Noise on high speed ADC

Hi, in past months I designed a data acquisition circuit, in which a 100M 12 bits ADC was used. Now, I am debugging it. I captured the output data of ADC and found some noise in it. The ADC is AD9628. Its input is differential. I connected the positive and negative input of ADC. The data captured was very good. Only one bit variation existed in ADC output. There are two stage amplifier before ADC. The first one is a VGA VCA824, and the second is a full differential amplifier ADA4927. So the noise in the output of ADC should come from the two stage. when I connected the input of the first stage to ground, noise is about 10 bits. It means that the noise is about 1-2mV. Can the noise generated by amplifiers reach this value? or the noise comes the crosstalk of nearby ICs? Are there some methods to reduce the noise of amplifiers?
Thank you!

Are you saying that the noise output of your 2 chip amp is 1-2 mV, what is their total gain? If you look at their data sheet there will be a formula to find the total noise voltage from each.
Are you looking at repetitive noise - clock transients etc, if so could be cross talk from digital circuits, make sure your grounds and Vcc are OK. Try feeding the amps with an extra 100 ohm in their +- Vcc lines with 100 MFD + .1MF down to earth from each(added decoupling).
Frank

Careful layout and good decoupling of the circuit to an analog ground plane is needed for low noise. Without seeing your circuit and layout it's difficult to comment further.

yes, you are right. Careful layout and good decoupling of the circuit to an analog ground plane is required. the noise is not repetitive. It does not seem that noise comes from digital circuit. I saw a reference design yesterday. I found that a inductance was connected between the power supply and the power pin of its analog circuit. it seems that my power supply is not decoupled very well. but can it result this 1-2mv noise?

gongyuwei said:

yes, you are right. Careful layout and good decoupling of the circuit to an analog ground plane is required. the noise is not repetitive. It does not seem that noise comes from digital circuit. I saw a reference design yesterday. I found that a inductance was connected between the power supply and the power pin of its analog circuit. it seems that my power supply is not decoupled very well. but can it result this 1-2mv noise?

Click to expand...

Yes. Digital circuits often have several tens of mV of noise on their grounds so 1-2mV of noise coupled into the analog circuits is quite possible.

I can use a signal copper area for analog power, and connect this copper to power through a inductance. However, when I design PCB, should I use a whole ground plan for both analog and digital circuit? or split the plane for the two?

gongyuwei said:

I can use a signal copper area for analog power, and connect this copper to power through a inductance. However, when I design PCB, should I use a whole ground plan for both analog and digital circuit? or split the plane for the two?

Click to expand...

Definitely a split plane. Here's some general recommendations:

Keep the analog and digital circuits separate from each other as much as possible, and use a split ground plane for the analog and digital circuits.

The two planes should be tied together at only one point, likely at the ADC since the ADC has both analog and digital signals. The digital side of the ADC should be over the digital plane, and the Analog side over the analog plane.

Tying the two planes together through a ferrite bead filter can help prevent noise from the digital getting into the analog.

Never run any digital signals over the analog ground plane or analog signals over the digital ground plane.

Use a ferrite bead fllter between the analog and digital power lines, and have separate electrolytic decoupling capacitors (10µF or greater) on the digital and analog side of the power. 0.1µF ceramic decoupling caps should be connected directly from all IC power pins to their respective ground planes.

Separate analog and digital grounds may be a solution in some cases.

The general problem with separate GND planes is where to connect them. You'll notice, that data sheets of mixed signal devices, e.g. ADC, DAC, µPs with analog functions usually claim to get the separate GNDs connected directly at their respective pins. It's a clever suggestion for trivial evaluation boards, that utilize only one mixed signal chip. For complex boards, a common ground plane is often the appropriate solution.

Separate grounds involve voltage differences between both planes and differential currents injected between them. Both tend to make the board act as an antenna.

- - - Updated - - -

Referring to the original question, I must confess that the problem description is too vague to understand what kind of "noise" (true amplifier noise, digital and switched mode power supply interferences) is observed and where it can be expected to enter the analog signal path. I'm unable to discuss a solution without understanding the problem nature.

I would prefer a schematic that allows to estimate gains and voltage levels.

yes, the most troubled problem is that I can not find whether the noise or interference is true amplifier noise or interference from other circuit. I do not understand how to calculate the output noise of amplifier, so I do not know whether the true amplifier noise reach the value sampled by adc. If the noise is true amplifier noise, I may replace the amplifiers with lower noise amplifiers. If it is interference, I will improve my pcb design and add more decouple C. Can the ture amplifier noise of my analog circuit reach a enough value and degrade my ADC's performence?

What are the gains of the two amplifiers?

You didn't report if the noise level varies with VGA gain.

Even with minimal gain (Vg=-1V), there's a certain output noise density of about 20 nV/√Hz. The largest noise contribution (at least at low VGA gain) is however created by the 2 kΩ ADA4927 feedback resistors in combination with the 19 pA/√Hz current noise density. Assuming 100 MHz noise bandwidth, you'll end up with about 0.5 mV,rms noise voltage.

oh my god, it seems that the true amplifier noise has reached one bit of ADC. The 0.5 rms noise is the result of only ADA4927? It seems that the noise of VGA is even biger. It means that my analog circuit has a larger noise, which is enough to degrade ADC's performence. The gain of the first amplifier is 5V/V, the gain of the second one is 0.5V/V. Is it necessary to replace the two amplifiers with lower noise ones?

ADA4927 is in fact low noise, but you should follow the component values suggested in the datasheet.

A multiplier based VGA won't expose lowest noise. It's a question of application requirements, do you need continuous or step-wise gain control? I presume there are lower noise methods.

What is the maximum signal input frequency you need to convert (50MHz is the maximum Nyquist frequency)?
You should limit the bandwidth of the second stage output to slightly over that frequency to minimize the noise.

crutschow said:

What is the maximum signal input frequency you need to convert (50MHz is the maximum Nyquist frequency)?
You should limit the bandwidth of the second stage output to slightly over that frequency to minimize the noise.

Click to expand...

the maximum signal input frequency is about 10MHz. There is a low pass filter after the second amplifier. Its -3dB frequency is about 10-30MHz

Its -3dB frequency is about 10-30MHz

Click to expand...

I don't know if you have more capacitors that are missing in the schematic, but the shown low-pass filter has about 100 MHz cut-off frequency. If the signal band width is only 10 MHz, this would be a bad choice in terms of noise bandwidth.

As FvM noted the filter shown is much higher than 30MHz. My simulation showed a -3dB point of 95MHz. If you increase C172 to 180pF and reduce R76 and R78 to 7.5Ω each, you will lower the -3dB point to 26MHz with about 0.1dB rolloff at 10MHz. This will reduce the thermal noise by about a factor of two.

Yes, you are right. the -3dB point of low pass filter is about 100MHz. The reason why I used this filter is that I want to get a good gain flatness below 10MHz, but it seems that it is not necessary and changing its -3dB point to 26MHz can achieve lower noise. In past two days, I simulated the PCB structure using Full-wave 2.5D solver. I found my analog power design is not good too. The 26 digital outputs of AD create about 5mV interference in VGA's power and a lower interference in ADA4927's power. the decoupling of ADA4927 is better than VGA. The powers of the two amplifiers are obtained from the two LDOs which generate 1.8V from 5V. In my simulation, I simplify LDO with a C directly. Then, I increase its parasitic inductance to 10uH to add a inductance isolation between digital 1.8V and analog 5V. Simulation result indicates that noise in the two amplifiers power is greatly reduced to below 0.5mV, so can I think that adding a inductance between power and analog +/-5V powers of amplifiers can greatly reduce interferences? Now I think that interference is the most important factor, and true amplifier noise contributes a lot too. The noise of ADA4927 is enough low, but I maybe need to replace the VGA with a lower noise one due to its noise performance in datasheet. In my design, I need to adjust the gain of system step-by-step, so I used a VGA.