regarding 1/f measurements

            I am a MTech student. I read your presentation on noise measurement. I am doing similar kind of noise measurement. My task is to measure 1/f noise of various opamps. So I started with uA741 opamp. I followed your steps as described in presentation. I am using LabView 11 for automation and Agilent's Dynamic Signal Analyzer 35670A for measurement. Data sheet value of 741 opamp noise is 32 nV/rtHz at 100Hz and values which I obtained(in one set of readings) is 98nV/rtHz at 100Hz.

 Comment:

 The 1/f region is the most susceptible to process variation.   So seeing a difference between measured and data sheet specifications in this region may not be an error.  Also, the 741 is a very old device.  This device has been released on multiple processes for different manufactures.  I would suggest a more modern device for your experiments.  Some examples are OPA827, OPA2188, and OPA211.  Really the list is endless.  Just look at a device that has been released in the past 5 years.  

              Regarding this experiment I have some questions...

1. Right now I am doing all my calculations on circuit assembled on breadboard. I am using battery as a power supply. Main problem I am facing is 50 Hz noise. I am getting similar peaks even at 100 Hz, 150Hz, 200 Hz (multiples of 50Hz). I think these are coming through Agilent's 35670A. How to get rid of these? I did measurement by ignoring these peaks. But sometimes it affects the adjacent frequencies (sometimes from 35 to 55 Hz). So I have to ignore all this span of readings. 

 Is your amplifier in gain?  If the amplifier is in gain the amplifier noise will be higher, so the 50Hz noise pickup will be smaller by comparison.  Set the gain as high as possible while keeping that bandwidth greater the 100kHz.  Some times a cascaded amplifier works best.  The first (the DUT) gain set to 100 and the second with gain set to 10.  Also, are you using a shielded enclosure?  I like the paint can method (shown in the presentation).  Finally, some times it is hard to completely eliminate the noise.  We have a joke that says that the best way to eliminate 50Hz noise is with an eraser.      

 2. Data sheet graph of 741 shows calculation from 100 Hz frequency where as 1/f noise is dominant at lower frequencies. I did the measurement starting from 100 Hz so that I can verify the values. How can I get more accurate values of 1 by f noise?

 I don’t see the calculation for noise in the uA741 data sheet.  I’m probably looking at a different data sheet.  Nevertheless, I recommend using a different device.  Modern devices will better specify noise.  Including 0.1Hz to 10Hz scope plots, spectral density plots, and noise tables.

 3. I am planning to do the measurement on the circuit assembled on PCB with BNC cables. Also will the 0.1 Hz to 10 Hz filter you specified in presentation help in getting accurate values? Are such filters commercially available??? or do i will have to assemble it on PCB??

 No.  They are not commercially available. 

 4. Sometimes the noise level shoots up to few micro volts at low frequencies. I searched the local market for 20 uF ceramic capacitor(required at the input onf analyzer)  but couldn't find one. So i bought the electrolytic one.  Then i replaced the same with 47nF ceramic capacitor. Then again repeated the measurement by replacing the capacitor with LPF (single RC with R = 1M ohm, and C= 47 nF) then with HPF with same RC. So can you advise me which way is correct?? 

I normally connect several ceramic caps in parallel.  You need the 10uF to set the cut frequency low (0.001Hz).

Best regards,

Art Kay

Below are links to an article series and a book that I have written on noise.  These reference have much more detail then the presentation.  Details on measurement of 1/f noise are discussed.  Also, process variation of broadband and 1/f noise are covered.

 http://www.en-genius.net/site/zones/audiovideoZONE/technical_notes/avt_012312

 http://www.amazon.com/gp/product/0750685255/ref=s9_simh_gw_p14_d12_i1?pf_rd_m=ATVPDKIKX0DER&pf_rd_s=center-2&pf_rd_r=1WKXS5NEAW890KTTJYFW&pf_rd_t=101&pf_rd_p=1389517282&pf_rd_i=507846

Best regards,

Art

You should consider averaging also.  I think some of this is measurement variation.  Keep in mind that noise is specified with respect to the input.  So you will need to divide your results by the gain before comparing to the specification.  Shielding should help.  Also a post amp in the gain of 10 may help. 

best regards,

Art

Yes.  Noise specified in the data sheet is input referred.  The output noise will always be multiplied by the gain of the amplifier so depending on the gain you will get different noise outputs.  An amplifier with 10nV/rtHz input noise will have 100nV/rHz out in a gain of 10 and 1000nV/rtHz out in a gain of 100.  You can see this in simulation also.

 As a side note.  Broadband noise will normally closely match the op-amp specification.  This specification doesn’t very much from device to device for a given model of op-amp.  So your measured results should closely match the data sheet.  I would, however, recommend that you change to a more modern amplifier for your experiments.  But the point is that you will know when the set up is correct because it will closely match the data sheet specifications.  

Art

Semiconductor devices are produced in large batches (wafer lots). Each lot has somewhat different characteristics then the other batch. The low frequency noise (1/f noise) is the most susceptible to process variation. In other words, you will see the greatest variation from lot-to-lot in the low frequency region. Looking at your data it appears that the broadband region (greater then 100Hz) is close to the simulated results, but the 1/f region (low frequency) is less accurate. This is to be expected. See article 7 for more details.

Another point is that this is these are old products. It is possible ( likely ) that the process has changed and probably improved since the product was released. These improvements would likely also be in the 1/f region. I recommend looking at a more modern device for your experiments if you want to see good agreement between simulation and measurement.

One final comment, you should always take data one decade lower then the decade that you want to look at. So for example, if you want to collect data at 1Hz then measure down to 0.1Hz and discard from 0.1Hz to 1Hz. There is a phenomena called a “low frequency” tail that can give you an error at low frequencies. This is also described in my articles and book.  See article 6 figure 6.17.

Art

Yes.  Make the amplifier following the DUT the same amplifier.  The second amplifier will not add any significant noise because the DUT is in gain.  See calculation below:

Assume

en1 = 10nV/rtHz (Dut noise)

en2 = 10nV/rtHz (post amp noise)

Gain_Dut = 100

Gain_Post_amp = 10

e_n_out = (Gain_Dut*en1 + en2)*Gain_post amp

e_n_out = (100*10nV/rtHz + 10nV/rtHz)*10

Look in the parenthesis.  Note that the noise from the second stage 10nV/rtHz can be ignored compared to the noise from the first stage 100*10nV/rtHz.

One more comment.  The TL084 is the measurement that shows the issue with the low frequency tail.  Set the start frequency to 0.1Hz and collect the data.  Discard data from 0.1Hz to 1Hz.  I think if you do this the TL084 will compare well with the simulation.

Also, collect data to 100kHz.  I think you will see good agreement between measured and simulated between 100Hz and 100kHz.

Art

Nice results!  A few other comments:

 Some additional averaging might further improve your results.  Averaging at low frequency can take a long time; think of time as the inverse of frequency (time = 1/f).  Temperature variations during this time can cause errors because temperature variation causes offset drift which looks like low frequency nose.  So, the point is that if you want to do additional averaging make sure you keep the devices at constant temperature.  Even a few degrees of room temperature shift can cause variations in measured results.  When I am trying to get the best performance I use a thermal bath to get good temperature control.  Thermal baths can control temperature to 0.01C.  Ovens don’t work for this purpose because they have temperature variations of +/-2C (typically).  Simple insulation around the device helps.

This phenomena is discussed in noise article 9 of the series.  I think you have some good material for you paper.  Good luck!

Art

Thanks Sir for your valuable advises, suggestions and articles. Without your help it wouldn't be possible for me to get these results.

Thanks once again.