EMRFD Message Archive 3027

Message Date From Subject
3027 2009-05-13 12:11:09 Glen Leinweber xtal filter tuning
Please keep the mixer discussion going - excellent.

Was re-reading EMRFD again (hey, I'm slow, and it
takes multiple attempts to sink in). In Chapter 3 where
crystal filters are covered, linear-phase filters are said
to be more difficult to build than Butterworth or
Chebychev. Got to wondering how folks decide whether
a filter they build is usable or not, and if tweaking it
is necessary?
I might sweep a filter and take a subjective look at
the passband shape near the top. Frequency response
graphs in EMRFD imply that this is an important
parameter. But flat group delay is also mentioned as a
worthy design goal for CW or for instrumentation filters.

EMRFD suggests careful crystal measurement in the first
place. Crystal Q and frequency tolerance isn't pushed too
hard, so filter results should be close to design equations.
Few suggestions are given as to how a filter might be
tweaked to shift it closer to what is expected.
I've tried a 4-crystal ladder filter where all coupling elements
and source/load resistances were variable - just to see how
swept filter response is affected by tuning. Quite instructive.
But I haven't tried a test fixture to probe for group delay.
I'd imagine that a carrier at the filter's center frequency,
modulated with a low-frequency square wave might be
useful - hoping that the transient response would tell
something about group delay.
Seems to me that linear-phase filters have such smooth
passband tops that it is hard to tell from a swept display if
the filter is properly tuned. I'd be interested in comments
about how you tell if a filter you build is "good enough",
or is properly tuned.
3028 2009-05-13 13:05:30 ehydra Re: xtal filter tuning
Glen Leinweber schrieb:
> Please keep the mixer discussion going - excellent.
>
> Was re-reading EMRFD again (hey, I'm slow, and it
> takes multiple attempts to sink in). In Chapter 3 where
> crystal filters are covered, linear-phase filters are said
> to be more difficult to build than Butterworth or
> Chebychev. Got to wondering how folks decide whether
> a filter they build is usable or not, and if tweaking it
> is necessary?
> I might sweep a filter and take a subjective look at
> the passband shape near the top. Frequency response
> graphs in EMRFD imply that this is an important
> parameter. But flat group delay is also mentioned as a
> worthy design goal for CW or for instrumentation filters.
>

linear-phase ist VERY important for avoiding ISI (inter-symbol
interference) in digital modes. If it is an audio listener, it is not
important. The human hearing is not sensitive to passband phase
information other than the difference between sound ariving at the two ears.
In fact, it is annoying to find that humans cannot locate a clean
sinusoid sound source in space!

More linear-phase means less filter-steepness in general. More
grade/poles filter can improve it but will introduce more delay.

As always, we say in german "Einen Tod muß man sterben". I think this is
equivalent to "There is no free lunch".


regards -
Henry


--
www.ehydra.dyndns.info
3060 2009-05-19 12:16:09 timshoppa Re: xtal filter tuning
3061 2009-05-19 12:17:14 timshoppa Re: xtal filter tuning
Oh -
Also wanted to add, the real advantage for my CW use of a Gaussian or Linear Phase filter, is that the ringing from a noisy 80M band is far far more easy on the ear than using the Chesbyshev-type filters.

Tim.
3064 2009-05-20 08:15:30 Glen Leinweber Re: xtal filter tuning
Tim wrote:
>I built an 8-pole Guassian-to-12-dB CW filter and am
>very very happy with the way it sounds. Much less
>ringing than a Chebyshev-type filter of similar bandwidth.

Tim also mentioned that he built a "ditter", described in
Chapter 7 of EMRFD - exactly the kind of circuit useful
for probing a filter's transient response. And yes, it does
produce a signal just like Morse code.

Have looked at a filter's output to see transient response
of Chebyshev, Butterworth, Gaussian-to-6dB, and
Flat group-delay filters. Chebyshev (or elliptical) have
"bad" ringing, Butterworth is better, flat group delay is
best. Although Tim built a test fixture to see his filter's
transient response, he didn't mention if this fixture helped
him to tweak the filter to achieve good results. Like most,
he describes qualitatively that his filter sounds better than
others. An 8-pole xtal filter is not easy to build close to
optimum! I was hoping for more than a qualitative answer -
has anyone a procedure to optimize filters for good
transient response?
BTW: Thanks very much Wes, for including as default
the "Gaussian-to-6dB" filter coefficients in XLAD08.
This kind of filter seems close to optimum for CW use,
and should give results close to those Tim enjoys.