Discrete DI-Box

[Samuel Groner]

Hi

Just a quick idea for an active DI-box I scratched together while curing that cold last week:
DI-Box_r1.pdf

Would make a good figure for the next advertising brochure: discrete, class A and transformer balanced output. :grin:

Any thoughts welcome!

[Edit--here's the PCB layout: DI-Box_r1_PCB.pdf. And THD+N measurements: DI-Box_r1_thdn.pdf. And pictures: DI-Box.html]

Samuel

 

[JohnRoberts]

[quote author="Samuel Groner"]Hi

Just a quick idea for an active DI-box I scratched together while curing that cold last week:
DI-Box_r1.pdf

Would make a good figure for the next advertising brochure: discrete, class A and transformer balanced output. :grin:

Any thoughts welcome!

Samuel[/quote]

I have done a few of those, all different approaches than you chose. Ignoring the general approach, I do have some specific comments.

1. I would bias the JFET stronger than the diode string. As designed the JFET goes all the way to pinchoff to deliver full negative drive to output stage.

2. Ripple filter (Q9) output voltage may vary with Hfe of Q9. If the voltage drop was set by a Vb-e multiplier (instead of R15 to gnd, connect smaler value to emitter). This way the bias string current isn't wasted but goes back into circuit.

3. I'd connect JFET source to base of either Q7 or Q8 and cap couple between two bases, but this is just a personal preference to eliminate a changing term in the path.

4. Rather than tapping power from transformer center tap, I'd just use precision resistors from +/- outputs. This simplifies output transformer design.

This looks like a rather parts intensive (expensive) approach that won't be extremely linear. Should sell well. :roll:

JR

 

[bcarso]

[quote author="JohnRoberts"]

1. I would bias the JFET stronger than the diode string. As designed the JFET goes all the way to pinchoff to deliver full negative drive to output stage.


JR[/quote]

It doesn't work that hard unless the output load is brutal though---the output stage input Z is essentially the reflected transformed load Z divided by beta, since the 1mA through the bias diodes comes from pretty high-Z current sources.

For example, neglecting C loading by the cable, and supposing that the (perfect transformer) transformed load is 25k, the peak-to-peak current change at the JFET is about 7uA for a 2V p-p 1 kHz input voltage. With no load it's almost the same, with 4.3uA of that just from the output stage and the rest from the output Z of the FET I source.


There are some enhancements I would make. A complementary follower arrangement rather than the diodes would work a tad better, for one thing. Then I would arrange to bootstrap the FET drain.

 

[mikep]

cool

I would expect this to have a "sound". did you breadboard it? how would you describe it subjectively?

 

[JohnRoberts]

[quote author="bcarso"][quote author="JohnRoberts"]

1. I would bias the JFET stronger than the diode string. As designed the JFET goes all the way to pinchoff to deliver full negative drive to output stage.


JR[/quote]

It doesn't work that hard unless the output load is brutal though---the output stage input Z is essentially the reflected transformed load Z divided by beta, since the 1mA through the bias diodes comes from pretty high-Z current sources.

For example, neglecting C loading by the cable, and supposing that the (perfect transformer) transformed load is 25k, the peak-to-peak current change at the JFET is about 7uA for a 2V p-p 1 kHz input voltage. With no load it's almost the same, with 4.3uA of that just from the output stage and the rest from the output Z of the FET I source.


There are some enhancements I would make. A complementary follower arrangement rather than the diodes would work a tad better, for one thing. Then I would arrange to bootstrap the FET drain.[/quote]

I'll take your word for it that the FET is not doing any heavy lifting. I am too lazy to look up device betas and such, but I did look up the transformer and as wired it looks to me like a 2.5x step down, so the nominal 2k mic preamp input termination would reflect to 12.5k in ideal sim world. So I would add to my previous suggestion to not draw power from the center tap, I would also put those windings in parallel instead of series to attain the more preferential impedance and voltage ratios.

I designed a very simple DI back in the '80s that I could literally build into a 1/4" plug barrel. Two input JFETs connected differentially, biased appropriately, the differential current output from the two sources was converted back to voltage by a 200 ohm resistor connected between them. I never reduced this to a commercial product (I would have called it a DIC "direct injection cord" make up your own joke), but I did hand build a few that were used in a friends studio to DI bass and keys into his console for tracking.

JR

 

[bcarso]

One caveat after looking at some additional sim results: the output stage quiescent I is awfully sensitive to diode type. That's another reason to replace them with transistors, whether diode-connected ones, complementary drive ones (which does replicate some of the currents so has a slight disadvantage in this phantom power scenario), or possibly the best of both worlds (where the new collectors are returned to the output rather than the rails---see Feucht, HB of Analog Circuit Design IIRC for an example).

The bootstrapping of the drain using one scheme or another BTW helps immensely in reducing distortion for relatively high-Z sources due to FET capacitance modulation.

What would we want to consider as the maximum input voltage swing?

I concur with JR's estimate of output loading---I had based mine on 4k load, and the transformed Z based on Groner's statement about -8dB gain, so for 2k 12.5k is spot on. It doesn't change, to speak of, the conclusion about FET current modulation, nor does the addition of 1nF to represent 2.5nF of cable capacitance.

One could contrive to use the FET drain current to modulate the upper current source and get push-pull drive.

 

[clintrubber]

[quote author="bcarso"] helps immensely in reducing distortion[/quote]
In general, maybe Samuel could chime in with what his original intentions were for this design?

Ultraclean ? Somewhat dirty ?
I assume the former, but the absence of global feedback suggests the latter*.

Best regards,

Peter

/ edit: corrected typo /




*: obviously more open-loopgain would be required than the near unity of the present topology, so that might mess things up a bit.

 

[bcarso]

In sim at least it's way too clean to be dirty :razz:

 

[clintrubber]

[quote author="bcarso"]In sim at least it's way too clean to be dirty :razz:[/quote]
You've ever been to Europe ?
What could be called dirty for the Swiss is quite clean for Amsterdam :wink:

 

[bcarso]

This is daily-step-scrubbing good-citizen army-knife clean.

 

[clintrubber]

[quote author="bcarso"]This is daily-step-scrubbing good-citizen army-knife clean.[/quote]
Cool :wink:

So the previously mentioned idea of being not clean enough was a dirty attack at this circuit :thumb:

 

[PRR]

Looks like a lot of transistors for an easy job. It's just a low performance condenser mike buffer. We do that with 1 to 3 transistors just fine.

AKG 414 head amp has about this many transistors, but does a lot more than just buffer.

We need a power gain of 1,000. (2,500:1 impedance ratio, 2.5:1 voltage loss.)

You threw 9 transistors at it. Potential power gain of a healthy BJT is 1,000. The average power gain you got is is 2.

Have you gained anything through extravagance? I'm not seeing it. Some looks like parts-tossing.

Driving the center of D3 D4 from a single-ended source, and adding two single-ended current limiters to keep the diodes alive, seems wrong. You can set the current in Q1 and D3 D4 with one resistor (or current-limiter, if you feel the signal amplitudes deserve it).

What's with the hefty filtering? Phantom power is normally very clean. Your amplifier is all Collectors between rail and signal, large PSRR. And nothing is critical. It will work at 1mA, or 3mA, or 0.5mA all the same. And if you think about it, the device currents should be proportional to the supply voltage, proportioned to the Load Impedance. Phantom won't be over 53V, and at any decent current will be under 40V, and all these parts will take that fine. Smooth the hiss which the Zener may be adding, smooth turn-on/off ramps, and be done with it.

 

[PRR]

Oh, maybe someone spotted this, but D1 D2 won't work as expected as drawn. You maybe thought they were on the other side of C1.

And Q1 being a fat junction part, you may not be able to kill it even in studio use.

 

[JohnRoberts]

[quote author="RogerFoote"]Samuel

Thanks for posting that! I hope I get time to try it (after I build all the Samuel op amps )

John, can you post a schem for that DIC DI?

I have built a DI for my basses using Bo Hansen's impedance converter, and I would like to compare a few other of the unexplored (well, unexplored for me) designs.[/quote]

I don't remember the actual values, that was probably 25 years ago, but I can scratch out the general topology (just a pair of N ch JFETS biased up to maybe 10V or so)... I'll try to do one this weekend.

JR

 

[bcarso]

Buried in a previous post and uncommented on as yet, so asking again: What's the desirable, or at least the tolerable, input voltage range?

 

[mikep]

[quote author="JohnRoberts"]
I designed a very simple DI back in the '80s that I could literally build into a 1/4" plug barrel. [/quote]

sounds cool. though you gave enough info for us to figure something out, Id like to see your circuit.

the other day I was thinking about making a screw-on 1/4" female to neuman km-80 series capsule mount, so you could unscrew the capsule and use it as a DI. Id put a blocking cap and protection diodes inside. could also try it with an oktava so that something a little less valuable and fragile as at the bass player's feet. whose got a lathe?

has anyone tried making a DI that is basically a DIY version of a famous mic head amp? 47FET electronics could be cool. the U67 high end rolloff might sound good as a bass DI.

mike

 

[Gus]

http://www.groupdiy.com/index.php?topic=1195&highlight=tube

 

[JohnRoberts]

[quote author="RogerFoote"]Samuel



John, can you post a schem for that DIC DI?

I have built a DI for my basses using Bo Hansen's impedance converter, and I would like to compare a few other of the unexplored (well, unexplored for me) designs.[/quote]

OK, this is one I found in one of my old notebooks. I suspect the final values I ended up with were probably different. I loaded this into a schematic capture this morning and only put in values that were in my old notebook schematic. The schematic shows 4093s only because they were all I could find in the library. In my working DICs I probably used PF5103s which I had tons of matched pairs laying around from my phono preamp kits, or the 2SK117 even lower noise JFET I used in my later pre.

The caps were most likely .1uF polyester (green Mylars), and the design built into a 1/4" plug didn't use a gain pot but a fixed value resistor. Gain is simply R4/R1.

For the record this was not intended as an ultra pristine, uber-linear DI, just something simple and easy for tracking. I expect the FETs if overdriven to not sound too nasty but I never critically tested that.

The design criteria for this are several. For optimal noise and loading of the following mic preamp the R connected between the two drains should be more like 200 ohms than 2K. This will be in parallel with the mic preamps input termination so as drawn the effective R there is more like 1K.

The value of the R between the two sources should be selected based on expected p-p input voltage that you want to reproduce cleanly, to not exceed the bias current established by the source to ground resistors. In this case my choice of 38.3K to set operating current density was pretty arbitrary but must likely because that was a value I used in RIAA playback EQ so handy (this also suggests I used the PF5103 devices which were lower Idss than 2SK117s).

In general JFETs will give good noise performance when biased for 0 Vgs although for some very low noise FETs you won't be able to pull 2x Idss from phantom.

Note: if these FETS aren't matched there will be a DC offset at the output that might exceed Wayne's servo pull in range. For most cap coupled or transformer inputs it will be a non-issue. You could add another cap in series the resistor between the two sources to reduce this output offset but due to lower impedance there this would probably end up an electrolytic. If you add one there orient it to be properly biased.

This is probably a good simple circuit to mess around with (maybe even sim if you're so inclined). It's simple, but everything interacts so educational in that respect. For the record, for a bigger than 1/4" plug barrel, but arbitrarily clean DI just power a TL072 (or newer equivalent) from phantom and pad the output to mic level. I don't have a schematic handy for that, but the last one I designed over 15 years ago is still in production.

Have fun, and I make no warranty that this circuit as drawn will do anything. I don't even have one of the few I built to open up and look at. I remember making one for friends studio and another to be used on a Rhodes. YMMV.

JR

 

[Gus]

The Voltage drain to gate is something that looks like will matter.

Maybe a BJT on "top" for a cascode to reduce the V drain to gate?

 

[JohnRoberts]

[quote author="Gus"]The Voltage drain to gate is something that looks like will matter.

Maybe a BJT on "top" for a cascode to reduce the V drain to gate?[/quote]

The 5103 is a 40V part so it should be OK, and I didn't lose any, while admittedly I only built a handful of units, not exactly production qty.

Agreed that is a potential concern if using low voltage parts, but if you look at how it biases up they don't ever see full 48V except perhaps when first powered up and then it's current limited by 6.81K phantom Rs. If you decide to cascode them, you need to make sure the cascode devices keep out of the way of signal swings. I guess it would be no big deal to hang some off the resistor bias string, but it adds parts/complexity and might reduce headroom.

JR