Ferrite Beads on Output Pins 2, 3

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
Without the bootstrap, each 470 pF cap has an impedance around 16 kΩ at 20 kHz. This makes the CM input impedance of the circuit 1,000 times lower than it is at mid audio frequencies. This makes CMRR exquisitely sensitive not only to source impedance imbalances but also to the matching of those capacitors - even if the driving source had perfectly matched 50 Ω impedance in each leg. Remember that to get a 90 dB CMRR figure, the impedance of the two lines must match within 0.01% or better. But the imbalancing effect on the receiver's own CM impedances can be reduced by effectively reducing the capacitor size (raising it impedance) by bootstrapping. I don't recall the exact numbers, but the bootstrap reduces the 470 pF down to the area of a few pF in the audio frequency range, making their impedance rise to over 1 MΩ. So they make the degradation of CMRR due to source impedance imbalance at 20 kHz roughly comparable to the 10 MΩ figures at mid audio frequencies. One of the side effects is that a 5% tolerance at 470 pF has about 100 times less effect - making its effect on the circuit about the same as if it had 0.05% tolerance. There's a lot more to the circuit behavior than meets the eye ... and, as I've demonstrated here, it's also hard to explain. But, under identical conditions of source impedance, the InGenius circuit will, above about 5 kHz, have better CMRR than the best of Jensen's input transformers. The THAT website has some good references explaining why extremely high CM input impedances (10 MΩ) work to vastly reduce the CMRR-degrading effects of source imbalances. It's nearly 1,000 times less sensitive than an ordinary diff-amp built with an op-amp and matched 25 kΩ resistors. The bootstrapping of the RF filtering capacitors is a way to add RF attenuation while keeping the CM input impedances very high.
 
To echo Bill's comments, in broadcast, it was common practice to make sure the CMRR was adjustable so it could be trimmed. With the older API 318 distribution amp, it had a cap and res that was adjustable, we had a CMRR of -105dB at 90Khz. This was because many of the trucks or stations were very near antennas. When they tested the racks, they would stuff a Walkie Talkie into the rear wires and click it on and off. If they could see it on a scope, you failed. What's even more important in audio is phantom power protection if you aren't using a transformer. The series resistors (3) can be from 10R to 100R. The attached pic also has caps for the HF rejection, 1000pf may be high, 100pf would be fine. The phantom circuit has the 510R so their is a slight ramp and to it doesn't arc the switch. I always put the phantom power LED (in series with 20K) on the cap side, to show real 48V, and to show that if 48V is coming from another location, like a split, it will light up. If that happens, you push it in there as well. Thats mostly for live, but it saved my ass once at a Dave Matthews gig. Ferrite beads don't give you much unless you're trying to get CE approved...

Of course, if you have the CMRR in pin, incorporate the earlier circuit into it.
 

Attachments

  • cmrr and 48.gif
    cmrr and 48.gif
    41.8 KB · Views: 54
Well this is the part I didn't get but indeed it is 16k and I must admit I don't fully understand why. I guess I'll just keep leaning on LTSpice. Great info. Thanks.

View attachment 81551

Z = 1 / wC (where w = 2piF). Or -j / wC to indicate the complex / vector nature of the result.
Okay - I haven't got a 'scientific' keyboards set up on this PC atm.
 
And I'm still struggling to see how two 470p caps are going to be that different. Use 1% NP0 (GCM31A5C3A471FX01D). If you REALLY want to get pedantic about it, use a capacitor array (CA064X102K2RACTU). Even though it's 10%, the difference between multiple elements of the same part should be very low I presume.
The strange truth is all alleged "non-polar" capacitors do in fact have a polarity. All of the famous dipped capacitors (Red, Orange, Yellow, Green, etc) and their axial cousins all have a polarity. The markings on them (stripes, spots, etc) are only there to assist to ensure transport to packaging (vibratory bowl feeders, etc) are facing the same way. Via photo eyes and the like provide the delivery systems that read/see these makings and correct any misalignment. All this to make ensure all the capacitors face the same way on tape reels and the like used in automatic insertion systems. A good video that explains all this and how to build an inexpensive tester. Which will identify the "groundier" side of the capacitor. That side needs to be aimed at and/or through to the path to ground. The video:

"Are Your Capacitors Installed Backwards? Build this and find out." by Mr. Carlson's Lab

at

This may be part of or all of your problem.
 
Last edited:
The strange truth is all alleged "non-polar" capacitors do in fact have a polarity.
A bit of an overstatement. Some "non-polar" caps are built from stacked layers (or a single layer). MLCCs, stacked plastic film/foil and most silver micas don't have an outside foil and thus orientation doesn't matter. What you're referring to here are caps with "wrap and fill" type construction. When you hook the cap up to the scope and struggle to see a difference in noise between leads, you're likely dealing with a stacked construction cap.
 
The strange truth is all alleged "non-polar" capacitors do in fact have a polarity.
Calling a dissymetry "polarity" is not only scientifically incorrect, but also misleading.
Seems to me this guy in the video does not mention polarity, he just mentions backwards.
He's correct, but does he really need to display his entire collection of T&M to be taken seriously? For me it has exactly the inverse effect.
 
Well this is the part I didn't get but indeed it is 16k and I must admit I don't fully understand why.
What do you mean by "did'n get it"? I know you know how to calculate the reactance of a capacitor. I guess you don't understand that Bill is comparing the CM Z between a standard arrangement where the caps go to "ground/chassis/earth", and the arrangement where they are boostrapped by the CM pin.
If I'm wrong, say it and I'll let someone else answer. Anyway I think Bill is the most apt at answering that.
 
I don't fully understand why.

The impedance of capacitors decreases with frequency, lower and lower with higher and higher frequency. With real capacitors (as opposed to model capacitors described in a beginning textbook, or in SPICE without including parasitic behavior) there is a limit, because any device with finite physical size will have an inductance associated with the geometry of the current flow through the device. That means capacitors have an effective inductance in series with the capacitance, in rough general terms higher inductance the larger the package, so at some frequency the inductance will become dominant and the impedance will begin to increase again.
For frequencies below that point the impedance is pretty close to 1/2*pi*f*C where f is the frequency you care about and C is the capacitance.
Plug that in for 20kHz and 470pF cap, 1/2*3.14*20*10^3*470*10^(-12)=16939 ohms.
Those are in series for a differential signal, but parallel for common mode signals.
 
What do you mean by "did'n get it"? I know you know how to calculate the reactance of a capacitor.

Honestly I don't really love getting deep into the mathematical models and such. I suppose I lean on LTSpice a little too much sometimes. But as long as I stop and think when I come across something new, it's ok.

I thought that the impedance would be proportional to the reactance. Meaning if I do a simple RC plot with 100R, Fc is 3MHz and so I assumed the AC impedance of the cap at 20kHz would be "high". But clearly that is not at all the case. If I adjust the source impedance to 16.9k, then Fc is 20kHz. So source impedance has no effect on the impedance of the capacitor.

Ho hum. Live and learn.
 
Last edited:
The impedance of capacitors decreases with frequency, lower and lower with higher and higher frequency. With real capacitors (as opposed to model capacitors described in a beginning textbook, or in SPICE without including parasitic behavior) there is a limit, because any device with finite physical size will have an inductance associated with the geometry of the current flow through the device. That means capacitors have an effective inductance in series with the capacitance, in rough general terms higher inductance the larger the package, so at some frequency the inductance will become dominant and the impedance will begin to increase again.
For frequencies below that point the impedance is pretty close to 1/2*pi*f*C where f is the frequency you care about and C is the capacitance.
Plug that in for 20kHz and 470pF cap, 1/2*3.14*20*10^3*470*10^(-12)=16939 ohms.
Those are in series for a differential signal, but parallel for common mode signals.
I wrote about this in my old magazine column back in the 1980s. Capacitors have multiple non-ideal characteristics. ESR is the equivalent series resistance and modeled like an R in series with the ideal C. Likewise ESL is the equivalent series inductance, modeled similarly. It is not much of a stretch to consider real world caps like little tuned circuits. If you look at an impedance plot vs frequency you will see a zero (minimum) caused by the ESR, and a pole caused by the ESL. Capacitors also have other harder to model non-ideal behaviors Like DA, DF, voltage coefficient, etc .

JR
 
Honestly I don't really love getting deep into the mathematical models and such. I suppose I lean on LTSpice a little too much sometimes.
Clearly I was optimistic about your understanding of EE fundamentals.
How you can digress about CMRR without this knowledge is beyond me.
I thought that the impedance would be proportional to the reactance.
For a single perfect capacitor or inductor, yes.
Meaning if I do a simple RC plot with 100R, Fc is 3MHz and so I assumed the AC impedance of the cap at 20kHz would be "high".
Remember we're dealing with CM impedances of 10 megohms there.
 
Last edited:
I'm glad you understand that the filter bootstrapping is there only to minimize the drop in CM impedance at high audio frequencies. I should have made that more clear. Once the low-pass filter makes its transition, radio frequencies are attenuated completely passively via all three capacitors. Those capacitors, as well as the 4.7 kΩ resistor should be mounted as close as possible to the connector. This prevents radiation of RF into the device interior and, of course, provides a low impedance path to the panel for RF. The bootstrapping doesn't participate in RF attenuation at all, so it's not a "cancellation" scheme. I find it interesting that, for its newest switching regulator ICs, TI is promoting an "on-chip" which is basically the old capacitance multiplier technique using an op-amp - which incidentally works at frequencies where op-amp gain is >1).

The "Y" configuration of the RF bypass capacitors is a way to reduce the capacitance mismatch in the two legs. The matching is improved by approximately the ratio of the across-the-lines pair to the grounded one - in this case about 5 to 1. So two 5% parts will behave as 1% parts in their effect on CMRR. It could be argued that the effect of mis-match in these caps is trivial, but when trying for 90+ dB of CMRR, little things matter. The "Y" capacitor configuration also reduces part count - each of the caps could be bootstrapped individually, but it would require one additional C and one additional R.

It could also be reasonably argued that the mis-match of these caps is trivial when you consider a common, and much worse, degradation of CM impedance balance caused by cable capacitances. As I pointed out in my 1995 "Balanced Lines ..." AES paper, the conductor-to-shield capacitances of common STP cables are typically mis-matched by about 5% (indirectly because they're different colors). These capacitances are effectively in parallel with the receiver's RF capacitors discussed above. But, in a purist world, if the cable shield is grounded only at the send end, they don't do this because the cable capacitances become part of the source impedances. I talk about the reasons for this in a topic called "Which end to ground" in my seminars. Again, these degradations are relatively small and would only become audible under some special conditions - so industry "standard practice" is to ground the shields at both ends. But, to get every last dB of CMRR in any balanced interface using a shielded interconnect, the shield should be grounded only at the line driver end and left floating at the receiver end.
 
I meant to include in my previous post that this issue of RF suppression capacitor placement was discussed in our AES working group as we created AES48. I argued that the capacitors could be made integral to the XLR connector body to keep attenuation very high well beyond 1 GHz (which seemed very high at the time - but now we have 5G upon us ....). And, since a representative from Neutrik was also part of the working group, the seed was planted for the Neutrik "EMC" series of XLR connectors. They use a tiny PCB with an array of surface-mount capacitors that connect pin 1 to the housing. They also have an integral ferrite bead to help even more to keep RF on the outside of the enclosure.
 
Now, in order to reduce diff-mode RFI, one has either to increase the series resistors or replace them with inductors, I believe.
Am I right thinking the 120x can withstand RFI without rectification, so filtering could be placed after it?
 
The "Y" configuration of the RF bypass capacitors is a way to reduce the capacitance mismatch in the two legs.

There is a style of cap sold under the trademark "X2Y" originally sold by Johanson Dielectrics but now also licensed by at least Yageo, which has a 3 terminal package, with the capacitor plates interleaved so that you get capacitance between all three terminals. If you connect the outside terminals to the signals, and the middle terminal either directly to shield or to a third capacitor to shield you have the two caps in the "Y" figure in a single package. Because the two capacitor structures are made together and interleaved, the matching between sides is better than the capacitor tolerance. The information I had a few years ago was that the caps sold as 20% tolerance were matched better than 5% side to side. I'm not sure how much "better" was, I think they didn't want to guarantee any closer than that, but could be quite a bit closer depending on how consistent the manufacturing process is.
I have not seen them used in any commercial gear that I have looked in, but seems like it would be worth the few extra cents to try out.
Something like this would add about 10 cents per input, so Behringer isn't going to do it, but I don't see any reason that you would care on a low volume product or DIY project:
470pF 50V C0G X2Y cap at Digikey
The X2Y info starts on page 10 of the datasheet linked from there:
Johanson cap datasheet
You have to be a little sensible with layout to get the optimum from that package, but that is true of layout for single cap packages as well. Johanson has a good app note on optimizing layout for best results, nothing surprising if you have even basic understanding of inductance of different layout patterns.
They have an online tool to plot the impedance for a particular part. That 470pF part I linked has a self-resonance at 410MHz according to their plotter. This is the page for the plotter, put in the part number (500X15N471MV4E for that example I linked) and click the "add" button and it will plot the insertion loss, presumably with 50 Ohm source impedance.
S21 plotter for Johanson caps
 
Last edited:
I don't think that is the case for any device with semiconductor junctions, the physics of PN junctions is always going to rectify RF that you put in the input pins.
PN junctions in active circuits that are fast enough to amplify/pass RF cleanly do not rectify (decode) the RF. Rectification occurs when circuits slew limit. Rectification is common in high gain input stages that could ignore a few mV of RF but when the gain stage tries to boost those mV of RF 100x, some slower circuitry can't keep up.

JR
 
There is a style of cap sold under the trademark "X2Y" originally sold by Johanson Dielectrics but now also licensed by at least Yageo, which has a 3 terminal package, with the capacitor plates interleaved so that you get capacitance between all three terminals. If you connect the outside terminals to the signals, and the middle terminal either directly to shield or to a third capacitor to shield you have the two caps in the "Y" figure in a single package. Because the two capacitor structures are made together and interleaved, the matching between sides is better than the capacitor tolerance. The information I had a few years ago was that the caps sold as 20% tolerance were matched better than 5% side to side. I'm not sure how much "better" was, I think they didn't want to guarantee any closer than that, but could be quite a bit closer depending on how consistent the manufacturing process is.
I have not seen them used in any commercial gear that I have looked in, but seems like it would be worth the few extra cents to try out.
Something like this would add about 10 cents per input, so Behringer isn't going to do it, but I don't see any reason that you would care on a low volume product or DIY project:
470pF 50V C0G X2Y cap at Digikey
The X2Y info starts on page 10 of the datasheet linked from there:
Johanson cap datasheet

Thanks for the tip! Better matched caps here for the InGenius will help the 20 kHz CMRR. I remember seeing caps like this decades ago but didn't realize they'd come back in SMD form! And I love that the self-resonance is so high - high enough that even a small ferrite bead on the inboard side would help with RFI at GHz frequencies. Hopefully, there won't be much up there because typical STP cable is pretty lossy at those frequencies - but it also depends heavily on how good the XLR pin 1 to chassis connection is! The Neutrik EMC connector is pretty good up to a few GHz I'm told by colleagues in the AES standards group (I don't own gear that gets up to those frequencies myself). And speaking of gear, I'd take anything Mr. Carlson says with a huge block of salt! Sadly, he's very typical of hundreds of self-appointed "experts" on the web - but followed by high numbers of "cut and paste" "engineers."
 

Latest posts

Back
Top