Interpreting SCUBA noise observations

This page is designed to assist in the interpretation of SCUBA noise measurements and the diagnosis of problems associated with excess noise levels.

1. Introduction: Why take SCUBA noise measurements?

In most cases the first thing to check, once the TCS and SCUBA software is up and running, is that SCUBA is performing well. This is done in the first instance by the NOISE observation. The observation definition file is called noise_ref.obs (usually it is contained in the startup.m macro) and measures the "system" noise for each pixel looking at a cold load (actually a reflecting mirror in front of the cryostat window). Removing the mirror allows the "sky" noise level to be measured (noise_sky.obs) which is useful indicator of the stability of observing conditions. The rest of this report is mainly concerned with the interpretation of system noise data, and how to diagnose problems when excess noise levels are encountered.

2. Interpreting noise observations

A noise observation will display the rms noise levels for each pixel after 64 seconds of integration time. The on-line display takes the form of a plot of noise voltage on the y-axis against pixel number on the x-axis. Since this display will autoscale dependent on the most noisy pixel, it is often advantageous to run the SCUNOISE utility (part of OBSDESK) after the observation has completed to look more closely at the mean noise level and identify where the noisy pixels are on the arrays.

The case shown below is an example of a good noise performance (excellent in fact!). On the left is a plot of noise voltage against pixel number. The first 91 pixels constitute the SW array and the next 37 the LW array. On the right is a map of the array showing the layout of the pixels. As can be seen there are only 3 pixels above the user-selected threshold of 78 nV. The mean level for the LW array should be around 40 nV, and for the SW array around 60 nV. By design (and some luck...) the most noisy pixels are often to be found on the edge of the array, which is a far better situation than having one of the more central pixels affected. (Incidentally, the difference in the mean levels for the two arrays is due to the passband of the filters - for the SW array the passband is almost twice the LW. The background photon noise, which dominates the overall level and arises from the cold optics within the cryostat, is proportional to the square-root of the filter bandwidth, and so the measured noise is higher for the SW array).

3. What constitutes an acceptable performance?

This is a good question (and may depend on the observation being performed and just how fussy the observer wants to be...). Since the SCUBA bolometers have a high impedance (some 20 MOhms under optimum bias settings) they are quite susceptible to changes in the RF environment of the observatory. Associated with this is the long cable length between the high-impedance bolometer and the low-impedance first stage amplifiers. This makes the bolometers susceptible to electrical pick-up, changes in the grounding etc. The upshot of this is that the noise levels can vary substantially on a day-to-day timescale. Some bolometers (e.g. I9 and G13 shown above) are almost always consistently noisy. Their noise characteristics in the frequency domain show a distinctive 1/f-type spectrum. This is believed to be caused by a poor contact somewhere in the signal chain. Over the years we have had some success at repairing these contacts, and the recent ribbon-cable upgrade has resulted in a significant reduction (on-average) of the number of noisy pixels.

Pixels exhibiting very high levels can easily dominate the dynamic range of a map, and so in these circumstances consideration should be given to switching them off for the on-line display. See the SCUBA Observing Guide for details on how to do this, but note that they continue to record data and can, if desired, be switched back on for the off-line data reduction. It should also be noted that pixels that are only moderately noisy (say, a few times the baseline level) can often still be included in the dataset by using the SETBOLWT feature in SURF. This calculates the noise weights of each bolometer with respect to the central pixel and can essentially be used to "noise flat-field" your map (although care must be used in the presence of strong sources). See the SURF manual for more details.

4. Identifying problems

There are quite a few factors that can contribute to excess noise on the SCUBA arrays. These range from the presence of superfluid helium films to ground loops to low battery supplies. If excess noise is observed it is generally very good practice to repeat the measurement a few times just to make sure that a short-lived transient effect (such as a ground spike or cosmic ray...) has not compromised the data. If the excess noise persists then the following examples can be used to identify the problem and help to find a solution.

Case 1: Noisy pixels are seen in a small group at the start of the LW array

Symptoms: Perhaps the most common excess noise problem is that associated with contamination caused by superfluid helium films. This is most likely caused by small amounts of liquid/gaseous helium penetrating the seal between the LHe can and the vacuum space which houses the arrays. It is believed that the helium manages to get to the ribbon-cable connectors which carry the bolometer signals to the outside world. This has been seen on many occasions, and invariably the same set of symptoms are always observed. The following plots show how to recognise this noise effect when it first appears.

A small group of very noisy pixels is seen at the start of the LW array. The noise level can vary significantly but is typically many 100's to 1000's of nV. On most occasions the film will move with gravity - i.e. it starts at pixel G1 (which is closest to the seal) and moves along the connector line. After a period of time the film moves along the array as shown below (it usually takes about 1 day to move this amount, although faster changes have been seen in the past).

In this case the noisy pixels would start to have a serious affect on any mapping observation (particularly for jiggle-map). Photometry using the central pixel would still be possible, although the loss of 6 or more pixels would start to decrease the effectiveness of the sky-removal algorithm. Eventually, the situation becomes serious enough that the centre of the array is affected (see below), and although it is clearly still possible to use the other parts of the array for small, compact source observations, the performance is seriously degraded and the effectiveness of the array compromised. It will also be seen that the problem has materialised on the SW array as well. The reason for the "time-lag" between the first appearance on the LW and subsequently on the SW is not well understood.

Investigation/cure: The standard cure for this problem is to carry-out a "mini-warmup". This raises the temperature of the array to about 8 K forcing the helium film off the connector pins. The whole procedure takes about 4 hours, and although there are detailed instructions on how to carry this out, it should not be done by anyone unfamiliar with the SCUBA cryogenic system. The "mini-warmup" has found to be effective at temporarily curing this problem for periods of 3-6 days.

In summary:

Case 2: Groups of up to 16 noisy pixels are observed

Symptoms: When groups of noisy pixels are seen, which are not necessarily at the start of the array, then it is likely that the problem is not associated with a superfluid film, but most likely is an electrical problem. A possible exception to this is when SCUBA has not been used for a few days, and during this downtime the superfluid film has appeared and already moved some way across the array. The example below shows an excess noise problem that is obviously associated with an electrical problem.

Looking at the SW array it can be seen that almost all the B and F-card bolometers are showing excess noise, and that the majority of the remainder are okay. The key thing to note here is the 16-channel grouping. The SCUBA bolometer signals go into groups of 16 at the first-stage amplifier i.e. 16 pixels on a card (hence the A-I lettering). This problem is most likely caused by a loose connector - probably (by past experience) on the Tempest analogue electronics box. Other possible causes include a faulty pre-amplifier card, DAQ card or grounding problem. The situation is not so clear on the LW array in this case but a similar problem could be in evidence.

Investigation: This can be tricky to diagnose. In the past the 100-way D connector cables coming out of the Tempest box have occasionally been knocked (e.g. by the LHe dewar during a transfer). Re-seating the affected connector has cured the problem. On a few occasions the green ground wire (there is one on each connector shell) has been disturbed. If this is not the cause then it is possible that we have a faulty preamp or DAQ card. This can be investigated by swapping the suspect card and investigating whether the excess noise moves with the card change. (Note that some knowledge of the SCUBA electronics is needed to carry out work of this type i.e. how to use static protection precautions, as well as procedures for power cycling of various components in the analogue and digital electronics). A less probable explanation is a fault inside the cryostat (i.e. at the first-stage amps). To fix this would need a full warmup to ambient temperature, and hence quite serious downtime of the instrument.

In summary check the following in this order:

Case 3: High mean noise level observed

Symptoms: In this case the average noise level is much higher than the nominal level. All pixels on both arays are affected by about the same amount. An example of this is shown below.

The mean noise levels are now about 50% too high for each of the arrays. Although this could be caused by batteries that are "just-on-the-edge" of being able to supply sufficient power to the FETs, it is more likely that the background power on the arrays is higher than it should be. Other possibilities include a low-frequency temperature drift/variation.

Investigation: The first thing to check is whether the reflector blade of the chopper wheel was in place. Noise levels of this magnitude could be explained by SCUBA viewing the zenith sky and not the reflector. This means a trip to the Nasymth platform (with a flashlight if it is dark). A large "R" should be visible on the back of the blade near the window. If the blade is in the wrong position check the Berger Lahr power supply and cables to the chopper wheel unit. Make sure it has power. If this is not the problem check for moisture or ice on the window of the cryostat. Again, this could cause an elevation in the mean noise level (it's a bit like looking at cold or "damp" sky...). A moist window may mean that the LN gas hose was not re-connected after the daily LN fill. If there is moisture wipe carefully with a tissue. If there is ice, use a low-powered heat gun to melt (take extra care not to get the window too hot!). Another possibility that could cause these symptoms is a systematic, low-frequency variation in the base temperature. Look at the AVS readout and see if there are any low-frequency drifts. The worst case scenario is that there is "wide-spread" contamination on the arrays (could be caused by a vacuum leak, for example). In this case a full-warmup will be required to rectify the problem.

In summary:

Case 4: Variable (high) noise levels are seen between successive measurements

Symptoms: In this case successive noise measurements may have little similarity, but in general will show high, random noise across the array. An example is shown below.

As can be seen there is no obvious pattern to the excess noise, as in the cases of the superfluid He films or a suspect cable or electronics card. The fact that some pixels are close to the nominal value suggests that this is NOT a cryogenics problem (e.g. fluctuating fridge temperature). The most common cause of this noise trace is either an underpowered FET battery or a major electrical ground loop.

Investigation: Check the FET batteries, and also the bias batteries. One problem that has been encountered in the past has been cabling or metal pipes that cause an additional ground loop. The SCUBA electrical system has a single-point ground (back at the arrays) and any electrical connections to, for example, a shell of a cable or a metal pumping line can compromise this. A recent example was that the spare AVS bridge cable was found to be touching the top plate of the cryostat.

In summary:

Case 5: Marginal SW array noise performance

Symptoms: The number of noisy pixels on th SW array is high (say, greater than 20%). If the LW array is not seriously affected then it is likely that this is caused by an electrical problem associated with only the SW array.

Investigation: The most likely cause is that the FET battery supply is not supplying enough voltage to power the SW array first-stage amps. The SW supply usually runs down quicker than the LW supply (simply because there are more channels to draw current).

In summary:

Check the FET battery supplies. Replace them if low voltage.

Case 6: Marginal SW & LW array noise performance

Symptoms: There are 20% or more excessively noisy pixels on each array. This starts to have a serious impact on the science achievable with the instrument. An example of this is shown below.

Investigation: There are no obvious patterns in the noise levels in this case. Although the majority of pixels are close to the nominal baseline level, there are a significant number that do not meet the specification. The most likely cause of this is low FET or bias batteries, and although the SW array draws more current than the LW (and should therefore "fail" earlier...) there have been occasions when both supplies have begun to fail at the same time (battery "fatigue"). In this example the level of the excess can also give a clue. A ground loop tends to give higher levels (e.g. 100's of nV), and the superfluid He film symptom even higher (1000's). On two other occasions excessive vibration was found to be a problem (oscillating antenna drives and excess vibration from the RxB3 compressor). Both these events caused pixels that have a known microphonic sensitivity to vibration to have higher noise levels. The worst case scenario is some kind of non-uniform (random) contamination on the array. A mini-warmup is recommended at this stage, but if unsuccessful a full warmup and extended pump-out would be required to cure.

In summary:

Case 7: Noise levels are too low...

Symptoms This happens maybe a few times a year, and the symptoms are that the noise levels for both arrays are much lower than the nominal level. Typical levels might be 10 nV or less.

Investigation:Usually this is caused by the FET batteries (i.e. the big rechargeable battery packs) being replaced but not switched back on. The measurement is then that of the second stage amplifier noise (about 5-10nV). Another more serious situation may be that the fridge has warmed up, causing the bolometers to have very low impedance (little sensitivity) to incoming radiation. Contact the designated SCUBA system engineer as soon as possible if this has occurred.

In summary: