CMRR, what is it?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
In laymans terms, CMRR is the ability of a stage to not be influenced by a signal that is common to both the signal and ground* inputs to it. So something with poor CMRR, if the signal and ground both have say 1 V RMS sine on them, the output of the stage wil 'bleed through' a small amount of that signal. Which is obviously unwanted.

*Yes I know it may not strictly be ground, but I'm trying to keep it simple :)
 
Think of the amplifier as an opamp with an inverting and noninverting input. The master equation (ideal) is Vo = A (V+ - V-). So ideally, if the inverting and noninverting input were tied together and driven by a signal, the output would be zero. CMRR is a measure of how close to zero the actual amplifier is with this connection.

For a transformer, the CMRR is pretty much the same thing- connect the primary ends together and drive them with a signal. Whatever appears on the secondary is an error voltage, and reflects the imperfect common-mode rejection.
 
Hi,
take any normal two tone input signal. Apply the big signal to the front end and then add the small signal on top of it.
The output will not be an exact replica of the input. Some of this diistortion is due to common mode error at the input and some from later stages in the amplifier.

If one can reduce the common mode error (by increasing the CMRR) then there is a good chance that the output distortion will also be lower.

Leach and Self have a good explanation on solutions for the front end reduction of Common Mode errors.
 
lumanauw said:


How do CMRR affect audible sonics? By what mechanism?


Such questions are never easy, but maybe this one is more profound than it appears at first sight. It is easy to just parrot the textbook answer that high CMRR is desirable since it is the differential voltage we want to amplify, not the common voltage. However, that is the general answer, that goes for diff pairs in general, with no particular application in mind. I do not claim to answer the question below, but just elaborate a little bit on what theory seems to tell us. And please pardon me for the amount of text. I have deliberately tried to avoid math in order not to scare people away.

I often find it useful to introduce two new variables, Vc and Vd, and rewrite the two input voltages to the diff pair as:

Vi1 = Vc + (Vd/2)
Vi2 = Vc - (Vd/2)

This equation system always has a solution. Obviously, Vc is the common voltage and Vd the differential voltage. (How fortunate I happened to use subscripts c and d. :) ) We typically want to amplify the Vd/2 components, but not the Vc components.

However, if we consider the application where the diff pair is used as the input of an amplifier with a single, non-differential input connected as Vi1 and the feedback signal connected as Vi2, the we note something interesting. Vc and Vd are very strongly correlated. If we assume an ideal amplifier with no noise or distorsion, then Vd will be a scaled (actually much smaller) and possibly phase shifted version of Vc. And Vc will mimic the input signal at just a slightly lower amplitude and possibly a iota of phase shift. In principle, it seems that amplifying Vc too wouldn't cause too much harm, since both Vc and Vd are basically the signal we want to amplify. However, if CMRR is too low, the Vc component will no longer be a negligible part of the output voltage of the amp, and will affect the feedback since only the differential voltage contributes to correcting errors in the amp. Usually, even a bad CMRR should be sufficient for this not to be a problem, I think.

What else might happen? Vd is amplified differentially and the two halves of the diff pair cancel each others errors to some extent. That manifests itself in the well-known S-shaped transfer curve of diff pairs, which is very linear for small differential voltages (at the expense of quickly becoming very nonlinear for higher voltages). So for small voltages, a diff pair is much more linear than a single-transistor stage. Now consider the Vc components of the inputs. Clearly the diff pair is identical to two separate CE stages for common mode voltages. Hence, the Vc components are amplified according to the usual transfer curve of the device (eg. exponential for BJTs). So one thing that seems to happen is that even if amplfying the Vc components seem not to cause much harm, there will be much more distorsion from amplifying Vc than from amplifying Vd (in relative terms). On the other hand, the CMRR is usually high enough to make the fundamental caused by amplifying Vc start out at a much lower level than the fundamental caused by amplifying Vd, so in absolute terms, we can probably expect more distorsion from Vd. Maybe the distorsion from Vc is not negligible, however, and maybe it has a more intrusive spectrum? Maybe this could be a starting point for understanding if and why CMRR is important?
 
Hi, Christer,

I will try to understand your post above. I don't get much from first reading :D.

Concerning PSRR, there a schematic that is not often used, but I think it is usefull. The VAS is differential Q66-Q67, and the most interesting part that it is supplied by a CCS (Q68).

http://www.diyaudio.com/forums/attachment.php?s=&postid=1220180&stamp=1180358772

Many uses differential VAS, but only put one resistor between emitors of Q66-Q67 to rail.

Only one resistor to power Q66-Q67 has a slight different operation than if we put CCS (Q68) for powering the VAS differential, because the common drop of R69-R70 will not change the VAS current (like if we use only one R in place of Q68 CCS).

This way, the VAS will only sense the differential voltages of R69-R70, and not amplifying the common voltage (depending on the quality of Q68-CCS)

Why this kind of CCT (differential VAS with CCS powered) seldom used?
 
Christer said:

snip .....Maybe the distorsion from Vc is not negligible, however, and maybe it has a more intrusive spectrum? Maybe this could be a starting point for understanding if and why CMRR is important?


Questions like this got me thinking about using seperate feedback loops for the CM signal(Vc) and the differential signal(Vd). It seems to me that a common mode signal error at the output would mis-balance the input differential by way of the NFB loop.:dodgy: :smash:
 
lumanauw said:

I will try to understand your post above. I don't get much from first reading :D.

I admit it is not the clearest thing I have written. :)

Would you prefer it in mathematical formulas rather than text?


CBS240 said:

Questions like this got me thinking about using seperate feedback loops for the CM signal(Vc) and the differential signal(Vd).

I cannot really see how it would be possible to distinguish the two components and even less how to feed them back separately. If there is actually a problem of the kind I speculated about, then the remedy is to have a high enough CMRR. My previous post was just an attempt to understand how a lowish CMRR might possibly impact on the sound. However, it is well known how to achieve high CMRR, so it shouldn't really be a problem. The exception might be that some people insist on using a resistor only instead of a CCS. On the other hand, some claim it sounds better that way, which is further food for thought: Does that perceived improvement have anything to do with lower CMRR?
 
Hi Christer

It is odd sometimes what people think sounds good as it is really left up to opinion. Beauty is in the eyes of the beholder, some say. If only there was a perfect amplifier, perfect speakers, and a perfect listening room we all could use as a control for comparison. :rolleyes: It seems certain music sounds better with certain distortions to certain people.:D IMO, very low distortion in general and lots of headroom is desirable for the classical genre.:dodgy:

In the differential VAS amp with a CCS bias differential input, any common mode signal, due to phase shift or whatever, should change the amount of current in both transistors of the differential VAS. IOW, the tail current of the VAS is dependent on common mode and tail current bias of the input stage. If the VAS tail current is biased with a CCS, then the voltage change on that CCS(emitters of VAS) should be dependent on the common mode current change & tail current of the input stage, leaving the circuit with no real stable reference for a DC Q point. Sensing this change in voltage with a J-fet gate, so as not to interfere with the bias current of the VAS, a feedback loop could be established by referring this signal back to the input CCS so it has reference to the common mode operation of both stages, and the VAS CCS which is a constant...
http://www.diyaudio.com/forums/attachment.php?postid=1214613&stamp=1179682044 ...

My :2c:

On this subject I'm beginning to feel like I'm getting stranded on thin ice, and it's starting to crack.:smash: :) There are quite a few, more knowledgeable folks here than me and I haven't seen much that resembles this in a power amp topology....:dodgy: CMRR may not be a major issue as reason it is not discussed much here. In reading some of your posts, it is obvious you are perspicacious in the subject of SS, your thoughts?
 
This is a clever circuit. It is from patent #5,568,561. In left picture, if the +in and -in are not in the same magnitude (180deg phase difference), there will be difference in drop on R1 and R2.
Making the CCT like in the right picture, the drop on R1 and R2 is adjusted to be the same, no matter if +in and -in are not in perfect balanced magnitude.
 

Attachments

  • jensen.jpg
    jensen.jpg
    28.3 KB · Views: 180
"Is that CMRR is partially responsible for some of the sonic quality terms (beautiful words for describing how a power amp sounds) like we used to read in reviews?"

yes beautiful words :D but I had my enough share of reviews, following what the reviewer says which gear sounds good, but turns out to be major dissapointment. Review is still useful .though ..sometimes.
 
using op-amp inside the feedback loop is ok as long as proper compensation/ caution is used. The circuit shown didn't show all the compensation details though.

"a discrete follower with high input impedance and quite low output impedance will do the work"

problem is matching the capacitance in the input up to high frequency.
Including stray capacitance, so differential amp is better used with matched transistor in one package, or differential (instrument) op-amp.

EDIT : By the way, op-amp in feedback loop increase distortion, compared to resistor.


Cheers,
Hartono
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.