Please let me ask about this without showing any circuit, YET.

There is an active 4-poles Sallen Key lowpass filter. Cut off freq (-3 db) = 740 Hz. Both LM 358 opamps fed with +/- 9 V and duly filtered.

At the input, the signal from a network of 12 resistors driven by CMOS shif registers generating sine sequence of 24 voltage steps. Their output is referred to ground and so input to filter properly centered on 0V.

To my surprise, from 120 Hz to 500 Hz (upper limit of this design), filter output is remarkably smooth and looks OK, but going down from 120 to 10 Hz, the filter output is more and more jagged (one peak for every step at the input).

My question= is it any typical reason for this filter to behave like this? What would you check first in a case like this?

Just in case, please note that I tried two designs already: one with equal value resistors and the current one designed with Filterlab from Microchip.

I am far from my bench so posting values/circuits is not feasible by now.

My intention was to generate those steps in LT Spice to see what could be the reason but (see my post somewhere below) could not find how yet.

Do lowpass filters behave like this?

That sounds like the system is operating exactly as designed

The Fourier series for a square wave is sin(x) + 1/3*sin(3x) + 1/5*sin(5x) + ..... for all odd numbers. You sine wave is a stepped function, which is similar to a square wave insofar as the incremental steps of the sine wave function will have higher frequency content than the fundamental and your output will have jaggies. You could actually calculate the DFT of your signal mathematically if you really wanted to to get its Fourier series and figure out what harmonics are/aren't getting through your lowpass filter, but that's getting way to deep into the problem. If you double the output rate (samples/second) and add 1 bit depth (twice as many steps) you will keep the fundamental constant but smooth out the output sine wave.

Question is, do you want us to help you fix it? Increasing the output rate (number of steps per second) and bit resolution (number of steps your amplitude is divided into) will help you smooth it out while keeping the frequency constant.

Apologies if you already know this stuff, you likely do since you're working on something like this anyway.

As I mentioned on another forum, if the filter is a Butterworth type then it boosts frequencies near its cutoff frequency.

With a 10Hz input and only 24 steps then the first harmonic is only 230Hz which is well within the passband of the filter.

It needs more steps.

Four separated single pole filters have a loss of 12db at the cutoff frequency. A Butterworth filter has positive feedback at the cutoff frequency which boosts the level up so the loss is only 3dB. So the boost is 9dB.
The boost causes ringing at the cutoff frequency.

A true Butterworth has no peaking in the Bode plot. Any lowpass filter with abrupt cutoff will exhibit Gibbs phenomenon, which exhibits itself as ringing in the time domain.

Originally Posted by Roff

A true Butterworth has no peaking in the Bode plot.

A single RC network has a Q of 0.5 so it doesn't ring. A Butterworth filter has positive feedback which creates a Q of 0.707 and boosts the loss at the cutoff frequency to only 3dB.

I use a second-order Sallen and Key highpass filter with too much boost to make a bass boost circuit. The pot adjusts how much positive feedback there is. I set them for +10dB at 30hz.

Thanks to both for replying.

Speakerguy79

The first file is the circuit and the other four are pictures taken when output frequency was appr 77 Hz.

http://img374.imageshack.us/img374/5...circuitdk1.png

Network not loaded by filter
http://img514.imageshack.us/img514/3...networkvv4.jpg

Network loaded by filter
http://img404.imageshack.us/img404/4...networkob0.jpg


After filter, before RC passive highpass
http://img412.imageshack.us/img412/2...eforercvo2.jpg

After RC passive
http://img413.imageshack.us/img413/3928/afterrcdv2.jpg

I can see that I would end redesigning everything. I prefer not to increase the number of steps. Do I have other chances even if using a more complex filter?

Maxim make some simple to use switched-capacitor lowpass filter ICs that have up to 8 poles. You feed in a square-wave and they output a low distortion sine-wave.

I made a very low distortion sine-wave generator with a stepped sine-wave filtered by a switched capacitor filter.