标题: 【原创】《物理层和链路层认知技术》 [打印本页]
作者: fight_boy 时间: 2009-3-25 12:36 标题: 【原创】《物理层和链路层认知技术》
《认知无线电技术》学习三——
《物理层和链路层认知技术》
by Thomas W. Rondeau and Charles W. Bostian
(Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA, USA)
概念和观点理解：
case-based theory -- 案例理论；genetic algorithms (GAs) -- 遗传算法（GA）
It is not a purely selfish adaptation where the radio seeks to maximize its own consumption of resources. CR不是只追求自身资源消耗最大化的自私自适应参数调整。
“tragedy of the commons” -- 共有悲剧. 2.4GHz ISM频段，IEEE 802.11设备和蓝牙设备相互干扰，前者有更高的发射功率，后者具有不断重发数据包直到传输成功的协议。
This chapter addresses methods of how to find a local or global optimum for the current channel environment (当前信道环境下的全局或局部最优).
最优的定义：a radio is optimized when it achieves a level of performance that satisfies its user’s needs while minimizing its consumption of resources such as occupied bandwidth and battery power. 满足用户需求的同时最小化其带宽和功率等资源消耗。
Cognitive radios merge AI and wireless communications. -- CR是AI与无线通信相结合的产物。认知引擎是其智能核心。
The cognitive engine takes in information from the user domain, the radio domain, the policy domain, and the radio itself. The user domain (用户域) passes information relevant to the user’s application and networking needs to help direct the cognitive engine’s optimization. The radio domain (无线电域) information consists of radio frequency (RF) and environmental data that could affect system performance such as propagation or interference sources. The policy engine (政策引擎) receives policy-related information from the policy domain (政策域). This information helps the cognitive radio decide on allowable (and legal) solutions and blocks any solutions that break local regulations.
In radio, we can think of the classical transmitters and receivers as having adjustable control
parameters (knobs) that control the radio’s operating parameters. Knob -- 旋钮，文中用其代表无线电的可调节参数。
Radio performance metrics are referred to as meters. Meter -- 文中用来作为衡量CR结果的指标。
The knobs of a radio are any of the parameters that affect link performance and radio operation. -- Knob：影响链路性能和无线电操作的任何参数。物理层中，中心频率、符号速率、发射功率、调制类型和调制阶数、PSF（脉冲成型滤波器）类型和阶数、扩频类型、扩频因子等；链路层中则为各种可以改进网络性能的变量，信道编码与交织的类型和速率、接入控制方法（如流量控制、帧大小以及多址接入技术）等。
[物理层&链路层&MAC层知识——
物理层的主要任务是实现通信双方的物理连接，以比特流(bits)的形式传送数据信息，并向数据链路层提供透明的传输服务。物理层是构成通信网络的基础，所有的通信设备、主机都需要通过物理线路互联。物理层建立在传输介质的基础上，与传输媒介密切相关，是系统和传输介质的物理接口，是OSI模型的最低层。 物理层有关的连接设备有：集线器、中继器、传输媒介连接器、调制解调器等。物理层主要解决的问题是：连接类型、物理拓扑结构、数字信号、位同步方式、带宽使用、多路复用等。
数据链路层的功能就是利用物理层提供的比特流传输功能，实现在相邻节点(node)间的透明、可靠的数据传输，具体要实现下列功能：链路管理、帧同步、差错控制（CRC，ARQ，信道编码，交织）、流量控制。 根据网络规模的不同，数据链路层的协议可分为两类：一类是针对广域网（WAN）的数据链路层协议，如HDLC、PPP、SLIP等；一类是局域网（LAN）中的数据链路层协议，如MAC子层协议和LLC子层协议。
MAC(子)层位于OSI七层协议中数据链路层的下半部分，主要负责控制与连接物理层的物理介质。在发送数据的时候，MAC协议可以事先判断是否可以发送数据，如果可以发送将给数据加上一些控制信息，最终将数据以及控制信息以规定的格式发送到物理层；在接收数据的时候，MAC协议首先判断输入的信息并是否发生传输错误，如果没有错误，则去掉控制信息发送至LLC层。]
Performance is a measure of the system’s operation based on the meter readings. In optimization theory (最优化理论), the meters represent utility and cost functions (效用或代价函数) that must be maximized or minimized for optimum radio operation. All of these performance analysis functions constitute objective functions (目标函数).
Modeling Outcome as a Primary Objective -- 将结果建模为主要目标
The basic process followed by a cognitive radio is that it adjusts its knobs to achieve some desired (optimum) combination of meter readings. Rather than randomly trying all possible combinations of knob settings and observing what happens, it makes intelligent decisions about which settings to try and observes the results of these trials.
BER与SINR的观测例子
The radio observes the BER and SINR value. If these are consistent according to the above formulas, the radio can assume that the channel is behaving predictably. It can then turn knobs that directly affect SINR, for example starting with the easiest, transmitter power. If the transmitter power is already at the allowable limit, the radio may lower the data rate to change the occupied bandwidth and therefore increase the average energy per bit. If the BER and SINR are
not consistent with the known formulas, the radio might assume, for example, that the channel is dispersive and opt to change the carrier frequency rather than the transmitter power.
MODM (Multi-Objectives Decision Making, 多目标决策) 理路[或称MOP (Multi-Objective Programming, 多目标规划；Multi-objective Optimization Problem, 多目标最优化问题）理论]
In an MODM problem space, a set of solutions optimizes the overall system, if there is no one solution that exhibits a best performance in all dimensions. -- 此话怎讲，貌似不通？
理解Pareto前沿最重要的概念是所有的解都是各个目标函数的折中。只有很小一部分多目标问题可以使其解同时优化所有目标，这个概念称为乌托邦点。
[多目标规划中，由于存在目标之间的冲突和无法比较的现象，一个解在某个目标上是最好的，在其他的目标上可能比较差。Pareto 在1986 年提出多目标的不受支配解(Non-dominated set)的概念。其定义为：假设任何二解S1 及S2 对所有目标而言，S1均优于S2，则我们称S1 支配S2，若S1 的解没有被其他解所支配，则S1 称为非支配解（不受支配解），也称Pareto解。这些非支配解的集合即所谓的Pareto Front。所以座落在Pareto front 中的所有解皆不受Pareto Front 以外的解（以及Pareto Front 以内的其它解）所支配，因此这些非支配解较其他解而言亦拥有最少的目标冲突，所以提供决策者一个较佳的选择空间。在某个非支配解的基础上改进任何目标函数的同时，必然会削弱至少一个其他目标函数。]
不同的目标对不同的用户需求和信道条件有不同的意义和重要性。目标与Knob间的相互作用很复杂，所以很难定义搜索空间，而且搜索空间不是线性的，甚至不是凸的。这些相互作用很难特征化和预测，所以必须独立分析各个目标，用MODM理论来寻找参数集合。
重要结论：
®Many objectives exist, creating a large N-dimensional search space.
®Different objectives may be relevant for only certain applications/needs.
®The needs and subjective performances for users and applications vary.
®The external environmental conditions determine what objectives are valid and how they are analyzed.
®We may search for regions where multiple performance metrics meet acceptable performance, rather than searching for optimal performance.
MODM求解的GA方法
类比：染色体中的每个基因与无线电的某个特性（Knob）相对应。无线电参数的演进就像生物进化一样通过后续子代来改善无线电的“种群”，用基于性能的选择来指导无线电的进化。
GA是一种受生物和进化行为启发的搜索技术，实现步骤如下：
（1）对染色体种群进行初始化
（2）重复执行以下步骤直到达到停止准则：①选择父体染色体；②对父体染色体进行交叉得到子代；③对子代染色体进行变异；④对父代染色体的适应度进行评估；⑤替换适应度低的父代染色体。
（3）在最后一代中选择最佳染色体。
在实时的约束下，希望优化过程提供更好的相应而不是 最佳的相应，即没有必要完全实现最优化。Looking at what we need out of the performance of a radio instead of just what we want (着眼于所需要的而不是所想要的), the GA can give us very usable solutions very quickly.
Modeling means that the machine must have some representation of the outside world to which it can respond (其能相应的外部世界).
Actions are taken by the WSGA (无线系统遗传算法) in creating a new radio configuration.
Unsupervised learning is accomplished through feedback and a series of rewards and punishments (奖惩).
Knowledge is represented as a database of past models and actions to take, along with past actions taken, and any estimates of the level of success associated with these actions.
The core of the decision-making process is rooted in CBDT (案例决策理论), a technique that grew out of economic decision-making research and is similar to the CBR (案例推理) used in some AI implementations.
Learning is the aggregation of the above processes. -- 学习是以上过程的总和。This learning machine mechanism is similar to processes known to occur in human learning: sensing, acting, reasoning, feedback, and accumulating knowledge and experience.
[ 本帖最后由 hangasyougo 于 2009-3-25 23:06 编辑 ]
作者: fight_boy 时间: 2009-3-25 12:40
染色体的多维分析
With cost and efficiency multiplied together, we get an efficiency-weighted cost value useful in comparing different solutions. 目标函数结果的组合。
In communications, a typical data link should have around 10^-6 BER, but an audio link is sustainable up to and around a BER of 10^-3. In a cognitive radio, we have to account for all of these applications and their different performance demands on the radio to properly normalize them. 对目标函数结果归一化以支持其组合。
Niching (小生境) is a popular technique to ensure population diversity (确保种群多样性) throughout the GA. In a population, we wish to spread our chromosomes around the solution space, especially if the space is multimodal (has many local optima).
Dividing a total population into some smaller subset of populations (将总的种群划分成小的种群子集) is a technique used to reduce the computational complexity (减少计算复杂度) of each algorithm. These techniques have been tested to run different groups of solutions simultaneously to find the same optimum.
高层智能需求
自动调整参数来实现目标。希望认知引擎有自主设置GA权重的能力。
奖励好的性能以及惩罚差的性能。Rewards and Punishments Can Be Radio Algorithm-Inflicted; Rewards Can Be User-Inflicted.
Ultimately, all cognitive radio work really comes down to the multi-objective optimization problem, which is that the MODM search space: -- 所有认知无线电工作都归结为多目标优化问题：
1. Has no utopian point; that is, it has no point that fully optimizes all objectives.
2. Is nonlinear and nonconvex; instead, it is a very complex plane to model with complex relationships between all inputs and outputs.
用户意识：识别用户需求。
环境意识：理解信道和外部环境强加给无线电的操作限制。神经网络以其分类信息和模式识别的能力而著称，可用来识别调制类型。
权重和目标函数
The cognitive radio bases its decisions on a set of metrics to best represent the requirements of the whole radio system, the network, and the user. To do this properly, the decision-maker must value objectives differently. In the optimization process, different needs of the radio should be handled by using different objectives. The objectives used for a given optimization problem can be learned
through successes and failures, and the weights associated with each objective can likewise be altered to best represent the situation.
小结（非常好！）
The goal of a cognitive radio is to optimize its own performance and support its user’s needs. A radio is optimized when it achieves a level of performance that satisfies its user’s needs while minimizing its consumption of resources such as occupied bandwidth and battery power. The intelligent core of a cognitive radio exists in the cognitive engine, which performs the modeling, learning, and optimization processes necessary to reconfigure the communication system in which
the radio operates.
The first problem in dealing with cognition in a system is to understand (1) what information the intelligent core must have and (2) how it can adapt. In radio, we can think of the classical transmitters and receivers as having adjustable control parameters (knobs) that control the radio’s operating parameters, and observable metrics (meters) that measure its performance. The knobs are any of the parameters that affect link performance and radio operation. Meters are indicators
of performance on a particular level; thus, at the link layer, PER is an important metric.
The basic process followed by a cognitive radio is this: it adjusts its knobs to achieve some desired (optimum) combination of meter readings. Rather than randomly trying all possible combinations of knob settings and observing what happens, it makes intelligent decisions about which settings to try and observes the results of these trials. Based on what it has learned from experience and on its own internal models of channel behavior, it analyzes possible knob settings, predicts some optimum combination for trial, conducts the trial, observes the results, and compares the observed results with its predictions as summarized in an adaptation loop.
Without a single-objective function measurement, we cannot look to classic optimization theory for a method to adapt the radio knobs. Instead, we can analyze the performance using MODM criteria. MODM theory allows us to optimize in as many dimensions as we have objective functions to model. Cognitive radio operation requires an MODM algorithm capable of robust, flexible, and online adaptation and analysis of the radio behavior. The clearest method of realizing all of these needs is the GA, which is widely considered the best approach to MODM problem-solving. In it, we represent the radio parameters as genes in a chromosome and select the fittest chromosomes through a process called relative tournament evaluation (相对锦标赛评估).
The primary goal of the cognitive engine is to optimize the radio, and the secondary functions are to observe and learn in order to provide the knowledge required to perform the adaptation. A cognitive radio becomes a learning machine through a tiered algorithm structure based on modeling, action, feedback, and knowledge representation, as shown in the cognition loop of Figure 7.10. In the Virginia Tech cognitive engine, these functions are realized through the algorithmic structure of Figure 7.11. Its main parts are the CSM, responsible for learning, and the WSGA, which handles the behavioral adaptation of the radio based on what it is told to do by the CSM. The modeling system observes the environment from many different angles to develop a complete picture. The CSM holds two main learning blocks: the evolver (进化器) and the decision-maker, which takes feedback from the radio that allows the evolver to properly update the knowledge base to respond to and direct system behavior.
[Pareto（帕雷托）相关知识——
1879年，经济学家意大利人维弗雷多·帕雷托(Villefredo Pareto)提出：社会财富的80%是掌握在20%的人手中，而余下的80%的人只占有20%的财富。渐渐地，这种“关键的少数（vital few）和次要的多数（trivial many）”的理论，被广为应用在社会学和经济学中，并被成之为Pareto原则（Pareto Principle）。Pareto原则也常被称为80/20效率法则（the 80/20 principle），或称帕累托法则、帕累托定律、最省力法则或不平衡原则、犹太法则。80/20法则认为：原因和结果、投入和产出、努力和报酬之间本来存在着无法解释的不平衡。一般来说，投入和努力可以分为两种不同的类型：多数，它们只能造成少许的影响； 少数，它们造成主要的、重大的影响。
Pareto最优是以提出这个概念的维弗雷多·帕雷托的名字命名的，他在关于经济效率和收入分配的研究中使用了这个概念。 帕累托最优（Pareto Optimality），也称为帕累托效率、帕累托改善，是博弈论中的重要概念，并且在经济学， 工程学和社会科学中有着广泛的应用。 帕雷托最优的定义：帕雷托最优是资源分配的一种状态，在不使任何人境况变坏的情况下，不可能再使某些人的处境变好。
Pareto解又称非支配解或不受支配解（nondominated solutions）：在有多个目标时，由于存在目标之间的冲突和无法比较的现象，一个解在某个目标上是最好的，在其他的目标上可能是最差的。这些在改进任何目标函数的同时，必然会削弱至少一个其他目标函数的解称为非支配解或Pareto解。一组目标函数最优解的集合称为Pareto最优集。最优集在空间上形成的曲面称为Pareto前沿面。
另一个相关概念是Pareto改进(Pareto Improvement)，帕累托改进是指一种变化，在没有使任何人境况变坏的前提下，使得至少一个人变得更好。一方面，帕累托最优是指没有进行Pareto改进的余地的状态；另一方面，Pareto改进是达到帕累托最优的路径和方法。]
[数学规划（最优化）相关知识——
（变量）可行域：所有满足约束条件的点（即可行点）的集合。[与“定义域”等价？]
目标可行域：目标函数在定义域上的值域。
对于单目标优化而言，人们比较关注决策变量空间（即变量可行域）；而对于多目标优化，人们关注的焦点是目标向量的空间（即目标可行域）。
向量大小的比较，即是比较向量对应元素的大小：x = y; x \prec y；x \leqslant y; x < y。
多目标优化通常不存在绝对最优解（每一个目标相对于其它解都是最优的）；有效解x是指不存在另一个解y，满足y生成的每一个目标都比x生成的对应目标更优或等优，且确实有一个目标是更优(若存在这样一个y，则y显然比x更有效，x就不是有效解；允许存在y其所有目标都等优，即等有效的y)；弱有效解x是指不存在另一个解y，满足y生成的每一个目标都比x生成的对应目标更优（条件比有效解弱。可能存在某个（更有效的）y，生成的某些（担不是全部）目标更优，但x一定有某个(些)目标是不可超越的）。[不知我这些理解对否？]
有效解即是Pareto(最优)解，并非所有的目标都是最优的，但从这个解出发，要想改进某一个目标，必然会损害到其它某些目标。]
作者: 扬帆远航 时间: 2009-4-4 01:19
真有趣，这才像真正的科学，呵呵！
| 欢迎光临 研学论坛 (http://bbs.matwav.com/) |
Powered by Discuz! 7.0.0 |