Jump to content


Photo
- - - - -

Crossover mods for the AR4x


  • Please log in to reply
94 replies to this topic

#1 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 09:02 PM

My local vintage audio dealer had a nice pair of AR4x's a few months back. Grilles and cabinets were in great shape and the drivers looked unmolested, but the tweeters were dead. Thanks to JKent and Zilch I now have new tweeters (from opposite coasts) and rather than put the system back together stock I want to see what I can do to improve the crossover network. Before all that lets take a look at the stock system to see what our $58 bought us in the late 60s.

This is my first run with the Holm measuring system mentioned in the kitchen. So far it is a very handy FFT (with MLS) measuring system. Better than the PC RTA system I've been using, it gives true phase and lets you time window in a number of ways. Very cool.

The first curve is the system at 3 angles, 30 degrees left, 0 degrees and 30 degrees right. This is from about 1 meter away with the grille off (grille on was surprisingly similar). I've seen other curves on the web and they are similar. Response is level from 50 Hz or so up to a "significant" bump at 1000 to 1500, followed by a 2kHz dip and finally a significant bump above 10k. I was hoping that the 2k hole was cabinet edge related because then some discrete damping under the grille could improve it. The fact that it doesn't change with lateral shift suggests it is inherent in the tweeter.

AR4_stock_net_30l_0_30r.jpg




This curve shows the individual parts, the woofer and tweeter. The woofer curve (orange) looks great and has a clean rolloff even with a simple network of inductor only. AR obviously worked hard to get a woofer with a smooth roll-off rather than the typical peak. For the tweeter the 2k dip is much more obvious here in the individual sections curve.

LF_vs_HF_w_phase.jpg

It's easy to see where the 1200Hz peak comes from. For 2/3 of an octave the woofer and tweeter both have full strength. Overlap between units is not a good thing.

The dashed curves are the phase curves for the two units (orange dashed = woofer, green dashed = tweeter). The actual phase curve of the units doesn't matter but the realative phase between the units is very important, especially in the crossover range where they both have about the same energy. The system response is the vector sum of the two sections so if there phase curves are close, their outputs add fully. If there phase curves approach 180 degrees apart then they will cancel rather than add.

In our case the curves are within 30 degrees of each other in the 1 to 2k region so we will get full addition, i.e. a bump. If we could somehow spread the phases we might getl less bump but that isn't a good solution because at other vertical listening angles they would come back into phase.

I'm guessing the original designers struggled with a compromise between the bump at 1200 and the dip at 2k. Pulling down 1200Hz would make the range above sag lower. This is compounded with a first order network (series cap) where the unit may roll off as intended, but at resonance where the driver impedance bumps up, the voltage after the cap bumps up as well. A higher order network would give more degrees of freedom hence better control of shape.



David

#2 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 09:20 PM

I ran across this curve on the Stereophile website. It illustrates the problem of a first order network on the tweeter. While first order (a series capacitor) should give a 6dB per octave electrical rolloff, this only works well when the load is a resistor. When the load is a tweeter and its realworld impedance curve, then things won't be so tidy. At the tweeter's resonance its impedance will go quite high. This is a "light load" on the network and the voltage will rise also (in the graph the bump at 500Hz). At frequencies above that, the tweeter impedance will drop to near its DC resistance and voltage at the tweeter terminals will fall. Finally, at highest frequencies the inductance will bring the tweeter impedance up again and the voltage will rise again.

One solution for this would be a Ferrofluid of the heavy damping type. It would drop the impedance rise at resonance (lower Qm) and give a more classic highpass.

For these reasons text book crossovers are never very successful. A software based approach must always bring driver impedance into the calculations as the load on any proposed network.

first_order.jpg



Anyhow, this shows the individual curves again plotted with the combined response of the whole system. The dark curve is the system response, the combination of the green and magenta woofer and tweeter.

Stock_net_vs._individuals.jpg

Again you can see that where the individual curves overlapped excessively the summed response shows the midrange peak.

David

#3 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 09:37 PM

I thought I would try reversing the tweeter polarity to see what the sum is like. Note that the stock network has the tweeter out of phase with the woofer. Best phasing in a system depends on a lot of variables including woofer depth, crossover order, crossover frequency, etc. you can't generalize and generally must use the phasing that gives the best addition.

As I mentioned it isn't good to achieve flat response via cancelation but it was worth a look. Here it might be worth listening to as it knocks about 7dB out of the peak and leaves a less offensinve trough around 2k.

20uf_twt_cap_norm___rev_phase.jpg




Here is a look at the effect of the grille. It is surprisingly minor with just a little bit of loss at 5k and 8k with the grille on. This indicates that the standard cloth is very transparent. Also that reflections off the side of the grille are minor. I suspect that the tweeter being fairly directional helps reduce the side reflections.

Grille_off_grille_on.jpg

David

#4 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 09:41 PM

The stock network for an AR4x is 20uF on the tweeter, followed by the wirewound pots. The pot has the usual problems so I've bypassed it and plan to use fixed resistors for treble level. When I connected an external substitute for the internal network I noticed significant differences between the measured response of the internal 20uF and an external 20uF. Since the cap is 40 or so years old this isn't too surprising. Note that the capacitance seems to be down (loss at the low end of the tweeter) and there is loss up to 10k also. This would have to be due to higher ESR (effective series resistance).

I'd note that many who restore these old speakers rush to replace all the caps. In this case a new cap will make a significant change but you have to wonder if it really is an improvement? If the pot is working you can have whatever treble level you like, but the replacement cap will raise the 1200 Hz bump, arguably not an improvement. (something to think about).

external_vs_internal_network.jpg



Here I played with the external network values until I got a reasonable fit to the response of the internal part. 14uF and 2 ohms got the two curves to overlap so I have to assume that is the value, due to aging, of the AR unit.

best_fit_to_internal_cap.jpg

David

#5 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 09:53 PM

If you wanted to stick with a first order network, this was a little better than stock. It is 10uF and reversed phase (reversed from as sold, but actually in-phase with the woofer). The overlap between drivers was less and and the 1200 Hz bump is gone but there is still a broad depression above. We can do better.

Next time: a second order network.

David

Attached Thumbnails

  • Stock_net_vs._10uF_and_polarity_flip.JPG


#6 Pete B

Pete B

    Advanced Member

  • Members
  • PipPipPip
  • 1,557 posts
  • Gender:Male
  • Location:Connecticut, NE USA
  • Interests:Interest in Audio for over 30 years, got started early
    Electrical Engineer (see my web site link)

Posted 14 February 2010 - 11:16 PM

Nice work David!
I see your measurements are full bandwidth, what method
did you use for LF? Did you splice a nearfield measurement,
if so at what frequency?

#7 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 14 February 2010 - 11:30 PM

Nice work David!
I see your measurements are full bandwidth, what method
did you use for LF? Did you splice a nearfield measurement,
if so at what frequency?


They are full bandwidth unspliced. Holm seems to give you either the choice of time windowing or frequency smoothing, not both. I would think that time windowing with a little smoothing would be useful but it isn't an option. At 1 meter away, a 6th octave full range curve looked pretty reasonable. I also use the highpass function liberally to clean up hash in the tweeter curves.

I think you've been working with this longer. Is there a way to splice with Holm?

David

#8 Pete B

Pete B

    Advanced Member

  • Members
  • PipPipPip
  • 1,557 posts
  • Gender:Male
  • Location:Connecticut, NE USA
  • Interests:Interest in Audio for over 30 years, got started early
    Electrical Engineer (see my web site link)

Posted 15 February 2010 - 12:44 AM

They are full bandwidth unspliced. Holm seems to give you either the choice of time windowing or frequency smoothing, not both. I would think that time windowing with a little smoothing would be useful but it isn't an option. At 1 meter away, a 6th octave full range curve looked pretty reasonable. I also use the highpass function liberally to clean up hash in the tweeter curves.

I think you've been working with this longer. Is there a way to splice with Holm?

David


Most of my experience is with LAUD which I run on an old PC
because it requires a particular sound card in an ISA slot.

I'm setting up a Laptop mainly for in room measurements,
with RTA software and this has led to taking a look at a few
programs. I like the free RTA feature in Praxis best so far.
Room EQ Wizard is also decent. I did take a look at ARTA,
which is a full MLS type system and I might use this to
replace LAUD once I use it enough since so far I've only
had a quick look at it.
Ken and many others seem to like Holm. Have you seen
the threads at DIY Audio?

#9 Carlspeak

Carlspeak

    Forum Moderator

  • Administrators
  • PipPipPip
  • 1,877 posts
  • Gender:Male
  • Location:North Central Connecticut

Posted 15 February 2010 - 03:08 PM

This shows the individual curves again plotted with the combined response of the whole system.

Again you can see where the individual curves overlapped excessivley the summed response shows the midrange peak.

David


Dave:
Have you tried NF measurements on the woofer? I've generally found with a number of AR4x's there is a hump in the response typical of a high AS box Q which I've measured in the 1 to 1.5 range. IMO, it was put there by design to give those little guys a bass boost which many restorer's seem to like.
IT'S ALL ABOUT THE MUSIC!

Carl
Carl's Custom Loudspeakers

#10 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 15 February 2010 - 07:53 PM

Dave:
Have you tried NF measurements on the woofer? I've generally found with a number of AR4x's there is a hump in the response typical of a high AS box Q which I've measured in the 1 to 1.5 range. IMO, it was put there by design to give those little guys a bass boost which many restorer's seem to like.

Hi Carl,

I'll add a nearfield curve. The first box seems to leak considerably and I'll have to track that down. Unless that's just the way it is with an open dustcap?

Regards,
David

#11 Zilch

Zilch

    DECEASED

  • Members
  • PipPipPip
  • 1,056 posts
  • Gender:Male
  • Location:Berkeley, CA

Posted 16 February 2010 - 12:23 AM

The first box seems to leak considerably and I'll have to track that down. Unless that's just the way it is with an open dustcap?

If it's like the ones I have here, the voice coil is essentially exposed in back but for some masking tape. With no dust cap, I'd say that's a leak path.

Play 30 Hz and with a length of tubing in your ear, listen for the "motorboat" at potential leak locations.

[OR, as a professional, a stethoscope with the bell removed.... ;) ]

ZilchLab Loudspeaker Recycling Center - Crank up the SIGNAL!


#12 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 17 February 2010 - 03:56 AM

Here are some looks at a second order (12dB/octave) network for the tweeter section.

Remember that the tweeter response is presenting some problems. It is a little underdamped giving a significant bump at 1200 Hz. A little higher at 2kHz there is a significant hole. With a simple network you have to find some type of compromise between too much 1200Hz energy, leading to a "midrange snarl" type of balance, or a hole at 2k. A higher order network will allow a squarer electrical corner that might give a better combined shape.

Below is a comparison between a first order and second order network. The green curve is a 14uf (1st order) network similar to the aged cap in my system. The red curve is a second order network of about the right values to match the woofer. Note that I was able to bend the 1200Hz response down considerably while actually gaining a little at 2500. Not enough to fill in the 2k hole but at least moving in the right direction.

The dashed curves show the effect of a higher order network on phase. Phase at the bottom corner of the tweeter bends upwards with the higher order network. Generally you should get 90 degrees more ultimate phase shift below the corner and about 45 at the corner for every order you add (1st to 2nd, 2nd to third, etc.). Because the rolloff corner is shifting upwards as well, we have about 60 degrees shift relative to the capacitor only network.

Whats the right phase curve? Depends on what the woofer is doing.

Attached Thumbnails

  • 1st_vs_2nd_ord_tw_net.jpg


#13 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 17 February 2010 - 04:15 AM

Here are the same two tweeter curves compared to the woofer. The orange phase curve for the woofer is below the green phase curve of the first order tweeter network. What we would really like to see is the phase curves being close over the range where both units (woofer and tweeter) have significant output. Say 700 to 3000Hz. Beyond that one unit or the other has rolled off to the point where its phase doesn't matter. Orange and green curves are about 30 degrees apart so those sections would add fairly well in phase. Changing to the second order network pushes the phases apart to the point where they average about 90 degrees.

Thats about the worst case possible. Flipping the phase of the tweeter won't help, due to the 180 phase reversal it would just become 90 degrees apart in the other direction. Note that these phase curves only apply for the exact microphone position they were measured at, in this case half way between the woofer and tweeter and about 1 meter out. If you move the microphone up from that position then the woofer phase will bend down and the tweeter phase up. Move the mic down and vice versa. The 90 degree phase discrepency would therefore cause unsymetrical lobing.

The solution? Maybe a third order network would get the units near 180 degrees apart and then a polarity flip would get them quite close (tune in to part 3).

Attached Thumbnails

  • 1st_vs_2nd_ord_tw_net___woof.jpg


#14 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 17 February 2010 - 04:26 AM

Here was about the best I could get with a 2nd order tweeter network. It used a 22uF input capacitor with a .274mH inductor to ground and finally a 4 ohm resistor to pull down the top end. The exactness of those values isn't important. It's just what I had on hand.

So that is the hot tip if you want to try a simple improvement to your AR4x: add an inductor from between the 20 uF capacitor and the level pot, to ground. Use any inductor between .25 and .3mH and you should get reasonably close to the curve shown.

Note that the bump at the top of the woofer range is gone and the 2k hole is a little better, but by no means dead flat. I'm really starting to think that a third order network is what is called for.

David

Attached Thumbnails

  • Best_2nd_order_net.jpg


#15 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 17 February 2010 - 04:38 AM

Just a reminder, here is the 2nd order network vs. the stock network. The 1200Hz bump is about 5dB better than before and the midrange prominance is gone (and this is with a diminished value capacitor, a new 20uf cap would make it a couple of dB worse). The phase curves show that the tweeter polarity is the same as the stock network (otherwise they would jog 180 degrees above 1kHz, rather than the two curves tracking together).

David

Attached Thumbnails

  • Best_2nd_vs_stock.jpg


#16 genek

genek

    Advanced Member

  • Administrators
  • PipPipPip
  • 1,604 posts
  • Gender:Male
  • Location:San Marcos, CA

Posted 18 February 2010 - 02:51 AM

Dave, any thoughts on whether the AR-6 with its different tweeter and crossover and the 4xa that used the same new parts as the 6 was AR's fix for these issues? Or is there no way to tell without getting a 6 and a 4xa in front of some instrumentation?

#17 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 18 February 2010 - 04:36 PM

Dave, any thoughts on whether the AR-6 with its different tweeter and crossover and the 4xa that used the same new parts as the 6 was AR's fix for these issues? Or is there no way to tell without getting a 6 and a 4xa in front of some instrumentation?


Hi Gene,

Good questions. I would need to measure, or see measurements, of an AR4xa to see how it was really improved.

Putting myself in the shoes of an engineer there (and then) I would be hoping to improve the top octave (bump at 12k) and the broad 2kHz dip in any upgrade to the AR4x. Knowing that dispersion was a key element of AR's philosophy, then the smaller size of the AR4xa tweeter, and subsequent wider dispersion, was probably the main incentive for the change, but I can only speculate here.

On the 4xa, a brochure in the library AR says; "incorporates a new tweeter design and a modified crossover network. This new tweeter provides improved high frequency response and dispersion." They also mention power handling which we know is an issue and might already have been revealed to them as a problem. Part of the power handling problem is the large value (20 uF) 1st order crossover.

I'm still looking for the cause of the 2k dip. I think it is purely cabinet related but can't find a time path that corresponds to it. That is, it doesn't correspond to the reflection time from either side edge or the top edge, might be a double bounce? Of course, if it is due to the cabinet then the 4xa would have a similar problem.

Interesting stuff,
David

#18 genek

genek

    Advanced Member

  • Administrators
  • PipPipPip
  • 1,604 posts
  • Gender:Male
  • Location:San Marcos, CA

Posted 18 February 2010 - 06:02 PM

Good questions. I would need to measure, or see measurements, of an AR4xa to see how it was really improved.

I've never seen or heard the 4xa, but I do have a pair of 6's. I was wondering if any guesses could be made about the 6's based on the crossover schematic and the FR curve, which I have from an old High Fidelity article, but the 6's don't use the same cabinet as the 4's, so maybe not.

#19 rl1856

rl1856

    Advanced Member

  • Members
  • PipPipPip
  • 69 posts

Posted 18 February 2010 - 07:15 PM

Dave, you have achieved something interesting and of course useful.

Through your experiments you have created a 40-45yr old speaker that is +/-5db from ~45hz-19khz. Preatty heady stuff that allows the inherant brilliance of the original AR design to shine through.

Next test should be a listening comparison between your modded 4 and a stock 4. One concern is whether to use a stock 4 as found or use a 4 in which the caps have been verified to be operating within spec. Your measurements pointed out the response differences that may be attributed to long term cap deteoration. Looked at another way- is what we hear now the same as what the designer intended and the first buyers experienced ?

Thanks for the hard work !

Best,

Ross

#20 speaker dave

speaker dave

    Advanced Member

  • Members
  • PipPipPip
  • 468 posts
  • Gender:Male
  • Location:The Mountain
  • Interests:Vintage audio, psychoacoustics, bicycling, shortwave radio, photography, record collecting (mostly classical piano).

Posted 18 February 2010 - 07:32 PM

I've never seen or heard the 4xa, but I do have a pair of 6's. I was wondering if any guesses could be made about the 6's based on the crossover schematic and the FR curve, which I have from an old High Fidelity article, but the 6's don't use the same cabinet as the 4's, so maybe not.


There was a time when High Fidelity was using the CBS test chamber and running axial, hemispherical and spherical curves. Those were pretty revealing and would at least tell how the tweeter performed in terms of better dispersion. They would also show whether the bump above 10k was improved, although I would consider that a minor fault (especially with 54 year old ears!) Do you have a way of scanning and attaching?

David




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users