I have designed analog filter for two years, I designed gm-C bandpss filters working at frequency about serveral MHz.

I am now amazed at how to design bandpass filter at serveral MHz with small ripple. For example, I choosed filter architecture, synthesize it and designed it , the simulation results show it has ripple with 1dB, but after I finished layout and do postsimulation, I found the ripple will changed into 2dB, and the filter bandwidth will also changed.

I think the change in bandwidth and ripple is because of parasitics, but I do not know how to compensate these parasitics  during the design process.

My question is : Is there any methods that can compensate the parasitics of the filter during the the design process of the filter? Or
please tell me where I can find papers or book talking about it.

Thank you very much.

Thank you.

I am speaking of gm-C filters, is active filter.

Yes, I tried to reduce its influence, but because of the uncertaily of the parasitics and the limitation of the process, it is very difficult for me to fabricate a filter which have excellent performance.But when I see the
datasheet of some integrated bandpass filter (for example, IF filter in  receiver chip) which designed by some famous comany, its perfomance(bandwidth and ripple) is very good.

I am amazed at how can they fabricate filter with such good performance with the parasitics?

Hi, with circuit optimization techniques you can push back the changes of the real gmC filter compared to the idealized filter-prototype  :). Optimize the element values, i.e. gm and C. This way you should be able to get a near-ideal performance even with nonideal elements.
This is more difficult with op-amp based active filters because they really suffer from addional poles and zeroes. But also here there are some tricks, like adding a series R to the integration caps.

Bye Stephan

unaffected wrote on Oct 7^th, 2012, 7:37pm:


Does your foundry provide a rule deck for parasitic extraction? Then extract your layout and just re-simulate and check how the parasitics influence your characteristics.

buddypoor wrote on Oct 14^th, 2012, 3:31am:



I do agree with that -

Beyond the less than ideal performance of the
transconductor, you can impedance scale  the
device so that parasitics are a smaller amount
of the performance criteria

Parasitics remain constant, and frequency defining
elements become a large part of the equation.

Even that fails however if you have to size scale up
your transistors to deal with higher currents required
to do impedance scaling.