A new signal processing technology to eliminate range sidelobes in meteorological radar data

传统气象雷达的工作原理是向大气层发射微波脉冲，并测量云层和降水粒子的回波。然后，这些回波在许多脉冲上的总和（“雷达反射率”）可以用来推断这些粒子的特性，例如降雨率。雷达的反射率在很小的距离上往往会有很大的变化，例如层积云可能只有200米厚，而雷暴云的大雨可能只有几公里宽。因此，需要高分辨率的测量。这取决于两件事。首先，通过使用大型天线，雷达的笔形波束可以变得非常窄，这对于大多数云和降水雷达来说是简单的。第二，我们需要沿着该笔形波束的长度的高分辨率沿着。这种“距离分辨率”取决于雷达发射脉冲的持续时间。短脉冲导致高距离分辨率，这正是我们想要的。长脉冲导致粗略的距离分辨率，这将无法分辨我们希望探测的云和雨细胞。不幸的是，这是一个权衡：长脉冲对弱回波的灵敏度更高，而短脉冲的灵敏度较差。脉冲压缩旨在绕过这种权衡。为了高灵敏度，长脉冲被传输，但是额外的信息在短时间尺度上被编码到这些长脉冲中。反射回天线的回波然后被“解码”成所需的高距离分辨率。这种同时具有高灵敏度和高分辨率的能力使得脉冲压缩成为气象雷达的一项极具吸引力的技术。然而，它有一个主要的实际缺点，这是距离旁瓣的形成。由于解码过程永远不会完美，因此一些信息会分散在范围内。这导致雷达数据的损坏以及云的特性和降雨率的错误测量。如果要充分利用脉冲压缩，这个问题必须解决。我们最近开发了一种简单的新技术，可以解决这个问题。这种方法在统计学和信号处理方面有着严格的基础，使我们能够识别距离旁瓣发生的位置，甚至纠正损坏的数据。事实证明，大自然为这个问题提供了解决方案，它通过使云和降水中的粒子每隔几毫秒就相对于彼此“重新洗牌”来解决这个问题，这是湍流、风切变和下落速度变化的结果。这会导致雷达测量到的回波波动--而每一次波动都是独一无二的，就像指纹一样。为了确定来自一个范围的回波在哪里泄漏到另一个范围的数据中，我们在不应该出现的地方寻找指纹的痕迹（我们将两组波动关联起来）。如果存在显著的相关性，我们可以确定数据已损坏。此外，两组波动之间的相关程度告诉你有多少能量从一组泄漏到另一组，这一知识使我们能够纠正这种泄漏。到目前为止，我们的工作一直是理论分析，并将其应用于两个短样本的数据从云雷达。拟议的项目将发展的想法，我们可以证明，这可能是一个实用的技术，用于研究和国家气象服务雷达。我们将通过计算相关性测量中的误差来支持我们的理论-这决定了我们识别和纠正数据的准确程度。其次，许多雷达都是24/7工作的，因此我们将证明该技术的计算速度足够快，可以实时操作。第三，气象服务使用扫描雷达，发射“双极化”脉冲，这给这项技术带来了一些挑战：我们将演示如何将其应用于此类系统。

A conventional meteorological radar works by transmitting a pulse of microwaves into the atmosphere, and measuring the echoes from cloud and precipitation particles. The sum of these echoes over many pulses (the "radar reflectivity") can then be used to infer properties of those particles, such as rainfall rate. Radar reflectivity often varies substantially over small distances - for example stratocumulus clouds may only be 200m thick, while the heavy rain from a thunderstorm cell may be only a few kilometres wide. Because of this, high resolution measurements are required. This depends on two things. First the pencil beam from a radar can be made very narrow by using a large antenna, and this is straightforward for most cloud and precipitation radars. Second, we need high resolution along the length of that pencil beam. This "range resolution" depends on the duration of the pulse which the radar transmits. A short pulse leads to high range resolution, which is what we want. A long pulse leads to coarse range resolution which will not resolve the clouds and rain cells we wish to probe. Unfortunately there is a trade-off: long pulses give greater sensitivity to weak echoes, while short pulses give poorer sensitivity. Pulse compression aims to bypass this trade-off. Long pulses are transmitted for high sensitivity, but extra information is encoded into these long pulses on short time scales. The echoes reflected back to the antenna are then "decoded" into the desired high range resolution. This ability to have both high sensitivity and high resolution makes pulse compression an extremely attractive technology for meteorological radars. However it has a major practical drawback, which is the formation of range-sidelobes. Because the decoding process never works perfectly, some information is spread out in range. This leads to corruption of the radar data and erroneous measurements of cloud properties and rainrates. This problem must be solved if pulse compression is to be fully exploited.We have recently developed a simple new technique which may solve this problem. This method, which has a rigorous grounding in statistics and signal processing, allows us to identify where range sidelobes are occurring, and even correct the corrupted data. It transpires that nature provides the solution to the problem, and it does so by causing the particles in clouds and precipitation to "reshuffle" relative to one another every few milliseconds, as a result of turbulence, wind shear, and variations in fall speed. This causes the echo measured by the radar to fluctuate - and every fluctuation is unique, like a fingerprint. To determine where the echo from one range has leaked into data at another range, we look for traces of that fingerprint where it shouldn't be (we correlate two sets of fluctuations). If there is a significant correlation we can identify that corruption of the data has occured. Furthermore, the degree of correlation between the two sets of fluctuations tells you how much power has leaked from one to the other, and this knowledge allows us to correct for that leakage. So far our work has been a theoretical analysis, and its application to two short samples of data from a cloud radar. The proposed project would develop the idea to the point where we can demonstrate that this is likely to be a practical technology of use for both research and national weather service radars. We will underpin our theory with calculations of the errors in our correlation measurements - this determines how accurately we can identify and correct our data. Second, many radars operate 24/7 so we will demonstrate that the technique is computationally fast enough to operate in real-time. Third, weather services use scanning radars which transmit "dual-polarised" pulses and this creates some challenges for the technique: we will demonstrate how it can be applied such systems.