Design review of my new discrete audio buffer

[Lars Nielsen]

Hi all,

I have been working for some years now, on a new audio buffer, for use in preamps and d/a converters. It is going to be af complete DIY project, posted on my website.

All of the selected solutions have been tested in listning tests and measered with serious equipment.

Before I continue with the final PCB layout, I would like to invite you all, to give comments, tips and ideas, to the design..

See schematic here: http://www.diyhifi.dk/1A.pdf

Best Regards
Lars

 

[Lars Nielsen]

No comments..

Really?

 

[DF96]

Is the DC servo stable? It appears to have a large number of low pass filters with similar time constants - exactly the conditions for loop instability. I would expect at least a gain peak at a few Hz, or maybe oscillation.

 

[bear]

Specs?

Real world vs. SPICE plots?

A look at the square wave?

I dunno...

As Count Basie said "...if it sounds good it IS good..."

_-_-bear

 

[aparatusonitus]

How about output noise filter instead single R57 at integrator input...also you have to protect integrator opamp J-fet input against the rails.

 

[Lars Nielsen]

Is the DC servo stable? It appears to have a large number of low pass filters with similar time constants - exactly the conditions for loop instability. I would expect at least a gain peak at a few Hz, or maybe oscillation.

I have been working very much with the DC servo, because i have experienced that many of the typical servoes in other constructions have influence on the sound quality.. and THD.

The 680nF is selected because of high capacity at best price, and the 221K resistor is selected to have minimum influence on dc offset error that will appear because of the input current in LF412.

The servo takes about 6-7 seconds to get the output below 0,1mV offset, so in real life very precise. 20Hz is below -120dB.

I have been trying to use other type of opamps, but then I see some unstability in the first servo buffer where no parallel resister are on the feedback capacitor..

But do you think it would be better to have differing filter values?

PS: Tonight i will post some prototype pics on my website..

/Lars

 

[Lars Nielsen]

Specs?

Real world vs. SPICE plots?

A look at the square wave?

I dunno...

As Count Basie said "...if it sounds good it IS good..."

_-_-bear

Well.. the Spice plots are very similar to real life, the bandwith is a little bit lower in real.. I think it is something about the transistors..

I have mostly been doing spice work when i made the base work arround temparature stability in the stage..

The square wave.. i will work on a plot of that, but with a bandwith of 0,05 to 5Mhz i would expect to see af perfect picture..

/Lars

 

[Lars Nielsen]

How about output noise filter instead single R57 at integrator input...also you have to protect integrator opamp J-fet input against the rails.

I do not understand what you mean with output filter?

Yes.. very good point, the opamp needs to be protectet.. I can see that maximum input voltage is 15V.

2 pcs. of 1n4148 across C29 would do the job, and they would actually also make the DC servo faster at startup, because the rail would not charge C29 at ramp up.

Another solutions would på 2 pcs. of zenerdiodes of 15V, but I am not happy with the zener at that point in the circuit.. What do you think?

 

[DF96]

Does the output settle monotonically to around zero volts, or does it oscillate? The simple way to loop stability is to have no more than two roll-offs. Failing that, have one dominant roll-off, with the others have frequencies much lower/higher (for a DC servo, higher). Failing that, do a full stability analysis. Having a string of similar roll-offs is about the worst thing to do!

"20Hz is below -120dB" means what? I assume this is the result of simulating/calculating the DC servo in isolation, as it is too precise to have been measured. Look at 1-5Hz; this is where the trouble may be. You need to consider phase as well as amplitude.

 

[Lars Nielsen]

Does the output settle monotonically to around zero volts, or does it oscillate? The simple way to loop stability is to have no more than two roll-offs. Failing that, have one dominant roll-off, with the others have frequencies much lower/higher (for a DC servo, higher). Failing that, do a full stability analysis. Having a string of similar roll-offs is about the worst thing to do!

"20Hz is below -120dB" means what? I assume this is the result of simulating/calculating the DC servo in isolation, as it is too precise to have been measured. Look at 1-5Hz; this is where the trouble may be. You need to consider phase as well as amplitude.

When DC servo adjust the output to zero it goes like this example:
Startup output = +1V, servo correct to -0.5V, servo correct to +0.25V, servo correct to -0.12V, servo correct to +0.06, and so on...

I think that it is the right way to do it? If I make the servo so fast, that it doesn't make the swing, the servo will have influence on the sound..

Yes, the 120dB is the calculated value of the servo filter.. Yes the gain of the servo is high at 1hz.. But is the way it adjust okay?

 

[peranders]

Is the DC servo stable? It appears to have a large number of low pass filters with similar time constants - exactly the conditions for loop instability. I would expect at least a gain peak at a few Hz, or maybe oscillation.

You should have max two poles in the DC servo loop and if they suffiently apart you will get a microscopic hump in the frequency response. Good work Lars, but why don't you make the schematics in colour? Easier to read.

Oops, forgot to read following posts. When I look closely I see that there are two integrators. Only one is needed.

 

[peranders]

I have been working very much with the DC servo, because i have experienced that many of the typical servoes in other constructions have influence on the sound quality.. and THD.

/Lars

Lars, the DC servo opamp should be a good one and the cut off frequency shouldn't be too low. 1-5 Hz is recommended (by me) because if you have a lower frequency you will have fluctuations due to temperature shifts. The DC servo opamp must be good up to 1 kHz at least since you have it in the audio path.

 

[peranders]

A DC servo can look like this. The example is a non-inverting circtuit and an inverting one is even easier.

 

[Lars Nielsen]

You should have max two poles in the DC servo loop and if they suffiently apart you will get a microscopic hump in the frequency response. Good work Lars, but why don't you make the schematics in colour? Easier to read.

Oops, forgot to read following posts. When I look closely I see that there are two integrators. Only one is needed.

Thanks.. Schematics in colour.. hmm I thougt that schematics were suposed to be boring..

Yes.. there are two opamps.. The LF412 is a dual opamp, so it is there anyway.. But if I remove the second opamp, will the filter not be able to control the current generator, and the filter will also be loaded, so bigger caps and smaller resistors will be needed..

At the beginning I started with the filter of C26/C27 at the input of U1B, but then will the input leak current of the opamp have influence on the overall DC precision.

 

[Lars Nielsen]

Lars, the DC servo opamp should be a good one and the cut off frequency shouldn't be too low. 1-5 Hz is recommended (by me) because if you have a lower frequency you will have fluctuations due to temperature shifts. The DC servo opamp must be good up to 1 kHz at least since you have it in the audio path.

You are right about the temparature drift can have an influence when the servo is slow.. but hopefully most stages are so stable that this not can be an issue.

Up to 1Khz? Why? The signal most never be able to go back through the servo, that would make af large amount of THD.

 

[Lars Nielsen]

A DC servo can look like this. The example is a non-inverting circtuit and an inverting one is even easier.

Well.. It is basicly the same circuit as mine.. and I also started with this more simple servo.. But listening tests showed me that much more filtering is needed. My transistor stage can run without a servo, so it is easy to check..

In my tests I discovered that a 3 pole of 221k x 680nF is not enough.. it can be heard.. the 4 pole could not be heard, but to be absolutly sure I inserted the 5.

 

[Lars Nielsen]

Some pictures of the prototypes is now on my website: DIYhifi.dk - Forside

 

[peranders]

Up to 1Khz? Why? The signal most never be able to go back through the servo, that would make af large amount of THD.

You have full signal at 1-5 Hz and 10 Hz 20 dB lower, 100 Hz 40 dB etc. so at 1 kHz you will have a tiny amount of the signal left and at 10 kHz you are near the noise level. Why don't use LTSpice and do some probing at the opamp output?

 

[peranders]

At the beginning I started with the filter of C26/C27 at the input of U1B, but then will the input leak current of the opamp have influence on the overall DC precision.

The input bias currents are so small that you won't notice them. You may choose AD8610 with 25-100 uV offset. Excellent opamp for audio and even better for a DC servo.

 

[Lars Nielsen]

You have full signal at 1-5 Hz and 10 Hz 20 dB lower, 100 Hz 40 dB etc. so at 1 kHz you will have a tiny amount of the signal left and at 10 kHz you are near the noise level. Why don't use LTSpice and do some probing at the opamp output?

If the servo only lower 100hz to -40dB it would be a direct global negative feedback, which is out of phase..

I am not sure I see your piont here..?