An Ultra-Efficient Radio Correlator Architecture

作为干涉仪工作的大基线射电望远镜阵列可以产生非常高的空间分辨率和动态范围的图像，但这通常是一项计算要求很高的任务。 特别是一个步骤，在阵列中的每对望远镜之间形成互相关以给出vibrance，具有随着望远镜数量N的增加而增加的令人印象深刻的缩放行为，随着该数量的平方增加。 信号处理链中的其他步骤只与望远镜的数量成比例，因此在计算上往往要便宜得多。 因此，随着望远镜阵列数量的增加，相关步骤占主导地位，这正是许多当前和计划中的仪器的趋势。 尽管现代计算机的运算能力有了很大的提高，但用于超大型干涉仪阵列的相关器电子设备预计将非常昂贵，并消耗大量的电能。华盛顿大学的Morales及其同事说。 该方法通过空间傅立叶变换来代替经典的相关步骤，并且在此之前进行网格化操作，该操作将数据置于规则间距上，从而允许使用快速傅立叶变换（FFT）。 众所周知，FFT的计算速度与N ln（N）成比例，因此当望远镜的数量很大时，FFT的计算速度比N的平方节省了很多。 网格化和其他计算操作引入的先决条件的问题只是线性的N，所以整体成像计算的干涉仪与许多元素应该减少，也许显着。新的，更快的相关器架构将探索与实验室原型，建立在FPGA（浮点门阵列）固件运行在标准的射电天文学计算硬件。 这一努力的最终目标是建立一个完全并行化的端到端演示装置，利用长波阵列和探测再电离时期精密阵列射电天文阵列的实际天空数据进行操作。 这项工作的资金由NSF的天文科学部通过其先进技术和仪器计划提供。

Large-baseline arrays of radio telescopes operating as interferometers can produce images of remarkably high spatial resolution and dynamic range, but it is in general a computationally demanding task. One step in particular, the formation of cross-correlations between each pair of telescopes in the array to give visibilities, has an imposing scaling behavior with number of telescopes N, increasing as the square of that number. Other steps in the signal processing chain scale only in proportion to the number of telescopes, and so tend to be computationally much cheaper. The correlation step therefore dominates as arrays grow in number of telescopes, which is precisely the trend for many current and planned instruments. Despite vast increases in the power of modern computers, the correlator electronics for very large interferometer arrays are projected to be very expensive and consume vast amounts of electrical power.A potential breakthrough in computational efficiency for interferometer data processing is being investigated by Dr. M. Morales of the University of Washington and coworkers. The approach replaces the classic correlation step by a spatial Fourier transform, and precedes this with a gridding operation that puts the data on regular spacings, thereby allowing the use of the Fast Fourier Transform (FFT). As is well known, the FFT computational speed scales as N ln(N), and therefore represents a substantial computational savings over N-squared when the number of telescopes is large. Gridding and other computational operations introduced to precondition the problem are only linear in N, so the overall imaging computation for an interferometer with many elements should be reduced, perhaps dramatically.The new, faster correlator architecture will be explored with laboratory prototypes that build on FPGA (Floating Point Gate Array) firmware running on standard radio-astronomy computational hardware. The culmination of this effort is intended to be a fully parallelized end-to-end demonstration unit operating on actual sky data from the Long Wavelength Array (LWA) and the Precision Array to Probe the Epoch of Re-ionization (PAPER) radio astronomical arrays. Funding for this work is being provided by NSF's Division of Astronomical Sciences through its Advanced Technologies and Instrumentation program.