Is it possible to reduce noise after ADC

[nadabro]

Hi.

I´m using a sensor with a 24bit ADC (Sensor -> Zero-Drift Op Amp -> ADC) sampling at 41SPS (23bit -3dB at 11Hz, 40Hz cut off), i´m happy with the results, but i´m having a background noise:

ImageShack® - Online Photo and Video Hosting

The results in the image are divided by 12, so it gives a noise count of +/- 5000 (in 23bits).The noise is kind of wavy, do you think this noise is from the circuit or from the ADC? The circuit is already mounted in the PCB so i cant change nothing on it..but using oversampling or other technique is it possible to reduce the noise?

Never used ADC at so high resolution, dont know what to expect from noise count.


Thanks.

 

[mtwieg]

That's a ton of noise, even for a 24 bit ADC. It's very likely a circuit issue in your preamplifier or layout or signal source. The fact that you describe the "noise" as having a shape like "wavy" means it's probably not noise at all, and is just some sort of interference.

It's only possible to filter out signals that are outside your frequency range of interest. If that "noise" overlaps your signal bandwidth, then you're stuck with it for now.

 

[chaojixin]

what kind of sensor? is correlated doubling sampling available?

 

[poorchava]

How is your circuit powered? from "wavy shape" I'd suspect mains frequency in the signal. Did you try to implement some kind of metal shielding to the circuit?

Another thing: what is voltage range of this signal you are measuring? 24 bits at 5V for example equals LSB of roughly 0.3 uV. Noise of +/-5000 counts means around 1.2 mV in either direction. That is entirely possible unless exceptionally good shielding, layout and EMC.EMI design is used.

 

[FvM]

The noise waveform looks like generic electrical noise, true random with low frequency (1/f) enhancement. ADC intrinsic noise looks similar, but also low frequency amplifier noise and many kinds of sensor noise. You didn't mention the ADC full scale input range or gain of external preamplifiers, so I don't want to guess about the details.

As a first point, you would want to calculate the ADC's and other involved component's noise specification. Without modifying your PCB, shorting the signal at different stages for test should be possible anyway. For the ADC, it's usually a software option.

Averaging respectively low-pass filtering in the digital domain is always an option, depending on the signal bandwidth of interest.

 

[nadabro]

Hi.

I suppose i should give more details on the circuit before i post it:

-The sensor is a geophone, its like a accelerometer, it has two analog pins, which generate a very low voltage. The frequencies i want to record are all below 8Hz.
-The ADC is the ADS1211P, 4 channel multiplexer. At that configuration, the internal filter of the ADC is set to 40Hz, with a -3dB at 11Hz.
-The amplifier is the ADA4528-1 and i think is set to max (?).
-About the electronics i can only show the schematic in private message (contact me if you are interested to understand the noise). The PCB was design and made in a shop, so i´m pretty null in electronics.
-The readings with the AMP, starts at +4000000, but since i´m using calibration option on the ADC, the readings start at 0.

I´m using 2 channels, one with Amp and one directly from the sensor to the ADC. I suppose to show better the noise, is to post the readings without the sensor.

This is the readings using on the channel without amplification (the sensor is not connected)
**broken link removed**

Reading on the channel with amplification (the sensor is not connected)
**broken link removed**

Both readings has different lenghts, so the top one looks like is more compressed. So no AMP - Up to +/- 75 counts, With AMP Up to +/- 800 counts.

@chaojixin, its a Geophone, according to wikipedia "is a device which converts ground movement (displacement) into voltage".

@poorchava, the layout of the PCB was ordered in a store, so i dont know the details (but i have the schematic if you wish to see it), i power the board with a 9V supply. Im using a shielded box but it doesnt do much difference (but i can try to improve it). The best results from the sensor i achieve in terms of readings (sensibility) is at 23bits with amplification.

@FvM,
In ADS1211 datasheet
**broken link removed**
The software i use has Low/High Pass filters, it helps, but its not enough..

Any suggestions i appreciate

 

[poorchava]

What kind of supply are you using. If you want exceptionally noiseless power supply (at least to confirm/eliminate power supply as a source of noise) you may use gel rechargeable batteries (example: **broken link removed**).

Other thing: what level of voltage does the sensor actually output? You may want to amplify it more to use whole range of ADC.

You can also check if noise increases when you touch the boards or even more your hand near it.

Also when the board was ordered, did you specify that this line is critical to system accuracy and should be routed to achieve lower noise/interference possible?

 

[FvM]

If I understand your initial post right, noise is one order of magnitude higher with sensor connected?

The both waveforms clarify at least, that most noise is contributed by the amplifier (or interferences coupled to the amplifier input). For a detailed analysis, one would need to know the input referred resolution (1 LSB = xx nV at the amplifier input). But I fear, it's effectively impossible to narrow down the problem without putting hands on the real circuit.

I agree, that ADA4528 has remarkable low noise for a chopper stabilized amplifier, some bipolar OPs still expose lower noise with a low impedance source. I must admit, that I'm generally cautious with "zero-drift" OPs because of their capability to demodulate higher frequency interferences into the baseband. Hopefully ADA4528 is better in this regard.

My general doubts, besides possible power supply inteferences, has already addressed by others, are related to noise generated by thermal fluctuations. Do you have the amplifier and all input circuitry shielded against enviromental influences? Also RF may be an issue, so electrical shielding is suggested as well.

 

[nadabro]

What kind of supply are you using. If you want exceptionally noiseless power supply (at least to confirm/eliminate power supply as a source of noise) you may use gel rechargeable batteries (example: SONNENSCHEIN|A506/1.2S|BATERIA, 6V, 1,2*AH | Farnell Polska).

Other thing: what level of voltage does the sensor actually output? You may want to amplify it more to use whole range of ADC.

You can also check if noise increases when you touch the boards or even more your hand near it.

Also when the board was ordered, did you specify that this line is critical to system accuracy and should be routed to achieve lower noise/interference possible?

I´m using a regular AC adaptor 9V (from a wireless router), i tested with a 12V from different brand but it didint do any difference. About the Level voltage output i really dont know, i only have this specs.

About making noise on touching the board it doenst do any difference.

When i ordered this, i specifiy that i needed the lowest noise possible (they did in a double PCB)..the pcb well designed and soldered.

If I understand your initial post right, noise is one order of magnitude higher with sensor connected?

The both waveforms clarify at least, that most noise is contributed by the amplifier (or interferences coupled to the amplifier input). For a detailed analysis, one would need to know the input referred resolution (1 LSB = xx nV at the amplifier input). But I fear, it's effectively impossible to narrow down the problem without putting hands on the real circuit.

I agree, that ADA4528 has remarkable low noise for a chopper stabilized amplifier, some bipolar OPs still expose lower noise with a low impedance source. I must admit, that I'm generally cautious with "zero-drift" OPs because of their capability to demodulate higher frequency interferences into the baseband. Hopefully ADA4528 is better in this regard.

My general doubts, besides possible power supply inteferences, has already addressed by others, are related to noise generated by thermal fluctuations. Do you have the amplifier and all input circuitry shielded against enviromental influences? Also RF may be an issue, so electrical shielding is suggested as well.

Indeed, the results with the amplifier has much more noise. About the noise of the sensor, is quite difficult to say..because it is always recording something (like ocean waves).

I´ve switch the boards to a new case, but it didint make much difference..its well shielded (closed aluminum box, GND shared from the microcontroller to the ADC board to the aluminum box) and protected from the environment, in a styrofoam box.

This is a picture of the ADC (the channel with Maxim filter is not being used)
ImageShack® - Online Photo and Video Hosting

A quick question, when i know that amplification is enough? I also thought on amplify it even more, the ADC has a internal PGA up to 16, but when i set it to more than 2..i cant get any reading from the ADC, i think with one amplifier i get 4 million value, with PGA set to 2 i will get 8 million value, with 4 i should get 16 million value, which is over the ADC can handle (16 million = 24bits, the ADC reads up to 23bits ~8 million). So if i add another AD4528 (AD4528+AD4528) what is the value i will get on the ADC? Will i get better results from the sensor?


Thanks for the help.

 

[FvM]

Assuming the input amplifier has some gain, e.g. > 20 or 50, total noise will be effectively determined by the amplifier and it's surrounding circuitry only. Cascading amplifiers doesn't help if the total gain is sufficient, redesigning the circuit for lowest noise or possibly changing the amplifier of course can. Can you give an idea of the exrtenal components around the ADA4528?

I'm still interested to hear gain respectively input referred resolution and noise numbers.

A detail, that I won't like is the power supply connector directly beneath the µV input signal connector. In this regard, I appreciate the suggestions to use a battery for test to exclude any power supply interferences.

 

[poorchava]

How is your board designed? As i understand, you provided only schematic to pcb house and they've designed and manufactured the PCB, right? Did you specify which signals are critical? There are special rules when routing for example signals which have very low amplitude. I don't know specifics, but I'm gonna ask pcb design guys at my company.

Anyway, how many layers does the board have? I know that there are some circuits which require signals to be literally surrounded by either VCC or (better) GND. This means that you need at least 3 if not 5 layers. Then you route signal with ground on the sides and also above and below. IIRC USB High Speed specification mentions something like this (of course that's whole different design, I'm only giving examples how sophisticated methods are used to maintain signal integrity)

 

[dave9000]

Hi nadabro,
did you give a look at the spectrum of your data?
I am working on accelerometers too, and that's the only way for me to make any sense out of the raw data.

You should be able to see a broad peak of seismic noise in the range 0.1-1Hz - the ocean waves are visible indeed!
Above 1Hz it is mainly anthropic noise, i.e. motors, cars, etc...
Below 0.01Hz I get mainly thermal and 1/f noise from a number of circuit sources.
Also, if your sensor is enclosed in a box not ermetically sealed, you are going to get a lot of acoustic noise: the same effect you get putting a shell over your ears (not joking).

 

[nadabro]

Sorry for only reply know, too much work lately

Assuming the input amplifier has some gain, e.g. > 20 or 50, total noise will be effectively determined by the amplifier and it's surrounding circuitry only. Cascading amplifiers doesn't help if the total gain is sufficient, redesigning the circuit for lowest noise or possibly changing the amplifier of course can. Can you give an idea of the exrtenal components around the ADA4528?

I'm still interested to hear gain respectively input referred resolution and noise numbers.

A detail, that I won't like is the power supply connector directly beneath the µV input signal connector. In this regard, I appreciate the suggestions to use a battery for test to exclude any power supply interferences.

This is the part of the amplifier
**broken link removed**
The REF is from the reference LM4050 (used to REF in ADC).
The 2.7k resistor shared between two inputs is optional, but now i cant test it without the resistor to check the results.
Simple low pass filter at 7Hz (i think).

About that conector so close of the inputs, i agree with you.

How is your board designed? As i understand, you provided only schematic to pcb house and they've designed and manufactured the PCB, right? Did you specify which signals are critical? There are special rules when routing for example signals which have very low amplitude. I don't know specifics, but I'm gonna ask pcb design guys at my company.

Anyway, how many layers does the board have? I know that there are some circuits which require signals to be literally surrounded by either VCC or (better) GND. This means that you need at least 3 if not 5 layers. Then you route signal with ground on the sides and also above and below. IIRC USB High Speed specification mentions something like this (of course that's whole different design, I'm only giving examples how sophisticated methods are used to maintain signal integrity)

I only had the amplifiers/filter part (simple hand made circuit), the ADC/power supply/reference was all designed by the company. I choose the ADS1211P ADC because i know how to programme it. Its a double layer PCB. For the price (really low) i didint expect great results, i´m only trying to check if its possible to low the noise.

I understand that more 2 layers is necessary to get good results, but the the price is way too much for me.

Hi nadabro,
did you give a look at the spectrum of your data?
I am working on accelerometers too, and that's the only way for me to make any sense out of the raw data.

You should be able to see a broad peak of seismic noise in the range 0.1-1Hz - the ocean waves are visible indeed!
Above 1Hz it is mainly anthropic noise, i.e. motors, cars, etc...
Below 0.01Hz I get mainly thermal and 1/f noise from a number of circuit sources.
Also, if your sensor is enclosed in a box not ermetically sealed, you are going to get a lot of acoustic noise: the same effect you get putting a shell over your ears (not joking).

With a natural frequency of 4.5Hz i dont expect to get so low frequencies (<0.1Hz) with the geophone (i wish i could!). I live 100m from the ocean, so i understand what are you saying about the ocean waves, the noise gets much higher when the waves are >3m.
What makes really noise is trucks, specially the garbage truck, it gives very high spikes..but usually a low pass filter applied in the software remove most of the "urban" noise and i can see low seismic events.

About the hermetically sealed box, the geophone comes with its original box (round, waterproof) and its completely closed.

But i´m trying to fix the electronic noise, than i will study more the geophone noise.

This weekend i will test with a battery and post back the results.

 

[FvM]

I won't be surprized if the REF connection of the preamplifier is the main source of excessive noise. You sill don't give a complete schematic, so wed need to guess about possible filtering of the REF voltage. But you'll need very effective low frequency filtering to avoid noise injection from this side.

As another point, what's the geophone output impedance, what's the designed gain of the input amplifier? The circuit looks mixed-up.

 

[shahbaz.ele]

This is important point that FvM raised, what is o/p impedance of the geophone.
I think you should post the circuit diagram of geophone + amplifier + adc .
Its a helping forum, do disclose your circuit si that experts can help you better way.
Buy the way i am also working on geophone. I am using OP177 amp in simple inverting stage having good results.

 

[dave9000]

From the short specs of the geophone posted by nadabro the coil resistance is between 150 and 380 ohm.
As for the inductance we can only make an educated guess; up to a few Henry(s) it shouldn't matter that much, given the limited frequency range.
Also we can see that the 2.7K input resistor has been choosen to resonably damp the mechanical resonance.

Assuming a 240 ohm sensor, the voltage gain is around 4100.
The opamp has got a voltage noise of 97nV p-p in the band 0.1-10Hz, so we get ~400uV p-p at the ADC input.
At a resolution of 24 bit, and using the internal 2.5V reference, this noise alone makes up for 2670 ADC counts!

If these quick calculations are correct (someone please proof-check) it means that you are already close to the limits of the circuit: some polishing is possible, miracles aren't :-(

P.S. I assume you are not trying to compensate the sensor response at low frequencies, that would only make the problem worse...

 

[nadabro]

This is the schematic:

**broken link removed**

I had a simple schematic for this geophone (same brand and model), it was designed for what i want (record seismic events): so insted of the old ADC on the schematic i replace with the ADS1211P and keep the rest of the circuit..
**broken link removed**

The idea was to have 3 channels and compare to see which channel has best results. The sensor have a natural frequency of 4.5Hz but its possible to get lower frequencies electronically, the idea is get all frequencies from the sensor below 10Hz.:
1x Channel without filtering/amplification..geophone connected directly to ADC. Results: need amplification.
1x Channel with amplification (ADA4521-1): good results, noise issue.
1x Channel with amplification and Low Pass Filter (ADA4521-1 + MAX7401): MAX7401 introduce too much noise.

But in the original schematic the OP-AMP was the AD8628, Analog mention in the website the new ADA4528-1, so i order this one (just plain and simple:lol
The Low Pass filter is to filter >10Hz (its missing the capacitor in CLK). The ADC does a good job using the internal filter at 60Hz (-3dB at 12Hz)

I´m pretty sure the sensor have 380ohm.

So, what do you think is the cause of the noise?
When amplification is enough?
Is it possible to improve the schematic? ex: replace the amplifier with a better one, remove add components, etc?

I can also change the ADC, but i prefer using the ADS1211 or similar, because i´m more comfortable on programming it.

My intention is to make a second board with improvements..but i will order in the same company so i will have a 2 layer pcb limit.

I appreciate any help..this system is working ok, but if its possible to improve, i´m willing to try.

 

[dave9000]

Minor points:
- I'd connect the negative inputs of the ADC to REF instead of AGND.
- if you intend to keep the MAX701, add a jumper to disable the clock when not in use, to minimize supply noise.

 

[FvM]

Using the ADC in differential mode referred to Vref (your virtual ground) should help to reduce the impact of reference noise. But you still get noise above the ADC resolution due to the unfiltered reference.

Using a virtual ground for a circuit dealing with uV levels isn't really a good idea. With uV and below 1 Hz signal frequencies, it's even worse. I would provide blocking of the virtual ground in a 1000 uF range.

I don't know the geophone specification, but I doubt that's it's intended for operation with zero ohm load impedance of a I/V converter. If the sensor impedance is mostly resistive, it possibly works without serious problems, but I would expect a voltage amplifier circuit.

MAX7401 dynamic range is miles away from the application requirements, I think. It rather fits a 12 bit ADC.

 

[dave9000]

Vref in this circuit is used for two logically different tasks:
- as a virtual ground for the amplifiers and the input sections of the ADC
- as a voltage reference for the A/D conversion process
Given the high PSRR/CMRR of the ICs at low frequency, the quality of the virtual ground doesn't look critical to me.

As for the voltage reference: any unwanted noise or drift on this signal will result in gain modulation (and common mode noise, mostly rejected).
In other applications this could lead to severe problems, because of intermodulation distortion. In this case it's utterly negligible (I extimate ~30uVpp in 0.1-10Hz for the LM4050, i.e. some 12ppm gain modulation).

I agree with FvM on trashing the MAX7401 and on using a voltage-amplifier topology for the sensors - maybe leaving the input load of 2.7K for damping control.