Balanced Outputs

Where we discuss new analog design ideas for Pro Audio and modern spins on vintage ones.
Post Reply
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Balanced Outputs

Post by mediatechnology »

This thread is a collection of data about line driving and outputs culled from another thread.
I chose to use 50 ohms for my standard buildout R, because I read somewhere that audio cable's characteristic impedance was something like that.
What are the characteristics of common audio cable?

Belden 9451 signal specifications:

Image
Belden 9451 Signal Characteristics

Early THAT1646 and SSM2142 Line Driver Comparison

I also did this comparison some time ago: viewtopic.php?f=6&t=141&p=1156

A good reference on line driving requirements is by Dennis Bohn: http://www.rane.com/pdf/ranenotes/Pract ... ements.pdf

Driving Cables with Square Waves Tells Us Something About The Cable

I always like to look in the time domain and I think it might prove instructive here.

The following image is a comparison between one leg of the THAT1646 differential drive feeding ~400 feet of Belden 9451 and the received output from a THAT1246. (The SSM and DRV look virtually identical.)

The circuit is fully-balanced from send, through the cable and to the receiver. The opposing drive output, not shown, is complementary.

Image
THAT1646 feeding 400 feet of 9451 recovered by a THAT1246. 10 kHz, 1 V P-P. Line out top, recovered bottom.

Note the stepped "hook" on the leading and trailing edge. This is not cross-over distortion but simply the time-domain response of the wire. More on that.

Observe how the fully-balanced signal recovered by the line receiver has eliminated the step in the response. Although there is a mild overshoot, what do you expect from 400 feet of wire?

About those "hooks:" Call it time domain reflectometry. Take a closer look at the expanded waveform at 500 nS/division.

Image
THAT1646 feeding 400 feet of 9451 recovered by a THAT1246. 10 kHz, 1 V P-P, expanded 500 nS/division. Line out top, recovered bottom.

At the top we see the THAT1646 output. Almost immediately we see the signal rise to a plateau at about 60% of it's final value.

This tells us one thing from the beginning: The output impedance is not exactly equal to the characteristic impedance of the cable. If it were, the plateau would be at 50%. But, we're close.

On the bottom trace we see the 1246 start to respond approximately 600 ns later. This is the cable delay. With the delay of 9451 being 1.5 nS/foot it is apparent we have about 400 feet of cable.

The incident wave from the 1646 output is initially terminated by the cable characteristic impedance. Our cable delay is about 600 nS. But the plateau is twice that. Why?

The incident wave has to travel 400 feet down the cable and back another 400 feet. The delay of the reflected component is 2T because it has to make a round trip. Thus, the plateau is approximately 1.2 us.
After 1.2 us the reflected wave sums with the incident to converge upon the final value.

Though I have not proved it yet by testing the single-ended case, we have visual evidence to support:
Use of differential circuits at both ends can insure signal integrity at least with regard to those two ends.
Fully balanced transmission and reception permit greater time-domain waveform fidelity. Though we usually think of music as sinewaves, we occasionally have to transmit timecode...

What we see in the time domain is also in the frequency domain. It's just another way of looking at things.

Induced Currents, High Output Common Mode Impedance and Emulating Transformers

One of the things I noticed during testing was that when I grounded one leg at the receive end, I could see on the transmitting end (on that same conductor) a hum potential develop.

The 1246 output remained clean. My test circuit has common grounds for both transmit and receive. They are connected by a huge circuit board plane.

From the waveform shape it appeared to be current from a capacitor input supply. As I suspected, the 400 foot spool of cable was receiving an induced current resulting in a peak-to-peak voltage of 10 mV at the end opposite the short. Moving the spool, I can dial in up to 10 mV of peak current or null it. With it sitting on a wooden floor above a basement, my hunch is I'm right over the Romex feeding the workbench or above the fluorescent light in the basement below. It's easy to forget that there's stuff below you. This is real-world.

At 14.1 Ohms/1000 feet I should have a conductor resistance of about 6 Ohms plus some stray. That's 1.6 mA p-p in a signal line.

The 1.6 mA of current returns to an output.
Due to OutSmarts being transformer-like, the hum potential also appears in common mode on the active output. CCOS circuits such as the DRV134 and SSM2142 behave similarly.

Image
Oscillogram showing transformer-like operation of THAT 1646 with OutSmarts. 10 mV per division.
The top trace is active output. Signal is approximately 10 kHz.
The bottom trace is the 1646 output that is shorted to ground at the far end of 400 feet of cable.
The output that is distally shorted to ground has 10 mV of induced hum picked up by the 400 feet of cable.
The hum component then appears in common mode on the active output.


Other Floating Outputs: Cohen, Pontis, Strahm, Porter and Hebert

I had completely forgotten about Cohen's figure 6 which provides a floating balanced output similar to THAT's OutSmarts.

Though there are differences, Cohen's use of an HP floating output may make them the grandaddy of the floating balanced line driver.

Image
Cohen's proposed figure 6 cross-coupled floating output based on an HP article.

Image
Cohen's proposed figure 6 cross-coupled floating output schematic.

The more I look at Cohen's work the more I appreciate it. We sometimes give him credit for the wrong things and not give him credit where it's really due. I'd like to find that HP Journal citation.

Just found this 1980 HP Journal article that Cohen cited. I think I may actually have a copy of this HP Journal. I remember it.

Looks like sometime between 1980 and 1984 Cohen or someone (Porter?) figured out how to add a center tap.

So far the cross-coupled output timeline is:
Pontis, August 1980.
Thomas Hay, October 1980.
Cohen with output center tap and cross-coupled input, 1984. *
PMI SSM2142, 1989.
Strahm, 1990. *
Hebert, 2000. *
Porter?

I think U2's inputs are mis-drawn....

Image
G.D. Pontis, "Floating A Source Output," HP Journal, August 1980.

https://www.ka-electronics.com/images/p ... df#page=12 (pdf 7.7 MB)

THAT1646 and SSM2142/DRV134 Sense Lead Capacitors
Caps are evil & huge. NPs even more so.
Use a 1606 and film C.

The 10 uF capacitors used with the 1646 are in the common mode feedback path.
Only recently was the datasheet changed to add "NP."
Under normal balanced loads, the capacitors sum into a node driving it to 0V with differential outputs.
Both capacitors could be really, really bad capacitors, and if they were reasonably equal in their "badness," the differential sum at the node would still be 0V.
With the 1646 outputs each swinging 20V P-P at 20 Hz, the terminal voltage across these caps are only 2.25 V P-P. Midband it's only tens of millivolts with the outputs swinging 20V P-P.
Only when the output frequency is very low and one output shorted to ground do these caps contribute to the "differential" component.

I suspect you could probably bridge the caps with back-to-back diodes to make them really bad actors and it would still be unmeasurable until you reached the sub-sonic. And then, only if one output was shorted to ground.

Image
THAT1646 Block Diagram

I submit that there's no meaningful or measurable difference, in this application, whether polarized or NP caps are used.

For reference, a polarized electrolytic capacitor I-V curve.

Image
Capacitor I-V Curve Courtesy Nichicon

Barry Porter Output

I found a (now 404'd) post from Ted Fletcher at soakmag.com
Barry Porter was the original designer of the Trident 'A range' mixer, and this was the one that sounded best of all for the rock and roll records of the time. He was just a lad at the time, and most of the circuits were taken from hobby magazines, modified to be able to use components available from the 'surplus' stores in the Soho area of London. The mic amp was awful. but that didn't matter too much because everyone played so loud. The EQ was one of those happy accidents. it was based around 'Baxendall' type circuits using descrete transistor amplifiers, and their class A operation and minimal feedback gave the whole thing a glorious sound.

The circuit that Dave points to is a very different animal using lots of electronic switching, and five sections of 'state variable' filtering.
Sadly, I must say that this circuit will sound about as exciting as the inbuilt equaliser in Cubase.

Barry was a good friend of mine, after his time at Trident he came and worked with me at Alice on the development of 'on air' mixers, but later in his career he spent all his time grappling with the last decimal points of distortion in everything he designed. (He did some very good power amplifiers).

It's a fascinating thought that it was the imperfections of the old 'A' design that made it sound so good. I have kept to that philosophy in my own P9 mastering EQ design... old fashioned hand wound inductors and soggy class A amplifiers.
I don't want to put a general link up for the circuit of my EQ (as it's a current, and fairly new product) but if anyone is really interested in having a look at it, I will be happy to respond to an email.
I also found a Sound On Sound Review from 2001 where a Trident A-range module had been introduced as a rack-mount but it didn't say by whom. Odd for a product review. I think it may have been the Porter re-issue. The whole netEQ thing is new to me though I do recall seeing that particular output stage drawing cited before. The Porter EQ is a nice design and it looks well thought out. I suspect that output stage got grafted on when he re-launched it in '01.

Porter is one of those figures that reminded me of Flickenger.

I love this quote from Ted Fletcher:
The mic amp was awful. but that didn't matter too much because everyone played so loud.
Rock on.
For those who haven't seen the output stage in the Porter netEQ it's very similar to Cohen's proposed figure 6.

Image
Porter netEQ floating output schematic.

I suspect that the adjustment of P2 not only has an effect on LF balance but also the amount of current dumped into ground when shorted.

Having built something today like Cohen's (below) I do know that it works.
I did take the common mode feedback from the right-hand side of the build-outs as in Porter.

Image
Cohen's proposed figure 6 cross-coupled floating output schematic.

SSL Balanced Output

The SSL cross-coupled output stage "CCOS," was made using discrete resistors. This example is from the main programme output, an 82E195 card.

It has one difference I don't think I've seen before in a CCOS which is a current sense connection on the build-out resistors similar to OutSmarts.

Image
SSL 82E195 Balanced Line Driver

Driving Cables With Square Waves for Fun and Amusement

I ran some bench tests driving cable and found something interesting.

Image
Line Driver Test Circuit Current Boosted THAT1646

I used a test circuit in which had a cross-coupled THAT1286 to provide push-pull drive.
The 1286 outputs drove 2X THAT1646 in a current-boosted configuration to provide drive current up to 70 mA/leg.
(Each 1646 had both sections in parallel using "common mode" drive. There were two 1646.)
The native output impedance of each leg (the 1646 internal 25 Ohm ballast resistors are in parallel) are approximately 12.5 Ohms.
Additional external build-out resistance was added; see table for total value.
The test frequency was a 10 kHz squarewave. See table for levels.
The cable was ~400 feet Belden 9451.
The signal was recovered by a THAT1246. There is 6 dB attenuation in the recovered level.

The input level was increased and the recovered output was monitored until the point just before waveform break-up occurred due to current limiting. The p-p level was recorded.

Results:

Rbuild-out = 12.5 Ohms/leg; max recovered p-p level 7.5V. Large overshoot in recovered waveform.
Rbuild-out = ~34.5 Ohms/leg; max recovered p-p level 10V. Mild overshoot in recovered waveform.
Rbuild-out = ~45.5 Ohms/leg; max recovered p-p level 11V. Critically damped, perfect RC exponential shape.
Rbuild-out = ~45.5 Ohms/leg; with 600 Ohm far-end termination; max recovered p-p level 12V. Critically damped, perfect RC exponential shape.

(Note that the actual differential swing at the input to the 1246 in the above table is 6 dB higher due to the -6 dB attenuation of the part.
Thus, the actual differential voltages are 2X the recovered value.)

Note that increasing the build-out resistance (to approach the line's characteristic impedance) allowed more overall undistorted (in the time domain) signal to be recovered by the line receiver. Adding termination improved it slightly as well.

Decreasing the build-out resistance (at least in this test) places greater demands on the current driving requirements and actually penalizes the maximum level that can be faithfully transmitted and recovered.

Matching the characteristic impedance of the line does seem to help overall performance. Terminating it also makes a small improvement.

This may not apply in every case, but it appears to, with a long line, improve things here. And I think what we see in the time domain will translate to measurement in the frequency domain.

(Though not a fair comparison to the above due to current limiting, the 5534 in a single-ended output - even with a 47R build-out - started losing waveform fidelity with an output level of about 6 V p-p. Since it's SE, this would be a recovered level of around 3V after being attenuated 6 dB by the typical line receiver. The 5534 had a Cc of 22 pF and was connected as a follower.)

Some Photos of the THAT1646 With Various Build-Out Resistors:

The top trace is one leg of the line driver output.
The bottom trace is the line receiver output having 6 dB attenuation.
Both photos were taken at the point right before the driver ran out of current.

In the photo below, using only the internal 12.5 Ohm build-outs, the maximum level is about 5V p-p. Above this level, current limiting sets in.
In this particular case, the 2X 1646 begin to slightly thermal limit and the input reduced before finally being recorded. A LOT of capacitance is being driven with a high dV/dt. A DIP 1646 was used.
Note the overshoot in the 1246 recovered output.

Image
Maximum Faithful Output; Line Driver; 13R Buildout; 5V-Div 10kHz; Top Line Driver Output; Bottom Recovered

In the photo below, the slight "hook" you see at the waveform's rise and fall time mid-point is cable reflection.
The transmitted and recovered waveforms have high fidelity in the time domain.
It appears that 45R build-outs (33R external, 12R internal) match the cable's published characteristic impedance fairly well.

Image
Maximum Faithful Output; Line Driver; 45R Buildout; 5V-Div 10kHz; Top Line Driver Output; Bottom Recovered
Layman's guess; in this case if you look at buildout+cable C, is the build-out successfully current limiting the bits that are over tasking the 1646 with no build-out? And thereby allowing full drive within the audio bandwidth of interest? ?
Well, your layman's guess is pretty close to my layman's guess.
I have to wonder of a power transistor or power amp safe operating area is a good analog.
It's also the build-out plus the complex cable impedance which is series resistance, shunt capacitance and series inductance. An interesting reference follows.
Clearly the larger build-out, to a point, is giving us more. Some of it seems to be current reduction.
In any case it does allow more drive and bandwidth.
So in the case of a long line, proper matching becomes every bit as important as it is with transformer coupling, theoretical specs be damned?
This is where it gets confusing.

Terminating it in 600 Ohms - the value has historical significance - seemed to allow slightly more drive. (11V p-p vs. 12V p-p once the attenuation due to loading was compensated for.)
600 Ohms however is not the characteristic impedance of the line.
Belden quotes ~ 45 Ohms and that seems to be the impedance for a single conductor.
Typically, twisted pair ranges from ~ 90 to 120 Ohms.
110 Ohm Shielded twisted pair is used for digital audio to maintain impedance control and provide a nice open eye pattern.
I think the actual "tip-to-ring," black to red of 9451 is about 90 Ohms. Many manufacturers don't spec it.

400 feet is not a long "transmission line" considering audio bandwidths. It's a short line.

A short Google search in audio DIY type fora seem to indicate that people are confusing impedance-matching, characteristic impedance and transmission lines.
In these fora, the fact that most short lines are not transmission lines are immediately used to slap down any serious discussion about this.

My 400 foot piece of cable is not a transmission line.
It does not need to terminated in it's characteristic impedance of 90 Ohms.
But, adjusting the total build-out resistance - in most cases increasing resistance - seems to increase available level by decreasing peak current.
Counter-intuitive to say the least.

Impedance-Matching the Driver Build-Out Resistors to the Cable. Really?

The SSM2142 and DRV134 have 50 Ohm/leg build-outs.
But, with CCOS outputs, those 50 Ohm per leg resistors are electrically 25 Ohms.
Just about every non-SSM2142 line driver circuit Jung has published has 50 Ohm build-outs.
If you're going to have build-out resistors it seems worthwhile to scale them accordingly.

I'm going to grab some other spools of cable.
I'm also going to see what effect, if any, this has on OutSmarts.

Please visit "Transmission Lines at Audio Frequencies, and a Bit of History," Jim Brown, Audio Systems Group, Inc.:

http://www.audiosystemsgroup.com/TransLines-LowFreq.pdf

I promised the test circuit schematic for the Current-Boosted Line Driver.

Image
Line Driver Test Circuit Current Boosted THAT1646

Image
Maximum Faithful Output; Line Driver; 45R Buildout; 5V-Div 10kHz; Top Line Driver Output; Bottom Recovered

The differential voltage in the above photo is ~22 V p-p, 10 kHz driving 400 feet of cable.
Bottom trace is the received output with 6 dB attenuation.

THAT1646 Driver with Added Build-outs.

I had a chance to try a single THAT1646 with OutSmarts enabled in the previous test circuit with added 22R per leg build-outs.
The differential Zout is 94 Ohms. (25 internal per leg + 22R external per leg)
The sense capacitors are connected on the left-hand side of the added build-out so the added resistance does not become part of the OutSmarts bridge.

In the top photo you can see the cable "hook" at mid-point. The recovered signal, when fully-balanced does not display the hook.

Image
THAT1646 with added 22R build-out.
Driving 400 feet 9451. 1V/div, 10 kHz, Balanced transmission, Received by THAT1246.
Top is drive, bottom -6dB received output.


In the bottom photo it appears that the single-ended configuration is not impedance-matched if you look at the output.
But, looking at the recovered signal, it appears that it is.
I'm curious about this.

Image
THAT1646 with added 22R build-out.
Driving 400 feet 9451. 1V/div, 10 kHz, Single-ended far end short, Received by THAT1246.
Top is drive, bottom -6dB received output.


Is More Available Output Current Better? Maybe.

Here's a current-boosted THAT1646 using two devices in parallel with OutSmarts enabled.
Comparative results both with and without added build-out resistors are also shown.


Image
Current-boosted THAT1646 using two devices in parallel with OutSmarts and added build-out resistors.

A few notes...

The THAT1646 has internal build-outs of ~25R. Paralleled, these serve as ballasting resistors and the native Rout/leg is 12.5 Ohms.
Adding 36R to 12.5R provides a Rout of ~48.5 Ohms and a ZoutDiff of ~ 97 Ohms.
Csense capacitors should be used to reduce each THAT1646's common mode offset.
Without Csense, VosCM could be as high as 250 mV and any output current gained by adding the second device would be wasted by it tugging against the other output.
Using Csense, VosCM is typically <<5 mV.
The value of Csense is doubled because it is driving two OutSmarts bridge networks.

Comparison of waveform fidelity vs. build-out values.
Note that in every example, the actual drive voltage is twice the recovered output.
A recovered signal of 10V p-p is actually 20V p-p differentially.


No external build-out, small-signal response, 2V p-p recovered. ZoutDiff ~ 25 Ohms.

Image
THAT1646X2, 0R Build-out, Small Signal 1V/div, 10kHz, 400ft 9451, Recovered -6dB

36 Ohm external build-out, small-signal response, 2V p-p recovered. ZoutDiff of ~ 97 Ohms.

Image
THAT1646X2, 36R Build-out, Small Signal 1V/div, 10kHz, 400ft 9451, Recovered -6dB

No external build-out, large-signal response, 10V p-p recovered. ZoutDiff ~ 25 Ohms.

Image
THAT1646X2, 0R Build-out, Large Signal 5V/div, 10kHz, 400ft 9451, Recovered -6dB

36 Ohm external build-out, large-signal response, 10V p-p recovered. ZoutDiff of ~ 97 Ohms.

Image
THAT1646X2, 36R Build-out, Large Signal 5V/div, 10kHz, 400ft 9451, Recovered -6dB


36 Ohm external build-out.
Single-ended termination, one leg shorted to the shield at the far end.


Image
THAT1646X2, 36R Build-out, Large Signal 5V/div, 10kHz, 400ft 9451, Recovered -6dB, Single-ended.

This all seems like it's worth the extra two resistors, or if one is rolling their own output, making the resistors the optimum value.
A ZoutDiff of about 90-100R seems about right for this cable and the majority of 2C shielded based on published characteristic impedances.

Moving the Position of the "Hook."

What I haven't shown are intermediate values of Zout...

By varying Rbuild-out I can slide the step up or down....

Image

I can tell when Zout = Zcharacteristic when the "hook" is at midpoint. (Balanced)

If Zout << Zcharacteristic the hook is > mid-point on the rising edge. (It also rings.)

If Zout >> Zcharacteristic then the recovered waveform is overdamped and the hook is < mid-point on the rising edge.

By varying Zout, the location of the hook is moved. This is not pure capacitance.

There is a sweet-spot where Zout ~ Zcharactersitic. This gives the cleanest overall response in the recovered waveform.

You may also not recall it earlier but I did test it with the cable terminated in it's Zchar of ~100 Ohms.
(The drive level of course was reduced significantly due to the low Zterm.)
The step goes away and the final value is 1/2V.

For an unterminated cable with Rsource = Zcharacteristic:

0, 1/2V, then 1V - 1/2V, then 1/2V + 1/2V = 1/2, then 1/2, then 1.
At t=0, t, 2t, then t final.

Please visit "Transmission Lines at Audio Frequencies, and a Bit of History," Jim Brown, Audio Systems Group, Inc.:

http://www.audiosystemsgroup.com/TransLines-LowFreq.pdf

Image
"Transmission Lines at Audio Frequencies, and a Bit of History," Jim Brown, Audio Systems Group, Inc.

Our 10 kHz square wave has components >100 kHz where the line reaches it's characteristic impedance.

THAT publish in the 1646 datasheet:
As an example, Belden 8451 is specified as having with (sic) 34 pF/ft of inter-conductor capacitance and 67 pF/ft of conductor to "other conductor + shield capacitance". Thus, we can assume a single conductor-to-shield capacitance of 33 pF/ft (the difference between 67 and 34) for each conductor. For balanced signals, the load capacitance across the 1646 outputs will be 34 pF/ft + 16.5 pF/ft = 50.5 pF/ft.
I need to sketch this but it looks like the load capacitances are in a "delta."

Driving Long Lengths of Wire is Fun

On another front I was able to "save" about 1000 feet of Brand-Rex 2C shielded that had been on a cardboard spool. One end came off, it spilled over the core, and it developed a knot somewhere in the last 250 feet or so preventing it from being unreeled for the last 20 years. I was able to string it back and forth throughout the front yard, removed the knot and was going to leave it laid out to run a few tests with it as an antenna before re-spooling it. As I thought this I noticed it had gained the attention of the neighborhood squirrels. I decided I like the cable too much to have it chewed somewhere in the middle and re-coiled it. But I can now get to both ends. This cable is about the same overall dimension and construction as 8451/9451 but a lot more flexible. SSL used it for their internal patchbay wiring. I may distract the squirrels with some sunflower seed roll it out again and see if I can make it into an antenna. I'll have to run it the full length of the property and then double it back once. This is one long piece of wire and I should be able to get lots of common-mode RF.

I decided to check the Brand-Rex cable of unknown length using the generator and 'scope.

I suspected the cable was 1000 feet (from having re-coiled it) and that appears to be the case based on overall capacitance measurements and TDR.
I have to assume a velocity of propagation and delay using 9451 values which work out to 1.5 ns/foot.

The top trace is the signal at the far end of the cable terminated only by the scope probe.

The bottom trace is the generator drive which has a 50 Ohm source.
The driven conductors were red and shield showing a rough Zcharacteristic of 50 Ohms or 100 Ohms red-black.
The cable delay is ~1.5 us which is ~1000 feet.
I suspect I was sent a new spool of wire...

Image

Take-aways:

So far:

1) Raising the build-outs to match the Zcharacteristic of the line seem to have benefit in the time domain and permit another dB or two to be obtained with much higher recovered waveform fidelity.
2) I have not seen any added benefit of "impedance matching" with mid-band sine wave testing.
3) Having said that there appears to be no harm in making the Zdifferential out close to 100 Ohms.
4) Exploration of RFI filtering at the outputs needs to be explored.

And it's looking more and more like that bandwidth limiting edge rates with a simple RC at the input of the line driver (or within it) to limit output di/dt provides a huge improvement.

FFT measurements of IM both with and without added build-out resistance to match cable Zcharacteristic.

The test circuit. In the following tests, 1000 feet of Brand-rex "9451 equivalent" was driven.

Here's a current-boosted THAT1646 using two devices in parallel with OutSmarts enabled.
Comparative results both with and without added build-out resistors are also shown.


Image
Current-boosted THAT1646 using two devices in parallel with OutSmarts and added build-out resistors.

A few notes...

The THAT1646 has internal build-outs of ~25R. Paralleled, these serve as ballasting resistors and the native Rout/leg is 12.5 Ohms.
Adding 36R to 12.5R provides a Rout of ~48.5 Ohms and a ZoutDiff of ~ 97 Ohms.
Csense capacitors should be used to reduce each THAT1646's common mode offset.
Without Csense, VosCM could be as high as 250 mV and any output current gained by adding the second device would be wasted by it tugging against the other output.
Using Csense, VosCM is typically <<5 mV.
The value of Csense is doubled because it is driving two OutSmarts bridge networks.


Two Tone Tests

I tried a number of two-tone combinations first considering 19 and 20 kHz sinewave.
Ultimately, for reasons I'll go into later, decided on 10 kHz squarewave and 11 kHz sinewave at 1:1 ratio.
This produced the "best" 1 kHz IM product.


This is a very difficult test and a single THAT1646 or DRV134 could not drive this cable, at these rates edge rates, to 30V p-p.
But, having said that, without a grueling - and perhaps non-realistic - test such as this the 1646 (X2) 1246 combination pretty much gives you out what you put in even 1000 feet later.
I suspect that with a single 1646 drive capability will be cut in half.

First I tested the soundcard to see if it would handle this without excessive IM. I did OK. No needles were found in the haystack.
Initially I chose 48 kHz since I was going to be doing 16 Kpt FFT and averaging over a large number of samples.
I later switched to 96 kHz and the measurements correlated.
I also checked for aliasing.

Bear in mind that these measurements may not have the most accurate absolute results - we're looking for relative changes both with and without added build-out resistors.

Comparison of waveform fidelity vs. build-out values.
This is what the test waveforms look like recovered by the THAT 1246 at the far end of the cable. The scale is 5V/division.
Note that in every example, the actual drive voltage is twice the recovered output.
A recovered signal of 15V p-p is actually 30V p-p differentially.


Zoutdiff = 25 Ohms.
Image
Two Tone Balanced 25 Ohm Zout Recovered Waveform

Zoutdiff = ~98 Ohms = ~Zcharacteristic.
Image
Two Tone Balanced 98 Ohm Zout Recovered Waveform

Zoutdiff = 25 Ohms. Unbalanced.
Image
Two Tone Unbalanced 25 Ohm Zout Recovered Waveform. Shorted end was at the receiver.

Zoutdiff = ~98 Ohms = ~Zcharacteristic. Unbalanced.
Image
Two Tone Unbalanced 98 Ohm Zout Recovered Waveform. Shorted end was at the receiver.

FFT results show that added build-out reduce the IM by about 8-10 dB:

Zoutdiff = 25 Ohms.
Image
Two Tone Balanced 25 Ohm Zout Recovered FFT

Zoutdiff = ~98 Ohms = ~Zcharacteristic.
Image
Two Tone Balanced 98 Ohm Zout Recovered FFT

Zoutdiff = 25 Ohms. Unbalanced.
Image
Two Tone Unbalanced 25 Ohm Zout Recovered FFT. Shorted end was at the receiver.

Zoutdiff = ~98 Ohms = ~Zcharacteristic. Unbalanced.
Image
Two Tone Unbalanced 98 Ohm Zout Recovered FFT. Shorted end was at the receiver.

Not sure what the spur is at ~5800 Hz.

This is a wideband display FFT using a 96 kHz sample rate.

Image

I compared the out-of-band results with all the various combinations of build-out and termination to see if my in-band spurs were the result of aliasing.
I saw no major changes to suggest that this is anything but IM.
48 kHz sample rate provided almost exactly the same results.

A single THAT1646 under the same test conditions shows a bigger difference.
Though not clipping, the 1646 is starting to run out of output current.


Zoutdiff = 50 Ohms. Single THAT1646.
Image
Two Tone Balanced 50 Ohm Zout Recovered FFT. Single THAT1646.

Zoutdiff = ~94 Ohms = ~Zcharacteristic. Single THAT1646.
Image
Two Tone Balanced 94 Ohm Zout Recovered FFT. Single THAT1646.

Though it's starting to run out of current, the added build-out does reduce IM allowing a 1646 to perform well under some grueling test conditions.
Let's not forget that we're driving 1000 feet of cable.

These are my conclusions thus far:

What it boils down to is that the added build-out resistance to match Zcharacteristic seems to reduce current slewing requirement and overshoot.
It reduces distortion and improves waveform fidelity every time it's tried and it seems to do it without hurting anything else.
If an op amp stage is driving the line the added cost is zero.
If a THAT1646 is driving the line the added cost is about 2 cents.
I think you could probably do the same thing rise-time limiting the input to the driver.

The bottom line is if you're going to pick a resistor value for a resistor you need anyway, why not pick the "right" one?

A Zoutdiff of 90-120 Ohms seems about right.

How does a 5534 line driver perform?

The alleged "da best" line driver was made from a 1977 date code Signetics NE5534.

"da best" is either a unity-gain follower, Cc=22pF, with a 47R build-out, or, a gain of two to provide outputs of either ~11 V p-p or 22V p-p.
Zout is 47 Ohms driving one conductor. The other conductor is grounded.

Image
This is the 5534 just beginning to poop out. It's driving 1000 feet of cable. Output is 11 V p-p.

Image
This is the 5534 coming unglued. It's driving 1000 feet of cable. Output is 22 V p-p.

Conclusion:

The THAT1646 has 'nads.
The 5534 is for girly-girl line drivers.
Current limited indeed.

I found this, and some very interesting arguments, in Burdick's "Clean Audio Installation Guide."

Our tests indicate that an 80 to 90 Ohm source yields a maximally flat response. Lower source
impedances will result in high frequency peaking because of series inductance in the cable, not
accounted for in our simplified equivalent circuit. However, the best compromise between
bandwidth and high end response, even though some peaking takes place at the very highest
frequencies (above 100 kHz), is 60 Ohms.


I wrote the manual for Benchmark's first product, the DA-101, and I thought I remembered 30 Ohm/leg build-outs providing a Zoutdiff of 60 Ohms.
I remember looking at the board and it's 2W build-outs and recall seeing some orange.
This was more than 25 years ago. (Too bad I can't remember what I ate last night. :roll: )

One of Al's first customers was ABC. ABCs requirements included wide band time code transmission through his DAs.
Thus, his compromise of 60 Ohms with its' slight peaking makes sense.
His line drivers were small power amps. In fact, that was one of the DA-101's additional applications.

I'm still going to recommend build-outs of 47R or 49R9 (total) per leg. I prefer the maximally flat response and it seems to reduce the current drive requirements.

Al also makes a good case for +26 dB line receiver capability (and IMHO the 20 dB mic preamp pad) in legacy broadcast environments:

Another benefit of the 60 Ohm output impedance is the 0.8 dB amplitude difference between a
bridging input and residual 600 Ohm inputs. The advantage of this may not be immediately
obvious, but let’s consider a not unusual TV situation where the nominal system reference (O VU)
is +8 dBu.

Next, we must realize that the normal peak to average ratio (crest factor) of most
audio is 8 to 10 dB, but, in fact, may be as high as 16 dB, or even higher on very percussive
material.

Now, let us assume that we are feeding a bridging input from a 600 Ohm system (the output voltage will be
6 dB higher without a 600 Ohm “termination”). And lastly let’s add up our amplitudes:
+8 dBu (system avg.) + 16 dB (peak to avg.) + 6 dB (for no termination) = +30 dBu peak out.

When you understand that a unity gain differential (op-amp) input stage running from ± 15 volt
supplies (typical for many pieces of equipment) clips at about +21 dBu, and the input clip of a
properly designed input stage operating from the same ± 15 volt supply rails is +26 to +27 dBu,
you begin to see the need for reducing that last 6 dB term.


Eliminating that last 6 dB term still leaves us with +24 dBu peak out.

(With a 100 Ohm Zoutdiff the insertion loss with a 600 Ohm load is a tad higher at 1.3 dB.)

Comparing the THAT1646, DRV134 and SSM2142.

I ran some tests checking the relative distortion levels of the THAT1646, DRV134 and SSM2142 both with and without 22R/leg build-out resistors.

I decided to give the DIM method described by Leinonen, Otala and Curl a try.
See: http://www.ka-electronics.com/images/pd ... rement.pdf

Based on the high performance of all the line drivers I modified the test to increase the ratio from 4:1 (3.18kHz, 15 kHz) to 1:1.
There were DIM-30 and DIM-100 (30 and 100 kHz) measurements which limited the risetime. I did not apply rise-time limiting.
The line drivers are driving the 1000 feet of Brand-Rex cable, received by a 1246 and an FFT performed.
The differential drive was 24 V p-p.
The peak slew rate appears to be 6V/us.
Tests were run with the native 50 Ohm Zoutdiff and with 22R/leg added to provide an ~94 Ohm Zoutdiff.

The displayed FFTs show only two distortion products at ~2.28 kHz (f2-4f1) and ~4.08 kHz (f2-6f1).
I did this to simplify interpretation.

SSM2142 No Added Build-Out
Image
SSM2142_DIM-30M_0R

SSM2142 Zoutdiff ~ 94 Ohms
Image
SSM2142_DIM-30M_22R

DRV134 No Added Build-Out
Image
DRV134_DIM-30M_0R

DRV134 Zoutdiff ~ 94 Ohms
Image
DRV134_DIM-30M_22R

THAT1646 No Added Build-Out
Image
THAT1646_DIM-30M_0R

THAT1646 Zoutdiff ~ 94 Ohms
Image
THAT1646_DIM-30M_22R

The added build-out resistors seem to reduce the IM sidebands around 3.18 kHz in each case.

Note that the 1646 appears to out-perform the DRV134 by 6-8 dB when build-out resistors are not used.

DRV134 distortion tends to increase after about 30-60 seconds after the die temperature rises.
I can heatsink the DRV134 with my fingertip and lower the distortion.
I did not see thermal drift in the THAT1646.

I ran some measurements using two THAT1646 operating in parallel under identical test conditions.
Outsmarts was enabled.

The first is 24V p-p showing the reduced sidebands from having twice the available drive current.
The differential output impedance is ~97 Ohms.

Image

The following image is 35 V p-p with a ~97 Ohm Zdifferential:

Image

Under identical test conditions two 1646 have approximately 15 dB better performance than a single THAT1646 or DRV134 and about 25 dB lower distortion (these two sidebands) than an SSM2142.

Comparing the above 24V p-p examples using two THAT1646 to a single THAT1646, DRV134 or SSM2142 in "stock" circuits without added build-out resistors the improvements are 25, 31 and 33 dB respectively.

High Current Dual Class-A Board Tests

I wanted to post one more set of FFTs.

This one is the "Dual Class-A Output" we've shown previously as a headphone amplifier.
The cable load and line receivers are the same test circuits as before.
The build-outs are 47R/leg ~94 Ohms differential.
Drive for the dual Class-A is from the cross-coupled "double-balanced" 1286.


The Dual Class-A output at 24V p-p.
Compared to the SSM2142 with no build-out the DIM products are about 35 dB lower.

Image
Class-A 24V p-p 1000 feet

This is at 55V p-p differential.
None of the other amps could handle this test without serious distortion.
The performance shown below is similar to a THAT1646 (with no build-out added) but at a level that is ~7dB hotter. (55V p-p vs. 24V p-p.)

Image
Class-A 55V p-p 1000 feet

THD Driving the line differentially at +24 dBu is nearly equal to the generator residual.
Image
THD Class-A +24dBu 1000 feet
Post Reply