Adjustable Linkwitz-Riley filter: Help needed with SPICE - Page 2 - diyAudio
Go Back   Home > Forums > Source & Line > Analog Line Level

Analog Line Level Preamplifiers , Passive Pre-amps, Crossovers, etc.

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 15th April 2012, 09:41 AM   #11
godfrey is online now godfrey  South Africa
diyAudio Member
 
Join Date: Nov 2009
Location: Cape Town
Hi NebuK

It sounds like it worked fine as soon as soon as you left out the input buffer opamp. I've no idea why, because that shouldn't make any difference at all.

Quote:
Originally Posted by NebuK View Post
i've corrected the 49K9 to 49K8, sadly the rise at ~18k in the lowpass curve didn't vanish
I think you're needlessly worrying here. Let's get a couple of things in perspective.

Firstly the "bad behavior" at -120dB above 18KHz: That really is a very, very small amount of crosstalk. -120 dB = 0.0001% = 1 part per million. That's far lower than the theoretical minimum possible noise from a CD. It's also less than either the crosstalk, noise, or distortion of most amplifiers.

When you actually build the circuit, the crosstalk will probably be worse than that due to stray capacitance, coupling through the ground wiring and power supply rails etc.

Secondly, the resistor values: As Andrew pointed out, if you use two 24k9 resistors in series instead of 49.9, there is an error of 100 ohms. So what? If you're using 1% tolerance parts, the value could be out by up to 500 ohms anyway.

IMHO, the resistor selection in that circuit is idiotic. The actual values aren't important, it's the ratio between them that counts, so there's no reason to use non-standard or hard-to find values throughout.

For example, I got the almost perfect results below as follows:
A) Replace all the 24.9K resistors with 27K.
B) Replace the 49.9K resistor with 56K.
C) Replace the 5.23K resistor with 6.75K.

The only non-standard value there is 6.75K, but that's the resistor that needs to be adjustable, so you could use e.g. 4.7K in series with a 5K trimmer, or 5.6K in series with a 2K trimmer, depending on how much adjustment range you want.

Cheers - Godfrey


Edit:
Quote:
Originally Posted by NebuK View Post
(but the strange hump in my original schematics lowpass remains )
That's probably just due to the simulator taking into account that real opamps don't have infinite gain and their output impedance isn't zero. Nothing to worry about. As mentioned above, real life will be worse.
Attached Images
File Type: gif resp2.GIF (13.2 KB, 148 views)

Last edited by godfrey; 15th April 2012 at 09:46 AM.
  Reply With Quote
Old 15th April 2012, 10:52 AM   #12
AndrewT is offline AndrewT  Scotland
diyAudio Member
 
Join Date: Jul 2004
Location: Scottish Borders
Quote:
Originally Posted by godfrey View Post
If you're using 1% tolerance parts, the value could be out by up to 500 ohms anyway.
But I would select resistors to better than 0.2% total difference for a single channel. And probably try to get the second channel within that same <0.2% total difference.
__________________
regards Andrew T.
  Reply With Quote
Old 15th April 2012, 11:09 AM   #13
godfrey is online now godfrey  South Africa
diyAudio Member
 
Join Date: Nov 2009
Location: Cape Town
Fair enough. I wonder how sensitive it is to component tolerance. (I've just managed to talk myself out of wasting an hour of my life trying to find out.)
  Reply With Quote
Old 16th April 2012, 07:29 AM   #14
NebuK is offline NebuK  Germany
diyAudio Member
 
Join Date: Jan 2011
Hey,

i've tried to resimulate a bit, toy around here, toy around there, change the resistors to worst-case-5%-values ... everything still looked okay. Except that strange hump ;P.

Once again, i'll post the picture:
Click the image to open in full size.

This is now using godfreys values for the resistors. LTSpice has been told the resistors are 5% tolerance parts (i dunno how it takes this into account because ... nothing changed - same with manually adding 5% tolerance btw! I still planned on 0.1% resistors). Well, Even though you tell me there will be higher crosstalk in the real circuit (probably...!), this hump somehow worries me. At 100kHz its at some -87dB, while that is well more silent and high than what we do hear, wouldn't it be better just to get this stuff out as much as possible, as to save the amplifier having to amplify some senseless ghost frequencies?

Thanks a bunch! I'll update my schematic and PCB later and post them for further critique ;P. Thanks so much!

- NebuK

*EDIT: Ps: Capacitor tolerance values seem to be much much worse! I can't seem to find 3.3n caps at under 5% tolerance... but simulating this doesn't - once again - show any difference. So ... i'm on the safe side? (Still trying to get best matched parts possible...)

Last edited by NebuK; 16th April 2012 at 07:34 AM.
  Reply With Quote
Old 16th April 2012, 09:03 AM   #15
Calvin is offline Calvin  Germany
diyAudio Member
 
Calvin's Avatar
 
Join Date: Nov 2004
Location: close to Basel
Hi,

the topology was afaik introduced by Prof. Hawksford in a AES paper "M. 0. J. Hawksford, A family of circuit topologies for the Linkwitz-Riley (LR-4) crossover alignment, Audio Engineering Society preprint No. 2468 (82. Convention, 1987)
In 1995 the german magazine Elrad published a 2-way LS-project featuring a Xover using this schematics.
On demand I could mail both, the Hawksford and the Elektor article.

The basic idea behind this topology is to create a family of subtracting filters, where HP and LP follow each other with great precision.
This special topology allows for lowest component number count of the frequency decicive -low tolerance, high-cost- parts.
Mathematically they exploit the fact with filters of even order (2nd, 4th, etc.) the denominator of the transfer functions of Highpass and Lowpass are the same. The TFs only differ in the numerator.
The Lowpass function may be realized by an equal number of integrations (2x, 4x, etcx) of the Highpass function. This is why we find 2, 4 , etc. integrators in this topology.
(In Nebuks schem these are U1, U2, U3 and U6 together with the RC Networks R2/C4, R3/C3, R4/C2 and R5/C1).
Each integrator can be calculated after the RxC formula. This allows to optimize the R and C values for lownoise or after component tolerance.
As such the requirement for tight component tolerance is relaxed.
The RC-time constants (T=RxC) need to be of distinct (and mostly different) values to achieve the desired character, say B4, LR4, Bessel4, etc.
The output signals of the inverting integrators are summed up (U4, R1, R8-R15, beware of the inversion of the signal from each integrator) and at the same the difference of Input signal and summed-up integrator-signals is formed with U4. All other resistors are just used for gain settings and correct summation of signals. As such all these resistors may be of equal value, say 10kOhm-27kOhm. U5 is just a inverting Input buffer stage and not needed if the source has a low-impedance output.

Compare: A 2nd order Hawksford-Riley-Filter
Click the image to open in full size.
This is a 2nd order filter which may be expanded to a fourth-order filter by simply adding two integrator stages.
Here, U4 takes up the summing functions of the two integrator signals (U4 is also U4 in Nebuks schematics, so maybe You like to redraw it).
U1 takes the difference of the Input- and the Integrator-signals and forms the HP-output at the same. So the summing and differencing functions are split between U1 and U4. And U1 presents the signal an non-inverting Input.

While the filter worked well at frequencies of a couple of hundreds Hz on, I sometimes got oscillation effects with low crossover-frequnecies.
I finally dropped this filter, because it is an "ideal" filter, if the load is an ideal load like a resistor. It does not take into account the drivers own amplitude and phase-response. To achieve the required acoustic response each driver demands additional equalization, which turns the low part number count advantage of this filter to the opposite (The Elrad article also describes the design of such equalizer-filters) .
I therefore came back to design each filter specially after the requirements of its LS-driver, as it is usual practise with passive filters. This resulted in the end with superior results and lower parts number count.

jauu
Calvin

Last edited by Calvin; 16th April 2012 at 09:06 AM.
  Reply With Quote
Old 16th April 2012, 09:34 AM   #16
AndrewT is offline AndrewT  Scotland
diyAudio Member
 
Join Date: Jul 2004
Location: Scottish Borders
Quote:
Originally Posted by NebuK View Post
Capacitor tolerance values seem to be much much worse! I can't seem to find 3.3n caps at under 5% tolerance
you, as builder, must do the matching of components where matched components are required by the circuit.
I would aim for <1% on capacitors. Many caps come in 10% tolerance and worse.
__________________
regards Andrew T.
  Reply With Quote
Old 16th April 2012, 09:38 AM   #17
AndrewT is offline AndrewT  Scotland
diyAudio Member
 
Join Date: Jul 2004
Location: Scottish Borders
Calvin,
the subtractive filters that I have so far looked at, had a subtractive "error" in their operation. They did not give the equal slope in the low pass and high pass sections.

Does the Hawksford implementation overcome this problem?
__________________
regards Andrew T.
  Reply With Quote
Old 16th April 2012, 10:13 AM   #18
godfrey is online now godfrey  South Africa
diyAudio Member
 
Join Date: Nov 2009
Location: Cape Town
Hi NebuK

I'm not too familiar with LTSpice but I think if you want it to tell you the effects of component tolerance, you have to do a "Monte Carlo" analysis.

I tried changing the values manually. The first pic below shows the effect of increasing the 105K resistors one at a time by 20%. The second pic shows the effect of increasing all the other resistors one at a time by 20%.

It seems like a 20% change in any value can throw the response out by up to about 2dB, so I guess 1% tolerance => up to 0.1dB error, with the errors randomly accumulating. The variable resistor (the 6.75K one) should allow you to tune out the errors fairly effectively.

Re capacitor tolerance:
You can always use a different value of capacitor, and adjust the value of the 105K resistors accordingly. I'd be tempted to increase the capacitance and reduce the resistance, rather than the other way round, to keep the noise down. Having said that, RS components seems to have some suitable caps:
polystyrene capacitor,3300pF 63Vdc 1%
polyprop cap,3.3nF 63V 1%

Re the bump:
All my "pretty" curves were made with "ideal" opamps in the simulation. As soon as I substitute "real" opamps, things look worse. Replacing the opamp on the far right with a LT1013D or TL071 gives me a similar bump to yours. Replacing the opamp second from the right with a LT1013D or TL071 messes up the curves so badly, I think the simulator may be hallucinating.

For best accuracy, I would guess the most important features to look for when choosing the opamp are FET input and high gain.

Cheers - Godfrey
Attached Images
File Type: gif tol1.GIF (14.7 KB, 105 views)
File Type: gif tol2.GIF (14.9 KB, 105 views)

Last edited by godfrey; 16th April 2012 at 10:18 AM.
  Reply With Quote
Old 16th April 2012, 10:27 AM   #19
AndrewT is offline AndrewT  Scotland
diyAudio Member
 
Join Date: Jul 2004
Location: Scottish Borders
low pass active filtering demand adequate opamp performance.

The filter stops filtering if the opamp approaches it's bandwidth limit.
__________________
regards Andrew T.
  Reply With Quote
Old 16th April 2012, 10:35 AM   #20
godfrey is online now godfrey  South Africa
diyAudio Member
 
Join Date: Nov 2009
Location: Cape Town
Quote:
Originally Posted by AndrewT View Post
Calvin,
the subtractive filters that I have so far looked at, had a subtractive "error" in their operation. They did not give the equal slope in the low pass and high pass sections.

Does the Hawksford implementation overcome this problem?
I'm not Calvin, but...
This type of filter doesn't seem to suffer from that problem. I don't think it's correct to call it a "subtractive" filter (but I also don't want to get into an argument about semantics).

Looking at the outputs from the five opamps gives a good intuitive idea of what's going on. Essentially the output of a 2'nd order high pass filter is passed through four integrators to get the low-pass output.

The only problem is that at very low frequencies the integrators run out of gain, which is probably the cause of the trouble Calvin had.
Attached Images
File Type: gif family.GIF (14.8 KB, 20 views)
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
Adjustable Linkwitz Transform filter spot Subwoofers 10 24th October 2011 11:20 AM
Linkwitz-Riley slopes and crossovers Andres Silva Multi-Way 6 9th August 2011 03:16 AM
A Challenge to Linkwitz-Riley ? jackinnj Multi-Way 17 14th March 2006 11:50 PM
Linkwitz Riley Filter Implementation Roysyboy Subwoofers 1 19th September 2005 03:40 AM
linkwitz-riley crossover arthur Solid State 0 28th April 2003 10:26 AM


New To Site? Need Help?

All times are GMT. The time now is 02:05 AM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright 1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2