DRS4 Forum
 DRS4 Discussion Forum Not logged in Message ID: 692     Entry time: Tue May 8 14:43:03 2018     In reply to: 691
 Author: Stefan Ritt Subject: Peak at 0 mV in traces

The DRS chip is read out with a 12 bit ADC, thus the phyical resolution is roughly 1V/4096 = 0.24 mV. I say roughly since the DRS has an analog gain of 0.98, which is corrected for. Now you have integer values which are converted into floating point numbers my multiplying them with ~0.24mV. If you then do histogramming with different bin sizes such as 0.1 mV and 0.35 mV , you get aliasing effects. The code truncates the result to 0.1 mV, which can give you also rounding artifacts. You will probalby see the same if you generate random 12 bit values and do the same histogramming. The 0.35 mV are not the RESOLUTION of the board (this is 0.24 mV as written above), but the Signal-To-Noise ratio of the DRS chip. If you measure zero volts at the input, and you make statistics over the distribution, you get an RMS of 0.35 mV.

Stefan

Alessio Berti wrote:

Hi Stefan,

following your example, we tried to perform the same measurement, using drs_exam and taking 1000 events. The results we obtained are in the plots attached (both in log and linear scale). We tried two different binnings:

- the first is the same as the one used in your example, that is 0.1 mV (corresponding to the plots having 81 bins)

- the second is a more wide binning equal to 0.35 mV, that is (2^(-11.5)) mV, 11.5 being the effective number of bits given in the DRS4 spreadsheet (corresponding to the plots having 23 bins)

With the fine binning we see that in the bin centered around 0 there is a little excess of events (the effect is more visible in the log scale histograms). This excess is not present in the wide binning case.

Is the problem we had before (and also here in the fine binning case) lying in the fact that we were trying to have bins with a width smaller than the effective resolution of the instrument (0.35 mV)?

We also noticed that in drs_exam, the values for the waveform are printed in the ASCII file with 1 digit after the decimal point, but when trying to print more digits the resolution is not improved (i.e. the decimal digits from the second one on are 0). This means that the values are rounded to a resolution of 0.1 mV when they are saved through the GetWave() routine (and in fact the member fPrecision is set to 0.1 -mV- in DRS.cpp, line 7502, and also in DRS.h, line 757, GetPrecision() returns 0.1). Why is that so? How does it reconcile with the effective number of bits giving a resolution of 0.35 mV?

Thank you,

Alessio & Davide

Stefan Ritt wrote:

I tried the following:

- trigger on a 10 MHz sine wave on CH0, CH1 was open

- run drs_exam.cpp program and write data.txt with a few events

- imported the event into Excel

- did a histogram on (empty) CH1

What I see is a nice Gaussian distribution centered around 1mV, but with no spike around zero. See attachment. So I still believe that you have either a binning or a rounding problem. Like you round value -0.99 to +0.99 all to zero mV, and 1.00 to 1.99 mV to one mV.

Stefan

Alessio Berti wrote:

Hi,

thank you for the quick reply. All the bins in the previous histograms have the same width. We also tried to plot the noise histogram for channel 2 with more bins (i.e. 1000, so that we can see almost discrete values), and the peak is still there.

Alessio & Davide

Stefan Ritt wrote:

I note that your peak at zero is exactly twice as high as the bins left and right, so this looks to me like a binning problem in your histogramming. Maybe your bin #0 goes from -1mV to +1mV, which all other bins are just 1mW wide. Can you check that?

Stefan

 Alessio Berti wrote: Hi, we modified drs_exam.cpp to read all 4 channels from the DRS4 and apply directly the spike removal (taken from Osci.cpp) during the acquisition phase. For test purposes, we don't save the data showing spikes and we focus on the data not having spikes (even if at the end we end up having triple and quadro spikes which are not removed by the spike removal routine, but they are rare). With this modified program we wanted to characterize the noise of the DRS4, so we took 30000 events at 5GSPS, triggering on channel 1 with a 10 MHz sine wave with 100 mV_pp (trigger level set at 10 mV), while channels 2,3 and 4 were left open without any input. We then took a look at the data and plotted the noise histograms for channels 2,3 and 4, which you can find attached (without offset correction, named zero_peak_after_spike_removal_ch*.png). For completeness, we also attached the plot from ch1 (the sine wave). The selections in time and amplitude we applied had the goal to remove the high oscillations in amplitude occurring in the first and last samples and to discard the quadro spikes we had in the data. We see that there is a peak at 0 mV in all histograms from all channels and scanning through the data, we saw that indeed the value 0 mV is stored many times for each event, thus originating the peak we see in the histograms. We also applied an offset correction to the data (taking the average of the first three most occuring amplitudes) of channels 2 (as an example) and the problem seems to be only partially removed. We also noticed that this peak at 0 mV is present also when we acquired the data from the DRS4 with DRSosc saving the data in binary format. So we had the following questions: - why is the DRS4 saving so many times the value 0 mV (exactly 0 mV)? - is there any way (in our case through software, preferably at acquisition time) to solve this problem? Thank you for the help and best regards, Alessio & Davide

ELOG V3.1.4-80633ba