Noise in L1Cal ADF Input signal



1. Introduction

1.a. Raw ADC

The ADF cards digitize the incoming BLS signals at 4 times the 132ns Tick rate, i.e. every 33 ns.
The ADF cards thus sample each EM or HD Trigger Tower BLS signal with 4x159=636 Raw ADC samples per Turn in the Tevatron.

1.b. Output Et

For each of the 36 Live Crossings (LiveX), one specific ADC sample is used and scaled to become the Output Et value sent to the TAB cards for L1Cal processing.

1.c. Capturing Whole Turns

The ADF cards include Monitoring circuitry to capture a whole turn of Raw ADC data and Output Et data.

2. Test Tools

The L1Cal Trigger Control Software (L1Cal TCS) includes expert tools based on collecting a number (1,000) of separate turns of monitoring data and calculating statistical quantities.

2.a. Profiling Raw ADC

Build 636 separate histograms (1,000 points each), and for each histogram calculate its Average and Standard Deviation (StdDev).

Also derive from the 636 Averages and 636 StdDev:

  • The Average of the 636 Averages
  • The StdDev of the 636 Averages
  • The Average of the 636 StdDev
  • The StdDev of the 636 StdDev

From the subset of the 36 histograms corresponding to those Raw ADC samples used to determine the Live Crossing energy deposit, again derive:

  • The Average of the 36 Averages
  • The StdDev of the 36 Averages
  • The Average of the 36 StdDev
  • The StdDev of the 36 StdDev

2.b. Profiling Output Et

Build 36 separate histograms (1,000 points each), and for each histogram calculate its Average and Standard Deviation (StdDev).

Also derive from the 36 Averages and 36 StdDev:

  • The Average of the 36 Averages
  • The StdDev of the 36 Averages
  • The Average of the 36 StdDev
  • The StdDev of the 36 StdDev

3. Observed Noise Components

We observe three components in the noise:

  • a) Random Noise, i.e. not correlated with time of sample
  • b) 132ns Syncrhonous Noise, i.e. noise with 132ns period
  • c) Turn-wise Syncrhonous Noise, i.e. noise repeating every turn

3.a. Random Noise

This noise is apparent in the standard deviation of any one histogram built from samples collected from the same relative time within a turn. One can also take the average of multiple such histograms.

Random noise measured in Raw ADC data:
HD EM

Random noise measured in Ouput Et data (note different color scale):
HD EM

The amount of random noise decreases as expected with higher eta values except for two moderate hotspots in negative eta values.

3.b. 132ns Synchrnonous Noise

Looking at the 636 separate averages of Raw ADC data for any given Trigger Tower produces a 33ns digitally sampled record of the incoming BLS signal with the random noise component averaged out. The resulting representation includes both the 132ns noise and the Turn-wise noise, but the 132ns noise, when present, is overwhelmingly larger in amplitude than the Turn-wise noise.

Different Trigger Towers have a different amount of 132ns noise e.g.
one HD tower with little noise,
one with a medium amount of noise,
and one of the worst towers.

Note: when the Raw ADC value is scaled to Output Et (via memory lookup), the 50 counts of Zero Energy Response (seen in above plots) is also recentered to 8 counts of Ouput Et Trigger Tower Energy before being sent to the TABs.

The 132ns noise is not, in itself, preventing the determination of stable pedestals in the Output Et data, since the ADF always uses, for all Live Crossings, a sample at the same relative phase of the 132ns noise, 3x4=12 samples apart. One can plot those 36 samples used for the 36 Live X, e.g. the worst example above (~60 counts of noise swing) still produces a uniform pedestal for all Live Crossings:

The existence of a large 132ns noise is however putting the stability of the ADF Output Et pedestal in a precarious situation and any small change in the phase or amplitude of the noise has a direct and drastic impact on the zero energy response. The detector areas with the worst 132ns noise are also the areas where the pedestal has been observed to drift around with high luminosities, and drift over time.

The 132ns noise is the overwhelming component of noisy towers, and a measurement of the amount of 132ns noise can be represented by calculating the standard deviation of the 636 averages. We have data for the full HD trigger tower coverage (not EM) and can see the geographic distribution of this 132ns noise.

132ns synchronous noise measured in Raw ADC data:
HD

3.c. Turn-wise Synchronous Noise

In addition to the 132ns noise, we also see localized disturbances within the accelerator turn, with one disturbance "chirp" before each super-bunch plus one at the end of the turn.

On some channels, these chirps extend into the beginning of the supper bunches, and the effect is to alter the pedestal value of the Output Et of those Live Crossings. The end result is a bunch dependent pedestal for the Output Et value of the 36 Live Crossings.

Plotting the standard deviation of the average of the 36 histograms from the Output Et for the 36 live Crossings shows the geographic distribution of the non-constant pedestals. This does not show the full extent of the presense of turn-wise noise, but only where this noise has an impact on the pedestals.

Effect of turn-wise synchronous noise measured in Raw ADC data:
HD
Effect of turn-wise synchronous noise measured in Ouput Et data:
HD EM

4. Hints on Noise Origin

4.a. Location

The areas with highest synchronous noise are in the TT_Eta(-6:-13) HD (note: only fine hadronic used for TT). The Trigger Tower coordinate system and the Calorimeter to Trigger Tower Eta Map show that this corresponds to the area towards the top (around TT_Phi=8) and bottom (around TT_Phi=24) of the north end cap.

4.b. Iron Position

On Sept 26th, data was collected for one eta ring while the iron was closing (but the exact position of the Iron is not known, unless someone can be recovered it from knowing the time the data was taken).

The 132ns synchronous noise increased (example):

The turn-wise synchronous noise increased (example):

The turn-wise synchronous noise increased (overall):

The random noise did not change increase (except for the two mild hotspots):

4.c. SMT and Muon turned off

On Oct 21, data was collected with the Muon electronics for layer A, then A&B, or the SMT electronics turned off.

Turning off Muon Layer A has an impact on the 132ns synchronous noise:


Turning off Muon Layer A has less impact on turn-wise syncrhonous noise:

Turning off SMT changed (peculiar!) the turn-wise syncrhonous noise landscape (with iron open data for comparison):

These plots illustrates both (1) the bunch dependence of the zero energy response, and (2) the average drifting up or down with changed conditions:

5. References

Updated 08-Dec-2006
mailto:laurens@pa.msu.edu