Maker Pro
Maker Pro

Distortion from audio power amp

J

John Popelish

Jan 1, 1970
0
Terry Pinnell wrote:
(snip)
I intend to replace the two presets for setting the output DC level
and quiescent current respectively. The first is obviously easy to set
(half supply voltage), but how do I determine optimum setting for
quiescent current please?

The voltage between the bases of Q3 and Q4 have to not only force
those two transistors into slight conduction, but Q5 and Q6, also. I
would connect a millivolt meter across R11 and raise the bias voltage
until a few millivolts drop was visible. But, as I said in an earlier
post, this will require at least 3 diode drops between the basses of
Q3 and Q4. After you set the quiescent current through R11, go back
and measure one of the diode drops (D1 or D2) and the total voltage
between the bases of Q3 and Q4 to see if I am right about this.
 
T

Terry Pinnell

Jan 1, 1970
0
John Popelish said:
Terry Pinnell wrote:
(snip)

The voltage between the bases of Q3 and Q4 have to not only force
those two transistors into slight conduction, but Q5 and Q6, also. I
would connect a millivolt meter across R11 and raise the bias voltage
until a few millivolts drop was visible. But, as I said in an earlier
post, this will require at least 3 diode drops between the basses of
Q3 and Q4. After you set the quiescent current through R11, go back
and measure one of the diode drops (D1 or D2) and the total voltage
between the bases of Q3 and Q4 to see if I am right about this.

OK, many thanks, will do.
 
T

Terry Pinnell

Jan 1, 1970
0
Thanks to all those contributing. I'm pleased to report what seems a
successful working circuit.
http://www.terrypin.dial.pipex.com/Images/PushPullR2=72k.gif

The notes on that are hopefully self-explanatory. I've also attached a
simulated waveform
http://www.terrypin.dial.pipex.com/Images/PushPullR2=72kWave.gif
showing the theoretical output with Vpk-pk = 20.8 V (2.6 V RMS). A
little cross-over distortion is visible.

This circuit is deliberately almost exactly the same as the original
design by Torrens (posted earlier), which I used as part of this
project (A Dog Barking Alarm) in 1981. (Later I plan to try
variations; I've had no shortage of recommendations!)

The only alterations so far are

- the trivial one of R0, a 1k preset to adjust input amplitude

- the small resistors R10 and R11 as recommended by several here
(although I gather R11 could be in a better place in some opinions? I
have not yet had time to try that. In fact I'm beginning to think
Torrens probably left out such resistors purposely!)

- the frequency limiting cap C4 suggested by Win (I've seen no visible
effect throughout tests so far, but these have been limited to 1 kHz
sine.)

Note that I've shown the circuit with R2 at a different value, 72k5,
to the one *actually* in my circuit at present, 47k. My circuit gives
a DC output level of 12.0 V and an output signal similar to that in
the simulation, despite the difference in R2. And I thought it would
be useful to show the theoretical node voltages with the circuit
performing at its present optimum.

With R2 = 47k, the simulation sets the DC output level at 17.1 V, and
clips the +ve peaks of the output. The few changes I did try last
night (including swapping 2N3055s for 2N3053s) made no difference. So
that's one side issue puzzle on which I'd appreciate comments please.

This circuit is currently on breadboard, with croc-clips connecting
case-mounted Q5 and Q6. I'll probably convert that to stripboard
shortly and test again. I'll make R2 and R7 presets at that stage, as
eventual power supply (20-28V) is another issue. Then I can return to
breadboard for further experiments with design changes.
 
T

Terry Pinnell

Jan 1, 1970
0
Jan Panteltje said:
I would connect left side C4 to base Q2, and make it 330pF at least.
(Without spice, without testing without thinking).

Thanks. I haven't altered actual circuit but did try a series of
simulations incoporating that recommendation. As you see in this first
one
http://www.terrypin.dial.pipex.com/Images/PushPull-C4Q12.gif
the difference is visibly hard to detect, at least with existing value
of 68pF. But then thankfully the cross-over distortion is small to
start with.

I then tried a series of values with C4 in your suggested loaction.
http://www.terrypin.dial.pipex.com/Images/PushPull-C4Q2CB.gif
I kept the scale identical and zoomed in on each result. Maybe there's
a flaw in my interpretation, but I reckon a very low value like 10pF
*looks* best. Certainly 330pF was noticably much worse.

BTW, as an aside, (admittedly partly prompted by recent sneers about
overusing simulation), I'd be interested if anyone can *calculate*
this optimum any faster? Even with the benefit of knowing the result
ahead. Took me about 15 minutes using Circuitmaker. Strikes me that
simulation (or even trial and error on the bench) is often the most
productive way of arriving at an optimum design.
You should adjust for a few mA quiecent current.
The cross-over should beless, and, since you have 4 be junctions, why not
use 4 diodes, and a trimpot.

or like this
--------------- base q3
| |
/ | c
\ |/
/<---------|
\ I0 |\/ NPN e
/ |
| |
-----------------base q4

Thanks, I'll try that. See my latest summary and plan of action.
You did not reverse emittor base did you?

No.
 
W

Winfield Hill

Jan 1, 1970
0
Terry Pinnell wrote...
- the frequency limiting cap C4 suggested by Win (I've seen no
visible effect throughout tests so far, but these have been
limited to 1 kHz sine.)

You'll want to test with square-waves, creating a 2 to 10V p-p
output into the load. Be sure to add say 1000pF of capacitance
across the load, simulating a long cable, etc. Evaluate the
output waveform for any ringing. You may want to add a Zobel
damping network (series R - C, say 0.22uF + 10 ohms, 1/2W to
ground) at the output.

Thanks,
- Win

(email: use hill_at_rowland-dot-org for now)
 
J

Jan Panteltje

Jan 1, 1970
0
Thanks to all those contributing. I'm pleased to report what seems a
successful working circuit.
http://www.terrypin.dial.pipex.com/Images/PushPullR2=72k.gif
OK, one thing, why is Q1 needed, note you create a 70 kOhm input
impedance, but the input put circuit is very low, like 4 k.
Better would perhaps be a small emittor resistor in Q2, say 100Ohm,
resulting in an input impedance for Q1 of > 10k (depending on beta).
That would be enough, have better linearity in Q1, so better RF stability.
The feedback network R3 - R2 could then be 20 times smaller values.
There is no need for a very high value for R3.
Finally the usual 22nF 10Ohm series network in parallel with the output
in recommended.
I even think an emittor resistor in Q2 is essential.
JP
 
T

Terry Pinnell

Jan 1, 1970
0
Lacking a distortion meter, I'm looking for ways of roughly assessing
the total harmonic distortion in my output signal. For example, at
what point does harmonic distortion become visible? Assume say a
screenshot captured from my PC-based 'scope and displayed on a typical
17" PC screen. I presumably would not see 0.1%, and I definitely would
see 10% - but where is the threshold?

Generalising from that, anyone know a source of illustrations of a
range of values of distortion in a 1 kHz sine wave? I could then
possibly make my estimate by visual comparison.

Alternatively, my 'scope can also show a Fourier Analysis. Would I get
a reasonable estimate if I went through the chore of identifying the
significant components, converting them to voltages, squaring them,
adding them, and taking the square root?

Are there any other relatively methods anyone can recommend please?
 
J

Jan Panteltje

Jan 1, 1970
0
Thanks. I haven't altered actual circuit but did try a series of
simulations incoporating that recommendation. As you see in this first
one
http://www.terrypin.dial.pipex.com/Images/PushPull-C4Q12.gif
the difference is visibly hard to detect, at least with existing value
of 68pF. But then thankfully the cross-over distortion is small to
start with.

I then tried a series of values with C4 in your suggested loaction.
http://www.terrypin.dial.pipex.com/Images/PushPull-C4Q2CB.gif
I kept the scale identical and zoomed in on each result. Maybe there's
a flaw in my interpretation, but I reckon a very low value like 10pF
*looks* best. Certainly 330pF was noticably much worse.
As for cross-over distortion, of cause strong feedback DECREASES its
effect in the output.
But best is to reduce the cross-over by biasing right.
Like with the transistor circuit I published here.
If you reduce open loop high frequency gain, by using a bigger value for
that C, distortion like that crossover, mainly in the high frequncy
spectrum, will increase.
BTW, as an aside, (admittedly partly prompted by recent sneers about
overusing simulation), I'd be interested if anyone can *calculate*
this optimum any faster? Even with the benefit of knowing the result
ahead. Took me about 15 minutes using Circuitmaker. Strikes me that
simulation (or even trial and error on the bench) is often the most
productive way of arriving at an optimum design.
The graps are interesting, it is not clear to me why it should zero,
it should increase as C gets larger.
Very complex phase shifts may happen for low values of C, and it seems
something cancels.
So to get crossover low in the first place, is the best bet in my view.

That does not mean there are no ways to reduce crossover even further,
I used that in the seventies in, it was popular then, but can't remember
how the system was called.
Sort of a bridge design, with a capacitor trimmer, to cancel the crossover.
Was it in the Quad amps?
Maybe you accidently re-invented or re-discovered it :)

--------------- base q3
| |
/ | c
\ |/
/<---------|
\ I0 |\/ NPN e
/ |
| |
-----------------base q4

JP
 
J

Jan Panteltje

Jan 1, 1970
0
Generalising from that, anyone know a source of illustrations of a
range of values of distortion in a 1 kHz sine wave? I could then
possibly make my estimate by visual comparison.

Alternatively, my 'scope can also show a Fourier Analysis. Would I get
a reasonable estimate if I went through the chore of identifying the
significant components, converting them to voltages, squaring them,
adding them, and taking the square root? Yes

Are there any other relatively methods anyone can recommend please?
Substract divided output from input, should be zero for zero distortion.
But the numbers..
JP
 
W

Winfield Hill

Jan 1, 1970
0
Terry Pinnell wrote...
Here's the output waveform from circuit as it stood yesterday
http://www.terrypin.dial.pipex.com/Images/PushPullOutMay22-1.gif

I have yet to test the very latest circuit, incorporating a couple
of small changes:
http://www.terrypin.dial.pipex.com/Images/PushPull-May22-1.gif

Getting better, you moved R11. Now let's look at the C5 return,
which should be to the power ground, or more accurately, close to
the speaker return-wire node.

You can also add a Zobel network (the simple series R-C form)
returned to that node as well. This makes the circuit robust
by reducing opportunities for high-frequency instability (like
you've already seen) under real-world load conditions, which
includes the speaker's series inductance. A Zobel Rz-Cz also
assists in dealing with another real-world load condition,
parallel capacitance, by insuring that the amplifier load is
no higher than Rz at high frequencies where it can get into
trouble. But another common addition further eliminates this
issue, a parallel L-R in series with the output. The use of a
small inductor isolates the load capacitance at high frequencies,
and the parallel resistor establishes a maximum load impedance.
At low (audio) frequencies the small inductor looks like a wire
and has no effect.

Adding robustness with these four components may allow you move
the amplifier's compensation to higher frequencies (reduce C4),
thereby providing improved transient response.

Thanks,
- Win

(email: use hill_at_rowland-dot-org for now)
 
R

R.Legg

Jan 1, 1970
0
F

Fred Bloggs

Jan 1, 1970
0
Winfield said:
Terry Pinnell wrote...



Getting better, you moved R11. Now let's look at the C5 return,
which should be to the power ground, or more accurately, close to
the speaker return-wire node.

You can also add a Zobel network (the simple series R-C form)
returned to that node as well. This makes the circuit robust
by reducing opportunities for high-frequency instability (like
you've already seen) under real-world load conditions, which
includes the speaker's series inductance. A Zobel Rz-Cz also
assists in dealing with another real-world load condition,
parallel capacitance, by insuring that the amplifier load is
no higher than Rz at high frequencies where it can get into
trouble. But another common addition further eliminates this
issue, a parallel L-R in series with the output. The use of a
small inductor isolates the load capacitance at high frequencies,
and the parallel resistor establishes a maximum load impedance.
At low (audio) frequencies the small inductor looks like a wire
and has no effect.

Adding robustness with these four components may allow you move
the amplifier's compensation to higher frequencies (reduce C4),
thereby providing improved transient response.

Thanks,
- Win

(email: use hill_at_rowland-dot-org for now)

He needs to increase that Q2 bias current by a factor of 10 minimum- and
chuck using any 3055 that has HFE<25 at 1.5A- wonder what the insistence
on leaving that Q4 an unspecified mystery component-also that ac-beta of
0.01 together with a forward gain of something around 300 gets a
T=1+beta*G of something like 3-5 which is only marginally better than no
distortion compensation at all.
 
J

Jim Thompson

Jan 1, 1970
0
Und zo - ze qveschun: vy ist ze zmaller vafeform more distorted zan ze
bigger vafeform?

Diz iz amblivyer, yah?

Id has negadiv distorzhun?

Such genyus!

RL

Sno-o-o-o-ort! ROTFLMAO!

...Jim Thompson
 
T

Terry Pinnell

Jan 1, 1970
0
Winfield Hill said:
Terry Pinnell wrote...

Getting better, you moved R11. Now let's look at the C5 return,
which should be to the power ground, or more accurately, close to
the speaker return-wire node.

You can also add a Zobel network (the simple series R-C form)
returned to that node as well. This makes the circuit robust
by reducing opportunities for high-frequency instability (like
you've already seen) under real-world load conditions, which
includes the speaker's series inductance. A Zobel Rz-Cz also
assists in dealing with another real-world load condition,
parallel capacitance, by insuring that the amplifier load is
no higher than Rz at high frequencies where it can get into
trouble. But another common addition further eliminates this
issue, a parallel L-R in series with the output. The use of a
small inductor isolates the load capacitance at high frequencies,
and the parallel resistor establishes a maximum load impedance.
At low (audio) frequencies the small inductor looks like a wire
and has no effect.

Adding robustness with these four components may allow you move
the amplifier's compensation to higher frequencies (reduce C4),
thereby providing improved transient response.

Thanks, will follow through. Latest at 24/0V gave me 7V rms into 8R,
with clean looking output.
 
T

Terry Pinnell

Jan 1, 1970
0
He needs to increase that Q2 bias current by a factor of 10 minimum- and
chuck using any 3055 that has HFE<25 at 1.5A- wonder what the insistence
on leaving that Q4 an unspecified mystery component-also that ac-beta of
0.01 together with a forward gain of something around 300 gets a
T=1+beta*G of something like 3-5 which is only marginally better than no
distortion compensation at all.

Q4 is a TO5 but just couldn't properly identify it from worn markings.
Before finalising on stripboard I'll probably ring the changes with a
few others, and all things being equal I'll use a known type.
 
T

Terry Pinnell

Jan 1, 1970
0
Und zo - ze qveschun: vy ist ze zmaller vafeform more distorted zan ze
bigger vafeform?

Diz iz amblivyer, yah?

Id has negadiv distorzhun?

Such genyus!

Good point - hadn't noticed that. Signal is from my Philips PM5134
function generator. To rule out its being some artifact of my PC-based
PicoScope, I'll view with my analog 'scope too.
 
T

Terry Pinnell

Jan 1, 1970
0
Terry Pinnell said:
Good point - hadn't noticed that. Signal is from my Philips PM5134
function generator. To rule out its being some artifact of my PC-based
PicoScope, I'll view with my analog 'scope too.

As I suspected, that was merely a characteristic of my ADC 'scope at
that particular scale.

I've just finished getting the circuit back onto stripboard and it
looks fine. About 7.3 V rms into 8 ohms.
http://www.terrypin.dial.pipex.com/Images/PushPullOutMay23.gif

On my Hameg analog 'scope, both input and output look even cleaner of
course.
 
Top