Hi,

I'm in the process of designing a Chebyshev bandpass filter (gm-C) with a passband ripple of 0.1dB.

It's however very difficult to come with the device sizes that make sure that the (typical) transistor-level simulation
has the same gain response as e.g. the MATLAB toolbox generated filter gain response. To this end, I'm now resorting to optimisation  :)

Anyway, I was just considering that - in general- it's very important to be on that curve? Or is close enough good enough?  :)

Kind Regards

Peter

I agree, so long as you can meet the system requirements (ie inband ripple & gain, groupdelay, signal handling, noise, out-of-band rejection, power, area etc.) then it'll be fine.

I have designed a few gmC filters for wireless transceivers in the past, and have always envitably had to tweak gm values and C values to "optimise" (mostly to correctly centre the response and "square up" the shoulders) when using real kit devices to get nearer the ideal mathematical response.

as you are using gmC, I guess you have some tuning mechanism to overcome PVT spread.