<font size=+2><b>TTF: A Flexible Approach to Narrowband Imaging

TTF: A Flexible Approach to Narrowband Imaging

Joss Bland-Hawthorn and D. Heath Jones Anglo-Australian Observatory, P.O. Box 296, Epping, NSW 2121, Australia Mount Stromlo & Siding Spring Observatories,
Private Bag, Weston Creek P.O., ACT 2611, Australia


The Taurus Tunable Filter ( TTF) is a pair of tunable narrowband interference filters covering 3700-6500 Å  (blue `arm') and 6500-9600 Å (red `arm'). The TTF offers monochromatic imaging at the cassegrain foci of both the Anglo-Australian and William Herschel Telescopes, with an adjustable passband of between 6 and 60 Å. In addition, frequency switching with the TTF can be synchronized with movement of charge (charge shuffling) on the CCD which has important applications to many astrophysical problems. Here we review the different modes of TTF and suggest their use for narrowband imaging.


The Taurus Tunable Filter (TTF), manufactured by Queensgate Instruments Ltd., has the appearance of a conventional Fabry-Perot etalon in that it comprises two highly polished glass plates. Unlike conventional Queensgate etalons, however, the TTF also incorporates very large piezo-electric stacks (which determine the plate separation) and high performance coatings over half the optical wavelength range. The TTF is used in the collimated beam of the TAURUS-2 focal reducer available at both the 3.9 m Anglo-Australian (AAT) and 4.2 m William Herschel (WHT) telescopes. Field coverage is 10 arcmin at f/8 or 5 arcmin at f/15.

For the first time, TTF provides the capability to synchronize frequency switching with the movement of charge on a CCD (charge shuffling). This has important benefits for many astrophysical experiments, not least for averaging out temporal variations due to the atmosphere or measurement apparatus. This instrument is an important step in changing the way that intermediate to narrowband imaging is performed at observatories.

The TTF has largely removed the need for buying arbitrary narrow and intermediate interference filters, as one can tune the bandpass and its centroid by selecting the plate spacing. The spacing of the plates is controlled to extremely high accuracy with capacitance micrometry[]. This approach to tunable imaging has existed since the instrument of Ref. , although TTF is the first of its kind in terms of both wavelength and bandpass accessibility. Since tunable filters have a periodic transmission profile, the instrument requires a limited number of blocking filters. At low resolution (R = 300), conventional broadband UBVRIz filters suffice. At high resolution (R = 1000), eight intermediate band filters are used to sub-divide the wavelength range of each arm.

The highly polished plates are coated for optimal performance over 3700-9600 Å. The coating reflectivity (96 %) determines the shape and degree of order separation of the instrumental profile. This is fully specified by the coating finesse, N, which has a quadratic dependence on the coating reflectivity. The TTF was coated to a finesse specification of N = 40 which means that the separation between periodic profiles is forty times the width of the instrumental profile. At such high values, the profile is Lorentzian to a good approximation. For a given wavelength, changes in plate spacing, L, correspond to different orders of interference, m. This in turn, dictates the resolving power R = Nm according to the finesse.

The flexibility of TTF is well-suited to narrowband astronomical imaging in lines such as [O II]l3727, [O III]l5007, Ha, [N II]l6583, [S II]l6727 and [S III]l9069. Furthermore, we have the capability to obtain images of spectral lines at arbitrary redshifts (see e.g. Ref. ). Figure gives a good demonstration of the power of TTF. We observed seven 14 Å sequential bands near 8100 Å for 20 mins each, illustrated in the inset. Typical images and spectra for individual sources are shown in Fig. (a) and (b). We used faint-object finding to identify many sources in each of the seven images. The fluxes in the B2 on-bands were summed and compared with the summed off-bands, B1 and B3. The main figure illustrates what happens when you compare (B2-B1) with (B3-B2). Sources with excess signal (emission lines) or a lack of signal (absorption lines) at the cluster redshift lie close to the diagonal line.

This technique is immensely powerful for identifying emission-line and absorption-line galaxies. Most of the absorption-line galaxies are elliptical galaxies with absorption lines characteristic of A stars, so-called `E+A' galaxies, first identified in clusters by Ref. . Our method goes more than a factor of five deeper than all previous spectral line surveys at red wavelengths for several reasons: the TTF is able to tune to very dark regions between the bright, atmospheric emission bands; the system has a high total throughput (30 %); our detector is one of a new generation of highly red-sensitive, deep depletion CCDs.

In the following sections we summarise the different observing modes of TTF. Discussion is also made of phase effects in the field and their influence.

Figure 1: Left: Each point is a cluster galaxy in A 3665 (z = 0.237) identified by TTF from comparing Ha images at the cluster redshift with two neighbouring bands (see inset). Many absorption-line (E+A) galaxies (bottom right quadrant) and emission-line (star forming) galaxies (top left quadrant) are clearly seen. Right: (a) 10 arcsec subsets of 7 TTF images sequential in wavelength for three emission-line galaxies. (b) TTF spectra derived from (a) showing clear Ha emission (1.1 arcsec seeing).


Central to almost all modes of TTF use is charge shuffling. Charge shuffling is movement of charge along the CCD between multiple exposures of the same frame, before the image is read out[,]. An aperture mask ensures only a section of the CCD frame is exposed at a time. For each exposure, the tunable filter is systematically moved to different gap spacings in a process called frequency-switching. This way, a region of sky can be captured at several different wavelengths on the one image (Fig. ). Alternatively, the TTF can be kept at fixed frequency and charge shuffling performed to produce time-series exposures. Each of these modes is described in the following section. The details of charge-shuffling are described in Sect. 5.

Figure 2: Simplified view of how the CCD chip is illuminated and shuffled. The mask aperture size controls the amount of the CCD exposed and therefore the number of different images fit on to an individual frame. Charge can be shuffled in both up and down the chip.

The TTF plates can be switched anywhere over the physical range 2 to 12 mm at rates in excess of 100 Hz, although in most applications, these rates rarely exceed 0.1 Hz. If a shutter is used, this limits the switching rate to about 1 Hz. Charge on a large format CCD can be moved over the full area at rates approaching 10 Hz: it is only when the charge is read out through the amplifiers that this rate is greatly slowed down. The TTF exploits the ability of certain large format CCDs to move charge up and down many times before significant signal degradation occurs[]. In this way, it is possible to form discrete images taken at different frequencies where each area of the detector may have been shuffled into view many times to average out temporal effects.

The field of view available in shuffle mode depends on the number of frequencies being observed. When we move the charge upwards, say, information in pixels at the top of the field is rolled off the top and lost. For example, two frequencies requires that we divide the CCD into three vertical partitions where information in one of the outer partitions is lost in the shuffle process. In the limit of n frequencies, where n is large, only half the available detector area is used to store information. The MIT-LL 4096 ×2048 (rows × columns) format CCDs with 15mm pixels increase the detector area available for shuffling by threefold compared to the Tek 1024 ×1024 CCD (24mm pixels) previously used. This is because the instrument field of view projects to an aperture 1024 pixels in diameter and shuffling is only possible in the vertical direction.

One application of charge shuffling that does not sacrifice detector area is time-series imaging (Sect. 3.3, below). This is because the imaging region is relatively small and able to be read out as the time-series progresses.


There are several technical problems driving the development of tunable filters for narrowband imaging over standard fixed interference filters. First, it is difficult and very expensive for manufacturers to produce high performance narrow passbands, particularly at resolving powers approaching 1000. This problem is largely circumvented by use of TTF in conjunction with a 5-cavity, intermediate bandpass blocking filter. Secondly, the TTF instrumental function has identical form at all wavelengths and all bandpasses. A comparison of two discrete bands at different wavelengths is moderated only by the blocking filter transmission at those wavelengths (which is normally flat in any case), and the ratio of the gap spacings. Thirdly, the same optical path is used at all frequencies. Furthermore, the ability to shuffle charge allows one to average out all temporal variations: atmospheric transparency, the contribution from atmospheric lines, seeing, detector and electronic instabilities.

We now describe some of the advantages of the TTF shuffle capability in imaging emission-line sources.

3.1  Tuning to a Specific Wavelength and Bandpass

This combines the versatility of the charge-shuffle technique with our ability to tune-in on obscure spectral lines. We can also optimize the bandpass to accomodate the line dispersion and to suppress the sky background. The off-band frequency can be chosen to avoid night-sky lines and can be much wider so that only a fraction of the time is spent on the off-band image. Figure (a) shows a charge-shuffled image of the planetary nebula NGC 2438 in [S II] lines at 6731 and 6717 Å. During this 12 min exposure, TTF was switched between the two frequencies 18 times while the charge was shuffled back and forth accordingly. In this way, temporal variations in atmospheric transmission are equally shared between each passband over the entire exposure time.

(a) Figure (b) Figure
(c) Figure

Figure 3: (a) Charge shuffled image of the planetary nebula NGC 2438 in [S II]l6371 (top) and [S II]l6717 (bottom). The TTF was frequency-switched 18 times during the 12 min exposure and the charge shuffled back and forth in concert. A 10 Å bandpass was used. (b) Charge shuffle time-series imaging of V2116 Oph, an x-ray pulsar in orbit around a red giant. V2116 Oph is to the right, reference stars are at the centre and left. TTF was tuned to a 7 Å bandpass centred on the [O I]l8446 line, which is thought to vary with the pulsar period of 120 s. Each exposure was 12 s. (c) Charge shuffling either side of an emission-line to average out the effect of a steeply sloping continuum. Equal time is spent collecting continuum from both sides of the emission-line (left). The line only contributes to the top portion of the shuffle frame while both sides of the continuum contribute to the bottom (right).

3.2  Shuffling Between On and Off-band Frequencies

With the new MIT-LL chips, we are able to image the full 10 arcmin field for two discrete frequencies. We can also choose a narrow bandpass for the on-band line and a much broader bandpass (factor of 4-5) for the off-band image so that we incur only a 20 % overhead for the off-band image. As with specific tuning (Sect. 3.1, above), multiple frequencies can be imaged in a single frame, the number of which is entirely arbitrary.

We are currently adapting the TTF system to bring both the large and small- scale tuning under complete software control. The present arrangement limits software control to only small-scale tuning. Having the tuning fully integrated under software control will allow us to alternate between broad and narrow bandpasses with twice the speed of that at present. Such a capability is useful in a wide range of astronomical observations that require averaging of varying atmospheric conditions over time.

3.3  Time-Series Imaging

For time-varying sources, we can step the charge in one direction while only switching between a line and a reference frequency. For example, a compact variable source imaged through a narrow aperture in the focal plane forms a narrow image at the detector. We can switch between the line and a reference frequency many times forming a set of narrow interleaved images at the detector. The reference frequency is a measure of the atmospheric stability during the time series. Alternatively, if nearby reference stars are available for photometric calibration, then charge shuffling can be done at a single fixed frequency as demonstrated in Fig 3(b). For example, some x-ray binary stars produce strong emission lines that vary on 0.1 Hz timescales. With a slit only 4 pixels wide, we can obtain 500 images in the emission line, interleaved with a further 500 images at a reference frequency. In this example, all of the CCD area is utilized because the charge is clocked in only one direction. The vertical shift takes about 50 ms per row. Strictly speaking, this operation constitutes a `charge shift' rather than a charge shuffle. Once the chip is full, it takes ~ 3 mins to read out the CCD.

If the read-out time of several minutes is critical to the measurement, an alternative method is time-series read out. For the example above, the four CCD rows comprising the slit are clocked downwards at the end of each exposure. However, the next exposure is delayed by the time it takes to read out the bottom four rows ( ~ 200 ms). In practice, the slow shutter means that the time series mode is to be preferred for most applications over the `charge shift' mode.

3.4  Adaptive Frequency Switching with Charge Shuffling

When imaging spectral lines that fall between OH night-sky emission lines or on steeply rising continua, a powerful feature is the ability to shuffle between on and off bands but where the off band is alternated between two or more frequencies either side of the on-band frequency. As illustrated in Fig. 3, this can be used to average out either rapid variations in blocking filters or the underlying spectral continuum.

3.5  Shuffling Charge in One Direction with a Narrow Focal Plane Slit

This allows us to produce a long-slit spectrum of an extended source. While this is vastly less efficient than using a long-slit spectrograph, the capability is a fundamental component of establishing the parallelism of reflecting mirrors at few micron spacings[]. We do this by taking long-slit TTF spectra of an arc-line, with a pupil-plane mask to isolate a particular section of the beam (Fig. ). The plates are adjusted until the line shows peak transmission at a common plate setting for all regions of the beam (Fig. ). It is then that the plates are parallel. Using this technique we are also able to observe phase reflectance effects at very narrow gap spacings. These are also described in Ref. . Conventional methods of aligning plates are an order of magnitude slower by comparison.


Figure 4: Schematic showing a simplified view of how the CCD chip is illuminated and shuffled to obtain a long-slit spectrum.


Figure 5: TTF charge-shuffle long-slit imaging of a deuterium lamp (top) and a copper-helium lamp (bottom), before (top) and after (bottom) plate parallelism has been established. The lines are discrete images of the focal plane slit. Note the slight curvature in the lines over the 10 arcmin field at the largest plate spacings.


The primary goal of a tunable filter is to provide a monochromatic field over as large a detector area as possible. With the present TTF, however, the field of view is not strictly monochromatic. The effect is most acute at high orders of interference. Fig. (a) shows how wavelength gradients (or phase effects) are evident from a ring pattern of atmospheric OH emission lines across the TTF field. In this particular case we see rings at different wavelength appearing within the one order. The circular pattern is not centrally aligned due to tilting of the plates ( ~ 10 to 15) to deflect ghost images from the beam.

Wavelengths are longest at the centre and get bluer the further one moves off-axis. For instruments such as TTF, the wavelength as a function of off-axis angle q is1

l(q) = lcentre ·

1 - q2

Here lcentre is the on-axis wavelength, equal to 2 L/m (for an air gap), where L is the physical plate spacing and m is the order of interference.

It follows from Eqn. (1) that the change in wavelength lf across an angle q from the centre is given by

lf (q) = lcentre · q2
Since q 0.14 for the current Tektronix CCD, then lf / lcentre is always   ~   10-2. It also follows that lf / lcentre remains fixed over a given radius, irrespective of order m. Reference derives related equations in the context of a higher order conventional etalon (Sections II(c), (e)).

We define our monochromatic field by the size of the Jacquinot spot, the central region of the ring pattern. By definition, the Jacquinot spot is the region over which the wavelength changes by no more than 2 of the etalon bandpass, dl. For wavelength, l, the bandpass relates directly to the order m such that

dl = l/ Nm = l/ (40 ×m).
Here, N is the effective finesse of the etalon, which in the case of TTF is approximately 40. Combining Eqns. (1) and (3) we find that the angle subtended by the Jacquinot spot is
qJac2 = 2 2/ N m    = 1/(14.1 ×m) .
For a particular etalon, the size of the Jacquinot spot depends on order m alone. Equation (4) shows how the spot covers increasingly larger areas on the detector as the filter is used at lower orders of interference. The absolute wavelength change across the detector remains the same, independently of order. However, its effect relative to the bandpass diminishes as m decreases.

Figure (b) demonstrates how the effects of atmospheric emission can be removed during reduction. The software creates a background map by median-filtering copies of the original image, each one offset from the other by a small amount. The result is smoothed and subtracted from the original, leaving little or no night-sky residual. This technique exploits the fact that at low orders of interference, the night sky rings are lower frequency structures than the objects.


Figure 6: Images showing the removal of the atmospheric OH emission lines from a raw TTF image before (a) and after (b) cleaning. Only half the full TTF field is shown in each case.

The TTF is the most straightforward application of tunable filter technology. Other, more sophisticated techniques such as acousto-optic filters exist (see Ref.  for a review), although all are currently considerably more expensive. In future TTF-type instruments, phase effects will be eliminated from the outset by bowed plates. One advantage of such a design is that the TTF will no longer have to be tilted to deflect ghost reflections. Furthermore, interference coatings are notorious for bowing plates and this can be factored into the plate curvature specification. Other possible improvements are additional cavities to square up the instrument profile. We are also investigating the use of commercial acoustic-optic and liquid crystal tunable filters in place of our current blocking filters. Further advances of this technology will see such an arrangement become reality.


Modern CCDs are buried channel devices where charge packets are confined to a channel that lies beneath the surface of the silicon. An extra n-dopant is applied to the surface to reshape the potential well so that electrons are forced to collect below the Si-SiO2 interface. Thus interface traps which drastically lower the efficiency of surface channel devices are largely avoided. This gives the buried channel device a much higher charge transfer efficiency (typically 99.999 %). However, electron traps do exist in buried channel devices and are the main obstacle to our new charge shuffling mode (see Fig. 2). For ground-based observations, the most important trapping site is the `bulk trap'[]. These are due to deep-level metallic impurities or lattice defects associated with the silicon material on which the CCD is built. Bulk traps that happen to lie within the charge transfer channel will trap only single electrons.


Figure 7: A 20 min dark frame shuffled (a) 10 times, (b) 100 times, (c) 1000 times, (d) 10000 times. The effect of bulk trapping sites becomes apparent after shuffling more than 100 times.

5.1  Electron trapping sites and Charge Shuffling

The charge transfer efficiency (CTE) is a measure of the ability of the device to transfer charge from one pixel to the next[]. For well-made buried channel devices, the CTE will be in the range of 0.99999 to 0.999999 for strong signals ( ~ 103 e-). If we assume a CTE of 0.999999, for a CCD of 1000 pixels on the side, 99.4 % of the charge will remain in the pixel farthest from the output after it has been transferred to the output (2000 pixel transfers where each pixel requires three transfers in a three-phase clocking scheme). We find that the CTE for our Tek chips may be significantly better than 0.99999 since shuffling allows us to examine the loss of charge over many shuffles. The sharp response of a bright cosmic ray suffers substantial leakage into neighbouring pixels after 100 shuffles or more.

More generally, the charge shuffling mode enables the astronomer to build up a CCD exposure comprising several on-chip charge movements made, generally, into and out of an aperture of a masked CCD. The state of an external device (etalon, filter wheel, polariser, telescope position, chopping secondary, etc) may be synchronised to these movements (shuffles) as can the opening and closing of a shutter. Alternatively, no mask may be used, resulting in superimposed images from each of the phases, these images being separated spatially by the vertical shuffling. Also, the shutter, opened at the start of the exposure, may be left open throughout the charge shuffling process and closed at the end of the exposure prior to reading out the CCD.

Charge-shuffled exposures may range in complexity from the simplest two-image shuffle comprising only 2 phases per cycle, through to a large number of image phases each comprising a chopping between two conditions. A very fast flash facility (which fills the CCD in a single burst over a short interval) is also available to sample transient phenomena, e.g. occultations at rates almost down to 1ms, if required.

The external device (etalon, filter wheel, polariser, telescope position, chopping secondary, etc.) responds to sync pulses from the micro and must be programmed and set up in such a way that the series of sync pulses will cause the external device to cycle through a known and predetermined sequence of conditions corresponding to each shift of the charge on the CCD.

5.2  Shutter operation

A choice between shuttered or unshuttered exposures is available. A shuttered exposure enables unsmeared exposures to be made with a maximum phase time of 650 seconds. However the shortest opening times and maximum repetition rates (max may be 10Hz for physically small shutters) of the particular shutter being used and even perhaps how rapidly the shutter might wear out may limit phase times to relatively long values.

Unshuttered exposures open the shutter at the start of the exposure and close it at the end just before the CCD is read out. These exposures offer phase times that may be very short (down to about a millisecond) but may suffer from objectionable image smearing, depending how rapidly the CCD charge shifting occurs (CCD dependent) and how much of the phase or cycle time is used in moving the charge.

A charge shuffling sequence is specified by a phase definition table which has been generated by the observer in an editing session. The table is downloaded to the CCD controller when setting up a shuffle readout. The table comprises a number of phase definitions. Each line of the phase definition table enables or disables the following operations to be performed for that particular phase:

On issuing the run command several other parameters are sent to the controller. These are (i) the number of repeat cycles to be done (the repeated part of the phase table is the running sequence portion; the start and end sequences are not part of the cycle); (ii) the resolution of the phase time and exposure time clock can be set from 1us to 10ms; (iii) a minimum phase time to be used in bias frames; (iv) a delay time associated with the settling of the external device; (v) a selection of either the internal clocks or hardware sync inputs (two off) or a combination of both used to start and stop the exposure and to time the phases; (vi) a control byte defining the shutter action and whether a normal exposure, dark frame or a bias frame is requested (the shutter action determines whether the shutter opens continuously - unshuttered phase - or whether it is opened and closed each phase - shuttered phase).

A laboratory demonstration of charge shuffling is shown in Fig. 7. The integrity of the data is preserved by the shuffle process for less than 100 shuffles. This technique will allow us to reconstruct low-resolution spectra, with photometric integrity comparable to or exceeding that of purpose-built spectrographs.


We have discussed details of the Taurus Tunable Filter (TTF) instrument and its use. When used in conjunction with a CCD charge shuffling technique, TTF is well-suited for multi-narrowband imaging of emission-line sources. The analysis of tunable filter data is more straightforward than traditional Fabry-Perot imaging as all regions of the field are close to a common wavelength.

The present TTF instrument exhibits some phase effects at high resolving powers. We have demonstrated that these effects are tolerable for most applications (lf / lcentre   ~   10-2) and the side-effects correctable through software. The commissioning of TTF has marked the start of an exciting period of new imaging instruments for optical astronomy.

We are indebted to J. R. Barton for his design and development of the AAO-1 CCD controller which has enabled the shuffle mode described here. The extraordinary versatility, fast read out and low read noise of the AAO-1 controller have been the cornerstones of many successful astronomical experiments at the AAT. Thanks also to L. G. Waller, T. J. Farrell, E. J. Penny, C. McCowage and D. J. Mayfield for technical input during TTF and charge-shuffle implementation. DHJ acknowledges the assistance of a Commonwealth Australian Postgraduate Research Award.


1 Equation () makes use of the small angle approximation for cosq.

File translated from TEX by TTH, version 0.9.