Thanks Neil and Joe. I need to clarify. I think I need something that plots the frequency response, given the component values, rather than a component value calculator. The schematic does not give the intended filter parameters, just the component values, and these seem strange. From memory, pairs of Rs and Cs are supposed to have identical values, but not in this schematic, so I'm wondering why… Sorry to be vague, but electronics isn't my field, and I'm, perhaps rather cheekily, only posting because the original post piqued my interest.

Last time I looked at Fourier analysis adding in-phase odd harmonics of reducing amplitude to a sine wave approximated a square wave.

I had a quick go with the simulator; compared to some it is easy to use. I didn't register or login, but I didn't try and save anything. I have an AD login, so it's possible my computer remembered it anyway.

Warning: Skip this bit if maths isn't your thing!

It's interesting the way the simulator switches between filter characteristics without so much as a by your leave. Like all simulators they can be useful if one understands the theory, but if one doesn't then they can lead one into a cul de sac. Via Wikipedia I've just read the original paper by Butterworth from 1930. Of course that was long before opamps and RC filters, so everything is valves and L, C and R. The intent of the original paper was to design filters that were maximally flat in the passband without compromising the roll off. In other words the amplitude characteristic is monotonic and flat in the passband, which it appears previous filters were not. It's interesting to note that the original article only refers to second order sections. The concept of single real poles isn't mentioned, so the original Butterworth filters could only be even order.

In order to understand filters an appreciation of the complex s-plane, and poles and zeros is useful. On the s-plane the x-axis is sigma, a measure of how a signal decays and the y-axis is j times omega, complex frequency. A pole is a point of inifinite value and a zero is just that, zero. If a rubber sheet is stretched over the poles, and nailed down at the zeros, then a section along the y-axis will give the frequency response of the filter. There are some constraints, for stability all poles must be in the lefthand half, ie, sigma is negative. Poles always come in complex conjugate pairs. A special case is when omega is zero in which case the two poles are coincident, and real, on the x-axis.

The filter simulator starts off with Bessel filters, which are maximally flat in group delay, ie, signal distortion in time is minimised. This is at the expense of roll off. Pole position is determined from Bessel functions, hence the name. As more roll off is required the program switches to Butterworth which as stated is maximally flat in amplitude in the passband. So no amplitude distortion and faster roll off, but at the expense of non-linear group delay giving time distortion of the signal. A Butterworth filter only has poles, and those poles are equi-spaced, and lie on, a circle. As yet more roll off is required the filter characteristic becomes Chebyshev, which allows ripple in the passband in return for faster roll off while still being monotonic in the stop band. Technically these are type 1 Chebyshev fillters. Type 2 Chebyshev filters are flat in the pass band but have ripple in the stop band – never seen them used. A type 1 Chebyshev filter is also pole only, and the poles lie on an ellipse. In the limit as the ellipse becomes a circle the passband ripple reduces to zero and the filter becomes Butterworth. Although not used in the simulator the fastest roll off is with an elliptic filter that has poles on an ellipse giving ripple in the pass band and zeros on the y-axis giving ripple in the stop band.

It took me a while to find the design where the capacitor values are uniform; in the multi-feedback circuit. The Sallen-Key arrangement is the classic filter circuit mentioned in all text books. It's ok but can have problems implementing high Q sections. There are better, but more complex, circuits available such as the biquad.

The tolerance feature of the simulator was useful, and very quick. Not sure how it worked, seemed a bit fast for a proper Monte Carlo anaylsis?

I had to look up the Linkwitz circuit. Looks a bit odd with T-sections in the input and feedback paths. It's not high enough on my priority list to do a proper analysis. In the past I've used Tina from Texas Instruments as a simple circuit simulator. I expect it is based on Spice internally but has a graphical input interface. Ah, I see Kiwi Bloke is using Linux, in which case he's probably out of luck for a simple simulator.

I'm with Bob Pease on circuit simulators. They have their uses, but are only as good as the component models. Ultimately prototyping trumps the theory and simulation.

Andrew

Elsie also runs under Wine on Linux.

Playing with this wideband Software Defined Radio, I found it overloaded at night due to very loud MW broadcast signals and powerful Long Wave time standard signals. I used Elsie to design a filter to attenuate the whole MW band and to notch out TDF (France) on 162kHz and BBC R4 on 198kHz, while leaving everything else open.

Unfortunately, the filter only revealed the next problem! That's the appalling radio racket created by internet users with ADSL connections, plus the generally high level of horrible noises from unsuppressed domestic equipment.

Not quite on topic because Elsie does Inductor-Capacitor rather than Resistor-capacitor filters, but some may be interested. These days we're spoilt for choice!

Dave

Thanks Russell, I'll look up QUCS. Thanks Andrew for your input – I wish I understood the theory…

Having somewhat hijacked this thread, I've felt obliged to try to follow some of the suggestions – make some effort to solve my own problem. Crikey! The maths is frightening, but not as opaque as the underlying concepts hinted at by Andrew J. OK, I think I should give up, my poor aged brain won't cope with going off in such a complex new direction. Current mental activity is centred on wondering how to manipulate a CertiFlat welding table kit: not disturbing the fit of the components whilst invering it for final TIG welding. It will weigh >100kg.

Before I give up on this filter thing entirely, however, the problem is (if I remember correctly) that the circuit contains various Sallen-Key-like filters, built around op-amps. The usual implementation of these seems to use two same-value Rs and two same-value Cs. It looks as though this may be because it simplifies the maths. The schematic uses different values for the Rs, and I wondered why. I can't remember whether the C values were identical, and now I can't remember where the wretched schematic is. The perils of working from memory…