Maker Pro
Maker Pro

High pass bessel?

M

Michael

Jan 1, 1970
0
Is it possible to generate a high pass filter with a constant group
delay in the pass band (bessel response)?

This is for car audio, although people claim that phase error cannot
be heard, I believe it can at low F, where the delay created by the
phase shift becomes more significant.

I have seen methods for calculating values on the net but they don't
seem to have a bessel phase response.
I am capable of analysing active filter circuits and deriving
transfer function, I then write simple basic programs to plot the
response. I have verified my program by matching values and creating
the same plots as Microchips filter lab. I have also noticed that
filter lab 2.0 will not generate high pass bessel.

If this is impossible I will be forced to continue with a DSP
approach, which is turing out to be a lot of work.
 
M

Martin Riddle

Jan 1, 1970
0
TI has a program FilterPro. You might want to look into it.

Cheers
 
M

maxfoo

Jan 1, 1970
0
Read page 275 from the 2nd edition of "The Art of Electronics"
Available at Borders or Barnes & Noble.
 
M

Michael

Jan 1, 1970
0
Martin Riddle said:
TI has a program FilterPro. You might want to look into it.

Cheers

Thanks Martin, thats a nice program. Much better than Microchips
filterlab. It looks as though hi-pass filters can have a flat group
delay only in the cutoff band, and drop to a delay of zero through the
pass band. I was able to achive a suitable alignment that only has a
delay of 4-5mS under the crossover point.
 
R

Rene Tschaggelar

Jan 1, 1970
0
Michael said:
Is it possible to generate a high pass filter with a constant group
delay in the pass band (bessel response)?

This is for car audio, although people claim that phase error cannot
be heard, I believe it can at low F, where the delay created by the
phase shift becomes more significant.

I have seen methods for calculating values on the net but they don't
seem to have a bessel phase response.
I am capable of analysing active filter circuits and deriving
transfer function, I then write simple basic programs to plot the
response. I have verified my program by matching values and creating
the same plots as Microchips filter lab. I have also noticed that
filter lab 2.0 will not generate high pass bessel.

If this is impossible I will be forced to continue with a DSP
approach, which is turing out to be a lot of work.

The choice of bessel or something else has nothing to do
with analog or digital. A DSP can do both.
A somewhat better understanding of the subject than just
using an application may be necessary though.

Rene
 
B

Bill Sloman

Jan 1, 1970
0
Is it possible to generate a high pass filter with a constant group
delay in the pass band (bessel response)?

This is for car audio, although people claim that phase error cannot
be heard, I believe it can at low F, where the delay created by the
phase shift becomes more significant.

I have seen methods for calculating values on the net but they don't
seem to have a bessel phase response.
I am capable of analysing active filter circuits and deriving
transfer function, I then write simple basic programs to plot the
response. I have verified my program by matching values and creating
the same plots as Microchips filter lab. I have also noticed that
filter lab 2.0 will not generate high pass bessel.

If this is impossible I will be forced to continue with a DSP
approach, which is turing out to be a lot of work.

Get hold of a copy of the "Electronic Filter Design Handbook" by
Arthur B.Williams and Fred J.Taylor.
My copy is the second edition, ISBN 0-07-070434-1.
The third edition, ISBN 0-07-070430-9, has been claimed to be even
better.

The book seems to be out of print a the moment, and Amazon has three
buyers waiting. You may be able to find it in a university library.

Very useful for precisely your sort of problem - and it also covers
linear phase filters with equi-ripple phase error, whihc you might
find useful.
 
J

Jeroen Vriesman

Jan 1, 1970
0
Is it possible to generate a high pass filter with a constant group
delay in the pass band (bessel response)?

This is for car audio, although people claim that phase error cannot
be heard, I believe it can at low F, where the delay created by the
phase shift becomes more significant.

Below 600Hz the output signals of the neurons responsible for sound
processing are completely in sync with the audio signal, therefore phase is
important. Between 600Hz-1.8kHz the output signal of our neurons is still
synchronous with the audio signal, but since our neurons cannot fire that
fast, not every top in the audio signal is a neuron output peak, so phase
sensitivity becomes less. Above 1.8kHz there is no synchronicity between
the signal and the output of our neurons.

So a constant group delay below 2kHz should be ok.

But since it's for car-audio, try to calculate the effect of the car
itself, reflections against the windows etc. on your sound, much bigger
than a little filter phase shift.
 
P

Phil Hobbs

Jan 1, 1970
0
Michael said:
Thanks Martin, thats a nice program. Much better than Microchips
filterlab. It looks as though hi-pass filters can have a flat group
delay only in the cutoff band, and drop to a delay of zero through the
pass band. I was able to achive a suitable alignment that only has a
delay of 4-5mS under the crossover point.

The classical design procedure for filters is to start with a low-pass
prototype, and apply a conformal mapping in the complex frequency plane.

For a high pass, this means transforming f ->fc**2/f. Doing this and
keeping all the reactances the same in magnitude (inductors become
capacitors, capacitors become inductors) gives you a highpass.

This works great for the magnitude response, but since the frequency
stretch is nonlinear, your nice linear phase response has suddenly
become a nice approximation to a particular nonlinear curve, instead.

Bandpass has the same trouble.

A bit of Matlab code will optimize one for you.

Cheers,

Phil Hobbs
 
Z

Zak

Jan 1, 1970
0
Michael said:
Thanks Martin, thats a nice program. Much better than Microchips
filterlab. It looks as though hi-pass filters can have a flat group
delay only in the cutoff band, and drop to a delay of zero through the
pass band.

Makes sense - how can a few capacitors store hundreds of cycles of 20
KHz signal? That would be needed to delay these as much as a 90 degree
shift at low frequency.


Thomas
 
K

Ken Smith

Jan 1, 1970
0
Jeroen Vriesman said:
sensitivity becomes less. Above 1.8kHz there is no synchronicity between
the signal and the output of our neurons.

So a constant group delay below 2kHz should be ok.

A does not prove B in this case. A rapid change in phase VS frequency can
still be detected above 2KHz. The simplest proof is to imagine an AM
modulated 2KHz sine wave. If the modulation frequency is low enough, you
can hear the signal decrease to zero amplitude. If this signal is then
run through a circuit with a very rapid change in phase, the amplitude
will no longer decrease to zero.

But since it's for car-audio, try to calculate the effect of the car
itself, reflections against the windows etc. on your sound, much bigger
than a little filter phase shift.

Besides, if it is for a car the environment adds a lot of noise to the
music. This makes errors in the sound system harder to hear.
 
V

Vladimir Vassilevsky

Jan 1, 1970
0
Rene said:
The choice of bessel or something else has nothing to do
with analog or digital. A DSP can do both.

It is not so easy.
The DSP response is limited to Fsa/2, versus the analog response which
goes to the infinite frequency. The frequency warping and/or aliasing
effect is introduced. The DSP can never have the exact Bessel,
Butterworth or whatsoever transfer function. However you can do a lot of
other different things with the DSP.

A somewhat better understanding of the subject than just
using an application may be necessary though.

That is very true :)

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com
 
K

Kevin Aylward

Jan 1, 1970
0
K

Kevin Aylward

Jan 1, 1970
0
Michael said:
Is it possible to generate a high pass filter with a constant group
delay in the pass band (bessel response)?

This is for car audio, although people claim that phase error cannot
be heard, I believe it can at low F, where the delay created by the
phase shift becomes more significant.

Well, this is a bit misleading. Static phase shift, is indeed, to all
practicable intents and purposes, not detectable by the ear. That is,
put a signal through an all pass filter, i.e constant magnitude and
varying phase with frequency, and you simply wont be able to tell the
phase shifted signal from the unshifted signal.

However, two signals mixed together with differing static phases can
result in huge audible differences. Such a combined signal will result
in cancellations and/or peaks in the frequency response.

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

Active8

Jan 1, 1970
0
And note my one:) It does do all the main HP and LP filters, LC and
opamp versions.

oh yeah. don't forget kevin's filter designer in SS :)
 
B

Ban

Jan 1, 1970
0
Zak said:
Makes sense - how can a few capacitors store hundreds of cycles of 20
KHz signal? That would be needed to delay these as much as a 90 degree
shift at low frequency.


Thomas

Zak,
the delay drops to zero in the pass-band, which is in our case the high
frequencies.
The transfer function for a 2nd order high-pass with gain(oo)=1 will be:

A(P)= 1/(1+ a1/P+ b1/P^2) with P= j*Omega= j*f/fg
for a Bessel characteristic a1= 1.3617 and b1= 0.618.

Since our transfer function returns complex values, we can calculate
amplitude and phase with it.
Let's calculate a few values:

P= j0.1 A(P)=1/(1+ 1.3617/j0.1+ 0.618/0.01j^2)
=1/(1- j13.617- 61.8) = -1/(60.8+ j13.617)
when transformed into polar we get: |A|= 0.01605 and phi= -7.8°
with 100Hz the group delay would be 0.2167ms

P= j1 A(P)=1/(1+ 1.3617/j1+ 0.618/1j^2)
=1/(1- j1.3617- 0.618) = 1/(0.382- j1.3617)
when transformed into polar we get: |A|= 0.7071 and phi= -74.3°
with 1000Hz the group delay would be 0.2065ms

P= j10 A(P)=1/(1+ 1.3617/j10+ 0.618/100j^2)
=1/(1- j0.13617- 0.00618) = 1/(0.994- j0.13617)
when transformed into polar we get: |A|= 1.00 and phi= -12.6°
with 10000Hz the group delay would be 0.0035ms

ciao Ban
 
R

Rene Tschaggelar

Jan 1, 1970
0
Vladimir said:
It is not so easy.
The DSP response is limited to Fsa/2, versus the analog response which
goes to the infinite frequency. The frequency warping and/or aliasing
effect is introduced. The DSP can never have the exact Bessel,
Butterworth or whatsoever transfer function. However you can do a lot
of other different things with the DSP.
You're right.
But the analog filters do not go to infinity, neither the ones
with passive elements only, nor the one with opamps.

I was thinking of using the DSP for frequencies far below
the sampling frequency. Yes, much is possible.

Rene
 
M

Michael

Jan 1, 1970
0
Rene Tschaggelar said:
You're right.
But the analog filters do not go to infinity, neither the ones
with passive elements only, nor the one with opamps.

I was thinking of using the DSP for frequencies far below
the sampling frequency. Yes, much is possible.

Rene

So was I, my dsp approach was such:
Keep in mind this is a one off project for personal use, and I have
limited resources.

I have an MSP430F147 processing two 147 tap Low pass FIR filters,
sample rate is 2kHz.
Analog filters remove alias F's over 1k and remove step noise from the
output.
A digital delay (bunch of ram chips, a2d's and d2'a, and a pic to run
it all) will then delay the original signal 36mS to bring it in phase
with the FIR filter output.
Because the low pass and original signal will remain in phase I will
be able to use opamps to subtract the bass from the original to get my
mids & highs.
I will use digital attenuators after the filters to control overall
volume of the 9 channels.

However:
Last night I measured the phase response of the vented sub in my car,
it was filthy and I'm not so worried about the phase response of my
xover anymore, so I'm just gonna use some analog filters.

Thanks for the help,

Regards, Michael.
 
Top