So you've got a stereo ribbon mic - how do you choose between Blumlein and Mid-Side?

[David Spearritt]

Daniel, with two figure 8's, the subject of this thread, you don't reduce any gains on the preamp because the matrixed levels are exactly equal to the unmatrixed levels and pattern distribution. Remember its not a cardioid in the middle, its an 8. MS Blumlein matrixed is mathematically identical to pure blumlein in pattern distribution and gain, you leave the gains on the matrix amp in a ratio of 1:1, assuming the preamp gains for pure blumlein are 1 and 1 also.

So the analysis a post up from csbarnet is correct. All that happens is the noise adds.

You may not have heard or had problems with self noise, but I have heard it many times as explained above in this thread with harp, lute, classical guitar. Perhaps you have quieter mics than me.

[Martin Kantola]

Originally Posted by csbarnet

1. the sound would be 45 degrees off-axis to both M and S mics (positive lobes), and would arrive at each with the same intensity.

Yes.

Originally Posted by csbarnet

2. since off-axis response of an ideal figure-8 at 45 degrees is -3dB, adding M and S results in no net loss or gain of signal as compared to a Blumlein array in the same spot, since the Blumlein L mic would be pointing exactly in the direction of the source.

Think that since the signal in M and S are correlated (the same) the increase from summing is 6dB, not 3dB. The response goes down -3dB as you say, but we end up with a net gain of +3dB.

Originally Posted by csbarnet

3. the self noise from M and S must be added in each channel

Yes, but the noise is uncorrelated, which gives only 3dB of gain.

Originally Posted by csbarnet

4. since the signal does not increase, yet noise does, the signal to noise ratio does go down for an M-S array, as compared to Blumlein.

As I see it, signal does increase by 3dB (see above), while noise increases by the exact same amount 3dB. So, no change in S/N, only 3dB more total output. Which can be confusing and sound like an increase in noise if not compensated for.

Originally Posted by csbarnet

The 45 degree angle doesn't matter so much...it's just to make the math easier. Obviously, an oversimplification, but I think results could be generalized.

Very good thinking if you ask me, simplified always works best for my mind :-).

Martin

[Martin Kantola]

Originally Posted by David Spearritt

Martin, your Ad hominem attacks on me in this thread are getting tiresome and are not helping your cause.

David, I apologize if you feel I've attacked your person, as that was not my intention. But I simply can't agree with all of your statements, sorry.

Originally Posted by David Spearritt

Here are my credentials.
Bachelor Engineering Mech (Hons), Major Acoustics
Master Engineering Science, (Pure Research) Vibration of Flexible Structures.
12 years Professional Consulting Engineer, Acoustics Noise and Vibration Control.
15 years Professional Software Architect, Engineering and Mining applications.
15 years professional audio engineer and company director
Member Australian Acoustical Society.
Member Audio Engineering Society.

This is very impressive. Did not know your background (on these forums one seldom does), but understand now why you felt personally attacked. Again, I'm sorry. These discussions can get heated, but trying to stay polite and on topic.

Originally Posted by David Spearritt

Please, from now on, address the issue and leave my credentials out of it.

David, with all due respect, think you just crossed a line here...

Originally Posted by David Spearritt

And please do not direct me to Google MS Matrix.

Yes, why don't you direct me to a good third party reference that clarifies and proves your point of view instead?

Peace,

Martin

[mohthom]

Martin, David,

While I can see some of the logic in both of your posts, I would have to say that I agree with Martin here;

Yes, why don't you direct me to a good third party reference that clarifies and proves your point of view instead?

David, quite rightly, suggests a more robust reference. For example, the AES article that forms the core of the reference you give (which could, incidentally, be found just a few hits lower on the same google search) can be found at the following link. Academically speaking, this is much stronger for many reasons (aside from its clarity and more thorough style, it far more closely resembles the points made in the argument and is published; thus at some point it will have been checked for errors and irregularities at some point - unlike a self-published webpage). It is far too easy to tarnish the name of academia with assumptions backed up by wikipedia references and celf-cert websites.

It does seem that this discussion has turned pretty patronising at points, and I can see why David may feel that he has been treated in this way.

Excellent, that's a good way to discuss this.

Martin; the way in which you've taken the role of 'moderator' of the discussion (and, in doing so, agreeing with or dismissed the posts of others - in the position of a reviewer) has afforded you the ability to be the centre of each and every discussion - the way in which you either endorsed or refuted individual claims is not supportive to your argument, and I can wholly understand why David is not happy about this.

David, with all due respect, think you just crossed a line here...

I don't understand how this statement caused David to cross a line - he is merely requesting that you don't refer to his posts as 'embarrassing' - particularly in binary subjects (as this is, in which there clearly is a right and a wrong answer), but in academic debate on the whole - it's either right or it's wrong. You told David to check with someone more qualified (which I have to say, I'd be pissed at you for), and it seems that qualification isn't an issue - it's who is right. It's maths!

David, I don't know that I agree with your maths.

Martin, I don't know that I agree with the way in which you've gone about this discussion.

Can't we have a 24hr cooling off and come back to the table without the pettiness and with some maths instead of insults? I've always found the posts of the two major posters here pretty refreshing, and I don't like the way that this has turned so personal.

MohThoM

[Martin Kantola]

Originally Posted by mohthom

I don't understand how this statement caused David to cross a line - he is merely requesting that you don't refer to his posts as 'embarrassing'

Never said that David's posts are embarrassing! It was the discussion that was getting embarrassing, for both of us. Don't want to keep repeating myself in this manner, but I don't feel I should change my opinion just because he hints I'm wrong and don't understand this issue. David crossed a line for me by first posting his credentials and then asking me to leave them out. I have addressed the issue itself all along.

Originally Posted by mohthom

You told David to check with someone more qualified (which I have to say, I'd be pissed at you for)

Why? I told David to please check with some more qualified than me. Someone he trusts. Realize there might be a misunderstanding here.

Originally Posted by mohthom

Martin, I don't know that I agree with the way in which you've gone about this discussion.

For what it's worth, I've tried to keep it civilized all the way. At one point I teased David a little, but it was only meant to be in a most friendly manner. I'm glad you brought up the two things I said above, as they clearly needed an explanation. English is not my first language, which sometimes shows.

Martin

[aracu]

Originally Posted by d_fu

I don't think I've ever heard mic self noise in a real-world recording situation.

This is the essence of this type of conflict, very little to do with math. Who are you kidding, it comes down to I haven't so you haven't either. You can find this in any thread having to do with noise.

[csbarnet]

Think that since the signal in M and S are correlated (the same) the increase from summing is 6dB, not 3dB. The response goes down -3dB as you say, but we end up with a net gain of +3dB.

OK. This makes sense to me.

[d_fu]

Originally Posted by David Spearritt

Daniel, with two figure 8's, the subject of this thread, you don't reduce any gains on the preamp because the matrixed levels are exactly equal to the unmatrixed levels and pattern distribution. Remember its not a cardioid in the middle, its an 8. MS Blumlein matrixed is mathematically identical to pure blumlein in pattern distribution and gain, you leave the gains on the matrix amp in a ratio of 1:1, assuming the preamp gains for pure blumlein are 1 and 1 also.

Sorry David, I have a strong feeling that you are contradicting yourself here. Your original blanket statement, which brought up the entire discussion, was this:

Originally Posted by David Spearritt

1. MS'ing is noisier than pure Blumlein as a matrix is involved.

This refers to the existence of a matrix per se (in your argumentation, the fact is that the mic's self noise etc. of the S channel is added to the final LR signal twice by the matrix - I'm leaving the question of noisy analog circuitry aside here).

This point (whether significant to the final signal's S/N ratio or not) is IMHO not dependent on whether the mid mic is a cardioid or Fo8 or omni, or is it?
You said above that "matrixed levels are exactly equal to the unmatrixed levels". How so? The only signal that is matrixed/matricised is the S channel, and that does not change WRT the M mic.
If you add S to M for L (and add reverse-polarity S for R), the signal level of the post-matrix L/R channel goes up, does it not?
(BTW, M is also spread across two channels, as opposed to AB, XY, or Blumlein... So in fact both channels are doubled )
If, theoretically, M alone reached 0 dBFS post-AD, you'd have to lower levels to avoid clipping of the resulting L channel, would you not?

You may not have heard or had problems with self noise, but I have heard it many times as explained above in this thread with harp, lute, classical guitar. Perhaps you have quieter mics than me.

Perhaps you get to record in quieter locations than I do...
The only time I have had to exchange a mic tha I had set up was when I tried an Earthworks SR-78 as a spot for a harpsichord. It was unbearable...
While I don't think your comment on my mics was serious, most of what I've got would be in the 10-to-below-20 dB category, but I don't use any 5 dB Rode NT1a mics...

Daniel

[David Spearritt]

Doing a lot of reading, some surprising results, back later today with a summary.

[opegas]

So I am convinced there is no noise added in MS other than that induced by the electronics of the matrix.

I also asked a few key people who couldn't give me a reason why theoretically the two mic techniques would have different noise figures. Both told me the noise could come from the matrix electronics, but thats about it.

So I ran some tests with a R88 (stereo ribbon) in a extremely quiet room with no source, just quiet. with the gain full on output at max on an AEA 380TX preamp/MS box I measured 1.2 dB difference in noise with the matrix switched in or out. Still unsure if it was the electronics or somehow the mic I ran the test again with some 156 ohm resisters in place of a mic. With this test I saw about the same, a 1.1 dB difference in the noise.



Paul

[Corran]

Originally Posted by opegas

So I am convinced there is no noise added in MS other than that induced by the electronics of the matrix.

I also asked a few key people who couldn't give me a reason why theoretically the two mic techniques would have different noise figures. Both told me the noise could come from the matrix electronics, but thats about it.

So I ran some tests with a R88 (stereo ribbon) in a extremely quiet room with no source, just quiet. with the gain full on output at max on an AEA 380TX preamp/MS box I measured 1.2 dB difference in noise with the matrix switched in or out. Still unsure if it was the electronics or somehow the mic I ran the test again with some 156 ohm resisters in place of a mic. With this test I saw about the same, a 1.1 dB difference in the noise.



Paul

Just out of curiosity, why don't you run these tests again but matrix in a DAW. Theoretically there shouldn't be any electronics noise with this correct?

[Ozpeter]

I think you'd have to be careful about drawing too firm a conclusion from a DAW test. For instance, if I take 12 instances of a MS decoder VST and play a normal stereo file through that, inverted against the original (applying 36dB of gain offset) I don't get a perfect inversion - I get a peak difference of about -42dB. On the other hand, if I have 6 pairs of instances of encode/decode and invert that against the original, I get a perfect inversion. The imperfectly inverted audio sounds like a low level version of the original - it's not just noise.

(The point of running it through mulitple instances was to see whether there would be a cumulative effect of some sort which would be hard to spot with a single instance).

[Martin Kantola]

Originally Posted by opegas

So I ran some tests with a R88 (stereo ribbon) in a extremely quiet room with no source, just quiet. with the gain full on output at max on an AEA 380TX preamp/MS box I measured 1.2 dB difference in noise with the matrix switched in or out. Still unsure if it was the electronics or somehow the mic I ran the test again with some 156 ohm resisters in place of a mic. With this test I saw about the same, a 1.1 dB difference in the noise.

With either resistor noise or microphone noise (doesn't matter really) the matrix should have brought up the claimed added noise by at least 3dB, or even slightly more because of internal noise sources. But it didn't, as you noted.

Originally Posted by Ozpeter

For instance, if I take 12 instances of a MS decoder VST and play a normal stereo file through that, inverted against the original (applying 36dB of gain offset) I don't get a perfect inversion - I get a peak difference of about -42dB.

Think the 36dB gain difference is the problem, there is not enough resolution to preform such a huge change without loss of information. Do I understand you correctly that your MS decoder seems to give 3dB gain for each instance?

Asked for help with a better explanation, and got this from Paul Hodges with his kind permission to quote it here:

"Because the noise is uncorrelated, the addition (if they are equal) only raises the level by root 2; but the signal has increased by addition too, so the s/n stays the same. This is how you can do the sum and difference again and regain the original signals.

The way I think of it (saves thinking about correlation and stuff) is that two successive sums and differences bring you back to the same thing - noise and signal - so long as you drop dB each time. So the s/n just /has/ to stay the same overall or the world would fall apart (it might vary slightly between channels though). Or you can think of the MS using two fig-8s, and then the processing (with -3dB) simply rotates the fig-8s by 45 degrees - just a different presentation of the same data in effect."

Hope this helps.

Martin

[d_fu]

Originally Posted by Martin Kantola

Asked for help with a better explanation, and got this from Paul Hodges with his kind permission to quote it here:

"Because the noise is uncorrelated, the addition (if they are equal) only raises the level by root 2; but the signal has increased by addition too, so the s/n stays the same. This is how you can do the sum and difference again and regain the original signals.

Makes perfect sense. If (as David also observed) "MS Blumlein matrixed is mathematically identical to pure blumlein", then this must apply to the entire signal, both the music and whatever noise it contains, IMHO.

[David Spearritt]

Yes, after much reading of papers yesterday, and some more discussions with my custom MS matrix builder and an electrical engineering friend, the seeds of doubt were planted in my mind about my stubborn, overly confident, but incorrect position on this issue. I want to apologise to Martin and Daniel for my resistance to the explanations offered, but they didn't make sense to me. The observation, given to me by my electrical engineering friend, that planted the seeds of doubt in my mind, is also contained in the words of Martin's friend, ie. In a linear operation, like MS summing and difference, no energy or noise is created or destroyed. If the noise could change, it would have to be "absorbed" or created somehow in a nonlinear fashion, and this is plainly not happening, in an ideal lossless matrix anyway.

The explanations using equations and analysis like M+S and M-S are just that, too simplistic, due to music being a mixture of correlated and uncorrelated signals. My reading discovered that this topic is not well understood by most published authors either. Also, it depends on the type of matrix being used, an active or passive matrix. Active matrixes are adding gain during the summation, passive are reducing gain by sqrt(2), as also indicated by Martin's friend. A further complication is that for uncorrelated signals you must "power" add, for correlated, you must voltage add, taking phase into account. This is why there is disagreement on pan laws in mixers, should it be 3dB, 4dB etc.

This point (whether significant to the final signal's S/N ratio or not) is IMHO not dependent on whether the mid mic is a cardioid or Fo8 or omni, or is it?
You said above that "matrixed levels are exactly equal to the unmatrixed levels". How so? The only signal that is matrixed/matricised is the S channel, and that does not change WRT the M mic. If you add S to M for L (and add reverse-polarity S for R), the signal level of the post-matrix L/R channel goes up, does it not?

Daniel, what I meant here is this. If one has an MS config of fig 8's and we assume that the mid 8 gain is set so that its output is 0dB and that the side gain is identical as it must be for Blumlein, then adding the side 8 to the mid will not increase the result to greater than 0dB, because the directivity rolloff of the mid 8 exactly matches the directivity "rollon" of the side 8, so as the source moves from centre to side, the sum always stays at or under 0dB. At 45 off the middle, the two eights are 3dB down and add together to give 0dB, thus being a identical to pure Blumlein.

If there is a cardioid mid, it's directivity rolloff is slower than the side 8 rollon, so the summed result exceeds 0dB, around the middle of the stereo field.

I examined the simplistic equations again after reading a few references. For example, these two references:
http://www.dxarts.washington.edu/cou...son_Stereo.pdf
http://www.neumann.com/download.php?...d=info0039.PDF

show the decoding equations of MS to LR to be L = 1/sqrt(2)*(M+S) and R = 1/sqrt(2)*(M-S).

This is what I have certainly observed with the three matrixes I have, ie. I can leave the gains identical and get the same levels and recording whether the mic is arranged pure or MS Blumlein. There is no doubling of level and its reversable.

However these references, for example, are less than rigorous.
http://www.waves.com/Manuals/Plugins/S1.pdf
http://www.wesdooley.com/pdf/technique.pdf

I have thought long and hard about this issue for the last ten years mainly in trying to explain the things I observed in practice with a Royer SF12, SF24, and Coles 4040 mics into a custom matrix, an AMEK 9098 and a TC Gold Channel, namely that noise and distortion increase with matrixing over pure blumlein. In particular I have had lots of selfnoise issues (so I thought) with these arrangements for quiet classical music. This thread, while somewhat humiliating for me, ultimately has been very welcome to force me to really converge on the issue and sort it out once and for all. Its also taught me to be much clearer when discussing these issues or Daniel will pounce. Thanks again to Martin and Daniel in particular for keeping that hammer going and again I apologise to readers for being so stubborn.

I have just recently traded the AMEK and TC GC for a Forssell SMP-2. Last night I connected the two Coles 4040's up in pure Blumlein, played my classical guitar into it and now cannot hear any self noise at all. The SMP-2 is quite astonishing. Someone further up said this thread is academic. Now I have to wholeheartedly agree.

As you were.

[Ozpeter]

So the s/n just /has/ to stay the same overall or the world would fall apart

My relief that the world isn't going to fall apart is profound! Thanks for the detailed account of your change of position, David. It's been a most interesting discussion, parts of which I almost understood...

[csbarnet]

Thanks to Martin and David: I have learned a lot from this thread. The discussion might be an academic one, but certainly nothing can be lost from better understanding.

I think that it's great to have a forum to discuss issues like these, technical or otherwise. If we are wrong, we learn something new. If we are right, we solidify our understanding by learning how to present what we do know.

I look forward to more in the future. In the meantime, I think I'll pick up a book or two on acoustics. Suggestions welcome!

[Martin Kantola]

David,

there's absolutely no need to apologize for staying firm in your opinion, and I'm glad that you understand now why I didn't change mine. We had an interesting discussion and all learned something new hopefully, I certainly did.

Originally Posted by David Spearritt

...show the decoding equations of MS to LR to be L = sqrt(2)*(M+S) and R = sqrt(2)*(M+S).

You missed a minus sign there, but for clarity we should be able to re-write that as:

L = (M+S) + 3dB
R = (M-S) + 3dB

So I would still drop the level by -3dB to arrive at the final L and R with the same level as the original pure Blumlein. Or?

Originally Posted by David Spearritt

This is what I have certainly observed with the three matrixes I have, ie. I can leave the gains identical and get the same levels and recording whether the mic is arranged pure or MS Blumlein. There is no doubling of level and its reversable.

As you said, it depends on the type of matrix you are using. Some certainly seem to give exactly that +3dB for every plugin instance as noted, some not. +3dB is of course acoustic power doubling, not voltage doubling. But the process is theoretically 100% reversible, yes.

Martin

[David Spearritt]

I think I have had enough of this thread. I am still mistyping things, my quoted post was incorrect. The correct equations should be:

L = (M+S)/sqrt(2)
R = (M-S)/sqrt(2)

Reversability occurs because you matrix back the same way.

M = (L+R)/sqrt(2)
S = (L-R)/sqrt(2)

You missed a minus sign there, but for clarity we should be able to re-write that as:

L = (M+S) + 3dB
R = (M-S) + 3dB

So I would still drop the level by -3dB to arrive at the final L and R with the same level as the original pure Blumlein. Or?

So your equations should be:

L = (M+S) - 3dB
R = (M-S) - 3dB

Not sure why you need to drop level after matrixing. All of my active matrixes make up the three dB each time.

See this reference:
http://www.dxarts.washington.edu/cou...son_Stereo.pdf

and your friends statement about "so long as you drop dB each time"

[David Spearritt]

Originally Posted by Martin Kantola

Some certainly seem to give exactly that +3dB for every plugin instance as noted, some not. +3dB is of course acoustic power doubling, not voltage doubling. But the process is theoretically 100% reversible, yes.
Martin

I think its because the DAW developers read the wrong published papers on MS theory as well and use the straight M+S and M-S calculation method, therefore suffering an unwanted gain change that one then has to deal with in mastering to avoid overs. The Waves reference cited above indicates that this is so.

[Martin Kantola]

Originally Posted by David Spearritt

So your equations should be:

L = (M+S) - 3dB
R = (M-S) - 3dB

Yes, the equation was bothering me ever since my last post, so I looked in the Anderson paper at the equation for rotating a L-R microphone by 45 degrees instead, which is practically speaking what we are doing anyway. By substituting our initial L & R with M & S instead, we get:

L = cos(45) * S + sin(45) * M
R = -sin(45) * S + cos(45) * M

Which gives approximately:

L = 0.707 * (M+S)
R = 0.707 * (M-S)

Looks like our MS matrix, doesn't it? And multiplying voltage by 0.707 is the same as a -3dB drop, isn't it?

Originally Posted by David Spearritt

Not sure why you need to drop level after matrixing.

If the matrix is done correctly, we don't. But the matrix needs to sum and drop -3dB internally. That should be it I hope, phew...

Martin