmitting it interferes with the reception at B. In this case C is termed the hidden terminal to A. The hidden terminal problem can be minimized with the use of the request-to-sent/clear-to-send (RTS/CTS) handshake protocol (to be explained later) before the data transmission starts.
Exposed terminal. An exposed terminal is one that is within range of the transmitter but out of range of the receiver. In Fig. 19.51, if terminal B is transmitting to A and terminal C senses the channel it perceives it as busy. However, since it is out of range of terminal A it cannot interfere with the current conversation.
Therefore it can utilize the channel to establish a parallel link with another terminal that is out of range of B, for instance, terminal D. In this case C is termed the exposed terminal to B. Exposed terminals may result in under-usage of the channel. As in the hidden terminal problem, this also can be minimized with the use of the RTS/CTS handshake protocol.
Capture. Capture at a given terminal occurs in case among several simultaneous signals arriving at it the signal strength of one of them prevails over all of the others combined. In Fig. 19.51, terminals C and E are both within range of terminal B. If C and E are transmitting the interference may result in a collision at B. However, B may be able to receive successfully if one of the signals is much higher than the other, for instance, the signal from E. Capture can improve throughput because it results in less collisions. However, it favors senders that are closer to the intended destination, which may cause unfair allocation of the channel.
From the above considerations, it is promptly inferred that the design of a MAC protocol for ad hoc networks requires a different set of parameters must be considered as compared with those of the wireline systems.
Protocols Categories
Jurdak et al. (Jurdak, 2004), after conducting a survey and analysis of a number of current MAC protocol proposals, offer a set of key features that may be used in order to classify MAC protocols for ad hoc networks.
Channel separation and access. The way the medium is organized is an important issue in the protocol design. For instance, all stations may share a single channel, which they use for control and data transmissions. On the other hand, the medium may be divided into multiple channels, in general one for control and the others for data. The single channel approach was favored in earlier MAC designs because of its simplicity. However, it is intrinsically subject to collisions and it does not perform well in medium to heavy traffic conditions. Particularly at heavy loads, simulations show that single channel protocols are prone to increased number of collisions of control packets, for example, RTS and CTS, which cause increased back off delays while the medium is idle (Tseng, 2002). The choice for multiple channels brings the issue of how to separate these channels. The most common ways of separating channels make use of FDMA, TDMA, and CDMA technologies. Frequency division multiple access (FDMA) uses multiple carriers to divide the medium into several frequency slots. It allows multiple transmissions to occur simultaneously although each sender can use only the bandwidth of its assigned frequency slot. Time division multiple access (TDMA) divides the medium into fixed length time slots. A group of slots forms a time frame and defines the slot repetition rate. Because of its periodic nature, TDMA protocols are suitable to delay sensitive traffic. In TDMA, a sender uses the whole available bandwidth for the duration of a slot assigned to it. In addition, to access the medium terminals need to keep track of frames and slots and, as a result, TDMA protocols require synchronization among terminals. Code division multiple access (CDMA) allows
senders to use the whole available bandwidth all the time. Each sender is assigned one of several orthogonal codes and simultaneous transmissions are possible for users are identified by their unique code. A general requirement in CDMA is for power control. The reason behind it is that an unwanted signal that is stronger than the desired signal may overwhelm it at the receiver’s antenna. This is known as the near-far effect. space division multiple access (SDMA), similarly to CDMA, aims at allowing senders to use the whole available bandwidth all the time. However, the terminals use directional antennas and are allowed to start transmission only if the desired transmission’s direction does not interfere with an ongoing conversation. RTS/CTS handshake. Many MAC protocols for ad hoc networks use variants of the RTS/CTS handshake. The original three-way handshake minimizes both the hidden and exposed terminal problems. A terminal wishing to send data first senses the channel. If the channel is idle for the appropriate amount of time, the terminal sends a short request-to-send (RTS) packet. All terminals on hearing the RTS defer their transmissions. The destination responds with a clear-to-send (CTS) packet. All terminals on hearing the CTS also defer their transmissions. The sender, on receiving the CTS assumes the channel is acquired and initiates the data transmission.
Topology. Ad hoc networks have a large degree of flexibility and uncertainty. Terminals may be mobile and have distinct capabilities and resources. The network must take this into account and adapt dynamically while optimizing performance and minimizing power consumption (Jurdak, 2004). A network topology can be centralized, clustered, or flat. Centralized topologies have a single terminal or base station that controls and manages the network. The central terminal may be responsible for broadcasting information relevant to the operation of the network. In addition, terminals may only communicate through the central terminal. Clustered topologies create a local version of a centralized network where one terminal assumes some or all of the duties expected from the central terminal. Flat topologies implement a fully distributed approach where all terminals are at the same level, and central control is not used. Flat topologies are further divided into single-hop and multiple-hop. Single-hop assumes that the destination node is within range of the sender. Multiple-hop assumes that the destination node may be beyond the sender’s reachable neighbors. In this case, intermediate terminals are responsible for relaying the packets until they reach the intended destination. Single-hop protocols are simpler but pose limitations on the size of the network.
Multiple-hop adds scalability to the network at the expense of higher complexity.
Power. Power consumption is a relevant issue for all wireless networks. Power conservation is particularly influential for the mobile terminals because of the limited battery power available. An efficient power conservation strategy involves several aspects. The energy used to transmit the signal represents a large share of the power consumption. Ideally the transmit power used should be just enough to reach the intended destination. Another source of wasted energy is the long periods of time terminals need to spend sensing the channel or overhearing irrelevant conversation. If terminals are able to learn in advance about when the medium will be unavailable they may decide to go into a sleep mode for that period of time in order to save energy. The network behavior may be influenced by the terminals’ battery power level, for instance, in the selection of a cluster head or in assigning transmission priorities. Terminals aware of their battery level may adjust their behavior accordingly. The exchange of control messages before the data transmission phase also represents power wastage. Reduced control overhead should therefore be pursued for the sake of power efficiency.
Transmission initiation. Intuitively, it is expected that a terminal wishing to start a conversation must initiate the transmission. And, in fact, most of the protocols are organized this way. However, a receiver- initiated protocol may be more suitable to some specialized networks, for instance, a sensor network. In receiver-initiated protocols the receiver polls its neighbors by sending a ready-to-receive (RTR) packet, which indicates its readiness to receive data. If the receiver is able to know or successfully predict when a neighbor wishes to send its data, this class of protocols actually produces better performance. However, for generalized networks and unpredictable traffic, sender-initiated protocols are still a better choice.
Traffic load and scalability. Protocols are usually optimized for the worst expected scenario. Sparse node distribution and light traffic conditions do not pose a challenge for the implementation of ad hoc networks. The protocols are optimized for high traffic load, high node density, and/or real-time traffic, depending on the intended use. Protocols that offer the possibility of channel reservation are those with best performance on both high load and real-traffic situations. Receiver-initiated approaches also tend to work well in high load conditions because there is a high probability that RTR packets reach terminals wishing to send data. If the network ranks terminals and traffic, then it is able to assign priorities based on the traffic nature. Therefore, it can offer favored handling of real-time traffic. Dense networks tend to suffer from higher interference because of the proximity of transmitting nodes. For this reason the use of power control makes a significant difference in the performance of the network.
Range. Transmission range is the distance from the transmitter’s antenna that the radio signal strength still remains above the minimum usable level. Protocols can be classified (Jurdak, 2004) as very short-range (range up to 10 m), short-range (from 10 up to 100 m), medium-range (from 100 up to 1000 m), and long-range (from 1000 m). There is a trade-off between increasing the transmission range and achieving high spatial capacity that needs to be negotiated during the protocol design.
Industry Standard Protocols
IEEE 802.11
The family of IEEE 802.11 standards (IEEE, 1999a; IEEE, 1999b; IEEE, 1999c) can be viewed as a wireless version of the local area network (LAN) protocol Ethernet. The 802.11a standard operates in the unlicensed 5 GHz band and offers data rates up to 54 Mb/s. The commercially popular 802.11b operates in the industrial, scientific, and medical (ISM) band at 2.4 GHz and offers data rates up to 11 Mb/s. The current activity of the 802.11-working group is toward quality of service (QoS) (802.11e, described later) and security (802.11i). The 802.11 standards focus on the specification of the MAC and physical (PHY) layers. While their PHY layers differ, existing 802.11 standards rely on the same medium access mechanisms. The basic (and mandatory) access mechanism is referred to as distributed coordination function (DCF). The optional point coordination function (PCF) is an access mechanism in which a central node (the access point) polls terminals according to a list. DCF is available for both flat ad hoc and centralized topologies whereas PCF is only available in centralized configurations. MAC offers two types of traffic services. The mandatory asynchronous data service is based on the best effort and is suited to delay insensitive data. The optional time-bound service is implemented using PCF.
DCF uses the listen-before-talk scheme based on carrier sense multiple access (CSMA). A terminal wishing to transmit a data packet first monitors the medium activity. If the channel is detected idle the terminal waits for a DCF interframe space (DIFS) time interval (34 us in 802.11a). If the channel remains idle during the DIFS period, the terminal starts transmitting its packet immediately after DIFS has expired. The transmission is successfully completed when the sender receives an acknowledgement (ACK) packet from the destination. However, if the channel is sensed busy a collision avoidance procedure is used. In this procedure, after sensing the channel idle again for a DIFS period, the terminal wishing to transmit waits an additional random backoff time. The terminal then initiates its transmission if the channel remains idle during this additional time. The backoff time is a multiple of the slot time (9 us in 802.11a) and it is determined individually and independently by each station. A random number between zero and contention window (CW) is selected for any new transmission attempt. The back off time is decremented while the medium is in contention phase and frozen otherwise. Thus the backoff time may be carried over for several busy cycles of the medium before it expires. Refer to Fig. 19.52 for an example of the backoff
procedure. The initial value for CW is CWmin (15 for 802.11a) and since all terminals operate with the same CWmin value they all have the same initial medium access priority. After any failed transmission, that is, when the transmitted packet is not acknowledged, the sender doubles its CW up to a maximum defined by CWmax (1023 in 802.11a). A now larger CW decreases the probability of collisions if multiple terminals are trying to access the medium. To reduce the hidden terminal problem, 802.11 optionally uses the RTS/CTS handshake. Both RTS and CTS packets include information on how long the data frame transmission is going to last, including the corresponding ACK. Terminals receiving either the RTS or CTS use this information to start a timer, called network allocation vector (NAV), which informs the period of time the medium is unavailable. Between consecutive frames RTS and CTS, and a data frame and its ACK, the short interframe space (SIFS) (16 us in 802.11) is used. SIFS is shorter than DIFS and therefore gives the terminals sending these frames priority to access the medium.
HIPERLAN 1
High Performance LAN type 1 (HIPERLAN 1) is a wireless LAN standard operating in the 5 GHz band, which offers data rate up to 23.5 Mb/s to mobile users in either clustered ad hoc or centralized topology. HIPERLAN 1 offers asynchronous best effort and time-bound services with hierarchical priorities. There are five priority values defined, from zero (highest) to four (lowest). Each individual MAC protocol data unit (PDU) is assigned a priority that is closely related to its normalized residual lifetime (NRL) value. The NRL is an estimation of the time-to-live the PDU has considering the number of hops it still has to travel. A PDU is discarded if its NRL value reaches zero. In addition, some terminals are designed forwarders and are responsible to relay data to distant nodes in a multi-hop fashion. HIPERLAN 1 allows terminals to go into sleep mode in order to save energy. These terminals, called p-savers, inform support terminals, called p-supporters, of their sleep/wake-up patterns. p-supporters then buffer packets directed to p-savers terminals, as required. Although it has some interesting features, HIPERLAN 1 has not been a commercial success. The channel access mechanism used in HIPERLAN 1 is the elimination-yield non- preemptive priority access (EY-NPMA). It comprises three phases: prioritization (determine the highest priority data packets to be sent); contention (eliminate all contenders except one); and transmission. During the prioritization phase, time is divided in five minislots, numbered sequentially from zero to four. A terminal wishing to transmit has to send a burst during the minislot corresponding to its MAC PDU priority. For example, a terminal with a priority two PDU monitors the medium during minislots zero and one before it can assert its intention by transmitting a burst during minislot two. If the medium becomes busy during either minislot zero or one this terminal defers its transmission. Once a burst is transmitted, the prioritization phase ends and only terminals having PDUs at the same priority level remains in the dispute. The contention phase follows. It starts with the contending terminals transmitting an elimination burst. The individual terminals select the burst length, varying from 0 to 12 minislots, at random and independently. After transmitting the burst the terminals sense the medium. If it is busy they defer their transmissions. Otherwise, the remaining terminals enter the yield listening period. They select at random and independently a value between 0 and 9 and start monitoring the medium. If at the end of this period the medium is still idle the terminal assumes it has won the contention and is allowed to transmit its data. Otherwise, it defers its transmission. It is clear that the mechanism does not allow any lower priority packet to be sent if another with higher priority packet is waiting. At the same time the mechanism does not totally eliminate the possibility of collision but reduces it considerably. Similarly to IEEE 802.11, if the medium has been idle for a time longer than the interframe period a terminal wishing to transmit can bypass the EY-NPMA and transmit immediately.
Bluetooth
Bluetooth (Bluetooth, 1999) is a wireless protocol using the license-free ISM band to connect mobile and desktop devices such as computers and computers peripherals, handheld devices, cell phones, etc. The aim is to produce low-cost, low-power, and very-short range devices able to convey voice and data transmissions at a maximum gross rate of 1 Mb/s. Bluetooth uses frequency hopping spread spectrum (FHSS) with 1600 hops/s. For voice, a 64 kb/s full-duplex link called synchronous connection oriented (SCO) is used. SCO assigns a periodic single slot to a point-to-point conversation. Data communication uses the best effort asynchronous connectionless (ACL) link in which up to five slots can be assigned. Terminals in Bluetooth are organized in piconets. A piconet contains one terminal identified as the master and up to seven other active slaves. The master determines the hopping pattern and the other terminals need to synchronize to the piconet master. When it joins a piconet, an active terminal is assigned a unique 3-bit long active member address (AMA). It then stays in either transmit state, when it is engaged in a conversation, or connected state. Bluetooth supports three low-power states: park, hold, and sniff. A parked terminal releases its AMA and is assigned one of the 8-bit long parked member address (PMA). Terminals in the hold and sniff states keep their AMA but have limited participation in the piconet. For instance, a terminal in the hold state is unable to communicate using ACL. A terminal not participating in any piconet is in stand-by state. Bluetooth piconets can co-exist in space and time and a terminal may belong to several piconets. A piconet is formed when its future master starts an inquiry process, that is, inquiry messages are broadcast in order to find other terminals in the vicinity. After receiving inquiry responses the master may explicitly page terminals to join the piconet. If a master knows already another terminal’s identity it may skip the inquiry phase and page the terminal directly. Bluetooth uses time division duplex (TDD) in which master and slave alternate the opportunity to transmit. A slave can only transmit if the master has just transmitted to it, that is, slaves transmit if polled by the master. Transmissions may last one, three, or five slots although only single-slot transmission is a mandatory feature.
IEEE 802.11e
The IEEE 802.11e is an emerging MAC protocol, which defines a set of QoS features to be added to the 802.11 family of wireless LAN standards. Currently there is a draft version of the specifications (IEEE, 2003). The aim is to better serve delay-sensitive applications, such as voice and multi-media. In 802.11e, the contention-based medium access is referred to as enhanced distributed channel access (EDCA). In order to accommodate different traffic priorities, four access categories (AC) have been introduced. To each AC corresponds a backoff entity. The four distinct parallel backoff entities present in each 802.11e terminal are called (from highest to lowest priority): voice, video, best effort, and background. For the sake of comparison, existing 802.11/a/b standards define only one backoff entity per terminal. Each backoff entity has a distinct set of parameters, such as CWmin, Cwmax, and the arbitration interframe space (AIFS). AIFS is at least equal to DIFS and can be enlarged if desired. Another feature added to 802.11e is referred to as transmission opportunity (TxOP). A TxOP defines a time interval, which a back off entity can use to transmit data. It is specified by its starting time and duration, and the maximum length is AC dependent. The protocol also defines the maximum lifetime of each MAC service data unit (MSDU), which is also AC dependent. Once the maximum lifetime has elapsed, the MSDU is discarded. Finally, the protocol allows for the optional block acknowledgement in which a number of consecutive MSDUs are acknowledged with a single ACK frame.
Other Protocols
PRMA—Packet reservation multiple access (Goodman, 1989). In PRMA, the medium is divided into slots and a group of N slots forms a frame. Slots are either reserved or available. The access to the medium is provided by means of the slotted-ALOHA protocol. Data may be either periodic or sporadic, and this is informed in the header of the packet. Terminals are allowed to reserve a slot when they have periodic data to transmit. Once the central node successfully acknowledges the periodic packet, the terminal assumes the slot is reserved and uses it without contention. When the terminal stops sending periodic information then the reserved slot is released. PRMA assumes the existence of a central node but the mechanism can be adapted to other topologies (Jiang, 2002).
MACA-BI—Multiple access with collision avoidance by invitation (Talucci, 1997). In MACA-BI, the receiver polls a prospective sender by transmitting a ready-to-receive (RTR) packet. (This is an example of a receiver-initiated protocol.) In order to perform the polling in a timely fashion the receiver is required to correctly predict the traffic originated by the sender. Periodic traffic makes this task easier. In case either the data buffer or the delay at the terminal increases above a certain threshold this terminal may trigger a
conversation by transmitting an RTS packet. Improvements to MACA-BI are proposed in (Garcia, 1999), in which RIMA-SP–receiver initiated multiple access with simple polling –, and RIMA-DP–Receiver initiated multiple access with dual-purpose polling–are introduced. Both protocols render the RTR-data handshake collision free. RIMA-DP gives an additional purpose to the RTR packet: a request for transmission from the polling terminal. After a reservation phase both terminals can exchange data between them.
DBTMA—Dual busy tone multiple access (Haas, 2002). In DBTMA, the RTS/CTS handshake is replaced by two out-of-band busy tones, namely: BTt (transmit busy tone) and BTr (receive busy tone). When a terminal has data to transmit, it first senses the presence of the BTt and BTr tones. If the medium is free (no busy tone detected), the terminal turns on the BTt, sends an RTS packet, and turns off the BTt. As in other protocols, there is a random backoff time if the medium is busy. The destination terminal, upon receipt of an RTS addressed to it, turns on the BTr and waits for the data. Once the BTr tone is sensed, the sender assumes it has successfully acquired the medium. After waiting a short time (for the BTr to propagate) it transmits the data packet. On successful reception of the data packet the destination terminal turns off the BTr tone, completing the conversation. If no data are received, the BTr tone is turned off after a timer expires at the destination terminal.
Fitzek et al. (Fitzek, 2003) proposes a multi-hop MAC protocol based on the IEEE 802.11. A common channel conveys signaling and dedicated channels carry the data traffic and the ACK packets. Figure 19.53 presents the proposed MAC handshake. The first RTS packet is used to contact the destination and assess its willingness to receive data. The sender includes a list of idle dedicated channels, which is used by the destination terminal to select the dedicated channel. It then transmits this information to the sender in a CTS packet. If no suitable dedicated channel is available the handshake ends. After receiving the CTS packet, the sender transmits a PROBE packet on the dedicated channel. The destination terminal uses this packet to test the channel conditions. It then sends a second CTS packet on the common channel informing about the chosen coding/modulation scheme. The sender to confirm the parameters chosen transmits a second RTS packet. Although at a higher complexity cost, the authors claim that the proposed scheme outperforms the original 802.11.
LA-MAC—Load awareness MAC (Chao, 2003). In LA-MAC, the protocol switches between contention- based and contention-free mechanisms depending on the traffic volume. Contention-based mechanisms are best suited to light traffic conditions in which the probability of collision while attempting to gain the medium is small. For heavy traffic a contention-free mechanism allows higher and more evenly distributed throughput. In (Chao, 2003), the IEEE 802.11 DCF is adopted during contention-based periods while contention-free periods use a token passing protocol. The traffic load is measured by the delay packets are experiencing. Each terminal for the packets it has to transmit computes such delay. During a contention- based period, before a terminal transmits its data packet, it checks the packet’s current delay. If the delay is greater than a pre-defined threshold A the terminal creates a token and transmits it attached to the data packet. This indicates to all terminals the start of a contention-free period. Once the delay has fallen below another pre-defined threshold B, the terminal about to transmit removes the token. This indicates the end of the contention-free period and the start of a contention-based period. Threshold A is chosen to be greater than B to give the switching decision some hysteresis.
PCDC—Power controlled dual channel (Muqattash, 2003). In PCDC, the objective is to maintain network connectivity at the lowest possible transmit power. PCDC is a multi-hop protocol that uses the RTS/CTS handshake found in IEEE 802.11 with some modifications. Each terminal is required to keep a list of neighboring terminals and the transmit power needed to reach them. When a packet is received, the list needs to be visited. If the sender is not known an entry is added. Otherwise, the existing entry is updated. In any case, the receiver needs to re-evaluate its connectivity information and confirm that it knows the cheapest way (in a transmit power sense) to reach all terminals that appears in its neighbor list. For instance, for some terminals it might be cheaper to use an intermediate terminal instead of the direct route. At heavy traffic loads there exist enough packets transiting to keep terminals well informed of their neighborhood. For long idle periods terminals are required to broadcast a ”hello” packet periodically for this purpose. PCDC achieves space efficiency and simulations carried by the authors indicate an increase in the network’s throughput.
MAC ReSerVation—MAC-RSV (Fang, 2003). In MAC-RSV, a reservation-based multihop MAC scheme is proposed. The TDMA frame consists of data and signaling slots. Data slots are marked as follows: reserved for transmission (RT), reserved for reception (RR), free for transmission (FT), free for reception (FR), or free for transmission and reception (FTR). The signaling slot is divided in minislots with each minislot further divided in three parts: request, reply, and confirm. A terminal wishing to transmit sends an RTS packet. In the RTS, the sender informs its own identity, the intended receiver’s identity, and the data slots it wishes to reserve. The intended receiver replies with a CTS if any of the requested slots is among its FR or FTR slots. Otherwise, it remains silent. It is possible that the CTS packet accepts reservation of only a subset of the requested slots. Terminals other than the intended receiver replies with a Not CTS (NCTS) if any of the requested slots is among its RR. Any terminal that detects an RTS collision also replies by sending an NCTS. Otherwise, it remains silent. Finally, if the sender successfully receives a CTS it confirms the reservation by sending a confirm packet (CONF). Otherwise, it remains silent. RTS packets are transmitted in the request part of the minislot; CTS and NCTS use the reply part; and CONF packets use the confirm part. Data slots are divided in three parts: receiver beacon (RB), data, and acknowledgement (ACK). A terminal that has a data slot marked RR transmits an RB with the identity of the active data transmitter. In addition, the receiver acknowledges the correct data reception by transmitting an ACK at the end of the data slot. Simulations carried out by the authors indicate that the proposed protocol outperforms the IEEE 802.11 at moderate to heavy traffic loads.
Comments
In (Jurdak, 2004) a set of guidelines is provided that a suitable general-purpose MAC protocol should follow. In particular, it is mentioned that the use of multiple channels to separate control and data is desirable in order to reduce the probability of collisions. The need of flexible channel bandwidth, multiple channels, and the high bandwidth efficiency suggests that CDMA is the optimal choice for channel partition. Multi- hop support is recommended to ensure scalability with flat or clustered topologies depending on the application. In order to favor power efficient terminals, protocols need to be power aware, must control transmission power, and allow for sleep mode. To complete the set of recommendations, the authors include, for the sake of flexibility, short to medium range networks and a sender-initiated approach.