NDFL DOA?

[nickt]

Just paid $29.95 (ouch!) for EM Cherry's 1977 paper that introduced his "Nested Differating Feedback Loop" amp design. I actually built one of these critters in the 1980's and still have it (minus my Tannoy's unfortunately ).

NDFL amps have higher open loop gain at high frequencies than "normal" amps. This is done by rolling off open loop frequency response higher than 6db per octave (eh 18db). This means lower distortion at high frequencies (because there's more loop gain to cancel it out via negative feedback - Yay!)

What I'm thinking about is:
- would this NDFL topology scale down from a power amp to a DOA?
- if so would simple gain adjustment be possible?
- could the resultant circuit be treated as a normal (discrete) opamp?

Any thoughts?
Thanks!

 

[mikep]

not familiar with that circuit. can you post a schematic?

 

[JohnRoberts]

Yes, NF amplifiers are NF amplifiers so it should apply.

DOA don't have to do as much heavy lifting as power amps so I'm not sure how much benefit to expect from this approach, versus making the DOA very linear and very fast. Some other power amp techniques may also be useful. One consideration that seems overlooked about DOA design is tweaking the DOA for the application. While we have become lazy about just plugging in good GP opamps, the DOA can tweak input stage current density for optimal noise at impedance used, and decompensated for actual closed loop gain.

I wonder if there's a collection of all Cherry's papers? I recall seeing several that struck me as good information about power amp design.

JR

 

[Samuel Groner]

While attempting to reduce distortion you need to have a very (very!) clear idea of what your dominant distortion source is and how it works (i.e. how it generates distortion). For operational amplifiers the dominant source can vary for different configurations/applications; e.g. common-mode distortion and input impedance nonlinearity is usually dominant as a follower, while crossover distortion will be far more important for a high noise gain inverting configuration and heavy output loading.

The NDFL implementations I have seen nest feedback around the second stage, i.e. the first stage (usually a differential pair) and the output buffer do not benefit from additional feedback. This is a IMO unhappy idea for typical topologies as it is just the second stage which usually is the last one to dominate distortion performance; output stage distortion (for class B power amps at least) and first stage slew-induced distortion is much more of a problem. NDFL is often (and successfully) used for CMOS ICs but you can't compare these implementations with bipolar audio circuits.

Of course you could implement a DOA with this technology but I doubt you'll end up with something usefull; likely you get a complicated design at half the performance of a standard implementation. My latest discrete amplifiers using standard topologies have distortion in the order of -150 dB within the audio frequency range for an inverting configuration and full voltage swing--I think that should do the trick for most situations. If you want the same performance for noninverting configurations the input stage get's complicated, but it can be done (or more carefully: reasonably well approximated).

Samuel

 

[nickt]

Thanks Samuel:

you could be right which is why I'm askin :grin:

This is the basic NDFL topology and below is the schematic from the paper. The schematic of my own amp is different (but similar) to the below. Unfortunately the magazine is in a box in storage. Anyway Cherry's circuit complexity is a worry as is the fact that the schematic he presents has "extras" and doesn't follow his written design process (eg where are the three nested loops in the below diagram?? )

Could be me of course - I recall scratching my head over the same things 25 years ago when I built my power amp.

My idea around this topology is that it allows more loop gain while avoiding instability. Instability seems to be an issue for DIY DOA's so my goal is a design with very high performance and good stability that even an eegit like myself can put together.

Thanks so much for responding in detail. I'm sure you're right but it might be fun trying to prove you wrong :grin:

cheers
Nick

 

[Samuel Groner]

Thanks for the schematic. That's not the one I've seen I think--it does have local loops including the output stage, so this surely is beneficial with respect to distortion for a class B design. If you can make it stable...

For a small-signal amp things are nonetheless different as typical loads require much less current than speakers do; it is easy to have a class A output stage which doesn't really need local feedback to be pretty linear.

Samuel

 

[bcarso]

[quote author="nickt"]
My idea around this topology is that it allows more loop gain while avoiding instability. Instability seems to be an issue for DIY DOA's so my goal is a design with very high performance and good stability that even an eegit like myself can put together.
cheers
Nick[/quote]

Ed might be persuaded to give his take on this approach based on current practices---although getting up there in years I think he is still affiliated with Monash U. There was a guy I corresponded with, Paul Rossiter, after his audioXpress article on a MC phono pre-preamp, that knows him well.

I would say that the nested diff approach was successful, clearly, but it is not for the faint of heart. However, in these days of pretty good simulators you could probably take a stab at it. However, the "issue" of instability for DIY DOAs is I think NOT a good motivation---if you can't use the same tools to get a relatively simple topology to be robust, you are even less likely to use Cherry's approach to do the same.

But to boldly go where no man etc. and make an additional performance enhancement by leveraging both recent practices and nested loops, might be fun. I think Halcro Candy has probably done some of this (and has the patents to prove it), but I haven't delved into his work that much as to say.

 

[JohnRoberts]

Another thing that Cherry wasn't afraid of, that most power amp designers generally avoid, is common emitter output stages.. Fun to stabilize and maintain good PSR.

JR

 

[nickt]

[quote author="bcarso"]However, the "issue" of instability for DIY DOAs is I think NOT a good motivation---if you can't use the same tools to get a relatively simple topology to be robust, you are even less likely to use Cherry's approach to do the same.[/quote]
Haha - you're probably right!

This is my first stab at a DOA and I'm by no means a guru. There's an emotional attachment to the NDFL topology as I'm getting back into music, electronics etc after a long break and haven't really seen much regarding NDFL's in the last 25 years (not that I've looked very hard).

WRT simplicity. The Cherry schematic is a bit twisted round itself. That's definitely NOT what I want as a result. Anyway the worst that can happen is I waste my time but learn something along the way.

Thanks for the advice - I'll definitely start out with spice. I'll post the results (don't hold your breath though, I'm pretty slow at DIY)

 

[Samuel Groner]

I think Halcro Candy has probably done some of this.

As far as I've studied the according patents and understud the terminology Halcro does not use nested differentiating feedback loops. He surely uses nested feedback loops, but they tend to be entirely local; e.g. the output stage in his latest class B patent has a second-order local feedback loop but signal gain of the composite output stage is unity.

Another thing that Cherry wasn't afraid of is common emitter output stages.

Indeed; might be interesting if wrestling with efficieny. Just unhappy that he needs both emitter and collector resistors--perhaps we could re-arrange the protection circuitry to omit the former. What I particularly like about his design is that the output stage bias is defined by a feedback loop and not by some potentially lousy thermal tracking.

Samuel

 

[Samuel Groner]

I quickly derived a Spice implementation which appears to basically work well. With the output stage running in class A distortion is fine (though nothing spectacular); surprisingly it is drastically worse compared to standard implementations when switching to class AB--perhaps a result of the common-emitter output topology. It looks like there is local instability during switching of the output transistors.

Samuel

 

[bcarso]

I see that Cherry uses the actual darlington connection for his common-emitter output devices. This of course cuts down on how close you can get to the rail, but he probably found it expedient to avoid the bigger devices' saturation, which would be nasty to disasterous.

Duke Aguiar warned me about common-emitter outputs, which in the old days would have increased my resolve to use them. But for Epson's Ensemble HD center channel and sub amps, I stuck with e-followers.

 

[JohnRoberts]

One of my first DIY amp projects was a variant on using opamp PS terminals to drive a pair of common-emitter power devices. Only 35 or 40W built into an older junk amp chassis to make a Hifi (that didn't suck) for my little sister's apt, on a pocket change budget, back in early '70s.

I don't recall many details but it was slower than dirt to keep it stable. It worked serviceably for quite a few years, but I didn't pursue that path again. There are plenty of quasi-complementary amps that used a CE NPN to make a usable power PNP, but this is not the same as a CE overall topology.

Cherry is surely correct that's it's six of one half-dozen of the other, but e-followers are one less thing to worry about from my POV.

JR

 

[bcarso]

[quote author="JohnRoberts"]One of my first DIY amp projects was a variant on using opamp PS terminals to drive a pair of common-emitter power devices. Only 35 or 40W built into an older junk amp chassis to make a Hifi (that didn't suck) for my little sister's apt, on a pocket change budget, back in early '70s.
JR[/quote]

If they made opamps with guaranteed and fairly-tightly-distributed quiescent currents that approach would probably be adopted more. I've had some designs cross my bench using that, and I warn people that working now with a given date code opamp is no assurance that it will work in the future. You can of course adjust on the production line, which is almost a guarantee of misadjustment sooner or later.

 

[JohnRoberts]

It might be interesting to consider a OTA like the CA3080, where you could directly control operating current. I'm not sure why bother, it's pretty cheap and easy to use the grounded emitter driven rail approach where opamp can run from normal low voltage. I guess this is CE topology too.

More common than I though.

JR

 

[Samuel Groner]

I studied this approach somewhat more and came up with the following schematic: NDFL-DOA-1_r1.pdf

As one can easily see these are just two structurally identical opamps in series, with the first one omitting a common collector output stage. If you try this in real world I strongly recommend to independently test the two opamps for unity gain stability first. If this works well I think the composite opamp should not provide substantial additional problems.

The open loop gain from such a beast is impressive: NDFL-DOA-1_r1_gain.gif

GBW is 10 GHz at 10 kHz and 10 THz at 1 kHz.

I can't help repeating though that the resulting distortion performance will only be superb for configurations without common-mode swing. For others a driven rail concept will probably be needed to gain the advantages of increased loop gain.

Samuel

 

[nickt]

Looks pretty interesting. 

What are your thoughts on the phase response?

 

[Samuel Groner]

Phase and frequency response are inevitably linked in a minimum phase system (which this opamp hopefully is); it simply means that the unity loop gain frequency needs to be above ~3 MHz for stability, i.e. in most applications you'll need a well-sized feedback capacitor.

Samuel