EMRFD Message Archive 6932

Message Date From Subject
6932 2011-11-28 21:28:13 ehydra filter design with standard n-multiple parts
Hi!

I'm looking for resources where filter design with the constraint that
parts are unified is shown. I mean something like the Hagerworth LPF.
This filter is made of just two part variants.

A picture is worth a thousand words:
http://www.hagtech.com/theory.html

Any suggestions? Preferable for data-communication so constant
group-delay in pass-band is important.

Thanks!
- Henry

--
ehydra.dyndns.info
6935 2011-11-30 00:44:17 Kerry Re: filter design with standard n-multiple parts
Lokks somewhat like a constant-K design; ELSIE can do these.

You can tinker with the component values to see what happens with your example filter.

The group delay is reasonable but not perfectly-flat across the passband.
6939 2011-12-03 08:15:02 ehydra Re: filter design with standard n-multiple parts
Thanks Kerry!

Well, better than nothing but ELSIE uses {C, 2*C, L} only. The
Hagerworth filter uses additional 2*L (or 1/2*L).

So I keep searching.

- Henry


Kerry schrieb:
> Lokks somewhat like a constant-K design; ELSIE can do these.
>
> You can tinker with the component values to see what happens with your example filter.
>
> The group delay is reasonable but not perfectly-flat across the passband.

--
ehydra.dyndns.info
6942 2011-12-03 12:31:43 Kerry Re: filter design with standard n-multiple parts
G'day Henry.

You can use the Edit function in Elsie to change a design that you have done.

So design a capacitor-input constant-k filter with Elsie then change the C & L values with Edit to the 2/0.5 values used in the Hagerworth filter.

Kerry.
6943 2011-12-03 19:34:23 ehydra Re: filter design with standard n-multiple parts
Hm Kerry -

For what is this good for?
I found the Hagerworth filter and the site provides the formulas. I can
use LTspice to plot the filter curves and insert parasitic elements.

But it lacks the theory behind. I want to learn...

- Henry


Kerry schrieb:
> G'day Henry.
>
> You can use the Edit function in Elsie to change a design that you have done.
>
> So design a capacitor-input constant-k filter with Elsie then change the C & L values with Edit to the 2/0.5 values used in the Hagerworth filter.


--
ehydra.dyndns.info
6953 2011-12-05 16:27:56 wb6tpu Re: filter design with standard n-multiple parts
Henry-

I think the "Hagerworth" filter is really just a relabeled constant-K type filter. These filters are based on the old image method of design developed at AT&T back in the early '20s.

Elsie can indeed design them. Take a look here (and lots of other places) for some theory
6954 2011-12-05 16:57:08 wb6tpu Re: filter design with standard n-multiple parts
I may have posted my earlier response without looking closely at the "Hagerworth" filter.

It is related to a constant-K filter but is decidely different.

Here is a snippet from the Javascript code that runs the little design applet on the hagtech.com website. It calculates the L and C values for their Hagerworth filter.

-Ray WB6TPU

function HagerWorth()
{
// get values from form

var omega = 6.95 * document.hagerworth.frequency.value;
var impedance = document.hagerworth.resistance.value;

if(document.hagerworth.band[0].checked)
{
omega = omega * 1e3;
}
else
{
omega = omega * 1e6;
}

// frequency and impedance scaling

var C = 1 / (omega * impedance);
var L = 1 * (impedance / omega);

// display

document.hagerworth.c.value = Round(C);
document.hagerworth.l.value = Round(L);


6959 2011-12-06 06:49:39 ehydra Re: filter design with standard n-multiple parts
Yes, already seen and used. Note that the factor 6.95 is close to omega
but not exactly. I too think it is a derivative of constant-K. But I'm
not a filter expert and lost in the myriad of filter names on the Net.

The complication is what is required for group delay in absolute terms.
All is relative :-)
Time to let SPICE transient simulate a lengthy PRBS ...

- Henry


wb6tpu schrieb:
> I may have posted my earlier response without looking closely at the "Hagerworth" filter.
>
> It is related to a constant-K filter but is decidely different.
>
> Here is a snippet from the Javascript code that runs the little design applet on the hagtech.com website. It calculates the L and C values for their Hagerworth filter.
>
> -Ray WB6TPU
>
> function HagerWorth()
> {
> // get values from form
>
> var omega = 6.95 * document.hagerworth.frequency.value;

--
ehydra.dyndns.info
6964 2011-12-06 10:00:24 wb6tpu Re: filter design with standard n-multiple parts
Henry-

Yes, the answer always boils down to "it depends"...

As a matter of curiosity I designed a 10Mhz (-3db) Hagerworth filter (using the online applet) and a 10MHz (-3dB) Butterworth filter (using Elsie). Both were 7th order with 50 ohm in/out Z.

Butterworth:
-3dB freq: 9.93 MHz
peak group delay: 124.4nsec @ 9.5 MHz
(52nsec delay peaking)

Hagerworth:
-1dB freq: 10 MHz
-3dB freq: 16.47MHz
peak group delay: 95.26nsec @ 16.6 MHz
(45psec delay peaking)

Though the Hagerworth applet asks you to enter the desired -3dB freq it appears that it is creating a filter (at least in this test case) with a -1dB loss at that frequency.

Be that as it may, as you mentioned, you need to understand how much group delay you can stand in your application. 95nsec or 124nsec GD could be just fine, or it might be totally unacceptable.

Contrast those GD figures with an equivalent Bessel or Gaussian 7th order LPF. The roll-off of course is way slower, but there is no delay peaking (i.e. it is a constant 42nsec for a Bessel and a constant 47nsec for a Gaussian type).

Engineering tradeoffs are the name of the game....

-Ray WB6TPU




6969 2011-12-06 17:44:38 ehydra Re: filter design with standard n-multiple parts
wb6tpu schrieb:
Though the Hagerworth applet asks you to enter the desired -3dB freq
> it appears that it is creating a filter (at least in this test case)
> with a -1dB loss at that frequency.

Maybe he confusion of power, voltage or noise bandwidth? dB is always
power and at given Z0, but often misused for relative voltage.


>
> Be that as it may, as you mentioned, you need to understand how much
> group delay you can stand in your application. 95nsec or 124nsec GD
> could be just fine, or it might be totally unacceptable.
>

We're looking for constant _differential_ group delay. The common case.


> Contrast those GD figures with an equivalent Bessel or Gaussian 7th
> order LPF. The roll-off of course is way slower, but there is no
> delay peaking (i.e. it is a constant 42nsec for a Bessel and a
> constant 47nsec for a Gaussian type).
>
> Engineering tradeoffs are the name of the game....

No! There is always a ultimate S/N for a given complexity realizable.
The problem is how to find the relationship between wanted signal,
interferer signals, noise, linearity, price, etc. Rules of thumb apply ;-)


- Henry


--
ehydra.dyndns.info