Some LME49600 implementations

[sek@]

A very interesting discussion about the servo circuit. 
 
I read that as follows:
 
Under normal conditions the servo acts to minimize the output DC voltage occuring from both an offset inherent to the opamp/buffer circuit and the input signal. Depending on the choice of feedback and servo/gain resistors, the influence ("authority") can be adjusted, but is of course defined as a proportion of the signal, limited by the supply voltages. So far, so good.
 
Thus, if either a full scale DC voltage is applied to the input or any active device fails, introducing either rail voltage into the circuit's signal chain, the servo is incapable to effectively double as a safety feature in order to protect the headphones.
 
In order to account for this, among other things, wakibaki incorporates a relais based protection circuit into his design.
 
The question for me is:
 
Is the servo worth the trouble under normal conditions, i.e. is the nulling effect that significant to the sound out of the headphone (compared to, say, a fully DC coupled signal chain)?
Seeing that protection is covered by the relais circuit, isn't the servo expendable in all but the most delicate cases?
 
As of my understanding, the circuit is rather stable, using high quality/low drift components. All causes of output offset voltage that stem from the circuit itself should essentially be constant, making it easy to deal with them without a servo (i.e. selection, trimming) or to just live with it and leave well enough alone.
 
And in order to put things into perspective, I can report a maximum of approx. 50 mV DC out of THE WIRE, which is basically the circuit at hand minus the servo. That's about 80 uW into a 32 Ohm headphone or 5 uW into 600 Ohm.
 
Thanks,
Sebastian.

 

[Avro_Arrow]

Here is how I take it.
 
Since this is a DC coupled design from end to end, any DC offset
in the source will be amplified. The servo is able to null out any
reasonable DC offset at the input to keep the output near zero.
 
If the above fails, then Wakibaki's fail safe will kill the output.
 
The SE-SE Wire can not correct for DC offset at the input.
 
edit: spelling

 

[sek@]

Hi again,
 
Quote:

Originally Posted by Avro_Arrow /img/forum/go_quote.gif
 
The servo is able to null out any reasonable DC offset at the input to keep the output near zero.

 
Granted, but my question was actually wether this brings any advantage over taking care how you build and set up your components. Let's ignore sources with insensible DC offset for the moment.
 
 

The SE-SE Wire can not correct for DC offset at the input.

 
I was referring to wakibaki's circuit without the servo, which is essentially a SE-SE with protection circuit.

 

[Avro_Arrow]

If you remove source offset from the discussion,
I don't think the servo adds anything over and above what a properly designed circuit will do.
The only other thing I can think of is if your signal is AC offset (more swing on one side than the other)
the servo will push it back to the center. I'm not sure under what circumstances the above condition
happens, but I have seen it when editing audio files in Audacity.

 

[sek@]

Quote:

If you remove source offset from the discussion,
I don't think the servo adds anything over and above what a properly designed circuit will do.

 
The question that remains is wether it adds something you don't want.
 
 

The only other thing I can think of is if your signal is AC offset (more swing on one side than the other)

 
 
Different swing on either side by definition is a DC component.
 
Different wave form shape on either side, however, is part of the music.
 

 

[kevin gilmore]

there are some otherwise nice dac's out there with unusual i/v converters that put out more than .1 volts dc.
The resulting offset after amplification can damage some sensitive headphones.
My opinion is that the servo is absolutely required.

 

[Avro_Arrow]

The servo only works at low frequency, just a few Hz.
At audible frequencies, it should be invisible.
 
Probably the way to truly answer your question is to build the
amp and listen to it with the servo enabled and disabled to see if you can
hear the difference.
 
I don't want to seem like I am arguing for or against servos...I think
like any other circuit element, it has it's uses.
 

 

[wakibaki]

If we take a 16 ohm headphone with 100dB/mW and we want to drive it to 100dB then the voltage across it is sqrt(0.001 * 16) = 126mV.
 
Then 1 LSB would be 0.126/(2^16) = ~2uV. (rms), 2.73uVpk.
 
So I think ideally we would like to see 1/2 LSB offset, and 50mV is unacceptable. It's about ~18,000 times 1 LSB and  I was thinking of setting the protection to 50mV and AMB's E12 protection circuit operates @ 60mV IIRC.
 
If we use OPA132 with the 1M resistors and a gain of 2 we get 1050uV offset and with OPA277 we get 1020uV.and with LTC1050CS8, 160uV, all worst case. With LTC1050CS8 & 100k this comes down to ~25uV worst case (10 LSBs) and 2.5uV typical. Farnell stock LTC1050CS8
 
Feel free to disagree with my arithmetic.
 
These numbers look a lot better if we are using more of the amplifier's output range, of course, and it's capable of nearly 100 times this voltage swing.
 
w
 
LTC1050CS8 is really only good for +/- 16V

 

[sek@]

Thanks for replying to my questions everyone!
 
Quote:

there are some otherwise nice dac's out there with unusual i/v converters that put out more than .1 volts dc.

 
As I mentioned earlier, I'm leaving source DC offset out of the equation with my question.
But you're of course right in that this can't be a recommendation for everyone and any source.
 
If, though, the known source (like a DAC that is known to be DC free after it's reconstruction filter) is no issue...
Also, we don't quite cover the presumably relevant case of balanced operation.
 
I agree with you, Kevin, an amp like this better has the servo in it!
 
Quote:

listen to it with the servo enabled and disabled to see if you can hear the difference.

 
The more I think about it, the more I see this point and also find myself starting to think about incorporating it.
 
Quote:

So I think ideally we would like to see 1/2 LSB offset, and 50mV is unacceptable. It's about ~18,000 times 1 LSB

Well, we'd still have to distinguish between AC and DC components at the amplifier output.
 
The output offset will contribute to heating the voice coils, reducing the headphone's dynamic range through phenomenons like reduced dissipation, power compression or even biasing (depending on construction). It also poses the danger of overheating the phones even without a music signal present at all. And it causes nasty clicks at turn-on and turn-off...
 
A DC offset will not be noise or distortion in itself (depending on whether you measure for audible distortion), though. Nor will it alter any AC component (the music). If it did, it never were to be considered offset to begin with and a low frequency servo wouldn't help much anyway. The servo circuit discussed here is not intended to fight motorboating or unstable supplies, although it would do to that to a degree.
 

Then 1 LSB would be 0.126/(2^16) = ~2uV. (rms), 2.73uVpk.

 
Does it really make sense to refer output offset level to DAC resolution? What did I miss?
 
What if you increase your desired resolution to, say, 20 bit (or maybe even 24 bit)?
 
Then 1 LSB would be 0.126/(2^20) ≈ 120 nV. 
 
The worst case you mentioned above would have a totally different meaning now, perceiving this as a proportion of LSB gets difficult.
 
I gave an example calculation referring to (thermal) power levels above.
 
 

 ~25uV worst case (10 LSBs)

 
 
25 uV represented by 2.72 uV steps is a factor of ~9.2, which is inbetween bits 3 (2^3=8) and 4 (2^4=16) referred to LSB, so 25 uV would correspond to 4 bit (as 2^3=8 < 9.2 < 2^4=16).
 

Anyway, the 50mV I mentioned are a worst case scenario in itself, the real world offset without a servo can still be made much lower (approximately by a decade).
 
I'd like to add that a servo circuit should be a measure to make AC coupling (using series capacitors) unnecessary, not as an excuse to ignore inherent offset that could be minimized by design or adjustment.
 
Cheers,
Sebastian.

 

[wakibaki]

Quote:

50 mV DC out of THE WIRE, which is basically the circuit at hand minus the servo. That's about 80 nW into a 32 Ohm headphone or 5 nW into 600 Ohm. 

50mV in 32R is 78uW, not 78 nW.
 
25uV is between the 3rd. and 4th significant bits, but it is still 10 times the 1st. or lowest significant bit. Try to keep your eye on the ball instead of indulging in mindless nitpicking.
 
w

 

[sek@]

 

50mV in 32R is 78uW, not 78 nW.

 
Thanks for checking, I corrected my mistake in the original post above.
 
 

25uV is between the 3rd. and 4th significant bits, but it is still 10 times the 1st. or lowest significant bit.

 
Great, so you do agree with me on the math, I also corrected a typo in my last post, thanks. The reason I explicitly presented the numbers is because I still don't see how your attempt at producing as low as 2.5 uV DC offset brings us to your self-imposed goal of getting below 1/2 LSB. We're still off, even with your benevolent choice of 16 bit resolution and your best case calculation.
 
I'd rather like to discuss the relevance of relative DC offset level, assuming the absolute level is low enough not to cause damage or dynamic compression or bias, than get lost in the details of a certain example.
 

Try to keep your eye on the ball instead of indulging in mindless nitpicking.

 
No reason to get grumpy...
 
 
Speaking of getting the eye back on the ball, here's the question at hand again:
 

Does it really make sense to refer output offset level to DAC resolution?

 
Cheers,
Sebastian.
 
 
 

 

[wakibaki]

OK, I've been looking at this a lot, the first thing is LTC1050 is only good for 16V total, it'll have to be LTC1150. Which is expensive. It also means some modifications to the circuit due to the reduced range of output swing available, and the fact that it's a chopper-stabilized amplifier.
 
I started looking at the output filter I was going to use to feed the comparators. It has a rollof of 0.16Hz, same as the integrator. This is 2 decades down on 16Hz so a rail-rail output of 16Hz will only be suppressed by 40dB or 100 times, meaning that there will be a residual of 150mV pk. This means that the threshold of the comparator controlling the relay can be set no lower than 150mV or it will false trigger for a 16Hz full-scale input, This means that the DC level detectable can be no less than 150mV. The residual @ 20Hz will be somewhat less than the one @ 16Hz, but a little margin is required. Setting the threshold to 150mV simply means that no input at 20Hz could cause false triggering of the relay. 
 
It is important to understand what would happen in the circuit using the filter to trigger the relay. If the servo authority is set to 1/10th. of the full-scale voltage then a DC offset in the source of >1.5V will defeat the servo and the phones can be exposed to a continuous offset of up to 150mV before the relay operates. The DC servo nulls out any offset voltage in the source up to the range of its authority (up to 1.5V in this case). If the source offset exceeds 1.5V then the excess (the offset - 1.5V) appears at the amplifier output up to the point where the servo comparator sees 150mV (total offset = 1.65V), when it causes the relay to drop out.
 
If, instead of using the filter at the output, the relay comparator looks at the servo output, then the relay can be forced to drop out before the servo runs out of adjustment. Now what we have is an arrangement where the servo compensates for any DC offset in the source (or any arising anywhere) until it can no longer compensate, whereupon the relay drops out.
 
The LTC1150 has an output swing of +10.5V into 10k. It really requires a filter at the output to eliminate switching spikes.
 
If we take the 10k resistor that is supplying the servo feedback and split it into 2 * 4k7 resistors we can create a filter with a rolloff @ ~1.6Hz using a 20uF cap at the junction of the 2 resistors and also insulate (to a degree) the filter from the 1k resistors in the amplifier proper which will otherwise load the filter. This is an order of magnitude higher than the loop unity gain frequency and should not contribute to any instability. It has the added benefit that the LTC1150 now sees >10k, keeping its output voltage up.  If we set the relay comparator threshold to 10V (at the servo output) then the relay will operate for input offsets of just over +/-1V and the 20Hz residual if sampling at the junction of the 2 * 4k7s will be only 4V (settling to 2.5Vpk), so a 20Hz full-scale output will only cause triggering when an offset of >0.5V is already present. The servo authority is reduced by a factor of 9.4 instead of 10.
 

 
The last question to be resolved is how quickly the circuit responds to an offset. The time constant of the integrator is 1 second, but the comparator sampling point is now the junction of the 2 4k7 resistors, which is being pumped by the amplifier output via the feedback resistor. The sim says 566mS to reach 10V feedback. This is not ideal, but the integrator tau is the same as as that of the filter I was going to use and the response is better than the DC servo response was in the original circuit. 
 

 
As regards using 1 * 16-bit LSB as a standard for assessing offset performance; (taking +/-15V rails)
 
1) 16/44k1 is the limit of audibility. There's no point in using 24 bits, it's unrepresentative of what can be heard. Less than 16 bits is throwing away resolution that arguably can be heard  
 
2)There's no point working to get the offset down below 1/2 LSB, so it sets a limit on how hard we need to work. Once we get it down to 1/2 LSB we can stop trying to get it down any further.
 
3)What else are you going to use? Different phones have different sensitivities, but not many are more sensitive than 100dB/mW. If you get the offset down to 1/2 LSB of the most sensitive phones working at a 'normal' level then the offset will be insignificant for all other phones. It's a ballpark thing.
 
You can pick an arbitrary number out of the air, such as the 50mV that has been discussed and say, that'll be OK on the basis of experience, and quite honestly I doubt that even the most sensitive phones suffer any ill effects from a 50mV offset. I'm less sure that a 50mV offset has no audible effect in such phones, 1 mW in 16R is only 126 mV rms. I'd be a lot happier arguing that 2.5uV is not significant as opposed to 50 mV. Yes, 2.5uV is still not 1/2 LSB but it's only just over 1 LSB, which has to be a lot better than the ~1000uV that the OPA132 was going to get us. It is less than 1 LSB for a DAC having a 2V output.
 
I also decided to substitute the PIC for the AND gates, it's just better all round. It has 5 inputs, less pins, no cap, lower parts count. I've added a high-speed switch off so thumps at turn-on and turn-off should both be eliminated now..
 

 
OK, I'm going to run some more sims. If anybody sees any holes in my thinking, please point them out, there are few subjects with as many lurking traps as electronic design.
 
w
 

 

[kevin gilmore]

Lets start with C2 in the wrong place, needs to be after the resistor.
Correctly shown in the 3rd diagram. Need to run analysis again.
 
This seems to be an even more simple protection circuit with a pair of lm339.
http://gilmore.chem.northwestern.edu/protector.pdf

 

[wakibaki]

Thanks, I changed the sim. It makes little practical difference to the result, the 20 Hz residual is reduced to less than a volt, which is all to the good, since the threshold is 5V at the sampling point. Similarly the rise time at the sampling point for a step input is reduced. This means that the relay will drop out @ 220mS with a 15V step at the amp input. Because of the amplifier gain the output reaches 15V for a DC input of ~6.5V and the time to dropout is not greatly increased. 
 

 

 
Your protection circuit is a considerable improvement in component count and complexity, and while looking at it I realised that it could be modified to provide fast turn-off in the case of loss of AC supply.
 
For the reasons I have already pointed out, as drawn it would anyway only operate after the servo had been defeated, and it cannot detect an offset much below 150mV, so I have redrawn my circuit to incorporate yours, but sampling at the junction of the 2 * 4k7's in the servo loop.
 

 
I'm wondering now if I can get it into 100 * 50 mm.
 
w
 
Oh, anything else?

 

[G.Trenchev]

You'll get better results if running the buffers in high bias WB mode(with a jumper instead of resistor).
Otherwise,I'm pretty sure that offset will be lower that 1 mV even without the servo(I've got same topology amp,just with BUF634).