You're correct, and I apologize. In fact, the example I gave of a
"quarter wave" filter contradicts the statement about the attenuation. I
was thinking of a Butterworth when I wrote it, but as you point out and
as my own example shows, there are many other types for which the
statement is wrong.

I apologize for the error. Thanks for the correction.

Roy Lewallen, W7EL

Peter O. Brackett wrote:

Roy:
[snip]

The last sentence should read:

A lossless lowpass filter has zero attenuation only at DC.

:
:

Roy Lewallen, W7EL


Ummmm... no that statement is only true for one type of approximation
polynomial.

. . .

Roy:
[snip]
The last sentence should read:

A lossless lowpass filter has zero attenuation only at DC.
:
:
Roy Lewallen, W7EL

Ummmm... no that statement is only true for one type of approximation
polynomial.

A lossless low pass filter has zero attenuation at its' reflection
coefficient zeros.

If it is a maximally flat low pass. a.k.a. Butterworth. then all of the
reflection zeros
are located at DC, but for any other type, e.g. Chebychev, Cauer/Darlington,
General Parameter,
etc, etc... this is not true.

Such a filter will have zero loss at the designed reflection zeros which are
distributed at various
appropriate frequencies across the passband according to the dictates of the
approximation
polynomials.

Aside: Reflection zeros are also known as Return Loss [Echo Loss] poles.
These are the
pass band frequencies of zero loss for lossless LC filters designed
according to modern
insertion loss methods. No one really knows where the reflection zeros of
an image
parameter LC filter are, one has to find them by analysis after the design.
Whereas
with insertion loss design the frequencies of zero loss [the reflection
zeros] are specified
by the approximation polynomials, specifically the reflection zero
polynomial usually
designated by F(s). In fact modern insertion loss design begins with a
specification
of attenuation ripple between zero loss and the maximum loss in the pass
band. The
frequencies of zero loss then become the zeros of the reflection zero
polynomial F(s).
The attenuation in the stop band results in the specification of the loss
pole polynomial
P(s) whose zeros are the so called loss poles or attenuation poles. The
natural mode
polynomial of the filter E(s) whose zeros are known as the natural modes or
sometimes
just "the filter poles" is formed from the loss poles and reflection zeros
using Feldtkeller's
Equation.

E(s)E(-s) = P(s)P(-s) +k^2F(s)F(-s)

In the approximation process the stopband attenuation is set first by
"placing" the loss poles
in the stopband, i.e. determining the polynomial P(s). Then from the
desired passband
attenuation and type of approximation desired; maximally flat, equiripple,
etc... the
reflection zeros F(s) are determined and finally from Feldtkeller's Equation
and the ripple
factor k, the natural modes or E(s) is determined.

Then the LC filter is synthesized from either or both of the short circuit
or open circuit
reactance functions which are formed from even and odd parts of E and F, for
example.

X = (Eev - Fev)/(Eod + Fod), etc...

You can review all of this in the very practical and professionally oriented
textbook:

Adel S. Sedra and Peter O. Brackett, "Filter Theory and Design: Active and
Passive",
Matrix Publishers, Champaign, IL 1978.

Another good practical and professionally oriented textbook is:

Louis Weinberg, "Network Analysis and Synthesis", McGraw-Hill, New York,
1962.

If you can get a copy of:

R. Saal and E. Ulbrich, "On the design of filters by synthesis", IRE Trans.
Vol. CT-5,
No. 4, pp.284-327, Dec. 1958.

Bind it firmly and keep it in your library forever... you will have the
whole story in a nutshell.

Saal and Ulbrich is "the bible" on LC filter design.

--
Peter
Freelance Professional Consultant
Signal Processing and Analog Electronics
Indialantic By-the-Sea, FL

Correction:

Roy Lewallen wrote:
. . .
A passive lossless filter can't achieve any frequency selection by means
of loss, so it depends entirely on mismatch. Among other things, that
means that a passive filter works properly only when both the source and
load impedances are the ones it was designed for. A lossless lowpass
filter has zero loss only at DC. . .

The last sentence should read:

A lossless lowpass filter has zero attenuation only at DC.

The attenuation is often called "mismatch loss", but in the remainder of
what I wrote, I use the term loss only to mean dissipative loss -- which
"mismatch loss" isn't.

Roy Lewallen, W7EL

Paul Burridge wrote:
Hi all,

On page 57 of RF circuit Design, Chris Bowick sets out a filter design
example. I've posted this to a.b.s.e under the same subject header. He
claims that the filter in question - a low pass Butterwoth - matches
50 ohms source to 500 ohms load. However, having checked this out with
the aid of a Smith Chart, it appears there is some capacitive
reactance present that would require the addition of a shunt inductor
to neutralize. However, this would of course totally screw up the
filter's characteristics. Upon closer examination, it appears
impossible that this type of arrangement could ever be designed
without introducing some reactance into the signal path. Or am I nuts?
I'd always thought of these kind of filters as being purely resistive
overall at Fo but is that really the case? It don't look like it...


Of course it isn't. A 1st order butterworth is a simple RC. At its
characteristic frequency, i.e. its 3db point, it has 45 degs of phase
shift in its impedance. Note, centre frequency is meaningless for a LP
and HP. In general the input impedance can be all over the place for a
filter.


Kevin Aylward

http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

On Fri, 09 Apr 2004 18:38:30 +0100, Paul Burridge
wrote:

Hi all,

On page 57 of RF circuit Design, Chris Bowick sets out a filter design
example. I've posted this to a.b.s.e under the same subject header. He
claims that the filter in question - a low pass Butterwoth - matches
50 ohms source to 500 ohms load. However, having checked this out with
the aid of a Smith Chart, it appears there is some capacitive
reactance present that would require the addition of a shunt inductor
to neutralize. However, this would of course totally screw up the
filter's characteristics. Upon closer examination, it appears
impossible that this type of arrangement could ever be designed
without introducing some reactance into the signal path. Or am I nuts?
I'd always thought of these kind of filters as being purely resistive
overall at Fo but is that really the case? It don't look like it...

Design criteria:

Centre frequency: 35Mhz
Response -60dB at 105Mhz
zero ripple(!)
Rs 50 ohms
Rl 500 ohms

Most LC filters are designed to produce the advertised frequency
response, given a resistive generator and a resistive load, but they
don't generally look like nice resistors themselves, at either port.

There are filters designed to look like a resistor, or nearly so,
wideband. Picosecond Pulse Labs, among others, sell such. We had a
thread on s.e.d. a while back, and Jeroen Belleman posted some nice
work, some original techniques to make constant-resistance filters.

John

"Paul Burridge" schrieb im Newsbeitrag
...
Hi all,

On page 57 of RF circuit Design, Chris Bowick sets out a filter design
example. I've posted this to a.b.s.e under the same subject header. He
claims that the filter in question - a low pass Butterwoth - matches
50 ohms source to 500 ohms load.

Hello Paul,
to be honest, the circuit filters exactly as written. It is designed for
an input impedance of 50 Ohm and an output impedance of 500 Ohm.

However, having checked this out with
the aid of a Smith Chart, it appears there is some capacitive
reactance present that would require the addition of a shunt inductor
to neutralize. However, this would of course totally screw up the
filter's characteristics. Upon closer examination, it appears
impossible that this type of arrangement could ever be designed
without introducing some reactance into the signal path. Or am I nuts?

You are wrong here. I assume that the book doesn't claim to do an
impedance match to 50 Ohm input resistance for max. output power.
It's just designed as a passive lowpass filter with different source
and load resistor having a flat amplitude response. Nothing more.
The input resistance of this filter is for example 5 Ohm at f=12MHz.
According to the seven reactive parts, it has 3 notches and 3 resonances
for the input resistance over the frequency band from 0 to 200Mhz and
an additional zero at infinity frequency.


I'd always thought of these kind of filters as being purely resistive
overall at Fo but is that really the case? It don't look like it...


There is always a phase shift at the 3dB corner frequency
of any lowpass filter.

Design criteria:

Centre frequency: 35Mhz
This the -3dB corner frequency!
Response -60dB at 105Mhz
zero ripple(!)
Rs 50 ohms
Rl 500 ohms


Paul's filter from a.b.s.e.
---------------------------
A netlist file for the old SPICE people.

* Butter50_500_35MHz.asc
Rs N003 in 50
L1 in N001 152n
L2 N001 N002 323n
L3 N002 out 414n
RL out 0 500
C4 out 0 143p
C3 N002 0 153p
C2 N001 0 97p
C1 in 0 21p
V1 N003 0 AC 1
..ac dec 100 10k 200MEG
..end

Best Regards,
Helmut


PS: Such filters are analyzed with the .AC command in (LT)-SPICE.

The simulation type: .AC DEC 100 10k 1G
The SPICE voltage source: V1 0 AC 1

V(in) is the node after the source resistor Rs.
The frequency response: Logarithmic Bode Plot V(out)
The input impedance: Logarithmic Bode Plot V(in)/I(Rs)

If you want the output impedance, then you have to add a source in
series with the load RL and stimulate from the output side only.
The output impedance: Logarithmic Bode Plot: V(out)/I(RL)


For the newcomers to this group:
--------------------------------
LTSPICE is free SPICE with graphical GUI from www.linear.com/software
There is a special newsgroup for LTSPICE:
http://groups.yahoo.com/group/LTspice/




This is the schematic file of this filter for LTSPICE.
Save it in a file named "Butter50_500_35MHz.asc".


Version 4
SHEET 1 880 708
WIRE 0 176 32 176
WIRE 208 176 240 176
WIRE 368 176 400 176
WIRE 512 176 560 176
WIRE 688 336 560 336
WIRE 688 336 688 304
WIRE 400 224 400 176
WIRE 400 176 432 176
WIRE 240 224 240 176
WIRE 240 176 288 176
WIRE 64 224 64 176
WIRE 64 176 128 176
WIRE 64 288 64 336
WIRE 64 336 -80 336
WIRE 240 288 240 336
WIRE 240 336 64 336
WIRE 400 288 400 336
WIRE 400 336 240 336
WIRE 560 288 560 336
WIRE 560 336 400 336
WIRE 560 224 560 176
WIRE 560 176 688 176
WIRE 688 176 688 224
WIRE -80 176 -224 176
WIRE -224 176 -224 224
WIRE -80 368 -80 336
WIRE -80 336 -224 336
WIRE -224 336 -224 304
WIRE 32 176 64 176
FLAG -80 368 0
FLAG 688 176 out
FLAG 32 176 in
SYMBOL res -96 192 R270
WINDOW 0 32 56 VTop 0
WINDOW 3 0 56 VBottom 0
SYMATTR InstName Rs
SYMATTR Value 50
SYMBOL ind 112 192 R270
WINDOW 0 32 56 VTop 0
WINDOW 3 5 56 VBottom 0
SYMATTR InstName L1
SYMATTR Value 152n
SYMBOL ind 272 192 R270
WINDOW 0 32 56 VTop 0
WINDOW 3 5 56 VBottom 0
SYMATTR InstName L2
SYMATTR Value 323n
SYMBOL ind 416 192 R270
WINDOW 0 32 56 VTop 0
WINDOW 3 5 56 VBottom 0
SYMATTR InstName L3
SYMATTR Value 414n
SYMBOL res 672 208 R0
SYMATTR InstName RL
SYMATTR Value 500
SYMBOL cap 544 224 R0
SYMATTR InstName C4
SYMATTR Value 143p
SYMBOL cap 384 224 R0
SYMATTR InstName C3
SYMATTR Value 153p
SYMBOL cap 224 224 R0
SYMATTR InstName C2
SYMATTR Value 97p
SYMBOL cap 48 224 R0
SYMATTR InstName C1
SYMATTR Value 21p
SYMBOL voltage -224 208 R0
WINDOW 123 21 106 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName V1
SYMATTR Value ""
SYMATTR Value2 AC 1
TEXT -192 24 Left 0 !.ac dec 100 10k 200MEG

Filters can be lossy or, in theory, lossless. An example of a lossy
filter is a first order lowpass, consisting of a series R and shunt C.
But let's talk about the passive lossless variety, made solely of
inductors and capacitors, since I think that's what you're asking about.
In practice, inductors in particular can have appreciable loss, and this
complicates analysis a great deal. But for many applications, for
example HF filters that aren't too sharp, loss can be negligible for
practical purposes. So I'll further simplify things by talking about
only theoretically lossless LC filters.

A passive lossless filter can't achieve any frequency selection by means
of loss, so it depends entirely on mismatch. Among other things, that
means that a passive filter works properly only when both the source and
load impedances are the ones it was designed for. A lossless lowpass
filter has zero loss only at DC. At DC, or very low frequencies, then,
the input is matched to the output. If the filter is designed for 50
ohms in and out, for example, you'd see 50 ohms resistive at the filter
input when the output is terminated in 50 ohms. It can also be designed
for other transformation ratios -- imagine the same filter with a
broadband 10:1 impedance transformer at one end. There are other ways to
effect the transformation, but the end result is the same.

But as you go up in frequency, the attenuation of the filter increases.
In the case of an LC filter, that means -- it has to mean -- that a
mismatch is occurring. The attenuation typically rises slowly and not
too much until you approach the cutoff frequency, but there are an
infinite number of possible filter shapes, and some can vary pretty
wildly in the pass band (the frequency range from DC to cutoff).
Butterworth, Chebyshev, and a number of other canonical types have a
substantial amount of attenuation, and therefore mismatch, at
frequencies quite a bit below cutoff.

An interesting passive LC filter type is a "quarter wave" lowpass
filter. It's so called because it mimics a quarter wave transmission
line over a moderate range of frequencies. This is a pi section filter
(although like any other pi, it can also be realized as a tee)
consisting of a series inductor and shunt capacitors. Each has a
reactance at the operating frequency equal to the source and load
resistance, which for the simple form I'm describing, are equal. This
filter is unusual(*) in that it *is* perfectly matched at the operating
frequency, which is just below the cutoff frequency. The cutoff isn't
particularly sharp, but sections can be cascaded for better high
frequency attenuation without changing the impedance match at the
operating frequency. It's a really handy tool for homebrew transmitters,
where additional harmonic attenuation is needed, since sections can be
added without necessitating output circuit redesign.

(*) It's unusual in my experience with modern filter design, but I
suspect this might be a common characteristic in "image parameter"
designed filters -- a technique I never learned.

Roy Lewallen, W7EL

Paul Burridge wrote:

Hi all,

On page 57 of RF circuit Design, Chris Bowick sets out a filter design
example. I've posted this to a.b.s.e under the same subject header. He
claims that the filter in question - a low pass Butterwoth - matches
50 ohms source to 500 ohms load. However, having checked this out with
the aid of a Smith Chart, it appears there is some capacitive
reactance present that would require the addition of a shunt inductor
to neutralize. However, this would of course totally screw up the
filter's characteristics. Upon closer examination, it appears
impossible that this type of arrangement could ever be designed
without introducing some reactance into the signal path. Or am I nuts?
I'd always thought of these kind of filters as being purely resistive
overall at Fo but is that really the case? It don't look like it...

Design criteria:

Centre frequency: 35Mhz
Response -60dB at 105Mhz
zero ripple(!)
Rs 50 ohms
Rl 500 ohms