Phase Self-Correcting LP Filter

This sounds like a job for the PLL. If you have a signal you can lock it onto, they do this sort of thing easily.

If you want no phase shift in a linear filter, you need a band pass filter. You can contrive a low pass filter to have exactly 180, 360 or some multiple degrees of phase shift at your working frequency and call this no phase shift if you want but other than that the nature of the universe prevents you from making a low pass with no phase lag.

So, if I take one bandpass filter at 60 Hz, then connect another bandpass filter at, say, 55 Hz, in parallel, then another at 50, 45,40 and so on, so forth to zero Hz... this will make a lowpass filter with no phase shift :)))))

VLV

Actually, no it won't because you left out the bandpass filter at

The components get pretty big for the one at zero Hertz!

An ideal lowpass filter passes everything, unaltered, within its passband and nothing outside. That can be described mathematically but is physically impossible, because it's non-causal: its impulse response has outputs before the input, which means it predicts the future.

You can make an approximation to an ideal lowpass, but you have to add time delay to keep it causal. The better the approximation, the more delay you have to add, whether the implementation is analog or digital.

So you can make a lowpass filter whose phase changes little with frequency *after* you allow for the time delay. Even than it's a nuisance to do analog.

What's the application?

In addition to conservation of energy, our universe seems to have a law that prevents predicting the future. Both laws can be handy in short-cutting a lot of electronic analysis.

John

You've gotten the responses that explain why you can't have a zero phase delay filter in this causal world of ours.

Look for "Bessel" filter, which gives a pretty good approximation to a constant group delay. It won't be _no_ delay, which you can't get, but it'll be as close to _constant_ as you can get with a mininum-phase filter.

If you really build one be careful with component tolerances -- IIRC from the last time I considered using one (a long long time ago) the group delay vs. frequency relationship is fairly sensitive to component values. So you can't just take the nominal filter and run with it -- you have to be good and do your Monte Carlo analysis of the filter.

I thought this was interesting:

Williams' filter book has some interesting designs. Some are "transitional", Gaussian in the passband but rolling off faster than a Gaussian after 6 or 12 dB down, similar in concept to the LTC things. We use things like this in our waveform generators, where we want to lowpass a DAC output without making it ring a lot. He also has filters that have a Chenbychev-like equiripple pattern, but in phase, not amplitude.

John

Depends on the Q.

I inadvertently tried that once--in my first engineering job, when I was

I took the classical all-pass phase shift filter trick,

and convinced myself by doing some algebra that changing the capacitor for an inductor would change the sign of the phase shift, which would give me a bunch more loop bandwidth.

It failed to work, which I didn't understand until my boss, a very smart guy called Joe Fikart, pointed out to me that I'd made a math mistake--the phase shift has the opposite sign, but also goes the opposite way with frequency, so that the group delay d(phi)/d(omega) is always positive.

You can't go back in time, even if you are a young enthusiast. ;)

Cheers

Phil Hobbs

It is analog output direct from an Earth micropulsation sensor coil, all below 50Hz.

What is the best I can do to preserve the real time signal characterisitics with minimum phase delay and distortion?

Glenn Kenroy

Could you please provide a brief description of the design approach you would take using this technique (PLL)?

The object is minimum delay and distortion in 50/60Hz LP for real time analog data from an Earth micropulsation sensing coil.

Glenn Kenroy

You may want to consider a different sort of magnetometer than a coil.

Your biggest noise sources are the mains and the 3rd harmonic. I will assume that the mains are 60Hz. If you are in a place with

The 60Hz and 180Hz are fairly constant in most locations. They will be on the order of 100nT in amplitude. Your micropulsations will be on the order of 1nT. You need a filter that takes the 60Hz down by about 60dB and passes perhaps 10Hz with little distortion.

The trick is to make a very tight notch at the mains frequencies and use a normal low pass for the general junk.

The first step is to make a PLL that locks onto the 60Hz. You want the VCO in the PLL to be running at many times the 60Hz frequency. I am going to suggest 7200 times, but faster is likely better. 7200 times just makes the explanation easier.

Important frequencies:

7200 / 15 = 60*8

I will assume that you have the PLL locked to the 60Hz.

You will be making the same circuit 3 times. It uses the CD4051 The 8 times the frequency goes to a counter that makes the CD4051 scan through a group of capacitors.

Each of the 8 outputs of the CD4051 connects to one end of a capacitor. The other end of the capacitor is grounded.

If the common point is fed with a resistor. This makes a circuit that will charge up the capacitors until they match the 60Hz input waveform.

Or even a comb filter at 10 Hz, that will take out both 50 and 60 Hz (depending on location) and their harmonics.

The PLL would solve the network frequency drift during the day, which can vary more than 1000 ppm, depending on the load.

This appears to be the most promising response so far to the elusive zero-delay issue.

What might it take for you to draw this up as a conceptual circuit (untested OK) to get me started on the right track?

Glenn Kenroy