Detectors INTERNATIONAL. P = (r n t n-1 ) e -(r t) ) / (n-1)! (1) And for rt << 1 it may be simplified to. P = (r n t n-1 ) / (n-1)!

Transcription

1 Detectors Member Observatory Global Network of Optical SETI Observatories INTERNATIONAL November 20, 2016 Modest electronics and a small telescope are capable of detecting extremely faint, brief light pulses against a stellar background. Two detection methods are described in the following pages. Both methods use photomultiplier tubes (pmt) in the pulse counting mode as the means of detection. The first method is one that was used here for about two years and then discarded in favor of a simpler, more effective second method. But to appreciate the means by which a very short duration pulse, e.g., <10 nanoseconds, can be detected against a stellar background it is first necessary to have some understanding of Poisson statistics as it relates to random rate of stellar background photon flux. Poisson statistics may be used to calculate the number of clustered events that can be expected to occur during a selected time interval given an overall rate of random events. For example, if a photomultiplier total photoelectron pulse count rate were 200,000 counts per second, then in an interval of 5 nanoseconds, one would expect to observe 2 photoelectron pulses 200 times per second, 3 photoelectron pulses 0.1 times per second and 4 photoelectron pulses only times per second i.e., for n=2, 3, 4 events respectively in equation 1 below. (In practice, and for a small telescope, 200,000 counts per second may be approximately equivalent to a star of magnitude 6 or 7). Continuing with the example, if one expected 4 photoelectron pulses at a rate of times per second, but detected a significantly higher rate, that would fall outside of the natural random distribution and would be indicative of something unusual; perhaps even the sought for laser signals. More explicitly, Poisson statistics determine the rate of events occurring within an interval as given by: P = (r n t n-1 ) e -(r t) ) / (n-1)! (1) And for rt << 1 it may be simplified to P = (r n t n-1 ) / (n-1)! where: r = the rate of random events per unit time t = time interval over which the measurement is made n = number of events detected in the interval. e =

2 If we confine detection to stellar background rates <10 6 counts per second and the detection interval to 5 x 10-9 seconds, and evaluate equation 1 for n=2, 3, and 4, the probable number of n photon events per second that are available to be detected may be seen in Figure 1. Figure 1. In the pages that follow, both of the detectors to be described take advantage of instrument settings where there can be a low probability of false positives, i.e., random stellar photons being interpreted as an incoming pulsed laser signal.

3 Method 1, Coincidence detection through the use of a plurality of photo detectors. This method continues to be the one most often used by others. In this technique, a laser pulse a few nanoseconds or less in length can be detected as a pileup of photons. Pulse pileup clearly results in increased photoelectron pulse amplitude during the brief interval; however, simple pulse amplitude detection has not been favored because large amplitude signals are commonly produced in photomultipliers by gamma radiation, corona effects and Cerenkov radiation. To avoid those interfering phenomena, it is usual to employ optical beam splitter(s) to feed multiple photomultipliers or avalanche photodiodes. When each of the n photomultipliers or photodiodes detects photon(s) that are coincident in time, an output is produced. An example of a 4 photomultiplier arrangement with beam splitters and used here a few years ago is shown below. Left. A small, 1 cubic inch, beam splitter divided the beam into four equal amplitude and equal length rays. Each pair of rays spaced 1/2 inch apart to match the input characteristics of the dual photocathode pmts. Right. The overall assembly of the photometer optics box is shown. The electronics were mounted on a cover panel, side facing.

4 Referring to the graph in Figure 1, a total count rate of 5x10 5 Hz, and using 3 detectors (n=3), each responding to a single detected photon, there would be 1.5 coincidence detections/second from the stellar background. If one has a large enough telescope the individual detector s low level discriminator thresholds can be set such that they produce an output only when there are two or more (total n=6) photoelectrons per channel. In so doing, the false positive rate can be reduced to near nothing, but of course, sensitivity is sacrificed. The latter case has been most commonly used by the various institutions doing optical SETI. I ll skip over any further description of Method 1 because the advantages of Method 2 make it a less complicated yet, I believe, a far more desirable detection system. Method 2. Coincidence and group pulse detection. The main feature of this technique is to detect groups of photons closely spaced in time in addition to coincident pulses. No beam splitters are used and only a single photomultiplier is required. The optics and circuitry are less complex and more easily adjusted. For smaller telescopes the typical photomultiplier detection rates for random stellar background photons range from about 10 2 to 10 6 counts per second. As regards Poisson statistics, the minimum photometer detection interval is adjustable in the photometer circuitry. The detector is sensitive to photoelectron pulse groups within that interval (and coincident photons) while rejecting pulses that are not closely grouped. It is normally set to detect 2 or more pulses per group occurring within about 25 ns, (more on this later). In this manner the stellar background noise is greatly reduced regardless of the raw background pulse rate. The photometer is routinely tested to detect pulsed signals as might be expected of an actual laser signal, i.e., 2-5 detected photons, against stellar backgrounds and test lighted backgrounds. With that level of pulsed light "intensity", the filtering electronics along with a computer based spectrum analyzer readily detects LED test light pulse groups from coincidence to 100 ns in duration at a 0.05 counts per second repetition rate with stellar background noise counts up to 500,000 counts per second. Since 2014 there have been a number of upgrades that have simplified the circuits, improved the sensitivity and made calibration a much simpler task. Below is a simplified drawing of this type of detector.

5 Figure 2. The photomultiplier output pulses are amplified and inverted then fed to a low level discriminator and an adjacent high level discriminator. The low level discriminator passes only those pulses that are above the noise and pulse ringing levels. The high level discriminator passes only those pulses that are greater than about one and a half times the normal photomultiplier output pulse, i.e., coincident events and other spurious phenomena. Next, the pulses from the low level discriminator are integrated such that pulses closely spaced in time increase the input level to a comparator. Note that two closely spaced photomultiplier output pulses will produce the same response as three pulses somewhat more spaced out in time. Also note that with closely spaced pulses (<15 ns between adjacent peaks), the low level discriminator does not fully recover; its output pulses are often broadened to encompass several input pulses. This is advantageous at the integrator/discriminator to more easily detect pulse groups. If the pulses fall within the desired selectivity, the comparator sends a pulse to an output pulse stretcher. While the target spread for photoelectrons is from coincidence to ~25 ns, any group of closely space photoelectron pulses greater than the minimum will be detected; out to 100 ns for example. Coincident pulses, detected at the high level discriminator, bypass the integrator /discriminator and are fed directly to the coincidence output pulse stretcher. Coincident and group pulse detections are summed and fed to a pulse stretcher that conditions the signal for the spectrum analyzer input. A computer based audio, fast Fourier transform (fft) type, spectrum analyzer (Spectrum Lab software)

6 scrutinizes the few (.05 to 20 pps) 50 us photometer output pulses for periodicity. This particular post detection analysis method is especially advantageous when observing dim stars and operating with low discriminator threshold settings, (i.e., n = 1.5 to 2) to achieve the highest sensitivity. Tying it all together, a detector designed to take advantage of the Poisson distribution at group numbers between 2 and 4 has the following advantages over the coincident only method. It: requires only a single photomultiplier, works equally well with coincidence and multiple pulse detection, simpler setup and calibration and no concern for channel crosstalk, eliminates signal losses related to beam splitters, substantially reduces the affects of pmt dark noise, suppresses random stellar background noise to a low level, simplifies pmt cooling, (if needed), sensitivity can be adjusted according to the total count, e.g. n=1.5 for dim stars and n=3 or 4 for brighter stars, longer pulse width requires less laser peak power for a given pulse energy, with more energy per pulse, but with the same laser peak power, more photons per pulse may be available for detection. The disadvantages of this detection method appear to be few. The high count rate associated with bright stars raises the background noise and reduces the signal to background count ratio (SBC). This can be offset by increasing the 2 nd discriminator setting, but at the expense of sensitivity. Since early 2016, the threshold is set automatically according to the stellar magnitude by the control computer. Alternatively, neutral filters or several optical bandpass filters, used sequentially, would improve the SBC, but would also incur a loss of sensitivity. Figures 6 and 7 below are complete schematics of this detection method. Figure 6. Pulse Amplifier (front end) above, for the circuit in Figure 7.

7 Figure 7. Photometer circuit, 6/16. Addition info on request. With care and frequent ground compartmentalizing, <2 ns rise time can be attained where needed with little or no ringing, no oscillations, and good all around performance. Surface mount devices have been incorporated using Schmart Board interface adapters. Another problem for any photometer is that of PMT corona hash or micro discharges. These noise bursts produce effects that are similar to the sought after signal, but being aperiodic and not much of an issue. However, a continuous purge of dessicated air into the photometer housing has proven effective in eliminating this problem. Unusually large pulses from gamma particles and Cerenkov radiation can also produce event detections, but these are only minor aperiodic contributors to downstream processing background noise and are suppressed at the spectrum analyzer.

8 The version photometer head. The head is mounted on Thompson rods and motorized for focusing. A remotely movable right angle mirror is used to redirect the rays to a high sensitivity video camera (LG, LCB5300-BN, not mounted). A 0.01 (250 micron) aperture plate provides a FOV of ~12 arc seconds and a low sky contribution to the background. Not shown is the LED pulser that was later mounted onto the assembly. Returning to the argument regarding laser peak power and using the 1 Megajoule with a 1 ns pulse and where the peak power would be 1 petawatt. If the pulse were lengthened to 50 ns, the peak power is reduced by a factor of 50 to only 20 terawatts for the same energy expenditure and total photon flux per pulse. The use of lasers transmitting the same energy but with longer pulses and with less peak power can reduce the facilities costs, intelligent resources and maintenance costs. If there are technical or economic reasons to do so, multiple lower power lasers might be employed to fire sequentially - filling out long pulse lengths and perhaps having different wavelengths for encoded data transmission. It is also reasonable to expect that lower power lasers can function with greater energy efficiency. For both reception and transmission the use of longer pulse lengths appears to be a winwin strategy. See the Amateur OSETI Viable? section for more definitive examples. Non-Poisson Group Pulse Detector Experimental Results. A prototype group pulse detector has been placed in operation that takes advantage of the Poisson distribution characteristics at n numbers of 1.5 and greater. In this device, an adjustable integrator/discriminator can be set for different timing intervals. Because the discriminator uses simple analog integration, the n number settings are only approximate. That is, two coincident or closely spaced pulses will produce the same results as three pulses that are somewhat more spread out in time.

9 Lower trace: typical triple photoelectron pulse detection from an LED 50ns test pulse. Upper trace: Output of the pulse width discriminator indicating a group detection. Using a computer based audio spectrum analyzer, the test pulses at 0.06 pps are greater than 10 db above the simulated stellar background at count rates up to 500 khz. The Spectrum Lab software is normally set for an fft size of , decimated by 16 for a bin width of 10.5 millihertz and no bin averaging. The autocorrelator function of the SpectrumLab software significantly improves the minimum detectable limit. Notably, the pulse detection rate can be set reasonably close to that predicted by Poisson statistics. However, for total count rates greater than about 500k, e.g. stellar visual magnitude >6, the soup rises to near the signal level. The 2 nd discriminator threshold level is automatically adjusted by the software to the target star s visual magnitude, i.e. n<2 for m v >11 up to n=3.5 for m v = 6. Above. Telescope control computer display (cerca 2014). The "Log Amp" display is a logarithmic analog conversion of the total count. LED test pulses at 0.6 Hz are on with a 25 khz background (LED) noise. Note the "Group Hits" are nearly 100 percent detected while the "Coincident Hits are less often and irregularly detected. This is because the LED test pulse amplitude is set for marginal "Group" detection and there are only few coincident photons to detect.

10 Above. On the right is the Spectrum Lab output screen. The white lines are in the waterfall presentation and are representative of the 0.6 Hz LED test pulses with a 100kHz background (LED) noise. On the left is the Stellarium stellar navigation program screen. This will be updated soon to include the autocorrelation feature. In the past, I ve been a harsh critic of my crude method for detecting grouped pulses and considered that a digital approach might be preferred. But, having worked with this simple hybrid technique for a long time, I now feel assured that it is not only the simplest method, but perhaps it is as good as it gets. The method has great flexibility and range without compromises that might otherwise be necessary. The overall sensitivity for detecting periodic pulses embedded in very large noise levels has exceeded my expectations and clearly exceeds methods employed by others. Extensive data collection over the test pulse range of 0.05 to 10 Hz has proven this. Additionally, using a recorder having a slow recording and fast playback speed, the range of detectable periodicity has been extended to Hz with only a minor loss in sensitivity. Additional information. The photomultipliers used are Hammamatsu R1548, rectangular, dual photocathode head-on units. The dual pmt feature made them ideal for coupling with beam splitters in the Method 1, coincidence detection mode. I continue to use them even though only one of the photocathodes/anodes is needed. They work well having low dark noise typically volts and 60 F in the grounded photocathode configuration. They have a gain of over 10 6, a nm spectral response and a peak quantum efficiency of As I understand it, the used R1548 tubes on the market were removed from PET scanners and have been relatively abundant. They have been available on ebay for several years at about $50 each. I purchased three and have had good experiences with all three.

11 One of the photomultipliers installed in a water cooled Peltier cooling fixture. To date, cooling has not, nor is it likely to be needed. In the near future, this photomultiplier will be replaced with one having higher quantum efficiencies over a much wider optical range. Photometer Purge Air Dryer The air dryer is comprised of a silica gel desiccant with an aquarium air pump forcing air through the media and to the photometer housing. With a low air flow, the silica gel requires rejuvenating (baking at 250 deg.f for two hours) after about hours of observing time. Early tests with this little system used a Peltier cooler to condense water out of the airflow prior to entering the desiccant jar. This was later determined to be unnecessary. At Boquete, during the rainy season, the relative humidity is in the 90 percentile range and photomultiplier operation without a dry air purge is unsatisfactory. However, with the dryer, no corona has been be detected. During the dry season the relative humidity is generally around 50% and the dryer is not as necessary.