Mid Side M-S Matrix Uses No Precision Resistors

Where we discuss new analog design ideas for Pro Audio and modern spins on vintage ones.
Gold
Posts: 678
Joined: Sat Dec 01, 2007 5:20 pm

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by Gold »

In the above drawing it's not clear whether the AD797 is inverting or non-inverting. If inverting there should be another inverting opamp before the pot or some way to flip the polarity so absolute phase is maintained. It would take a 3 pole switch to flip the input to the 1240 along with the variable gain. This would be impractical to do if a rotary switch is used instead of a toggle and a pot. Sorry if I missed something or am stating the obvious.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

In the above drawing it's not clear whether the AD797 is inverting or non-inverting.
It's non-inverting Paul. So there's no additional re-inversion required.

Because it's non-inverting, the CM performance of that stage requires careful op-amp selection. That's why we looked at the AD797 vs. a National LMEXXXX. I've never used a 797 in commercial designs because they were expensive but I checked Self's site to confirm the performance of the 797. The LMEXXXX and most other 5534-class parts work best in inverting topologies. I'm gonna prototype that stage and I'll probably have to use null testing to see what the distortion contribution is. I suspect the 797 will be quite good.

WRT pots vs. switches. I would prefer stepped switches too but, as shown, a builder could do it either way.

Keep in mind everyone that +/-3 dB over 300 degrees rotation provides a very wide "0" dB center. If the pot was log/rev log around center, and it was linear dB, it would be about 50 degrees/dB. That taper only exists in our dreams. With it being linear taper however, the center will also be quite wide - providing a "wide bearth" around unity gain. Putting AOT (adjust on test) resistors on one side of the wiper to accurately set the center point one could easily have repeatability. A bigger concern is having the +1 and +2 settings being squashed too far near the ends. Again I think it would be worthwhile to build one with a linear pot and see if we can bow it enough so it has a good feel along with a wide 0 dB center.
Gold
Posts: 678
Joined: Sat Dec 01, 2007 5:20 pm

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by Gold »

mediatechnology wrote: Because it's non-inverting, the CM performance of that stage requires careful op-amp selection.
Although it's OT could you give a little lesson on why this is important and what about the 797 makes it a contender?
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

I decided to prototype a quick +3 dB gain stage with a 5534 to check the resistor values. With 6K19 and 15K at their ideal values we hit 3 dB within 0.08 dB. I confirmed exactly 3 dB via measurement. THD isn't bad although I really don't have the instrumentation to measure beyond 6KHz. Near the clipping point it measured around ~0.002%.

I didn't include the pot so the source impedance in these tests is significantly lower but still not bad for a 5534. As drawn, there is bias current in the pot, and if switches are used there particularly needs to be some Rbias to keep the input nailed down during switching. It needs to be fairly low to reduce clickiness. There's no real harm in AC-coupling it if it becomes a problem.

When I get the chance I'll do two more things. One will be to play with and check the pot's taper. For my firsts tests I didn't break out the expensive Champagne (an AD797) and used a humble 5534. An AD797 may require some compensation tweaks and resistor scaling (lower) from what we show here.

Image
https://www.ka-electronics.com/Images/j ... dB_Amp.jpg

Paul: To answer your question an inverting op amp avoids common mode linearity limitations because both inputs are at "ground." The non-inverting input is tied to ground (usually) and the inverting input is held at virtual ground by feedback. A voltage follower, a non-inverting stage with a gain of "1," requires both the non-inverting input and inverting input to track the entire range of the input voltage. With supply voltages of +/-15V, a 5534 has a CM range of about +/-12V and it's linearity over that range is pretty good. An LM4562 for example has a higher CM range, but it's linearity over that range isn't as good. An LM4562 has significantly higher distortion in non-inverting applications than inverting ones.

See: http://www.dself.dsl.pipex.com/ampins/webbop/LM4562.htm

Other op amps, the TL07X series for example, have even worse CM linearity (and polarity reversal) which increases distortion in non-inverting configurations. Most of the op amps Self tested show the non-inverting CM issue to one degree or another.

As gain is increased in a non-inverting topology the required common mode range is reduced because the inputs are operating at lower level. The feedback resistors are a voltage divider. At a gain of 10 for example the CM range is 1/10 what it would be compared to a follower. (edit-orig inverter. Hey -it was early.)

The AD797 (a bipolar part) has a lot of things going for it among them very good common mode linearity. In Self's test at a non-inverting gain of 3.2 (~10dB) the output of the AD797 is virtually indentical to the genertor.

I also looked at the AD823's (Bifet) CM performance comparing it to an OPA604/2604. The OPA2604 also had more pronounced THD than the AD823 in the non-inverting configuration. For reasons I don't understand, the AD823 isn't particularly popular in audio. I don't think it's quite been "discovered." Most of the online audio cites deal with the AD823 driving headphones directly - something it's output stage just doesn't have the capability to do. The AD823 however, for a Bifet, measures quite well.

John R can provide a better explanation than I can about this - and probably use less words doing it.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

Yes Roger Rbias will factor into the precision of the attenuator as the pot (or switch) is rotated towards the -3dB position. And, it will shift the 0 mark as drawn. Consider that portion visual short-hand.

I originally considered using two Rbias resistors. Both connect to the wiper. One connects to the CW side of the pot the other the CCW side. See my earlier drawing that was inverting for how that would work.

http://www.picocompressorforum.com/foru ... 2&start=42

For a bipolar part, such as the 5534 or AD797, the 220K would be too large. This keeps the 0 dB point consistent (and allows AOT trimming.)
Gold
Posts: 678
Joined: Sat Dec 01, 2007 5:20 pm

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by Gold »

Thanks Wayne. I'm glad best performance is the goal.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

Thanks Roger.

I still need to prototype the AD797 version to check stability. It will quite likely have a 619 ohm and a 1K5 Rfb network.
Mastertone
Posts: 11
Joined: Sat Jan 05, 2008 6:18 pm

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by Mastertone »

So, any progress on this one?
Jonas Ekstrom
Mastertone
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

Hi Jonas -

I got the prototype board from Roger on Thursday and hope to be stuffing it today. We've had some really good weather in Texas the last couple of days - 75 and sunny - and I needed to take advantage of that. Today it's 28 degrees with a high of 38 so doing work in the shop with a hot soldering iron and warm instruments seems just like the ticket. I'll post something here later today.

Wayne
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Mid Side M-S Matrix Uses No Precision Resistors

Post by mediatechnology »

When I originally designed the matrix I was concerned about gain build-up during encoding in the Mid (L+R) channel and also in the decoder which produces not L and R, but 2L and 2R. In anticipation of that I designed the following circuit and then allowed myself to get talked out of using it off-forum. This is the original concept for the decoder:

Image

Yes, 2L and 2R. The prototype, using the same circuit for encode and decode had 6 dB of gain end-to-end. So I modified it using THAT1246s in the decoder and an added inverter.

Image

The encoder is on the left. Balanced connections loop from the encoder Mid and Side outputs back into the decoder located on the right. The decoder has a THAT1240 added. Left and Right go in on the left-hand side and come out on the right-hand edge of the board.

As it turns out the encoder doesn't have headroom issues because the input is attenuated 6 dB by the input line receivers before being summed and producing, with fully-correlated mono inputs, +6dB gain in the L+R channel. I measured better than +20 dBu headroom end-to-end (single-ended I/O) using +/-18V rails. (At the output of the encoder Mid and Side levels are at their "natural" ratios and have plenty of headroom.)

The decoder, with unity-gain THAT1240s in the de-matrix produces a net 6 dB of gain. When encode is looped through to decode for an end-to-end connection, there is 6 dB of overall gain. This is unacceptable. The solution is to use THTA1246 -6dB parts in the de-matrix. The original decoder circuit originally posted elswhere in March 2007 still makes sense. I shouldn't have allowed myself to get talked out of using it. This restores unity gain operation.

There's just one small problem. A THAT1246 can be used to subtract inputs with a gain of 1/2 by using both inputs. The summation of inputs with a 1246 isn't directly possible. The non-inverting input resistor (internal) and the reference resistor have different values.

Summing inputs using a unity-gain THAT1240 is easy: Pin 3 and pin 1, the reference connection, can both be used because the resistors are of equal value. With a THAT1246 however, summation, to produce the "L" signal, requires using pin 2. Double-negation, by inverting the inverting input, thereby provides a sum. The prototype has a THAT1240 added to serve as a precision inverter. Roger has already done a new layout with the added IC.

The 2L term comes from left being derived from Mid+Side. 2L = (L+R)+(L-R). The R term drops out and we get 2L. We need one "L" so we use the 1246 gain of 1/2 to get 2L/2.

Likewise the 2R term comes from right being derived as the difference of Mid and Side, or Mid-Side. 2R=(L+R)-(L-R). The L term drops out leaving 2R. To get one "R" we use a 1246 with it's -6dB gain to get 2R/2.


Fortunately the right channel can be fixed by just using a THAT1246. Only one inverter, for the left channel output is required.

With the added THAT1240 I did some quick crosstalk tests and the results, looping through the less-gain-accurate 1646s was pretty good. L to R was something like -75 dB and R to L more like -60 dB. Swapping the 1646s did change the crosstalk - it followed it. My original method of precision bypass involved bypassing around the 1646s and 1246 on the decoder in single-ended. I'll run those measurements and I expect we'll see around -75 dB or greater.

RMAA Results:

This is the M-S matrix E-E performance. Most of what we're seeing here is soundcard, an EMU-0404...

Image

Crosstalk performance in the "non-precision" bypass mode end-to-end through the THAT1646s:

Image

This is the full html report:

https://www.ka-electronics.com/Images/h ... _RMAA6.htm
Post Reply