Keyboard DI/Preamp/Balancer

[abbey road d enfer]

But what are the benefits of using an opamp? Isn't it sufficient to take a jfet in, say, source follower configuration, just for impedance conversion?

The 1646 wants to see a very low drive impedance, like that coming from an opamp. The BH DI may not be optimum in that respect)
For KB, the best option is a good old 5532. However, for guitar, a good JFET opamp may be a better choice OPA2134 is a good candidate (it would be good also on KB's.

One more option would be to include a hi Z switch to be able to use the board as guitar/bass input as well.

You should not need that; KB's can see almost any load from 10k to infinity.

Btw.: in SPICE, the b.h. DI works flawless at 15V as well, bias voltage can be corrected by only changing the divider network.

Indeed, discrete electronics can often be adjusted for large variations in operating point.

 

[RFSiesta]

[The 1646 wants to see a very low drive impedance, like that coming from an opamp. The BH DI may not be optimum in that respect)
For KB, the best option is a good old 5532. However, for guitar, a good JFET opamp may be a better choice OPA2134 is a good candidate (it would be good also on KB's.

One more option would be to include a hi Z switch to be able to use the board as guitar/bass input as well.

You should not need that; KB's can see almost any load from 10k to infinity.

Thank you so much for your support!
Ok, so the OPA134 seems to be a good starting point. I would prefer to use one that I can easily adapt to guitar level.
I will maybe include a 100k shunt resistor to enable the coupling cap to unload. If I want to use it for guitar and have, say, 1M termination, it should be OK with 1/10 capacitance ahead. Correct, or am I missing something?
Is a coupling cap OK without any termination at all?

 

[abbey road d enfer]

If I want to use it for guitar and have, say, 1M termination, it should be OK with 1/10 capacitance ahead. Correct, or am I missing something?

That's correct, you could probably reduce the cap value to 4.7n and get a response that would be suitable for electric bass, but there is no advantage in using lower values.

Is a coupling cap OK without any termination at all?

I would stick a 4.7Meg resistor right at the input, just to make sure the cap discharges any inadvertent DC.

 

[RFSiesta]

Alright! Sounds like a plan
Hope to be able to prototype ASAP...

 

[RFSiesta]

So I successfully built the version I laid out a week ago, with opa 134 as differential input, gain jumper for 0/+12dB, followed by THAT1646 with output protection diodes.
Prototype with one channel works perfectly, very clean, no noise/hum even without shielding. Have to  check when all 5 channels are populated.
I used three grounds, one plane as pcb ground, one (stripe) for the inputs and one for outputs.

Maybe I will do two additional channels for DI purpose, with 134 as simple non-inverting amp, no gain jumper, inv.in connected directly to ground, 4.7M shunt across the input and a 4.7-10nF coupling cap. For the Fender Rhodes and guitar or bass
Two DI channels is plenty, since the device will be placed in the smaller control room B.

Thanks a lot for helping an audio gear dumpster like me save some headaches!

 

[RFSiesta]

So I built another prototype. This time, I used a simple non-inverting buffer, a >4MΩ shunt resistor after a small polyester cap. Schematic is simple, I just thought of isolating stages with a coupling cap and also coupled the shunt resistor in the feedback network. I thought it was only a "security" thing to make all opamp inputs and outputs floating and it couldn't do any harm, or could it?
The preamp sounded cool with keyboards, but guitar sounded a bit boomy... so we changed the shunt resistor to 1MΩ to be in the range modern pickups would expect, and it instantly was better, much better for the humbucker actually than my metric halo 2882 instrument inputs

Next I decided to give the laptop outputs a try and it turned out they distorted on full level. Wich would be perfectly OK, if it was the soundcard inputs - but they were at -4dB, so no danger here. Headroom should be enough to drive the metric into total distortion, with +-15V rail supply.
What is wrong here?

Then I shorted the decoupling caps C19 and C20 with a piece of wire and it was distorted all the time at much lower input level.
They should not be necessary, should they?
Disturbing...
*********
EDIT: It works both ways with keyboards and guitar. only the laptop outputs are crushed...

 

[abbey road d enfer]

So I built another prototype. This time, I used a simple non-inverting buffer, a >4MΩ shunt resistor after a small polyester cap.

I can't see the full schemo, but it's not only a buffer, it's a gain-stage with 21+6 dB gain. It's no wonder you have headroom problems. A Les Paul can produce up to 500mV rms (-3dBu); with 27dB gain, you can easily overload any input.

Schematic is simple, I just thought of isolating stages with a coupling cap and also coupled the shunt resistor in the feedback network.

I guess you mean "added lag-compensation caps across  the FB resistors"...?

I thought it was only a "security" thing to make all opamp inputs and outputs floating

Again I guess you meant "AC-coupling the IO's..? Using the correct (or at least sanctified-by-usage) terminology is part of the game, isnt'it?

The preamp sounded cool with keyboards, but guitar sounded a bit boomy... so we changed the shunt resistor to 1MΩ to be in the range modern pickups would expect, and it instantly was better, much better for the humbucker actually than my metric halo 2882 instrument inputs

I have never felt comfortable with input Z>1Meg. It is recommended for piezo's, but I always thought it brings out mud. Since you have input caps, I would think you should experiment with different values; a whole lot of the mud comes from the pick attack that contains VLF components.

Next I decided to give the laptop outputs a try and it turned out they distorted on full level. Wich would be perfectly OK, if it was the soundcard inputs - but they were at -4dB, so no danger here. Headroom should be enough to drive the metric into total distortion, with +-15V rail supply.
What is wrong here?

I don't understand your question there. Laptop outputs are generally powered with a single 5V rail, for a max output level of about +4 to +10 dBu. What do you mean by "distorted on full level"? Is it still distorted if you turn down the volume in the control panel?

Then I shorted the decoupling caps C19 and C20 with a piece of wire and it was distorted all the time at much lower input level.
They should not be necessary, should they?

As I mentioned earlier, I can't see the schemo in its entirety, there seems to be a problem with the server...

 

[RFSiesta]

I can't see the full schemo, but it's not only a buffer, it's a gain-stage with 21+6 dB gain. It's no wonder you have headroom problems

Of course I am aware it's an  amplifier. Amplification factor for different instruments is what I need to find out with tests.
With my Fender Rhodes, these 27dB are the least I need, for guitar it's already too hot. With a microkorg keyboard it was a little too hot, almost perfect (?), and for the AKAI mpc >15dB too high.
I would expect it to have too much gain for line level, but then the input of my converter should clip instantly.
! It didn't, it read -4dB, but the sound was  distorted. THAT's what puzzled me...
Could not reproduce this  with any other signal source.
We had some cable issues later (shortcut), so maybe it was the cable beginning to fail.

The pres are part of a big patchbay system and have no frontplate for switches or pots. we will do different fixed levels for different inputs, so there will be a 30dB gain "Rhodes" input, a couple of +10dB keyboard inputs and each two or so  for active/passive git&basses.
We will need some channels  with different gains in each of the studio's rooms as "utility" amps. I don't want any keyboard player in any room plug his instrument into >20m cables leading to the upstairs control room without buffering/amping/balancing.

Using the correct (or at least sanctified-by-usage) terminology is part of the game, isnt'it?

I am no native speaker, I have to improvise, there's many technical words and many are "slang" and still used everywhere. And in german, most are completely different. Since my little knowledge is from books (70%german) and internet (80%english), I sometimes get confused with words. Sorry about that.

With floating I wanted to say the whole opamp has DC reference to ground  only through the MΩ shunt resistor at the input. So I thought it's a bit like a transformer coupled input is "floating"- but it is actually not.
I replaced the AC-coupling cap C19 from R31 to ground with wire, I guess it does not make sense here. At testing, it made no difference whatsoever.
I left in the coupling cap to the THAT input, though.

I have never felt comfortable with input Z>1Meg. It is recommended for piezo's, but I always thought it brings out mud. Since you have input caps, I would think you should experiment with different values; a whole lot of the mud comes from the pick attack that contains VLF components.

Yes, guitar was indeed muddy with 4MΩ and much better with 1M. But it's the behaviour of the pickups, not the filter C, which is "huge" enough in both cases: 100nF.
We were discussing the filter C in the studio yesterday- For keyboards and MPC drumsampler (which sounds incredibly cool now I don't want to have a locut, but it does make sense on git and sometimes on electric bass.

Sorry for the incomplete schemo, but there's not much missing. Right from the 1646 are the output connectors and guarding diodes to rails, against phantom power "abuse". The bridging cap over R34 is 68pF at the moment. the OPA134 is  stable if I take it out, but Ii imagine it could be necessary with another opamp and different feedback networks, so I will leave it in the layout.

One last question: What values for a feedback network of f.ex 1:10. Should R34+R31  be in the area of the next stage's load impedance?
I know, too high values lead to more noise, filter effects through stray capacity, instability etc., too low values will lead to excess current and are to be avoided. But other than that..?
EDIT: found info, will test within 10k-100k range in my design, see if the sound changes at all..

 

[abbey road d enfer]

One last question: What values for a feedback network of f.ex 1:10. Should R34+R31  be in the area of the next stage's load impedance?
I know, too high values lead to more noise, too low values will lead to excess current and are to be avoided. But other than that..? I don't find much information on that topic...

You should make sure that the impedance seen by the inverting input is smaller than the optimum noise impedance. The impedance seen by the inverting input is equal to the resistor that goes from the output to the inverting input in parallels with the one that goes from inverting input to ground.
The optimum noise impedance is calculated from the datasheet by dividing the noise voltage density by the noise current density.
For example: NE5532 (BJT input)
Noise voltage density: 5nV/sqrtHz
Noise current density: 0.7pA/sqrtHz
Ropt: 5n/0.7p=7140ohms
So typically I would choose Req<1.5.k

Now if we take OPA134 (FET input)
Noise voltage density: 8nV/sqrtHz
Noise current density: 3fA/sqrtHz (0.003pA/sqrtHz)
Ropt: 8n/3f=2 660 000ohms
You may conclude that choosing Req<500k is acceptable, but in fact stray capacitances may create issues, so I would choose Req<50k.

Now the same rule applies to the source impedance that is connected to the non-inverting input; if the source is the typical high impedance coming from a piezo pick-up, I would choose a FET input.

 

[RFSiesta]

One last question: What values for a feedback network of f.ex 1:10. Should R34+R31  be in the area of the next stage's load impedance?
I know, too high values lead to more noise, too low values will lead to excess current and are to be avoided. But other than that..? I don't find much information on that topic...

You should make sure that the impedance seen by the inverting input is smaller than the optimum noise impedance. The impedance seen by the inverting input is equal to the resistor that goes from the output to the inverting input in parallels with the one that goes from inverting input to ground.
The optimum noise impedance is calculated from the datasheet by dividing the noise voltage density by the noise current density.
For example: NE5532 (BJT input)
Noise voltage density: 5nV/sqrtHz
Noise current density: 0.7pA/sqrtHz
Ropt: 5n/0.7p=7140ohms
So typically I would choose Req<1.5.k

Now if we take OPA134 (FET input)
Noise voltage density: 8nV/sqrtHz
Noise current density: 3fA/sqrtHz (0.003pA/sqrtHz)
Ropt: 8n/3f=2 660 000ohms
You may conclude that choosing Req<500k is acceptable, but in fact stray capacitances may create issues, so I would choose Req<50k.

Now the same rule applies to the source impedance that is connected to the non-inverting input; if the source is the typical high impedance coming from a piezo pick-up, I would choose a FET input.

Wonderful! Again, thank you for the insights! We've been discussing this topic from time to time ever since we began repairing stuff in the studio. There is no detailed information to be found anywhere, apart from: too high=noisy and unstable, too low=current issues and "well, you can't go wrong with...1k - 100k". The values we found almost anywhere in our TL071-driven Trident 80, are 4k7 and 47k, btw. 

FWIW I will stick with the JFET OPA134, I bought a whole bunch for some repairs and kind of have them lying around studio.
Project will be finished next week.

 

[abbey road d enfer]

FWIW I will stick with the JFET OPA134, I bought a whole bunch for some repairs and kind of have them lying around studio.

Indeed, OPA134 is an impressive opamp, very tolerant on its environment, and capable of superb performance when implemented correctly.

 

[RFSiesta]

Build a working 5-channel PCB.
EDIT: It's just a single ended non-inverting input, because other than I thought at the beginnning of this thread we won't use balanced sources at this section of the patchbay... so it got simple and "boring"
I chose different ratios for different gains and Rshunt - all seem to work perfectly, and seem stable without having a cap placed across the feedback resistor.  Will leave the pads open if no problems occur.
I will do some testing tomorrow, I'm especially interested if there's a difference between 10k/1M input Z with keys and MPC, and if there's a difference in sound between the 100/10k and 10k/1k version Just for fun.
Values (not easily recognized in the picture) are:
10k/1k - input Z= 1MΩ
47k/47k - input Z=10k,
100k/4k7 - input Z=1MΩ
100k/10k - input Z= 1MΩ
47k/10k - input Z=1MΩ
In any case, the preamp sounds very clean and better than 90% semiprofessional gear, which is already good for me.
We have coloration galore in the studio, so clean is different in our case
Hope to have some pics soon.