GD153 is a DA white dwarf with an effective temperature of 39,000K. The infrared portion of the spectrum of this object is approximately Rayleigh-Jeans. GD153 has a V-band Johnson magnitude of 13.35. The UKIRT faint standards list reports the star (FS33 in the UKIRT list) as 14.24 at K with H-K = -0.078 and J-K=-0.223. The figure at left below shows the counts in units of DN/s based on eight 120 seconds spectra of this object. In this figure the flux in different orders has been merged to give a total system response as a function of wavelength. The adjacent figure gives the S/N per spectral pixel for this object in the full 8x120= 960s integration. Evident is the efficiency function of each of the orders, peaking near the center of the order. The third figure shows the calbrated response for this object in Janskys.

Based on the observation of this 14th magnitude white dwarf SNR=5 can be achieved in one hour in a three pixel spectral bin at magnitudes K=15.5, H=16.0, J=17.0.

Response in units of DN/s for GD153 (J=14.0, Ks=14.2). Counts have been binned where the source is detected in multiple orders.	SNR per pixel in 8x120s of integration on GD153.	Calibrated GD153 spectrum in Janskys.

3.1.b. The limits of detectability

The images below shows the K-band detection and extraction of a faint red source in one hour of integration time. During that hour the source was observed in twelve five-minute integrations, alternating in slit position ABBAABBAABBA. The count rate of 0.07 DN/s corresponds to a K=17.5 mag continuum. With three pixel spectral smoothing the continuum has an SNR=0.5 per smoothed spectral bin. In order to produce a spectrum bright enough for extraction, the five minute integrations had to be binned into 30 minute stacks at the "A" and "B" positions. Overall, this spectrum may not be very useful scientifically, but it does illustrate the limits of detecting and extracting a source spectrum under good conditions at the 3.5-meter.

Note that although the extracted continuum K-band magnitude is 17.5, the reported source K-band magnitude was 15.8. The observers struggled to guide on this faint target and the lower than expected detected magnitude is likely due to losses experienced in attempting to keep the light going down the slit.

The image result of differencing two 6x5 minute stacks of "A" vs. "B" position frames. The extracted source magnitude is consistent with a 17.5 mag Ks-band continuum.	The SNR per sets of 3 binned spectral pixels in the H and K spectral orders. The K-band SNR here is about 0.5 per 3 pixel bin.	The extracted spectrum in the H and K-band orders in units of DN/s. The spikes in the spectra are residual airglow lines contaminating the spectrum.

The sensitivity of the slit viewer dictates the faintest source that can be observed directly and placed on the slit. Below this sensitivity threshold observers will have to depend on blind offsetting to place a source on the slit. In general, if a source is too faint to be seen in the slit viewer its continuum will be difficult to detect in the spectrograph. The integration time for the slit viewer is adjustable, but 30 seconds represents a maximum practical integration for positioning the source and for guiding. Longer integration times are possible, but the delay between exposures makes guiding and/or positioning the source on the slit tedious. In addition, since the slit viewer operates at Ks-band, significant thermal background accumlates during an exposure and will eventually saturate the array. The time to saturation is a strong function of ambient temperature. As a reference point, a recent observer found that the array was saturating in 45 seconds under warm (60F) conditions. Given blackbody emission at ambient temperature, the time to saturation will be four times longer at 32F. . The slit viewer is thus about a magnitude more sensitive under winter conditions than during the summer for the same integration time. Observers with extremely faint targets may wish to consider this fact in scheduling observations.

For bright sources, positioning a target on the slit is straightforward. Once a source is identified in the slit viewer, [ctrl-leftclick] on the center of the source will move the target to the "hotspot" location on the slit. After a few nudges to get the spilled light symmetrical guiding in boresite mode and integration can begin.

For faint sources attention will be required to maximize the source signal in the slit viewer images - which to 0^th order is accomplished by increasing the frame integration time. There are two routes to obtaining optimal SNR in the slit viewer. The best practice remains to be determined.

The HAWAII-2 spectrograph array can be read non-destructively multiple times during the course of an integration. Fowler sampling refers to conducting a burst of N readouts at the beginning of an integration and an equal burst of N readouts at the end of integration in order to suppress read noise (FowlerN). Read noise suppression can be important for TripleSpec, since the dark current is 0.05 e-/s and the inter-airglow line continuum is weak. For correlated double sampling (one read at the beginning of an integration and one at the end - also ``Fowler1") TripleSpec read noise is observed to be 5DN or 17 electrons. The TripleSpec TUI configuration menu permits the user to select a range for the number of Fowler samples. The TripleSpec read noise is observed to improve, as expected, with the square root of the number of Fowler samples. With Fowler8 the effective read noise is about 7 electrons. After the collection of 100 dark current or sky electrons these factors will dominate the read noise.

Matt Nelson has provided the following detailed breakdown of the exposure sequence:

1 -  ICC receives exposure request
2 -  ICC calculates and sets exposure time (SET cmd to controller)
3 -  Controller finishes loop in Continuous Reset, Processes command
     and replies to ICC.
4 -  ICC initiates exposure in controller
5 -  Controller finishes loop in Cont Reset, Breaks out of 
     loop to expose
6 -  Controller does full pixel by pixel reset of array
7 -  Controller Delays for reset settling
8 -  Read-1 reads are made 
9 -  Controller waits for calculated Integration time
10 - Controller waits for 400mS for Array outputs to stabilize
11 - Read-2 reads are made
12 - ICC finished scavenging last Read-2 read, builds frame and
     writes it to disc.
13 - ICC replies "done" to hub

Guesses about timing.

1-4 should be relatively fast.  The line loops in continuous reset are
quite quick so I would expect this sequence to finish in < 10mS

5-6 was never timed by me.  What I recall from the pixel clocks when I 
was developing the DSP code is that the reset pixel clock was running 
about 1/3 of a normal readout pixel clock.  So I'd guess ball park 
200-300 mSec for this

7 50mS

8 N*790mS

9 Exp Time - N*790ms

10 400mS

11 N*790mS

12-13 Unknown but fairly quick.  Most of the frame data are scavenged
   and averaged while the pixels are still being read.  it is just
   the recovery time of the last frame, subtraction of the Read1/2
   the writing of the frame to disc.  I'd estimate 100mS-200mS 
   nominal timing.  


Of course what is missing is the time required for the hub to cycle
back around to requesting the next in the frame series.  I'm certain
the APO staff would have a good estimate.  As a summary, the ICC
and controller are probably using up 250+50+400+150mS = 850mS of
time beyond the time spend during integration + readout.  So for
a rough estimate of instrument cycle time beyond the requested 
integration time  850+790*N mS should be close.  

Optical ghosting: The brightest emission, specifically the K-band thermal emission, can be reflected about a point that is somewhat close to the center of the 2048x1024 array. This reflection produces an inverted stripe of emission that is evident in the right-hand quadrant just above the 6th order. This reflected strip ends in an intense bar this is acutually an image of some of the surface components and wirebonds on the detector wafer. Fainter reflection stripes are visible at a couple of other locations on the array.

Electronic quadrant offset/shading: At the level of a few DN the "bias" level on the detector can drift or be offset from one quadrant to another, producing a faint discontinuity across the quadrant boundaries.

All of these effects are repeatable from frame to frame and largely subtract out in the difference between two frames. The one place where residual features remain is the "stripe" electronic ghosts produced when observing bright standards. In this case, the level of ghosting is small compared with the source intensity, so the ghosts are of little consequence.

Bright sources produce an after-image on the array. This "persistence" image can linger for up to an hour following the observation of an extremely bright source. If a faint target is observed directly after a bright calibrator (e.g. K=7) the first few 5-minute exposures on the faint target may be contaminated with the spectrum of the calibrator (as a faint positive source in both the "A" and "B" positions). Observers should be careful to select fainter standards prior to faint source observations and should be aware of possible persistence contamination during data reduction.

Typical sequence after slewing to a new source:

• Decide on your guider integration time and acquire a frame to serve as the background image with "Expose".
• Choose Bkgnd Sub = On by pressing the "On" button.
• Move the telescope >= 5 arcsec.
• Press "Expose" for a single frame or engage manual guiding to view a steady stream of incoming frames.
• Identify the target star and move it into the slit.

Important tip:

Background subtraction only works well if the saved background frame is "fresh". Even after a few minutes a mismatch can develop between the saved background frame and the actual background due to airglow, temperature, and flexure. Background frames must also have an identical integration time to the incoming frames. A new background frame is required (by either stopping and restarting background subtraction or by pressing the "New" button) if

• the integration time is changed.
• the telescope is slewed to a new source.
• significant non-celestial structure is visible in the guider frames due to a drift in the background level (usually after the saved background has become stale by several minutes).

Like any spectrograph, TripleSpec requires both continuum and line illumination for flat-field and spectral calibration. The Triplespectool data reduction code currently uses airglow lines for spectral calibration. Spectral lamp observations are not necessary, but many observers may like the security of having traditional lamp spectra in reserve. In order to use the airglow calibration method some of the spectra must be free of source flux in the middle of the slit. Since the natural observation procedure is to offset the source between two off-center slit positions, standard TripleSpec observations naturally provide for airglow wavelength calibration. In general, observers should be sure to have a few observations with only airglow at the slit center for wavelength calibration.

A typical calibration sequence to be conducted once each night (possibly once per run) consists of sets ten 60 second exposures that include:

• No illumination - only continuum thermal radiation from the telescope and dome evident at the longest wavelengths.
• Bright quartz lamp on for continuum illumination.

Not needed are:

7^th-9^th magnitude A0V stars are ideal calibrators for most TripleSpec observations. The Triplespectool pipeline and this calibration method are described in:

The TripleSpec slit wheel drive has 8 positions. Four of these positions have spectrograph slits. The other four positions in between the slits block all light from entering the spectrograph. Technically it is thus possible to take true dark frame, but in practice true darks are not necessary. Although this statement seems counter to standard observational practice, it turns out that dark current and frame bias are naturally accounted for in the differential procedures that underlie standard TripleSpec data acquisition. In particular, all spectra are obtained in differential mode - either A/B positions on the slit or on-slit vs. adjacent sky. In each case, data processing takes place by operating on the difference of pairs of frames. The underlying dark current and bias (and thermal background) are subtracted automatically. Similarly, when observing calibration lamps an equivalent length exposure is taken with the lamps off so that dome thermal emission is the only source of illumination. When pairs of lamp-on vs. lamp-off frames are subtracted the dark current, bias, and thermal backround all are removed leaving only the source spectrum.