(from an email) 17-aug-2007

The input filter I used is not really different from what we find in amateur books and magazines. It consists of two capacitively coupled resonating LC cells, input and output are coupled through inductors; Generally we see capacitive coupling throughout the filter because capacitors are cheaper and are readily available in a large list of values, inductors are not so.

http://py2wm.qsl.br/SDR/40m_filter.html

As the detector used in the receiver is sensitive to odd multiples of the designed frequency (for instance, the 7 MHz receiver detects 21MHz with just some dBs of loss, this is often called oversampling), we need a bandpass filter (to protect the input from strong out of band signals that could lead to intermodulation) which has its skirts tending to a low-pass filter.

Coupling by inductors has this effect. On the contrary, capacitive coupling makes the overall filter shape degenerate into a high-pass filter.  By using inductors to control input and output coupling into the filter we have an added high frequency attenuation.

I even tried to transform the design into an elliptic one by placing a 30pF trimmer in parallel to one of the coupling inductors, introducing a parallel resonance giving bith to a notch in the frequency response on 21MHz:
 http://py2wm.qsl.br/SDR/imgs/40m_filter-notch21.gif
This was for study, it is not really needed in the receiver.

For the filter I employed ready made inductors with Q around 80, there is no need to use higher Q inductors. Those small SMD inductors available today have Q around 10~30 so they are too lossy and unsuitable for use in this filter. The inductors were placed at right angles in order to lessen magnetic coupling.

I used the Ladpac software from Wes Hayward, W7ZOI (comes with his book EMRFD), and ELSIE, a free software. With Elsie I did a Monte Carlo sweep, a survey as to filter performance considering the inductors 10% and capacitors 5% tolerances.

Measured performance completely agrees with simulation.