用于多跳认知无线电网络的分布式网络编码控制信道

 Alfred Asterjadhi等 著   

大多数电磁频谱由政府机构长期指定给公司或机构专门用于区域或国家地区。由于这种资源的静态分配，许可频谱的许多部分在许多时间和/或位置未使用或未被充分利用。另一方面，几种最近的无线技术在诸如IEEE802.11，蓝牙，Zigbee之类的非许可频段中运行，并且在一定程度上对WiMAX进行操作;这些技术已经看到这样的成功和扩散，他们正在访问的频谱 - 主要是2.4 GHz ISM频段 - 已经过度拥挤。为了为这些现有技术提供更多的频谱资源，并且允许替代和创新技术的潜在开发，最近已经提出允许被许可的设备（称为次要用户）访问那些许可的频谱资源，主要用户未被使用或零星地使用。这种方法通常被称为动态频谱接入（DSA），无线电设备发现和机会性利用未使用或未充分利用的频谱带的能力通常称为认知无线电（CR）技术。

 DSA和CR最近都引起了无线通信和网络界的极大关注。通常设想两种主要应用。第一个是认知无线接入（CWA），根据该认知接入点，认知接入点负责识别未使用的许可频谱，并使用它来提供对次用户的接入。第二个应用是我们在这个技术中研究的应用，它是认知自组织网络（CAN），也就是使用

    用于二级用户本身之间通信的无许可频谱，用于诸如点对点内容分发，环境监控，安全性等目的，灾难恢复情景通信，军事通信等等。

 设计CAN系统比CWA有更多困难，主要有两个原因。第一是识别未使用的频谱。在CWA中，接入点的作用是连接到互联网，因此可以使用简单的策略来推断频谱可用性，例如查询频谱调节器在其地理位置的频谱可用性或直接与主用户协商频谱可用性或一些中间频谱经纪人另一方面，在CAN中，与频谱调节器或主要用户的缺乏直接通信需要二级用户能够使用检测技术自己识别未使用的频谱。第二个困难是辅助用户协调媒体访问目的。在CWA中存在接入点和通常所有二级用户直接与之通信（即，网络是单跳）的事实使得直接使用集中式媒体接入控制（MAC）解决方案，如时分多址（TDMA）或正交频分多址（OFDMA）。相反，预计CAN将跨越多跳，缺少集中控制器;而对于传统的单通道多跳自组织网络而言，这个问题的几个解决方案是已知的，因为假设我们处理允许设备访问的具有成本效益的最先进技术的状态，因此将它们重用于CAN是不直接的一次只能限制频谱的一部分，中间访问将在多个信道上执行，而且可用于二次通信的实际信道可能会随着位置和时间而变化。

     由于刚刚描述的两个问题，CAN中出现了几个实际的设计挑战，如实现控制信道，辅助用户对媒体接入的协调，实现用于检测未使用频谱的可靠方案等。在这篇文章中，我们将讨论这些挑战，我们显示，在以前的文献中，有几个很好的解决方案可以有效地解决一个或者一些这些问题。

     在讨论之后，我们提出了我们设计的方案，以克服CAN缺乏完整的解决方案。我们的方案是基于一个虚拟控制通道，利用用户以伪随机方式访问信道，并在任何频道遇到任何情况时交换控制信息。通过网络编码实现对所有用户的控制信息的高效传播。用户交换的控制信息包括根据预定义的确定性算法确定信道切换模式以及数据通信的资源分配所需的所有信息（带宽要求，主要用户存在和位置等）。我们通过提出和讨论模拟结果来讨论所提出的方案的性能，表明它是CAN实际实现的有效解决方案。

我们在CAN中遇到的第一个问题是鸡蛋问题：二次设备需要彼此协调来执行频谱接入，但是它们还需要访问频谱以便通信和实现协调。这个问题通常被称为控制频道问题，不幸的是，在与DSA相关的工作中往往被忽视。事实上，大多数DSA相关出版物更侧重于主要用户检测和/或高效频谱分配的问题，并且在这样做时，假设某些控制信道实现对于次要用户是可用的。

     为了实际实现控制通道，一些作者提出静态分配一些频谱带。这个实际提出了两个主要问题：一是需要静态频谱调节，这正是DSA旨在避免的一个问题。第二，选择的控制带可能很容易成为瓶颈。这在多跳场景中尤其如此，其中对控制信息交换的需求潜在地非常高（例如，不仅对于媒体访问，而且用于路由目的）。

沟通;然而，这些建议没有解决控制瓶颈问题。

     当然，CAN的理想解决方案不仅需要解决控制信息交换的问题，而且还要有效地实现对可用频谱资源的有效利用。在这方面，应该注意的是，先前讨论的多重会合策略最初被提出作为单通道技术的扩展，最着名的是IEEE 802.11;特别地，在这些解决方案中看到的优点是仅仅通过使用多个通道，可以在单通道情况下实现网络容量的显着增加。然而，要注意的是，多通道网络的容量限制还远远没有达到多重交会方案，这更是解决问题的实际方法，而不采取系统的方法来最大限度地提高信道利用效率。

     应该考虑到频谱有效使用的一个方面是在多跳网络中，通常只有一部分用户处于给定用户的干扰范围内。这通过频率重用来提高频谱利用率的可能性。不幸的是，在实践中，这需要更复杂的频谱分配策略，以及更多信息的可用性（例如每个用户的位置知识）。以分布式的方式是非常具有挑战性的。与此相关的问题是链路调度和路由问题：传统的自组织网络路由策略在多信道网络中是无效的，主要是因为给定的链路在任何时候都不能被激活，因为要求发送方和接收器在同一个通道上。理想情况下，应共同执行信道分配，链路调度和路由，以最大化频谱利用效率和网络性能。在这方面，已经提出了一些有趣的解决方案，但是它们具有要求集中式调度器的缺点。鉴于CAN的性质，需要一种分布式解决方案来实现实施。

     到目前为止，我们还没有处理可能最具特色的CAN的特征：适用于二次频谱接入的频谱的这些部分的识别必须由次要用户自己使用感测技术来执行。从最近的文献中已经深入研究了从单个二次用户的角度进行感测的主题，并且已经提出了从简单的能量或匹配滤波器检测到复杂的循环平稳特征检测技术的几种解决方案。然而，如对于无线电接入频谱的情况所讨论的，对主要用户的二次干扰维持在一定阈值以下的要求转化为对单用户检测策略的灵敏度要求高到不符合成本效益，或者甚至完全不切实际，用现有技术实现这种检测器。

     如上一节所述，这一领域的大多数以前的工作只解决了其中一个问题;相反，我们的方法旨在同时解决这两个问题。直观地，频谱分配和传输调度最好使用关于特定通信需求（例如，服务质量[QoS]要求）和频谱可用性的知识来执行（例如，由主用户检测信息）。将这些知识称为控制信息，通过收集所有用户生成的控制包获得。在文献中，当完整的控制信息用于资源分配时，

为了得出分布式方法，我们选择不同的策略：每个用户收集完整的控制信息，并为整个网络独立地确定资源分配。关键在于，如果相同的控制信息成功传播给所有用户，并且资源分配算法是确定性的，则每个用户将能够确定相同的资源分配，而无需用户之间的任何进一步的交互。这是我们首先提出的单跳多通道网络的多通道方案的基本原理，并在此讨论在多跳CAN中的使用。在本节的其余部分，我们提供更多关于我们的计划如何工作的细节;本文的其余部分更侧重于多机场和机场频谱接入问题。

     控制信息的确切性质由所选择的特定调度算法确定。作为一个例子，在我们讨论了一种相对简单的单跳网络统一资源分配算法。该算法仅需要参与参与分配的用户组的知识以及用于确定伪随机信道切换模式的随机数发生器的种子。因此，由每个用户生成的控制信息分组仅包括用户的唯一标识符（例如，其MAC地址）和使用的随机比特串以及所有其他用户的比特串来确定公共种子为随机数发生器。

     我们的方案正常工作的一个重要要求是控制信息的传播到达所有用户。每当特定用户在分配周期结束时无法检索控制信息时，该用户将潜在地确定用于后续分配周期的错误的信道切换模式和传输调度，可能开始使用资源（某些信道中的传输时隙）的传输将其分配给其他用户。在本文的其余部分中，我们将此事件称为频谱冲突，并参考无法将控制信息检索为误传用户的用户。一般来说，频谱冲突的机会，因此频谱资源浪费的平均数量随着用户数量的错误而增加。因此，我们想要一种传播方案，其中定义为普通用户从所有其他用户成功检索控制信息的概率的检索成功概率很高。

数据包引起额外的开销，这将需要在确定我们的DSA解决方案的总体控制开销时予以考虑;有关这个问题的详细讨论，请参阅读者。最后，为了实现网络编码的实际，我们采用缓冲模型。

     如我们以前的工作中所讨论的，网络编码大大优于其他策略，以便在单跳多通道网络中传播控制信息。换句话说，使用网络编码与伪随机信道切换模式相结合，为我们提供了一个虚拟控制信道，允许用户有效地共享控制信息。该网络编码的虚拟控制信道对于分组丢失和链路故障是鲁棒的，并且最重要的是不需要存在专用于交换控制信息的静态频谱资源。对于适用于二次接入的未使用频谱资源的检测，我们注意到，网络编码控制信道自然适合实施协同主用户检测解决方案。

4。结论

     在本文中，我们讨论了CAN中出现的主要挑战，并提出了基于虚拟网络编码控制通道的这些挑战的实际解决方案。我们提出模拟结果，证明在几种情况下如何实现控制信息的有效分散和有效的频谱利用。我们的解决方案显示出对主用户活动的鲁棒性，并且可以针对次级用户的数量进行扩展。未来的研究方向包括在提出的解决方案中整合更精细的频谱分配，传输调度和路由策略。

 

 

 

 

 

 

 

     most of the electromagnetic spectrum is assigned by government agencies to companies or insti-tutions for exclusive use over regional or national areas on a long-term basis. As a result of this static allocation of resources, several portions of the licensed spectrum are unused or underused at many times and/or locations . On the other hand, several recent wire-less technologies operate in unlicensed bands, such as IEEE 802.11, Bluetooth, Zigbee, and to some extent WiMAX; these technologies have seen such success and proliferation that the spectrum they are accessing — mostly the 2.4 GHz ISM band— has become overcrowded. In an effort to provide further spectrum resources for these existing technologies, as well as to allow the potential development of alternative and innova-tive ones, recently it has been proposed to allow unlicensed devices, called secondary users, to access those licensed spectrum resources that are unused or sporadically used by their owners, called primary users. This approach is normally referred to as dynamic spectrum access (DSA), and the ability of radio devices to find and opportunistically exploit unused or underused spectrum bands is normally called cognitive radio (CR) technology .

     Both DSA and CR have recently attracted significant atten-tion from the wireless communications and networking community. Two main applications are commonly envisioned. The first is cognitive wireless access (CWA), according to which a cognitive access point takes care of identifying unused licensed spectrum and uses it to provide access to secondary users. The second application, which is the one we investigate in this arti-cle, is cognitive ad hoc networks (CANs), that is, the use of

unlicensed spectrum for communications among the secondary users themselves, for purposes such as peer-to-peer content distribution, environmental monitoring, safety

communications in disaster recovery scenarios, military communi-cations, and many others.

     Designing a system for CANs presents more difficulties than for CWA, for two main reasons. The first is the identification of unused spectrum. In CWA the access point is by its role connected to the Internet, and therefore can infer spectrum availability using simple strategies, such as querying the spectrum regulator for spectrum availability at its geographic location or directly negotiating spectrum availability with the primary user or some intermediary spectrum broker . On the other hand, in CANs the lack of direct communication with the spectrum regulator or primary users requires secondary users to be able to identify unused spectrum by them-selves using detection techniques. The second difficulty is the coordination of secondary users for medium access purposes.In CWA the presence of an access point and the fact that commonly all secondary users communicate directly with it (i.e., the network is single-hop) makes it straightforward to use centralized medium access control (MAC) solutions, such as time-division multiple access (TDMA) or orthogonal frequency-division multiple access (OFDMA). On the contrary,CANs are expected to span over multiple hops and to lack a centralized controller; while several solutions to this problem are known for traditional single-channel multihop ad hoc networks, it is not straightforward to reuse them for CANs due to the fact that, assuming we deal with cost-effective state of the art technology that allows devices to access only a limited portion of the spectrum at a time, medium access is to be performed across several channels, and moreover the actual channels that can be used for secondary communications might vary with respect to location as well as time.

     Due to the two issues just described, several practical design challenges arise in CANs, such as the realization of the control channel, the coordination of secondary users for medium access, the implementation of a reliable scheme for the detection of unused spectrum, and so on. In this article we discuss these challenges, and we show that, while in the prior literature there are several good solutions that can effectively solve one or some of these issues.

     After this discussion we present the scheme we have designed in an effort to overcome this lack of a complete solution for CANs. Our scheme is based on a virtual control channel which exploits the fact that users visit channels in a pseudo-random fashion and exchange control information whenever they happen to meet in any channel. Efficient dissemination of the control information to all users is achieved by means of network coding . The control information exchanged by users consists of all the information (bandwidth requirements, primary user presence and location, etc.) that is needed to determine channel switch patterns as well as resource allocation for data communication according to a predefined deterministic algorithm. We discuss the performance of the proposed scheme by presenting and discussing simulation results which show that it is an effective solution for the practical realization of CANs.

     The first issue we encounter in CANs is a chicken-egg problem: secondary devices need to coordinate among themselves to perform spectrum access, but they also need to access the spectrum in order to communicate and achieve coordination.This issue is often referred to as the control channel problem,and unfortunately it is often neglected in work related to DSA. The fact is that most DSA related publications focus more on the problem of primary user detection and/or efficient spectrum allocation, and in doing so assume that some control channel implementation is available to secondary users.

     For the practical realization of the control channel, some authors  propose to statically allocate some spectrum band.This practice presents two major issues: first, it requires static spectrum regulation, which is exactly what DSA aims at avoiding; second, the chosen control band could easily become the bottleneck. This is especially true in multihop scenarios,where the need for control information exchange is potentially very high (e.g., not only for medium access, but also for routing purposes).

Some other solutions have been proposed that attempt to solve the first issue by dynamically choosing an unused licensed band to perform secondary user control

communications; however, the control bottleneck issue is not addressed by these proposals.

     Of course, the ideal solution for CANs needs not only to address the issue of the exchange of control information, but also to effectively enable efficient usage of the available spectrum resources. In this respect, it is to be noted that the multiple-rendezvous strategies discussed earlier were originally proposed as an extension to single-channel technologies, most notably IEEE 802.11; in particular, the advantage seen in these solutions was that just by enabling the use of multiple channels, a significant increase in network capacity could be achieved over the single-channel case. However, it is to be noted that the capacity limit of multichannel networks is still far from being reached by multiple-rendezvous schemes,which are more of a practical solution to the problem and do not take a systematic approach to maximizing the channel utilization efficiency.

     One of the aspects that should be taken into account for an efficient usage of the spectrum is that in a multihop network typically only a subset of the users are in the interference range of a given user. This opens up the possibility of higher spectrum utilization efficiency by means of frequency reuse.Unfortunately, in practice this requires more complex spectrum allocation strategies, as well as the availability of more information (e.g., knowledge of the location of each user).Doing this in a distributed fashion is very challenging. Coupled with this problem is the issue of link scheduling and routing: traditional ad hoc network routing strategies are not effective in multichannel networks, due primarily to the fact that a given link cannot be activated at all times because of the requirement that both the sender and the receiver are on the same channel. Ideally, channel allocation, link scheduling,and routing should be jointly performed in order to maximize spectrum utilization efficiency as well as network performance. In this respect some interesting solutions have been proposed , but these have the drawback of requiring a centralized scheduler. Given the nature of CANs, a distributed solution would be needed in order to allow practical implementation.

     So far, we still have not dealt with what is possibly the most peculiar trait of CANs: the identification of those parts of the spectrum that are suitable for secondary spectrum access must be performed by the secondary users themselves using sensing techniques. The topic of sensing from the point of view of a single secondary user has been intensively investigated in the recent literature, and several solutions have been proposed,from simple energy or matched filter detection to complex cyclostationary feature detection techniques. However, as discussed in  for the case of unlicensed access of TV spectrum, the requirement of maintaining secondary interference to primary users below a certain threshold translates into a sensitivity requirement for single-user detection strategies so high that it is not cost-effective, or even completely impractical, to implement such detectors with current technology.

     We consider the case in which each secondary user has a single transceiver, and thus can be tuned only on a single channel at any given time. We have a set of secondary users and a set of channels available for unlicensed access. In order to design a spectrum access scheme that is effective in this scenario, we need to solve the following two problems: how to make secondary users coordinate among themselves, and how to assign spectrum resources to these users in an efficient way.

     As discussed in the previous section, most prior work in this area addressed only one of these problems; in contrast,our approach aims at solving both problems simultaneously.Intuitively, spectrum allocation and transmission scheduling are best performed using knowledge about the particular communication needs (e.g., quality of service [QoS] requirements)and spectrum availability (e.g., expressed by primary user detection information) of all users. We refer to this knowledge as the control information, obtained by collecting the control packets generated by all users. In the literature, when complete control information is used for resource allocation purposes,

a centralized scheme is usually assumed. This means that there is a centralized controller that gathers the control packets generated by all users, determines the

global resource allocation, and then tells each user what resources to use for data communications.

     In order to derive a distributed approach, we choose a different strategy: each user gathers the complete control information and independently determines for the whole network the resource allocation. The key point is that if the same control information is successfully disseminated to all users, and the resource allocation algorithm is deterministic, each user will be able to determine the same resource allocation without any further interaction among users. This is the underlying principle of the multichannel scheme we first presented in for single-hop multichannel networks and discuss here for use in multihop CANs. In the rest of this section we provide more details on how our scheme works in general; the rest of this article focuses more in particular on multihop and oppor-tunistic spectrum access issues.

     The exact nature of the control information is determined by the particular scheduling algorithms chosen. As an example, in we discussed a relatively straightforward algorithm for uniform resource allocation in single-hop networks. This algorithm only needs knowledge of the set of users participating in the allocation and the seed for the random number generator used to determine the pseudo-random channel switch pattern. As a consequence, the control information packets generated by each user consisted of only a unique identifier for the user (e.g., its MAC address) and a random bit string used, together with the bit strings of all other users,to determine a common seed for the random number generator.

     An important requirement for our scheme to work properly is that the dissemination of control information reaches all users. Whenever a particular user fails to retrieve the control information at the end of an allocation period, that user will potentially determine a wrong channel switch pattern and transmission schedule for the subsequent allocation period,possibly starting transmissions using resources (transmission slots in certain channels) meant to be allocated to other users.In the rest of this article we refer to this event as spectrum collision, and refer to the users that failed to retrieve the control information as misinformed users. In general, the chances of having a spectrum collision, and hence the average amount of wasted spectrum resources, increase with the number of misinformed users. For this reason, we want a dissemination scheme in which the retrieval probability Pretr, defined as the probability that a generic user successfully retrieves the control information from all other users, is high.

     We propose the use of network coding in order to implement a reliable and efficient dissemination scheme for the control information. Network coding is a recently introduced paradigm for data dissemination, according to which the packets generated by multiple sources are jointly coded at intermediate nodes and decoded at the final destination. This coding strategy can be very effective in increasing throughput, reducing delay, and enhancing robustness. In order to have a practical implementation of network coding we refer to , where the authors proposed a distributed scheme for network coding that obviates the need for a centralized knowledge about the encoding and decoding functions, and at the same time allows asynchronous data exchange between nodes. According to this approach each node stores all incoming packets in an internal buffer and transmits an encoded packet that contains a random linear combination of all packets in its own buffer.At transmission time this packet is forwarded to all nodes situated within transmission range. Now, if the encoding vectors are generated randomly and the symbols lie in a finite Galois field of sufficient size, the information will be disseminated to all users with high probability . Based on this approach, every time a node receives an encoded packet, it has to know the coefficients used to perform the encoding in order to recover the original information packets. A simple solution consists of appending within each encoded packet the corresponding encoding vector that describes which linear combination of information packets it contains.This way, the encoding coefficients needed to decode the information stored in encoded packets can be found within the encoded packets themselves. Any node can thus recover the information packets

generated by all nodes simply by inverting the matrix that stores all the coefficients of the packets received during data dissemination. Appending the encoding vectors to the