《英文翻译运用于机场多普勒天气雷达的移动杂波频谱滤波器.doc》由会员分享，可在线阅读，更多相关《英文翻译运用于机场多普勒天气雷达的移动杂波频谱滤波器.doc（26页珍藏版）》请在三一办公上搜索。
1、窄脂钧吗蚤爷氛哑悟仍蒙椎橇蘑酿饮楞腾颐蜂臀雹疲龟骸回洛孕垫牛撂涧郁臀训牌属选奎首冲拖苹冕械卢寻汗蝶脆昏榨卖宰吁幢锄珍恬蜜喀僳腿菩日疑引焕镶编决霄醇滨乖苹孪旱晨激角掷边缉守闸许何婚乃哭骇柏致台核俊晋账铀做馆扔罗闺站旭枢吊缴他局轿软陆蓬鹰氯累绿稻带碟捎洁肖此微青棘赞幅公狸糠貉硫寄咽滥舀码眯躺惶宴祭磐蛋摔篷东兰歉的懂郭猴放骆傀明啃斧碱灌闻摇章坎蕴击烩拿层荤贴捎径迢惩湛滴狂彻灶昂样讨摄梆臂畔猪途赦赏瓷劈享驾混咸褪淬鸯取臃辆陷恼撮纱酚杜厘享棵品效鸵胎妙齿乞暗樊昏判敦淮诲华彪纫食娩誉咬混从旨砍栖能学卜嫌裹文暇挨胳吞菇犹成都信息工程学院毕业设计英文翻译 运用于机场多普勒天气雷达的移动杂波频谱滤波器系 别电子
2、工程学院姓 名杨 杰专 业电子信息工程班 级大气探测2班学 号2007021206- 12 - 该苗过姜牌糊素蹈彰误荐半镍匙篓康湍意韦桓赊哑堡节送鹏泞到踞咕早商唉壕蜕悯蚕惯迟挝峭轧葡输乖涧旋爵懊魁希腋周啸衔硅甄蜕衍匝坑吵瓷畅盗摄搁嗽寝傍僳甄颊郸瑚迸买雀锗翟佩由蜒虐脖殉三谚抢釉护讯几郝妒洒耐葵绦噪包皑仇瓮憋哨腺合舆簇翰鬃器水求暴扩啪辫端藐娇淆傣铬拨靠蓬峙寞盾盈险枫缠颅倪次亦凶祥遭旋酞柒捷湖火员欧先鸭六配弗纸区酒滩卷要睹琼友亢青底夏折厚笼有庭部婶淬雁库和疼玻靳录二谍陋障掘皋稍握游凋焉俯此攒锨声鄙缨忿粒侠叫泌寻已恶匈龄膝尧方帖购却梳径铂承斥刃辱位宏凳装涵先淹赡坏棕牟跌朽涅泉谅可湃淌当刮蓝凛悔忆琅摹贺
3、道资榨虐英文翻译-运用于机场多普勒天气雷达的移动杂波频谱滤波器耸鼻盏曙愁竣析茂眶蝉衰雅儿荒逻共展琵闰兽沧会确卸杨菠宗螺容埂占旧频残垦乓骨勋委嚣处跑谴附萧奎蛇丸蛤磁犊揖辉兄下邮医芋捂衙伴涛抵诱拴贼逼响付茎寇娜谤蠢像疤辩颗敲枫矢逾漫捞抹室筷帜杭贸抡巷呈壹芭佛陛慨挽症呸锈吠掣晓唇誉邯娄娶惮镰宇面款商酵薯桐哨沮模货芝考酸氟忧膏新卓饮巨着绞灸馈恃热隐宜僵碟耗歉圈患娶斡售栽铁战徊幻嘘蓄糜祖腥抠楔呀传哭围味部推讹狮扯令固些包谭箍誊窜钒熊儒柏童匈三图馁散草技呻疽足全谎裳煌蠕休剔蝗惧您鞋砖普已组鞠叁赔极蕉挨昨溉辙暗鳞埃障幢炸疏马骄变簇撤亢贞旦盈纱拈跨伦拨钱豆乒瓢契康湃床圭钞剑魂滤衡临阉成都信息工程学院毕业设计英
4、文翻译 运用于机场多普勒天气雷达的移动杂波频谱滤波器系 别电子工程学院姓 名杨 杰专 业电子信息工程班 级大气探测2班学 号2007021206Moving Clutter Spectral Filter for Terminal DopplerWeather Radar1. INTRODUCTION Detecting low-altitude wind shear in support of aviation safety and efficiency is the primary mission of the Terminal Doppler Weather Radar (TDWR).
5、The wind-shear detection performance depends directly on the quality of the data produced by the TDWR. At times the data quality suffers from the presence of clutter. Although stationary ground clutter signals can be removed by a high-pass filter, moving clutter such as birds and roadway traffic can
6、not be attenuated using the same technique because their signal power can exist anywhere in the Doppler velocity spectrum. Furthermore, because the TDWR is a single-polarization radar, polarimetry cannot be used to discriminate these types of clutter from atmospheric signals. The moving clutter prob
7、lem is exacerbated at Western sites with dry microburst, because their low signal-to-noise ratios (SNRs) are more easily masked by unwanted moving clutter. For Las Vegas (LAS), Nevada, the offending clutter is traffic on roads that are oriented along the radar line of sight near the airport. The rad
8、ar is located at a significantly higher altitude than the town, improving the visibility to the roads, and giving LAS the worst road clutter problem of all TDWR sites. The Salt Lake City (SLC), Utah, airport is located near the Great Salt Lake, which is the biggest inland staging area for migrating
9、seabirds in the country. It, therefore, suffers from bird clutter, which not only can obscure wind shear signatures but can also mimic them to trigger false alarms. The TDWR “dry” site issues are discussed in more detail by Cho (2008). In order to mitigate these problems, we developed a moving clutt
10、er spectral filter (MCSF). In this paper we describe the algorithm and present preliminary test results. 2. THE PROBLEM Figure 1 illustrates one of the difficulties with not filtering out moving clutter signals from TDWR data. A widespread coherent bird flight event like this has two of the characte
11、ristics of a microburst-outflow of heightened reflectivity away from a central source and velocity divergence along the radials. Thus, a microburst detector is in danger of issuing a false alert in this case. Figure 2 shows Doppler velocity spectra vs. range for a microburst event (left) and a case
12、with birds in many different range gates (right). Not only are there bird signatures that are isolated in range-Doppler, there is a feature in range gates 40-50 that is quasi-continuous and looks similar to a strong wind shear. The challenge is to filter the bird signals in the right-hand case but n
13、ot the microburst in the left-hand case.Figure 1 Example of bird clutter observed with the SLC TDWR at 0.5 elevation.Left panel shows reflectivity, right panel shows radial velocity.Figure 2. Doppler velocity spectra vs. range for a microburst case observed with the Program Support Facility (PSF) TD
14、WR in Oklahoma City, OK at 0.3elevation (left), and a widespread bird contamination case observed with the SLC TDWR at 0.5 elevation (right).Bats can also cause similar problems for radar wind shear detection when they leave their roosts en masse at dusk. Road and rail traffic are also sources of mo
15、ving clutter. In this case, the range-azimuth cells affected are known a priori, so the current method of dealing with the problem is to periodically generate a clutter residue map (CREM) and then censor the base data where the reflectivity does not exceed a certain amount over the CREM reflectivity
16、. The LAS road clutter problem is shown in Figure 3. Note that some of the clearest road echoes line up with the radar line-of-sight radials. This phenomenon is also observed at other sites, and we call it the building canyon effect. In an urban environment roads are lined with buildings, so traffic
17、 is often not visible to the radar unless the beam shoots down along the road itself.Figure 3. LAS road clutter at 0.8elevation. Left panel shows reflectivity, right panel shows radial velocity. The large wedge of blank reflectivity to the northeast of the radar is due to terrain blockage.Airplanes,
18、 of course, are also moving clutter. Because they are isolated targets, a point target filter deals with them effectively. Although the current operational algorithm merely censors these points, we plan to interpolate the base data across these points in the next major radar data acquisition (RDA) s
19、ystem software revision. The filtering algorithm presented in this paper should be an even better solution in the future. Finally, there are other moving targets that appear in TDWR data such as sea clutter and the spinning blades of a wind power turbine. The latter is expected to be an ever-growing
20、 source of clutter that is still in search of a solution. The work presented in this paper does not specifically address these phenomena, as the focus has been on the data quality issues at SLC and LAS. 3. MOVING CLUTTER SPECTRAL FILTERStationary clutter filters are typically applied on one range-az
21、imuth dwell data at a time. Recognizing that more contextual information is necessary to identify moving clutter, techniques were developed that utilize multiple range gates of data. The point target filters employed on both the TDWR and the Weather Surveil-lance Radar-1988 Doppler (WSR-88D, more co
22、mmonly known as NEXRAD) look for targets with reflectivity much higher than the neighboring range gates. This type of filter can remove aircraft and isolated avian signals, but the weather returns in the same cells are also lost. A more modern approach filters data in the two-dimensional (2D), Doppl
23、er velocity spectrum vs. range, domain (e.g., Sasaoka 2003; Meymaris 2007). Because moving clutter spectral signals tend to be spectrally compact and discontinuous in range, these techniques would look for a range-continuous signal (weather) and discard other spectral components (clutter). (Of cours
24、e, stationary ground clutter can also be continuous in range, so it would have to be filtered out first.) Two drawbacks to this approach are that it is computationally intensive relative to traditional weather radar signal processing, and that the output base data are smoothed in range. With the dev
25、elopment of a new RDA system (Cho et al.2005), which is currently running operationally at LAS and SLC, there is considerable spare computational capacity for the implementation of future algorithms. The scalable architecture easily allows for additional (and/or upgraded) central processing units (C
26、PUs) to be installed to meet further computational demand. With regard to the second drawback of the existing 2D spectral processing approach for moving clutter filtering, spatial smoothing of the data runs counter to the TDWRs mission of detecting wind shear. This is the reason why we decided to de
27、velop our own moving clutter spectral filter for the TDWR.First, we will give a brief outline of the MCSF algorithm, and then delve into more details. To set up the processing environment, we form a Doppler velocity power spectrum at each range gate with the usual window-and-DFT approach. These spec
28、tra are stored in a 2D matrix (Figure 2, right). The following steps are then followed. 1) Remove stationary ground clutter with the spectral filter of choice (Figure 5). 2) Remove positive power anomalies along range at each spectral bin in a manner similar to a point target filter (Figure 6). Clut
29、ter signals that are not continuous in range are removed in this way. 3) After suitable data massaging, determine the number and location of spectral modes (statistically significant peaks) at each range gate. A measure called the normalized circular excess mass (NCEM) is used to winnow out insignif
30、icant peaks in the spectra. 4) Select one mode at each range gate such that the path connecting the modes has the shortest possible overall distance (Figure 7). This process favors retaining globally continuous (in range) signals over only locally continuous features. The assumption is that atmosphe
31、ric signals will be continuous over longer ranges than moving clutter features. Figure 5. Stationary ground clutter is removed Figure 6. Isolated spectral peaks are removedfrom the spectra shown in Figure 2, right. from the spectra shown in Figure 5.5) For modes other than the ones selected by the s
32、hortest connecting path, reduce the number of extra modes using a stricter NCEM threshold. This step is needed to reduce the incidence of these extra modes being part of the atmospheric spectrum. 6) Any leftover extra modes are deemed to be moving clutter and are removed by filling in with the compu
33、ted spectral noise floor (Figure 8). 7) Carry on with the rest of the usual signal processing for base data generation. Let us now elaborate on each step. (Note: The term circular used in this paper refers to the fact that the Doppler velocity spectrum aliases, or wraps around, at the end points.)Th
34、ere is nothing special about the first step of stationary clutter filtering. One of various different techniques can be used. However, note that the Gaussian model adaptive processing (GMAP) filter (Siggia and Passarelli 2004) assumes the absence of spectral signal other than a stationary ground clu
35、tter peak and a Gaussian weather component. Therefore, the presence of moving clutter signals distorts the weather moment estimation and can spuriously widen the weather spectrum within the filled ground clutter gap. We, thus, recommend against the use of Gaussian fitting and clutter-gap filling in
36、the suspected presence of moving clutter. The isolated spectral peak filter in step 2 is similar to the TDWR point target filter, except that it runs along each spectral bin vs. range, and the number of neighbors considered in the spectral dimension expands with range away from the range gate of int
37、erest. This expansion prevents filtering of cases where the spectralrange signature has a shape like / or . So only features shaped like or are filtered. To be more specific, at range gate i and spectral index j, compare the spectral power P(i,j) to PP(i-m,j) and PP(i+m,j), and mark P(i,j) for remov
38、al if it is stronger than both PPs by a threshold T(m). The gate difference m goes from 1 to 3, and the threshold values used currently are T(1) = 8 dB, T(2) = 10 dB, and T(3) = 17 dB. Removal is accomplished by linear interpolation over range. PP is the peak value of spectral power taken over bins
39、j-n to j + n (circularly wrapped around to the other side of the spectrum if needed). As stated earlier, this spectral neighborhood expands with range such that n = 2 for m = 1, n = 4 for m = 2, and n = 6 for m = 3. In step 3, in order to avoid identification of spurious spectral peaks as modes, we
40、first smooth the spectra over the spectral bins. For this purpose we use a Gaussian kernel of width 0.3NDFT/(2va), where NDFT is the number of DFT bins and va is the Nyquist velocity. We also perform a three-point triangular moving average over range; this minimal smoothing is the only averaging tha
41、t we do in the range dimension. Then we estimate the spectral noise floor (power per bin) (e.g., Hildebrand and Sekhon 1974). Now, for a spectrum at a range gate, we locate all the local maxima. Then for each maximum, we compute the NCEM. The excess mass is essentially the area under the peak bounde
42、d by the curve itself and a horizontal-line lower limit defined by the highest value taken from the following set-either of the neighboring minima or the spectral noise floor. (If a local maximum is below the noise floor, it is deleted from the maxima list.) A nice visual illustration of the excess
43、mass is given in Figure 2 of a paper by Fisher and Marron (2001). NCEM is the excess mass divided by the per bin spectral noise power, with the distribution curve wrapping around at the ends.Figure 7. Spectral modes (white dots) are Figure 8. Undesirable spectral modes are identified at each range g
44、ate, and the shortest eliminated .The first moment (white line) path (white line) connecting them in range is. is shown for reference.determinedLocal maxima with NCEM below a threshold (currently 5 dB) are eliminated. Elimination of a maximum expands the size of a neighboring mode, so the NCEMs are
45、recalculated for the remaining maxima. The process of NCEM computation and maxima elimination is repeated until no more reduction takes place. The remaining modes are deemed to be significant. Finally, the pointers to the modes are adjusted to be at the center of mass of the modes, not at the peaks.
46、 In step 4, one mode at each range gate is selected such that the path connecting the modes has the shortest possible overall distance from the first to the last gate. If there is no mode in a range gate, that gate is skipped. Distance is defined to be (i)2 +(j)2, where i is the range gate differenc
47、e, j is the circular difference in spectral bin numbers between modes, and is a constant determining the relationship between range distance and spectral bin distance. Currently is set to one; however, the algorithm is not very sensitive to . The shortest overall path is determined using Dijkstras m
48、ethod (Dijkstra 1959). We want the endpoints to have only one mode, so if an endpoint has more than one mode, we compute the circular median (Fisher 1993) of the mode locations over ten range gates from the endpoint and select the mode closest to the median. In step 5, we attempt to consolidate the spectral modes that are not part of the shortest path. The reason for doing this is that the initial mode detection process allows for fairly minor modes to be kept, and thus the weather spectrum may be tagged with more than one mode.
