LPF-HPF Box. Design help and a few questions.

[jwhmca]

Hi Guys,

So, I'm needing to build an LPF, HPF filter with seletable filters and slopes. A client is "sick of using plugins." And, it's something that would work great in one of my other projects...

So, as I read and research, these are the questions/thoughts that I have come up with.

1. Seems like for sure I will need to de-balance, then re-balance going out(?) As opposed to trying to find caps and coils EXACTLY the same for a balanced circuit?
2. Active vs passive. Client doesn't mind either way... "whatever sounds good."  See #3,4 for more consideration?
3. Selectable filters. This seems easy... just different value caps etc...
4. Selectable slopes. 6db - 12db, 18db seem straight forward... but we are thinking about 3db, 9db.  Anyway?
5. Butterworth? Bessel? others? etc?
6. Input/Output impedance? The LCR circuit will depend on whats plugged into it?
7. Other considerations I'm missing?

 

[joaquins]

When mixing will be always connected at line level, and in any case, as recording, you would insert it after the mic pre, so 10k for input, as low as you can go for output is a good start, I said as low, but without care much, around 100Ω is fine, for output of an opamp with some resistance is fine, you could go lower using some inductance in parallel for get it able to work with capacitive loads but still low output impedance.

If going active where you'll find a lot of flexibility but something more complex, yes, you want to make the input balanced, de-balance it and then make a balanced output. If going to passive and you wan't to keep good CMRR you can use RL filters in series, but this will do two bad things, first is insert losses, and second bad impedance behavior and will depend the frequency for first and last stage mainly on driving and loading impedance, and get higher impedance at output which isn't a nice thing.

If you are going for active filters you could get a continuos changing frequency. 3 or 9 dB/oct aren't easy to get, if you sum a 6dB/oct with a 12dB/oct you'll have one slope at the beginning (6 or 12 depending on what freqs you choose for each) and 6dB/oct at the end. Look for thrust filter for example for something similar, you can have it, is more complex than simple filter and still isn't pure 3dB/oct but it does something different than normal filters.

I'd probably go for Butterworth, but others may be possible depending on what you are looking for, even a variable Q filter may be useful for some cases to create effects or something.

You could use a filter already builded and tested by any big company and recreate it if you like it, a lot of work already done in the filters of an eq for example, just add input and output stage, probably line receiver and driver ICs.

JS

 

[jwhmca]

The purpose would be a "Junk dumper." removing 10hz-20hz... 20-40khz etc...

 

[Gold]

I'd ask the client what types of filters he uses most and start with that. It's hard to be a mind reader. There are of course lots of ways to do it. I'd suggest taking a look at "The Active Filter Cookbook". It has everything you want to know.

 

[Gold]

It might be a good approach to put a few Salen-Key first order filters in series. If you put the same frequency points on each one, by engaging more than one you increase the filter slope. There are issues of over or under damping when doing this. You might give each filter a different Q to help this.

 

[JohnRoberts]

For dumping junk you can make an adjustable 2 pole Sallen and Key or whatever with a dual pot.

Back in the '70s I sold a kit for my old kit company that was a fixed 2 pole HPF at 50 Hz, and fixed 2 pole LPF at 12kHz, In the kit magazine article I told the builders how to change the cut off frequency with different value resistors. IIRC that old kit sold for $15.  ;D

JR

 

[L´Andratté]

The Drawmer 201 key filter might be an idea, I find it pretty useful. The HP and LP could also be switched parallel/serial for variable bandwith bandstop/bandpass.
http://groupdiy.com/index.php?topic=44990.msg563199#msg563199

 

[zamproject]

you can also have a look at the studer x89 series filter module  with extensive HP-LP-BP-BR from 20 to 20K hz
(I mean the full module, not the filter in channel module but you can have a look at this one too, discret HP-LP 6dB/oct)
bad point for you, it's a fixed 24dB/oct (48 if you chain the two channels )
Look easy, just have to source a quad log pot  :-\

Zam

 

[jwhmca]

I'll be going with all switches. No pots.

**edit**

Not planning on having any variable amount of cut. Just switch in the frequency and slope and away you go..

 

[zamproject]

the quad pot is for frequency, you can use resistor network in a 4 row rotary switch

 

[abbey road d enfer]

1. Seems like for sure I will need to de-balance, then re-balance going out(?) As opposed to trying to find caps and coils EXACTLY the same for a balanced circuit?

Unless you have vowed to stay away from any kind of versatility, you'll have to use active stages, so adding inpurt and output balanced connections is a no-brainer.

2. Active vs passive. Client doesn't mind either way... "whatever sounds good."

For a design engineer, "sounding good" is never part of the design brief. There are requirements, slope, frequency, levels, noise... and when the circuit is built, it is measured, and listened to. If it does not sound good, the causes are analysed, the circuit is changed, and perhaps the design brief modified.

4. Selectable slopes. 6db - 12db, 18db seem straight forward... but we are thinking about 3db, 9db.  Anyway?

Intermediate-slope filters are built using shifted-frequency counteracting filters. Since there is an infinity of propositions, there's an infinity of solutions. Any such filter must be individually calculated. Indeed, in the analog domain, making them variable is complex and tedious.

5. Butterworth? Bessel? others? etc?

You should look at the definitions of all the different filters. Butterworth is maximally flat, Bessel is minimal overshoot/best transient response. Which one do you need? Anyway, using the Sallen & Key or the SVF structure, you can build almost any type of filter just by changing the values of the components.

 

[jwhmca]

I've been looking but can't seem to find an example of how to make a 9db per octave slope... anybody have a link or description?

What about 3db per octave. Is that even possible?

 

[abbey road d enfer]

I've been looking but can't seem to find an example of how to make a 9db per octave slope... anybody have a link or description?

What about 3db per octave. Is that even possible?

Yes.
http://sound.westhost.com/project11.htm
It's a low-pass filter. You can make a reciprocal high-pass by putting the multiple RC network at the input and a resistor in the feedback loop.
Once you have a 3dB/octave filter, you can combine it with a 6, 12, ...type in order to create 9dB/oct, 15dB/oct,...
I would say a 3dB/octave filter is quite useful for taming the aggressivity of some sources, but the type that is described here has its corner frequency very low (about 10Hz!); you must scale the components in order to set a more usable frequency.
As you can imagine, due to the number of interacting components, it's impossible to make an analogue variable version, but it's relatively simple to make a switchable version. Since the corner frequency depends only on the RC network, it is possible to make a 12-position switchable version with a common SP12T rotary.
A 3dB/oct hi-pass is not as useful IMO, except when inserted in the side-chain of a compressor (so called "thrust").

 

[Fenris]

The Neve 1073 and 1081 use third-order passive filters with 2 caps and 1 inductor. Probably the best sounding steep filter.

I'm working on something like the filters in the Undertone Audio console, which have adjustable resonance so you can add a bump at the transition frequency to put back some of the lost energy. I don't know what circuit they used, but I'm using a Sallen-Key circuit. The Q is tied to gain, so this would require a dual pot that reduces the input level as it increases Q.

 

[abbey road d enfer]

I'm working on something like the filters in the Undertone Audio console, which have adjustable resonance so you can add a bump at the transition frequency to put back some of the lost energy.

I did that in this commercial product
http://www.webencheres.com/Vente-aux-encheres-de-la-vichy-394/fiche-du-materiel-occasion/-Egaliseur-paramEtrique-scv-pfl-52-56792

I don't know what circuit they used, but I'm using a Sallen-Key circuit.

S&K is not the easiest implementation. I did it with SVF.

 

[jwhmca]

...you can combine it with a 6, 12, ...type in order to create 9dB/oct, 15dB/oct,...

In series or something?

So make the 3db/oct circuit then run it straight into the 9db or whatever?

 

[VictorQ]

The mildly cryptic terminology is that a 'pole' in the 'transfer function' contributes a slope of 6 dB per octave to the frequency response. Physical systems generally only have an integer number of poles, so the natural slopes are multiples of 6 dB per octave.

A 'zero' contributes with a rising slope and is in a vague sense the inverse of a 'pole'. So if you alternate between poles and zeros as you increase frequency, you can achieve a wiggly line around a 3 dB per octave slope. It's only an approximation, but with a large number of poles and zeros (ie. a large number of RC combinations) you can get close.

In digital, poles and zeros (not exactly the same things in a digital context, but close) are reasonably cheap, so you can get a good approximation more easily.

 

[jwhmca]

The mildly cryptic terminology is that a 'pole' in the 'transfer function' contributes a slope of 6 dB per octave to the frequency response. Physical systems generally only have an integer number of poles, so the natural slopes are multiples of 6 dB per octave.

A 'zero' contributes with a rising slope and is in a vague sense the inverse of a 'pole'. So if you alternate between poles and zeros as you increase frequency, you can achieve a wiggly line around a 3 dB per octave slope. It's only an approximation, but with a large number of poles and zeros (ie. a large number of RC combinations) you can get close.

In digital, poles and zeros (not exactly the same things in a digital context, but close) are reasonably cheap, so you can get a good approximation more easily.

Got it.

What is/does a "ZERO" look like? Is it a component or circuit that does not contribute?

I'm looking at the ESP circuit "Figure 1A - Alternative 3dB/ Octave Filter" that Abbey posted about... which is which in that?

 

[jwhmca]

On functionality and practical implementation:

It seems to me if I have frequency choices A- B - C - D - E etc. Slope choices 1 - 2 - 3 - 4 - 5 etc...

And, if I wanted to be able to select between them all... there are 25 combinations.... So, that is either 25 discreet circuits or fewer if I can come up with a fancy way of only switching in or out certain parts, duh...

Seems the easiest would be to have 5 frequency circuits and have the poles switched in or out to get the slopes... That would mean then; on the faceplate there would need to be a control for each filter frequency and then a master slope control for the channel(?) I'm a thinking right?

Can anyone think of a way to do it more efficiently?

 

[VictorQ]

What is/does a "ZERO" look like? Is it a component or circuit that does not contribute?

I'm looking at the ESP circuit "Figure 1A - Alternative 3dB/ Octave Filter" that Abbey posted about... which is which in that?

A zero usually makes the frequency response tilt upwards for rising frequency.
A high pass has a zero at zero frequency, so the response starts at zero and slopes upward for low frequencies. A pole then breaks the response down to flat for higher frequencies.
C4 and R4 form a high pass, with the pole at a low frequency given by their product.


Assuming an ideal opamp the zeros are exactly at the frequencies corresponding to the feedback series RCs (f=1/(2*pi*R*C) or more precisely s=-1/(R*C)).

The poles are more involved, but for these values they are well approximated by the frequencies corresponding to the R*C products of a given resistor and the 'next' capacitor when rising in frequency. So R8*C1, R5*C2, R6*C3.

That's really confusing, I know. The point is clearer with some plots and drawings and stuff. Maybe I will make some tomorrow...
Edit: phew.. I don't think this is right... Don't drink and derive. I'll correct it after work.
Edit 2: done. I switched poles and zeros. Now it should make sense.