Very strict band-pass filter design

Discussion in 'General Electronics Chat' started by richiechen, Apr 4, 2014.

  1. richiechen

    Thread Starter Member

    Jan 1, 2012
    Dear all
    I am here looking for help on amplifier+bandpass filter design. We have built up a 6 order band pass filter with two stage amplifier. The performance, however, was not satisfactory.

    The design goal is to measure a very weak 2kHz signal from photodiode. The current will first go through a Transimpedance Amplifer(1MOhm) then go to bandpass filter and then be amplified by 77dB.

    The procedure is as following:
    PreAmplifier (Made by an OpAmp)-> 4 order BPF(active filter, no gain) -> 34dB Amplifier(inverting OpAmp) -> 2 order BPF(active filter, no gain) -> 43dB Amplifier(inverting OpAmp).
    We put one stage Amp between BPF to avoid saturation caused by BPF OpAmp bias.

    Design requirements:
    1. very low noise, especially at 1980Hz--2020Hz.
    2. 77dB of gain. (under such gain, noise floor may lead to OpAmp saturation..)
    3. stable, of course
    4. Need less tune of center frequency, or preferable no tune.

    What are not satisfactory for our current implementation:
    1. Gain can only achieve 63dB. If gain is further increased, circuit will oscillate.
    2. Too large noises. noise floor at 2.3k is -15dBu, which can be found in eclosed pircutre.
    3. center frequency of BPF shifts by 300Hz, which is unacceptable since it leads to a -15dB gain.

    My questions are:
    1. Should we use passive filters to reduce noises? If so, I am afraid that 1kHz is too low for inductors..
    2. If active filters are used:
    a. Shall we use bpf with gain to reduce stages, therefore to reduce noises?
    b. Shall we buy commercial active filter IC to avoid tuning? Such as this one: UA42:
    c. Shall we use high pass+ low pass, instead of single frequency bp? Maybe they can be more stable? And no need to tune that much?
    3. For the amplifier choices, can we use BJT/MOSFET? All the amplifiers I built are based on OpAmp, since most amplifiers on the internet are about OpAmp.. What are the tradeoffs of BJT/MOSFET amplifiers and OpAmp?

    The enclosed pictures:
    1. the noise floor at 2.3kHz(center frequency of BP) reaches -15dB, which is too large. No signal has been applied at this time.
    2. When a 2kHz signal is applied, strange harmonics showed up. Where are they from? The OpAmp is far from saturation and there are no harmonics at the very beginning signal source.
    Moreover, what are the "hills" besides the harmonics? We are using a single pass band BPF...

    Thanks for reading up all this. :p Any comment will be appreciated.

  2. joeyd999

    AAC Fanatic!

    Jun 6, 2011
    A lock-in amplifier is likely a better approach than filters.
    richiechen likes this.
  3. MrChips


    Oct 2, 2009
    Try lock-in amplifier.
    richiechen likes this.
  4. richiechen

    Thread Starter Member

    Jan 1, 2012
    Thanks for your recommendation and Mr. Chips.
    Yes, after filters Lock-in Amplifiers are used. We want to amplify the signal because a 16 bit ADC is used to adopt data..

    And in our environment, signal at 1kHz is 60dB larger than our target signal of 2kHz. Therefore, we need bpf to reject 1k first and then amplify 2k to 8Vpp and be measured by a lock-in Amp based on a 16 bit ADC.
  5. AnalogKid

    Distinguished Member

    Aug 1, 2013
    A lock-in amp might be the way to go (+0.1), but sticking with your original signal path for a moment...

    In my personal experience, high-order bandpass is easier to achieve with successive highpass and lowpass sections rather than true bandpass sections because there are equal-component circuit topologies that make component matching and tuning much easier. And for your application, component matching is everything. The tradeoff is that there are more amplifiers in the string, each one adding a little more noise.

    Oscillation at 77 dB gain but not at 63 dB sounds like a layout problem. You'll need ground planes, power planes, guard bands, and lots of decoupling. Analog Devices frequently shows the opamp power pins decoupled through a 10 ohm resistor for low noise and high gain applications. Because your frequency of interest is so low, another option is to put a pair of voltage regulators at each amplifier chip. A 78L12 and 79L12 in TO-92 or SMT packages will give you some serious isolation.

    Even with equal-component designs, for the frequency accuracy you want you will have to hand-select the capacitors and adjust each resistor. 0.1% components will not be good enough. Plus, the temperature coefficients will have to track.

    Rather than quadratic filter chips like the TI part, consider the high-order switched-capacitor chips. 8-pole filters are relatively easy if your application can handle a little switching noise. Each filter section would have to be hand-tuned, but the tuning is done by varying a reistor or a clock frequency, way easier than a dozen capacitors. Linear Tech and Maxim are the big players.

    The options for the lowpass section are much better than for hipass:

    The ratio of bandwidth to center frequency is 50:1, a very high value. Have you tried a resonant approach, such as a tightly-tuned twin-T circuit or even an L-C circuit with a gyrator acting as a synthetic inductor?

    Also, if you have a defined noise problem at one frequency, such as 1 KHz, a twin-T trap can eliminate it.

    Of course, all of these filter type create phase shifts in the original signal. What is the application, and can it handle frequency-dependent phase shifts?

    richiechen likes this.
  6. richiechen

    Thread Starter Member

    Jan 1, 2012
    Thanks for your reply, we are actually using 1ock-in amplifiers already. However, since we need 16-bit precision, the 2k signal has to be amplified first. And the 2kHz signal is going with very large 1kHz signal(60dB larger, comparing with 2kHz signal), therefore, it needs to go through a BPF first.

    About phase requirement: There are no restrictions on phase. We only need to measure the 2kHz amplitude。

    Could I ask some questions?
    1.What do you mean by component matching?Could you please illustrate more about it?
    2.Why 0.1% precision of resistors/caps is not enough? What if we use high pass+low pass where the pass band will be larger?
    3.How to deal with temperature drift? We currently are not going to deal with this problem.. Just curious.
    4."The ratio of bandwidth to center frequency is 50:1." Do you mean Centerfrequency/bandwidth, which is Q?
    5. Could you explain more about twin-T circuit? Is it similar to a lock-in Amp?
    6. Could you please explain more about L-C circuit with a gyrator? What is the benefit of such design comparing with active filters?

    1. Thanks for the recommendation of decoupling. We only used ferrite beads with five caps for each OpAmp.. But I think the oscillation does not come from resonance of Power line, since it happens at the exact center frequency of BPF.
    2. Thanks for the switched capacitor IC. I will check about it. We did not use it for noises and I will further study it.

    I really appreciate your reply, which provides me more possibility to research on..