1
\$\begingroup\$

I would appreciate suggestions or best practices for designing a stable amplifier. My goal is to design a 94 dB gain low-noise op-amp amplifier for a 100 kHz signal with a bandwidth of 10kHz.

At this point I have made several designs but, in the end, the circuit does not work consistently. It sometimes shows the expected gain (even when only 47 dB is used), but then, after a few seconds, the gain drops significantly or the noise floor increases. Find my circuit below using the LTC6226 (1nV/√Hz 420MHz GBW):

enter image description here

Some of the things I have tried already are (note: the circuit was built on a breadboard):

  • Very short wire connections.
  • Supply capacitor directly to terminals. Big supply (680 μF) capacitors.
  • Single or double op-amp.
  • Separate boards and supplies for each op-amp.
  • Increase gain to reduce BW.
  • Use lower GBW opamp.
  • Match input and output to 50 Ω.

I have also found some great articles 1 2 3 about this but my problem persists.

One of the things I have not clear yet is the influence of the input bias current on the gain and stability of the amplifier. I thought the GBW was the primary factor. I thought the input bias current was mainly related to the minimum detectable signal.

Any suggestions on these topics would be great.

\$\endgroup\$
11
  • 2
    \$\begingroup\$ This seems like it may be a good question -- but there's not actually a question stated anywhere. Please consider adding a question that's clear. \$\endgroup\$
    – JYelton
    Commented Nov 29, 2021 at 18:01
  • \$\begingroup\$ Specify supplies please. And correct purported 220 F capacitors. \$\endgroup\$
    – tobalt
    Commented Nov 29, 2021 at 18:36
  • 1
    \$\begingroup\$ Even if you are satisfied with a circuit built on a breadboard, you'll likely find that a circuit-board version gives different results. For such high-gain circuits, ground paths must be very carefully considered - it is easy to get gain variations when output currents feed back to input via ground currents. Power supply GND affect this too. \$\endgroup\$
    – glen_geek
    Commented Nov 29, 2021 at 19:43
  • 3
    \$\begingroup\$ 94 dB gain using only 2 opamps sounds to me like asking for trouble. I would stay below 20 - 30 dB per opamp and that's already pushing things. Big supply (680 μF) capacitors. Are those caps any good at ~100 kHz? I would choose 22 uF and a 10 nF in parallel. Watch this video by Dave from the EEVBLog about bypass capacitors: youtube.com/watch?v=BcJ6UdDx1vg Then on the same channel watch the videos about opamps. \$\endgroup\$ Commented Nov 29, 2021 at 20:49
  • 1
    \$\begingroup\$ Showing gain for a second or two and then dying sounds like a thermal issue. Are they getting hot? If they are, such high GBW op-amps on a bread board are just begging the Circuit Gods to gift you with unwanted oscillation. Have you checked? I suspect oscillation leading to heat leading to poor performance. \$\endgroup\$
    – TimWescott
    Commented Nov 29, 2021 at 20:52

1 Answer 1

2
\$\begingroup\$

Few know that GBW demands increase with the square of Q, thus you need a GBW = 10^2 * 100 kHz = 10 MHz at 0dB gain. With an overall closed loop gain of 94 dB ~ 2.5e4 you need more low gain stages with a total GBW = 2.5e4 * 10e6 = 2.5 e10 total.

Thus using inverting input for a BPF with 420 MHz GBW or 4.2e8 and a max gain of 420MHz/10MHz = 42 you need 2.5e4 /42 = 595 stages. Clearly using integrator internally compensated Damm Fast OA, is not the solution and rather you need video amps that are not unity-gain stable.

However reducing the filters to unity gain and using wideband high gain amps can reduce the \$Q^2\$ to a linear GBW issue. (Added see my pro bono cct design below).

With only 330 MHz non-inverting GBW, at 100kHz with have only a gain of 3.3e3 of open loop x 2 stages = 6.6e3 gain reduced in both Q and gain with load effects of 10 ohms relative to open loop output impedance to get far less gain of 47 dB or ~ 2e2 or reduced to 3% of avail BW limited by current.

Open loop Z, is approx Voh/Isc=5V/60 mA typ= 83 ohms which in 2 stages limits it by ~39 dB.

Using the same Op Amps but with metal foil shields (cans over PCB gnd plane), this might be stable and give 94 dB gain Bessel shaped Q=10 @ 100 kHz with GBW=330 MHz OA's and gain stage at front and back ends depending on interference sources to prevent saturation but problematic if cascaded raising sensitivity to <0.01 pF positive feedback.

enter image description here

\$\endgroup\$
9
  • \$\begingroup\$ I agree about the need to shield individual stages. Also need to make sure the isolation between output and input is greater than your gain, or else you could end up with an oscillator. Output/input isolation probably needs to be >120 dB with 94 dB of gain. Tricky to achieve unless you pay close attention to layout. Even a low-leakage path on your PCB could cause you trouble, \$\endgroup\$
    – SteveSh
    Commented Nov 29, 2021 at 23:04
  • \$\begingroup\$ The isolation required must be much less 0.1 pF and even < 0.01 pF which imposes special double Faraday shielding and semi-rigid coax inputs. Been there done that. @SteveSh FWIW Did you know about Q^2? \$\endgroup\$ Commented Nov 30, 2021 at 0:35
  • \$\begingroup\$ Yes isolation from output to any input is critical in all 4 stages as well as between stages with the exception of the dual filters as they are unity gain in order to achieve the GBWQ^2 requirements. Does anyone need proof ? ask a better question. \$\endgroup\$ Commented Nov 30, 2021 at 0:37
  • 1
    \$\begingroup\$ Tony I am interested why you posit such high GBW product. Can you expand a little about this "Q²" rule ? I have built noninverting sine wave amplifiers with single stage gain over 1000 and results were always entirely as expected from the usual GBW curves in datasheets, i.e. max frequency is approximately GBW divided by closed loop gain. \$\endgroup\$
    – tobalt
    Commented Nov 30, 2021 at 7:44
  • \$\begingroup\$ Although I have never seen it written anywhere of this term, but it makes sense to me.and both simulators are in agreement on this relationship, but implemented quite differently. \$\endgroup\$ Commented Nov 30, 2021 at 10:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.