what are OSI for this SSL_82E26 mix amp?

[gnd]

Hi.
I'm putting together mix amps for my console, and am wondering about optimum source impedances for the circuit. It is basically identical to this one:
http://www.tenmilecreek.net/images/ssl/ssl_82E26.pdf
I just used SSM2210 instead of LM394, and added some NE5532 output A biasing.

In SSL4000 this mix amp is fed via 15k resistors. Was it 64 channels going on SSL4000 into this mix amp? If so, then for 64 channels 15k gives source impedance of 234ohm, is it? (96 channels - 156ohm, 32 chan - 468ohm). How many channels was usually connected to this mix bus in SSL consoles?

I have 24 channels, and would like to go with optimal mix resistors, so that noise is minimal. I have NE5532 buffers on channel outputs, and could go down to 1k with mix bus resistors. Would anybody have info what is OSI for this amp? Or how to calculate it? And how low can I go with mix resistors for 24 channels, to keep currents within normal?

And what about ground mix resistors on channel grounds (the ones going from channel grounds to mix buss 0V)? Should they be the same value as signal mix resistors?

What would be best values for summing resistors (signal and ground) from 24channels into this SSL_82E26 mix amp? How to calculate them?

Thnx
gnd

 

[Samuel Groner]

A rough calculation indicates that it is around 500 ohm. But fortunately enough that doesn't matter at all--life (in first approximation) is simple here: the lower the resistor value, the lower the noise. Below some point (say below 1/5 the OSI) the improvement is marginal as noise will be completely dominated by amplifier noise. This leads to the conclusion that 2.4k ohm is about as good as it gets. I'd not go below this as with about any IC amplifier distortion will become higher with lower values.

[Edit: of course as pointed out in the previous post, you'll need to adapt the schematic for different resistor values]

Samuel

 

[JohnRoberts]

The 2210 data sheet doesn't show the typical family plot of NF curves re: source impedance. The curves it does show does not go up to 3 ma but it looks like optimal impedance there would be well below 1K.

I would be inclined to go with 15k resistors feeding the bus, and then look at tweaking opamp input current density for minimum noise (I suspect less than 3ma per device, but why not measure actual result).

wrt resistors feeding the - bus, these should IMO be at least an order of magnitude lower, but the actual value is not critical. Anything from a few hundred ohms to a couple k.

For dialing in the bus sum amp, simulate the effective bus resistance with low value single resistors to eliminate layout issues during that tweak. The 2210 looks like a nice part so will probably have a wide sweet spot.

JR

 

[gnd]

Thnx all.

MT:
We discussed that 15k+1k R50 resistor before, and you mentioned that it is 15k. Now I have it 15k on my board. --- Regarding ground to 0V resistors, I will use them anyway, no problem putting them in at this stage. Although 15k seems fine, I would try to reduce them, maybe 1k. Would I have to modify anything in schematics to use 1k resistors from channel grounds to 0V mix amp? Has that 100k R1 anything to do with this? --- How about if I use 2.4k for summing signal, like Samuel suggested, what changes would be neccesary in schematics for unity gain of signal? What problems could I expect with going as low as 2.4k on signal sum resistors?

Samuel:
I was also thinking about 2.4k for signal. Maybe 3.9K... But then gain of amp would be too high? I need unity gain from channel buffers. I don't know how to calculate gain (noise gain) in this circuit, those ssm2210 in front drive me nuts... :? . Which resistors are actually setting the gain of this mix amp? Is formula the same as for differential opamp?

John:
You suggested 15k for signal summing, but somehow I'd rather go with lower resistor values, like Samuel suggested. Do you see any problem in going as low as 2.4k with signal summing resistors? --- Regarding dialing in, simulating with single resistor, i thought about it, but am not sure how do i do it? Do I just connect 1k trimmer to input, between signal input and 0V input (or between signal input and ground???), and then play with it untill i find minimum noise at output? Is it simple like this, or do I need to feed signal into input, or do something else?

thnx
gnd

 

[Svart]

i use 3k3 for my summing resistors with 24 channels. No problems here.

 

[Samuel Groner]

Which resistors are actually setting the gain of this mix amp? Is formula the same as for differential opamp?

The LM394s together with T1 and some of the surrounding parts form a composite opamp. The bases on the "a" side of the transistors are the noninverting input and the "b" bases the inverting. Output of the composite opamp is the T1 output.

If we ignore the ground summing resistors (which is a good assumption as they are hopefully small enough to not contribute significant noise) we get a simple inverting topology--the noninverting input is grounded, the feedback resistor is R50 and the source resistor the summing resistors in parallel. The rest is simple text book math...

So for lower value summing resistors you might want to change R50 as well. C1 and C2 need to be proportionally larger as well and L1 can propably get proportionally smaller, though this should not be critical but might depend on bus design (stray capacity).

I don't think you need to modify anything in your "26 card" to use the 1k. R1, 100k, is there as a bias return resistor and it wouldn't affect your use of 1ks.

I'd say R1 should become a 1k as well, otherwise the output of the summing amplifier is not properly reference to local ground. See e.g. Summing Buss2.pdf.

Samuel

 

[gnd]

Thnx for great posts.

I did a quick simulation in multisim. I'm starting to figure out schematics. :grin: :idea:

With 15k on input gain is exactly unity.
R50 is gain, and 2.4k puts it to unity with 2.4k on inputs.

As I see it, second NE5534 is just unity gain inverter, so those 15k are probably not critical, and can be changed, is it?

:?: With 2.4k signal suming resistors, could higher currents be a problem for mix bus amp input?

Sam, thnx for explaining circuit. It is really simple, the way you put it.
We discussed Forsell differential sum bus some days ago. I will make R1 to be 1k too, same as ground resistors at channels.

C1 and C2 (and R24, R25 biasing) are a bit in question. Why do I need them at all? DC decoupling? Especially bipolar 1000uF may be hard to get over 16V. On my channel boards I planned capacitors on outputs, dual 220uF, minus tied together, plus to outside, without DC bias. So it would be 110uF on each channel output. Do they add together for 24 channels? If yes, that would make 2500uF C on bus, distributed over channels?

Would it be better to skip those C's on channels, and feed bus without them, and then use 1000uF on bus input as in schematics? Or better leave those 220uF back/2/back duals on channels, and skip C1 and C2 on mix bus? Or better use both, on channels and on bus input?

Regarding C28 100p, for now I put in 47p, and will check stability with 10kHz square wave. Usually I reduce those capacitors to minimum before ringing on square starts, because to my ear they make sound a bit dull.

gnd

 

[gnd]

[quote author="mediatechnology"]gnd: Is this going in a box, or is it a traditional desk with a frame (and long bus) sitting in a flux loop? If the bus is only 17" long (rack width) I'm not sure that 1K or any of them are going to make too much difference. SSL built consoles for years without them. It was only after they started to build huge frames that they saw a problem and it was usually in facilities that had lots of steel in them carrying circulating currents within the building structure.[/quote]

It is a bigger than 19" rack, more like two racks. Buses are 1 meter long. But thats more like small console. Still, while I'm working on it, pulling all channels out, better I stick that R 1k on each channel, just to have peace that I did my best. I'm sure some tiny bit of 50Hz hum will come in somewhere, and I do not want to think later on that it is because I didn't put in those 1k's. Can those 1k's make things worse? If not, they will go in, and even if they do nothing, at least I will feel good. :grin:

...

 

[JohnRoberts]

John was out cutting grass...and am now in :guinness: mode..

Regarding the merit of dropping summing resistor impedance to reduce noise, I understand the desire to optimize each and every circuit element, but if you do a sensitivity analysis of total noise to changing this one factor, with practical sources connected to the bus you will find this of minor significance. Perhaps if you used 1 meg resistors, but I suspect the contribution from 15ks will be less than 1 open mic channel, etc.

That said the customer is always right, and you are the customer here so do what you want to do. I have gone as low as 2.4k in a summing amp but for different reasons. I was getting electrostatic crosstalk due to layout constraints in the input modules of a large console. Dropping the bus impedance scaled down the crosstalk a like amount, without need for a new physical layout.

------

My comment about using scaled down single resistors to dial in noise vs. input stage current density was to keep layout issues and other noise sources out of the picture. While I don't think it matters in the grand scheme of things whether you use 20k or 2k summing resistors it will impact the optimal current density for the long tail pair. So pick something and proceed to dial in for that.

------
I'm not sure I follow the cap question :guinness: :guinness: , but if you are not switching inputs in and out, or using grounded wiper pan circuits you might just ignore DC issues until after the summer and just block any DC once with a nice film cap.

I just went back and looked at more of the schematic lots of parts there. Some I'm not sure I'd want in my audio path. I am not prepared to design something different for you (I used to get paid for doing that) , but I would suggest experimenting with leaving some parts out, unless you find you need them to solve some problem later.

JR

 

[gnd]

Hi again.

Just for info, I did some measurements, and found out OSI for this SSL mix amp to be arround 310ohm.

This makes sense, since it would make optimum mix resistance for 48 channels to be 15k. Maybe it was calculated for 48chan console.

For 24 channels, with existing topology, optimal value for summing resistors wouls be arround 7k to 8k. Using 15k resistors instead of 7k would result in 3 to 4 dB higher noise floor.

So I think my few hours of playing arround with it were worthed, I just gained 3dB on SNR. :grin:


Again, thnx all for the valuable info. I will adjust circuit a bit. I realised that those chokes and capacitors need to be there, otherwise there is plenty of oscilations and ringing. But those biased electrolytic caps I may throw out, I will decide after I see what will be DC when all is put together.....


gnd

 

[gnd]

MT:
I have a question on SSL_82E26 mod. Say I want to make it fully balanced. Using 7k summing resistors, would I do right to do the following:
1. feed pusitive bus signal to pin8 (mix bus 0) via 7k resistor (and 220uH)
2. feed negative bus signal to pin5 (mix bus signal) via 7k resistor (and 220uH)
3. change negative feedback R50 to 7k
4. change R1 to 7k.

Would this procedure give me proper differential input amp for balanced bus?

I'd also like to ask you for your opinion on input diodes D8 and D9. Why are they there? If I remove them, could it cause any problems?

thnx
gnd

 

[JohnRoberts]

[quote author="gnd"]MT:
I have a question on SSL_82E26 mod. Say I want to make it fully balanced. Using 7k summing resistors, would I do right to do the following:
1. feed pusitive bus signal to pin8 (mix bus 0) via 7k resistor (and 220uH)
2. feed negative bus signal to pin5 (mix bus signal) via 7k resistor (and 220uH)
3. change negative feedback R50 to 7k
4. change R1 to 7k.

Would this procedure give me proper differential input amp for balanced bus?

I'd also like to ask you for your opinion on input diodes D8 and D9. Why are they there? If I remove them, could it cause any problems?

thnx
gnd[/quote]

D8 and D9 are there to protect input devices (from zenering) they are harmless in circuit but might help parts in event of PS failure or strange hook up mode. Only reason to maybe lose them is if 2210 has them built in.

Not sure I follow what you mean by making it balanced? (which is probably why you didn't ask me :roll: ). If you're talking about differential summing bus, there is merit in keeping the - bus lower impedance. If you are talking about feeding an opposite polarity audio signal from each input channel into a symmetrical (balanced) - bus, you will need to look at the topology of the summing amp. This DOA is presumably designed as a virtual earth summer so I don't know how much input compliance or CM range it would tolerate. A topology similar to seen in some of the mic preamps to bring an opposing polarity feedback loop back to the - input so we are back to a quasi virtual earth, might work.

I would take the more conventional differential bus approach, namely low impedance resistors from channel grounds to - bus. Regarding optimal impedance I might consider going a little lighter on input stage current density, but it looks like you got it pretty good. When you measured your optimal impedance was that at both + and - inputs? Bus 0V grounded? For noise analysis it's the impedance of what's connected to both inputs. Making resistors from channel ground references to bus 0 larger value than needed, would increase noise.

JR

 

[gnd]

MT, thnx.

John, thnx for jumping in.

Only reason to maybe lose them is if 2210 has them built in.

SSM2210 has some diodes in, but I think they are for other purpose, as they say, to prevent degradation of beta and matching characteristics due to reverse biasing....., whatever that means. :? I'll keep those protection diodes at input.

Not sure I follow what you mean by making it balanced?

Balanced, I mean positive polarity signal fed to one mix amp input, and opposite polarity signal fed into other input. Full signal, opposite polarities, fed to both inputs of diff sum amp.

There is similar topology in ssm2210 datasheet, page 9:
http://www.analog.com/UploadedFiles/Data_Sheets/SSM2210.pdf
Except there is no input protection diodes, and no feedback resistor. But there is (+) and (-) input. Thats why I thought I'd skip input diodes in my case too, and by adding 4 resistors I'd get proper differential amp with (+) and (-) inputs.... Do I properly see this circuit from ssm2210 datasheet as similar to SSL mix bus?

I would take the more conventional differential bus approach

I know differential approach would be just fine, with resistors from channel grounds to (-) bus. But if I go balanced, I gain 3dB SNR just from balancing, and I like that. Plus buses would be fully balanced, signal in both, and it seems more fool-proof to me. Those grounds connected via resistors are a bit puzzling to me.... :shock: Anyway, I haven't deciced yet whether I will go quasi-balanced or fully balanced. But I will need to make output boards for channels in any case, and it really makes almost no difference if I add one 5532 and few resistors, to make it fully balanced.

When you measured your optimal impedance was that at both + and - inputs?

When measuring OSI, it was measured just for (-) input. (+) was connected to ground via 1k resistor. That would be value for differential sum bus anyway. For balanced it would be 7k, so noise would be a bit worse, but probably much less than 3dB SNR improvement from real-balanced summing.
...Or maybe with 7k on (+) noise would be even better than with 1k. Because, at (-) input, going with source impedance below 300ohm (7k from 24 channels), noise started increasing rapidly. If sum resistors would be 3k (instead of 7k), noise would be 3dB louder. And looking at my OSI graph, extrapolating down to 1k, SNR would be 6dB worse than with 7k sum resistors. I don't know if it would be similar on (+) input, for test I had it fixed at 1k. But since input has identical ssm2210 on both (+) and (-), could I conclude that noise figures would be the same for (+) input? If yes, then 7k is best option for both (+) and (-) in case of 24 channels.

...

 

[JohnRoberts]

[quote author="gnd"]MT, thnx.


SSM2210 has some diodes in, but I think they are for other purpose, as they say, to prevent degradation of beta and matching characteristics due to reverse biasing....., whatever that means. :? I'll keep those protection diodes at input.[/quote]

That is "zenering". When low noise transistors reverse bias their b-e junction around 7V or so, they get noisy and their beta deteriorates. They can be repaired by annealing with high current but it's not trivial and easy to avoid. With those internal diodes the external ones are redundant.

Balanced, I mean positive polarity signal fed to one mix amp input, and opposite polarity signal fed into other input. Full signal, opposite polarities, fed to both inputs of diff sum amp.

There is similar topology in ssm2210 datasheet, page 9:
http://www.analog.com/UploadedFiles/Data_Sheets/SSM2210.pdf
Except there is no input protection diodes, and no feedback resistor. But there is (+) and (-) input. Thats why I thought I'd skip input diodes in my case too, and by adding 4 resistors I'd get proper differential amp with (+) and (-) inputs.... Do I properly see this circuit from ssm2210 datasheet as similar to SSL mix bus?

Yes you could drive both inputs with opposite polarity signals from each channel but I'm not sure it buys you much if any benefit, especially considering the incremental cost (extra inverting drivers in each channel). Another thing I'd need to think about, the bus is no longer a virtual ground, so there may be consequences from that.

A more complicated additional feedback loop could be configured in the master amp to hold both the buses at virtual grounds. Then those two bus outputs would then need to be single ended into a differential amp.

I know differential approach would be just fine, with resistors from channel grounds to (-) bus. But if I go balanced, I gain 3dB SNR just from balancing, and I like that. Plus buses would be fully balanced, signal in both, and it seems more fool-proof to me. Those grounds connected via resistors are a bit puzzling to me.... :shock: Anyway, I haven't deciced yet whether I will go quasi-balanced or fully balanced. But I will need to make output boards for channels in any case, and it really makes almost no difference if I add one 5532 and few resistors, to make it fully balanced.

Not sure that you get only 3dB. You pick up 6 dB of signal (2x). The noise is probably similar. So improvement could easily be more than 3 dB but like so what. If this isn't defining your noise floor it's wasted cost/complexity.

When measuring OSI, it was measured just for (-) input. (+) was connected to ground via 1k resistor. That would be value for differential sum bus anyway. For balanced it would be 7k, so noise would be a bit worse, but probably much less than 3dB SNR improvement from real-balanced summing.
...Or maybe with 7k on (+) noise would be even better than with 1k. Because, at (-) input, going with source impedance below 300ohm (7k from 24 channels), noise started increasing rapidly. If sum resistors would be 3k (instead of 7k), noise would be 3dB louder. And looking at my OSI graph, extrapolating down to 1k, SNR would be 6dB worse than with 7k sum resistors. I don't know if it would be similar on (+) input, for test I had it fixed at 1k. But since input has identical ssm2210 on both (+) and (-), could I conclude that noise figures would be the same for (+) input? If yes, then 7k is best option for both (+) and (-) in case of 24 channels.

...

If 1k in + input and 300 in - input to ground resistance (for noise purposes was 1k + 310 in parallel with 15k (feedback R). I notice from the data sheet they run the 2210 around 2 ma in application circuits, and many data sheet specs stop at 1 ma so I take that a strong hint about what they were designed for (per device).

Keep in mind, just one input channel noisier than this sum amp will moot all this effort. Get it good, but don't increase global warming for little benefit.

I remain unconvinced to push the complexity and suspect the 2210 might need to be operated a little cooler (2ma per pair). I vaguely recall there was room for improvement if you want to add active parts in the pan circuit but i don't care to revisit that tonight.

JR

 

[Svart]

Wayne, I never noticed that way of using those ICs like that. It makes sense though now that I've seen it..

:shock:

 

[gnd]

Hi.

John:

I'm still thinking about your points.

So improvement could easily be more than 3 dB but like so what. If this isn't defining your noise floor it's wasted cost/complexity

Yes, you are right. Single channel is now modified and noise is at -115dbA. And summing amp is at -112dBA. Those 3 dB will be filled with noise from 24 channels. If I calculate properly, 24 channels will increase noise by 14db, so final noise contribution from channels will be arround -101dbA. And mix bus with -112dBA will add nothing, no use to try to gain 3 dB by balancing.

Of course, at measurements load was 350R resistor, and then will be 24 resistors of 8k, so resistor noise will be higher on bus. Is there a way to predict how much noise will be aded by 24 8k resistors? Could it be so much, that mix bus will become noisier than channels? How much would you say that noise will be increased using real load of 24x8k?

I notice from the data sheet they run the 2210 around 2 ma in application circuits, and many data sheet specs stop at 1 ma so I take that a strong hint about what they were designed for (per device).

Simulation of ssl mix amp shows 5mA current to all four transistors, so it is 1.25mA per transistor. Would it need to be reduced? By increasing resistor r48 (82R) to 104R (82R + 22R) i can get current down to 4mA, which is 1mA per TR. But is it necessary, is this 1.25mA really critical comparing to 1mA in datasheets?

I vaguely recall there was room for improvement if you want to add active parts in the pan circuit but i don't care to revisit that tonight.

Yes, thats being done, I followed your suggestions. Active gain makeup after pan circuit, to bring level to 0dB before mix bus. And I also made new input boards (two opamp superbal with variable gain, as per Ted Fletcher of TFPro). So channels are all new at front and back, and I modified several things inbetween.

gnd
...

 

[gnd]

Hi.

MT:
SSM2017 was my second option all the time. I have 10 of them lying in drawer, so I could use them. But I also had SSM2210, so I thought I'd rather use 2210 in mix bus, and keep 2017 for mic pres or something.

You can still use the SSM2019, THAT1510 or THAT1512 in this configuration.

Are you saying, that I could use both 2017 and 2210 in same mix amp? That would be cool. But how to do it, I cannot design such thing.... But would surely try it if somebody would provide some design ideas, now is perfect time.

I must say I was a bit unsure regarding ssm2017 on mix bus because of its high noise at lower gains. It sure has 0.95nV at 60dB gain (1000). But 24 channels mix bus has gain of 25. At gain of 10 SSM2017 has noise of 12nV, which is plenty comparing to 3.5nV of NE5534. Maybe at gain 30 it drops to some 5nV.... But still I don't see that much of improvement over simple NE5532. Of course 2017 is optimized for lower impedances, but has 2pA noise current at gain 1000, whila 5534 has 0.4pA.... Am I missing something or...?

I mention the above because there really isn't much, if any, "magic" in the 82E26. It's pretty much a normal discrete op amp.

There may be magic or not, but it is well tested in practice. Possibly SSL became big name for its automation and routing, and not necessary for audio path filled with 5534's. I mean, SSL audio path is not esoteric like NEVE or API or similar - it is quite simple (cheap??) on audio path. So it cannot be that much "the sound" that made SSL famous... But this 82E26 seems to worked for serious SSL consoles, so I guess I'm preety safe by copying it.

But I wouldn't mind experimenting with ssm2017. Any suggestions?

thnx
gnd

 

[JohnRoberts]

[quote author="gnd"]Hi.

John:

===

Of course, at measurements load was 350R resistor, and then will be 24 resistors of 8k, so resistor noise will be higher on bus. Is there a way to predict how much noise will be aded by 24 8k resistors? Could it be so much, that mix bus will become noisier than channels? How much would you say that noise will be increased using real load of 24x8k?

gnd
...[/quote]


Not sure I follow your question. 24x8k in parallel are close to 300 ohms so any difference will be slight. The noise sensitivity to resistor value has several moving parts. The Johnson or thermal noise of the resistor (that includes a term for actual component temperature), the noise voltage of the transistors or ein (equivalent input noise) of the compound amplifier, and the noise current of the input transistors times the resistance using good old Ohms law to turn that into a voltage. These three voltages are combined in a square root of the sum of the squares calculation then multiplied by the noise gain N+1 for an N input virtual earth summer.

An operative point to consider here is that increasing the current density of the input transistors typically decreases their noise voltage while increasing their noise current. Thus my original caution about too much operating current. What wasn't clear to me is that you appear to be using 4 devices. I was thinking 5 mA feeding a single pair. The difference between 1.25 and 1 mA is probably minor. {revisiting the 2210 data sheet it looks like more than 1 mA per device might be optimal for 300 ohm total source impedance so don't feel compelled to pursue my 1mA suggestion. There would be a certain amount of guesswork involved to do a formal calcilation since the graph for noise voltage vs. current doesn't show above 1 mA. The noise current increases some 2.5x between 1 mA ns 10 mA so none of this is obvious. The noise plot for 10k source looks optimal at 50 uA, 1k looks happy at 1 mA, so I can speculate from that 300 ohms might benefit slightly from more current. }

If you really want to dial this in and have adequate instrumentation, tweak the current density and see how it measures. The noise figure plots looked pretty broad so probably exhibit wide sweet spot.

One pretty significant point that I don't know is your current plan for the impedance of the resistors feeding the bus ground from the individual channel grounds. The combined parallel resistance on this bus is added in series with the combined parallel impedance of the signal summing bus. So using the same value large resistors for both buses could effectively double resistance for noise contribution. (I would suggest something in the 100 ohm ballpark from each channel ground.

NOTE: just for making noise measurements you could increase the value of summing amp's feedback resistor 10x or 30x to make the measurement easier to see on test equipment. Since this feedback resistor is in parallel with input resistors for noise contribution, the noise you measure with this scaled up will be representative of what you would get in use only 20db or 30db larger.

JR

 

[JohnRoberts]

Just for a little historical perspective the Transamp as summing amp made sense because there weren't many low noise options in it's heyday. Those capable probably rolled their own DOA with a LM394 (I know I did), the TA offered a somewhat off the shelf solution with respectable noise performance.

I have previously stated that summing bus noise is perhaps over analyzed as a limiting factor for console performance wrt noise but is IMO more often limiting in gain bandwidth and/or phase shift effects due to high noise gain with unity gain compensated opamps. The TA nicely addressed these shortcomings while delivering low noise too.

JR

 

[Samuel Groner]

I have previously stated that summing bus noise is perhaps over analyzed as a limiting factor for console performance wrt noise but is IMO more often limiting in gain bandwidth and/or phase shift effects due to high noise gain with unity gain compensated opamps.

I second that. To help the author of this thread with deciding on the proper topology we'd need to know if the number of channels (24 it is) is fixed or variable (perhaps I missed that information?). For the former I think I'd use a voltage feedback design (the SSL design discussed here) while the latter calls for current feedback (TransAmp et al).

Samuel