Balanced Inputs

[mediatechnology]

"Dobrev?" Summing (Mix) Amp Topology Using Common Mode Drive to the Reference Leg's Non-Inverting Summing Node.


"Dobrev?" Summing (Mix) Amp Topology Using Common Mode Drive to the Reference Leg's Non-Inverting Summing Node.

[JR.]

OK, if you are asking about sum buses, this I have spent some time thinking about.

First there is a critical distinction between I/O and sum buses... I/O generally, (not always on the "O" side) only interface between two local grounds. Summing a multiplicity of signals requires dealing with a multiplicity of different local grounds. It is possible to use two virtual earths where a + and - or + and 0ref is summed side by side, then later differentialled, or just make the sum amp a compound differential where the multiplicity of locals grounds are summed in the + input , and the signals in the - input. It's getting hard to still call this a virtual earth since the reference it is following is the composite off all the sundry channels.

I have given this some though over the decades and have a few refinements up my sleeve (some discussed in another thread here) viewtopic.php?f=6&t=211 and some since then. The bottom line for me, as I become more knowledgeable about things digital, digital really does promise number prefect summation, while the marketplace does not embrace it as superior.

I believe I could probably make a better analog summer than what is out there, but not better than a good digital summation. So what's the point? It's a little like making the best buggy whip.

JR

[JR.]

"Dobrev?" Summing (Mix) Amp Topology Using Common Mode Drive to the Reference Leg's Non-Inverting Summing Node.


"Dobrev?" Summing (Mix) Amp Topology Using Common Mode Drive to the Reference Leg's Non-Inverting Summing Node.

OK, I'm a little lost? Non-inverting sum? Rsource=10k?

JR

[mediatechnology]

OK, I'm a little lost? Non-inverting sum? Rsource=10k?

I think you got it. Rsource would be 10K build-outs from the channel. Whether it works with non-floating sources dunno. Seems like it should.

"Non-inverting sum." Hmm. I think I said non-inverting summing node. (And meant the one at the 741 input.) It's not actually a current summing node in the virtual ground sense the inverting input is. What it would serve as is the reference connection for the mix amp. The virtual ground node is clearly the inverting input.

But what is the non-inverting input doing? It looks like it's going to be a voltage node with the common mode impedance actively driven by Vcm. Do I see that right?

With no Vcm, eg Vcm=0V, then the common mode impedance is 10K/leg. (Edit: 10K non-inverting leg, impedance is virtual ground inverting leg.)

When Vcm doesn't=0V, then it looks like the CM error is applied as a current back into the non-inverting input. I need to re-read the IFD after re-drawing it.

You tell me what's really going on here. What I see is something different than what I've seen before.

EDIT: He wrote "G1 balances the common mode interference currents." (Emphasis not in original.) G1 is the common mode amplifier and 10K resistor forming the current source.

EDIT: It looks like CMR is realized at the 741's non-inverting input where Rsource and the 10K sum. (741 LMAO )

EDIT: Birt/Fletcher are impedance-equal to make them current-equal per leg. Dobrev is not impedance-equal per leg but current equal per leg.

[mediatechnology]

I believe I could probably make a better analog summer than what is out there, but not better than a good digital summation. So what's the point? It's a little like making the best buggy whip.

Tell that to the people that make them. And I belive that you don't think there's a point in exploring this any further. Maybe the entire analog processing industry should just pack it in then. I don't think so.

So what's the point? It's a little like making the best buggy whip.

...the marketplace does not embrace it as superior

Well, there you have it folks...

[JR.]

I believe I could probably make a better analog summer than what is out there, but not better than a good digital summation. So what's the point? It's a little like making the best buggy whip.

Tell that to the people that make them. And I belive that you don't think there's a point in exploring this any further. Maybe the entire analog processing industry should just pack it in then. I don't think so.

So what's the point? It's a little like making the best buggy whip.

...the marketplace does not embrace it as superior

Well, there you have it folks...

Sorry if I sound dismissive or condescending... I am just fast forwarding to my conclusion after decades in this pursuit, not laying out my full list of pro and cons. I obviously still have an intellectual interest in this, and somewhat emotional connection as I have friends still making and selling analog consoles (while I give them this same advice).

There is clearly a niche for polished buggy whips (i.e. "Dangerous" whatevers) and I have other friends who are not technical but do recording and lust after dangerous and the like. So there is a niche, but I don't like engineering for niches, it too much like fashion and I don't do well with the BS involved. I like solving simpler more universal (and real) problems, not imagined ones (there I go again.. my apologies again).

I have even tried to steer my die hard analog friends, who make very good analog product, toward the more esoteric customers who deserve to get good analog product for the unusually good money they spend.

FWIW I have hypothesized all kinds of circuitry that I have never reduced to even breadboards, i reserve that kind of effort for products that even I could sell telling only the truth (what I should be working on right now, but this is more entertaining).

So please understand where I am coming from, this is a well explored subject (from my perspective). I first wrote about it 1980, and that was after plenty of work already. I haven't really had another ah-hah moment since then, while today I would take advantage of newer technology to improve execution of my old ideas (current source summing).

JR

[JR.]

OK, I'm a little lost? Non-inverting sum? Rsource=10k?

I think you got it. Rsource would be 10K build-outs from the channel. Whether it works with non-floating sources dunno. Seems like it should.

I will try to keep this investigation in the context of console summing buses. Sources are not floating, but relative to each give or take local ground potential differences. Further bus interference from crosstalk, noise et all, is mechanically different that conventional I/O.

"Non-inverting sum." Hmm. I think I said non-inverting summing node. (And meant the one at the 741 input.) It's not actually a current summing node in the virtual ground sense the inverting input is. What it would serve as is the reference connection for the mix amp. The virtual ground node is clearly the inverting input.

The upper opamp is inverting so I will consider it the inverting summing node. The lower non-inverting opamp the a ground reference sum node. Summing X channels through 10k Rs into top opamp will generate -1X at - output, so far so good.

Ignoring the 1240 for now. The lower + sum will generate a sum of all local grounds so useful to subtract from the top opamp sum, but typically it gets multiplied by the noise gain of the top sum amp. As drawn this ground reference gets the loss from a passive combine but no gain, so will not correct ground differences.

The 1240 looks like it is trying to dump the current going into the - sum bus, back into the + low sum bus.

But what is the non-inverting input doing? It looks like it's going to be a voltage node with the common mode impedance actively driven by Vcm. Do I see that right?

With no Vcm, eg Vcm=0V, then the common mode impedance is 10K/leg. (Edit: 10K non-inverting leg, impedance is virtual ground inverting leg.)

When Vcm doesn't=0V, then it looks like the CM error is applied as a current back into the non-inverting input. I need to re-read the IFD after re-drawing it.

You tell me what's really going on here. What I see is something different than what I've seen before.

EDIT: He wrote "G1 balances the common mode interference currents." (Emphasis not in original.) G1 is the common mode amplifier and 10K resistor forming the current source.

EDIT: It looks like CMR is realized at the 741's non-inverting input where Rsource and the 10K sum. (741 LMAO )

EDIT: Birt/Fletcher are impedance-equal to make them current-equal per leg. Dobrev is not impedance-equal per leg but current equal per leg.

I see enough to see that it is not converging on a working sum bus structure.

It is simpler (not cheaper) to cancel currents in both legs other ways. like a dual polarity differential output feeding dual virtual earth buses. I have run my current source summing with just a simple brute force ground under the premise that current source compliance will ignore ground errors, it would take 2x the current sources to cancel bus currents while the second current source could be lower performance, if used at all.

I can imagine this proposed topology netting out the current going into and out of the master section, but to share nicely and not interfere with the forward sum seems like asking too much from this topology (because of multiplicity of inputs).

JR

[mediatechnology]

OK, thank you for the constructive input. I did say from the get-go that it was "out there."

I think Dobrev's circuit is somewhat similar to what SSL tried with the quasi-balanced topology modification. I don't exactly recall what they did in the master section but their options were limited because it was a retrofit.

I do agree that a fully differential input with virtual ground nodes for each input is a better way.

[JR.]

If I were trying to find other things useful to do with THAT chips, adding two external precision resistors to a 1240 could turn it into a bilateral current source (or just add 5 precision resistors to a 741). But sticking to an all THAT all the time theme, here is a question for Wayne to ask his buddies at THAT. What does the 1240/3/6, 1280/3/5/ chip die look like inside?

While I don't feel that lucky. if they include all of the precision resistors to make all 3 versions on the die, and just wire bond them up at final assembly for the different gain configurations, it would really be sweet if we could reconfigure those internal resistors to make a precision bilateral current source. The output impedance of those is the product of resistor value and importantly, how well they are matched. This very good internal matching could deliver relatively high output impedance with relatively low noise, sweet for my current source summing.

I fear even if a large (analog) console maker were to adopt this approach the numbers would not be enough to justify even a semi-custome part. Perhaps if they just brought out the unused resistors to uncommitted pins on a higher pin count package, customers could program with PCB layout for several different applications.

Of course this is all based on the premise that they did not tool unique silicon for every version (I know I wouldn't ).

JR

[mediatechnology]

I built Dobrev and tried it as a line receiver with a floating low-impedance source. Rs = ~50 Ohms.



I replaced the 741s with an NE5532. Rfb on the top op amp has 100 pF added for stability. I also added a 100 pF across the 10K feeding the non-inverting input to restore balance.

The generator "low" was tied to the inverting virtual ground node; the "high" to the non-inverting input. The 10K resistors around the 5532s were 1%. The "10K" resistors around the 1240 were actually 9K nominal and internal to it.

The power connection to the generator had a floating third pin which has always produced a large amount of leakage current that had previously prevented it from being floated.

The following shows rejection from the common mode interference from WBAP 820 kHz.
The scale for both traces is 500 mV/div. The upper trace is output, the lower the point labeled "Vcm." Apparently it's not really Vcm, But Vin plus Vcm (trash).


Dobrev_WPAB_820kHz_Rejection_Out_Top_Rejected_Bottom_500mV_Per_Div.JPG

The next photo displays reduced amplitude input on the top. The top vertical scale is 50 mV/div. The input signal is ~100 mV p-p.

The bottom scale is Vin+Vcm at 500 mV/div. Note the 20 dB difference in vertical scaling, the top (output) is 20 dB lower!


Dobrev_Top_Out_50mV_Per_Div_Bottom_Rejected_500mV_Per_Div.JPG

Note the slight amount of signal buried in RF, hum and impulse triac/LED lighting dimmer buzz in the "trash" trace.

How does it work with grounded SE sources? Very good though the hum level does appear to rise slightly when it's grounded. There is of course a huge amount of DC-gain in the inverting input with it grounded. AC-coupling both inputs would deal with that. (EDIT: And the noise gain in the inverting leg would be sky high if it was grounded.)

Typically, with the generator AC third wire floating, I see reduced performance with the generator loaded fully-floating. The output is 50 Ohm BNC > coax. (Not shielded twisted pair balanced interconnect.) There is a lot of AC leakage current. Usually the hum is just lower with the generator grounded to the DUT.

Using Dobrev's circuit, the opposite appears to be true: Floating performance is better. This is what we want from a line receiver - at least one from this single-ended generator.

I need to setup a test circuit to provide a balanced transformer with CT producing equal but opposite drive with a variable CM.

My initial impression is that Dobrev's circuit does work "as advertised" and quite well. Whether it's a suitable line receiver or mix amp remains to be seen.