[Slightly different - threads ?] Let's...

[bcarso]

[quote author="clintrubber"][quote author="mikep"][quote author="bcarso"]
What occurred a while back and I've never seen, but may well have been done: have a sample-hold for the needed correction voltage. Update at prolonged times of zero signal, unless another comparator detects something outside of a permissible window. If one uses an infinite-hold-time S/H (a/d to memory to d/a) you don't have to worry about hold capacitor droop.

Then one could truly say that the servo is out of the signal path most of the time.
[/quote]

precisely what I have been postulating for some time as well. has a digital servo been implemented in a commercial audio product yet?[/quote]
Not that I know of, but it's at least been patented:

http://www.pat2pdf.org/patents/pat6605990.pdf

(if this is what you mean)

Regards,

Peter[/quote]

Not exactly what I describe, although the Philips patent is interesting in the technique of using digital integration in a counter.

What I'm talking about is having a more-or-less conventional analog-domain servo, then using a zero-droop sample-hold for the servo-determined correction variable (voltage or current). The Philips invention (based on a quick glance-over) is integrating and correcting all the time.

 

[clintrubber]

[quote author="bcarso"]Not exactly what I describe, although the Philips patent is interesting in the technique of using digital integration in a counter.[/quote]
Thanks, I'll pass that on (I just did :wink: ).

[quote author="bcarso"]What I'm talking about is having a more-or-less conventional analog-domain servo, then using a zero-droop sample-hold for the servo-determined correction variable (voltage or current).[/quote]
OK, I see, no 'digital factory'.

[quote author="bcarso"]The Philips invention (based on a quick glance-over) is integrating and correcting all the time.[/quote]
Sorry, might have missed that, why wouldn't you want a continuous correction ?

If that system is 'kept running' it does indeed correct all the time, but deliberately in a very lazy manner. Intended for tackling slowly developing offsets (like for instance for gradually increasing capacitor-leakage etc) so a quick correction not needed.

System obviously adaptable for applications with other requirements, and the correction could obviously be frozen: ignore any new incoming info and maintain the correction that's been determined so far.
It's obviously intended for integration without any external components, so not the best approach for an application that's to be made with off the shelf parts.

Regards,

Peter

 

[JohnRoberts]

[quote author="clintrubber"]
Sorry, might have missed that, why wouldn't you want a continuous correction ?

If that system is 'kept running' it does indeed correct all the time, but deliberately in a very lazy manner. Intended for tackling slowly developing offsets (like for instance for gradually increasing capacitor-leakage etc) so a quick correction not needed.

System obviously adaptable for applications with other requirements, and the correction could obviously be frozen: ignore any new incoming info and maintain the correction that's been determined so far.
It's obviously intended for integration without any external components, so not the best approach for an application that's to be made with off the shelf parts.

Regards,

Peter[/quote]

The distinction is like the difference between a fixed trim pot adjusted periodically and a HPF. While with digital integration you could make the pole frequency quite low, and if low enough the distinction becomes moot.

I don't find the digital counter as integrator all that interesting, that's just how you do it on the dark side.

I use a variant on that to estimate AC zero so I can rectify audio in the dark world.

JR

 

[bcarso]

[quote author="clintrubber"][quote author="bcarso"]And for other uses, feeding multiple destinations is not easy without some very odd input stages.[/quote]
In principle the system in the link above is able to do that, see Fig.3.[/quote]

The replication remark was about current drive a la Krell, and nothing to do with the servo. Sorry to try to cover too many bases in a single post.

 

[bcarso]

[quote author="JohnRoberts"][quote author="clintrubber"]
Sorry, might have missed that, why wouldn't you want a continuous correction ?

Peter[/quote]

The distinction is like the difference between a fixed trim pot adjusted periodically and a HPF. While with digital integration you could make the pole frequency quite low, and if low enough the distinction becomes moot.
JR[/quote]

And yes, that's the idea---along the lines of "There's no capacitor like no capacitor."

EDIT: So the determination of net offset goes on continuously, but the update to correction occurs only sporadically, in pauses, and one hopes "in a silent way" (apologies to Miles D.)

 

[clintrubber]

[quote author="bcarso"]EDIT: So the determination of net offset goes on continuously, but the update to correction occurs only sporadically, in pauses, and one hopes "in a silent way" (apologies to Miles D.)[/quote]
No, 'clicking' of the correction wouldn't be acceptable and can be made to be 'silent' by doing the correction in the form of a number of smaller steps. That system is intended to measure and correct all the time, and this during the presence of signal.
And doing its thing without being noticed.

IIRIC it were the Japanese (Panasonic ?) who have various patents on related systems but these lack the 'all the time' & 'during signal' properties.

In an application with fixed/known value coupling cap & known preceeding source-impedance, these R & C can be incorporated in the determination of the allowed stepsize (C has a smoothing action, more or less effectively depending on R).

 

[clintrubber]

[quote author="JohnRoberts"]The distinction is like the difference between a fixed trim pot adjusted periodically and a HPF. While with digital integration you could make the pole frequency quite low, and if low enough the distinction becomes moot.[/quote]
Thanks, I see the principal difference now.

I don't find the digital counter as integrator all that interesting, that's just how you do it on the dark side.

Sure, the world hasn't significantly changed because of that patent, but as you know that's how the game is played (unfortunately):
using a certain technique for a certain application can be patented. And while 'nifty' and 'revolutionary' is obviously welcomed, patents are there for economical reasons, not necessarily for bringing the reader into a state of euphoria.

There's more to that system though, the digital counter is just one of the techniques that constitute it. The others are 'known' techniques as well, but I'd say they co-operate fairly elegantly to realize the required functionality. And it can all be slammed onto the same die as the power-amp(s) live(s) on, so it has an application for 'line-level' opamps as well.
-end of advertisement!-

I use a variant on that to estimate AC zero so I can rectify audio in the dark world.

Sounds interesting to me (but maybe boring to you :wink: )

Regards,

Peter

 

[mikep]

[quote author="clintrubber"]
No, 'clicking' of the correction wouldn't be acceptable and can be made to be 'silent' by doing the correction in the form of a number of smaller steps. [/quote]

you need to filter the output of the D/A. there are a couple of ways to do it. an interesting possibility is to couple the servo signal into the amplifier with a vactrol. should yeild low noise contribution at low frequencies compared to an integrator. the vactrol LED could be driven with the PWM output of a low-cost microcontroller. the vactrol time constant will filter the PWM noise as well as the servo action steps.

mike p

 

[Samuel Groner]

An interesting property of a digital servo would be the possibility to provide easy implementation of fast settling after startup.

The vactrol time constant will filter the PWM noise as well as the servo action steps.

What is the time constant of a typical LDR? I do have my doubts if a digital servo can be made as low noise as a good analogue implementation, at least not without serious filtering.

Samuel

 

[bcarso]

The implementation I had in mind for my scheme was indeed to use some analog variable-gain device (could be analog opto, could be OTA etc.), not an abrupt switching operation. Of course the transitioning device has to itself have good accuracy and offset or it introduces more error than it's correcting.

 

[clintrubber]

[quote author="Samuel Groner"]An interesting property of a digital servo would be the possibility to provide easy implementation of fast settling after startup.[/quote]
Indeed.
Most (integrated) amps go through a power-on sequence of standy/->mute/->operating. During mute-mode (=absence of signal) a 'quick first estimate' of the required compensation can already be determined by a quick algoritm.

I do have my doubts if a digital servo can be made as low noise as a good analogue implementation, at least not without serious filtering.

Still thinking from inside an IC, you'll need a low-noise reference (that might require an external filtering-cap) and proceed wisely with the changes in the correction: small steps, lazy update rate.

Unless there's really something wild going on (say an unavoidable and leaky electrolytic input-coupling cap experiencing unreal & periodic temp-changes), most required corrections don't need to change a lot once some equilibrium has been obtained. That's also the reason such a system can be made lazy.

Criteria for the audibility of correction-steps are dV/dt and d2V/dt2.
The max allowable values depend on everything downstream.

Regards,

Peter

 

[clintrubber]

[quote author="mikep"][quote author="clintrubber"]
No, 'clicking' of the correction wouldn't be acceptable and can be made to be 'silent' by doing the correction in the form of a number of smaller steps. [/quote]

you need to filter the output of the D/A. there are a couple of ways to do it. an interesting possibility is to couple the servo signal into the amplifier with a vactrol. should yeild low noise contribution at low frequencies compared to an integrator. the vactrol LED could be driven with the PWM output of a low-cost microcontroller. the vactrol time constant will filter the PWM noise as well as the servo action steps.

mike p[/quote]
The mentioned system is obviously meant to be integrated, so a Vactrol will be a bit difficult (or as an external component too expensive for an application that wants lots of output-power for just a few dollars).

But if you know where the signal is coming from (so what's the coupling cap & source-impedance) you could involve these for the filtering. In other words: if C & 'R' (the source-impedance) are known & fixed as well as the path downstream (max amp gain, speaker sensitivity and whether the listener is from click-obsessed Japan :evil: or from the rest of the world), then you can determine a suited stepsize etc.
These 'knowns' may not be available for all applications, but they were for that system.

But keeping the transients unaudible is not a really too critical I'd say; the offset-problem solved this way can very well live with lazy intervention, hence spread the correction out over time in minute steps.
But you do need a low-noise reference to have a silent-enough steady-state correction.

Regards,

Peter

 

[mikep]

[quote author="clintrubber"]
The mentioned system is obviously meant to be integrated,
*snip*
for just a few dollars).
*snip*
But keeping the transients unaudible is not a really too critical I'd say; [/quote]

I appreciate that design, but those are rather different goals than the ones that brought me to this concept. it is physically impossible to make a compact, low noise analog servo with a ~10 minute time constant. I looked into switched capacitor filters and I concluded that you still need at least one electrolytic to get it very slow. you can do it digitally and only use ceramic caps. this is great for board space and MTBF. In my view the biggest drawback of bringing a vactrol to the party is lifetime. the aging effects are significant. the limited max temp typ 70C is also problematic.

coincidentally, the slowest vactrols are also the ones with the lowest temperature coefficient and the lowest light history effects. 5C3

 

[JohnRoberts]

Digital A/D seems like bringing a cannon to a knife fight, but for serious long time constants digital is the appropriate technology.

PWM to LDR seems like a plausible way to control a wide range of interfaces.

JR

 

[clintrubber]

[quote author="clintrubber"]
But keeping the transients unaudible is not a really too critical I'd say; [/quote]
I may have been not clear there, maybe in context it was less confusing. Either way, of course I meant that the transients may not be heard at all, even for applications on a budget. And that this shouldn't be to difficult to accomplish (given a few known conditions).

[quote author="mikep"]I appreciate that design, but those are rather different goals than the ones that brought me to this concept. it is physically impossible to make a compact, low noise analog servo with a ~10 minute time constant.[/quote]
Fully agreed, and as said, that system meant for integration. But you need as much as minutes ?

Regards,

Peter

 

[Samuel Groner]

For serious long time constants digital is the appropriate technology.

One should note though that implementing very-low frequency (relative to the sampling frequency) digital filters is not a trivial business; standard topologies (as unfortunately used in most EQ plug-ins) are not suitable as they show horrifying high noise and distortion as well as serious numerical stability problems (which of course is related to noise and distortion), especially if implemented in fix point numerics.

Samuel

 

[JohnRoberts]

[quote author="Samuel Groner"]

For serious long time constants digital is the appropriate technology.

One should note though that implementing very-low frequency (relative to the sampling frequency) digital filters is not a trivial business; standard topologies (as unfortunately used in most EQ plug-ins) are not suitable as they show horrifying high noise and distortion as well as serious numerical stability problems (which of course is related to noise and distortion), especially if implemented in fix point numerics.

Samuel[/quote]

Indeed low audio frequency RC mimics in digital domain can run into filter coefficient resolution issues at high sample rates, while there are techniques to deal with that too (besides floating point).

However a DC auto-zero tweak, that more resembles a S/H or "measure and hold" can hold a correction setting arbitrarily long with precise periodic updates far more easily in the digital domain than analog. Sorry if I wasn't adequately specific.

JR

 

[mikep]

[quote author="Samuel Groner"]
One should note though that implementing very-low frequency (relative to the sampling frequency) [/quote]

true. but with appropriate analog anti-alias filtering and a suitable sample rate it works out nicely. you dont need to do it like an EQ plugin either. the output sample rate of my implementation is not equal to the input rate.

one issue with any digital servo arrises when the LSB of the input (a/d) is smaller than the LSB of the servo output (d/a, pwm, etc). it will tend to "hunt", or oscillate around the desired point. this can be easily delt with in software. the problem is much more difficult to solve though if the closed loop gain of the amp being servoed is variable (in a mic amp you might have 0-60dB gain range) because now the LSB of the servo output has a variable effect on the output offset depending on the gain. your firmware needs to take the gain into account, which means you need to know the gain, which means you might as well use the microcontroller to help do the gain control. Thats my thinking anyway.

high loop gain also implies the need for quite a small LSB step size for the servo output, depending on the maximum input offset you need to be able to cancel. for example: to be able to handle 20mV input offset with 60dB gain and keep the output DC offset at 0V +/- 1mV requires an output D/A with at least 15 bits resolution. it gets alot better if you only have a miniscule offset in the first place. a 10 bit D/A is sufficient to cancel 1mV input offset with the same +/- 1mV output criteria.

mike p

 

[clintrubber]

[quote author="mikep"]you dont need to do it like an EQ plugin either.[/quote]
Right, the amp itself doesn't experience it like a filter being wrapped around it. After reaching equilibrium there's just a steady state DC-correction being applied, with sometimes a few steps to follow eventual changes.

one issue with any digital servo arrises when the LSB of the input (a/d) is smaller than the LSB of the servo output (d/a, pwm, etc). it will tend to "hunt", or oscillate around the desired point. this can be easily delt with in software.

Yes, you need a 'dead zone'. And not want to reduce output-offset to 1 uV.

the problem is much more difficult to solve though if the closed loop gain of the amp being servoed is variable (in a mic amp you might have 0-60dB gain range) because now the LSB of the servo output has a variable effect on the output offset depending on the gain. your firmware needs to take the gain into account, which means you need to know the gain,

True; the kind of amp that system intended for is easy in that respect, either 16 or 26 dB IIRIC. A much larger gain-range involves some more work, but doable.

Regards,

Peter

 

[PRR]

> Cherry, Hooper .... highly recommended Amplifying Devices and Low-Pass Amplifier Design, which has among other things a unified treatment of hollow- and sand-state devices. It is also infamously expensive---you would think by now that more ex-library and estate copies would be showing up, as it dates to 1968!

Darn you and your good ideas!!!!!! Made me $330 poorer.

OTOH.... why didn't someone tell me about this book 40 years ago? It says the things which need to be said, which most authors don't say..... perhaps because they assume you will meet AD&LPAD some day.