method for power control is proposed for broadband, packet-switched TDMA wireless networks. By observing the temporal correlation of cochannel interference when transmitters can send data contiguously, a Kalman filter is used to predict interfence power in
received power and the power of the desired signal,where the latter can be measured by ltering basedon the training symbols for the signal. In fact, suchmeasurements can be involved and challenging,especially when time duration is short; see e.g.,[AS95], [AMY96] and [A97].
3.2 Interference Prediction by Kalman Filter
As mentioned above, despite of the burstiness ofpacket traf c, interference is correlated in time in wirelessIP networks. This temporal correlation enables us toemploy a Kalman lter to predict the amount ofinterference to be received at a base station in theimmediate future.
LetInbe the actual interference-plus-noise power indBm received in time slotnat a given base station. Inother words,Inis the "process state" to be estimated by theKalman lter. We assume that the noise power, whichdepends on the channel bandwidth, is given and xed. Forbrevity, unless stated otherwise, the interference plusthermal noise is simply referred to asinterferencein thefollowing. The dynamics of the interference power isdescribed by
In=In 1+Fn
(1)
whereFnrepresents the uctuation of interference powerfor slotnas terminals may start new transmissions and/oradjust their transmission power in the time slot. In theterminology of Kalman lter,Fnis the "process noise."LetZnbe the measured interference power in slotn. Then,
Zn=In+En
(2)
whereEnis the "measurement noise." Eq.(1) and (2) arecommonly referred to as thesignal generation model.Bythe Kalman lter theory [BH97], the time andmeasurement update equations for the interference powerare:
1_____Qn=_ W 1Rn= ηQn
Σi=n W+1
n
_
[Zi Zn]2
(9)(10)
whereηis a constant between 0 and 1. Strictly speaking,Qnin (9) is an estimate of the variance of the sum of theprocess and measurement noise because measurementsZn’s include the uctuation of both interference andmeasurement errors. However, since the standarddeviation of the interference power can reach as much astens of decibels, which is much higher than typicalmeasurement errors, (9) yields a good variance estimationfor the process noiseFn. In addition, the choice ofRnaccording to (10) withηless than 1 is reasonable becausethe measurement noise (error) is likely to be proportionalto, but smaller than, the uctuation of interference power.Furthermore, the sliding window sizeWshould be at leastseveral times the average message length so that multipleterminals are likely to have transmitted during the timewindow, thus capturing changes of interference power.It is noteworthy that the process noiseFnin (1) doesnot have a normal distribution, a requirement for theKalman lter to minimize the mean square error.However, in special cases when thermal noise is notconsidered and the interference uctuation is dominated byshadow fading, which is known to have lognormaldistributions [R96],Fnin dBm is indeed normallydistributed. For example, such domination by shadowfading can occur for static channel assignments with highfrequency reuse factor. In this case, cochannel cells are faraway from the target cell such that the path loss fromvarious transmitting terminals in the same cochannel cellto the target cell very much remains constant. Thus, the uctuation ofFnis mainly determined by the shadowfading between the transmitting terminal in the co-channelcell and the receiving base station in the target cell.AlthoughFnis not normally distributed in general,especially for frequency assignments with low reusefactors, our results show that the Kalman lter givesexcellent interference predictions when a message consistsof more than several packets on average.
For each slotn, the interference measurements areinput to (9) and (10) to estimateQnandRn. Using thesevalues and the current measurement, (5) to (7) yield theKalman gain, and the a posteriori estimates forInandPn,respectively. The a priori estimates for the next time slot
n+1in (3) is usedare given by (3) and (4). In particular,I
as the predicted interference power in slotn+1 for powercontrol as follows.
3.3 Determination of Transmission Power
Let us begin with some notation. Letβ*be the targetSINR,pnthe transmission power andgnthe path gainfrom the transmitting terminal to the base station for slotn,
nrepresent the actual andrespectively. WhileInandI
ntoestimated interference power in dBm, we useinandi
denote the respective values in the linear scale. ByAssumption 1, the base station can measure and estimategnaccurately. Based on this and the predicted interference
n+1=I nI
(3)(4)(5)(7)
n+1=P n+QnP
n(P n+Rn) 1Kn=P
n=( 1 Kn)P nP
n=I n+Kn(Zn I n) (6)I
nandI nare the a priori and a posteriori estimate ofwhereI
nandP nare the a priori and a posteriori estimateIn,P
error variance,Knis the Kalman gain, andQnandRnarethe variance for the process noiseFnand measurementnoiseEn, respectively.
Clearly,QnandRnneed to be estimated appropriatelyas input to (4) and (5). For that purpose, we propose thefollowing estimations based on the interferencemeasurements in a sliding window of the lastWslots:
n_1_ __Zn=Zi(8)
Wi=nΣ W+1
(0); //initialize with a measurement x_est_last = z_real + frand()*0.09; float sum_error_kalman = 0; float sum_error_measure = 0; for (...
% [x, V, VV, loglik] = kalman_filter(y, A, C, Q, R, init_x, init_V, ...) % % INPUTS: % y(:,t) - the observation at time t % ...
源于1957年他在哥伦比亚大学的一篇博士 论文和1960年发表的论文《A New Approach to Linear Filtering and Prediction Problems》,60年代Kalman 把状态空间模型引入滤波...
y〔n〕=Cx〔n〕 为 A = 〔1.1269-0.49400.1129 1.0000 0 B = 〔-0.3832 0.5919 0.5191〕; C = 〔1 0 0〕; 我们的目标是设计 Kalman 滤波...
卡尔曼全名 Rudolf Emil Kalman,匈牙利数学家,1930 年出生于匈牙利首都布达佩斯。...4. 简单例子 (A Simple Example) 这里我们结合第二第三节, 举一个非常简单...
(cm == 'A') mtype = 2; else error('Unkown measurement model.'); ...(sd,T.nmeas); % Kalman gain % Initialize filter L_init = L_max/2.0...
Kalman Filter介绍_数学_自然科学_专业资料。卡尔曼滤波(kalman filter)最简单...4 简单例子(A Simple Example) 这里我们结合第二第三节,举一个非常简单的例子...
(A,[B B],C,0,-1,'inputname',{'u' 'w'},'outputname','y'); Q = 1; R = 1; [kalmf,L,P,M] = kalman(Plant,Q,R); kalmf = kalmf(...
Kalman 滤波 在学习卡尔曼滤波器之前,首先看看为什么叫“卡尔曼”。跟其他著名...(k-1k-1)对应的 covariance,A’表示 A 的转置矩阵,Q 是系统过程的 ...
Kalman滤波中文版 17页 1下载券 kalman滤波的C与C++实现 10页 1下载券 ... Kalman滤波的公式推导...
热门文档