Balanced Output Circuit Variations versus Noise

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
These days, my audio circuits usually need a balanced output for connection to my high-end ADC (Metric Halo Labs LIO-8 and ULN-2). That means a single-ended to balanced converter stage. It might be unity gain, or it might involve approximately +12 dB of gain for -10 dBV to +4 dBu level adjustment.

I prefer to design with op-amps, and have found at least four variations on the circuit. I have attached a schematic showing the four circuit variations. Can anyone comment on the noise performance aspects of these?

I am primarily curious about noise, because it seems that any conversion that involves two stages in series would necessarily have more noise, since the noise added by the first stage would be present in the second.

Specifically, variation 1 and 3 below would seemingly have twice as much noise on the output of the second op-amp stage, because the noise from the first op-amp stage appears at the input of the second op-amp. I'm a little unclear about variation 2, assuming it would suffer from noise, although I might be analyzing it incorrectly.

I've decided on using variation 4, where the input feeds both op-amps in parallel, one inverted and one not. This means that the noise from one op-amp should not be present in the other op-amp (except for noise injected in the ground reference, or something like that). Ideally, the balanced single will have 3 dB better signal-to-noise because the noise in each half is uncorrelated, and thus will only be 3 dB louder compared to the signal, which is 6 dB louder when combined at the balanced receiver.

Regarding gain, variation 1 does not allow gain (e.g. from -10 dBV to +4 dBu) because the inverted output would not match the non-inverted output if any gain were applied with R11 and R12. Variation two can apply any desired gain via R31 and R32, while R33 and R34 would be set for unity gain. Variation 4 requires identical gain settings for both halves, but my opinion is that matching the gain is not critical because balanced signals do not really require perfect signal matching between the two lines, so long as their impedance to ground is identical.

I think that variation 3 is really elegant. Thanks to ideal op-amp conditions, pin 2 follows the signal while pin 6 is held at ground. The gain for each output is determined by the shared R22. It is necessary to match R21 and R23.

Speaking of impedance matching, I've omitted the series output resistors and optional DC-blocking capacitors because I assume that those could be identical for all variations.

p.s. My recent circuits include an active stage for my 13-pin guitar, with each string on its own channel, on a +/-7 V supply, and a phono preamp with both consumer unbalanced and pro balanced outputs on a +/-17 V supply.
 

Attachments

  • variations.png
    variations.png
    23.5 KB · Views: 354
There is a common misconception of what defines a balanced output, that frequently arises. In most cases, converting an unbalanced output to balanced could hardly be simpler, not requiring any phase inverting active circuitry - as is shown by Bill Whitlock in Jensen transformer app. note AN003, page 3. Simply connect the existing single-ended active output to XLR pin-2. Then, connect XLR pin-3 to signal ground via whatever passive network matches that connecting the active single-ended output to XLR pin-2.

So, for example, if your current active output is based on an op-amp with an 100R resistor connecting it to XLR pin-2, then connect XLR pin-3 to signal ground via an identical 100R resistor. If it also features an output coupling cap. in series, then insert an indentical cap. in series with the other line to signal ground. That's essentially it. This gives fully the common-mode noise rejection of the balanced interface.

Remember, do not connect XLR pin-1 to signal-ground, connect it to chassis-ground.
 
Member
Joined 2011
Paid Member
The rightmost two circuits load down the input with their low value gain-setting resistors (R32 and R44 respectively). The leftmost two circuits don't have that problem.

Personally, I favor the leftmost circuit. It lets you use your 6.5 digit benchtop DVM to get a very accurate match between the two equal-value resistors. If you believe Douglas Self's book, which states that a 5532 can drive a 1K load resistor, then set R11=R12=1.000K and get very low noise. If you are conservative, set them to 1.5K and get slightly higher noise but sleep better at night.
 
There is a common misconception of what defines a balanced output, that frequently arises. In most cases, converting an unbalanced output to balanced could hardly be simpler, not requiring any phase inverting active circuitry - as is shown by Bill Whitlock in Jensen transformer app. note AN003, page 3. Simply connect the existing single-ended active output to XLR pin-2. Then, connect XLR pin-3 to signal ground via whatever passive network matches that connecting the active single-ended output to XLR pin-2.
I am fully aware that the signal can appear on only one of the balanced pairs. However, there is a 3 dB improvement in signal-to-noise if both are driven.

In other words, I'm not willing to take the shortcut and sacrifice 3 dB S/N, even though that has identical performance with regard to noise pickup in the cable itself.

So, for the cases where you are driving both (+) and (-) on a balanced pair, I'd like to investigate the problems of the circuit variations, both shown above and any others that would work.
 
The rightmost two circuits load down the input with their low value gain-setting resistors (R32 and R44 respectively). The leftmost two circuits don't have that problem.
Thanks for the reminder! I happen to be driving the balanced section from a NE5534, which should be able to handle the 1K load just fine.

Personally, I favor the leftmost circuit. It lets you use your 6.5 digit benchtop DVM to get a very accurate match between the two equal-value resistors. If you believe Douglas Self's book, which states that a 5532 can drive a 1K load resistor, then set R11=R12=1.000K and get very low noise. If you are conservative, set them to 1.5K and get slightly higher noise but sleep better at night.
The disadvantage of the leftmost is that I would want to apply gain in an additional stage before the voltage follower. Although it's not strictly required to run balanced outputs at +4 dBu, I'd still like to aim for that level.

I was tempted to use the output of the 5534 directly as one of the balanced pair, but that would tempt me to increase the gain in that stage. But I could theoretically add just a single inverting op-amp stage at unity to complete the balanced pair. My thinking is to split the gain between the 5534 and 5532, which has the benefit of allowing an unbalanced output at -10 dBV as well as a balanced output at +4 dBu.
 
Member
Joined 2014
Paid Member
https://www.diyaudio.com/archive/bl...d1460406090-bruno-putzeys-micropre-g-word.pdf always worth (re) reading this.

If you must go balanced AND differential then (and sounding like a stuck record) the cross-coupled output floats my boat. Use an integrated diff amp such as THAT sell and the laser trimmed resistors are already there. If you want to drive long cables THAT1646 has a lot going for it AND you can upset delicate audiophiles by using Pro audio chips :)
 

Attachments

  • SSM2142-fbl.jpg
    SSM2142-fbl.jpg
    89.2 KB · Views: 256

PRR

Member
Joined 2003
Paid Member
In any rational Line Level case, "noise" is not about the line stage but the preamp(s) in front. If the line level is 300mV, coming from a microphone delivering 3mV, there is 100X of gain in front of the line stage. Assuming =equal= noise figures at mike input and line input (but you already gave this up with inverter resistors) then hiss is raised 0.09dB.
 
In any rational Line Level case, "noise" is not about the line stage but the preamp(s) in front. If the line level is 300mV, coming from a microphone delivering 3mV, there is 100X of gain in front of the line stage. Assuming =equal= noise figures at mike input and line input (but you already gave this up with inverter resistors) then hiss is raised 0.09dB.
Are you saying there'd be no noise disparity between the series and parallel differential circuit variations?
 
Last edited:
I should probably explain that I'm considering both general and specific cases, but not quite every possible variation.

In one sense, I'm wondering whether series op-amps result in a higher noise level, even if very small, because the second op-amp sees both signal and noise from the first op-amp. I'm thinking of the four variations that I showed plus any other variations that are similar in complexity.

In another sense, I'm not looking for the best balanced output circuit that can adapt to single-ended connections equally well. There are certainly cases where good product design that provides flexible options for customers would be very important. In my case, I have short cables and will be connecting to a specific piece of gear, so I don't need to add complexity or adaptability.

Basically, I'd just like to confirm that the theoretical 3 dB advantage of driving differential signals over balanced lines, due to uncorrelated noise combining at +3 dB and the dual differential signals combining at +6 dB, would be thwarted if all of the noise of one half is fed into the input of the other half.
 
Member
Joined 2014
Paid Member
Beware that the cross coupled outputs are only conditionally stable. There have been bad experiences in the PA world of the cross coupled specialty chips being unstable driving long cables, albeit some chips are better than others.


Hence why I said that for long cables the THAT1646 has a lot going for it. For domestic use I don't see a problem. But worth pointing out that pros have their own problems, sometimes big ones :).


Thanks for that link. Quoting from page 8, "Unless there are pressing reasons to build an electronically floated output I shouldn’t bother."

There are certainly many cases where the reason is pressing, but not when you can control what device is connected to the output.
Bruno has a nice way with words. Of course in your case the pressing need might be that you cannot drive your ADC to 0dBFS without a differential drive.
 
It works!

I built the most recent schematic shown, along with a +/-17.5V supply, powered it up and am enjoying great sound!

I don't own the kind of test equipment that would be needed to confirm the S/N ratios, but I might try to do some testing.

If there's interest, I can try to take some photos and post them here.
 
It works!

I built the most recent schematic shown, along with a +/-17.5V supply, powered it up and am enjoying great sound!
Uh, I posted this to the wrong thread. I was talking about a complete phono preamp design, not four variations of balanced outputs. Not sure how I managed to get lost.

The phono preamp does have one variation - #4 - on the final balanced output, but that hardly allows comparison of the options. :)
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.