Our application is driving the AD9240 14 bit 10MHz A/D.This is an instrumentation application so settling time, low noise and low drift are the more critical parameter. The goal is to have switchable gains up to 1000x using multiple stages. The input could be a square wave or pulse mixed with random noise.
Using the parametric search the ADA4895-1 and/or AD8021 appear to be the top choices with limits of 3nV/rtHz, GBP > 130M, The AD8099 and AD829 met those but the VosTC was 10x or more higher than the others.
Suggestions and comments welcome.
Note: The AD9240 was selected for compatibility with the test platform. Ultimately one of the newer serial A/Ds will be used, 20MHz or faster so an amplifier chain that can support the higher speeds is of interest.
I would like to be able to limit the bandwidth of the ADA4895-1 op-amp. Is there a limit to how large the feedback capacitor can be used on an op-amp that is not unity gain stable? For example using the ADA4895-1, minimum stable gain of 10x, could a feedback capacitor cause the op-amp to oscillate?
226 ohm feedback resistor
25.5 ohm gain resistor
35pF feedback capacitor
Sorry for the late response.
I'll take a look at this and will get back to you soon.
There are five gain amps and a unity gain buffer in a chain driving the AD9240. Relays switch the gain amps into and out of the chain. Starting from the AD9240 they are set for gains of 2.5, 5, 10a, 10b and 10c. The input is shorted to ground. The three first amps use an AD8021. A 10pF compensation capacitor was added to 10a to see if that would improve the histogram results over the ADA4895-1 that was there before. Amps 10b and 10c use the ADA4895-1. All use the recommended resistor values from the datasheets. Using the AD9240 histogram data is possible. The amplifier stage is also viewed on a scope (350Mhz 10x probe with the ground lead connected).
In the most recent test 10a has a 10pF compensation capacitor to provide unity gain stability. Comparing the three 10x gain amps and the 2.5x amp they each have +/- 4 counts of histogram spread. The 5x is +/-8. Possibly due to the 8pF feedback capacitor. The next test will be to reduce the capacitor to 2pF and then try a 10pF compensation resistor if that doesn't help.
If the 2.5x amp is added to the chain with any of the three 10x amps the histogram is +/-6 counts.
If any two 10x amps are cascaded the histogram is +/-10 counts. The trace on the scope is noticeably fuzzier but is random.
The schematic of amp 10a is attached. 10b and 10c are similar with the ADA4895-1 recommended resistor values.
The pcb layout is attached. The green shows the power plane voids (they should have been larger). The supply bypass caps are 0.1uF. The relay is visible at the left. Since U4 was originally an AD4895-1 the 10pF capacitor, C160, is on top of the AD8021.
Other than making the power plane voids larger what is the likely cause of the instability that appears at gains greater than 25x?
The AD8021 data sheet recommends a 1k feedback resistor. Even at 1x the value is 75 ohms. Driving 50 ohms to 5V will exceed the amplifiers output current.
If you were referring to the ADA4895-1 the data sheet for that amp recommends 226 ohm and says lower values will increase the distortion.
The 2pF feedback capacitor should compensate for the input capacitance.
If the feedback capacitor roll-off frequency is higher than the op-amp 10x gain at the frequency wouldn't that prevent oscillations?
The AD9240 is differential. It is being driven single ended. There is another AD8021 buffering the AD9240 input.
The AD9240 data sheet says this: "Note that the addition of a small series resistor of 30 ohms to 50 ohms connected to VINA and VINB will be beneficial in nearly all cases." The 51 ohm output resistor on each stage was added to isolate the amp output from the filter capacitor on the next stage (current 0pF).
The 1k feedback on the follower was chosen to more closely match the non-inverting input impedance. I suppose that isn't essential since the non-inverting input is used for offset and is unchanging after that. The non-inverting follower inputs will eventually be driven by D/As but still with an unchanging output while the A/D is measuring..
I tried taking the feedback capacitor off when U4 was an AD4895-1 it still oscillated.
As to noise, I was hoping for better than that, The ADA4895-1 has lower noise than the AD8021 but seems to be more tempermental.
I have produced 16 bit boards in the past with histograms matching the A/D datasheet but only with input frequencies under 1MHz.