Elecraft

K3 roofing filter group delay

Classic

List

Threaded

9 messages Options

Brian Lloyd-6

K3 roofing filter group delay

A number of messages have gone back and forth here about roofing
filters. We did mention group delay but I wonder if Elecraft can
provide group delay characteristics for the various filters offered
for the K3.

Seems to me that, in general, fewer poles tend to provide better
group delay in a filter at the expense of the skirts. OTOH, if the
skirts in the roofing filter are sufficient to attenuate a strong,
undesired signal so that it cannot cause desense, then it strikes me
that the 5-pole filters might actually provide superior performance
for digital communications.

73 de Brian, WB6RQN
Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com

_______________________________________________
Elecraft mailing list
Post to: [hidden email]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft

Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com

Don Wilhelm-3

Re: K3 roofing filter group delay

Brian and all,

The number of poles is not the primary deciding factor. It is the type
of filter (Cohn, Chebychev, Butterworth, Gaussian to 6 dB, Gaussian to
12 dB, etc.) that will influence the group delay. See the discussion on
filters in Experimental Methods for RF Design for further information.

In general, those filters with a 'rounded nose' will have the best group
delay characteristics - but that is only a generalization, the details
will tell "the rest of the story".

73,
Don W3FPR

Brian Lloyd wrote:

> A number of messages have gone back and forth here about roofing
> filters. We did mention group delay but I wonder if Elecraft can provide
> group delay characteristics for the various filters offered for the K3.
>
> Seems to me that, in general, fewer poles tend to provide better group
> delay in a filter at the expense of the skirts. OTOH, if the skirts in
> the roofing filter are sufficient to attenuate a strong, undesired
> signal so that it cannot cause desense, then it strikes me that the
> 5-pole filters might actually provide superior performance for digital
> communications.
>
> 73 de Brian, WB6RQN
> Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com
>
>
> _______________________________________________
> Elecraft mailing list
> Post to: [hidden email]
> You must be a subscriber to post to the list.
> Subscriber Info (Addr. Change, sub, unsub etc.):
> http://mailman.qth.net/mailman/listinfo/elecraft
> Help: http://mailman.qth.net/subscribers.htm
> Elecraft web page: http://www.elecraft.com
>
>

_______________________________________________
Elecraft mailing list
Post to: [hidden email]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft

Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com

John, KI6WX

Re: K3 roofing filter group delay

Don;
Don't forget Bessel filters. These are maximally flat for group delay, but
have a much slower rolloff.

A 5-pole 500-Hz 0.5 dB Chebyshev will be down 80 dB at 1 kHz and will have a
group delay of 16 ms at band center and 44 ms at -3 dB. A 5-pole 500-Hz
Bessel filter will be down 40 dB at 1 kHz and have a constant group delay
across the passband of 10 ms.

You can have a flat group delay or a fast rolloff, but you can't have both.

-John
KI6WX

> Brian and all,
>
> The number of poles is not the primary deciding factor. It is the type of
> filter (Cohn, Chebychev, Butterworth, Gaussian to 6 dB, Gaussian to 12 dB,
> etc.) that will influence the group delay. See the discussion on filters
> in Experimental Methods for RF Design for further information.
>
> In general, those filters with a 'rounded nose' will have the best group
> delay characteristics - but that is only a generalization, the details
> will tell "the rest of the story".
>
> 73,
> Don W3FPR
>
> Brian Lloyd wrote:
>> A number of messages have gone back and forth here about roofing filters.
>> We did mention group delay but I wonder if Elecraft can provide group
>> delay characteristics for the various filters offered for the K3.
>>
>> Seems to me that, in general, fewer poles tend to provide better group
>> delay in a filter at the expense of the skirts. OTOH, if the skirts in
>> the roofing filter are sufficient to attenuate a strong, undesired signal
>> so that it cannot cause desense, then it strikes me that the 5-pole
>> filters might actually provide superior performance for digital
>> communications.
>>
>> 73 de Brian, WB6RQN
>> Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com
>>
>>

Don Wilhelm-3

Re: K3 roofing filter group delay

John,

Thanks for that additional info - I had forgotten about the Bessel filter.

There is a distinct advantage with the K3, the 'roofing filter' is not
the major element in setting the rolloff of the total filtering and the
ultimate receiver selectivity because that aspect is handled by the DSP.
That makes the use of roofing filters with a slow rolloff (not steep
skirts) quite feasible.

I believe we should be looking at group delay filter characteristics
more closely for use with data modes, especially with the advent of
digital voice mode.

73,
Don W3FPR

John, KI6WX wrote:

> Don;
> Don't forget Bessel filters. These are maximally flat for group delay,
> but have a much slower rolloff.
>
> A 5-pole 500-Hz 0.5 dB Chebyshev will be down 80 dB at 1 kHz and will
> have a group delay of 16 ms at band center and 44 ms at -3 dB. A 5-pole
> 500-Hz Bessel filter will be down 40 dB at 1 kHz and have a constant
> group delay across the passband of 10 ms.
>
> You can have a flat group delay or a fast rolloff, but you can't have both.
>
> -John
> KI6WX
>
>
>> Brian and all,
>>
>> The number of poles is not the primary deciding factor. It is the
>> type of filter (Cohn, Chebychev, Butterworth, Gaussian to 6 dB,
>> Gaussian to 12 dB, etc.) that will influence the group delay. See the
>> discussion on filters in Experimental Methods for RF Design for
>> further information.
>>
>> In general, those filters with a 'rounded nose' will have the best
>> group delay characteristics - but that is only a generalization, the
>> details will tell "the rest of the story".
>>
>> 73,
>> Don W3FPR
>>
>> Brian Lloyd wrote:
>>> A number of messages have gone back and forth here about roofing
>>> filters. We did mention group delay but I wonder if Elecraft can
>>> provide group delay characteristics for the various filters offered
>>> for the K3.
>>>
>>> Seems to me that, in general, fewer poles tend to provide better
>>> group delay in a filter at the expense of the skirts. OTOH, if the
>>> skirts in the roofing filter are sufficient to attenuate a strong,
>>> undesired signal so that it cannot cause desense, then it strikes me
>>> that the 5-pole filters might actually provide superior performance
>>> for digital communications.
>>>
>>> 73 de Brian, WB6RQN
>>> Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com
>>>
>>>
>
>
>

Jack Smith-6

Re: K3 roofing filter group delay

Don:

Another interesting classical design is the equal-ripple-delay filter.
It uses the Chebyshev concept of allowing ripple in the passband to
provide increased skirt selectivity, except that the ripple that is
allowed is in the delay function, not the amplitude. By allowing a
slight amount of ripple in group delay of an equaphase filter, the
skirts can be improved over similar order filter with a flat group delay.

Also -- since the K3's roofing filters are backed up with a DSP, could
not the DSP implement a delay equalizer, so that the net delay is flat?
I also don't know whether the roofing filters are repeatable enough to
permit a single equalization setting or if each filter will require
custom coefficients. (If so, I suppose each filter could be equipped
with a small EEPROM chip with its equalization parameters, to be read
when a new filter is installed into a K3.)

I don't know if the K3's DSP implements IIR or FIR filters, but an FIR
can be implemented with linear phase (flat group delay) Hence, if FIR
filters are used, the only group delay requiring equalization will be
the crystal roofing filter.

Jack K8ZOA
www.cliftonlaboratories.com

Don Wilhelm wrote:

> John,
>
> Thanks for that additional info - I had forgotten about the Bessel
> filter.
>
> There is a distinct advantage with the K3, the 'roofing filter' is not
> the major element in setting the rolloff of the total filtering and
> the ultimate receiver selectivity because that aspect is handled by
> the DSP. That makes the use of roofing filters with a slow rolloff
> (not steep skirts) quite feasible.
>
> I believe we should be looking at group delay filter characteristics
> more closely for use with data modes, especially with the advent of
> digital voice mode.
>
> 73,
> Don W3FPR
>
> John, KI6WX wrote:
>> Don;
>> Don't forget Bessel filters. These are maximally flat for group
>> delay, but have a much slower rolloff.
>>
>> A 5-pole 500-Hz 0.5 dB Chebyshev will be down 80 dB at 1 kHz and will
>> have a group delay of 16 ms at band center and 44 ms at -3 dB. A
>> 5-pole 500-Hz Bessel filter will be down 40 dB at 1 kHz and have a
>> constant group delay across the passband of 10 ms.
>>
>> You can have a flat group delay or a fast rolloff, but you can't have
>> both.
>>
>> -John
>> KI6WX
>>
>>
>>> Brian and all,
>>>
>>> The number of poles is not the primary deciding factor. It is the
>>> type of filter (Cohn, Chebychev, Butterworth, Gaussian to 6 dB,
>>> Gaussian to 12 dB, etc.) that will influence the group delay. See
>>> the discussion on filters in Experimental Methods for RF Design for
>>> further information.
>>>
>>> In general, those filters with a 'rounded nose' will have the best
>>> group delay characteristics - but that is only a generalization, the
>>> details will tell "the rest of the story".
>>>
>>> 73,
>>> Don W3FPR
>>>
>>> Brian Lloyd wrote:
>>>> A number of messages have gone back and forth here about roofing
>>>> filters. We did mention group delay but I wonder if Elecraft can
>>>> provide group delay characteristics for the various filters offered
>>>> for the K3.
>>>>
>>>> Seems to me that, in general, fewer poles tend to provide better
>>>> group delay in a filter at the expense of the skirts. OTOH, if the
>>>> skirts in the roofing filter are sufficient to attenuate a strong,
>>>> undesired signal so that it cannot cause desense, then it strikes
>>>> me that the 5-pole filters might actually provide superior
>>>> performance for digital communications.
>>>>
>>>> 73 de Brian, WB6RQN
>>>> Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com
>>>>
>>>>
>>
>>
>>
> _______________________________________________
> Elecraft mailing list
> Post to: [hidden email]
> You must be a subscriber to post to the list.
> Subscriber Info (Addr. Change, sub, unsub etc.):
> http://mailman.qth.net/mailman/listinfo/elecraft
> Help: http://mailman.qth.net/subscribers.htm
> Elecraft web page: http://www.elecraft.com
>

John Magliacane

Re: K3 roofing filter group delay

In reply to this post by John, KI6WX

> Don't forget Bessel filters. These are maximally flat for group delay, but
> have a much slower rolloff.

John,

What penalty (if any) would occur in cascading identical Bessel filters to
improve roll-off and stop-band attenuation?

Would the overall response still be considered Bessel, or will a point be
reached where the response begins to take on Chebyshev characteristics?

Thanks.

73, de John, KD2BD

Visit John on the Web at:

http://kd2bd.ham.org/
.
.
.
.

____________________________________________________________________________________Looking for a deal? Find great prices on flights and hotels with Yahoo! FareChase.
http://farechase.yahoo.com/
_______________________________________________
Elecraft mailing list
Post to: [hidden email]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft

Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com

Brian Lloyd-6

Re: K3 roofing filter group delay

In reply to this post by John, KI6WX

On May 13, 2007, at 11:18 PM, John, KI6WX wrote:

> Don;
> Don't forget Bessel filters. These are maximally flat for group
> delay, but have a much slower rolloff.
>
> A 5-pole 500-Hz 0.5 dB Chebyshev will be down 80 dB at 1 kHz and
> will have a group delay of 16 ms at band center and 44 ms at -3
> dB. A 5-pole 500-Hz Bessel filter will be down 40 dB at 1 kHz and
> have a constant group delay across the passband of 10 ms.
>
> You can have a flat group delay or a fast rolloff, but you can't
> have both.

Right. TANSTAAFL (there ain't no such thing as a free lunch). Given
that we can get very steep skirts in DSP at the expense of
introducing more overall latency (delay) in the filter (usually not a
problem for what we are doing), then we can probably accept analog
roofing filters that don't provide the same ultimate attenuation in
the skirts.

Given the choice between the two filters I would probably opt for the
Bessel filter myself.

So, Elecraft, will you be offering a group of roofing filters
optimized for digital communications? How about your "brick wall"
filters in DSP?

73 de Brian, WB6RQN
Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com

_______________________________________________
Elecraft mailing list
Post to: [hidden email]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft

Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com

Brian Lloyd-6

Re: K3 roofing filter group delay

In reply to this post by Don Wilhelm-3

On May 14, 2007, at 5:32 AM, Don Wilhelm wrote:

> John,
>
> Thanks for that additional info - I had forgotten about the Bessel
> filter.
>
> There is a distinct advantage with the K3, the 'roofing filter' is
> not the major element in setting the rolloff of the total filtering
> and the ultimate receiver selectivity because that aspect is
> handled by the DSP. That makes the use of roofing filters with a
> slow rolloff (not steep skirts) quite feasible.
>
> I believe we should be looking at group delay filter
> characteristics more closely for use with data modes, especially
> with the advent of digital voice mode.

That is especially true if you are trying to put all your bits on one
carrier. If you opt for something like OFDM (multiple carriers) and a
lower symbol rate, each carrier and its sidebands will be narrower
and will not have as much delay skew across that carrier and its
sidebands. In that case center-to-edge group delay spread is less
harmful.

OTOH, it will be critical for trying to dig a really weak narrow
signal out.

Yeah, I agree with you. I really want to see what the filters are
going to look like.

73 de Brian, WB6RQN
Brian Lloyd - brian HYPHEN wb6rqn AT lloyd DOT com

_______________________________________________
Elecraft mailing list
Post to: [hidden email]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft

Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com

John, KI6WX

Re: K3 roofing filter group delay

In reply to this post by John Magliacane

You can cascade a pair of Bessel filters to get better frequency response
while maintaining a flat group delay, but you do have to isolate them so
that they both are seeing a resistive source and load impedance. The
cascaded filters would show a 6 dB rolloff at nominal cutoff frequency. The
3 dB point will occur at about 70% of bandwidth. A pair of 5-pole Bessel
filters cascaded with a 500 Hz 3 dB bandwidth will be down about 55 dB at 1
kHz.

The better method is to build a 10-pole 500-Hz Bessel filter which will be
down nearly 60 dB at 1 kHz.

There was also a question about a linear phase filter with equiripple error.
A 0.5 degree error is about +/-5% variation in group delay. A 5-pole 500-Hz
0.5 degree equiripple filter will be down 50 dB at 1 kHz and have about 11
ms of group delay.
-John
KI6WX

>> Don't forget Bessel filters. These are maximally flat for group delay,
>> but
>> have a much slower rolloff.
>
> John,
>
> What penalty (if any) would occur in cascading identical Bessel filters to
> improve roll-off and stop-band attenuation?
>
> Would the overall response still be considered Bessel, or will a point be
> reached where the response begins to take on Chebyshev characteristics?
>