Measure noise at Analog Input : NI PCIE 6361

[ZeusKight]

Hello,

I'm at the beginning of my PhD thesis in physic, and we are now developping a new measurement setup based on a NI PCIe 6361 and a BNC-2120. We use Labview 2012 SP1 on a Windows 7 pro 64 bit computer.

Our setup is a special kind of interferometer. I have to design a photodiode front end that will transduce a light signal into an analog voltage, which may be digitized using one of the 8 Analog Input avalaible on my DAQ board. My problem is that I want to measure the noise added by the DAQ system himself, because I need this information before to design any electronic circuitry. In fact, I'm not a expert in electronic, data acquisition, or data processing, meaning I have a lot of remaining question for which I was'nt able to find good answers on google.

So:

 - what is the proper way to measure the noise of my NI PCIe 6361 and BNC-2120 acquisition system? I've managed to put  AI 0 input in short circuit and acquire some data, and compare PSD estimate of these data for various voltage range, acquisition time and frequency; but I dont know if it's OK.

 - It seem that the noise level on the previous experiment depend only on the sampling rate (see file attached). Where in the datasheet may I look for proper information about the DAQ system noise density, or any related noise quantity?

We think that the noise of our DAQ is the fundamental lowest limit in term of noise, and we want to make any signal from our instrument above any electronic noise through adequate amplification. I have to compute and PSD estimate of my time serie data before to extract valuable information, so any information about noise/PSD estimate is invaluable!

I have a last question, not about noise : our setup will have 4 detector, and we need to acquire the 4 analog signal simultaneously. Do I need to do some kind of synchronous aquisition?

Best best best regard and thank you for your answer

RP ZK

[johnsold]

In the Specifications on page 6 is the AI Absolute Accuracy Table.  One column shows random noise ranging from 21 to 315 uV rms, depending on input range.  Those numbers are very simliar to the resolution as shown in Table 4-1 of the X Series User Mnual.

For comparison the input bandwidth is specified as 3.4 MHz. The input impedance is 10 Gohms. The thermal noise in that resistance over that bandwidth is 12.9 uV rms.

So it appears to me that quantization noise dominates except at the most sensative ranges.

I think you can use those numbers as guidelines for your design. They also seem to be consistent with the measurements you made. I would guess that the spikes around 10E5 Hz are due to switching power supplies.

What are the sensitivity and bandwidth requirements for your optical detector?

Lynn

[ZeusKnight]

Thank you very much for your reply.

My optical detector may provide shot noise limited measurement in a 1Mhz bandwidth, with a photon budget of 3*10e-2 to 3*10e-5 mW at 532nm (1*10e-2 to 1*10e-5 A of photocurrent from my photodiode). The shot noise of the photoccurrent appear to be the lowest fundamental noise limit in our optical measurement, so we want it to be above any other source of noise in our setup. This is the reason why I've tried to measure the noise of my NI PCIe 6361 and to evaluate it's PSD. Because it is easy to calculate the expected shot noise PSD for a given quantity of photocurrent, I would be able to settle my op amp gain for this shot noise to be above the noise of my DAQ board. But I still dont understand why my DAQ PSD line level depend on my acquisition rate?

Following your suggestion, in the AI Absolute Accuracy Table page 6, I can see that all the random noise specification are given in uVrms. Does it mean that the total random noise on an acquisition is always the same? If its true, it would be an explanation for the result I've obtained on my figure: the noise PSD line level is multiplied by 2 every time I divide my banwidth (i.e my sampling rate) by 2. Has I dont know exactly what kind of ADC thecnology is involved in the NI PCIe 6361, I'm not able to be confident in that result, along the fact that I'm not sure If my method for the noise measurement of my DAQ board is truly correct.

Otherwise, If it's true, it is very important for my application, as I may always use the fastest sampling rate in order to spread my DAQ noise on a larger bandwidth...

[johnsold]

I suspect that part of the reason for the rms noise specification is that it is much easier to measure than PSD.

While I do not have a theoretical basis for this, it seems reasonable that quantization noise could be tied to sampling rates. The amplitude is probably 1 LSB most of the time, but the number of possible transitions will be proportional to the sampling rate. 

All the noise in the analog bandwidth of the converter and preamplifier of the PCI 6361 (3.4 MHz) will be aliased into the output digitized data. Reducing the sampling rate will not change the input bandwidth but it would make the noise look higher at lower frequencies.  I have been looking at your graphs to see if I could identify something as aliasing in the high frequency components. The apparent frequency shifts with sampling rates are suggestive of aliasing, but without any idea of what the actual signal frequencies might be, it is difficult to be sure.

One good thing is that your signal is large enough that a low-noise signal conditioning circuit should not be too difficult.

It might be worthwhile for you to contact your local NI sales representative. They can sometimes get more information about the details of the internal workings of the DAQ devices than are published.

Lynn

[ZeusKnight]

Following your piece of advice, I've contacted my local NI sale representative. After that, I'm able to say my problem is solved.

As an answers for my questions:

-Yes, in order to measure the noise at the analog input of a DAQ board, you may put any of its channel in short circuit. You may use a impedance matching termination (50 ohm most of the time) only if you work in the RF frequency range.

-The effect observable on my graph is well known as a property of the quantization noise. This noise in directly dependant of the ADC resolution and set the lowest noise limit of a virtually perfect DAQ system. It is constant in term of voltage RMS, and assuming some property of your signal, behave like white noise after PSD estimation. A faster acquisition rate simply spread this noise on a wider frequency range.

-The quantization noise appear lower on a wider spectrum. If you want to maintain the noise line at the same level but with a lower sampling rate, the answer is decimation. Decimation is the combination of downsampling and low pass filtering, and is readily implemented in some NI digitizer, e.g. NI PXI/PCI 5922.

Quantization noise and decimation appear to be a quite complex subject. This is the reason why I will not give any formula or reference about it now, in order to stay synthetic. I think that for any further information wikipedia can be a good starting point, and a specialized book the final answer.

I am sad not to have been able to find usefull information about decimation at NI.com. Is there any VI dedicated for this task?