piglet <erichpwagner@hotmail.com> wrote:
> On 19/12/2023 5:20 pm, Phil Hobbs wrote:
>> 
>> I think the novel part of JT's design is that it controls what you 
>> actually care about, namely the minimum quiescent bias under transient 
>> conditions.&nbsp; Linear methods only get you the average, which doesn't 
>> guarantee what happens with a large output signal.&nbsp; Mike's method has a 
>> large transient error at the zero crossing, which gets worse at smaller 
>> inputs.
>> 
> 
> No sure about that: JT's design has two big capacitors (100uF and 22uF I 
> recall) in the bias control so it will always be sluggish on transients. 
> My second version has no capacitors in the bias control so should be 
> very adaptive?
> 
> piglet
> 
> 
> 

It&rsquo;s not as good, though. JT&rsquo;s gizmo samples the quiescent bias only at the
zero crossings, and accumulates the results on the big cap on the base of
the rubber diode. 

That takes account of large signal effects, capacitance, and so on. Jim&rsquo;s
100 uF is too big&mdash;something around 1-5 uF is better. You want just enough
to avoid distortion in the low bass. 

The current version of the zip file I posted has a plot file and screen
shot that shows an honest 0.0005 % THD while putting 8 V pp of 20 kHz into
4 ohms. 

Swapping out the $7 THS4631 for a 50-cent OPA172 still does 0.01% THD,
which is much better than your average car speaker. 

Cheers 

Phil Hobbs 

-- 
Dr Philip C D Hobbs  Principal Consultant  ElectroOptical Innovations LLC /
Hobbs ElectroOptics  Optics, Electro-optics, Photonics, Analog Electronics 

On 12/14/2023 9:20 PM, Fred Bloggs wrote:
> On Thursday, December 14, 2023 at 6:05:59 PM UTC-5, John Larkin wrote:
>> On Thu, 14 Dec 2023 22:40:23 +0000, piglet <erichp...@hotmail.com>
>> wrote:
>>> Inspired by the late Jim Thompson's autobias design that we saw in the
>>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang
>>> comparator in setting bias was non-optimal but I liked his idea of
>>> running the pa in pure class B for large signals and setting a minimum
>>> quiescent class A current for small signals.
>>>
>>> My idea uses JT's way of sensing output device current falling below
>>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias
>>> point apart with resistors and a pair of current mirrors. As an IC
>>> designer JT was liberal with current mirrors so I hope he would approve.
>>>
>>> I expect over the decades many clever ways have been devised at
>>> eliminating crossover distortion and autobiasing class A/B so I have no
>>> shame in adding another. Does this look interesting or has it been done
>>> already?
>>>
>>> Here is a pdf schematic:
>>>
>>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1>
>> Most of the audio amp circuits posted here seem to have been designed
>> before the invention of the opamp.
> 
> They're a step up from tubes...

Thank goodness the thread seems to have designed a satisfactory audio 
amp using a 250 MHz 900 V/usec op amp.

On 12/14/2023 6:05 PM, John Larkin wrote:
> On Thu, 14 Dec 2023 22:40:23 +0000, piglet <erichpwagner@hotmail.com>
> wrote:
> 
>> Inspired by the late Jim Thompson's autobias design that we saw in the
>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang
>> comparator in setting bias was non-optimal but I liked his idea of
>> running the pa in pure class B for large signals and setting a minimum
>> quiescent class A current for small signals.
>>
>> My idea uses JT's way of sensing output device current falling below
>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias
>> point apart with resistors and a pair of current mirrors. As an IC
>> designer JT was liberal with current mirrors so I hope he would approve.
>>
>> I expect over the decades many clever ways have been devised at
>> eliminating crossover distortion and autobiasing class A/B so I have no
>> shame in adding another. Does this look interesting or has it been done
>> already?
>>
>> Here is a pdf schematic:
>>
>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1>
> 
> 
> Most of the audio amp circuits posted here seem to have been designed
> before the invention of the opamp.
> 

U were designed before the invention of the opamp

On 19/12/2023 5:20 pm, Phil Hobbs wrote:
> 
> I think the novel part of JT's design is that it controls what you 
> actually care about, namely the minimum quiescent bias under transient 
> conditions.  Linear methods only get you the average, which doesn't 
> guarantee what happens with a large output signal.  Mike's method has a 
> large transient error at the zero crossing, which gets worse at smaller 
> inputs.
> 

No sure about that: JT's design has two big capacitors (100uF and 22uF I 
recall) in the bias control so it will always be sluggish on transients. 
My second version has no capacitors in the bias control so should be 
very adaptive?

piglet

On 2023-12-19 09:56, Phil Hobbs wrote:
> legg <legg@nospam.magma.ca> wrote:
>> On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
>> <pcdhSpamMeSenseless@electrooptical.net> wrote:
>>
>>> On 2023-12-14 17:40, piglet wrote:
>>>> Inspired by the late Jim Thompson's autobias design that we saw in the
>>>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang
>>>> comparator in setting bias was non-optimal but I liked his idea of
>>>> running the pa in pure class B for large signals and setting a minimum
>>>> quiescent class A current for small signals.
>>>>
>>>> My idea uses JT's way of sensing output device current falling below
>>>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias
>>>> point apart with resistors and a pair of current mirrors. As an IC
>>>> designer JT was liberal with current mirrors so I hope he would approve.
>>>>
>>>> I expect over the decades many clever ways have been devised at
>>>> eliminating crossover distortion and autobiasing class A/B so I have no
>>>> shame in adding another. Does this look interesting or has it been done
>>>> already?
>>>>
>>>> Here is a pdf schematic:
>>>>
>>>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1>
>>>>
>>>> <snip>
>>>
>>> I spent a bit more time on JT's design, just for interest's sake.  I
>>> replaced the op amp with a UniversalOpAmp2 that mimics a THS4631 (210
>>> MHz GBW, 1000 V/us), and used the RH111 (an old rad-hard AD version of
>>> the LM311).
>>>
>>> <https://electrooptical.net/www/sed/JimThompsonAudioAmpWithWeirdBiasing.zip>
>>>
>>> The bias doesn't become Class B for large signals--it's solid as a rock,
>>> weirdly.  You just get a 300-ns-ish bias current pulse near the output
>>> zero crossings.  With a nearly full-scale output at 20 kHz, its
>>> distortion is only 0.04%, not bad for having no local feedback.
>>>
>>> The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its
>>> NPN complement, the D44H11.  I put both libraries in the zipfile.
>>>
>>> It works exactly the same with BCV61C duals for the  mirrors, which
>>> would save a couple of packages.
>>>
>>> Cheers
>>>
>>> Phil Hobbs
>>
>> Phil,
>>
>> What's the point of sticking an ideal fixed offset voltage
>> in a circuit that is attenpting to demonstrate practical
>> circuitry? It's what mirrors are for . . .
>>
>> With all later revs of LTspiceXVII, you have to be pretty
>> careful when pulling parts from cmp\standard.xxx. This is
>> particularly true if demonstrating temperature effects.
>> I'm not sure what an ako:2n2904 (or even a stray 2n2906)
>> looks like on your machine.
>>
>> The second sim in your zip produces an error in a not-recently-
>> updated LTSpiceXVII installation ;
>>
>> Missing closing '}' in "{cb)"
>>
>>
>> . . . and won't start.
>>
>> I see no obvious editable brackets to correct this.
>>
> 
> Sorry about that. Just replace the closing paren with a right brace.  I
> obviously didn&rsquo;t save the file between fixing that and zipping it.
> 
> Re The bias voltage source: it makes zero difference to the performance of
> the circuit. It&rsquo;s the comparator loop that interested me, and I thought it
> might interest others too.
> 
> I really like things that work very well using cheap components in
> non-obvious ways.  I wouldn&rsquo;t have expected 0.0004% THD for an 8V p-p
> output into 4 ohms.
> 
> Cheers
> 
> Phil Hobbs
> 

I updated the zipfile with the correction and an auxiliary source to 
cancel the fundamental part of the SJ error voltage.  Use the plot 
settings file to look at the THD waveform.  (The aux source is 
calibrated by eye.)

I put in an annotated screen shot that shows an honest 0.0005% (5 ppm 
amplitude) p-p harmonic distortion. The actual THD is even better than 
that, because the p-p calculation is a bit pessimistic.)

I think the novel part of JT's design is that it controls what you 
actually care about, namely the minimum quiescent bias under transient 
conditions.  Linear methods only get you the average, which doesn't 
guarantee what happens with a large output signal.  Mike's method has a 
large transient error at the zero crossing, which gets worse at smaller 
inputs.

Cheers

Phil Hobbs

-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

legg <legg@nospam.magma.ca> wrote:
> On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
> <pcdhSpamMeSenseless@electrooptical.net> wrote:
> 
> <snip>
>> The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its 
>> NPN complement, the D44H11.  I put both libraries in the zipfile.
>> 
>> It works exactly the same with BCV61C duals for the  mirrors, which 
>> would save a couple of packages.
>> 
>> Cheers
>> 
>> Phil Hobbs
> 
> You can get some idea of the flavors of different bipolar models 
> there are floating around in the LTSpice arena, often with the 
> same ident, from a spreadsheet thrown together last year:
> 
> http://ve3ute.ca/query/bjt_spice_parameter_221218.zip
> 
> RL
> 

Thanks, I&rsquo;ll check it out. 

Cheers 

Phil Hobbs 

-- 
Dr Philip C D Hobbs  Principal Consultant  ElectroOptical Innovations LLC /
Hobbs ElectroOptics  Optics, Electro-optics, Photonics, Analog Electronics 

legg <legg@nospam.magma.ca> wrote:
> On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
> <pcdhSpamMeSenseless@electrooptical.net> wrote:
> 
>> On 2023-12-14 17:40, piglet wrote:
>>> Inspired by the late Jim Thompson's autobias design that we saw in the 
>>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang 
>>> comparator in setting bias was non-optimal but I liked his idea of 
>>> running the pa in pure class B for large signals and setting a minimum 
>>> quiescent class A current for small signals.
>>> 
>>> My idea uses JT's way of sensing output device current falling below 
>>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias 
>>> point apart with resistors and a pair of current mirrors. As an IC 
>>> designer JT was liberal with current mirrors so I hope he would approve.
>>> 
>>> I expect over the decades many clever ways have been devised at 
>>> eliminating crossover distortion and autobiasing class A/B so I have no 
>>> shame in adding another. Does this look interesting or has it been done 
>>> already?
>>> 
>>> Here is a pdf schematic:
>>> 
>>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1>
>>> 
>>> <snip>
>> 
>> I spent a bit more time on JT's design, just for interest's sake.  I 
>> replaced the op amp with a UniversalOpAmp2 that mimics a THS4631 (210 
>> MHz GBW, 1000 V/us), and used the RH111 (an old rad-hard AD version of 
>> the LM311).
>> 
>> <https://electrooptical.net/www/sed/JimThompsonAudioAmpWithWeirdBiasing.zip>
>> 
>> The bias doesn't become Class B for large signals--it's solid as a rock, 
>> weirdly.  You just get a 300-ns-ish bias current pulse near the output 
>> zero crossings.  With a nearly full-scale output at 20 kHz, its 
>> distortion is only 0.04%, not bad for having no local feedback.
>> 
>> The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its 
>> NPN complement, the D44H11.  I put both libraries in the zipfile.
>> 
>> It works exactly the same with BCV61C duals for the  mirrors, which 
>> would save a couple of packages.
>> 
>> Cheers
>> 
>> Phil Hobbs
> 
> Phil,
> 
> What's the point of sticking an ideal fixed offset voltage 
> in a circuit that is attenpting to demonstrate practical 
> circuitry? It's what mirrors are for . . .
> 
> With all later revs of LTspiceXVII, you have to be pretty 
> careful when pulling parts from cmp\standard.xxx. This is 
> particularly true if demonstrating temperature effects.
> I'm not sure what an ako:2n2904 (or even a stray 2n2906) 
> looks like on your machine.
> 
> The second sim in your zip produces an error in a not-recently-
> updated LTSpiceXVII installation ;
> 
> Missing closing '}' in "{cb)"
> 
> 
> . . . and won't start.
> 
> I see no obvious editable brackets to correct this.
> 

Sorry about that. Just replace the closing paren with a right brace.  I
obviously didn&rsquo;t save the file between fixing that and zipping it.

Re The bias voltage source: it makes zero difference to the performance of
the circuit. It&rsquo;s the comparator loop that interested me, and I thought it
might interest others too.

I really like things that work very well using cheap components in
non-obvious ways.  I wouldn&rsquo;t have expected 0.0004% THD for an 8V p-p
output into 4 ohms.

Cheers

Phil Hobbs

-- 
Dr Philip C D Hobbs  Principal Consultant  ElectroOptical Innovations LLC /
Hobbs ElectroOptics  Optics, Electro-optics, Photonics, Analog Electronics 

On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

<snip>
>The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its 
>NPN complement, the D44H11.  I put both libraries in the zipfile.
>
>It works exactly the same with BCV61C duals for the  mirrors, which 
>would save a couple of packages.
>
>Cheers
>
>Phil Hobbs

You can get some idea of the flavors of different bipolar models 
there are floating around in the LTSpice arena, often with the 
same ident, from a spreadsheet thrown together last year:

http://ve3ute.ca/query/bjt_spice_parameter_221218.zip

RL

On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>On 2023-12-14 17:40, piglet wrote:
>> Inspired by the late Jim Thompson's autobias design that we saw in the 
>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang 
>> comparator in setting bias was non-optimal but I liked his idea of 
>> running the pa in pure class B for large signals and setting a minimum 
>> quiescent class A current for small signals.
>> 
>> My idea uses JT's way of sensing output device current falling below 
>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias 
>> point apart with resistors and a pair of current mirrors. As an IC 
>> designer JT was liberal with current mirrors so I hope he would approve.
>> 
>> I expect over the decades many clever ways have been devised at 
>> eliminating crossover distortion and autobiasing class A/B so I have no 
>> shame in adding another. Does this look interesting or has it been done 
>> already?
>> 
>> Here is a pdf schematic:
>> 
>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1> 
>> <snip>
>
>I spent a bit more time on JT's design, just for interest's sake.  I 
>replaced the op amp with a UniversalOpAmp2 that mimics a THS4631 (210 
>MHz GBW, 1000 V/us), and used the RH111 (an old rad-hard AD version of 
>the LM311).
>
><https://electrooptical.net/www/sed/JimThompsonAudioAmpWithWeirdBiasing.zip>
>
>The bias doesn't become Class B for large signals--it's solid as a rock, 
>weirdly.  You just get a 300-ns-ish bias current pulse near the output 
>zero crossings.  With a nearly full-scale output at 20 kHz, its 
>distortion is only 0.04%, not bad for having no local feedback.
>
>The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its 
>NPN complement, the D44H11.  I put both libraries in the zipfile.
>
>It works exactly the same with BCV61C duals for the  mirrors, which 
>would save a couple of packages.
>
>Cheers
>
>Phil Hobbs

Phil,

What's the point of sticking an ideal fixed offset voltage 
in a circuit that is attenpting to demonstrate practical 
circuitry? It's what mirrors are for . . .

With all later revs of LTspiceXVII, you have to be pretty 
careful when pulling parts from cmp\standard.xxx. This is 
particularly true if demonstrating temperature effects.
I'm not sure what an ako:2n2904 (or even a stray 2n2906) 
looks like on your machine.

The second sim in your zip produces an error in a not-recently-
updated LTSpiceXVII installation ;

Missing closing '}' in "{cb)"


. . . and won't start.

I see no obvious editable brackets to correct this.

RL

<sunaeconospam@gmail.com> wrote:
> On Mon, 18 Dec 2023 16:00:09 -0500, Phil Hobbs
> <pcdhSpamMeSenseless@electrooptical.net> wrote:
> 
>> On 2023-12-14 17:40, piglet wrote:
>>> Inspired by the late Jim Thompson's autobias design that we saw in the 
>>> thread "Power Amplifier for 100kHz" I thought his use of a bang-bang 
>>> comparator in setting bias was non-optimal but I liked his idea of 
>>> running the pa in pure class B for large signals and setting a minimum 
>>> quiescent class A current for small signals.
>>> 
>>> My idea uses JT's way of sensing output device current falling below 
>>> threshold but instead of pumping up a "rubber Vbe diode" I pull the bias 
>>> point apart with resistors and a pair of current mirrors. As an IC 
>>> designer JT was liberal with current mirrors so I hope he would approve.
>>> 
>>> I expect over the decades many clever ways have been devised at 
>>> eliminating crossover distortion and autobiasing class A/B so I have no 
>>> shame in adding another. Does this look interesting or has it been done 
>>> already?
>>> 
>>> Here is a pdf schematic:
>>> 
>>> <https://www.dropbox.com/scl/fi/x1tw0nj4d44pgmutxw36p/classAB_rubber_mirrors_autobias.pdf?rlkey=0xr3fv4l2qswyhrombqtanwvo&raw=1>
>>> 
>>> <snip>
>> 
>> I spent a bit more time on JT's design, just for interest's sake.  I 
>> replaced the op amp with a UniversalOpAmp2 that mimics a THS4631 (210 
>> MHz GBW, 1000 V/us), and used the RH111 (an old rad-hard AD version of 
>> the LM311).
>> 
>> <https://electrooptical.net/www/sed/JimThompsonAudioAmpWithWeirdBiasing.zip>
>> 
>> The bias doesn't become Class B for large signals--it's solid as a rock, 
>> weirdly.  You just get a 300-ns-ish bias current pulse near the output 
>> zero crossings.  With a nearly full-scale output at 20 kHz, its 
>> distortion is only 0.04%, not bad for having no local feedback.
>> 
>> The LTspice 17.1 library has the D45H11 low-sat PNP, but weirdly not its 
>> NPN complement, the D44H11.  I put both libraries in the zipfile.
>> 
>> It works exactly the same with BCV61C duals for the  mirrors, which 
>> would save a couple of packages.
>> 
>> Cheers
>> 
>> Phil Hobbs
> 
> Why not try JT's updated design posted by myself and legg?
> 

I can see JT&rsquo;s works but it seemed ugly to me , I wanted to see if there
was a wholly linear way.

My second version (bias is set by lesser of the two currents) looks very
promising and now turns out has indeed been done before, I just found a
Bang & Olufsen patent from 2014 using similar approach.

Dang - fame and fortune elude me again :)


piglet