Passive "Notch-Pass" Filter 2

[logbook]

Is there such a thing as a "notch-pass" version of the Twin-T notch filter? I want to make a passive narrow band-pass filter (notch-pass) using only R’s and C’s for audio frequency use.

Input will be from a low impedance source and output will drive a high impedance, so load/sourcing are not a problem. Also power levels are not a problem.

 

[cbarn24050]

Hi, you can make a passive RC band pass filter, but it wont be very good. Much better to use an active filter.

 

[ScottyUK]

I agree with cbarn, use an active filter.

It is possible to make a good passive filter, but you will need very close tolerance (read: expensive) components, and the wound components will be 'specials'. It is an interesting exercise if you have time and are doing it to learn, but for a real-world design where cost and size and design time all need to be minimised passive filters are not practical.

What's your budget and performance requirement? You can get exceptional performance from a DSP-based digital filter, but they are more complex and costly than an op-amp based type, and designing such a filter is not a trivial task.

 

[logbook]

The requirement is to filter a signal generator output at 1kHz to produce harmonics below -140dBc.

Active filters are out.

Close tolerance components are not a problem.

DSP filters are not relevant.

Inductors would need to be air-wound to avoid non-linear cores. I don’t think an inductor at this frequency would be feasible at the kilo-ohm impedance levels that would be required.

 

[cbarn24050]

Hi, you will never achieve that with just R C. You might get near with a crystal filter.

 

[logbook]

A good generator will produce –120dBC. I only need another 20dB-30dB.

 

[cbarn24050]

Hi, so now your saying you want a RC bandpass filter which is down 20db at 2kHz?

 

[logbook]

Relative to the fundamental, yes. Obviously the more the better. A simple RC bandpass consisting of a series RC feeding a parallel RC, both R’s equal and both C’s equal, gives only 1dB separation between the fundamental and the second harmonic. That was why I was wondering about a “backwards” Twin-T, if such a thing exists. I don’t see that you can “design” something like a twin-T. You can probably just hit upon it and see that it works. In fact I have no idea who invented the twin T or when.

Anyone know?

Obviously I could make an RC low pass filter rather than a notch but the discrimination between the fundamental and second harmonic would not be that great. Also a notch-pass gives the lowest possible noise performance.

 

[IRstuff]

Having read all these posts, it seems to me that you've eliminated all the possible ways of realistically achieving your requirements.

This suggests that you need to re-visit your constraints and determine if there is sufficient basis for eliminating the only approaches that can realistically solve your problem.

TTFN

 

[nbucska]

Is this a school project? You could make an LC filter
with iron core inductors . With the low power level the linearity shouldn't be a problem for a single fix frequency.


<nbucska@pcperipherals.com>

 

[dspDad]

A notch **reject** filter with 40 dB or so built as a twin-T is a piece of cake. They work because one side of the T causes phase lag and the other phase lead. The two outputs are added together and at some frequency they exactly cancel out. Because the signals cancel, the depth of the notch is close to zero, which is pretty impressive on a log scale. Without some positive feedback, the input impedance is low, and the circuit can only be driven by a similarly low output driver, and the output impedance is high. The width of the notch is also extremely wide. None of these are nice charactersitics. Usually you want a high input and low output impedance, and a narrow notch.

A notch **pass** filter can be built, but the output is only slightly peaked, because doubling two signals gives a 3 dB improvement, whereas subtracting them gives an infinite dB improvement. The pass version doesn't have a similar frequency where everything cancels except the frequency of interest.

As noted above, you could build a passive resonant L-C circuit that would provide reasonable functionality, but without any active devices would still have input and output impedance problems, and the amount of peaking would be limited by the drive and load characteristics.

If your goal is -140 dB, you need to consider the non-linearity of the capacitors and/or inductors. The design issues as you go past 60-100 dB in any area become those of managing second order effects. I suspect a good op-amp would end up being one of the more predictable components in a design with this many constraints requesting this much performance.

DspDad

 

[logbook]

Thanks for the useful insight, dspDad.

The non-linearity of capacitors at these levels is quite well documented in a recent series of articles by Cyril Bateman in Electronics World. Unfortunately standard 50ohm BNC attenuators are also non-linear at these levels, as I discovered today. Some bad ones are at the level of HD3= -100dBc, and these are expensive Tektronix attenuators.

 

[cbarn24050]

Hi, you're begining to sound like one of those audio nuts, let's hope you're not.

 

[logbook]

No, I design high resolution acquisition systems: scopes, recorders, etc.

 

[ScottyUK]

logbook,

Glad to hear that you don't have 'golden ears'.

I knew I had seen a good example of a multi-stage passive AF bandpass filter somewhere. I suggest you look at Fig. 5.3 on page 265 of 'Art of Electronics' 2nd Edition. The authors borrowed it from IEEE Journal of Solid State Circuits, Vol SC-5, No. 3 (1970).

The circuit is fully passive and, for a passive filter, gives incredibly steep attenuation around the passband. It looks like a pig of a thing to design from first principles, unless you happen to like complex two-port networks - if you have PSPICE or similar, that should make things easier. Have fun.


Scotty.

ps. How's my grammar?

 

[cbarn24050]

Hi logbook, sorry i didn't think you were a grammer nut as your op has two grammer errors.

 

[ScottyUK]

Hey cbarn,

You need to buy a spellchecker as well as grammar checker!

 

[cbarn24050]

I was cleaning my keyboard, I haven't got all the keys back in the right places yet.

 

[logbook]

Scotty,
Thanks for the suggestion. It is a bit of an overkill for my needs, given that I only need 20-30dB extra attenuation. And in many respects it would be counter-productive since more parts means more worries about non-linearities. I have worked out how I can make an LC filter using just one air-wound 1mH inductor, and it simulates as doing the job. This seems to be the best way to give reasonable confidence of the overall linearity.

Thanks to the many respondents for discussing this problem.