brick wall low pass filter

[vacosound]

I'm interested in building a brick wall low pass filter. The objective would be to take a frequency of say 8k and stop all frequencies above from passing. From what I have read a complete brick wall filter is not possible? But there are designs with very steep rolloffs that can be achieved. A Butterworth type filter with a 24dB rolloff would not be ideal for instance. Any advice of reading materials or ideas? Thanks.

[2N1305]

I think this is where digital technology would really shine. I've heard that DSP-based signal processign can do such things, because rather than acting on the signal directly, in an analog way, it calculates every component of the signal, and can manipulate it very rapidly in different ways. I am by no means able to help you with that as I've never done DSP stuff. But I am sure University students and other digital "gurus"can help you.
Anyone, jump in, feel free!

Also, if you think that 24dB per decade is not enough, (well you said not "ideal"), yes, it's true that it is not brickwall, but I'm wondering if you cascade two of these filters, you might get a 40dB per decade. Are you handy with electronics?

Now look at me, you made me want to build one!

2N

[Paul Frindle]

Originally Posted by vacosound

I'm interested in building a brick wall low pass filter. The objective would be to take a frequency of say 8k and stop all frequencies above from passing. From what I have read a complete brick wall filter is not possible? But there are designs with very steep rolloffs that can be achieved. A Butterworth type filter with a 24dB rolloff would not be ideal for instance. Any advice of reading materials or ideas? Thanks.

The trouble with such a filter is that the very steep roll-off causes very considerable ringing which can be set off by program and continues long after the event - and even before the event as well if it is a linear phase filter.

For DAC reconstruction filters the fact that this occurs around 20KHz or above means you don't hear it directly - so it's ok. But at 8KHz you will hear it and it will sound bad generally.

From the point of view of the perceived sound as we actually hear it, roll offs beyond around 30dB/octave do not result in any greater perception of freq blocking above the set passband. So the extra ringing from ever steeper filters is not woth it - and in the end becomes detrimental only.

[2N1305]

Hi again,
I've just did some serious reading on the subject of analog op-amp filters, to brush up my knowledge on the subject. I'm sure what I can tell you Jim Williams would be able to tell you with more clarity and certitude, but since I'm here first, then here I go.

What you'd need is a Sallen-Key type filter. It exhibits a 40dB/decade roloff curve, which is approximately 12dB per octave. Now you didn't specify the response type you wanted per decade, or per Octave? 24dB per Octave is definitely possible (as I'll show later) but requires a bigger circuit (about 16 components including opamps). Not a big deal if you've built stuff before. So in your case, that 8kHz signal would actually be 12dB lower at 16 kHz. So what are you after: better than 24dB per Octave or better than 24dB per decade? The two have very different response curves...

If you need an even more steep response, consider the following: a three-pole filter consisting of a Sallen-Key followed by a single-pole lowpass filter.

Still not steep enough?

How about another two-pole stage following the first two-pole stage, that's 80dB per decade! Roughly 24dB per Octave...
I can scan some schematics if you'd like.
I don't want to post them here because of copyright issues from the book my information came from. The book is entitlted "Electronic Devices" fourth edition, by Thomas L Floyd. A very well-written book.

[Jim Williams]

Find an old copy of Don Lancaster's Opamp Filter Cookbook. All the topologies are there.

If you're going for a fixed frequency, a single 12db/octave or 2 cascaded sections will give 24 db/octave slopes. Easy enough with either MFB or sallen key designs. Sallen key designs allow the use of transconductance or current feedback opamps like the new National LME49813 since there are no caps or integrator functions to deal with.

What's more important is the filter type. Maximum flat butterworth slopes should be avoided due to excess group delay and non-linear phase response. Bessel or linear phase slopes are prefered to retain harmonic balance. The trade-off is the slopes are less steep at the transition band. This sometimes will require an additional pole to retain the same amount of roll off as a butterworth slope. 18 db/octave filters can be assembled with a single opamp as the third pole is done passive in front of the opamp.

BurrBrown also has a nice filter pro design/calculator on line. That allows you to skip a page or two of math and see the results on screen.

Now if you need adjustable frequency roll-offs, it can get complicated. Tuning resistors can be subbed for pots, but tolerance gets washed out. A dual reverse log pot is needed to tune a 12 db octave 2 pole filter. A 3 section is needed for 18db/oct. and a 4 section is needed for a 4 pole 24 db/octave slope. Good luck getting a 4 section pot to track accurately.

Years ago in the 1980's I built voltage controlled filters using a SSM filter chip set. It allowed DC control of frequency. It had 12 and 24 db/octave slopes and the various filter functions could be selected. I added a circuit to sum both outputs so you could adjust the slope from 12 to 24 db/octave or anything in between.

Jim Williams
Audio Upgrades

[Gabriel Sousa]

i would love to buid one too

[JohnRoberts]

Originally Posted by vacosound

I'm interested in building a brick wall low pass filter. The objective would be to take a frequency of say 8k and stop all frequencies above from passing. From what I have read a complete brick wall filter is not possible? But there are designs with very steep rolloffs that can be achieved. A Butterworth type filter with a 24dB rolloff would not be ideal for instance. Any advice of reading materials or ideas? Thanks.

Without knowing the application or your constraints it is difficult to give the optimal solution.

Looking at conventional topologies you can get steeper roll offs by cascading more poles (-6dB/oct/pole). In addition to adding poles there are alternate topologies like Chebychev where you stagger the pole frequency and Q of the individual filter pairs to get a steeper rolloff immediately above cutoff, but you trade some increased ripple in the passband (I used Chebychev quite a bit to make steep anti-alias filters for old analog delay lines). There is also a topology (I haven't used) that looks a little like a notch filter at tuning, for very steep dip, but ultimately past the notch the response recovers somewhat.

So long story short with analog filters you can trade some pass band ripple and/or smooth phase response for steeper roll off just beyond tuning but they all ultimately converge back to 6db?oct/pole.

Operating in the digital domain you could perhaps execute a steeper filter but that is a trade off in itself just for the complexity involved. Simplest way there may be to use a codec IC designed for cellphone use that you can dial back the sample rate for the passband you want. Their built in digital anti-alias filter will be quite steep.

JR

[Gabriel Sousa]

mine its for analog, to connect a microphone :D

[dcollins]

Originally Posted by JohnRoberts

Looking at conventional topologies you can get steeper roll offs by cascading more poles (-6dB/oct/pole). In addition to adding poles there are alternate topologies like Chebychev where you stagger the pole frequency and Q of the individual filter pairs to get a steeper rolloff immediately above cutoff, but you trade some increased ripple in the passband (I used Chebychev quite a bit to make steep anti-alias filters for old analog delay lines). There is also a topology (I haven't used) that looks a little like a notch filter at tuning, for very steep dip, but ultimately past the notch the response recovers somewhat.

In the olden days PCM LPF filters used Cauer or Elliptical structures. Similar to Chebyshev but the passband and stopband ripple are the same.

They used a minimum of 9 poles, some were 11. The designs I saw were arrived at by optimization (where the magnitude response [the answer] is given to the SPICE program and it iterates the values to arrive at the closet approximation of the response).

They rang and overshot like crazy, the component sensitivity was high, the channel-to-channel matching was poor, the noise and distortion was high, etc............

Then came oversampling.

If anyone is really interested I'm sure you could find some old Apogee or Murata filters to play with.


DC