Trade on behalf of binary signals bts
Since the node is not allowed to transmit anything on the channel before its NAV counts down to 0, it will not transmit anything to collide the on-going transmission. In particular, in the IEEE If an intended transmitter does not receive a CTS 62 message or ACK before it times out, it will double its CW value, and repeat the above handshaking process.
If the node succeeds in the intended transmission, it resets its CW to CW min. On the other hand, if the intended transmission is still unsuccessful after a certain number of retrials the associated data 56 packet will be discarded. As a result, transmissions during CFP are collision free. As a result, PCF transmissions have higher priority for channel access than DCF transmissions, and may be used for real-time packets.
However, PCF has not been implemented in any commercial In such a multiple stream model , The first packet in each queue counts down independently of each other. However, if the counters for more than one queue count down to 0 at the same time, a virtual collision occurs. The queue with the highest priority then has the right to send the data 56 packet or the associated RTS 60 message, while the other queue s backoffs and repeats the countdown process.
Third, the rule for countdown is also different in EDCF: In particular, higher-priority traffic category can increase the CW by a persistent factor smaller than 2. This allows the CW to be increased at a slower rate, thus reducing the delay and increasing the transmission rate of the traffic category as compared to DCF.
We refer readers to  for further details. In particular, the collision problems constitute a major issue that is inevitable in ad hoc networks and will degrade the throughput and QoS capability of multihop networks if they are not carefully handled.
In addition to significant reduction in network throughput, this phenomenon has two important implications to QoS provisioning in ad hoc network and multihop wireless LANs. The first implication is that QoS cannot be guaranteed since packets with reservations may still be collided with high probability during the reserved slots.
The second implication is that the contention window CW will be increased exponentially for unlock nodes that experience a number of collisions, which in turns leads to unbounded delays and lower throughput for the nodes. As a result, the collision problem also has significant implication to fairness in such multihop wireless networks since nodes that experience a number of collisions will be treated unfairly.
The interference problems constitute a major reason for collision rates in multihop networks to be high. More precisely, according to the current technologies, the interference range is typically larger than the transmission range. When there are multiple interfering sources, the additive interference will cause collisions at even larger distance.
This, however, will introduce a new form of the exposed terminal problem in ad hoc networks. Moreover, a new form of the hidden terminal problem will exist when there are obstructions blocking the signals from senders so that CSMA with sensitive carrier sensing hardware does not work well in multihop networks. The second major issue is that the energy and spatial reuse efficiency of IEEE The third issue is the well known exposed terminal problem , when IEEE The fourth major issue is that IEEE A reason is that a high-priority packet may be blocked by a nearby low-priority packet, and then blocked by another low-priority packet on the other side, and so on.
With a nonnegligible probability, such a situation can go on for a long time when the traffic is heavy and the network is dense. As a result, high-priority packets may still experience unacceptable delay. This problem cannot be solved by the current version of IEEE Moreover, a wireless device can transmit at different power levels according to the physical distance between the transmitter-receiver pair as well as the noise and interference level.
Note that the latter is a typical capability, rather than an exception, when the IEEE In fact, the majority of The reason is that carrier for the low-power transmission cannot be detected by wireless stations at moderate distance, so those wireless stations may transmit at a higher power and collide the low-power transmission.
This is the CSMA form of the heterogeneous hidden terminal problem. If the hardware for carrier sensing is made very sensitive so that a low-power transmission can be detected by wireless stations at moderate distance to mitigate or solve the aforementioned heterogeneous hidden terminal problem, then the exposed terminal problem  will deteriorate considerably. More precisely, the carrier for a transmission at high power will be detected by all wireless stations within a very large area i.
All these wireless stations will then be blocked from transmissions unnecessarily, significantly reducing the network throughput in multihop wireless networking environments. We refer to this problem as the CSMA form of the heterogeneous exposed terminal problem. As a result, the CTS 62 message of the intended receiver will either block some potential transmitters unnecessarily, fail to block some potential transmitters to protect its reception, or both, no matter how the range for the CTS 62 message is chosen.
For example, if the transmission radius of a wireless station is relatively small, and a wireless station only sends its CTS 62 message to wireless stations within a similar radius, then it is hidden from wireless stations outside the range see FIG.
Since these outside wireless stations do not receive CTS 62 from the on-going receiver, they will interfere with its reception if they decide to transmit data 56 packets with larger transmission radii. The intended transmitter A cannot sense the transmission of the on-going transmitter C or receives the CTS 62 message from the on-going receiver D. If the intended transmitter A sends its data 56 packet to the intended receiver B, the reception at the on-going receiver D will be collided.
This can be viewed as a new form of the hidden terminal problem that is unique in heterogeneous ad hoc networks. If the CTS 62 message of the on-going receiver D is sent to all wireless stations within the maximum transmission range, the intended transmitter A can be blocked successfully.
However, the intended transmitter E is also blocked from transmission to its intended receiver F unnecessarily, even though its transmission will not collide with the reception at the on-going receiver D. This can be viewed as a new form of the exposed terminal problem that is unique in heterogeneous ad hoc networks.
Most power-controlled MAC protocols reported in the literature thus far for ad hoc networks , , , , ,  require all transmitters and receivers to send their RTS 60 and CTS 62 messages at the maximum power, and transmit their associated data 56 packets and ACK 58 messages at the minimum power possible. Even if these exposed wireless stations want to send data 56 packets with a smaller transmission radius that will not interfere with the reception of the sender D of the CTS 62 message, they are still blocked unnecessarily.
The reason is that only wireless stations within a portion of the transmission range of a CSMA data 56 packet suffer from the conventional exposed terminal problem, while the transmission radius of a CTS 62 message transmitted at maximum power may be considerably larger that of its associated data 56 packet, and wireless stations within most part of the transmission range of the CTS 62 message will suffer from the heterogeneous exposed terminal problem.
Thus, an extension to IEEE In , , , , several power controlled MAC protocols were proposed to conserve energy consumption.
In all these protocols, which are based on the so-called BASIC scheme , require all transmitters and receivers to send their RTS 60 packets and CTS 62 messages, respectively, at the maximum transmission power, and send their associated data 56 packets and ACK 58 messages at the minimum power possible.
In , , a refinement were made to the BASIC scheme by more carefully selecting the transmission power levels of data 56 packets and ACK 58 messages according to their sizes. Although the error rates and resultant retransmissions can be reduced, the improvement in throughput is still limited.
Since these MAC protocols only attempt to reduce their power consumption, rather than utilizing smaller transmission power to increase spatial reuse and thus throughput, we categorize them as power controlled MAC protocols. In a single-hop wireless LAN, a node can employ a sufficiently sensitive CSMA hardware to make sure that it can hear transmissions from all other nodes. In this way, no hidden terminals exist as long as there are no obstacles and collisions will not be caused even if the interference radius is considerably larger than the transmission radius.
However, this is not the case for ad hoc networks and multihop wireless LANs. However, such an assumption does not hold in many ad hoc networking environments when IEEE Instead, the interference radius is typically larger than the associated transmission radius  e. In such an environment, a node A that does not receive an CTS 62 message from a node B may transmit a packet to collide with the reception at node B, since node B may be within the interference range of node A, while node A is outside the transmission range of node B.
Note that IEEE If we assume that the sensing radius is larger than, or equal to, the sum of the transmission radius and associated interference radius, then the hidden terminal part of IHET can be solved. However, the exposed terminal part of IHET will deteriorate in that many nearby nodes especially those near the transmitter's side will be blocked unnecessarily. As a result, no matter whether we assume IEEE More precisely, the authors proposed to augment a mechanism to flood CTS 62 messages of wireless devices with lower power capabilities.
They assume a device can only transmit at a constant power and fixed radius for all its lifetime , so the proposed scheme does not belong to power-controlled MAC schemes or variable-radius MAC schemes.
However, their own simulations results in  show that the proposed modification actually reduces network throughput due to the increased overhead in relaying CTS 62 messages, even when an enhanced version with precise GPS information is used. To the best of our knowledge, PCMA is the only previous protocol reported in the literature thus far that can solve IHET and both the hidden and exposed parts of the heterogeneous terminal problem.
In PCMA , a device senses the channel during its reception of data 56 packets, measure the current noise and interference level at its location, and then calculate the additional interference it can tolerate. It will then send its busy tone at a certain power level, which is a function of the additional interference it can tolerate. A device that intends to transmit a data 56 packet has to gather the busy tone signals sent by all nearby on-going receivers, and determine the maximum power level it is allowed to transmit according to the strengths of the busy tone signals it just received.
Although some ideas proposed in  are novel and interesting and PCMA can be classified as a power-controlled variable-radius MAC protocol, a main drawback of this protocol is that each device requires two transceivers. More precisely, one is needed for reception of data 56 packets, while the other is needed to measure the channel noise and interference and to transmit busy tone during its data 56 packet reception.
As a result, the hardware cost and power consumption of PCMA will be increased. Moreover, the aforementioned capability required by PCMA-based mobile devices may be expensive, if not impossible, to implement.
Prioritization-based techniques  as DiffServ  and reservation-based techniques  as IntServ  are the two main paradigms for provisioning QoS in practice or in the literature. Since reservations are very difficult to maintain in mobile ad hoc networks, and it is expensive, if not impossible, to police and enforce reservations in such networking environments, we focus on prioritization-based techniques in this application. Although these mechanisms can differentiate the delays between different traffic classes to a certain degree in single-hop wireless LANs, they are not adequate in a multihop environment such as ad hoc networks and multihop wireless LANs.
The reason is that in a single-hop wireless LAN, an However, this is not guaranteed in ad hoc networks or multihop wireless LANs. For example, in such networking environments, an The receiver D of the lower-priority transmitter C may then continue to block the high-priority intended transmitter A. With a nonnegligible probability, such a situation can go on for a long time for some high-priority nodes when the traffic is heavy and the network is dense i.
So high-priority nodes may still experience large delay in IEEE This problem cannot be solved by IEEE In order for killer real-time applications such as voice over ad hoc networks and multihop WLAN i. Accordingly, besides the objects and advantages of the flexible closures described in my above patent, several objects and advantages of the present invention are:. An EIM-based node only needs a single transceiver, and typically without requiring additional expensive or specialized hardware besides the standard hardware required by an IEEE However, multiple transceivers may also be employed to enhance the performance.
In this invention, a method of interference management to be referred to as the evolvable interference management EIM method, and the associated techniques and mechanisms are disclosed. In other words, they may compensate, support, or enhance each other for the purpose of interference control.
The presented method may be embodied in a way that different combinations of the presented or new processes, mechanisms, and techniques can coexist. However, a relatively inflexible embodiment of the presented method i. The triggered-CTS mechanism is employed. The transmission power level for the second CTS message is increased since the tolerable interference level is reduced so that the coverage range for the second CTS message has to be increased.
The detached dialogue approach is not employed. In this example, the control channel and data channel are separated, but a node only has a single transceiver so it cannot receive and transmit at the same time.
If node C intends to send a packet to D, it sends an RTS message to all the nodes within its interference or protection area. The scheduled receiver B then replies to C with an TPO or called OTS message via the control channel, if the request conflicts with its scheduled reception. Therefore, the reception scheduled for receiver B will not be collided even if C does not receive the DTR message from B.
The power level for the transmission from node B to node C or other transmissions toward nodes close to the BS should be raised. A nearby node can count the declaration pulses it receives to determine the maximum power level it can transmit without colliding the data packet reception at node B. For example, node C receives all 3 declaration pulses, so it cannot transmit during a packet slot overlapping with the one specified in the CTS message.
Node G is outside the protection area from node B, and can transmit data packets at any allowable power level e. Note that no specialized hardware is required by these nodes e. The RTS message specifies the dialogue deadline as 2 time units, the relative reception starting time as 3 time units, and the relative reception ending time as 4.
In what follows, various aspects of the invention will be described in greater detail in connection with a number of exemplary embodiments.
To facilitate better understanding of the invention, several components of the invention are described in terms of sequences of actions to be performed by elements in a plurality of communication devices.
In each of the presented embodiments, the various actions could be performed by specialized circuits, by program instructions executed on one or more processors, or by a combination of both. We generally refer to such an element as a node. Such combinations can be adaptive to the conditions of the environments, and changed through time based on optimization, heuristics, or other pollices.
Moreover, different nodes can utilize different combinations due to their limitations, available resources, preferences, current status, and the respective environment conditions where they are located. As a result, the various aspects of the invention may be embodied in many different forms, and all such forms are contemplated to be within the scope of the invention. However, to facilitate coexistence among a plurality of nodes, certain rules apply to restrict the permissible combinations, which can be very strict e.
Such rules are typically specified in the protocols, standards, regulations, or other policies the nodes follow, and can be further coordinated in a distributed manner as the efforts of a group of nearby nodes, in a centralized manner coordinated by clusterheads, elected coordinators, or a unit governing a wider range of other nodes. In this patent we disclose the evolvable interference management EIM approach for next-generation wireless networks. EIM is particularly designed to tackle the interference problems in multihop wireless networking environments.
EIM can solve various unique problems in multi-hop wireless networks including ad hoc networks, multihop WLANs, and sensor networks, but may also be applied to conventional single-hop networks. EIM is also developed with emerging and potential future technologies in mind, by developing a flexible, extensible, and consistent framework that can incorporate various possible advancements in the future and tolerate the co-existing of new and legacy deevices.
Some of these techniques and mechanisms can support or enhance each other, or compensate each other, while some of them can be alternatives to each other. However, typically, all of them do not need to be employed at the same time. Devices based on different but consistent subschemes e. In this section, we present several scenarios and promising EIM techniques along possible evolutionary paths for future wireless MAC protocols. We start with techniques that can work in combination with, or on top of, IEEE A Simple Solution to Interference Problems.
However, sender-based S-CSMA cannot completely eliminate the hidden terminal problem in such networking environments. One such scenario is introduced when obstructions block the signals between some active transmitters, resulting in hidden terminals.
Another kind of hidden terminals for CSMA exist in environments where the path loss is high. When power control is employed, this problem becomes even more severe since low-power transmissions tend to be hidden from far away high-power transmitters so that the latter may collide the former with a high probability. Note that S-CSMA is not effective in environements with high path loss since the signals of a transmitter may not be picked up by potential interferes far away from it but closer to its receiver.
However, there is a limit on the sensitivity of the sensing hardware, and such adaptive mechanism may require higher complexity and overhead. The reason is that a CTS message may not be decodable within a large portion of the associated maximum interfering range when the signal received at the associated receiver is not high.
As a result, irrelevant transmitters or receivers within that portion of the region may collide the associated reception by transmitting data packets or control messages at maximum or sufficiently high power levels.
Similarly, if an RTS message is transmitted at the same power level as the associated data packet, it is also undecodable within a large portion of the maximum interfered range unless appropriate accompanying mechanisms are employed. One way to solve this problem is to try to transmit the RTS and CTS meesages to most nodes within the associated protection ranges.
A simple but less efficient remedy in terms of radio utilization is to employ a relatively conservative algorithm for backoff control after collisions.
The relatively conservative algorithm can then be applied to such vulnerable transmitters or transmissions only, rather than to all colliding nodes in the network. Note that when power control is employed, whether a pair of nodes are hidden terminals depend on the associated transmission power levels and the received signal strengths.
Note also that hidden terminals may be unidirectional, rather than bidirectional or mutual, when one of them is transmitting at a higher power level. We categorize this accompanying mechanism to the reactive hidden terminal resolution paradigm, which typically requires additional mechanisms at a level higher than MAC, and may or may not require some simple changes at the MAC level.
Another important issue is that when the path loss is high, sensitive CSMA is not effective since the signals of a transmitter may not be picled up by potential interferes to its receiver. This problem can be mitigated by using ultra sensitive carrier sensing hardware, and adaptively adjusting the sensing threshold according to the environments. The prohibition-based patching approach PPA can increase spatial reuse at the expense of higher control overhead.
Under the reactive patching paradigm, nodes involved in collisions will go through a procedure to determine whether they are hidden terminals to each other. A threshold can be set for the procedure to be invoked.
When another higher threshold is reached, a regional detection procedure can be invoked to detect active hidden terminals within that region. These nodes will then avoid transmitting during overlapping times at forbidden power levels. As a result, collisions caused by interference-range hidden terminals can be prevented through a scheduling-based patch without modifying the MAC or PHY layer. Under the proactive patching paradigm, the regional or global hidden terminal detection procedure is provoked when appropriate e.
Then mutually exclusive schedules can be set up to prevent collisions at the first place, at the expense of higher control overhead and more complex schedules as compared to the reactive patching paradigm. The proposed patching approach also has other subclasses such as the encouragement-based patching approach EPA. Note that PPA, EPA, and other subclasses may work together for different pairs or groups of nodes and maybe in a hierarchical manner, rather than being mutually exclusive.
The aforementioned approaches rely on individual hidden nodes to negotiate for mutually exclusive schedules and are thus referred to as node-oriented PPA. The proactive patching paradigm is followed where a hidden terminal detection procedure is provoked whenever appropriate. Note that for this kind of groups, called harmonic groups, the members of a group typically cannot all transmit at the same time; otherwise a lot of collisions will be resulted.
Note also that a node can be affiliated with multiple groups, and a node that does not have any hidden terminals may be affiliated with every such group if so desired. Group-oriented PPA has many other important applications.
In particular, it can enable differentiated group channel by grouping transmissions of data packets or control messages at the same power range. Such a strategy can be expanded to assign nodes, transmissions, receptions, competitions, etc. PPA may be applied to achieve various other important objectives or to fix other problems. In what follows we use link-oriented PPA as an example to explain how it works. These links, if used, may consume too much resources e.
They may also considerably degrade the performance or quality of TCP-based or real-time applications, or even prevent them from working properly. As a result, it may be beneficial to prohibit these problematic links from being used in some applicatins or environments. This can be implemented in a variety of ways. Also, the protocol stack may be modified so that such problematic links are flagged and disabled for any uses or when associated with certain traffic categories or applications.
Link-oriented PPA can be applied to transform conventional routing protocols into protocols with desirable new features without having to modify the protocols or their implementations.
Consider an ad hoc network using a minimum-hop-based routing protocol such as AODV  or an evolved version due to its popularity. However, in some environments or certain regions a power-controlled routing protocol is actually favored due to energy or throughput concerns. A variety of other PPA variants are also possible, and can achieve different objectives. The details are omitted in this book chapter. We refer to this type of strategies as the prohibition-based patch approach PPA.
When power control is employed, sensitive CSMA without interference engineering or differentiated multichannel will not work for transmissions at very-low power levels. The reason is that such low-power signals e. Also, when a data packet is not large, the control overhead in terms of the consumed energy may not be acceptable. Moreover, the spatial reuse cannot be increased when CTS messages are transmitted at the maximum power level as power-controlled protocols based on the Basic scheme .
To solve these problems for reduced collision rate and increased throughput, we may emply interference engineering, which does not require much change to the MAC protocol of IEEE Power engineering can enable interference engineering, which appropriately control the interference generated for other nearby nodes and thus changing the maximum interfered range for a given interference threshold or the interference tolerable at the receiver and thus changing the maximum interfering range for a given interference threshold.
For example, by increasing the transmission power beyond the minimum required power level the required coverage area for the CTS message of the receiver can be considerably reduced. As a result, for transmissions that originally require lower power levels, the collision rate can be significantly reduced due to the appropriate protection from their CTS messages that reach the appropriate ranges.
Other advantages including that the power required and the blocked area for other nodes to transmit or receive can be better balanced, reaching a considerably better spatial reuse and energy consumption. A Robust Signaling Approach.
In the previous subsections we introduced several EIM techniques that can mitigate the interference problems without or with only minor changes to the IEEE In this and the next subsections, we introduce several powerful EIM techniques for tackling the interference, power-control, and QoS problems. In the following section, we will provide an example implementing several techniques and illustrating how they may work together. The central idea of CP is simple yet powerful.
When the resultant protocol is appropriately designd, collision-free transmissions of control messages and data packets can be virtually guaranteed. If centralized control is feasible e. Adoption of these mechanisms is relatively straightforward and the details are omitted here. However, when fully distributed MAC protocols are desired as expected in typical networking environments, the protocol design becomes more challenging.
In what follows, we briefly present such a fully distributed mechanism based on distributed multihop binary countdown DMBC. More details for DMBC and the prevention of collisions due to hidden terminals can be presented in Section??. To simplify the protocol description, we first assume that all competing nodes are synchronized and start competition with the same bit slot, and can sense the status of every bit-slots when they are not transmitting.
Such a surviving node whose i th bit of its CN is 1 transmits a buzz signal to all the nodes within an appropriate range e. A surviving node whose first bit is 0 keeps silent and senses whether there is any buzz signal during bit-slot i. If it finds that the bit-slot i is not idle, then it loses the competition; otherwise, it survives and remains in the competition.
More precisely, when the ID numbers of nodes are unique among all their possible competitors and all competitors within a radius of R can hear each other, there can be at most one winner within a radius of R from the winner. As a result, none of the control messages to be transmitted will interfere with each other at any nodes within their transmission ranges.
Various other ways to utilize prohibitive signals for competition and collision prevention are possible. Transmitters are prohibited by receivers with higher competition numbers CNs through prohibition signals in receiver prohibition slots; while receivers are prohibited by transmitters with higher CNs through prohibition signals in transmitter prohibition slots.
Transmitters sense prohibition signals in receiver prohibition slots in order to know whether its intended receiver survived the competition; while receivers also sense prohibition signals in transmitter prohibition slots in order to know whether its intended transmitter survived the competition.
The thresholds for sensing can change with slots to improve the performance. When the transmission is in multicast or broadcast mode, the transmitter can also act on behalf of its receivers for sending all the prohibitive signals, while using different power levels in the two slots corresponding to the same bit in order to facilitate power control.
Such an approach also works for the unicast mode, except that the efficiency for this combination can be improved. In this subsection we transform the interference, which was considered as among the worst enemies for wireless communications, to be a novel and power tool for wireless communications. Intermittent interference signals, possibly transmitted at appropriate power levels with other appropriate characteristics can be used to convey some information at the MAC layer even though they may not be decoded using the PHY-layer modulation hardware.
This approach may be considerably more robust as compared with other approaches when the environment is hostile e. However, whether this approach will be actually be deployed depends heavily on its performance and cost relative to its competitors, as well as whether other competing technologies such as spread-spectrum-based MAC techniques can mature in time and be implemented at low hardware cost. A receiver or transmitter can also periodically ,  or semi-periodically e.
As compared to conventional busy tone [? Moreover, this technique only requires a single transceiver per node though dual transceivers can improve the performance. In this subsection, we introduce several powerful EIM techniques for tackling the interference, power-control, and QoS problems.
To reduce control overhead for lower-power transmissions, we can limit the range of power levels that are allowed to be used for each channel in a multichannel environment. We refer to this strategy as the differentiated multichannel scheme. Moreover, this scheme does not rely on interference engineering or high-power CTS messages, considering reducing the control overhead. This scheme can be combined with TDMA-like phasing, group scheduling or patching.
Moreover, all the control messages and the associated data packet can be detached from each other. This figure illustrate the detached dialogue approach with a single shared channel for both control messages and data packets. It may also become a powerful tool, for example, for effectively differentiating service quality and quantity or providing QoS guarantees in a distributed manner. More details and performance evaluation results can be found in , .
This increases the robustness of the network, connectivity, quality of TCP or real-time applications, and so on. Combined with power control, spread spectrum techniques can enable effective interference engineering. There are also many other advantages for incorporating spread spectrum techniques at the MAC layer.
For example, power control can be efficiently supported by grouping transmissions with similar power levels into a code channel as a special case of the differentiated multichannel scheme. However, spread spectrum-based techniques and detached dialogues-based techniques are not mutually exclusive. They can in fact work together and combine into an effective and strong scheme for multiple access in multihop wireless networks.
Higher-layer Techniques to Interference Prevention. Routing-based techniques, such as radius-oriented ad hoc routing ROAR [? This strategy is in spirit similar to the link-oriented prohibition-based patching approach, but is different in that no additional PPA-like coordination or special operations like iptable are required, and problematic links are avoided naturally using appropriate routing metrics and policy.
Similarly, mobile wireless-MPLS can also avoid problematic links and solve the interference problems naturally. A difference from the routing-based techniques is that desirable links and routing paths are maintained through wireless LSP to maintain based on local label without having to rely on IP addresses or global IDs.
Clustering-based techniques that parititon the space for dynamic and adaptive TDMA or negotiate between nearby clusterheads for polling schedules or group i. Such an approach is particularly desirable if clustering is also in place for other purposes such as routing or maintaining a hierarchical architecture. Various other scenarios or evolutionary paths are also possible. We envision that the wireless MAC protocols will continue evolveing with the maturity and emergence of technologies and user needs.
In such environments, a consistent framework that allows various MAC and other techniques to coexist some time along the road will be desirable, and will be able to achieve better performance in the long run as compared to MAC protocols optiomized for each stage of technologies without such visions in mind and then extended upon necessarity, as is done in current practice based on the conventional wisdom.
We will also provide more details concerning the innovative detached dialogues approach. Detached dialogues, spread spectrum techniques, and individualized error control mechanisms are employed on a recommended, but optional, basis. As shown in FIG. When combined with appropriate accompanying mechanisms, detached dialogues can solve various problems including all the issues identified in Section??.
In GAP, an RTS message either implies the use of a default dialogue deadline or specifies a desired dialogue deadline, where the specified relative dialogue deadline DD time T DD is the maximum time allowed for the CTS message from the intended receiver to be received completely by the intended transmitter since the last bit of the RTS message is received by the intended receiver.
The RTS message requests for a data packet duration starting at packet lag PL time T PL after the dialogue deadline plus a turnaround time for the intended transmitter. Note that relative times are specified so that synchronization is not required and the number of bits required for such specifications is reduced as compared to the use of absolute times.
Moreover, the required duration for the receiver to be available and the duration for other third-party nodes to be interfered can then be specified with exactly the same relative time duration.
When the exact propagation delay between the transmitter-receiver pair is known, the exact value is used for T UP ; otherwise, the maximum propagation delay for the maximum coverage radius of the network or for the maximum transmission radius at the intended transmission power level is used for T UP. When the extension flag of a control message is set to 1, a larger-size format for the message will be used.
The message size may be further extended by setting another extension flag within the extended format, and so on. As a result, when a default value is used, smaller control message formats are used, and appropriate extended formats are used only when necessary.
In particular, when the CTS message is allowed to be replied till the last moment, then only one relative time i. In this way, the control channel overhead can be reduced.
The rational for detaching these control messages and the associated data packet are five-folds. First, detached CTS messages allow the intended receivers to reply at a later time if they are available during the requested duration but are currently not allowed to reply with a CTS message.
Second, the flexibility resulted from optionally detached data packets naturally avoids the exposed terminal problem from existing. Similarly, it also supports efficient power-controlled transmissions and interference-aware medium access, which considerably improves radio channel utilization .
This enables effective and efficient MAC-layer supports for differentiated service DiffServ  and fairness , . Fourth, by detaching acknowledgement messages, multicasting, power control and the exposed terminal problem can be supported or resolved without compromising reliability.
Fifth, the third-party opinion TPO mechanism is enabled by detached dialogues without requiring dual transceivers per node or dual channels. It can in turn be used to enable the preemptive mechanism. Such situations may occur in future high-speed wireless networks with small packets or in wireless networks with large coverage ranges such as satellite networks and future mobile wireless MANs. There are also various other advantages that may be achieved through the proposed detached dialogues.
Group Activation, Scheduling, and Competition. In the timing diagram example illustrated in FIG. Node A first employs a backoff control mechanism to count down to 0, and then gains the right to send the GA message.
Note that a node may also transmit the GA message after it successfully scheduled for a transmissionor reception. However, in many cases, we prefer to have a larger postponed access space PAS between GA and the associated coordinated starting time t 1. In a different scenario, the GA message may be initiated by nodes other than the first transmitter A, while node A may schedule for a coordinated time t 1 , t 2 or a subset of it, a superset of it, or simply an overlapping period of time, depending on the policy after receiving such a GA message.
Although flooding is most robust, its overhead may not be tolerable. Alternatively, a spanning tree can be used to execute the required geocast relaying. This is especially desirable when such spanning trees have already been made available for higher layer functions such as routing or clustering. Note, however, that it is not mandatory for a node belonging to a group to schedule around the recommended time period.
Note also that backoff time equal to 0 is allowed, especially when an optional prohibition-based competition mechanism is employed before the transmission of the GA message.
Note that the additive effect of multiple interfering signals is not linear so that some accompanying mechanisms or precautions need to be employed or made for the estimation.
Note also that typically the NAV is only set for the scheduled period possibly with some extension as safe margins for better protection. In some scenarios, an additional third-party opinion TPO control message may be sent. For example, consider a nearby irrelevant intended transmitter C that requests using an RTS message for a transmission duration overlapping with the scheduled transmission from nodes A to B, at a power level that will collide with the scheduled reception.
Node C will then reschedule the transmission or lower the requested transmission power level. Some additional CTS messages may follow the first CTS message to update important information such as the new tolerable interference level.
Also, after a number of unsuccessful RTS and CTS messages for the same packet, the handshaking may back off or be aborted. By allowing such PAS to be relatively large, various advantages can be achieved, such as strong service differentiation capability, better supports for power control and interference-aware multiple access, as well as better scheduling. Such larger PAS also enables the group action approach to more efficiently coordinate many group members e. Moreover, enabled by detached dialogues, TPO or OTS messages can be sent and received appropriately even when a node only has a single transceiver.
In GAP, control messages can be preceded with a prohibition-based competition phase. Such a mechanism when employed can reduce the collision rate of the associated control messages, and in turn reduce the collision rate of the data packets and bursts. Prohibition-based collision prevention can take on many different forms and formats for the competition phase.
Several examples are presented in FIGS. If a node receives a prohibiting signal before its own position for transmitting the prohibiting signal in a slot, it loses the competition. Candidate winners a survivor that survived all prohibition slots will transmit a declaration signal in the declaration slot.
When there are mutually hidden terminals, there is a good chance that other nodes will detect multiple declaration signals that are not likely to be from the same source according to certain criteria such as separation in time and the received signal strength.
These nodes will then send an OTS signal to block the transmissions so that the candidates fail to become winners. Mutually hidden terminal detectors which can be the receiver s or some irrelevant nodes within appropriate ranges send an OTS signal in the HTD slot to block their transmissions.
Various other ways to utilize prohibitive signals to avoid collisions are possible. For example, RTS, CTS, TPO, OTS messages, busy tone, and other messages and information may be replaced or conveyed by this kind of coded intermittent signaling when those control messages do not work well or are not supported. In ISS, the acknowledgements are not necessarily made on the per-packet basis.
Instead, during the error control phase, negative acknowledgement NAK -based implicit acknowledgement mechanism is employed in combination with other appropriate acknowledgement mechanisms, such as group acknowledgement passive acknowledgement, and group-coordinated acknowledgement based on the group action approach.
A large data packet can be segmented, with each segment accompanied with an error control code such as CRC, possible also with an error correcting codde. Other techniques from the previous section may also be employed. In particular, spread spectrum and power-control should be employed, if available, to better balance the resources consumed by different messages. Moreover, they may enable the control messages be transmitted to sufficiently large range to resolve the interference-range problems.
There are lots of alternative embodiments possible. For examples, we can use single channel, dual channel, 3-channel or multichannel for the control and data channels with lots of different combinations. Also, prohibition mechanisms and detached dialogues can be optional or removed. The network can be synchronized or asynchronous, and so on. To reduce the overhead for prohibition-based competition, the group action may be emplyed.
For example, in FIG. Other group members may rebraodcast such a group activation message, possibly with modifications for the timing information etc. Group members that have something to transmit can then compete at the same time if so desired using the same group competition number CN see FIG.
In the following sections, we present more possible embodiments and more details for the invention. In the following sections, more details or aspects for the description of the invention and more preferred or alternative procedures for embodiments of various phases, mechanisms, or aspects of the invention will be presented.
In the embodiment described herein, we consider a MAC protocol followed by all nodes in a plurality of wireless communication devices. For simplicity, this exemplary protocol is relatively restricted in terms of the flexibility to optionally use an optional mechanism. In some scenarios, an additional third-party opinion TPO control message may be sent by the receiver to a nearby irrelevant intended transmitter C if node C used a sender information SI 52 control message to request for a transmission duration at a power level that will collide with the scheduled reception at node B.
Some additional receiver information RI 54 control message may also be added to update the sender information such as tolerable interference level. Also, unsuccessful handshaking may be aborted. Node A first employs a backoff control mechanism to count down to 0, and then gains the right to send the GA 50 message. Note that a node may also transmit the GA 50 message after it successfully scheduled for a transmission or reception.
However, in many cases, we prefer to have a larger postponed access space PAS between GA 50 and the associated coordinated starting time t 1. In a different scenario, the GA 50 message may be transmitted by nodes other than A, while node A may schedule for a coordinated time t 1 , t 2 or a subset of it, a superset of it, or simply an overlapping period of time, depending on the policy after receiving such a GA 50 message.
Note that it is not mandatory for a node belonging to a group to schedule around the recommended time period. Note also that backoff time equal to 0 is allowed, especially when an optional prohibition-based competition mechanism is employed before the transmission of the GA 50 message. By allowing such postponed access space PAS to be relatively large, various advantages can be achieved, such as strong service differentiation capability, better supports for power control and solving interference problems, as well as better scheduling.
Our approach allowing detached control messages and the associated data 56 packet, burst, or its fragments is referred to as the detached dialogues approach DDA. An embodiment of DDA will be presented in the following subsection. Such a mechanism can reduce the collision rate of the associated control messages, and in turn reduce the collision rate of the data 56 packets and bursts.
Several examples and embodiments are presented in FIGS. The detached dialogues can considerably improve the spatial reuse when the prohibitive areas are considerably larger than the interference areas for data 56 packets and bursts. If a node receives a prohibiting signal before its own position for transmitting the prohibiting signal, it loses the competition. Candidate winners that survived all prohibition slots will transmit a declaration signal in the declaration slot.
When there are mutually hidden terminals, there is a good chance that that other nodes will detect multiple declaration signals that are not likely to be from the same source. These nodes will then send a signal to block the transmissions so that the candidates fail to become winners. The upper figure represent a scenario when there is only one candidate and it successfully becomes a winner and gain the right for transmissions.
The lower figure represent a scenario when there are two candidate within their prohibitive areas, and some other nodes send a signal in the HTD slot to block their transmissions. Transmitters are prohibited by receivers with higher competition numbers CNs through prohibition signals in receiver prohibition slots; while receivers are prohibited by transmitters with higher competition numbers CNs through prohibition signals in transmitter prohibition slots.
Software defined networking SDN is an emerging virtualization technology, which supports programmable interfaces to provide flexibility and agility on the network control management [ 1 — 3 ]. It basically consists of a number of network nodes such as switches, virtual switches, routers, and firewalls, which are automated, controlled, and reprogrammed through software commands.
Open source software such as Open Flow [ 4 ] can be used to dynamically reconfigure network elements, through an SDN controller which can handle multiple network switches at a time.
An SDN-based architecture allows dynamic and flexible network operations by decoupling the network control plane from the data plane , leveraging standard protocols which enable remote management and operation. The SDN controller can run on a commodity hardware and gives logically centralized control towards multiple switches.
This enables accurate monitoring and control of traffic load within the network and is also expected to minimize operational cost, while improving load balancing and data traffic handling at the edge, through the use of generic hardware [ 5 ]. Another virtualization technology is NFV, which has recently emerged to virtualize the EPC network functions and move them from proprietary to commodity hardware platforms, as the use of specialized hardware devices has been one of the limiting factors towards mobile evolution and the fast deployment of new services within the mobile space [ 6 ].
Network functions may be firewalls, domain name servers DNS , network address translation NAT services, intrusion detection systems, caching services, and so forth. These functions, which are of prime importance for the accurate operation of any network, are migrated into software and ran on top of general purpose servers. SDN tries to achieve a centralized control approach on switching and routing elements, thus allowing programmability of the network, while NFV moves NFs out of dedicated hardware into software that is imported into general purpose hardware.
While these technologies do not possess any intrinsic cognition, they will give rise to more flexible networks, where resources could be controlled and combined in a flexible manner. This is expected to facilitate network management and to make it more efficient, moving the intelligence that is required to manage network resources, such as load balancing, intrusion detection algorithms, and firewalls into NFs. As dictated by mobile edge computing MEC , an emerging architectural paradigm for the design and implementation of communication networks through NFV, the virtualized network functions VNFs can then be deployed at the BS , that is, at a MEC platform colocated with the BS, or at an aggregation point a central point that manages a set of BSs located close to each other.
MEC effectively moves the network intelligence towards the network edge: As a last consideration, 5G technology is currently adopting a so-called network densification approach, which involves the deployment of a large number of base stations BSs , to increase the network coverage and provide higher throughput to the users.
This however results in higher energy consumption, which is expected to considerably contribute to carbon emissions into the atmosphere [ 8 , 9 ]. In order to minimize the carbon footprint of 5G MNs, we advocate for the integration of energy harvesting EH into future base stations especially small cells. Besides helping to minimize the operational expenses OPEX , in terms of annual electricity bills, the use of renewable energy will help extend network coverage to areas where there is insufficient electricity, or to assist during the case of a natural disaster scenario, where the conventional electricity grid may become unavailable.
At the same time, the deployment of EH technology batteries, solar cells, etc. Nevertheless, current trends in battery and solar module costs are promising and suggest that in the future this equipment will be cheap enough. Further discussion and results on these aspects can be found in [ 10 ]. We stress that energy efficiency is a key consideration in future networks and can be addressed as follows.
First, the network procedures have to be streamlined and carefully orchestrated, and here is where virtualization technology also entailing new architectural designs will play a crucial role. This will allow for a more energy efficient network operation. Second, a modern and flexible management can be combined with EH technology to reduce the carbon footprint of communication networks.
In this paper, current softwarization technologies, architectures, and trends are reviewed with a special focus on energy efficiency. The rest of the paper is structured as follows.
In Section 2 we discuss the existing EPC architectural proposals, analyzing the following virtualization techniques: A grouping EPC functional entities Section 2. In Section 3 , we discuss the use of machine learning, data mining, and context-awareness within softwarized 5G networks. In Section 4 , we outline some challenges and open issues related to the EPC and MEC proposals in the state of the art and, lastly, in Section 5 we provide some final considerations.
The EPC network consists of a number of NFs, all interconnected through an Internet protocol IP infrastructure to provide packet data services to the access networks. This includes the packet data network gateway PGW , which is responsible for IP address allocation for the UEs, as well as for QoS enforcement and flow-based charging, according to rules from the policy control and charging rules function PCRF.
It is also responsible for the filtering of downlink user IP packets into different QoS-based bearers. It also retains the information about the bearers when the UE is in the idle state known as EPS connection management and temporarily buffers downlink data while the mobility management entity MME initiates paging of the UE to reestablish the bearers.
The non-access stratum NAS signaling terminates at the MME, which is also responsible for the generation and allocation of temporary identities to the UEs. The PCRF is responsible for policy control decision making, as well as for controlling the flow-based charging functionalities in the policy control enforcement function PCEF , which resides in the PGW.
Traditional EPC networks are complex and rather inflexible, use proprietary costly equipment, and incur high signaling overhead. To overcome these limitations, an architectural evolution that will permit dynamically scaling the EPC network functions while adapting to real world needs is in order. This can be achieved through the use of softwarization techniques , and such potential can be observed in the mobile network evolution trends [ 12 ].
These are illustrated in Figure 1 , where the changes in the access network and EPC are shown. The evolution in the access network involved the change from the use of base transceiver station BTS into Node Bs. In the last subfigure on the right, the data and control planes are decoupled and the control plane interfaces are handled by SDN controllers acting on the data plane, indicated by gray boxes.
In addition, the controllers handle network slices , which consist of a logical instantiation of a network, and enforce network management rules. Also, the BS in this last subfigure possesses energy harvesting capabilities, which is expected to reduce carbon emissions from mobile networks.
Therefore, it is difficult, if not impossible, to come up with a coherent system design that can act as a benchmark for future EPC designs. In this paper, we try to shed some light on the main architectural approaches, emphasizing their differences, pros, and cons.
The state-of-the-art EPC architectural proposals for next-generation networks can be categorized into the following strategies towards 5G network evolution: Virtualization in the EPC can be enabled by grouping the EPC network functional entities into different segments to attain less control, signaling traffic, and less congestion in the data plane [ 16 , 17 ].
It requests the user information from the user-data repository UDR, the central user information database and stores these data temporarily in its cache memory. In this way, authentication and authorization are processed internally, without performing any data transmission through the network. The idea behind this migration is that the PCRF requests user information in order to generate the required policies for each established bearer, and thus information exchange is minimized resulting in low latency for policy function generation.
The two approaches that we describe next [ 17 , 18 ] both group some of the EPC functionalities, although in different ways.
The controller also performs the MME functions, in its traffic management layer. The data bearers assignment, usually performed by the SGW, is implemented by the controller in advance, as soon as the UE moves near a new BS. The network entities are virtualized and deployed in one plane to achieve efficient interworking, and such allows independent networks to be reconfigured in a flexible manner and automatically on the same physical infrastructure. These two architectures are discussed in greater detail next.
The SoftCell architecture [ 18 ] consists of softwarized access switches that perform fine-grained packet classification on traffic from UEs located at the EPC network edge. In [ 18 ], researchers try to provide flexible policies in the EPC without compromising scalability. To come up with an efficient EPC design, the factors affecting the EPC scalability were considered and publicly available network statistics were utilized.
The EPC design consists of commodity middleboxes e. It computes and installs rules in the switches to direct traffic in both directions of a connection, thus minimizing the use of specialized network devices. The data traffic is then directed through a sequence of middleboxes optimized to the network conditions and UE locations, using the controller. In the data plane, hierarchical addressing grouped by BS and policy tags identifying paths through middleboxes are used in the EPC switches to forward traffic and the packet classification is pushed to the access switches, which are located at the EPC edge.
There, fine-grained rules are specified by the controller and applied to map UE traffic into policy tags and hierarchical addresses. The SoftCell architecture leverages some properties of the EPC, that is, by considering that traffic begins at the network edge. In this way, each BS has a serving access switch e. The EPC architecture proposed in [ 17 ] uses SDN, splitting the network into three planes; i application plane, ii control plane, and iii forwarding plane.
In the forwarding plane, the EPC coexists with the access network as all the control functions are moved into the control plane. By doing so, the forwarding plane consists of virtualized network devices that perform switching and packet forwarding, according to the SDN paradigm.
In addition, data caching strategies are exploited to minimize the traffic that goes through the EPC; that is, NFV techniques are used for caching popular content and store it on the EPC and access network when the network is idle, thus reducing the pressure in the data links and SDN controllers. In this way, latency can be minimized as content is cached locally, and better network management can be achieved through the use of virtualized EPC NFs.
The proposed architectures [ 17 , 18 ] differ from one another. However, both of them use an SDN controller for network management and policy enforcement, under different user mobility and traffic load variations.
That is, the infrastructures are deployed with simplified and virtualized network devices, whose software is decoupled from the hardware and centralized to the control entity. Entities such as switches solely take the strategy developed by the controller and forward traffic to access networks.
The architecture in [ 17 ] avails the potential of reducing latency through traffic offloading and caching in the RAN and EPC network. The challenge in [ 18 ] is that the centralized controller may become a source of bottleneck for the network operation when the network scales up, due to the fact that all the control functionalities are pushed towards it.
To reduce this burden, a wireless side of the network introducing the concept of Cloud-RAN C-RAN is proposed in [ 17 ], where the base band unit BBU pool has both control and data forwarding functions, in addition to the controllers deployed in the control plane. Through collaborative control, contents can be optimally distributed and stored in diverse devices of the EPC and the RAN via caching and broadcasting, thus overcoming the bottleneck problem of [ 18 ]. The emergence of a network paradigm supporting social requirements is one of the aims of 5G.
With softwarization, information centric networking ICN , as one of the candidates, can be enabled where in-network caching can be provided. The content cache server caches contents passing through the node and then autonomously selects which contents to cache based on the need of the mobile users accessing the node, that is, based on the content request frequency.
This approach can reduce the overall energy consumption within the network, since contents are cached and stored in close proximity to mobile users. In addition, in-network caching reduces the traffic content transmission within the network and also facilitates in-network data processing [ 14 ], whereby each network node carries out some data processing and service provisioning.
This leaves some of the nodes within the network inactive, thus enabling energy saving procedures e. The proposed architecture in [ 17 ] can improve energy efficiency as it allows content caching in the EPC and the RAN, while [ 18 ] does not employ any energy efficient EPC procedures, except for implementing policies for data traffic directions, which can also make some of the middleboxes inactive.
The virtualized EPC on VMware vCloud NFV platform is introduced in [ 20 ] to enable the degree of flexibility that will make it possible to deploy services closer to the edge, while managing, monitoring, and automatically scaling the heavier workload.
It mainly abstracts the EPC network functions, decomposing and allowing them to run as software instances virtual machines , on standard servers. This allows service providers to customize services and policies to design networks in new ways, to reduce costs and simplify operations. Another cloud-based approach that provides all network and access functionalities is proposed in [ 21 ], where the network cloud utilizes NFV for dynamic deployment and scaling of the NFs.
The key elements in this architecture are 1 a data-driven network intelligence for optimizing network resources usage and planning and 2 relaying and nesting techniques: The EPC is virtualized into three parts, namely, i control plane entity CPE , which is responsible for authentication, mobility management, radio resource control and non-access stratum NAS , and access stratum AS integration, ii the user plane entity UPE , acting as a gateway, mobility anchor, and over-the-air OTA security provisioner, and, lastly, iii the network intelligence NI plane is for the extraction of actionable insights from big data, orchestration, or required services and functionalities e.
The realization of the network cloud can be achieved by enabling virtual function instances to be hosted in data centers when needed. For example, in case of a natural disaster, with this technology the local data center is maybe unable to cope with the traffic upsurge; therefore, additional capacity can be sourced quickly from other data centers.
The proposed architectures [ 20 , 21 ] both use a cloud-based approach with NFV platform that enables a dynamic deployment of edge networks, the scaling of NFs, network monitoring, and load management. Also, they can pool capacity of resources when required, intelligently.
The difference between them is that, in [ 20 ], only the NFV platform is available for enabling network services provision while, in [ 21 ], the combination of SDN and NFV is utilized to provide network control and to host the network intelligence. The architecture proposed in [ 20 ] is commercially available. The driving force behind such architectures is the use of virtualization tools for instantiating each service when required, and these tools are discussed in the following.
Virtualization was introduced to primarily optimize hardware utilization by overlaying one operating system on top of another. Each of the systems consequently shares hardware resources to support underlying processes. The tools that enable virtualization can be categorized into i hypervisors and ii docker engine and are discussed next. These are functions which abstract or isolated operating systems and applications from the underlying computer hardware.
A new agnostic OS is generated to manage the underlying resources. For example, with a Windows system based hypervisor running on underlying physical hardware, another system running on virtual resources can be generated and Linux can be installed on it. This second OS will be the guest OS.
The base OS Windows in this example simply adapts the underlying physical hardware resources to accommodate the processing requirements of the guest OS. Since hypervisors sit between the actual physical hardware and the guest operating system OS , they are also referred to as virtual machine monitors VMMs. Xen is based on the open source Xen Project [ 22 ]. This hypervisor is a bare metal virtualization platform that has been included in the Linux kernel. It is used for a number of different commercial and open source applications, such as server virtualization, infrastructure as a service IaaS , desktop virtualization, security applications, and embedded and hardware appliances.
KVM is another hypervisor built into the Linux kernel, that is, a special operating mode of QEMU which is a generic and open source machine emulator and virtualizer that uses CPU extensions for virtualization via a kernel module. The kernel module provides the core virtualization infrastructure and a processor specific module.
Each virtual machine has private virtualized hardware: The kernel component of KVM is included in mainline Linux, as of 2. Another fundamental concept is that of a virtual machine VM , which is an operating system OS or application environment that is installed on software, which imitates dedicated hardware. Each VM includes a full copy of an operating system, one or more apps, necessary binaries Bins and libraries Libs taking up tens of GBs.
The hypervisor allows multiple VMs to run on a single machine. In Figure 2 a , we observe that each VM has a virtual OS of its own and the hypervisor provides the VMs with a platform to manage and execute multiple guest OS and allows host computers to share their resources among them.
A drawback of VMs is that they can be slow to boot. Docker is an open platform, or a software technology written in the Go programming language, and takes advantage of several features of the Linux kernel to deliver its functionality, for developing and running applications. It runs natively on Linux systems, where it uses Linux kernel features like namespaces , to provide isolated workspace, and control groups cgroups , a technology that limits an application to a specific set of resources, to create a loosely isolated environment called a container , thus avoiding the overhead of starting and maintaining VMs.
Mainly, it provides tooling, that is, software packaging tools that can package an application and its dependence in a virtual container that can run on any Linux server, and a platform to manage the containers lifecycle. Containers are abstraction units for isolating applications and their dependence, that can run in any environment.
They can run on the same machine, on top of the docker engine, sharing the OS kernel with other containers. They occupy less memory space than VMs, and this allows them to have a shorter start-up time. Mainly, they enable OS level virtualization whereas VMs provide hardware virtualization. However, they are similar as they also have a private space for processing, executing commands as root, and making use of private network interface and IP addresses. Figure 2 b shows that containers only package up the user space and not the kernel or virtual hardware, like a VM does.
Each container gets its own isolated user space to allow multiple containers to run on a single host machine. We observe that the entire OS level architecture is being shared across them. The only parts that are created from scratch are the Bins and Libs. This makes containers lightweight: Each application uses its own set of resources, without affecting the overall performance of the server.
They are therefore ideal for enterprises which concurrently run multiple processes on the same server. Despite the process isolation and lightweight character, containers are less secure and more vulnerable compared to hypervisors.
By only accessing a couple of namespaces through libcontainers , the default container format, and leaving out the rest of the kernel subsystems, it is possible to crack and hack through the OS. In [ 14 , 15 ], a dockerized EPC is presented as one architecture utilizing containers as a virtualization tool.
NFV is envisioned to be a key virtualization technology in 5G. One of the challenges to be faced by developers is the selection of the appropriate virtualization tool to use when developing the virtualized NFs platforms, that is, either hypervisors or the docker engine, or to simply let VMs and docker containers coexist in the same platform; see [ 13 ].
However, the future looks different; docker will probably coexist with hypervisors as the use of containers, running on top of the docker engine, speeds up innovation, requires less space and can be deployed across different platforms and hypervisors, with VMs on top, and allows running multiple applications on multiple VMs.
The combination of the virtualization tools can be beneficial as operators cannot be restricted to one infrastructure; instead they can simply develop applications once and then run them on any infrastructure [ 13 ]. The proposed architectures [ 20 , 21 ] both make use of NFV and cloud computing platforms. These technologies avail the possibility of scaling down resources when the demand is low and schedule resources based on demand; that is, resources can be outsourced through infrastructure as a service IaaS business models during peak hours.
The dynamic scaling of resources avails opportunities for improving energy efficiency EE within the cloud platform, as presented in [ 21 ]. Also, it allows the network cloud to collect user-centric, network-centric, and context-centric data.
Through this information centric approach, intelligent algorithms, mainly network optimization tools, can be applied to the aggregated data in order to provide useful input for network planning and resource management, thus improving EE. The virtualization tools used can also play a role towards EE improvement within the network. For example, the hypervisor can report resource usage to the orchestrator in order to trigger system automated sleep mode states and also to implement policies provided by management and orchestration, which includes power management and power stepping [ 23 ].
Since VNFs provide on-demand access to a pool of shared resources, where the locus of energy consumption for components is the VM instance where the VNF is instantiated. Therefore, the NFV framework can exploit the potential possessed by the virtualization technologies in order to reduce the energy consumption in future networks. Supporting the separation of the control and user plane functions is one of the most significant principles of the 5G EPC architecture.
With the advent of virtualization NFV, SDN, and cloud technology , it is now possible to build networks in a more scalable, flexible, and dynamic way. The concept of flexibility applies not only to the hardware and software parts of the network, but also to its management.
For example, setting up a network instance that uses different network functions optimized to deliver a specific service needs to be automated. With virtualization technology, resources can be isolated resulting in a so-called network slice , which refers to an isolated set of programmable resources to enable network functions and services.
With network slicing, one physical network is sliced into multiple virtual ones, each architected and optimized for a specific service or application. The dockerized EPC architecture using the FLARE node an open deeply programmable network node architecture is introduced as a network slicing architecture example in [ 14 , 15 , 24 ]. The EPC is decomposed into network slices , each implementing a network service as illustrated in Figure 3.
In [ 24 ], the FLARE architecture was used to resolve technical challenges that include ease of programming, reasonable and predictable performance, and the isolation among multiple concurrent logics for a faster and modular programming of the SDN data plane. To facilitate programming, the Toy-Block networking programming model [ 25 ] has been introduced and furthermore, the combination of computational resources was introduced to obtain a reasonable and predictable performance.
To improve performance, a lightweight resource virtualization technique, called resource container for isolation of multiple logics , was proposed. The cores were partitioned into groups, each with a resource container. The isolation used virtualization techniques. Another architecture based on network slicing, called mobile-central office rearchitected as a datacenter M-CORD , has been proposed [ 26 ].
M-CORD is expected to deliver the agility of a cloud provider, also featuring software platforms that enable rapid creation of new services. Its objectives are to enhance resource utilization, especially in terms of radio spectrum, to provide customized services and better QoE to customers and to offer an agile and cost-efficient deployment through the virtualization of the EPC.
It is expected that this architecture will involve the integration of vendor solutions within the CORD service framework. E2E QoS depends on the radio access, the EPC, and the wired part of networks, and 5G systems should have the capability to tailor it by organizing functions and connectivity so as to satisfy the system requirements, for example, mission critical applications.
Mobile users are supposed to be satisfied with the quality, when using any applications anytime, anywhere. There exists a gap between current SDN technology developments, as noted in current technology reports [ 14 ], and the functionalities that are required by 5G networks for E2E quality.
For example, current focus is towards coming up with robust SDN controllers for network control and rules enforcement in networks, that is, reducing bottlenecks, while the use of SDN controllers to manage network slices is overlooked, yet management of network slices, under latency constraints, is one of 5G requirements whereby each slice needs to be controlled in an efficient way for service provision, based on the quantity of data and quality requirements. Apart from slice s management, even the radio side of the network needs to be managed for resource reservation, especially in cases of disasters, thus guaranteeing end-to-end quality.
The observed similarities in the proposed architectures [ 14 , 26 ] are that both of them reorganize the network into network slices , respectively, consisting of data and control planes. The difference between them is that the MME and the integrated gateway, SP-GW, are implemented in the docker platform for control purposes [ 14 ], whereas, in [ 26 ], the controller manages the network as it performs control functionalities.
However, network slicing still poses a challenge as there are many dimensions and technologies included in this paradigm [ 27 ]. The challenges include RAN and EPC reconstruction to support end-to-end network slicing, slice management, and cooperation with other 5G technologies [ 27 ].
In addition, focus towards energy efficiency in network slicing is still lacking, as observed in [ 15 , 26 ], yet resources for the network slices can be set up based on various service characteristics, for example, bandwidth demand and latency demand, over the same or shared infrastructure. The slice manager, if present, can be able to allocate resources per slice and also trigger energy savings strategies in unused resources. The evolution towards 5G is expected to bring about several new ways of designing networks, so that the promise of always on, high-bandwidth, low latency, massive networks will become a reality.
MEC makes use of the large amount of power and storage space distributed at the network edge, which can yield sufficient capacities to perform computation-intensive and latency-critical tasks on mobile devices. Mainly, it aims at enabling cloud computing capabilities and information technology IT services in close proximity to end users, by pushing computation and storage resources towards the network edge i.
The direct interaction between mobile devices and edge servers through wireless communications brings the possibility of supporting applications with ultra-low latency requirements, prolonging device battery life and facilitating highly efficient network operations. This technology is expected to enable operators to better adapt traffic to the prevailing radio conditions, optimize service quality, and improve network efficiency [ 29 ].
Some of the computing functions that formerly only existed in the EPC are now moved out to the network edge. By disaggregating network services and functions out of the EPC, significant savings in cost, latency, round trip time RTT , traffic download time, physical security no need for security provision to facilities as the network consists of virtualized network devices , and caching efficiency [ 29 ] can be attained.
Energy efficiency is a major concern in the design of 5G systems and, as such, is also a prime concern for the design of MEC architectures [ 30 ]. In the following subsections, we provide an overview of 1 the reference ETSI MEC architecture, 2 the integration of renewable energy into MEC systems, 3 the optimization of MEC systems, and 4 we provide a discussion of relevant use cases.
MEC is a foundational network architecture concept which is expected to help 5G networks deliver the significant capability gains that are required by IoT, enhanced mobile broadband, virtual reality, self-driving vehicles, and many other applications. It will also provide a set of services that can be consumed by applications running on the MEC platform. These services will offer real time network information such as radio conditions, network statistics, and the location of connected devices to the running applications.
Different architectures are being proposed for future 5G MEC networks. In [ 31 ], a MEC NFV-based architecture is proposed and new APIs are opened, availing hosting environments for both mobile operators and external players, which can make use of the access network related information for their services. This architecture consists of the infrastructure plane, the control application plane, and the management plane.
In addition, there is an orchestration and management plane hosting MEC management activities. The hosting environment consists of hardware resources, a virtualization infrastructure virtual computation, storage, and network resources , and a set of associated management services for MEC applications. The major components of the MEC reference architecture [ 32 , 33 ] are the MEC application platform, providing infrastructure services, radio network information services RNIS, which provides radio network information systems features , and the user location services LOC, which provides UE location features.
Through the RNIS function, the radio network data and other real time context information can be exposed to authorized MEC network management applications. The TOF is responsible for service prioritization and routes selection, policy-based, and user-data stream to and from applications.
The overall view of the deployed MEC servers is maintained by the mobile edge orchestrator, which determines the optimum location s for instantiating a MEC application, and the virtualized infrastructure manager VIM , which is responsible for resource management of the virtualized infrastructure. In the orchestration and management plane, an additional manager is introduced, which is dedicated to managing the MEC platform, including its services and the respective APIs. An optimal deployment of MEC servers is key [ 34 ].
Such approach allows different vendors to develop applications and deploy them within the access network. To minimize latency, the MEC platform can be placed inside the BS as it is the first connection point for the mobile user. When considering colocated BSs, from one mobile operator, it may be beneficial to place the MEC platform at an aggregation point, a point within range to a set of BSs, as this can centralize resources and avails BS management without incurring significant amount of latency.
A new functional architecture that is worth mentioning is proposed within the COMBO project [ 35 ], where a new element called the universal access gateway UAG is introduced. They are located in a central office closer to end users so that they can access the national IP network and reach the Internet sooner, thus enhancing latency and saving transport resources. By placing the mobile gateways closer to end users, all traffic that does not need specific treatment is delivered locally to the operator IP EPC network.
A proper functional integration is of great importance to offer virtual resources at the mobile edge, while effectively adapting to the actual network load.
EH is orthogonal to what we have discussed so far. In fact, up to now we have elaborated on increasing the efficiency in the network management, whereas the aim of energy harvesting is to supply network apparatuses, reducing their carbon footprint.
Current mobile systems are powered using grid energy, which inevitably emits large amounts of carbon into the atmosphere. Recently, off-grid renewable energy sources such as solar radiation and wind energy have emerged as viable and promising sources for various IT systems due to the advancement of energy harvesting techniques [ 36 , 37 ]; see Figure 4.
The authors in [ 38 ] observed that solar energy is more suitable for workloads with high peak-to-mean ratio PMR , while wind energy fits better for workloads with small PMR. This avails the development of proper strategies for renewable energy provisioning for edge servers with the objective of eliminating any chance of energy shortage. This can be achieved by selecting the appropriate renewable energy source at each time instance taking into account current and forecast traffic loads.
Since MEC servers are small-scale data centers, each of which consumes less energy than conventional cloud data centers, it is expected that powering the MEC infrastructure with renewable energy sources will reduce the overall network energy consumption. This is important, especially in light of the dense deployment pattern that is foreseen in 5G systems. A challenge to be addressed for renewable energy powered MEC systems is the green-aware resource allocation and computation offloading, which should take the renewable energy constraints into account e.
Also, with renewable energy sources, the energy side information ESI , which indicates the amount of available renewable energy, will play a key role in decision making for storage and computing applications. MEC devices may also be energized through wireless power transfer WPT [ 39 , 40 ], when the renewable energy is insufficient. WPT may be exploited for computational offloading in mobile devices [ 41 ] or data offloading for MEC in future networks [ 42 , 43 ].
We stress, however, that the energy transfer efficiency of current WPT techniques is still very low, and that new methods are required to increase it and make WPT appealing in practice; see [ 44 ]. MEC servers can allow their resources to be jointly managed for serving a large number of mobile devices.
However, as the network size increases the resource management becomes a large scale optimization problem with respect to offloading decisions, radio, and computational resource allocation variables.
For energy efficiency reasons, it is desirable that MEC systems make use of low complexity optimization algorithms with moderate signaling overhead. Despite recent advancements in large scale optimization algorithms for radio resource management, these may be difficult to be verbatim-applied due to the combinatorial and nonconvex nature of computation offloading problems, which thus require ad hoc solutions [ 45 ], able to handle huge traffic volumes.
In the following paragraphs, some optimization examples are discussed. In [ 46 ], a reinforcement learning-based online algorithm was used to enhance decision making for EH powered MEC systems in determining the amount of workload to be offloaded from the edge servers to the central cloud, as well as the processing speed of the edge server, taking into account the congestion status in the EPC, the computation workload, and the ESI.
Furthermore, a Lyapunov optimization technique based on channel state indicator CSI and ESI was used to obtain dynamic offloading policies for EH powered mobile devices [ 47 ]. Both optimization techniques, reinforcement learning online and Lyapunov optimization based, were used to study small-scale MEC deployments, without taking into account large scale networks.
Further work shall be carried out to scale up and this still poses a challenge to researchers as data demand increases. Robust optimization algorithms for handling large scale deployments of MEC servers have to be developed under the ESI constraint.
In [ 48 ], a new energy efficient design principle for the BS colocated with the MEC server to minimize its energy consumption, while ensuring self-sustainable computation at the mobile devices through WPT , is investigated using the Lagrangian duality method. A multiuser MEC system consisting of a multiantenna access point and multiple users is assumed. Each mobile device is equipped with two antennas, one for WPT and the other for computational offloading.
The antennas operate over different frequency bands such that WPT and computational offloading can be performed simultaneously, without mutual interference. Users rely on their harvested wireless energy to execute the latency-sensitive computational tasks either via local computing or possibly partial offloading to the MEC server.
The optimal policy under energy harvesting constraints is obtained leveraging the Lagrange duality [ 49 ] and the ellipsoid methods [ 50 ].
During the MEC computation offloading process, the energy consumption for processing the task involves the energy spent by the mobile device to transmit the data to the MEC server and that involved in the computation at the server side.
To minimize the system energy consumption under latency constraints, a three-stage energy optimization scheme, that is, i mobile device classification, ii priority determination, and iii radio resource allocation, is proposed in [ 43 ]. This algorithm involves priority assignment and classification type for mobile devices to reduce the problem complexity.
The problem is formulated as a special maximum cardinality bin packing program [ 51 ], where mobile devices choose their task allocation mode through binary strategies, taking into account the transmission interference with other terminals and the limited radio resources.
The obtained numerical results demonstrate that the approach increases the energy efficiency of the MEC system. The MEC server is assumed to have knowledge of the local computation energy consumption, of channel gains and fairness indices for all users.
The resource allocation is formulated as a convex optimization problem, whose solution is obtained through dual-decomposition coupled with a relaxation constraint on the cloud capacity.