An Improved Dual Class-A Output: The "DCAO2"

Where we discuss new analog design ideas for Pro Audio and modern spins on vintage ones.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by JR. »

Gold wrote:
JR. wrote: PS I've been kicking the tires on Bruno Putzeys class D modules but 70EU per channel has me in sticker shock (for 6 channels of audio). They also sell a switching PS module so it should be like lincoln logs or lego blocks, just not my idea of cheap.
As I have recently found out prefab is much more expensive than DIY. It's a strong dollar. Now is the time to buy if you are. They do sound great. I guess it depends how busy you are and how much your time is worth to you. All you have to do is the casework which might not be your idea of fun.

OT: After getting a quote on a nice prefab cabin I've started researching house construction. Geez, framing doesn't seem too difficult or time consuming for a one story not too big structure. As I've been telling everyone, I seem to be incapable of doing things the easy way.
I'll start a new thread if I go down this road.
=======
re: building a house, i don't know anybody, who wanted to build a second house.. apparently not as much fun as it looks. :lol:
===========
@ Wayne, sorry if I appear pedantic about slew rate, but for years it was a design goal for power amp designers (and IC designers). In the bad old days, IC companies would publish small signal and large signal response. Generally the small signal was linear rise time limited, and large signal was with some internal path saturated (generally compensation node current limited).

I generally favor a LPF in the front end but realistically, any practical source material you feed this amp will already be band limited.

JR
Cancel the "cancel culture", do not support mob hatred.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by mediatechnology »

if I appear pedantic about slew rate, but for years it was a design goal for power amp designers (and IC designers). In the bad old days, IC companies would publish small signal and large signal response. Generally the small signal was linear rise time limited, and large signal was with some internal path saturated (generally compensation node current limited).
Well, in order to get the slew rate up to 100-1000 V/µs (though it doesn't need to be) I considered using comparators in place of all the op amps and THAT1240s.
Though open loop linearity would suffer, I figure that with enough open loop gain and feedback I can fix anything. :lol:

Instead, I ordered some INA134s from Mouser which I expect today.
The THAT1240s are stellar but have slightly lower 25 mA Isc output current capability and 36V maximum rail voltage.
The THAT1240s are perfectly fine for line/headphone applications up to 34V.

For the simple Dual Class-A 10W power amp space heater the INA134 may be a better choice.
The INA134's output current is 80 mA (Isc) and the maximum supply voltage is 40V.
It could be used with 36V rails and not be on the edge.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by JR. »

mediatechnology wrote:
if I appear pedantic about slew rate, but for years it was a design goal for power amp designers (and IC designers). In the bad old days, IC companies would publish small signal and large signal response. Generally the small signal was linear rise time limited, and large signal was with some internal path saturated (generally compensation node current limited).
Well, in order to get the slew rate up to 100-1000 V/µs (though it doesn't need to be) I considered using comparators in place of all the op amps and THAT1240s.
Though open loop linearity would suffer, I figure that with enough open loop gain and feedback I can fix anything. :lol:
Back in the day they designed extremely high slew rate op amps for sample and hold applications. They were designed such that whenever the input long tail pair saturated (maybe 50mV or so differential), additional circuitry would cut in parallel to the LTP to charge/discharge the internal compensation cap node faster. The trick of course is to have these recover and stabilize fast enough that you actually get some useful benefit from the higher slew rate.

Of course this is nonsense for linear circuits, but for the "more is always better" phools, some were attracted to these.

JR
Instead, I ordered some INA134s from Mouser which I expect today.
The THAT1240s are stellar but have slightly lower 25 mA Isc output current capability and 36V maximum rail voltage.
The THAT1240s are perfectly fine for line/headphone applications up to 34V.

For the simple Dual Class-A 10W power amp space heater the INA134 may be a better choice.
The INA134's output current is 80 mA (Isc) and the maximum supply voltage is 40V.
It could be used with 36V rails and not be on the edge.
Cancel the "cancel culture", do not support mob hatred.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by mediatechnology »

I received some INA134 from Mouser to compare them to the THAT1240 in the DCAO pre-driver circuit.

My initial concern is the potential of instability due to the INA134s lower small-signal bandwidth, 3.1 MHz (typ) vs. 8.6 MHz (typ) for the THAT1240.

The THAT1240/INA134 pre-driver stage is inside the feedback loop.
The INA134's "slowness," relative to the NJM2114 gain stage, does cause oscillation when a local Cfb of 22 pF is used.
The THAT1240 predriver is stable with 22 pF Cfb.
Raising Cfb back to 100 pF makes it work with INA134s.

The resulting small signal bandwidth of the entire amp is about 500 kHz with the INA134 vs. 1.4 MHz for the THAT1240.
The 170 kHz power bandwidth using the INA134 is slightly higher due to the INA134's 14V/µs slew rate.

The THAT1240 is a better choice for headphone and line amp applications owing to its higher bandwidth.

The Dual Class-A 10W version needs to be tested using the new DCAO2 circuit.
The INA134's 80 mA output current is more than sufficient to meet the base current needs of big TO-3P transistors when higher current is needed for driving speakers.
Having the higher output current of the INA134 available eliminates a second pair of pre-driver transistors.

It's good to know that both the THAT1240 and TI INA134 can be used.
rocksteady40
Posts: 2
Joined: Tue Apr 26, 2016 10:45 am
Location: Boston

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by rocksteady40 »

Wayne,
It would seem that a 500kHz bandwidth would be more than adequate for most of us :D Or does that result in less bandwidth for “large signals” (ie 10W version)? Sorry, my amplifier theory is alluding me :(

Thanks
Bob
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by mediatechnology »

Hi Bob - Thanks for joining us here at the forum!

I think the 10W version will have about the same 500 kHz BW when built with INA134s.

I've been measuring the THD with the INA134 pre-driver using BD139/140 outputs and it appears similar to the THAT1240 driving 30 Ohm loads.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by mediatechnology »

With the INA134s installed as the pre-driver I decided to look at 10 kHz THD.

The open loop THD performance of the DCAO2 (below 100 mW) isn't influenced much by the bandwidth of the pre-driver.
A THAT1240 or INA134 as pre-driver doesn't make much difference.

Closed loop however (for power levels from 100 mW to 2W) the bandwidth of the THAT1240/INA134 pre-driver does make a difference.
The THAT1240 has a typical 8.6 MHz unity gain bandwidth; the INA134 is 3.1 MHz typically.
For the INA134 to be as stable as the THAT1240, the local feedback capacitor around the error op amp/gain stage has to be increased from 22 pF to 100 pF.

As Jim Williams points out in his series on op amp boosters, the error amp needs to be slower than the output stage. (See part two: https://www.proaudiodesignforum.com/for ... ?f=12&t=13)
The THAT1240/INA134 stage is inside the feedback loop and is part of the output stage.
If the pre-driver stage is slow, the error amp must be slower.

The added compensation required reduces high frequency open loop gain and the amount of corrective feedback available.

The 10 kHz THD at elevated power levels show this: The THAT1240 has significantly lower THD because it has about 2.75X the bandwidth.

After discovering rising THD vs. frequency I decided to re-visit the use of op amps to act as pre-drivers.

THAT1240s were chosen over op amps because they have accurate gain.
The signal gain of the NPN/PNP pre-drivers should be closely matched to reduce even-order distortion.
This is particularly important when the DCAO2 is being used open loop.
THAT1240s eliminated the requirement for a trim that an op amp circuit based on 1% resistors would require.

The motivation to try INA134s instead of the THAT1240 was to obtain higher output current when the DCAO2 was used in 8 Ohm applications.
The lower bandwidth of the INA134 however requires the error amp to be more heavily compensated and the THD at HF not as good as the THAT1240.

The use of op amps as pre-drivers - something I originally tried - seemed to offer the best of both: High output current and higher pre-driver bandwidth.
The only downside seemed to be the requirement for a trim.
But this is DIY and trims aren't avoided like they would be in a production environment.

Since I was going to try using op amp drivers one more time I rebuilt one side of the protoboard and plugged in a couple of NJM2114s.

More Fun...

Stabilizing the NJM2114s - which are turbo charged NE5532s - turned out to be an interesting exercise.
The NE5532 was equally interesting.

The NJM2114 has a 15 MHz unity gain bandwidth and 15 V/µs slew rate versus the NE5532's 10 MHz BW and 8 V/µs SR.
Both have higher bandwidth than the THAT1240 or INA134s internal op amps.
But they performed worse.

To make a long story short it took a lot of tuning and the resulting Cfb turned out to be 100 pF + 1K.
The reason Cfb had to be so large is because the NE5532/NE5534/NJM2114 family have an interesting "resonance" in their out-of-band response around 500 kHz.
If you sweep slowly from 500-600 kHz there can be a peak in the response up to 6 dB.
Small-signal and large-signal square waves also reveal this.
With there being three NJM2114/5532 stages in total, with all three in the feedback loop, keeping overshoot and stability tamed required that the error amp be heavily "slugged."
The resulting 10 kHz THD with this resolved was still higher than the INA134s.

You can see the open loop response "hook" in several published 5534/5532 datasheets:

Image

The above appears to be drawn with the hook above 10^5 and several other datasheets also show the discontinuity.
Some do not.

When you sweep past this frequency the response looks like a snap oscillation but its actually a very narrow Q response peak.
It looks like the nested compensation internal to the 5532 has one stage where compensation seems to end and the next stage picks it up.
In most applications this discontinuity is of no consequence.
As John usually points out, out-of-band material should be filtered out at the input.
But because the feedback loop of the pre-driver are nested inside the DCAO2's global feedback the hook creates unnecessary problems.

Based on the results with the NE5532/NJM2114 I decided to try a different op amp family. My first choice was the OPA1612.

The OPA1612 is a Silicon/Germanium wonder.
The small-signal BW is 40 MHz with a SR of 27 V/µs.
It has 40 mA outputs.

The results are very, very good.
With the OPA1612 I didn't have to work around the 5532's 500 kHz weirdness.
I could reduce Cfb back to 10 pF and maintain stability into capacitive loads of 30R||1nF.

This is the DCAO2/OPA1612 THD at 1.5W, 10 kHz into 30 Ohms 1 nF:

Image
Dual Class-A DCAO2 using OPA1612 pre-drivers at 1.5W into 30 Ohms and 1 nF.
Red is Gen/Mon shifted by +12dB, Green is THD of the DCAO2


What I discovered is that several "fast" op amps worked well as the pre-driver. The key was to not use a 5532/2114 as the pre-driver.

For all but the last FFT the error amp and pre-driver are the same op amp type and the Cfb is 10 pF.

DCAO2/OPA2604 THD at 1.5W, 10 kHz into 30 Ohms 1 nF:

Image
Dual Class-A DCAO2 using OPA2604 pre-drivers at 1.5W into 30 Ohms and 1 nF.
Red is Gen/Mon shifted by +12dB, Green is THD of the DCAO2


The OPA2604 used a Cfb of 10 pF.
Curiously, the even-order null point was different for the OPA2604. The OPA2604's own even order may be getting nulled by the trim.

DCAO2/OPA2134 THD at 1.5W, 10 kHz into 30 Ohms 1 nF:

Image
Dual Class-A DCAO2 using OPA2134 pre-drivers at 1.5W into 30 Ohms and 1 nF.
Red is Gen/Mon shifted by +12dB, Green is THD of the DCAO2


The OPA2134 with its 8 MHz BW is still a respectable performer.


The NJM2114 works great as an error amp. This is the FFT of the NJM2214 driving an OPA2604 pre-driver:

Image
Dual Class-A DCAO2 using NJM2214 error amp and OPA2604 pre-drivers at 1.5W into 30 Ohms and 1 nF.
Red is Gen/Mon shifted by +12dB, Green is THD of the DCAO2


This particular configuration is stable with no Cfb because the input error amp is slower than the pre-driver.

When the NJM2114 is used both as error amp and pre-driver, the THD performance is worse due to the required heavy compensation of 100 pF:

Image
Dual Class-A DCAO2 using NJM2214 error amp and NJM2114 pre-drivers at 1.5W into 30 Ohms and 1 nF.
Red is Gen/Mon shifted by +12dB, Green is THD of the DCAO2


A 20 dB reduction in THD at 10 kHz is available simply by making the "correct" op amp choice for the pre-driver.

The overall performance improvement by using high-bandwidth op amps in the pre-driver have caused me to re-think the use of line receiver ICs.
rocksteady40
Posts: 2
Joined: Tue Apr 26, 2016 10:45 am
Location: Boston

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by rocksteady40 »

Wow, lots of great work there Wayne. Thank you for the very concise descriptions of your investigations.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by JR. »

5532.jpg
5532.jpg (43.54 KiB) Viewed 14544 times
This may be TMI but IIRC the 5534/32 used a novel "split pole(?)" compensation scheme, instead of the more typical dominant pole compensation. I suspect that compensation circuitry explains your above band kinks in the open loop gain plot.(see the 3 caps around the second gain stage stage). The open loop phase plot may look even worse. :oops: Dominant pole looks like a one pole integrator open loop (90' phase shift). That spit pole will have a more complex phase response between the poles.

In the context of the 1970s this was a kick ass op amp, for modest gain audio applications. This is probably still respectable today, but there are newer better amplifiers available now, some 4 decades or so later.

I've used truckloads of 5534/5532 over the decades with no regrets, but for price no object design, we can do better.

Thanks for sharing.

JR
Cancel the "cancel culture", do not support mob hatred.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Improved Dual Class-A Output: The "DCAO2"

Post by mediatechnology »

The light-bulb-turning-on moment was when I associated the "kink" I was seeing in the frequency response with the "kink" in the NE5532/NJM2114 open loop gain and frequency.

I revisited the post I did on the 5534 die photo and saw the LM318 has a similar kink in gain and phase also about 500 kHz. https://www.proaudiodesignforum.com/for ... 54&start=2

Though I'm not comfortable running the NJM2114 feeding the OP2604 without any compensation cap, it's on my bench in that configuration right now pumping 20V P-P into 30 Ohms in parallel with 10 nF.
The THD (including the generator's residual of 0.0007%) is 0.0033% loaded vs. 0.0014% unloaded at 10 kHz with 0R build-out.
With a 10 pF Cfb added for good measure the THD is still only 0.0064% loaded.

The all BIFET or all OPA1612 solutions seem to be the best.
The only downside to the OPA1612 is the requirement for an SMT/DIP adapter.
Brown Dog stock OPA1612 DIP adapters now.

Measures good - looks like hell:

Image
Post Reply