Design Considerations for Radio Astronomy. Rick Perley National Radio Astronomy Observatory Socorro, NM

Transcription

1 Design Considerations for Radio Astronomy Rick Perley National Radio Astronomy Observatory Socorro, NM

2 Who Am I? Staff Scientist at NRAO, located in Socorro, NM. A native of British Columbia Lived in many cites, High School in Kelowna BSc at UBC. In 1967 got NRC summer job in Ottawa Discovered radio astronomy Discovered wilderness canoeing (ARO, Lake Traverse, ON). MSc at new astronomy program at UBC, then decided I had to leave BC. PhD from U. Maryland in 1977 Used the E-W arm of the `TPT, at Clark Lake in southern California to do aperture synthesis at MHz Dry (usually) lake bed. One of the hottest, emptiest, and most forlorn places on earth I spent a lot of time here living in a trailer 2018 David Dunlap Summer School 2

3 Bill Erickson, and Clark Lake 2018 David Dunlap Summer School 3

4 Who Am I on to NRAO and the VLA My CLRO experience apparently impressed NRAO staff, so Postdoc at NRAO (Socorro) first postdoc at the new VLA! Staff scientist in Major research areas: radio galaxies, quasars Strong interest in calibration/imaging/polarimetry Driven by the need to improve VLA imaging capabilities Was EVLA Project Scientist from 1998 through Am now just trying to make the JVLA work as well as we designed it to be, and return to some science. VLA in `D configuration 2018 David Dunlap Summer School 4

5 What to Cover? The suggested title ( Design Considerations for Radio Telescopes ) is very broad. Cannot cover the entire field in 70 minutes Will instead discuss some key aspects of radio astronomy, guided by my own biases and experience: I m an interferometrist hence most examples are from that field I spend much time on calibration and imaging methodologies hence much emphasis here RFI is a major concern, hence some slides on this. I will briefly discuss upcoming Big Instruments esp. ngvla Interferometric polarimetry is really cool but I won t have the time to get that far. (slides are retained, however) David Dunlap Summer School 5

6 Top-Level Topics Sensitivity: Single Dish, Interferometer, Array Resolution, Field of View Brightness Sensitivity Survey Speed for interferometers Image Fidelity, Dynamic Range, Levels of Calibration Deconvolution Array Design Issues RFI Issues An Alternate View of the future for radio astronomy (Polarimetry) 2018 David Dunlap 6 Summer School

7 Radio Telescope Sensitivity The basic radiometer equation is derived in hundreds of references: 2kTsys p K Watt m -2 Hz -1 ha A D Dt Defined as flux density of a source whose output increment equals the rms noise. Important to understand the symbols: K is a system efficiency factor, defined w.r.t. a perfect analog system. k is Boltzmann s constant (1.38e-23 Watt/Hz/K) T sys is a measure of the noise produced by the system (K) A is the physical collecting area of the sensor (antenna). (m 2 ) h a is the aperture efficiency of the antenna. Dn is the bandwidth (Hz) Dt is the integration time (sec) David Dunlap Summer School 7

8 Single-Dish Sensitivity (cont.) In terms of the well-known unit Jansky, 1 Jy Watt m 2 Hz T p K ha A D Dt T sys comprises many contribution sys Jy T sys T rec T pol T atm T gnd T CMB T src Receiver Polarizer Atmosphere Ground 2.7K Source IMPORTANT: In nearly all cases, T src << T sys For a 25-meter antenna, with the best coherent receivers, a `typical 1 Jy source provides ~ 0.1% of the power! Heroic (a.k.a. expensive) efforts needed to improve this problem 2018 David Dunlap Summer School 8

9 System Equivalent Flux Density (SEFD) A useful concept is that of the SEFD the flux density of a source which double the system temperature (power). 2kTsys 26 SEFD K 10 Jy ha A The radiometer equation then looks like: SEFD p Jy D Dt The SEFD is a telescope constant: For the VLA s 25-meter antennas, at (say) X-band, SEFD = 250 Jy. For the 100-meter GBT, SEFD~ 15 Jy. So, to get high detection sensitivity, we need either/both: Low values of SEFD (good receivers, LOTS of area) High values of D Dt (wide bandwidth, long times). Example: For GBT, with SEFD = 15 Jy, Dn = 1 GHz, Dt = 3600 seconds, ~ 8 mjy David Dunlap Summer School 9

10 Sensitivity of a Two-Element Interferometer Presume two identical antennas, whose signals are correlated (multiplied). The rms fluctuations of the product (single correlation): p h SEFD 2D Dt Here, SEFD is for each antenna, and h c is the correlator efficiency, again defined w.r.t. the perfect analog multiplier. For the JVLA, h c ~0.95. If we have N antennas, then we have N(N-1)/2 unique baselines, so we get (N >> 1, assuming all antennas are identical), p And if we combine two correlators (i.e., form I, Q, U, or V), c SEFD h N D Dt c c SEFD I h N 2D Dt Jy Jy Jy 2018 David Dunlap Summer School 10

11 Array Sensitivity, Key Factors The last expression is the best you can get rarely attained. Presumes all pairs have equal sensitivity, and all pairs summed equally. We (almost) never do this. Obtaining a good PSF nearly always requires downweighting shortspacing data this decreases sensitivity. More on this later. Key Points: Most important factor is low SEFD This requires good receivers (low Tsys), lots of effective collecting area (high NhA) But, note that T rec for all frequencies < 100 GHz are about as low as they can be (near the quantum limit T = hn/k), so to get more sensitivity we can only add collecting area. Improving efficiencies helps, but not by an order of magnitude. Less important factors are Bandwidth and Integration Time David Dunlap Summer School 11

12 Diffraction and Resolution In radio astronomy, we are usually in the coherent regime phase paths are stable (to << l), so electric fields are summed coherently. In this case, the well-known, simple diffraction relation applies: l q D (radians) (where both l and D are in the same units) The interpretation varies a bit between single-dish and interferometer: For the single dish, this is ~ the FWHM of the antenna beam. For a single baseline, this is the fringe spacing, with D = baseline. For a synthesis array, this is ~ FWHM of the synthesized beam, with D = longest baseline (hereinafter = B). Example: For VLA, longest spacing = 35 km. X-band wavelength = 3.5 cm. Hence, q ~ 10-6 rad, or 0.20 arcseconds David Dunlap Summer School 12

13 Diffraction and Field of View single dish For a (steerable) single dish, the field of view is given by the range of angles over which the antenna points. Resolution given by the antenna primary beamsize small w.r.t. 4p steradians. Focal Plane Arrays (multiple horns) now common. NB radio telescopes are really fast f/d ~ 0.3, so coma is a problem. Shown here is the Parkes 13-beam focal plane array. Each horn responds to a different point on the sky. Speeds surveys by factor ~ 13. But sub-nyquist sampling. Can beat this with Phased Focal Plane Arrays at cost of computing David Dunlap Summer School 13

14 Field of View interferometers In synthesis imaging, one normally tracks the object of interest. q Field of view now given by the antenna primary beam Resolution given by the maximum baseline. Number of resolution elements within the FOV ~ (B/D) 2. Example: For VLA, N~(35000/25) 2 ~ 2x10 6. However, one can combine scanning and interferometry OTF imaging (super-sidereal tracking). On The Fly imaging combines array antenna motion with interferometry to permit much faster surveys at high resolution. Appropriate for wide, shallow surveys This is now implemented on the JVLA. The VLASS (VLA Sky Survey) employs this mode. q fov psf l/ D l/ B 2018 David Dunlap Summer School 14

15 Illustration VLA s Primary Beams at S-band These are voltage (E-field) responses. There are also phase images. First sidelobe (in power) about 5%, 2 nd sidelobe about 2%, of peak. Primary Beams from the JVLA from 2.0 GHz (top left) through 4.0 GHz (bottom right). Panels are 45 arcminutes on side. Circular symmetries from the antenna edge. Other symmetries due to quadruped legs, etc. Neat fact: FT of these complex images gives the electric field distribution over the antenna aperture David Dunlap Summer School 15

16 Surface Brightness Sensitivity The prior sensitivity argument is applicable to point sources. Point-source sensitivity is not a function of the location of your array elements. Synthesis arrays are designed to resolve objects. What is the sensitivity of an array to emission whose angular extent is larger than the resolution element? There is no exact answer here (it depends on array design and source distribution details) The following argument is good for pedagogical purposes, and permits decent estimates. But first, some review 2018 David Dunlap Summer School 16

17 Flux Density and Brightness (Intensity) The Flux Density, S n, of an object is given by: S B d Jy B n is the brightness, or specific intensity. As the units of S are Jy (= watt.m -2.Hz -1 ), then: The units of B are Jy/(solid angle). For example Jy/beam, or Jy/arcsec 2, or watt.m -2.Hz -1.ster -1. In radio astronomy, we commonly use brightness temperature the physical temperature of a perfect blackbody emitter which give the observed brightness. In the RJ approximation (good for cm wavelengths), T B 2 l B 2k K object 2018 David Dunlap Summer School 17

18 Brightness Temperature, cont. Use of Brightness Temperature is convenient and useful: `Degrees are simpler than watt.m -2.Hz -1.Ster -1 For thermal emission, the values relate directly to physical temperature and optical depth. T B T obj ( 1 e ) T if 1 obj 2018 David Dunlap Summer School 18

19 Returning to Brightness Sensitivity We now define the surface brightness sensitivity as: I B An important, and useful relation can then be easily derived by converting to temperature units, inserting the definitions of array sensitivity and synthesized beam, to get: Tsys T K harray 2D Dt Here, h array is the Array Covering Efficiency : h array 2 B It is the fraction of the array area covered by antennas. syn Nh a A 2018 David Dunlap Summer School 19

20 Some Examples For an array of fixed collecting area, brightness sensitivity is a strong function of antenna placement. Example: For VLA: h array ~ in `D configuration. VLA in D Configuration: h ~.017 Proposed E Config, h~ David Dunlap Summer School 20

21 Highly Packed Arrays Some close-packed arrays, offering high surface-brightness sensitivity. Upper right: ALMA Lower right: SKA-Mid Below: SKA-LFAA 2018 David Dunlap Summer School 21

22 Empty Arrays: These high resolution arrays have very high point-source sensitivity, but very low brightness sensitivity. VLBA has no chance of ever imaging a thermal object! VLA in `A Configuration h array ~ 0.002% VLBA (antennas not to scale!) h array ~ David Dunlap Summer School 22

23 Filled Arrays, and Single Dishes Arrays comprised of parabolic dishes cannot be packed too closely shadowing rapidly shrinks the available sky. If super-sensitive low-brightness emission is a key goal, there are two ways to get high (full) aperture coverage: A single dish (total power). Filling factor = 1. You can t do better. But, the cost is in resolution. Also, large single antennas (especially if moveable) are pricey. A filled aperture array. A sea of elements, packed tight, each seeing the whole sky. Elements can be cheap (dipole, horn), but electronics gets expensive and calibration can be complicated. Element cross-coupling is a major issue. Elementary array theory does not apply if elements are in communication. For simplicity, and clean optics, a parabolic antenna with a single feed can t be beaten! 2018 David Dunlap Summer School 23

24 The Ultimate Close-Packed Arrays Single dishes give (by far) the highest brightness sensitivity, but at a cost of resolution. GBT (100 m) (offset parabola) Arecibo (300 m) (fixed spherical) FAST (500 m) (deformable spherical) 2018 David Dunlap Summer School 24

25 Survey Speed Surveys have always been important to astronomy. Increased interest in transient phenomena, plus the renewal of low-frequency astronomy have increased interest in surveys. Surveys take a lot of time (sky is big, beams are small ). Key question is: How long does it take to survey a given sold angle, to a given point-source sensitivity? Easy application of the sensitivity and diffraction beam equations. The result: T Tsys hah cs lim 2 NA array 1 D h s 1 N beams 2018 David Dunlap Summer School 25

26 Surveying Speed (cont) T Tsys C hah cs lim 2 NA array 1 D h s 1 N beams There are many factors which influence survey speed: Major factors: Low Tsys (high sensitivity) critical Low frequencies much faster (beam size). Important factors For a given array area (A array ), more antennas preferable: => smaller! Multiple beams (N beams ) Survey efficiency (h s ) A note in survey efficiency: this includes calibration, setup, and moving the antennas from pointing center to pointing center. If the antenna move times become comparable to the integration time per pointing, the survey becomes highly inefficient. (Shallow surveys). For shallow surveys, OTF imaging becomes attractive David Dunlap Summer School 26

27 Speeding Up Surveys Multiple Feeds Wide-field surveys to reasonable depth can take years. Multiple fields of view offer a significant speed-up. Two ways to go: FPAs (focal plane arrays), and PAF (Phased Aperture Feeds) FPA: N independent feeds. PAF: N elements coherently summed to form M simultaneous multiple feeds David Dunlap Summer School 27

28 Phased Aperture Feeds (EVLA Memo 53) see htttps:library.nrao.edu/evla/shtml Traditional Single-Pixel feed must have aperture size appropriate to f/d and wavelength:. x~lf/d. This is larger than the Nyquist offset, meaning individual horns (FPAs) undersample the sky. To get everything at once, use an array of closely-spaced undersized elements, and phase them together to synthesize a large number of effective horns. Many cool features of this approach! Huge speed-up of surveying: ASKAP is ~100 times faster than VLA Image: SMC using 3 nights observing, 16 antennas, single pointing. ASKAP PAF 2018 David Dunlap Summer School 28

29 Speeding Up Surveys Multiple Feeds HERA (Hydrogen Epoch of Reionization Array). ~ meter fixed antennas, under construction in South Africa. Not an imaging interferometer! 2018 David Dunlap Summer School 29

30 Surveys with the VLA VLA has been used for many surveys Most famous are NVSS and FIRST (20cm, D and B configurations, respectively). Now in progress: the VLA Sky Survey (VLASS) All visible sky (d > -40), n = 2 4 GHz, q = 2.5 arcseconds. ~ 0.12 mjy, 5x10 6 objects Cadence: 3 times, each separated by 16 months hours for first 2 epochs. First epoch data taken. Unlike previous major surveys, VLASS is sponsored by NRAO we will be responsible for survey management, calibration, and delivery of data products. Science drivers include: Transients, episodic events, Faraday Tomography (wideband polarimetry), AGN, galaxy evolution in concert with new optical/ir surveys (e.g. LSST) David Dunlap Summer School 30

31 Image Fidelity How good is an image made by a synthesis array? How can we tell if it s right? Are low, flat noise residuals sufficient to assure us the image is good? Does self-calibration (radio adaptive optics) repair everything? When are heroic methods needed in calibration? These subjects are addressed in the next set of slides David Dunlap Summer School 31

32 Dynamic Range vs. Image Fidelity Dynamic Range is usually defined as: DR = Image Peak/Image Noise This makes sense for a field comprised of unresolved objects. For highly extended objects, we are more focused on the Fidelity. This a related, but different metric. Image Fidelity can be limited by a number of effects: Gaps in the spatial frequency ( u-v ) coverage Errors or insufficiency in the deconvolution/selfcalibration process Errors in the instrumental response and in calibration Whereas, Dynamic Range is much less strongly affected by the first two of these. Dynamic Range is primarily a diagnostic of instrumental performance and calibration methodology David Dunlap Summer School 32

33 Stepping through Calibration We now present a series of images of 3C147, using a single 1 MHz channel, at 1500 MHz, using 6 hours of VLA D configuration data (resolution ~ 45 ). 3C147 is a perfect point source makes interpretation easier. I show images of the field after various consecutive stages of calibration These indicate how much specialized calibration is needed to achieve a certain imaging fidelity David Dunlap Summer School 33

34 No Calibration! Suppose we don t calibrate at all D-configuration peak =.014 Jy, rms = 1.2 mjy/beam. DR ~ 12. This is not a random noise image because the (incorrect) amplitudes and phases for each baseline are quite constant over time. Each baseline contributes a rough Bessel-function, centered at the origin. No Calibration: DR ~ David Dunlap Summer School 34

35 Regular Antenna-Based Calibration Now we calibrate using the nearby point source. A nice image but the background sources are barely visible. Pk Jy/beam Rms 4.2 mjy/beam DR ~ 5000 This is typical for a regular calibration regimen. Regular Calibration: DR ~ David Dunlap Summer School 35

36 Antenna-Based Phase Self-Calibration The technique of selfcalibration is well-suited to the VLA. 3C147 so dominates the field that a simple point-source model is sufficient. Presume errors are associated with antennas (not baselines). Make a solution every second. Image much better but still far from what we aim for. Pk = Jy/beam Rms = 0.7 mjy/beam DR = 30,000 Phase Self-Calibration: DR ~ David Dunlap Summer School 36

37 Antenna-Based Amplitude Self-Calibration Normally, amplitude selfcalibration results in only small improvements But, at the current level of DR, amplitude variations important. Amplitude self-calibration again triples the DR: Pk = Jy/beam Rms = 0.31 Jy/beam DR = 68,000 More self-calibration loops, using this model, do not improve the image! Amp and Phase Self-Calibration: DR ~ David Dunlap Summer School 37

38 Baseline-based time-independent calibration The image is far from sensitivity limited the rms expected is about 0.09 mjy/beam we re short by a factor of at about 4. The swirly pattern provides a strong clue of what the problem is this looks like the pattern in the uncalibrated image. Such patterns occur when an error is specific to a baseline, and is constant over time. Postulate a small closure error, specific to each baseline, unchanging in time. AIPS provides a nice baseline-based calibration task: BLCAL Give it a try using the best image so far, and solving for a single timeinvariant multiplicative baseline-dependent solution David Dunlap Summer School 38

39 Baseline-Based Calibration Doing this makes a big improvement: Pk = Jy/beam Rms.09 mjy/beam DR = 230,000 This image is close to noise limited everywhere. So this really works! And if we have ~200 channels, we might expect to reach a DR of about 3,000,000. Can we? Closure Calibration: DR ~ David Dunlap Summer School 39

40 Rick s Best No DD gains, 64 MHz BW C+D config data 3C147 spectrum flattened. BW = 64 MHz Rms = 26 mjy/b Rms in corners: 12 mjy/beam DR = 1.76 million Note the disturbances at the bottom in the first sidelobe. DR ~ David Dunlap Summer School 40

41 Next Level Direction-Dependent Gains The process is like peeling an onion each layer removed reveals another layer. The next level is the variation of gain parameters with angle. Caused both by spatial variations in the screen (atmosphere), and by variations in the primary beam amplitude and phase. Correction greatly complicated by rotation of the beam on the sky with parallactic angle (alt-az telescope mounts). Oleg Smirnov (SKA-SA) has Meqtrees software package to manage this David Dunlap Summer School 41

42 Oleg s Best using 384 channels Rms in center: <10 mjy/beam Rms at edge: 7 mjy/b DR: 3.2e6 Easy seen the background sources in the first mainbeam sidelobe. Even some sources in the 2 nd sidelobe. Oleg s current best with these data: 5.5e6:1 Breaking News: We ve maxed out at about 8e6: David Dunlap Summer School 42

43 Rick s and Oleg s Best main beam Noise now so low that the image in the central beam is close to confusion limited. Higher resolution data taken to zoom in deeper. Noise at the center higher than on the edges of the map due to confusion noise David Dunlap Summer School 43

44 Why Bother with Macho Imaging? Are efforts to get > 10 6 fidelity a waste of time? For the VLA, most observing fields (especially at high frequency) are completely noise limited. For example at L-band with the VLA, a typical field has ~50 mjy maximum background source flux. With a (deep) noise of 1 mjy, only need > 50,000:1 DR. Easy! Only ~ 2% of the sky is within a VLA beam of the 1000 strongest sources. But imagine a `VLA on steroids (SKA?): 10 njy sensitivity Then a DR > 5x10 6 needed even for an average field. This issue is greatly exacerbated by smaller dishes larger fields of view will have more, stronger objects in them, requiring high fidelity. One of the critical factors is (u,v) coverage this is a design issue which we explore a bit here David Dunlap Summer School 44

45 Array Design, and U-V Coverage In synthesis imaging, the image (I) is related to the Visibilities (V) via Fourier Transform: I( l, m) V ( u, v) This is an analytic relation it presumes continuous knowledge of the visibilities over all distances, with no errors. (l,m) are direction cosines (not Cartesian coordinates!) for sky (u,v) are baseline coordinates, measured in wavelengths Real observations are always finite, and real data always have errors both (multiplicative) calibration errors, and (additive) noise. In reality, the visibilities are sampled by a function S(u,v), resulting in a dirty image. B(l,m) is the PSF I D Dirty Image What we got ( l, m) I( l, m)* B( l, m) V ( u, v) S( u, v) True Image What we want 2018 David Dunlap Summer School 45

46 How much is enough, and how to make the best of a given situation The conversion of I D (l,m) to I(l,m) is the subject of deconvolution a complex subject beyond the scope of this talk. Clearly, more visibility measurements are good for imaging. One gets more by: Adding more antennas (cost more $$). Observing over a long time (costs more time). Observing over more frequencies (but this introduces special problems) The sampling function (u,v plane coverage) at a given moment is the autocorrelation of the antenna locations distribution. The rapid improvement of imaging with number of antennas is shown in the next series of slides 2018 David Dunlap Summer School 46

47 Examples, using the VLA Two antennas -- single baseline, single observation, 1 Jy point source. UV-Coverage Image 2018 David Dunlap Summer School 47

48 More Antennas = Better Images! A single observation (with VLA) of a point source. 3 ant, 3baselines 4, 6 7, 21 10, 45 27, % 100% 93% 82% 41% = Pk. sidelobe 71% 41% 15% 11% 4% = rms. -280% -104% -32% -23% -10% = Pk. neg David Dunlap Summer School 48

49 To do better Observe over Time. UV Coverage and Images for a point-source for 0,1, 6, and 12 hours observing, with the 27-antenna VLA, at d = 60. Snapshot 1 Hour 6 Hours 12 Hours Pk Sidelobe 41% 14% 2% 1% rms 11% 4% 0.5% 0.3% Max. Neg. -10% -10% -5% -4% 2018 David Dunlap Summer School 49

50 Multi-Frequency Imaging One can dramatically improve the (u,v) coverage with a wide-band system and a wide-bandwidth correlator. Examples: 1 frequency channel at 5.5 GHz, and 8 frequency channels, spread over 1 GHz (from 5.5 to 6.5 GHz) -- 12% spread David Dunlap Summer School 50

51 Multi-Frequency Imaging 2:1 BWR The use of widebandwidth, multifrequency visibilities can dramatically improve the basic imaging characteristics. Shown is a 30 minute snapshot left side is single frequency, right side is 2:1 wideband. RMS narrow: 1.4% RMS wide: 0.2% Peak sidelobes: 12.3% vs. 2.1% 2018 David Dunlap Summer School 51

52 But Nothing Good Comes for Free Wide-band correlation has the potential for vastly increasing u,v coverage, and hence accurate imaging. But unless the sources have a completely flat spectrum everywhere there is a major cost in computing the deconvolution process. New software (multi-frequency CLEAN) under development at NRAO, and is available under CASA, to provide both total intensity and spectral slope (and even curvature, in some cases) David Dunlap Summer School 52

53 Wide-band C-band images of GRB080413A 75 minutes data, with 20-minute gap. Frequency span 4 to 8 GHz. Not enough time to effective reduce the Y arm sidelobes. RMS in Dirty Image: 33 microjy/beam. Dirty Image Dirty Beam (PSF) 2018 David Dunlap Summer School 53

54 Deconvolution Helps On left: Wideband deconvolution, with no accounting for spectral gradients. RMS = 9 microjy/beam. Not Pretty On right: Wideband deconvolution, including spectral gradients. RMS = 2.9 microjy/beam. Faint sidelobe residuals still visible David Dunlap Summer School 54

55 Array Design and Sidelobes Sidelobes. The characteristics of the dirty beam are critical in successful deconvolution. Height and distribution of sidelobes a reflection of the distribution of sampled points in S(u,v). The rays in the dirty beam are due to the oversampled rays in the (u,v) coverage. These greatly impact the efficiency and effectiveness of deconvolution. Reducing PSF sidelobes is a major goal of all array design. A completely uniform (u,v) distribution might be considered a worthy goal but in practice, this is not the best. (A T-Array design will give perfectly uniform sampling excellent for snapshots, not so for synthesis David Dunlap Summer School 55

56 Sidelobes, continued Since the worst sidelobes come from regular or organized structures in the antenna layout (think VLA s long arms), randomizing the antenna locations is clearly advantageous. This is used in nearly all modern array designs. (ALMA, for example). By all accounts, ALMA s imaging characteristics are excellent. Various algorithms have been applied to minimize sidelobe features. But not for the VLA why? The VLA s antennas are big and heavy -- > 200 ton. They are not easy to move around need a rail-borne transportation system. Also in 1975, nobody was thinking of snapshot imaging David Dunlap Summer School 56

57 Flaws in the VLA Array Design There are two serious (IMHO) issues with the VLA s array design: 1. Antennas are located along three straight lines. This gives a dreadful snapshot beam. VLA was never designed for this mode of operation. Reason: moveable 210-ton antennas need a railroad. 2. Antennas are centrally condensed. Combined with earth rotation, this causes an enormous oversampling of the short spacings. Nearly all VLA imaging has to downweight the central regions of the (u,v) plane to provide a decent PSF loss of sensitivity. Why? To preserve sensitivity with increasing taper (maximizes brightness sensitivity). 3. Question: How much central condensation is useful? Needed? 2018 David Dunlap Summer School 57

58 Array Design, and Surface Brightness I have shown earlier that surface brightness sensitivity is a strong function of areal coverage. If high brightness sensitivity is desired on a scale of q radians, then the region of high (u,v) plane sampling density must extend to r ~ q -1 wavelengths. Example: If the science case requires extra good surface brightness sensitivity for objects typically 1 arcminute in extent, then high sampling density out to r ~ 3600 l is desirable. The problem for any array designer is in balancing competing demands in the face of a finite budget. (i.e., the total number of antennas is a constant). You can t please everybody! Science cases featuring low-brightness extended emission, pulsars, and surveys, want the antennas close-in and tightly packed. High-resolution people want antennas spread out the compromise of keeping most in the center, with a few way out, is a poor one for them David Dunlap Summer School 58

59 Satisfying Everybody.? So how to make everybody happy (or, how not to drive your clientele out of your project)? Two approaches: 1. Reconfigurable Arrays. (VLA, ALMA, ATCA, WSRT, etc.) Move the antennas in and out in response to science needs. Advantages: You need fewer antennas to satisfy a wide range of resolution/brightness sensitivity needs. Disadvantages: Transporters, roads, and infrastructure costs go up. 2. Wedding Cake Array Design (SKA) Advantages: You sample all scales, all the time. No transporters are needed. Disadvantages: You need more antennas to span the spatial scales. You can also imagine a hybrid: Reconfigurable for short to intermediate scales (out to ~30km, say,) and fixed for larger. NG-VLA is considering this David Dunlap Summer School 59

60 Reconfigurable Arrays and their Transporters VLA ALMA ATCA 2018 David Dunlap Summer School 60

61 Fixed Arrays ASKAP ngvla LOFAR SKA-Mid (concept) 2018 David Dunlap Summer School 61

62 Design Considerations and RFI Radio Astronomy is not the only user of the spectrum! We are itty-bitty players in the spectrum utilization game. Old receiver systems were narrow-band Filters included which passed only safe bands with no RFI New systems are wideband 2:1 BWR ratios (e.g. 1 2 GHz) are the norm. This means that everything passes good and bad David Dunlap Summer School 62

63 Radio Telescopes Open to All A remarkable fact: All radio telescopes have about the same susceptibility to off-axis signals. Effective cross section is: A ~ l 2 /4p (to within a factor of a few). True for dipoles, whips, or 100 meter dishes! This is illustrated below (from Wikipedia). Beyond an angle of ~30 degrees, from the main beam, all antennas are about the same David Dunlap Summer School 63

64 RFI Limits for Radio Astronomy The traditional limit to tolerable external signals comes from a very simple argument: The interference power < 10% of the receiver noise fluctuations. This leads to an incredibly strict limit on external signals: A 1 watt isotropic transmitter, in 10 khz bandwidth would violate this limit at a distance of 6 x 10 6 kilometers (~10 x distance to the Moon). Interferometers, in general, are much less susceptible than single-dishes to RFI, due to fringe winding (source motion) but this is only effective for long baselines at high frequencies. There s no way radio astronomy will ever get protection like this over vast swaths of spectrum. The problem is now much worse than before partly because RFI is increasing, but mostly because we are using hugely wide bandwidths! 2018 David Dunlap Summer School 64

65 Managing the Radio Spectrum The radio spectrum is heavily utilized by a wide variety of services. Management of this limited resources is a complex, laborious, global business. The stakes are high. A good top-level summary is given in Harvey Liszt s RFI webpage: There are ~45 narrow slices of spectrum reserved (i.e., have some sort of legal protection against transmitters) for radio astronomy, between 13 MHz, and 272 GHz. Narrow is NARROW typically, <1% of the frequency. Notable exception is the MHz band. Protection is often very weak, and enforcement is difficult. The following slide gives an idea of the multiplicity of users David Dunlap Summer School 65

66 Chart Showing the US Frequency Allocations Showing the permitted spectrum usage within the USA Radio astronomy is bright yellow David Dunlap Summer School 66

67 RFI in the HI Band! From the SMOS (Soil Moisture and Ocean Salinity) satellite system. Operates in the globally protected HI band ( MHz). But RFI is seen as shown below David Dunlap Summer School 67

68 The Radio Spectrum (as seen from the VLA) So what s the neighborhood like? The VLA (not in any protected zone) conducts regular RFI Sweeps to monitor the external environment. There is little change from day to day, or month to month. Most significant RFI is from satellites, airplanes, and fixed communications. Some examples shown in the following slides. Good news is: Even the most congested bands are still remarkabaly empty David Dunlap Summer School 68

69 1 3 GHz Logarithmic Power Scale 1.0 GHz 2.0 GHz x Air Navigation DMEs GPS INMARSAT, Glonass GPS Cell Phones Sirius/XM Filter Roll-offs Even in this most heavily used spectrum, ~60% of the spectrum is useable. (Uncalibrated data) David Dunlap Summer School 69

70 5 7 GHz 10x Microwave Links These links are perfectly legal. They broadcast from towers located on surrounding mountains. The signal strength is a strong function of antenna location David Dunlap Summer School 70

71 What to do? We can t possibly evict all the other users of the spectrum. So we must: 1. Design for high linearity. High input powers drive receivers into a non-linear response this causes harmonic mixing. Digital equipment (samplers, correlators) may need high bit depth. (8 to 12 bit samplers are the normal for low frequency applications). 2. Locate observatories in remote locations, far from civilization as practical. SKA sites are in the most RFI-free locations that could be found. 3. Employ tight emission standards for local equipment. Telescopes and correlators (and everything else) are all digital now local emissions must be tightly controlled. All radio observatories have programs and RFI protection groups to establish and maintain emission standards for local equipment David Dunlap Summer School 71

72 RFI Excision or Removal Despite the best electronics design and observatory location, there WILL be RFI in the data. Can we do something about this, post-observing? Considerable work in this area. The following is derived from a presentation by Michael Kesteven (CSIRO), at Groningen in I divide this into two parts: 1. RFI Removal. We subtract the unwanted signals, leaving behind the good astronomical data. 2. RFI Excision. We get rid of the corrupted data including the astronomical portion. Obviously, the first is more useful and much more difficult David Dunlap Summer School 72

73 Time Excision Algorithms. Harmful RFI nearly always is sharply bounded in frequency and/or time. If so, it easily shows up in a high resolution frequency/time plot. A good example from VLA Low-Band ( MHz) data shown here. A number of good algorithims (e.g. AOFLAGGER) which scour spectra. 240 MHz 350 MHz David Dunlap Summer School 73

74 RFI Subtraction a Trial Test In some cases, can treat RFI as any other radio source. Image, and subtract! Cornwell 327 MHz, NGC6251, far northern source. RFI comes from ILS system in Albuquerque (airport radar). No Filtering Filter active 2018 David Dunlap Summer School 74

75 Some Conclusions about RFI Issuees Spectrum Management is important. Spectral awareness is crucial. Local management of RFI essential. Don t locate observatories near population centers. Our use of the total spectrum requires development of methodologies for removing (perfectly legal) transmission. Good excision software available now. Not a perfect solution, but is effective. Effective subtraction software/methods being developed. May be very effective in the future, but at a high computing cost. If this subject interests you, come to the RFI 2016 meeting, in Socorro, October David Dunlap Summer School 75

76 The Future of the VLA/NRAO Some facts: The US is not part of the SKA Consortium It is very unlikely the US will ever be a member. Likelihood of US support for SKA very low (and = 0 for Phase 1) The NSF (via NRAO) is: A major partner in operating ALMA The sole operator of the Jansky VLA. Both of these major facilities are new we are still learning to use them! Thus: Discussion of the future of the NRAO must focus on leveraging off these two major facilities high frequency radio astronomy. In 2002, we submitted a proposal (EVLA Phase II) for a modest expansion of the EVLA 10 more antennas (25% increase), 5 X better resolution. This was unsuccessful (too small, bad timing). NRAO now developing a much bolder visions the ngvla David Dunlap Summer School 76

77 The Future VLA (from a US/NRAO Perspective) 27 x 25m diameter antennas in New Mexico, USA Original construction completed in 1980 Major electronics upgrade completed in proposal to expand by 10 antennas turned down.

78 A next-generation Very Large Array (ngvla) Scientific Frontier: Thermal imaging at milli-arcsec resolution Sensitivity/Resolution Goal: 10x effective collecting area & resolution of JVLA/ALMA Frequency range: GHz Located in Southwest U.S. (NM+TX) & MX, centered on VLA Baseline design under active development Low technical risk (reasonable step beyond state of the art) 50% in core: b < 3km ~ 1 80% in mid: b < 30km ~ % in long: b < 500km ~ km Complementary suite from meter to submm arrays for the mid-21 st century < 0.3cm: ALMA to 3cm: ngvla > 3cm: SKA The Next Generation Very Large Array

79 ngvla Key Science Mission (ngvla memo #19) Unveiling the Formation of Solar System Analogues Probing the Initial Conditions for Planetary Systems and Life with Astrochemistry Charting the Assembly, Structure, and Evolution of Galaxies Over Cosmic Time Using Pulsars in the Galactic Center as Fundamental Tests of Gravity Understanding the Formation and Evolution of Stellar and Supermassive BH s in the Era of Multi-Messenger Astronomy Highly synergistic with next-generation ground-based OIR and NASA missions. ngvla 100hr 25GHz 10mas Ga s Decarli+2016 NGVL A HI + CO(1-0) CO(2-1) 0.1 = 13AU Isella Dust Model rms = 90nJy/bm = 1K ALMA The Next Generation Very Large Array

80 Current Reference Design Specifications (ngvla Memo #17) m offset Gregorian (feed-low) Antennas Supported by internal cost-performance analysis Fixed antenna locations across NM, TX, MX ~1000 km baselines being explored GHz; GHz Single-pixel feeds 6 feeds / 2 dewar package Short-spacing/total power array under consideration Continuum Sensitivity: 1cm, 10mas, 10hr => T B ~ 1.75K Line sensitivity: 1cm, 10 km/s, 1, 10hr => T B ~ 35mK Band # Receiver Configuration Dewar f L GHz f M GHz f H GHz f H : f L BW GHz 1 A B B B B B The Next Generation Very Large Array

81 ngvla Location Centered at VLA site. JVLA: elevation ~ 2200m Good 3mm site 550km

82 Notional Parameters Carilli et al. 2015, ngvla memo #5

83 Instrument Performance Comparison Angular resolution Effective collecting area

84 Current Status NSF has provided $11.7 M for design and development studies. Nominally for two years. Goal is to have a properly designed/cost/budgeted proposal, with a compelling science case, ready for the 2020 Decadal Panel. Current stated cost is ~$ B. Currently hiring engineers, scientists, project managers, programmers. Construction timescale has a start in Current funding potentials with NSF not good at this time, but opportunities expected to arise mid-next decade. This will be a US-centered, and US-led project. Interested parties will always be welcome to join as trusted partners. (Same model as EVLA). ASTRON Lunch Talk May