Nanovolt range amplifier

Konstantinos,

This is an extremely low signal voltage level, below the noise level of virtually all amplifiers. Can you explain the purpose of this amplifier. What is the signal source. What is the signal bandwidth.

Regards, Bruce.

Hello and thanks for the reply.

There are some sensors in the physics lab of the university. I actually don't know what they are measuring. They produce a slow variating DC signal in the nV range. I will have to measure 8 sensors one after the other, so I'm thinking I will have to make 8 amps and afterwards an 8-input analog switch.

I think that it might be possible to design a chopper amp. but instead of using a simple analog switch in the input, I could use an 8 to 1 analog multiplexer followed by an AC amp. and then a single analog switch.

I'm also thinking if it is possible to have a low voltage sinewave added to the signal, amplify the hole thing and then add an equally amplified but 180deg. phase-shifted sinewave to remove the original sine.

I also think if it is possible to just design a discrete jfet voltage amp with gain of about 100 followed by a chopper-stabilized opamp for every sensor.

An other difficulty is that there is no nanovoltmeter available to characterize and calibrate the amp.

Regards

Beiefly,  working at nanovolt levels will require at the minimum extreme thermal insulation to minimize temperature drifts. This is because resistor end-cap thermocouples and solder connection thermocouples have tempcos ranging from 50nV/Deg C to 50uV/Deg C.

The design of the amplifier is dictated by the bandwidth of the amplifier. Assuming DC operation, a chopper-type amplifier is indicated, probably discrete using, some big JFets. It is possible that the  ADA4528 might be sufficient to do the job using an IC op-amp.

If you will supply more details, I can comment further.

Hi Jerome,

If don't mind i would like to ask you some questions. I've been searching amp that amplifies auditory brainstem response signal (nano scale) to mili or micro scale.I am focus on precision instrumental amp. and its variations such as chopper, thermocouple. As you know this signal as small as noise, could you recommend anything else to remove this noise ?

Best wishes,

ERKUT

ERKUT -

     A  company  (EM Electronics) manufactures amplifiers which are claimed to be able to amplify down to pico volts. I believe that these amps use very large J-Fets (Interfet IF3601 or IF3602). The J-Fets are designed into a chopper amplifier. The EM Electronics amps'  specs are quite vague, especially the drift specs. Over what time period do you need the preamplifier to be stable with 1nV of offset? Do you need a DC amplifier ?  How much input capacitance can you tolerate ? (The big J-Fets will have several hundred pF of input capacitance.)

To see a circuit diagram of a nanovolt preamplifier using big J-Fets, try using the google patent search engine and look under Keithley nanovolt preamplifiers patents. Keithley has some great specs on their 8-digit volt meters, reading these specs was an education for me. Keithley does talk about drift in their specs.

One way to lower the noise of a nanovolt preamplifier is to simply use a very low-frequency filter. I am using 0.35Hz in my design. Keithly's nanovolt preamplifier claims 3.5Hz bandwidth and input voltage-noise of 6nV p-p, which is less than 1nV rms noise. This I believe is the best in the world.

The amplifier I am designing has a target input offset-voltage drift spec of about +/- 1nV every hour. This is achieved only with extreme thermal isolation, since the passive component thermocouple TC's are far, far greater than the offset-drift spec of the amplifier.front-end.

Thank you for your recomandations,

I am working on ABR (Auditory Brainstem Response) signals which is between 50Hz-3000Hz I am already gonna use filters.Actually, this is my undergraduate thesis so that I am not enough to analysis like a pro. So I dont know what period time i need or how many capacitance I can tolerate. I guess,I am on begginig of the way to create something useful.

I 'll take a look those, voltmeters and nanovolt preamp. 

ERKUT:

     I thought that you were dealing with a DC-coupled nanovolt preamplifier, therefore much of what I said will not apply to your design.

     Moving the lower cutoff frequency from DC to 50Hz will lessen the problems of noise and drift  that DC preamp designers must deal with.

     However one issue you might want to consider now is the issue of A/D conversion. The signal will have to be converted to digital. The presence of noise on the analog signal to be digitized will likely cause the lowest digit to dither at some rate. The question is, how much dither can you tolerate on the lowest digit of the digital readout ?

     For example, if the signal input is grounded, you might expect the digital readout to show all zeroes all the time. This is rarely the case on a practical signal-chain system. The price of all zeroes all the time is reduced sensitivity. Therefore I suggest that you look into signal estimator circuits, which may be used to enhance the signal-to-noise ratio of the analog signal without lowering the input noise or the filter cutoff frequency.

Jerome Friedman said:

One way to lower the noise of a nanovolt preamplifier is to simply use a very low-frequency filter. I am using 0.35Hz in my design. Keithly's nanovolt preamplifier claims 3.5Hz bandwidth and input voltage-noise of 6nV p-p, which is less than 1nV rms noise. This I believe is the best in the world.

I agree for limiting bandwidth like that is can reduce intrinsic noise of amplifier effectively.
But, with what ways the semiconductor company like TI or other measuring the input voltage noise (peak to peak) of their low noise Op Amp which can ranging from 0.1 Hz to 10 Hz range?
I has read some datasheet of low noise Op Amp gives the specification that input voltage noise of 0.1 Hz to 10 Hz is several tens nanovolt (peak to peak).

Thanks,

Saaddin

Jerome Friedman said:

The OPA111 datasheet shows a method of paralleling op amps to reduce the equivalent input noise voltage. Buy some low noise op amps and parallel them until you reach the required equivalent input noise voltage that you need. Use extremely small resistor values as well.

Hi Saaddin,

there two kinds og noise, thermal noise and flicker noise, also called 1/f noise, as it increases for low frequencies.
So you have to suggest, which type of noise effect your measurements mainly.

Thermal noise: Paralleling op amps could help.
Flicker noise: use a mux at the start of your Signal chain and alwasy Switch between a short at the Input of your amp chain and the real Signal.
With some math you get really pretty results.
Check out Cypress PSoC Chips for that solution, sorry can't find the document yet.  Note: FOUND. check attachment.Correlated Double Sampling to reduce low f noise.pdf

As I mentiones above, the best method is depending on the main type of noise you will find in your amplified Signal.

With best regards

Gerhard

With respect to the amount of noise-reduction to be gained from paralleling chopper op-amps, there is some question about this. It depends upon whether chopper op-amp noise below 10Hz is Gaussian or not.

Most application engineers will stick to the Gaussian equation for predicting the noise reduction, however this works only for Johnson-type noise, such as resistors, and noise which behaves like Johnson noise. 1/f noise may or may not observe this rule.

If I suppose a single noise pulse at one of the op-amp inputs, and there are 5 op-amps in parallel. then the said single noise-pulse will be attenuated by 5 at the array output, i.e. linearly.

The question is whether the totality of all of the chopper op-amp noise energy, including 1/f noise, pink noise, etc, obeys Gaussian rules. My experience is that the chopper op amp noise doesn't look like Johnson noise in the time domain, and for the op-amp array the noise is reduced more than Gaussian math predicts, however not as much as a straight linear reduction of 5:1 (for 5 op amps). 

To strictly prove the degree of noise-reduction, one would need to separately characterize each op amp for noise and then measure the noise of that particular array, a time-consuming process. Further, there are many sources of noise at 1nV rms noise levels, especially below 10Hz, which color the results.

There is a paper which discusses the noise of chopper op-amps, see "Unspecified Low-Frequency Noise in Chopper Op-Amps" - IFSA Journal.

Hello all.

The datasheet of the LT1028 gives some useful hints on paralleling OpAmps to reduce input noise.

See page 17.

Regards, Thomas

The LT1028 datasheet mentioned by Gay also has an important note for nV amp designers -

" RESISTORS MUST HAVE LOW THERMOELECTRIC POTENTIAL"

Some resistor end-cap thermocouple potentials can reach 40uV. With constant temperature,  the two end cap voltages tend to cancel. However, when there is a temperature gradient present, noise-voltage problems arise very quickly, since 40uV is 40,000nV.

It is worth noting that, for the case of the LT1028, where input noise-current is significant, the input noise-current flows mostly in the feedback impedance and not in the input or source resistor. This is because if the input noise current flowed in the input impedance, this action would create an op-amp input error-voltage which would and does draw the noise-current around to the output.

The chopper-assisted zeroing-circuit shown on page 20 of the LT1028 datasheet is a variation of the original idea of the LM669 IC. The LM669-method was also shown in one edition of "The Art of Electronics". However the original LM669 idea of using a second op-amp to correct the offset-voltage of a primary op-amp worked for the inverting configuration only, and not for using the high-impedance (+) input. The idea of the circuit of page 20 of the LT1028 datasheet is to extend the LM669 method to the use of the High-Z (+) input of the primary op-amp.

What is not made clear about the method of page 20 of the LT1028 datasheet is that the input noise of the (+) input of the compound amplifier varies between the noise spec of the LT1052 and the LT1028, depending on frequency. It is desired that at very low frequencies, the LT1052 noise will dominate. At some higher frequency, the fact that the LT1052 is connected as an integrator will cause the LT1052 noise to drop out and release the input node to the LT1028 noise. Since the LT1028 noise is about 1nV rms at 10Hz, this datapoint should be preserved for nV measurements.

The best chopper op-amp for use in this compound amp is the one that has the lowest input noise-voltage below 10Hz. The chopper op-amp integrator capacitor needs to be adjusted such that the input noise-voltage crossover frequency (from chopper noise to LT1028 noise) is about 0.2Hz. Since compound amplifiers may oscillate, the circuit needs to be checked for stability using the op-amp SPICE models.

Gerhard Kreuzer said:

With some math you get really pretty results.

Gerhard,

I have read about your attachement, it is a nice explanation. 

Can you tell the literature should I read about "some math" that can make us get really pretty results?

Thanks,

Saaddin

Saadin -

 Need for shielding - This may be discovered by building a test amplifier and then record the extraneous EMI signals picked up by various cables and.connectors. Hammond Company makes a line of small aluminum boxes which are excellent for enclosing modular amplifiers. The aluminum box may be grounded to form a first electrostatic shield. The aluminum box may then be wrapped in a thermal insulating material such as cotton fabric and then placed inside a thermos bottle, preferably metal. The thermos should be grounded also. This scheme reduces thermal drifting and EMI interference.

I would first use a recording voltmeter to measure the signals being picked up by the test amplifier so enclosed as above with no cable at all, just a resistor to ground to simulate the source impedance you will be measuring in the application. This will show the self noise. For the test amplifier I would recommend either the MAX4238/39, the LTC1250, or the ADA4528. These chopper amplifiers will show input noise voltages at or near 1uV p-p for a bandwidth of 10Hz. The ADA4528 has a p-p input noise-voltage of 100nV for DC - 10Hz and an input capacitance of 50pF. Five of these in parallel will result in approx 2nV/Rt-Hz input noise voltage for DC - 100Hz, which does not include any source, connector, wiring,  resistor or thermal voltages - these will all increase noise. There is a company in Canada, Proto Advantage, which will mount the IC SMD packages on an adapter board for about $20, which adapter will plug into a pluggable breadboard for experiments.

For very small voltages, a connector must be evaluated. Generally sliding contact connectors will emit spikes at uV levels, especially just after connecting. However after being allowed to sit undisturbed for several hours at constant temperature, they may settle down. This is where the recording voltmeter is most valuable, to find out what will happen. Cables must consist of matching solder connections and materials, and the signal wire and common wire must be very close/twisted and inside a shielded cable. The use of fluoresent lights and cell phones etc should be avoided, along with all radiators of RF energy. I have seen fluoresent lamps which will reset a PC when the lamp is turned on.

Refer to the Keithley 1801 Nanovolt Preamplifier data sheet/manual or the EM Electronics A10 Nanovolt Preamplifier data sheet for tips on how to connect the preamplifier to the source. Also refer to the app notes for any six- or eight-digit digital voltmeter, since these will have resolution into the nV region. The service manuals for many older six-digit DMM's are available on the internet, and these contain valuable tips and show circuits as well.

You definitely should read the Keithly Low Level Measurements Handbook.

www.keithley.com/.../LowLevMsHandbk.pdf

Saaddin -

     As far as using a single supply for a precision amplifier application, this would require some thought and analysis. I would ask the support engineers at T.I., Linear Technology, Analog Devices, etc for the pros and cons of doing this, and mention that nV sensitivity of the amplifier is required.

     Note that in your proposal, the precision amplifier common will be at +2.5V with respect to the power-supply common. I have never tried this.

     For my lab, I have gone to the trouble of sinking a rod into the earth and bringing this wire into the lab as earth ground. My 120V AC lab power is isolated from the mains by a 500 Watt AC isolation power transformer. The secondary of the 500 Watt transformer is used to create a quality 120V AC supply, with the earth ground of the quality 120V outlets being connected to the earth ground from the rod I sunk into the ground.

     All of my electro-staticshielding is connected directly to the earth ground, and there is a 1K resistor connecting the earth ground to the DC power supply common. In your system, the earth ground would have to be connected to +2.5V of the precision amplifier supply. I am not sure about this.