There's so much to comment on here! Perhaps the most glaring omission from your discussion is the importance of extremely high common-mode input impedances (why do you think transformer inputs work so well when diff-amps perform miserably at noise rejection?). They are the key to consistently high CMRR in the real-world where signal sources have significant impedance imbalances (I convinced the IEC to change its CMRR test because testing with a lab-generated zero imbalance source proves nothing). The InGenius IC manufactured by THAT (and patented by me) overcomes this very serious issue. You, like many practitioners, obsess over the non-match of single-ended input impedance for the ordinary diff-amp. It is a trivial issue - what matters is that the common-mode input impedances match - and they do (drive both legs to the same voltage and you'll see that the input currents in the two legs match). The asymmetry of input voltage with non-zero source impedances affects only headroom and crosstalk if cabling is unshielded. Concerning single-ended (unbalanced) sources: a shielded twisted-pair cable, with diff-amp "low" grounded ONLY AT THE SOURCE and the unbalanced signal to diff-amp "high" will eliminate common-impedance coupling in the unbalanced cable. The "benefit" of being able to drive only one leg of the diff-amp input and ground throws away at least 30 dB of noise rejection. There's lots more to criticize here, but you should sit in on one of my free tutorials at the next AES convention (or CEDIA or InfoComm) or read the chapter I wrote in Ballou's handbook or just visit the Jensen website and read the generic seminar handout there. It may be a real "eye opener". - Bill Whitlock, President & Chief Engineer, Jensen Transformers, www.jensen-transformers.com, AES Life Fellow, IEEE Life Senior
Wow, good article, but also very good first reply! In simulation and theory you need to take things into account explicitely. Whereas in reality "unexpected" things just happen. And indeed the proposed methods are not as good in the presence of impedance unsymmetries as what THAT has done. On the other hand the well-known 3-opamp solution is also good, and using 2 buffers instead is not that much a saving, but all in all THAT offers a real good practicable solution.
Bill - Thank you for your comments. I'm well aware of the benefit of high common mode impedance and the superiority of transformers. I'm also well aware of InGenius and have also developed some high CM impedance three op amp "active" approaches that do not use InGenius bootstrapping using "T-bias." I'm also not particularly obsessed with having equal input impedance from each leg to ground though I note in references 1 and 2 there are people who do, most notably Birt in his work with the BBC. What I do concern myself with is predictability in the field when unbalanced connections are fed into balanced inputs under what are often non-ideal conditions with no time for troubleshooting or optimization. I have tubes of THAT1200s, 1203s and 1206s InGenius ICs in my inventory and I am quite familiar with them. I've used them to make some excellent AC-coupled line inputs where the high CM impedance reduces capacitor mis-match effects on LF CMR. (The T-bias input also permits this.)
InGenius, and other inputs with high CM impedance are the most "transformer-like" input and with non-zero and unequal source impedances will provide greater CM rejection than those with low CM impedance. I think we agree on that.
There are a number of issues with high CM impedance inputs when they are improperly connected to unbalanced sources. A floating port connection to a transformer input will yield very little (or no) output because there is no primary current flow. An InGenius input with an open port will also produce no (or a highly attenuated) output. OK, we know this, but in the field I might have to grab another adapter to pull this off quickly.
(Comments are limited to 2000 characters so this reponse will be in two or more parts.)
I'd be very interested to learn more about your high CMZ circuits. Although classic instrumentation amps have extreme input impedances, there's the pesky matter of bias currents in a practical circuit. XLR inputs can't simply be tied to an IA input because many signal sources have no DC path for bias currents (transformers or capacitor-coupled outputs, notably). So the downfall of most IA implementations is the resistors from inputs to ground for bias. These now set the CM input Z ... and generally at a value far too low. I honestly don't believe that a balanced input should be required to perform with only one input leg connected. The fact that there's little or no output should alert the user that he's done something wrong (you wouldn't expect your toaster to work if you plugged in only one wire of its power cord, would you?). Any adapter that leaves either pin 2 or 3 of its XLR floating (I know of none) is simply built wrong. As an aside, I'm curious why my (late) friend and colleague Neil Muncy has his 1995 AES paper listed in your references and not mine (in the same issue of the AES Journal)? His paper deals almost exclusively with the "pin 1 problem" while mine specifically deals with the issues being discussed in this thread. BTW, that issue of the Journal, June 1995, has become the largest selling issue ever printed by the AES. If you don't want to spend the $5 for it, I'll e-mail a copy to any interested reader. I'm email@example.com. - Bill Whitlock, president & chief engineer, Jensen Transformers, AES Life Fellow, IEEE Life Senior
Bill - Thank you for your comments. An example of a "medium" CMZ "T-bias" circuit used in a mic preamp can be found in figure 5 of the THAT1510/1512 datasheet. R1 and R2 are the bias resistors, R7 a CM resistor. I provided that circuit to Rosalfonso Bortoni and he performed measurements with R7 up to 100K with no measurable increase in noise from it being 0R. The advantage of having the higher CM impedance is that it reduces the matching requirements of the input caps. http://www.thatcorp.com/datashts/THAT_1510-1512_Datasheet.pdf
For a relatively high CMZ line input using something with as "poor" Ibias and Inoise as a bipolar input 5532, one can scale R7 to very high values and allow both the Ibias Vos and Inoise to develop in common mode across R7. The CM rejection of the following stage removes both the DC-component and noise component. One can build a AC-coupled line input with modest value film caps that has virtually the same LF CM as mid-band and without significant noise penalty.
"I honestly don't believe that a balanced input should be required to perform with only one input leg connected. The fact that there's little or no output should alert the user that he's done something wrong (you wouldn't expect your toaster to work if you plugged in only one wire of its power cord, would you?)."
That may represent a difference in philosophy.
The fact is that the majority of active line inputs do produce output with an open leg, kinda sort of. Or, should I say in the -6dB case they work "halfway."
The fact that there's no output with a transformer or high CMZ input does let you know something's wrong but not necessarily where it's wrong since you don't often know that the cable you're holding in your hand has actually got signal on it to begin with. So in the interest of the show starting or the performance of a lifetime being recorded (and being the support person carrying a beeper) my hunch is that it's better to get signal under any circumstances at a reasonably consistent level no matter which port you're stuffing signal into. The use of combo TRS/XLR connectors in particular and wrongly-made adapters seem to almost require it.
With regard to the toaster example I'm pretty sure Sunbeam gets those calls.
"As an aside, I'm curious why my (late) friend and colleague Neil Muncy has his 1995 AES paper listed in your references and not mine (in the same issue of the AES Journal)?"
Not sure I follow you there. I didn't cite Muncy.
I've never heard that arrangement of bias resistors called a "T-circuit" but, in audio, it seems everything gets a pet name. Even if just two high-value resistors to ground supplied bias, I doubt they'd contribute significant noise because it will be seriously attenuated by the differential source impedance - somewhere in the 150 to 200 ohm range for a mic preamp. Also remember that i-noises of op-amp inputs are uncorrelated with each other. The i-related voltage noises in R1 and R2 will not be nullified by the following diff-amp. The two i-noises will add as the sq-rt of the sum of their squares in R7, creating a common-mode noise component that will be nullified in the diff-amp. Mic preamps are a bit of a special case. In the case of a passive (dynamic) mic, the source's CM impedances are extremely high, making conversion from CM to DM (noise into signal path) much less of a problem. With a phantom-powered mic, CM impedances are roughly comparable to those at the mic preamp (with phantom power turned on), making conversion more likely. But it's a rather moot point because ground voltage differences tend to be trivial, since the mic (source) is rarely, if ever grounded anywhere but (via the shield of the mic cable) the mic preamp itself. However, mic preamps driven from "mic splitter" outputs can sometimes be a problem (generally only when cheap split transformers without Faraday shields are used - Jensen models use dual Faraday shields for exactly that reason). It should also be obvious that a mic preamp using an input transformer requires no capacitors in the signal path, no transient phantom power protection (diodes, series resistors, ad nauseum), and no RFI filtering - since the transformer provides it all, as well as a match to the amplifier's optimum source impedance for minimum total noise.
"The i-related voltage noises in R1 and R2 will not be nullified by the following diff-amp."
Correct. That's why their values are relatively low.
"The two i-noises will add as the sq-rt of the sum of their squares in R7, creating a common-mode noise component that will be nullified in the diff-amp."
Also correct. That's why it's magic. Should also point out to our readers that a DC-term develops there also in common mode and is nulled out. For the THAT151X and NE5532 examples if Rcm is made too large there will be a reduction in headroom. That sets the upper limit. For a 5532 with a Ibias per input max of 1500nA, 3 uA total, a 1M can produce a Vcm of 3V. An LME49860, with an Ibias per input of 72 nA max (144 nA total) the Vcm with a 1M is less than 150mV.
I prefer to use the "T-bias" approach (and InGenius) with AC-coupled inputs to provide great LF CM because it reduces capacitor matching requirements considerably.
"Even if just two high-value resistors to ground supplied bias, I doubt they'd contribute significant noise because it will be seriously attenuated by the differential source impedance - somewhere in the 150 to 200 ohm range for a mic preamp."
In a mic preamp with coupling capacitors you'll never see a 150-200 source impedance at LF which gives rise to 1/f noise. At 20Hz 22uF (~47uF/2)you'll see an added 360 Ohms to the 150-200R source Z from capacitor reactance. Thus, in a preamp R1 and R2 still need to be kept relatively low.
Bootstrapping an active input to achieve high CM impedance and make it transformer-like has limits due to finite power supply voltages. The CM channel can only bootstrap the input bias resistors as long has it has the headroom to do so. An Ingenius input with the (-) input grounded and the (+) input driven has one-half the headroom compared to an InGenius input driven differentially. This is because in single-ended situations Vcm is equal to 1/2 Vdiff. Though the THAT1206 can accept (with 30V supplies) inputs up to +27 dBu differentially, the limit for single ended inputs is around +21 dBu because the Vcm bootstrap channel starts to clip. The problem increases with lower supply voltages.
Another issue with high CM impedance active inputs occur when single-ended equipment having a "two wire" ungrounded AC power cord is connected to a high CM impedance input that develops leakage current into the high CM impedance. As you well know, potential differences (in 120V countries) can often produce chassis potentials that are 60V (RMS) above ground. In that situation - and I've received support calls about it - the CM channel of the InGenius input clips. The solution in that case - to prevent CM channel overload - is to bond the two equipment grounds. It's not the best solution but we often have to take what we can get to stop hum.
One aspect that has not been explored - and you could provide some insight into this - is the effect on unequal port impedances to ground when they are connected to CT transformer outputs or sources whose op amp outputs may be heavily loaded internally. As the need for lower circuit impedances arise in balanced inputs to reduce Johnson noise in the era of 24 bit conversion, what is effect on unequal transformer loading on second harmonic distortion? With inputs built using 10K resistors I doubt this is a problem. But what about inputs made with 2K resistors? I've never explored it but a heavily-loaded balanced source (CT transformer or ground-referred "push-pull" op amp output) having 2-3 times the load on one leg might give rise to second-order distortion. Cohen in his 1984 mic preamp article used 300 Ohm feedback resistors around an NE5532. For a 5532 this is heavily loaded. His output stage (cross-coupled differential) used 2K resistors. Did Cohen's use of a cross-coupled output reduce THD-2? It's a question worth asking.
If I may be blunt Bill there's no substitute for one of your excellent transformers. Having said that for a low to medium CM impedance active input the double-balanced approach simply uses less "stuff" and works better than the Birt topology or other two op amp approaches.
Thanks for your very kind words about our transformers! As a related note, Jensen is celebrating its 40th year in 2013. First, I never recommend using a center-tapped transformer at an output (or an input for that matter). One of the problems with the simple diff-amp is that there's an intractable tradeoff between resistor noise and common-mode input impedance. Input buffers, as in the classic 3-op-amp "instrumentation" amp, eliminate this tradeoff and very low-noise designs are possible. Since we don't specify any of our transformers for CT operation (except with high-Z secondary loads in a few cases), I have no hard data to answer your question. That being said, small amounts of DC offset ("small" in this context depends on many things like winding DC resistance, number of turns, etc. so it will vary considerably among transformer types) will cause a "redistribution" of transformer distortion products. Without DC, transformer distortion at low levels is almost entirely 3rd harmonic. As DC is gradually introduced, there is an increase in 2nd (and sometimes a slight reduction in 3rd). At higher DC flow, 2nd harmonic dominates (and is responsible for that "warm" bass coloration typical of vintage class A vacuum-tube designs ... Jensen doesn't cater to that market BTW - our designs are as audibly transparent as we can make them). When it comes to inputs, we need to stay aware that a "pro" line-level input must keep its input Z at 10 k-ohm or more. The de-facto standard for mic inputs is 5 to 10 times the source (assumed to be a 150 to 200 ohm mic), making it 1 k-ohm or more typically. Consumer inputs must be 22 k-ohm or more, again by IEC standards. These impedances are *differential*. For balanced inputs, there are huge advantages to making the *common-mode* impedances much, much higher (for a typical input transformer, they are some 50 M-ohm at 60 Hz).
I went back and looked at some old data and the following statement isn't exactly correct regarding the entire 1200-series:
"An InGenius input with the (-) input grounded and the (+) input driven has one-half the headroom compared to an InGenius input driven differentially. This is because in single-ended situations Vcm is equal to 1/2 Vdiff. Though the THAT1206 can accept (with 30V supplies) inputs up to +27 dBu differentially, the limit for single ended inputs is around +21 dBu because the Vcm bootstrap channel starts to clip. The problem increases with lower supply voltages."
In a THAT1200 Vcm = 0.5*Vdiff with unbalanced inputs. Under the same unbalanced conditions a THAT 1203 and 1206 have Vcm = 0.354*Vdiff. The maximum output for a THAT1206 is +24.5, not +27 with 30V supplies which is 4 dB less than the 1246/1286 or cross-coupled circuit. At reduced supply voltages (think USB-powered products) unbalanced inputs produce significant Vcm bootstrap voltages and put the InGenius topology at a disadvantage with less headroom.
What is more important however, regardless of supply voltage, is the issue of AC leakage currents that can develop across the high common mode input impedance which saturate the CM bootstrap channel. So what high CM impedance giveth in terms of reduced source loading with unbalanced source impedances sometimes gets taken away due to AC leakage current. This is particularly true with consumer equipment having RCA outputs and two wire power cords.
Yes, "floating" or ungrounded equipment is problematic. But this problem, as well as the common-impedance coupling that's responsible for hum and buzz in unbalanced audio cables can be avoided in all but the most extreme electrical environments by simply making the correct interface cable. Caution - most commercial ready-made cables are made WRONG, so check them out before you buy one. A proper cable to feed a balanced input from an unbalanced source should always use a 3-conductor (twisted-pair plus shield) cable. The two pieces of equipment (the consumer source and the pro destination) have their chassis grounds tied via the cable shield, thus eliminating the "floating" CM voltage problem. The signal is now carried on the twisted pair and no noisy ground current flows in either signal wire, avoiding the common-impedance coupling of hum and buzz. A proper cable will have normal +, -, and shield connections at the balanced end and will tie - and shield to the RCA outer contact and the + to its center pin. This will, depending on the CM input Z of the balanced input stage, produce typically 30 to 40 dB of noise rejection compared to using an RCA cable (two conductors) and an RCA to XLR adapter. In extreme cases, you may need to add a Jensen (or other quality Faraday-shielded "input") transformer at the balanced end to improve CMRR of the existing input stage.
I have been wondering if there was a "just as good" substitute for a good input transformer, now I understand thatbthe answer is "not really, for all conditions, but yest for some", which is probably useful.
For certain, though, this is one 9of the best and most educational postings that I have come across. Thanks for that!
I have seen some very small input transformersthat made me wonder if they are really intended for auidio. Little tiny toroid transformers, not more than a quarter inch in diameter. Are those the current high quality audio imput transformers?
Incidentally, the difference between an "output" and "input" transformer is the inclusion of a Faraday shield in the latter. While the shield dramatically improves noise rejection (CMRR), it also makes the transformer sensitive to capacitive loading on its output. Therefore, "output" types can be used between a source and a long cable (i.e., capacitive load) and suffer little or no bandwidth reduction, "input" transformers must generally be physically located near the load (input) they drive since, in general, they want to see a load capacitance of a few hundred pF at most to preserve their high-frequency response. The vast majority of low-cost (and some not-so-low-cost) audio transformers are "output" types - they're very easy to manufacture and the best are bi-filar wound to get extremely tight magnetic coupling ... at the expense of noise rejection, of course. As in most of engineering, there is no "free lunch"!
In addition to Mr. Whitlock I wanted to thank the readers who have posted here. WKetel wrote: "I have seen some very small input transformers that made me wonder if they are really intended for audio. Little tiny toroid transformers, not more than a quarter inch in diameter. Are those the current high quality audio input transformers?"
Probably not. For examples of transformers intended for professional audio I'd visit Mr. Whitlock's site, Jensen transformers, or any of the manufacturers listed here:
The one outstanding attribute transformers provide that has not been successfully duplicated is galvanic (metallic) isolation. Though Wurcer and Kitchin wrote in their 1982 Design Idea (ref 7) that the cross-coupled circuit was an "electronic transformer" this wasn't exactly the case.
Tiny transformers are definitely NOT suited for pro audio applications - some, if not most, are suitable only for voice communication or P.A. systems (many actually are telecom or modem transformers). Most transformers have extremely sketchy so-called "specifications" and are very deceptive. The worst abuse is in specifying maximum signal level. The laws of physics dictate that the magnetic flux density in a transformer core is proportional to drive voltage and inversely proportional to frequency. Therefore, a transformer's biggest challenge is to handle large signals at low frequencies ... and this is precisely where the largest signals occur in real music. Jensen transformers are conservatively specified for level handling at 20 Hz, while most of our competitors specify at 50 Hz ... where, all else being equal, the level will be over twice as large. At low frequencies, there is no way around the fact that a larger transformer will handle larger signals at low frequencies. A good example is the widely used Sescon IL-19, a so-called "industry standard". With a 100-ohm driving source (typ balanced output) and a 20 k-ohm load (typ balanced input) and at 30 Hz (I'm feeling generous), its THD will reach 10% at a level of +2 dBu ... that's not even a reference level signal! Headroom? What headroom? Under the same conditions, a Jensen ISO-MAX isolator (such as our model PO-XX) exhibits 0.007% THD and it rises to only 0.2% at +20 dBu. Jensen is an engineer-owned company and we publish the most comprehensive specs in the industry. Most others hide the ugly truth by having either sketchy or no specs at all. There's a lot of junk out there that's unworthy of the "professional" label! - Bill Whitlock, president & chief engineer, Jensen Transformers, Inc.
(The prior post is continued due to character limitations.)
I had previously discussed the low supply voltage performance of InGenius vs. the simple or cross-coupled differential stages and wrote: "The maximum output for a THAT1206 is +24.5, not +27 with 30V supplies which is 4 dB less than the 1246/1286 or cross-coupled circuit. At reduced supply voltages (think USB-powered products) unbalanced inputs produce significant Vcm bootstrap voltages and put the InGenius topology at a disadvantage with less headroom."
To begin with I meant to use the word "input," not output, for maximum levels.
I had a chance to measure the maximum input levels for both a THAT1206 InGenius and THAT1246 with +/-5V supplies. These low supply voltages (low for pro-audio applications) are typically available in USB-powered devices where the +5V supply is inverted using a charge pump or simple switcher to provide -5V.
The THAT1206 operating on +/-5V allows a maximum unbalanced input level of +10 dBu before clipping. The THAT1246, in the circuit of figure 1 or cross-coupled topology (figure 4) allows up to +16 dBu unbalanced input (twice as much) before clipping. The circuit of figure 3, not tested, could also operate at these low voltages if a different dual op-amp were used.
I don't see the real-world need for concern about "headroom" for the InGenius (or any balanced input for that matter) with an unbalanced source. The standard signal reference level for unbalanced (i.e., consumer) sources is 316 mV rms (that's -10 dBV). The +10 dBu (that's 2.45 V rms) maximum unbalanced input level accepted by a THAT 1206 on bipolar 5V supplies now represents 18 dB of headroom ... which is generously adequate (most program material will remain undistorted through a channel with 12 to 14 dB of headroom). Further, operating practically any input stage on such low supply rails would make it incapable of dealing with normal pro levels, whether applied symmetrically or not. This is a bit of a "red herring" issue, since unbalanced outputs (even in so-called "semi-pro" gear) very rarely operated at "pro" levels.
As a point of reference I recently bought a Roland Quad capture USB-powered interface for use in instrumentation. I was quite disappointed to learn that the combo XLR/TRS would only accept levels on the XLR from -60 to -6 dBu. The TRS "line" input accepts -50 to +4 dBu. +4 was not the nominal, it was, unfortunately, the clip point. Though I wasn't surprised by the XLR range, I would have been much happier with it had the line input had greater than +10 dBu capability. Unfortunately, headroom is shrinking along with supply voltages.
We may have to agree to disagree on that one. For the record, I wish the RCA to XLR adapter didn't exist because it encourages folks to use an RCA cable (which intrinsically couples ground noise directly to the signal) for the hookup - and this is true regardless of how the adapter is wired (whether pin 3 is grounded or left floating). I was mistaken regarding my Muncy reference comment - it was another blog at another website! Sorry.
I'm aware of the CM i/p Z problem of convential audio difference amps and how the system CMR gets blatted by cable Z mismatch.
Can you explain the shortcomings of using e.g. INA317, its inputs buffered by 2 x op-amp stages like OPA1642 (to make a classic instrumentaion amp), plus say 4.7MOhm resistors to gnd on each op-amp input for bias?
"I'm aware of the CM i/p Z problem of conventional audio difference amps and how the system CMR gets blatted by cable Z mismatch."
Yes the effect is well documented though I think you may mean source Z mis-match which also includes the cable as well as what is driving it.
What I've never seen in the wild however is an audio tieline that develops (say 20 Ohms) resistive imbalance. I've cleaned a lot of dirty patchbays and cords in my career and replaced switches that may have developed 20 Ohms or more imbalance, or fixed a bad Elco or "DL" connector pin but I was never called to fix it because it hummed due to reduced CM. The call or note from the engineer was that the patchcord or switch sounded "crunchy" or the circuit was open.
I'm also not sure how a "proper" balanced output could develop a 20 Ohm imbalance. Typical build-out resistors might be 47R 1% and have at most a 1 Ohm error. One would have to use 1K build-outs with 1% resistors to approach 20R imbalance.
So my question is how do we arrive at such large imbalances? I realize that with 9-25K Zcm inputs 1 Ohm imbalance is significant but it generally doesn't produce session-stopping hum.
"Can you explain the shortcomings of using e.g. INA317, its inputs buffered by 2 x op-amp stages like OPA1642 (to make a classic instrumentation amp), plus say 4.7MOhm resistors to gnd on each op-amp input for bias?"
I'll let Bill take that question. The circuit you propose is similar overall to this one only the linked citation refers to a bipolar op-amp with T-bias and AC-coupling. It's a simplified circuit:
There's a lot of flexibility in the value choice of input T-bias values and Cin. As shown, the differential -3dB point is 4 Hz. For common mode signals the LF cutoff is approximately 0.1 Hz.
"The reason I singled out cable Z imbalance is because of the capacitance to ground variation of each wire in the cable. I'm thinking about impedance mismatch over frequency, not at dc."
Over long runs the capacitance to shield variations are significant. As a point of reference I measured approx. 1000 feet of "8451-type" two conductor foil shield with drain wire cable and found that the conductors measured 40.9 nF vs. 43.8 nF.
Bill Whitlock points out in Ballou "Handbook for Sound Engineers" 4th ed. that by grounding the driven end shield, and not the receiving end, common mode to differential conversion at the receiver due to capacitive imbalance from the cable is avoided.
But doesn't not grounding at both ends leave the system open to EMI issues?
Interesting info on the capacitive imbalance over 1km. At 1kHz, the imbalance in the two impedances to gnd is about 277Ohms. If you use that circuit you linked to earlier, that uses a 1MOhm to gnd T-network, then does that get you emough CMRR? Thinking the difference amp that follows the buffers can be anything you want, like good old INAxxx.
"But doesn't not grounding at both ends leave the system open to EMI issues?"
Please refer to Muncy "Noise Susceptibility in Analog and Digital Signal Processing Systems" http://www.aes.org/e-lib/browse.cfm?elib=7945
And Whitlock "Common-Mode to Differential-Mode Conversion in Shielded Twisted-pair Cables (Shield-Current-Induced Noise)" http://www.aes.org/e-lib/browse.cfm?elib=12594
As well as Whitlock's "Balanced Lines in Audio Systems: Fact, Fiction, and Transformers" http://www.aes.org/e-lib/browse.cfm?elib=7944
"If you use that circuit you linked to earlier, that uses a 1MOhm to gnd T-network, then does that get you enough CMRR?"
It depends on the source impedance imbalance (not the shunt capacitance to ground if the shield is grounded at the source) and how much CMRR is enough. Do realize that the 1M Ohm could be made larger (e.g. 4M7) approaching the CM impedance of InGenius bootstrapped approaches. The value is primarily limited by the op-amp bias current and the allowable reduction in DC common mode range. For an LME49860 with a worst-case I bias of 72 nA per input, the maximum CM Vos that would develop is approx. 680 mV. Typically at 10 nA per input it would be less than 100 mV. This CM Vos is rejected by the following diff amp and the Inoise that develops across it also appears in common mode.
"Thinking the difference amp that follows the buffers can be anything you want, like good old INAxxx."
Yes it could be an INA134/137 THAT1240/1246 or be cross-coupled (INA2137/THAT1286) to provide a differential output. Alternatively a conventional op amp and precision resistors could be used to lower the circuit impedances and Johnson noise though the resulting CMRR - and the noise performance with a high value of Rcm - might not be as good due to resistor mis-match.
The "Brand-Rex" cable I have measured is undocumented and I haven't measured resistance or inductance per foot.
I do have the published specs for Belden 9451 which is quite similar posted here: http://www.ka-electronics.com/images/jpg/Belden_9451_Signal_Characteristics.jpg
It shows 0.17 uH per foot and 14.1 Ohms per 1000 feet for the conductors.