BNC to XLR cables

[john12ax7]

THAT1646 would be an example of a cross coupled output stage. It's essentially two opamps driving + and - outputs, the cross coupling allows one side to be grounded without the usual issues of shorting an output.

 

[warpie]

THAT1646 would be an example of a cross coupled output stage. It's essentially two opamps driving + and - outputs, the cross couplong allows one side to be grounded without the usual issues of shorting an output.

I see. So, in this case I would need the configuration "6" with some sort of a "gnd lift" switch depending on whether the output is TX balanced or cross-coupled...

Or maybe two different cables since I can't see any way of having a switch on the XLR connector.

 

[moamps]

I have two types of bal-unbal cables in my set of measurement and interconnection cables. One type where the cold pin (pin 3 on XLR or ring on TRS) is connected to ground/common (pin 1 on XLR or sleeve on TRS) which I use to connect transformer balanced inputs/ outputs and to connect balanced/differential transformerless inputs. I use another type that has these pins not connected for connecting to balanced outputs (crosscoupled or not), which is clearly indicated on the cables. In the conditions of laboratory measurements where measuring devices are connected to one power point from the mains and these cables are very short (1 to max 2m) this way of cable design proved to be quite sufficient to me.

 

[moamps]

The figure




shows the bode plot of the frequency response of one transformer. Also the measurement results can be obtained in *.csv format (frequency, gain, phase) for further processing or display what Ii really need. The number of samples per octave can be changed, the maximum is 500!. Also, the horizontal line (frequency) can be displayed in the log scale.

FFT analysis is not very suitable for audio measurement, but regardless of that, I managed to set the minimum THD of THAT VCA in one compressor without any problems.

P.S.
This all applies to the R&S scope I use. I have no experience with other models from other manufacturers.

 

[ccaudle]

FFT resolution improves with frequency

FFT bin size is constant with frequency. FFT resolution improves with the number of points used, but there is the traditional time/frequency trade off, to get better resolution in frequency takes more time, i.e. more data, so if you are looking at a signal which is not periodic you have to trade resolution for visibility of short duration events.

(based on deductive reasoning anyway - I could be wrong about that).

Suffice it to say you have some misconceptions of how FFT analysis works. You have to study the math, I'm not sure deductive reasoning helps much for something which is completely a mathematical analysis technique.

I have to believe that it would not be usable for FFT of LF audio at all

Not relevant, no one brought up FFT at all, that is something you just injected into the thread. While most digital o'scopes can perform FFT on the captured data, scopes are inherently a time domain instrument, and the measurement being discussed is performed in the time domain, discussion of limitations of FFT is just not relevant at all to the particular case being discussed.

I can't imagine scenario where anyone would need to test an audio device doing anything at greater than 100 kHz

As I pointed out previously, I believe you do not understand the use case being discussed. I gave what I thought was a very clear example using an op-amp that has been discussed quite frequently on gdiy, and pointed out the parameters that would require extended frequency range to test or verify. The same thing would apply if someone were building their own discrete op-amp to use, especially if you were not using a canned design with someone else's PCB design.
This is basic to analysis of feedback systems, you can't really design feedback circuitry without understanding this stuff. In a lot of cases you can rely on the open loop phase/frequency data from the vendor datasheets, but when you understand those you see that a lot of modern op-amps could have the possibility of oscillations at very high frequencies (well into the MHz range). Not having test equipment which can display what is going on at those frequencies leaves you flying completely blind.

I think I will leave it here, the OP obviously knows what he is doing and why, I think discussion of basic electronic theory should probably go to the Drawing Board forum, this is becoming a distraction from the original question about best ways to connect the test equipment.

 

[ccaudle]

DUT-to-Scope wiring (i.e XLR balanced-to-BNC) might be more complicated though

Indeed, there really is no way around needing to know what kind of output you have.
There are four types to consider:
1. Transformer
2. asymmetric, balanced impedance
3. symmetric, independent output drive
4. symmetric, but cross-coupled output drive

Most of my gear is either transformer balanced of IC balanced (that12xx) so I believe that fig.6 would be the right solution.

Yes, and take note of the footnote that indicates pin 3 connects to pin 1 only for cross-coupled output stages. Transformers are fully floating, so it should just be pin 2 to center conductor, pin 3 to shell of BNC.
You can also do that for asymmetric, balanced impedance outputs. That type has pin 2 driven, and pin 3 connected to the output circuit local reference ground through components to match the impedance as closely as practical (so usually a resistor and capacitor to match the components on the output of the op-amp driving pin 2, and possibly EMI filtering components).

What I called type 3, symmetric independent drive is just two op-amps, one driving pin 2 and an inverted copy driving pin 3. In that case you don't want to ground pin 3, because it would just result in one of the op-amps driving into a short circuit. Excessive power supply draw in the best case, device overheating and distortion injected into the other side in some cases.

You wrote ThatCorp 12xx above; were you talking about inputs, or did you mean That 16xx output drivers?

what is "cross-coupled?

The 16xx drivers are a special case of cross-coupled outputs. The original simpler design is much easier to see. I believe the origin of the design was from an HP paper back in the 70's, but it became popularized with the integrated implementation from SSM/Analog Devices, and now the TI copy:
TI DRV134 product page

Note in the diagram the feedback connection from +Out back to the amp driving the -Out pin, and the feedback connection from the -Out pin back to the amp driving the +Out. The feedback connections "cross" (and of course there is the usual negative feedback as well).

That was an attempt to mimic the behavior of a transformer, where if you grounded pin 3, the signal at pin 2 increased in amplitude, so that the output amplitude was the same whether you were driving into a balanced or unbalanced input. It worked in that regard, but because of the feedback path back to the input of the driving op-amp, grounding at the far end of a long cable, which would add some delay and a lot of capacitance, could make the output go unstable, so it still wasn't a perfect substitute for a transformer.

API came up with a design that improved on that behavior, and That Corp licensed it for integrated device use. The head technical guy at That Corp presented an AES paper on it back in the day describing the problems with the original, and how the new circuit improved on that:
That 16xx AES paper

One option not discussed in the Rane paper because it isn't really relevant for sound systems, is using two inputs on your scope and using the math functions to create a balanced input. In that case you need two coax cables thin enough to fit into your XLR housing, and the center conductor of one goes to pin 2, the center conductor of the second goes to pin 3, and the shields of both connect to pin 1.
At the scope you then create an internal signal which is first input - second input, and use that synthetic signal instead of the direct signal for the rest of the measurements. If you are using some kind of automated Bode plot software package on the scope you have to make sure it lets you use the synthetic signal rather than direct input. If you are calculating by hand then it is just a couple of extra setup steps. Of course you need an extra 'scope channel to do that, so won't work with a two channel 'scope.

 

[Bo Deadly]

As I pointed out previously, I believe you do not understand the use case being discussed.

Clearly we're just pushing different points now so I won't hang on to this much more. Your point that a scope is needed for > 20 kHz is understood. Definitely. I never challenged that. My point is that an 8 bit (or even 10 bit) scope is not useful for LF frequency domain analysis. True I conflated FFT and Bode unnecessarily. But Bode and FFT share the same principal of comparing a sin to a signal to compute amplitude and phase. FFT just adds an algorithm to efficiently analyze arbitrarily complex signals. With both FFT and Bode the sampling window time is fixed and so at low frequencies the "information" in the signal is limited which equates to poor resolution [1]. This is exacerbated by the bit depth of a scope. To make a decent measurement you would have to use a really long sampling window. And it seems Bode is only sweeped which makes it even slower.

[1] At least I think it would be. As acknowledged previously, this reasoning could be flawed. Displaying frequencies as log probably contributes more to the appearance of poor LF resolution.

 

[warpie]

You wrote ThatCorp 12xx above; were you talking about inputs, or did you mean That 16xx output drivers?

I meant 16xx indeed.

 

[13engrsapper]

Bode plot? How do you use an oscilloscope to plot time domain data?

The best arrangement would probably be to use an XLR cable and connect BNC signal to hot and shield to cold and connect the XLR shield to pin 1 only at the XLR end.

But when you connect a bunch of clunky unbalanced earth grounded lab devices to studio gear, you run the risk of introducing ground loop currents. So any kind of measurement that requires a good noise floor is going to be limited. You could use cheater plugs to un-earth the lab gear but you're sort of risking shock. A better solution would be to use a USB audio device and a laptop running some kind of FFT software. The USB is floating and designed to be quiet so you'll get much better results. With the proper cables you can get a very good noise floor lower than the device you're measuring which is all that matters.

You need a bridging transformer. Google HP Bridging transformer (yes Hewlett-Packard made them) and read up on how and why to use them.

 

[13engrsapper]

You need a bridging transformer. Google HP Bridging transformer (yes Hewlett-Packard made them) and read up on how and why to use them.

Bode plot? How do you use an oscilloscope to plot time domain data?

The best arrangement would probably be to use an XLR cable and connect BNC signal to hot and shield to cold and connect the XLR shield to pin 1 only at the XLR end.

But when you connect a bunch of clunky unbalanced earth grounded lab devices to studio gear, you run the risk of introducing ground loop currents. So any kind of measurement that requires a good noise floor is going to be limited. You could use cheater plugs to un-earth the lab gear but you're sort of risking shock. A better solution would be to use a USB audio device and a laptop running some kind of FFT software. The USB is floating and designed to be quiet so you'll get much better results. With the proper cables you can get a very good noise floor lower than the device you're measuring which is all that matters.

Many scopes made over the last 30 years have had the ability to link to their brand signal generators. Bode plots can be done right in most modern scopes. the single ended BNC is the real issue... see my "Bridging Transformer" note.

 

[13engrsapper]

Not in any useful way for audio. Digital scopes use 8-10 bit AD converters that are spec'd for speed, not resolution. Even at 10 bits the plot is going to be very crude. There are 12 bit scopes that use software trickery to make a nice looking plot of audio. But for even a vaguely accurate audio measurement, you need a USB audio interface which has 24 bit AD converters that are spec'd for resolution, not speed.

I hate to disagree but if you are just looking at frequency response ALL of those scopes will do just fine. It is just rather tedious hooking them up. I can show you the difference in general frequency plots between the two mentioned umm "platforms" when I have the patience, and a smaller workload. If you can tell the difference I will buy you a beer next time we meet (which would be the first time we meet). We use HP, Siglent and Rhode & Schwarz stuff here - two shops worth.

 

[warpie]

I might be wrong but how a bridging audio transformer (with rather "limited" bandwidth) would show oscillations in the MHz range?

 

[warpie]

One option not discussed in the Rane paper because it isn't really relevant for sound systems, is using two inputs on your scope and using the math functions to create a balanced input. In that case you need two coax cables thin enough to fit into your XLR housing, and the center conductor of one goes to pin 2, the center conductor of the second goes to pin 3, and the shields of both connect to pin 1.
At the scope you then create an internal signal which is first input - second input, and use that synthetic signal instead of the direct signal for the rest of the measurements. If you are using some kind of automated Bode plot software package on the scope you have to make sure it lets you use the synthetic signal rather than direct input. If you are calculating by hand then it is just a couple of extra setup steps. Of course you need an extra 'scope channel to do that, so won't work with a two channel 'scope.

Do you know if Siglent scopes can do this "synthetic signal" for bode plots? With the math function I suppose you mean to invert one input and add the signal to the second input but I'm not sure how can I do this on a bode plot...

 

[Newmarket]

Do you know if Siglent scopes can do this "synthetic signal" for bode plots? With the math function I suppose you mean to invert one input and add the signal to the second input but I'm not sure how can I do this on a bode plot...

More time consuming - but if your scope can write the waveform data to eg a file on a USB (or transfer directly to a computer) then you can take that into Excel or similar and manipulate / automate / plot this however you wish.

 

[warpie]

One thing that isn't clear to me is why not probing the hot (probe's tip) and the cold (probe's gnd clip) of an an audio balanced signal (i.e XLR pin2 and 3) instead of combining two channels with a math function.

 

[Newmarket]

One thing that isn't clear to me is why not probing the hot (probe's tip) and the cold (probe's gnd clip) of an an audio balanced signal (i.e XLR pin2 and 3) instead of combining two channels with a math function.

I'm not clear - are you suggesting attaching the Probe's 'Ground Clip' to XLR Pin 3 ?

 

[warpie]

I'm not clear - are you suggesting attaching the Probe's 'Ground Clip' to XLR Pin 3 ?

Yes, is it not possible?

 

[Newmarket]

Yes, is it not possible?

Well you'd be grounding one leg of a 'balanced' signal.
As often discussed here wrt balanced/differential output stages that can be a problem depending on the detail of the output stage - transformer, impedance balanced, cross balanced, ground sensing, "Tascam Problem" etc.

 

[ccaudle]

Do you know if Siglent scopes can do this "synthetic signal" for bode plots?

I do not know. I'll skim through the user manual and see if I spot any terms that might indicate one way or the other.
I just pointed it out because I had used something similar previously with an Agilent 'scope, but not a Bode plot application, in that case it was other math functions, e.g. you could setup a synthetic signal that was comprised of math function on one or more channels, e.g. subtraction of two physical channels, then perform a different math function on that synthetic signal, such as FFT.

 

[warpie]

I do not know. I'll skim through the user manual and see if I spot any terms that might indicate one way or the other.
I just pointed it out because I had used something similar previously with an Agilent 'scope, but not a Bode plot application, in that case it was other math functions, e.g. you could setup a synthetic signal that was comprised of math function on one or more channels, e.g. subtraction of two physical channels, then perform a different math function on that synthetic signal, such as FFT.

Apparently it can't do it. You can do various MATH function but not on a bode plot.