Mark K's Speaker Pages
...when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely in your thoughts advanced to the state of Science...Lord Kelvin
Current Project-RST 2 way dipole
NOTE-this is a chronological blog, with the earliest entry at the bottom. The most sensible way to read this would be to start at the earliest date entry at the bottom, and work your way up.
PS-if you haven't noticed I tend to post updates on Wednesdays...
6/18/08
What was I saying last week?
In last week's preliminary measurements I was worried about suppressing the woofer breakup adequately.
But, it bothered me that the lspCAD simulation didn't really agree with the measurement all that well. So, let's take a look at the actual measured electrical transfer function transfer function with the loudspeaker attached. That is, instead of measuring acoustic output, I hooked up a probe at the passive crossover terminal output and looked at the electrical attenuation. The result is posted below along with the original outdoor, high quality measurement below it (with the active dipole equalization, but NOT the passive lowpass filtering at 1.4kHz).
The woofer is in black and the tweeter is in red.
If you look at the woofer acoustic curve first you'll see a broad breakup peak between 3.5kHz extending out past 10kHz. It peaks at marker 3, at 6.6kHz. Still, it's a low amplitude peak, being -1.5dB below the bandpass sensitivity.
Now, if you look at the electrical transfer function, the filter is down -47dB at 6.6kHz. So, the net effect should be roughly that the peak at 6.6kHz should be down -48.5dB below the bandpass sensitivity. However, my measurement last week showed the woofer to be down only ~35dB in the 5-10kHz range. I've reposted the salient curve below.
What gives? That extra 13dB could be important. I'm comfortable at ~45dB or lower for cone breakup, since that would be 0.5% or less. 30dB would be around 3%. Audible? I didn't hear any but still would like that 10-15dB cushion.
Well, the electrical transfer measurement is very accurate. There is no acoustic component to introduce error and the window is very long. The outdoor curve shown below the electrical curve should also be a better curve, being outside, with a much longer window, and the unit raised high in the air.
The aqua woofer curve was done with the panel in my garage. This means a shorter window, and the reality is that the panel is ~48" long in a garage with a ceiling height of 8 feet-this does not make for very accurate measurements. OK for crude, ballpark measurements, but not so good for low level details.
So, I'm willing to bet that the aqua curve is just not very good. I haven't given up on the relatively simple and straightforward crossover just yet and don't want to make it any more complex than necessary. I've made some minor changes to this simple 4rth order crossover, but I'm trying to avoid a notched filter or another trap if it's not necessary.
Anyway, the only way to settle this is to hook up the current active/passive crossover setup outdoors in the next month and measure it right.
The moral is-work really hard on getting high quality outdoor measurements!
It will take a bit of time to get those measurements. I'd like to spend a little time improving the construction quality of my active and passive crossover first. Ahem, see below.
PS-It took a lot of work in Photoshop just to get it to look this good!!!!
6/11/08
How much suppression is enough?
Below you'll see a 4th order LR crossover designed in lspCAD.
Below you'll find the crossover schematic.
This is more or less the same as the crossover below.
Now, let's look at the measured result of the active inline equalization and the passive crossover. The CD source is a Philips 963 and the amplifier is an Outlaw Audio RR2150. The active equalization just goes in the preamp out-amp in loop.
Pretty good for a first build iteration. This is done in my garage with a fairly short window, so the resolution is poor below 1k.
Off axis is below. Dipole below the 1.4k crossover, then a typical 1" dome response above.
How about an in-room RTA? Looks pretty good for its limited value.
Impedance of the passive crossover and the active equalization curve are below for verification. Everything checks out.
Now, how about some distortion measurements? I apologize for not having comparison graphs with my RS22528A. However, I did do the measurements and you'll just have to trust me the 5 tone pattern was better, the single tone was comparable, and the SL style 3 tone was somewhat worse. The moral seems to be that the distortion of the 4 driver open baffle is equivalent to a single sealed RS225. I did not do other distortion tests (this is preliminary after all!) Again, my guess is that 100-1k distortion will be markedly better.
So, everything is perfect. We can stop now. Well, not quite. Below you'll see a graph of individual driver response. The woofer breakup is not suppressed as much as I would like. Only ~35dB down between 5-10k.
So, how does it sound? You do know I listen, right...? Well, it sounds excellent. I had the RS22528A as a similar monopole reference. I verified and learned a couple of key points.
First, the active equalization circuit works well. No evidence of audible distortion. I also ran windowed bursts through the equalization circuit to make sure. Though not shown, these were fine. So, objective evidence agrees with subjective opinion, as it should.
The panel is clearly capable of satisfying full range behavior, although the last octave is not present. The active highpass filter at 45-50Hz keeps audible LF distortion down.
The dipole panels clearly sound different in their presentation, especially in the midrange and the bass region, as you would expect. Very natural sounding. Hard to hear any clear woofer breakup, though it's hard not to look at the above graph and not worry a little.
It's hard to make a lot of comments on the sound, since I'm only listening to one unit. (mono)
Next up will be a version with an additional notch filter, and a notched filter design. So far so good. I'm very impressed with the performance so far.
4/09/08
Ah, lspCAD
Well, I broke my SE dongle and Bohdan is giving me a hard time. ("I need to see proof of ownership!") Jeez, it's not like there are that many copies of SE. Let's see, how many upgrades have I bought, as well as new usb dongle. (I've got a parallel port dongle-but this pc doesn't have a parallel port. I know, buy a pci card-but they're finicky and I can't use my laptop!) Whine, whine, whine...
Anyway, lspCAD pro to the rescue. It's always worked before. Below is prototype passive midwoofer-tweeter fourth order LR crossover circuit with PE coil and cap values (with the active inline equalization present).
Don't worry about those dips at 2.5k and 3.5k. That's mostly diffraction and it's still mostly +/-1dB, with a slight droop at the high end. What about those traps? Well, it just makes it so much easier to get a good knee on the curve and let the rest of the network handle the asymptote. The woofer trap also helps the breakup which is 50+ dB down at 6k. Unlike others, I don't object to extra parts.
As an interesting note, the impedance rises inversely with frequency. Although the voltage gain is still ~14dB, the power required is less, more like 9dB, since the voltage will be into ~12 ohms at 50 Hz instead of 4 ohms at 1k. Overall it should be an easy load for any modern, decent 100 watt amp.
This is just the starting point for emulation. lspCAD pro emulates very nicely. Meantime, I'll order one of those Rosewill $13 PCI parallel cards from Newegg and hope Bohdan's old parallel SE dongle works with it...
4/2/08
The winds have calmed.
The winds have calmed and I was able to do some measurements. The graph above is a little busy, but it's worth taking a look at.
-The black curve is the same curve as in the bottom entry for 3/12. Its the measured, unequalized curve of the four RS225's (again, in a series parallel connection, with a 1.25mH coil on the outer pair to roll them off earlier and improve vertical lob
ing) multiplied by the measured transfer function.
-The red curve is an outdoors 3 meter, 8ms(true 8ms-total interval ~16-17ms with ~8ms time of flight) measurement of the dipole 4 RS225 drivers . As you can see, the black curve and the red curve are in excellent agreement, as they should be. It's nice to know the equalization circuit works as it should.
-The aqua curve is the Seas 27TBFC/G on the baffle.
-The blue curve is a Seas DXT horn. I tossed this on the baffle to see how it would work. It has more sensitivity in the 1-10k region, but would be difficult to get flat after 10k. Still, both are probably usable and I'll play around with both when generating a passive crossover.
For the sake of improved clarity, the curve below shows only the final measured woofer and tweeter curves that will be used for crossover modeling.
I will merge a nearfield curve into the woofer data above. (Actually, I will do it bother ways. I'll leave it as is for lspCAD and then generate a spliced version for SE. They should be very similar.) I'll work on the passive crossover in April and finally have a listenable single unit by late May.
A notable interesting point is that the tweeter will require no padding and the system sensitivity will be on the order 91-92 dB. This is a direct result of the increase in efficiency from both the MTM format and the extra 0.5 woofers, so to speak. You get 6dB from the TMM arrangement, then another 6dB from the MTM aspect. Not bad, eh? Now, it's not really 12 dB since the series arrangement nets extra excursion but does not raise the system sensitivity.
Also, while the SPL for 2.83 volts at one meter may be 91 dB at 1k, it will not be that at 50 Hz. Because of the gain circuit, it will require 14 dB more power, i.e. the amp will have to put out ~32 watts or so to maintain the same spl as one watt (assuming 8 ohms) at 1k. Hope that makes sense. Realistically it will take a 65-125 watt amp to drive this well into the 95-100dB range. With both, and room effects, I'm guessing that 100-105dB honest output (though 105dB at 50 Hz dipole is, well, stretching it...)
3/12/08
More Praxis gymnastics
Here's a graph of the true dipole measurement of the RS225's multiplied by the transfer function of the actual measured equalization circuit shown in black. (Praxis is nifty in that you can do math operations on curves in both the time and frequency domain. So you can take a raw measurement, and multiple it by some other curve. In this case, the equalization curve shown below. Very handy for simulating results.) The tweeter is in red.
If that all seems a bit confusing, hopefully this next, somewhat busy graph will clear things up. What I did was multiply the navy blue true measured response of the dipole panel by the true measured equalization circuit, shown in aqua and correct the gain. The result is the black curve. It's not quite the same as measuring the the panel with the equalization in place, but pretty close.
NOTE- the final black curve is not correct below ~85Hz since Praxis has no data below 85Hz in the navy curve. I can generate a more accurate curve below 85Hz, but due to technical reason, I lose sensitivity link with the tweeter. What I was most interested in these curves was making sure the sensitivity between the woofer and tweeter was ok.
So you really want to see a more accurate simulated low end? The true roll off is steeper as you can see below. However, I can't compare the relative tweeter level because of the way I post processed this graph. But it's the correct simulated low end, for what it's worth.
Could I use this to model a crossover? Probably. But the level is very critical. I don't have any extra sensitivity on the tweeter. If the gain on the woofer transfer function is off, even a dB or two-that's going to be a problem. I did this though, to see if the sensitivities would work out. It looks just perfect. The tweeter won't need much, if any padding at all.
So it's looking very good. Now, to just wait for a free Wednesday morning that's not windy so I can measure the panel with the equalization circuit.
I may also just increase the depth of the notch 0.5dB or so by tinkering with the circuit. If only I didn't break my usb SE dongle. I've got a parallel dongle, but no parallel port. Still, a parallel port adapter from Newegg is ~$15. A lot cheaper than a new usb dongle.
3/10/08
It's Alive!!!!!
Yes, yes, it looks messy. But, it works. I fired up the circuit using 18g thermostat wire for the power supply and 22g shielded HD security camera cable. You gotta love HD. Sure, the electrons would prefer driving in a Mercedes Benz. But they're ok with driving through budget wire for prototyping. That power supply is one of the common generic +/-15v ones available through Digikey for ~$45. Hard to make one for that.
Anyway, it works. Below you'll see the result using praxis to test the transfer function. This is the "true" measured result, NOT modeled. Compare this with the Soundeasy/John k spreadsheet simulated result at the bottom of this page. Seems to be accurate within 0.5dB.
Praxis will use a chirp up to 1.3 seconds long, so the resolution is better than 1Hz. The op amps held up well since I bent the pins, dropped them, rubbed them against my cat's fur, touched the Van de Graff generator, and they still seem to be work when I finally got them in.
Praxis makes the marker interpretation quite easy. You can set any marker as a reference. In this case, I set the 10k marker as zero. So there is a 4.4 dB dip for the dipole peak, a total of 14.6dB of dipole gain peaking between 40-50Hz, and then the 12dB roll off (actually, heading towards 18 with the 7 Hz input filter.
3/01/08
They're here
ExpressPCB has sent the boards. Prototype boards are 3/$51 plus shipping. Plus, handy software for board design. Perhaps not the cheapest, but pretty no nonsense for small board prototyping. You have to be able to fit it on a 2.5" by 3.8" board.
Here it is with the components. Only one channel for now. After all, what if it stinks? You can already see some eh, resistor changes. And, c19 and c0 are too close. I had to use a different brand film cap (the green ones) to fit. Live and learn. I figure it will go through a couple of iterations anyway before the final board. This is a small board. Good thing I'm nearsighted without my contacts. Care and a very small point soldering iron are required.
I may try to make a more generic board that would work for multiple active circuits, add an onboard power supply, etc. It depends on the final interest in this project. If I can get 10-20 folks to fork out $20, then we can get a pretty nice board. You can see from the expressPCB site that a nice, silkscreen/soldermask setup will be ~$350-400 for 20 boards. It's not worth it just to make 5 boards...
Now, I need some free time to measure the main panel woofer and tweeter with the inline equalization circuit above.
Incidentally, Context Engineering makes a pretty nice case for 2.5" boards. You can get these at Fry's for ~$10. The case is below the board in the photo above. Of course, you've got to think where you're going to put the +/-15v power supply. The nice thing about this case is that the board slides into machined grooves in the case and doesn't need any other mounting. It's a pretty well finished case for the money. Not like the cheap stamped aluminum project cases out there.
Here's a better picture (credit the gallery of context engineering) of the case style. Of course it's black and not hot purple-pink...
2/22/08
Genesis
It's time for a modestly priced full range dipole. Already, there are a number of high quality 3 way dipole plans available. What's not generally available is a dipole offering high quality, full range behavior, in a 2 way format, that's moderately priced and only requires 2 channels of amplification.
Of course, it depends on how you want to define it. I'm defining full range as true 50 Hz frequency extension, a two way as a WWtWW 2.5 way, and only two channels of amplification but inline low level active equalization (i.e. you need a preamp out-amp in loop or separates). And, of course moderately priced is ~$600 not counting the wood for the baffles. Still, this is cheaper than say, a typical 7" box two way with say, some Scan Speak drivers. And, as you'll see, at least in measured distortion performance, will significantly exceed a typical 7-8" boxed two way.
Why not a more typical 3 way with active equalization? Well, I think SL and John k have this covered. No reason to reinvent the wheel. What I've learned from my design work on the RS22528A has led me to believe that 4 of the RS225 drivers can give significant low end response, adequate for a full range dipole, and have just enough reach to match with the Seas 27TBFC/G. Why not the RS28A? Well, in some respects the RS28A is a better driver, but the Seas can more consistently reach 1.4-1.5k.
I've already been working on this for a good 6 months, and below is a review of the design process I've gone through. Hopefully this will give you a better idea of the project and how I've approached a dipole design.
The initial goal is to get at least the same level of distortion performance in a dipole at 50 Hz that I currently get out of the RS22528A. At least as a rough estimate, the fequal is approximately 200 Hz and to get to 50 Hz, an additional 12 dB of excursion is required. So, four RS225's in a dipole will give the same SPL for the same excursion as a single sealed RS225. Step one is to build a prototype baffle and do some preliminary distortion measurements and FR measurements.
There are four issues to consider at this point:
-First, will the measured distortion of 4 drivers on a baffle be acceptable?
-Second, will the on axis frequency response be relatively easy to equalize?
-Third, will the dipole behavior go high enough to cross at ~1.5kHz?
-Fourth, how to minimize the effects of the narrowing of vertical polar response with 4 drivers?
Above you'll see a prototype baffle. It's rough and ugly, but it is enough to answer these questions. We put it up and do some simple FR and distortion measurements.
For the FR curves, I looked at the inner pair, outer pair, all four, and the inner pair in series, then paralleled with the outer pair in series. Different inductors were placed on the outer pair to roll off the outer pair and improve vertical polar response.
First, all four without an inductor. Very preliminary curves looking at measurement distance, noise etc. This first one is a 50mS curve that's smoothed. Outdoors this works very well as the red is a 6.5mS window and the black is a 50mS smoothed curve. The agreement is excellent and allows a pretty good estimate of what's happening under 150Hz. You see the 6dB roll off and dipole peak. Still it looks very equalizable.
Next, I looked at the different curves for rolling off the outer two. I didn't consider these measurements very good, but they were a starting point.
Here I overlaid the Seas 27TBFC/G response on the raw woofer response. The sensitivity match is very close. This may be a problem later. Once I actively equalize the curve, hopefully the tweeter will be sensitive enough. This can be adjusted somewhat actively, but it ends up being trial and error through a couple of iterations to make sure it's correct.
Looking at the graphs, you might think that deep null at 3k is a bit puzzling. It's the confluence of a couple of issues. First, the RS225 does have a notch naturally there. Below is a curve from my RS22528A raw measurement files.
However, this is made worse by the 4 woofer configuration. In the measurements above, all four were connected without a roll off inductor on the outer pair.
And then, there is the issue of measurement distance. Below, you'll find an interesting set of simulation curves for the 4 driver setup. This shows what the experimental FR curve would be at 1m vs. 4m. The 1m curve looks suspiciously similar to the curves above. What this tells you is that 1m is an unacceptable measurement distance for this setup. I've also simulated 2 and 3m and ideally, 3m or greater is necessary for accurate measurements. Now it's not as bad as the edge would suggest, as a roll off inductor for the outer two woofers will improve things considerably. Still, this tells me that 3m+ is a necessary measurement distance. Any less will introduce significant errors in the crossover design between the woofers and tweeter.
On to distortion measurements. I compared SL style tone bursts between my RS22528A and the RST dipole.
Here, the RST dipole is better at 30Hz, despite being driven at 3dB higher. (Sorry about that-hit the max output on the preamp for the RS22528A and couldn't get to 84.6dB. No matter, the result is clear.)
Here, at 50Hz, the dipole again is in the lead.
A smaller lead for the dipole, but a lead nonetheless in all important higher order region.
At 100 Hz and 250 Hz, the dipole distortion is markedly lower.
At 800 Hz, I'd give the edge to the dipole because of its superior higher order distortion products. However, you can see a slight excess of 2nd order products for the dipole, as shown in the 800 Hz reverse graph. So it's close.
Same data, with the two units flipped. Now the excess red is excess RST distortion.
What's the conclusion? The RST dipole is hands down superior to the RS22528A in distortion, both in low end excursion, and in higher order products. And, the RS22528A is no slouch, if you've read the data on my RS22528A.
So this looks very promising.
Anyway, it's time to build a new baffle. (More measurements were done, but you get the idea.)
A new baffle is born
That first prototype baffle was pretty flimsy. So, time to build a new baffle. A little bit of simulation, a little bit of studying John k's and SL's site and some emails to John, then it's off to build a new baffle.
The baffle is constructed quite simply as follows.
A set of 1x2x6' maple is laminated to a 1x3x6' strip of maple and a 45 degree miter is cut as shown below.
Then, a main panel is cut out of mdf as below.
Next, the struts are rounded over and the main panel is glued into the struts.
Add some Krylon faux finish and there you have it. The maple laminate makes it quite sturdy, surprisingly so.
Measurements revisited
With the new baffle, it's time to redo the measurements at a distance of 2, 3 and 4m. The drivers were measured outdoors, with the inner pair of woofers in series, then paralleled with the outer pair using a 1.25mH coil. Yes, that's a series parallel connection. However, since the drivers are well matched and the drivers are rolled off before the impedance peaks, the series parallel connection is only marginally less desirable than the alternative. Unfortunately, a parallel-series connection doesn't allow for easy passive roll off of the outer pair.
First, a s nearfield estimate. Below you see a nearfield 4 driver average. The red curve is an estimated dipole response using the method documented on John k's website. The estimated dipole response is spliced to the 3m farfield gated curve.
Here we have the farfield 3m measurement above 150Hz spliced to the dipole estimate. This gets imported into John K's spreadsheets, then the circuit is modified and entered into SE for some fine tuning.
Below you see the results. The raw spliced curve is shown(yellow marker), then the equalized curve with the modeled inline filter(green marker). Also shown is the transfer function curve of the inline active equalization circuit(purple marker).
PCB
The next step is to design a pc board. Yes, you can just prototype it on a breadboard, but the cost of a prototype board is fairly inexpensive if you get just the basic board. The prototype board is shown below.
Now I'm just waiting on the prototype boards to show.
NOTE-this is a chronological blog, with the earliest entry at the bottom. The most sensible way to read this would be to start at the earliest date entry at the bottom, and work your way up.
PS-if you haven't noticed I tend to post updates on Wednesdays...
6/18/08
What was I saying last week?
In last week's preliminary measurements I was worried about suppressing the woofer breakup adequately.
But, it bothered me that the lspCAD simulation didn't really agree with the measurement all that well. So, let's take a look at the actual measured electrical transfer function transfer function with the loudspeaker attached. That is, instead of measuring acoustic output, I hooked up a probe at the passive crossover terminal output and looked at the electrical attenuation. The result is posted below along with the original outdoor, high quality measurement below it (with the active dipole equalization, but NOT the passive lowpass filtering at 1.4kHz).
The woofer is in black and the tweeter is in red.
If you look at the woofer acoustic curve first you'll see a broad breakup peak between 3.5kHz extending out past 10kHz. It peaks at marker 3, at 6.6kHz. Still, it's a low amplitude peak, being -1.5dB below the bandpass sensitivity.
Now, if you look at the electrical transfer function, the filter is down -47dB at 6.6kHz. So, the net effect should be roughly that the peak at 6.6kHz should be down -48.5dB below the bandpass sensitivity. However, my measurement last week showed the woofer to be down only ~35dB in the 5-10kHz range. I've reposted the salient curve below.
What gives? That extra 13dB could be important. I'm comfortable at ~45dB or lower for cone breakup, since that would be 0.5% or less. 30dB would be around 3%. Audible? I didn't hear any but still would like that 10-15dB cushion.
Well, the electrical transfer measurement is very accurate. There is no acoustic component to introduce error and the window is very long. The outdoor curve shown below the electrical curve should also be a better curve, being outside, with a much longer window, and the unit raised high in the air.
The aqua woofer curve was done with the panel in my garage. This means a shorter window, and the reality is that the panel is ~48" long in a garage with a ceiling height of 8 feet-this does not make for very accurate measurements. OK for crude, ballpark measurements, but not so good for low level details.
So, I'm willing to bet that the aqua curve is just not very good. I haven't given up on the relatively simple and straightforward crossover just yet and don't want to make it any more complex than necessary. I've made some minor changes to this simple 4rth order crossover, but I'm trying to avoid a notched filter or another trap if it's not necessary.
Anyway, the only way to settle this is to hook up the current active/passive crossover setup outdoors in the next month and measure it right.
The moral is-work really hard on getting high quality outdoor measurements!
It will take a bit of time to get those measurements. I'd like to spend a little time improving the construction quality of my active and passive crossover first. Ahem, see below.
PS-It took a lot of work in Photoshop just to get it to look this good!!!!
6/11/08
How much suppression is enough?
Below you'll see a 4th order LR crossover designed in lspCAD.
Below you'll find the crossover schematic.
This is more or less the same as the crossover below.
Now, let's look at the measured result of the active inline equalization and the passive crossover. The CD source is a Philips 963 and the amplifier is an Outlaw Audio RR2150. The active equalization just goes in the preamp out-amp in loop.
Pretty good for a first build iteration. This is done in my garage with a fairly short window, so the resolution is poor below 1k.
Off axis is below. Dipole below the 1.4k crossover, then a typical 1" dome response above.
How about an in-room RTA? Looks pretty good for its limited value.
Impedance of the passive crossover and the active equalization curve are below for verification. Everything checks out.
Now, how about some distortion measurements? I apologize for not having comparison graphs with my RS22528A. However, I did do the measurements and you'll just have to trust me the 5 tone pattern was better, the single tone was comparable, and the SL style 3 tone was somewhat worse. The moral seems to be that the distortion of the 4 driver open baffle is equivalent to a single sealed RS225. I did not do other distortion tests (this is preliminary after all!) Again, my guess is that 100-1k distortion will be markedly better.
So, everything is perfect. We can stop now. Well, not quite. Below you'll see a graph of individual driver response. The woofer breakup is not suppressed as much as I would like. Only ~35dB down between 5-10k.
So, how does it sound? You do know I listen, right...? Well, it sounds excellent. I had the RS22528A as a similar monopole reference. I verified and learned a couple of key points.
First, the active equalization circuit works well. No evidence of audible distortion. I also ran windowed bursts through the equalization circuit to make sure. Though not shown, these were fine. So, objective evidence agrees with subjective opinion, as it should.
The panel is clearly capable of satisfying full range behavior, although the last octave is not present. The active highpass filter at 45-50Hz keeps audible LF distortion down.
The dipole panels clearly sound different in their presentation, especially in the midrange and the bass region, as you would expect. Very natural sounding. Hard to hear any clear woofer breakup, though it's hard not to look at the above graph and not worry a little.
It's hard to make a lot of comments on the sound, since I'm only listening to one unit. (mono)
Next up will be a version with an additional notch filter, and a notched filter design. So far so good. I'm very impressed with the performance so far.
4/09/08
Ah, lspCAD
Well, I broke my SE dongle and Bohdan is giving me a hard time. ("I need to see proof of ownership!") Jeez, it's not like there are that many copies of SE. Let's see, how many upgrades have I bought, as well as new usb dongle. (I've got a parallel port dongle-but this pc doesn't have a parallel port. I know, buy a pci card-but they're finicky and I can't use my laptop!) Whine, whine, whine...
Anyway, lspCAD pro to the rescue. It's always worked before. Below is prototype passive midwoofer-tweeter fourth order LR crossover circuit with PE coil and cap values (with the active inline equalization present).
Don't worry about those dips at 2.5k and 3.5k. That's mostly diffraction and it's still mostly +/-1dB, with a slight droop at the high end. What about those traps? Well, it just makes it so much easier to get a good knee on the curve and let the rest of the network handle the asymptote. The woofer trap also helps the breakup which is 50+ dB down at 6k. Unlike others, I don't object to extra parts.
As an interesting note, the impedance rises inversely with frequency. Although the voltage gain is still ~14dB, the power required is less, more like 9dB, since the voltage will be into ~12 ohms at 50 Hz instead of 4 ohms at 1k. Overall it should be an easy load for any modern, decent 100 watt amp.
This is just the starting point for emulation. lspCAD pro emulates very nicely. Meantime, I'll order one of those Rosewill $13 PCI parallel cards from Newegg and hope Bohdan's old parallel SE dongle works with it...
4/2/08
The winds have calmed.
The winds have calmed and I was able to do some measurements. The graph above is a little busy, but it's worth taking a look at.
-The black curve is the same curve as in the bottom entry for 3/12. Its the measured, unequalized curve of the four RS225's (again, in a series parallel connection, with a 1.25mH coil on the outer pair to roll them off earlier and improve vertical lob
ing) multiplied by the measured transfer function.
-The red curve is an outdoors 3 meter, 8ms(true 8ms-total interval ~16-17ms with ~8ms time of flight) measurement of the dipole 4 RS225 drivers . As you can see, the black curve and the red curve are in excellent agreement, as they should be. It's nice to know the equalization circuit works as it should.
-The aqua curve is the Seas 27TBFC/G on the baffle.
-The blue curve is a Seas DXT horn. I tossed this on the baffle to see how it would work. It has more sensitivity in the 1-10k region, but would be difficult to get flat after 10k. Still, both are probably usable and I'll play around with both when generating a passive crossover.
For the sake of improved clarity, the curve below shows only the final measured woofer and tweeter curves that will be used for crossover modeling.
I will merge a nearfield curve into the woofer data above. (Actually, I will do it bother ways. I'll leave it as is for lspCAD and then generate a spliced version for SE. They should be very similar.) I'll work on the passive crossover in April and finally have a listenable single unit by late May.
A notable interesting point is that the tweeter will require no padding and the system sensitivity will be on the order 91-92 dB. This is a direct result of the increase in efficiency from both the MTM format and the extra 0.5 woofers, so to speak. You get 6dB from the TMM arrangement, then another 6dB from the MTM aspect. Not bad, eh? Now, it's not really 12 dB since the series arrangement nets extra excursion but does not raise the system sensitivity.
Also, while the SPL for 2.83 volts at one meter may be 91 dB at 1k, it will not be that at 50 Hz. Because of the gain circuit, it will require 14 dB more power, i.e. the amp will have to put out ~32 watts or so to maintain the same spl as one watt (assuming 8 ohms) at 1k. Hope that makes sense. Realistically it will take a 65-125 watt amp to drive this well into the 95-100dB range. With both, and room effects, I'm guessing that 100-105dB honest output (though 105dB at 50 Hz dipole is, well, stretching it...)
3/12/08
More Praxis gymnastics
Here's a graph of the true dipole measurement of the RS225's multiplied by the transfer function of the actual measured equalization circuit shown in black. (Praxis is nifty in that you can do math operations on curves in both the time and frequency domain. So you can take a raw measurement, and multiple it by some other curve. In this case, the equalization curve shown below. Very handy for simulating results.) The tweeter is in red.
If that all seems a bit confusing, hopefully this next, somewhat busy graph will clear things up. What I did was multiply the navy blue true measured response of the dipole panel by the true measured equalization circuit, shown in aqua and correct the gain. The result is the black curve. It's not quite the same as measuring the the panel with the equalization in place, but pretty close.
NOTE- the final black curve is not correct below ~85Hz since Praxis has no data below 85Hz in the navy curve. I can generate a more accurate curve below 85Hz, but due to technical reason, I lose sensitivity link with the tweeter. What I was most interested in these curves was making sure the sensitivity between the woofer and tweeter was ok.
So you really want to see a more accurate simulated low end? The true roll off is steeper as you can see below. However, I can't compare the relative tweeter level because of the way I post processed this graph. But it's the correct simulated low end, for what it's worth.
Could I use this to model a crossover? Probably. But the level is very critical. I don't have any extra sensitivity on the tweeter. If the gain on the woofer transfer function is off, even a dB or two-that's going to be a problem. I did this though, to see if the sensitivities would work out. It looks just perfect. The tweeter won't need much, if any padding at all.
So it's looking very good. Now, to just wait for a free Wednesday morning that's not windy so I can measure the panel with the equalization circuit.
I may also just increase the depth of the notch 0.5dB or so by tinkering with the circuit. If only I didn't break my usb SE dongle. I've got a parallel dongle, but no parallel port. Still, a parallel port adapter from Newegg is ~$15. A lot cheaper than a new usb dongle.
3/10/08
It's Alive!!!!!
Yes, yes, it looks messy. But, it works. I fired up the circuit using 18g thermostat wire for the power supply and 22g shielded HD security camera cable. You gotta love HD. Sure, the electrons would prefer driving in a Mercedes Benz. But they're ok with driving through budget wire for prototyping. That power supply is one of the common generic +/-15v ones available through Digikey for ~$45. Hard to make one for that.
Anyway, it works. Below you'll see the result using praxis to test the transfer function. This is the "true" measured result, NOT modeled. Compare this with the Soundeasy/John k spreadsheet simulated result at the bottom of this page. Seems to be accurate within 0.5dB.
Praxis will use a chirp up to 1.3 seconds long, so the resolution is better than 1Hz. The op amps held up well since I bent the pins, dropped them, rubbed them against my cat's fur, touched the Van de Graff generator, and they still seem to be work when I finally got them in.
Praxis makes the marker interpretation quite easy. You can set any marker as a reference. In this case, I set the 10k marker as zero. So there is a 4.4 dB dip for the dipole peak, a total of 14.6dB of dipole gain peaking between 40-50Hz, and then the 12dB roll off (actually, heading towards 18 with the 7 Hz input filter.
3/01/08
They're here
ExpressPCB has sent the boards. Prototype boards are 3/$51 plus shipping. Plus, handy software for board design. Perhaps not the cheapest, but pretty no nonsense for small board prototyping. You have to be able to fit it on a 2.5" by 3.8" board.
Here it is with the components. Only one channel for now. After all, what if it stinks? You can already see some eh, resistor changes. And, c19 and c0 are too close. I had to use a different brand film cap (the green ones) to fit. Live and learn. I figure it will go through a couple of iterations anyway before the final board. This is a small board. Good thing I'm nearsighted without my contacts. Care and a very small point soldering iron are required.
I may try to make a more generic board that would work for multiple active circuits, add an onboard power supply, etc. It depends on the final interest in this project. If I can get 10-20 folks to fork out $20, then we can get a pretty nice board. You can see from the expressPCB site that a nice, silkscreen/soldermask setup will be ~$350-400 for 20 boards. It's not worth it just to make 5 boards...
Now, I need some free time to measure the main panel woofer and tweeter with the inline equalization circuit above.
Incidentally, Context Engineering makes a pretty nice case for 2.5" boards. You can get these at Fry's for ~$10. The case is below the board in the photo above. Of course, you've got to think where you're going to put the +/-15v power supply. The nice thing about this case is that the board slides into machined grooves in the case and doesn't need any other mounting. It's a pretty well finished case for the money. Not like the cheap stamped aluminum project cases out there.
Here's a better picture (credit the gallery of context engineering) of the case style. Of course it's black and not hot purple-pink...
2/22/08
Genesis
It's time for a modestly priced full range dipole. Already, there are a number of high quality 3 way dipole plans available. What's not generally available is a dipole offering high quality, full range behavior, in a 2 way format, that's moderately priced and only requires 2 channels of amplification.
Of course, it depends on how you want to define it. I'm defining full range as true 50 Hz frequency extension, a two way as a WWtWW 2.5 way, and only two channels of amplification but inline low level active equalization (i.e. you need a preamp out-amp in loop or separates). And, of course moderately priced is ~$600 not counting the wood for the baffles. Still, this is cheaper than say, a typical 7" box two way with say, some Scan Speak drivers. And, as you'll see, at least in measured distortion performance, will significantly exceed a typical 7-8" boxed two way.
Why not a more typical 3 way with active equalization? Well, I think SL and John k have this covered. No reason to reinvent the wheel. What I've learned from my design work on the RS22528A has led me to believe that 4 of the RS225 drivers can give significant low end response, adequate for a full range dipole, and have just enough reach to match with the Seas 27TBFC/G. Why not the RS28A? Well, in some respects the RS28A is a better driver, but the Seas can more consistently reach 1.4-1.5k.
I've already been working on this for a good 6 months, and below is a review of the design process I've gone through. Hopefully this will give you a better idea of the project and how I've approached a dipole design.
The initial goal is to get at least the same level of distortion performance in a dipole at 50 Hz that I currently get out of the RS22528A. At least as a rough estimate, the fequal is approximately 200 Hz and to get to 50 Hz, an additional 12 dB of excursion is required. So, four RS225's in a dipole will give the same SPL for the same excursion as a single sealed RS225. Step one is to build a prototype baffle and do some preliminary distortion measurements and FR measurements.
There are four issues to consider at this point:
-First, will the measured distortion of 4 drivers on a baffle be acceptable?
-Second, will the on axis frequency response be relatively easy to equalize?
-Third, will the dipole behavior go high enough to cross at ~1.5kHz?
-Fourth, how to minimize the effects of the narrowing of vertical polar response with 4 drivers?
Above you'll see a prototype baffle. It's rough and ugly, but it is enough to answer these questions. We put it up and do some simple FR and distortion measurements.
For the FR curves, I looked at the inner pair, outer pair, all four, and the inner pair in series, then paralleled with the outer pair in series. Different inductors were placed on the outer pair to roll off the outer pair and improve vertical polar response.
First, all four without an inductor. Very preliminary curves looking at measurement distance, noise etc. This first one is a 50mS curve that's smoothed. Outdoors this works very well as the red is a 6.5mS window and the black is a 50mS smoothed curve. The agreement is excellent and allows a pretty good estimate of what's happening under 150Hz. You see the 6dB roll off and dipole peak. Still it looks very equalizable.
Next, I looked at the different curves for rolling off the outer two. I didn't consider these measurements very good, but they were a starting point.
Here I overlaid the Seas 27TBFC/G response on the raw woofer response. The sensitivity match is very close. This may be a problem later. Once I actively equalize the curve, hopefully the tweeter will be sensitive enough. This can be adjusted somewhat actively, but it ends up being trial and error through a couple of iterations to make sure it's correct.
Looking at the graphs, you might think that deep null at 3k is a bit puzzling. It's the confluence of a couple of issues. First, the RS225 does have a notch naturally there. Below is a curve from my RS22528A raw measurement files.
However, this is made worse by the 4 woofer configuration. In the measurements above, all four were connected without a roll off inductor on the outer pair.
And then, there is the issue of measurement distance. Below, you'll find an interesting set of simulation curves for the 4 driver setup. This shows what the experimental FR curve would be at 1m vs. 4m. The 1m curve looks suspiciously similar to the curves above. What this tells you is that 1m is an unacceptable measurement distance for this setup. I've also simulated 2 and 3m and ideally, 3m or greater is necessary for accurate measurements. Now it's not as bad as the edge would suggest, as a roll off inductor for the outer two woofers will improve things considerably. Still, this tells me that 3m+ is a necessary measurement distance. Any less will introduce significant errors in the crossover design between the woofers and tweeter.
On to distortion measurements. I compared SL style tone bursts between my RS22528A and the RST dipole.
Here, the RST dipole is better at 30Hz, despite being driven at 3dB higher. (Sorry about that-hit the max output on the preamp for the RS22528A and couldn't get to 84.6dB. No matter, the result is clear.)
Here, at 50Hz, the dipole again is in the lead.
A smaller lead for the dipole, but a lead nonetheless in all important higher order region.
At 100 Hz and 250 Hz, the dipole distortion is markedly lower.
At 800 Hz, I'd give the edge to the dipole because of its superior higher order distortion products. However, you can see a slight excess of 2nd order products for the dipole, as shown in the 800 Hz reverse graph. So it's close.
Same data, with the two units flipped. Now the excess red is excess RST distortion.
What's the conclusion? The RST dipole is hands down superior to the RS22528A in distortion, both in low end excursion, and in higher order products. And, the RS22528A is no slouch, if you've read the data on my RS22528A.
So this looks very promising.
Anyway, it's time to build a new baffle. (More measurements were done, but you get the idea.)
A new baffle is born
That first prototype baffle was pretty flimsy. So, time to build a new baffle. A little bit of simulation, a little bit of studying John k's and SL's site and some emails to John, then it's off to build a new baffle.
The baffle is constructed quite simply as follows.
A set of 1x2x6' maple is laminated to a 1x3x6' strip of maple and a 45 degree miter is cut as shown below.
Then, a main panel is cut out of mdf as below.
Next, the struts are rounded over and the main panel is glued into the struts.
Add some Krylon faux finish and there you have it. The maple laminate makes it quite sturdy, surprisingly so.
Measurements revisited
With the new baffle, it's time to redo the measurements at a distance of 2, 3 and 4m. The drivers were measured outdoors, with the inner pair of woofers in series, then paralleled with the outer pair using a 1.25mH coil. Yes, that's a series parallel connection. However, since the drivers are well matched and the drivers are rolled off before the impedance peaks, the series parallel connection is only marginally less desirable than the alternative. Unfortunately, a parallel-series connection doesn't allow for easy passive roll off of the outer pair.
First, a s nearfield estimate. Below you see a nearfield 4 driver average. The red curve is an estimated dipole response using the method documented on John k's website. The estimated dipole response is spliced to the 3m farfield gated curve.
Here we have the farfield 3m measurement above 150Hz spliced to the dipole estimate. This gets imported into John K's spreadsheets, then the circuit is modified and entered into SE for some fine tuning.
Below you see the results. The raw spliced curve is shown(yellow marker), then the equalized curve with the modeled inline filter(green marker). Also shown is the transfer function curve of the inline active equalization circuit(purple marker).
PCB
The next step is to design a pc board. Yes, you can just prototype it on a breadboard, but the cost of a prototype board is fairly inexpensive if you get just the basic board. The prototype board is shown below.
Now I'm just waiting on the prototype boards to show.