Spring 2010

The Xenover - Part II
Article And Design By Grey Rollins

Difficulty Level

  Last time, I laid out the basic topology for a JFET buffer and ingredients which, if stuffed between two of the buffers like cold cuts between slices of bread, would make a splendid active crossover sandwich. I offered two recipes — one for 6 dB/octave crossovers (both low and high pass) and another for 12dB/octave crossovers (again, both low and high pass). This time I’ll add 18 and 24dB/octave crossover fillings, but first I’d like to cover a couple of variations on the JFET buffers.

The original JFET buffer consisted of two N-ch and two P-chJFETs in a push-pull arrangement. I noted at the time that it wasn’t strictly necessary to use two pairs of JFETs. I just happen to have a bit of a fetish about supplying more current than the textbooks say is necessary. With that in mind, schematic 4a shows a single pair of JFETs to make that clearer. Note also that I’ve included 100Ω Gate stoppers for the JFETs just to be on the safe side. For some reason — I assume it’s because I have relatively low levels of radio frequency nonsense where I live — I’m able to do without such things, but it’s always a good idea to have stoppers if there’s even a skinny doubt about whether your area might be prone to RF-induced oscillation.

Schematic 4b shows a single-ended variation on the circuit for those who prefer current sources, although it will also allow those who can only locate one type of JFET to join in the fun. The J271 would be one possibility here; the same device I used in the P-amp headphone amplifier. Remember that the J271 is a P-channel device, so reverse the rail polarity to use them. The J310 would be another contender. This being an N-ch device, the topology shown will work as drawn. Match the top and bottom JFETs for Idssand adjust the Source resistors to give about 5 to 10mA of idle current. Higher values give less current—lower values give more. Neither of these devices is expensive. The 2SK170 or the 2SJ74s would work extremely well, but they have been discontinued and are somewhat harder to find, not to mention more expensive. For those fortunate enough to have their dual siblings, the 2SK389 and the 2SJ109, it would simplify matters greatly, but the cost of buying these is prohibitive. A more accessible alternative is Linear Integrated Systems’ LSK170, manufactured under license from Toshiba as a replacement for the 2SK170. Unfortunately, the last time I checked, they did not yet have a P-ch replacement for the 2SJ74. If you substitute parts, don’t forget to adjust the rail voltages accordingly and keep the power dissipation limits in mind when setting the bias current. I recommend running small signal devices at half or less of their rated power dissipation. If you chose to push things harder, use a heat sink and bench test the circuit thoroughly before putting it into your system.

Bipolar variations are possible too, but I prefer to use JFETs for inputs when possible. If you want to try a bipolar buffer, consider the one Walt Jung has posted on his website as seen here

Now let’s get back to crossovers. Although the 6dB/oct. crossover slope is ideal from the audio point of view, it doesn’t roll off quickly enough for many drivers. Woofers in particular tend to have nasty peaks an octave or two above their nominal roll off point. While it’s perfectly valid to move the crossover point down to a lower frequency so as to reduce the peak, it wastes valuable bandwidth and places greater demands on your midrange or tweeter to fill in the gap.

The 12dB/oct. crossover I offered last time is one option for dealing with this problem, but what if that isn’t steep enough? The obvious answer is to try a crossover slope that’s steeper still. The next step up from a 12dB/oct. crossover slope is an 18dB/oct. Not only does it offer a steeper slope, but it reverts to being non-inverting. Plus is plus and minus is minus again. The 12dB/oct. slope, at least in principle, requires hooking up the driver “backwards,” meaning that the plus lead goes to the minus terminal on the driver and the minus lead goes to the positive terminal. Don’t get too obsessed with this, however, as sometimes the speaker as a whole performs better with the driver hooked up the “normal” way. Sometimes you have to try it both ways to see which way gives a smoother response.

There are two ways to build an 18dB/oct. crossover. One way is to use a 6dB/oct. section in conjunction with a 12dB/oct. section. The other is to put both sections together; the benefit being the ability to dispense with one buffer section. The pitfall with the all-in-one approach is that the values can get pretty wonky, and if you’re interested in building an adjustable crossover—the kind where you twist a single knob to change the crossover frequency — it can complicate things to the point where it’s virtually impossible to implement in the real world. Given that the JFET buffers I’ve shown are about as close to neutral as any real-world circuit can be, there’s very little penalty for splitting the crossover into two sections.

Schematic 5 shows the topology for an 18dB/oct. high pass filter. Note that it is simply a 6dB/oct. filter added to a 12dB/oct. filter. The slopes are additive, so this yields an 18dB/oct. slope. The buffer sections can be either of the ones discussed above or the one shown in Part I of the Xenover write-up. The parts values change, depending on whether you want a Butterworth or Bessel crossover, but the layout remains the same.

Click here for PDF of the schematics.

For a Butterworth high pass filter, the formulas are:

C = C1 = C2 = C3

R5 = 1/(2*Π*F*C)

R10 = 0.5/(2*Π*F*C)

R11 = 2/(2*Π*F*C)

For a Butterworth low pass filter (see Schematic 6)

R = R5 = R11 = R12

C1 = 1/(2*Π*F*R)

C2 = 2/(2*Π*F*R)

C3 = .5/(2*Π*F*R)

Bessel filters are calculated in a similar manner, but the values will be slightly different.

For a high pass (Schematic 5)

C = C1 = C2 = C3

R5 = 1.3228/(2*Π*F*C)

R10 = 1.0474/(2*Π*F*C)

R11 = 2.008/(2*Π*F*C)

And for a Bessel low pass (Schematic 6)

R = R5 = R11 = R12

C1 = 0.7560/(2*Π*F*R)

C2 = 0.9548/(2*Π*F*R)

C3 = 0.4998/(2*Π*F*R)

And finally, we have the 24dB/oct. crossover. Butterworth and Bessel versions are possible of course, but the one that seems to inspire the most passion is the Linkwitz-Riley. Those who follow the Linkwitz-Riley crossover do so with a devotion usually reserved for rock stars or buxom movie starlets. But for those not so-inclined, I will also offer the formulas for Butterworth and Bessel. The 24dB/oct. topology remains unchanged; you only need to change the parts values to compare one crossover type to another. As you might deduce from the trend of the 6, 12, and 18dB crossovers, it is two 12dB/oct. crossovers in series. Again, the slope is additive, so this gives a 24dB/oct. slope.

Butterworth high pass (Schematic 7)

C = C1 = C2 = C3 = C4 

R5 = 0.9239/(2*Π*F*C)

R6 = 1.0824/(2*Π*F*C)

R11 = 0.3827/(2*Π*F*C)

R12 = 2.6130/(2*Π*F*C)

Butterworth low pass (Schematic 8)

R = R5 = R6 = R12 = R13

C1 = 1.0824/(2*Π*F*R)

C2 = 0.9239/(2*Π*F*R)

C3 = 2.6130/(2*Π*F*R)

C4 = 0.3827/(2*Π*F*R)

Bessel high pass (Schematic 7)

C = C1 = C2 = C3 = C4

R5 = 1.3701/(2*Π*F*C)

R6 = 1.4929/(2*Π*F*C)

R11 = 0.9952/(2*Π*F*C)

R12 = 2.5830/(2*Π*F*C)

Bessel low pass (Schematic 8)

R = R5 = R6 = R12 = R13

C1 = 0.7298/(2*Π*F*R)

C2 = 0.6699/(2*Π*F*R)

C3 = 1.0046/(2*Π*F*R)

C4 = 0.3872/(2*Π*F*R)

            Linkwitz-Riley high pass (Schematic 7):

C = C1 = C2 = C3 = C4

R = R5 = R11 = 1/(2.8284*Π*F*C)

R6 = R12 = 2*R

            Linkwitz-Riley low pass (Schematic 8):

R = R5 = R6 = R12 = R13

C = C2 = C4 = 1/(2.8284*Π*F*R)

C1 = C3 = 2*C

Should you need a band pass filter for a midrange, all you have to do is put a low pass filter and a high pass filter in series. The filter slopes do not have to match; you are perfectly welcome to use a 12dB/oct. filter between the woofer and midrange and a 6dB/oct. between the midrange and the tweeter. If that’s the best way to match the drivers you’ve chosen, then by all means mix and match to your heart’s content.