splitting a filter

tz

uple

y n

the

,

essel

oes

torize

in the

r

at

ed

The Eagleware is a freq based package ... so you will have to go back to spice for the time domain info. But your design time should go down immensely.

The big advantage is that you can add parasitics and get reliable results fairly quickly. The disadvantage is having to spend time learning Eagleware.

The time domain response is infinite in length, so clearly all you can do is approximate what you need if such software existed.

When you split a filter into two pieces with the intent of cascading them, then what you want to achieve is the square root of the cascaded response for each filter. Since you are concerned about time, you would want the group delay of each filter to be half of the cascaded (i.e prototype) filter. Find a prototype filter with the proper response. Create a new template for the optimizer that is the square root of the prototype. Try to synthesize that filter.

le

he

sel

s

rize

the

sorry I don't follow your logic.. adding a filter will not increase the energy below 100 MHz of the pulses, it will only decrease the energy above 100 MHz..

what is the order of your stages...

is it from left to right...

PD >>> 1/2 FILTER >>>CABLE >>>> 1/2 FILTER >>>> OP AMP>>>

if so then I don't see the point of splitting the filter, put it all to the right of the cable???

Mark

Hmm, but such software *does* exist. You can enter the time-domain response directly into most DSP software and see the frequency response and pole-zero plot, live. Even send audio through it simultaneously.

So you get the Z domain transfer function, which is just the sampling of the time domain response (FIR style). Do the Z-to-F conversion and you've got your transfer function. Then however the hell you convert that to a ladder of LCs, do it and you're done.

Since it's DSP and numerical methods stuff, things like "approximation" and "I know the real filter isn't FIR" are intrinsic to the process, and you'll get something 'close enough', which is all that any engineer really wants.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

John Larkin schrieb:

Hello,

my textbook here says cascading two 3-pole Bessel filters to get a

6-pole Bessel filter is the wrong way.

You have to take the coefficients of the 6-pole filter and transform it from v = v0/(1+c1P+c2P^2+c3P^3+c4P^4+c5P^5+c6P^6)

into

v = v0/(1+a1P+a2P^2)(1+b1P+b2P^2)(1+c1P+c2P^2)

each filter v = v0/(1+a1P+a2P^2) v = v0/(1+b1P+b2P^2) v = v0/(1+c1P+c2P^2) can be realized seperatly. You may split the 6-pole filter this way into a 4-pole filter and a 2-pole filter.

If you find a transformation like: v = v0/(1+a1P+a2P^2+a3P^3)(1+b1P+b2P^2+b3P^3)

the realization as two 3-pole filters is also possible, but only if such a transformation exists.

Bye

How many poles and zeros would a piece of transmission line add?

I want to say, just use a calibrated piece of coax, and include its characteristics in a sort of 6 1/2 pole filter. ;-)

Have Fun! Rich

Constant impedance filters exhibit a known (e.ge. 50R) impedance across the whole frequency axis. If you are able to make up your bessel filter from different constant-impedance sections you could split them up using

50R transmission lines between sections without changing the frequency response of the filter.

This may help

Pere

On a sunny day (Thu, 28 Oct 2010 14:36:27 -0700) it happened John Larkin wrote in :

Just a curious question, what happens if you simply add 2 of your existing filters in series, one at each end of the cable? Would that not work?

That's a very strange thing to want to do. Why not just move the existing filter to whichever box makes more sense? (There must be some unarticulated design requirement driving this!)

Any impedance mismatch between filter sections and the intervening transmission line are liable to mess up that nice clean impulse response, so I'd really be wanting to put the filter and buffer right at the PD.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

Approximation of the time domain response is intractable problem. Approximation of the frequency response obtained as Fourier transform of the time domain response is much simpler problem, but it is not equvalent to the time domain approximation.

This trivia will produce a delay line with taps which is neither the simplest nor the optimal way to get to the desired result.

So, why it is still impossible to have sex in the middle of the Red Square in Moscow?

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Since you can consider the transmission line to be an infinite set of RLC elements (drop "R" for lossless lines), they have infinite poles and zeroes.

You're probably aware that one can make a "simulated" transmission line out of a finite number of L's and C's and it works OK up to some frequency (the higher in frequency you'd like it to behave as one, the more L's and C's you need)... although you often end up needing rather more L's and C's than you might guess... or at least I always did...

One paper on this topic:

---Joel

Thanks, Joel!

PSpice's transmission line modeling is less than exemplary, so this paper will be handy when I need to ensure some realism in my simulations. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |

               I can see November from my house :-)

Because you weren't man enough to try it.

--
Politicians should only get paid if the budget is balanced, and there is
enough left over to pay them.

I was going to brag... but I won't... get me in trouble... restaurant parking lot at Scottsdale Airpark, mid '70's ;-) ...Jim Thompson

-- | James E.Thompson, CTO | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

| 1962 |

I can see November from my house :-)

Oh, we'd certainly use opamps to drive and receive the cable signals. I Have My Reasons for wanting filters in both boxes. I could just put

5-pole Bessels in each - that would work and the delay wouldn't hurt this application much - but it would save parts to split it up. And I run into a similar problem now and then: given a desired overall impulse response, and given a pole or two that's unavoidable somewhere, how does one design an LC filter to give the net desired response? In other words, how do you back one pole out of an LC Bessel filter?

It's easier with active filters.

John

text -

Imagine a lidar (which it actually isn't) with a photodiode seeing a big transmit pulse and weak return pulses, with the first echo coming about 10 ns after the transmit pulse. The optical pulses themselves are picoseconds. The impulse response of the photodiode is maybe a ns wide. To keep the amplifier linearity good, with affordable amplifiers, it's nice to lowpass filter the blips into maybe 3 ns impulses, so we're basically working at 100 MHz, where things are fairly easy. So the first echo peak, less than 1% of the main pulse, happens 10 ns after the main peak. The filters will delay both equally, so that's OK.

It's sort of the deconvolution problem all over again, only with passive LC elements.

I guess it's not a simple problem, so we'll have to do it some other way.

It would be easy with active filters: just split the filter somewhere in the middle. I wonder if a Sallen-Key is feasible at 100 MHz.

TI's FilterPro software happily designs a 100 MHz S-K 5-p Bessel active filter, using E96 resistors and E6 caps. I'll mosey on over to Digikey and order some 47 fF caps.

John

Yup, LC delay lines are messy. They ring badly, and the number of stages increases as the square of the delay/risetime ratio. That hurts.

The old Tek scopes, 545 vintage, used discrete delay lines for the vertical delay. They used t-coils, which work better than simple LCs. There were dozens of them, each with a tuning slug. The alignment procedure is a Halloween horror.

John

And you didn't have to be a party member, like in Moscow.

BTW: How do you starve a Libtard to death?"

You hide his food stamps under his work boots.

--
Politicians should only get paid if the budget is balanced, and there is
enough left over to pay them.

Neat... and at 30MHz no less!

I have a suspicion that if such technology were needed again today, the people developing it would feel quite confident it was cutting edge research and immediately file for patent protection...

Say... (and this is completely off-topic)... you're doing DDSes in FPGAs, right? How about PLLs? Integer, fractional, sigma-delta...? Any of that?

thanks,

---Joel