Site description

A blog for electronics professionals, amateurs, hackers, and anyone interested in the world of electronics.

2 September 2010

Filter Design (or, A Man With Two Watches Never Knows The Time)

There is an old saying; "A man with two watches never knows the time". Because they will never agree with each other. I feel like that today.

Theoretical 7-Pole Elliptic Low Pass Filter design, Fpass=450MHz, Fcutoff=500MHz

I have been designing a Low Pass Filter for a DDS. In order to get as sharp a cut-off as possible, I really need to use an elliptic (otherwise known as a Cauer) filter topology. There is always a tradeoff in electronics design, and the elliptic has considerable ripple in the pass-band, and relatively high peaks in the stop-band, due to the use of series capacitors. The ripple is fairly easy to remove with calibration. Even so, 4.6dB ripple for a 60dB stop-band attenuation is getting a bit 'iffy'.

However, when I come to model the filter with LTSpice, or the Altium DXP integrated Spice modeller, it seems to degrade the stop-band to around 40dB. Oddly, changing the component Q seems to make little difference.I wonder what it is doing differently to the simulators built into the filter design software.

Moving from the cheap (free) 'n' cheerful Elsie to Nuhertz Filter Solutions generates a very similar set of component values (as you would expect). It too claims 60dB stop band attenuation. I wonder what I am missing. Maybe Spice has a configuration option that I have missed.

*Scratches chin*

I'm not getting enough stop-band depth, so I may have to move up from a 7-pole filter to a 9-pole, or use a hybrid filter topology that gives me a reasonably fast cutoff with deeper stop-band attenuation.

In theory, with a limit of 450MHz, the DDS shouldn't be producing spurs in the 500-550MHz region anyway (for a clock of Fc=1GHz) so if I move the stop-band to start at 550MHz rather than 500MHz, that relaxes the cutoff speed constraint and so improves stop-band depth and ripple. Will 30dB cutoff at Fc/2 be enough though.

Or I could do the sensible thing and say "yeah that's fine".Of course, if I did that, it would come back to bite me...

It strikes me that all this filter software looks amazingly old-fashioned. Like it was originally written for Windows 3.1. I'm sure there must be something better for Mathcad. If that proves to be expensive, I might have a go at writing my own.


  1. I guess you have solved the problem by now. for what its worth: passband ripple and stopband attenuation are linked. Yes, many filter design programmes are from the DOS days before MS Windows came to life. Check the links at AWR have a good filter design wizard.
    Good luck

  2. Re-reading your post I wonder if the problem is caused by mismatch? The filter response depends upon having the correct source and load impedances.

  3. Thanks Dieter. I never did track down the bug, the source and loads were a standard 50R for each model. The AWR filter design wizard is the software they acquired when they bought out Nuhertz - which is the modeller I currently use!

    The final design I came up with was 9 pole and rather tricky to fabricate, but it does seem to be doing the job pretty well.