Devil's Staircase

On Saturday, June 1, 2019 at 11:44:56 AM UTC-4, John Larkin wrote:

On Sat, 01 Jun 2019 10:59:58 +0300, upsidedown@downunder.com wrote:

On Fri, 31 May 2019 15:36:39 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:

I want to make a class-D audio amp, 150 watts or so, using a TI
TPA3255 maybe. It's good for 600 watts mono!

I'll use it full-bridge to drive a step-up transformer, probably a
custom toroid. But toroids are especially unhappy with any DC drive,
and the class D part will surely have some DC offset. The TI spec is
60 mV max output offset, which could be a problem into a good
transformer. Speakers don't mind a little DC, but transformers do. DC
can cause stairstepped increase in circulating current, the Devil's
Staircase, until they saturate.

Is it necessary to use a toroid, why not ordinary EI transformer with
possibly air gap ?

A toroid would be about half the footprint and half the weight of a
regular transformer, and would have lower output impedance. I know a
couple of guys who do nice toroids. One did this for us, lower power.

https://www.dropbox.com/s/0bl6jdkw4ljx4e2/400_Hz_Toroid.JPG?dl=0




Or how about a high inductance but low resistance external solenoid
across toroid primary ? Apparently the frequency is in the 400 Hz
ballpark, so even less solenoid inductance should suffice, compared to
50/60 Hz.

I'm guessing that the TI amp would fight to maintain its output offset
voltage. A series cap would be a lot easier. I could quit bitching and
use some big caps. Probably don't need the diodes.

As just a half-baked idea, could you use two transformers? Keep
your toroid for the power, plus a second light-weight transformer
in parallel that's exposed to the toroid's same volt-seconds, to
sense any imbalance?

Then the sensing transformer could be anything you like (E-core,
pot-core, gapped, toroid, saturable cored, etc.), and you could
sense saturation, hysteresis, use a flux gate, Hall sensor, or
whatever crazy scheme you wanted to monitor any imbalance.

Just a notion for the notion bin...

Cheers,
James

 

[Lasse Langwadt Christense]

mandag den 3. juni 2019 kl. 05.24.59 UTC+2 skrev John Larkin:

On Sun, 2 Jun 2019 17:04:11 -0700 (PDT), Lasse Langwadt Christensen
langwadt@fonz.dk> wrote:

mandag den 3. juni 2019 kl. 01.36.05 UTC+2 skrev John Larkin:
On Sun, 2 Jun 2019 16:05:10 -0700 (PDT), Lasse Langwadt Christensen
langwadt@fonz.dk> wrote:

mandag den 3. juni 2019 kl. 00.23.03 UTC+2 skrev John Larkin:
On Sun, 2 Jun 2019 18:07:37 -0400, bitrex <user@example.net> wrote:

On 6/2/19 6:04 PM, bitrex wrote:
On 6/1/19 11:23 AM, John Larkin wrote:
On Sat, 1 Jun 2019 00:42:55 -0400, bitrex <user@example.net> wrote:

On 6/1/19 12:23 AM, bitrex wrote:
On 5/31/19 10:55 PM, John Larkin wrote:
On Fri, 31 May 2019 22:22:37 -0400, krw@notreal.com wrote:

On Fri, 31 May 2019 15:36:39 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:



I want to make a class-D audio amp, 150 watts or so, using a TI
TPA3255 maybe. It's good for 600 watts mono!

I'll use it full-bridge to drive a step-up transformer, probably a
custom toroid. But toroids are especially unhappy with any DC drive,
and the class D part will surely have some DC offset. The TI spec is
60 mV max output offset, which could be a problem into a good
transformer. Speakers don't mind a little DC, but transformers
do. DC
can cause stairstepped increase in circulating current, the Devil's
Staircase, until they saturate.

So I'm thinking I'll add a series blocking cap so I can ignore
any DC
problems. It will have to be big, 10s of millifarads at least.
Biggest
thing on the board. Maybe use a low voltage electrolytic with
antiparallel power diodes, or a shorted bridge, to protect it from
accidental forward or backwards over-voltage.

Lytics will be big, and supercaps don't seem to like ripple
current. I
think.

Any other ideas about driving a transformer from an audio amp?

Feed back a (heavily) filtered signal to the input?  The problem is
that they don't really tell you what the input of the 3255 looks
like,
IIRC (none do).  It's intended to be AC coupled.

TI does spec 60 mV max DC offset at the output, which is pretty good,
but even that could push a lot of DC into my transformer.

There is probably some way to tweak the input with a little DC.
Feedback loop or even a trimpot.



You can drive the transformer bridged but using current-sense feedback
instead of voltage feedback; put a small sense resistor in line with
each amp output going to each end of the transformer and take off
the DC
feedback to the opposite amp from the junction. That way the amps
should
act as their own servo to keep DC out of the transformer.

Lower power example like this for driving audio isolation transformer
for XLR cable:

https://www.dropbox.com/s/8d1flmr8lko2nf1/Screenshot_2019-06-01_00-12-49.png?dl=0




In the case of the TPA3255 I think you would put a low-offset op amp in
front of the inputs and AC couple into that and do something similar but
the 3255 just acts as a power buffer, I don't immediately see anything
in the datasheet that says you can't intentionally apply a small DC
feedback generated offset to its single input per channel as an error
signal.

It's not stated how the normally AC coupled inputs affect the DC
offset, but we have the eval board and I could have a scut bunny set
it up and try it.

One trimpot and 30 seconds of tech time, to turn it, is sure
appealing. If it turns out we don't need it, we can leave it off the
board.

The input resistance of the TPA3255 is 10K, and they want 10 uF input
coupling caps. Tau is 100 ms, which is 1.6 Hz corner. That suggests to
me that the input caps are also used as lowpass filters for the DC
feedback loop, which then suggests we could push the input pins gently
to change the output offset.

Datasheets tend to hide the good stuff.




Does LTC make any class D amp modules included with LTSpice? I'm curious
now about how using a DC servo loop into a first stage opamp driving the
D-amp as a power buffer, instead of a cap would work out in practice.

The problem with that XLR transformer driver circuit as drawn in a
modification is that while there are two feedback loops, one local to
the op-amps and one around the opposite side current sense resistor,
it's assumed that the phase of the AC signals on both inputs of the op
amps will be similar.

but the phase shift produced by the class D output filter complicates
things, where to put the sense resistor. If after the output filter need
to compensate the phase shift somehow and if before need to filter the
switching frequency down to DC. It may not be nearly as easily workable
in the class D bridged power-buffer topology as with linear amplifiers
driving the transformer, bridged.

Putting it after the LC also means have to take the (possibly not well-
defined) R of the L into account as a component of the current sense
impedance.


We plan to put the current sensor just before the customer output,
which is after the filter inductors and the isolation transformer.
Probably a resistive shunt, isolated by a small audio transformer.

It will be an interesting control loop, synthesizing the output
impedance.


use a h-bridge instead and do everything in software/FPGA ?


The TI switcher is a class-D h-bridge. I wouldn't want to try to go
linear.

We do plan to do all the processing in an FPGA.

I wasn't thinking linear but pwm directly from the FPGA to a powerstage, just like numerous variable frequency drives and brushless DC motor drives



We considered that, using GaN fets. The amp would be tiny. But the TI
thing has current limiting, thermal limiting, all sorts of
protections. We'd have to add that to a home-made H-bridge.

Are there any integrated h-bridges with all the suitable protections?
My FPGA guys wouldn't mind outputting PWM (or delta-sigma?). I should
look into that.

this is a bit slow https://www.infineon.com/cms/en/product/power/motor-control-ics/intelligent-motor-control-ics/single-half-bridge-driver/btn8982ta/

maybe? http://www.ti.com/product/TAS5631B

 

[Chris Jones]

On 03/06/2019 10:04, Lasse Langwadt Christensen wrote:

mandag den 3. juni 2019 kl. 01.36.05 UTC+2 skrev John Larkin:
On Sun, 2 Jun 2019 16:05:10 -0700 (PDT), Lasse Langwadt Christensen
langwadt@fonz.dk> wrote:

mandag den 3. juni 2019 kl. 00.23.03 UTC+2 skrev John Larkin:
On Sun, 2 Jun 2019 18:07:37 -0400, bitrex <user@example.net> wrote:

On 6/2/19 6:04 PM, bitrex wrote:
On 6/1/19 11:23 AM, John Larkin wrote:
On Sat, 1 Jun 2019 00:42:55 -0400, bitrex <user@example.net> wrote:

On 6/1/19 12:23 AM, bitrex wrote:
On 5/31/19 10:55 PM, John Larkin wrote:
On Fri, 31 May 2019 22:22:37 -0400, krw@notreal.com wrote:

On Fri, 31 May 2019 15:36:39 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:



I want to make a class-D audio amp, 150 watts or so, using a TI
TPA3255 maybe. It's good for 600 watts mono!

I'll use it full-bridge to drive a step-up transformer, probably a
custom toroid. But toroids are especially unhappy with any DC drive,
and the class D part will surely have some DC offset. The TI spec is
60 mV max output offset, which could be a problem into a good
transformer. Speakers don't mind a little DC, but transformers
do. DC
can cause stairstepped increase in circulating current, the Devil's
Staircase, until they saturate.

So I'm thinking I'll add a series blocking cap so I can ignore
any DC
problems. It will have to be big, 10s of millifarads at least.
Biggest
thing on the board. Maybe use a low voltage electrolytic with
antiparallel power diodes, or a shorted bridge, to protect it from
accidental forward or backwards over-voltage.

Lytics will be big, and supercaps don't seem to like ripple
current. I
think.

Any other ideas about driving a transformer from an audio amp?

Feed back a (heavily) filtered signal to the input?  The problem is
that they don't really tell you what the input of the 3255 looks
like,
IIRC (none do).  It's intended to be AC coupled.

TI does spec 60 mV max DC offset at the output, which is pretty good,
but even that could push a lot of DC into my transformer.

There is probably some way to tweak the input with a little DC.
Feedback loop or even a trimpot.



You can drive the transformer bridged but using current-sense feedback
instead of voltage feedback; put a small sense resistor in line with
each amp output going to each end of the transformer and take off
the DC
feedback to the opposite amp from the junction. That way the amps
should
act as their own servo to keep DC out of the transformer.

Lower power example like this for driving audio isolation transformer
for XLR cable:

https://www.dropbox.com/s/8d1flmr8lko2nf1/Screenshot_2019-06-01_00-12-49.png?dl=0




In the case of the TPA3255 I think you would put a low-offset op amp in
front of the inputs and AC couple into that and do something similar but
the 3255 just acts as a power buffer, I don't immediately see anything
in the datasheet that says you can't intentionally apply a small DC
feedback generated offset to its single input per channel as an error
signal.

It's not stated how the normally AC coupled inputs affect the DC
offset, but we have the eval board and I could have a scut bunny set
it up and try it.

One trimpot and 30 seconds of tech time, to turn it, is sure
appealing. If it turns out we don't need it, we can leave it off the
board.

The input resistance of the TPA3255 is 10K, and they want 10 uF input
coupling caps. Tau is 100 ms, which is 1.6 Hz corner. That suggests to
me that the input caps are also used as lowpass filters for the DC
feedback loop, which then suggests we could push the input pins gently
to change the output offset.

Datasheets tend to hide the good stuff.




Does LTC make any class D amp modules included with LTSpice? I'm curious
now about how using a DC servo loop into a first stage opamp driving the
D-amp as a power buffer, instead of a cap would work out in practice.

The problem with that XLR transformer driver circuit as drawn in a
modification is that while there are two feedback loops, one local to
the op-amps and one around the opposite side current sense resistor,
it's assumed that the phase of the AC signals on both inputs of the op
amps will be similar.

but the phase shift produced by the class D output filter complicates
things, where to put the sense resistor. If after the output filter need
to compensate the phase shift somehow and if before need to filter the
switching frequency down to DC. It may not be nearly as easily workable
in the class D bridged power-buffer topology as with linear amplifiers
driving the transformer, bridged.

Putting it after the LC also means have to take the (possibly not well-
defined) R of the L into account as a component of the current sense
impedance.


We plan to put the current sensor just before the customer output,
which is after the filter inductors and the isolation transformer.
Probably a resistive shunt, isolated by a small audio transformer.

It will be an interesting control loop, synthesizing the output
impedance.


use a h-bridge instead and do everything in software/FPGA ?


The TI switcher is a class-D h-bridge. I wouldn't want to try to go
linear.

We do plan to do all the processing in an FPGA.

I wasn't thinking linear but pwm directly from the FPGA to a powerstage, just like numerous variable frequency drives and brushless DC motor drives

The open-loop all-digital approach tends to produce less accurate output
waveforms than DAC + class D amplifier.

If the power supply rails are not perfectly smooth, or if the
propagation delays or rise and fall times or on-resistance of the output
devices have some dependence on the output current, then you will get
distortion, that can best be corrected with feedback.

The all-digital approach might be ok for a motor drive but you won't
reach the kind of distortion performance that can be achieved by a class
D amplifier that takes analogue feedback from the output terminals, e.g.
these Hypex ones:
https://www.hypex.nl/oem-audio-amplifiers/

 

[Chris Jones]

On 03/06/2019 13:31, John Larkin wrote:

On Mon, 3 Jun 2019 11:53:23 +1000, Chris Jones
lugnut808@spam.yahoo.com> wrote:

On 03/06/2019 08:22, John Larkin wrote:
On Sun, 2 Jun 2019 18:07:37 -0400, bitrex <user@example.net> wrote:

On 6/2/19 6:04 PM, bitrex wrote:
On 6/1/19 11:23 AM, John Larkin wrote:
On Sat, 1 Jun 2019 00:42:55 -0400, bitrex <user@example.net> wrote:

On 6/1/19 12:23 AM, bitrex wrote:
On 5/31/19 10:55 PM, John Larkin wrote:
On Fri, 31 May 2019 22:22:37 -0400, krw@notreal.com wrote:

On Fri, 31 May 2019 15:36:39 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:



I want to make a class-D audio amp, 150 watts or so, using a TI
TPA3255 maybe. It's good for 600 watts mono!

I'll use it full-bridge to drive a step-up transformer, probably a
custom toroid. But toroids are especially unhappy with any DC drive,
and the class D part will surely have some DC offset. The TI spec is
60 mV max output offset, which could be a problem into a good
transformer. Speakers don't mind a little DC, but transformers
do. DC
can cause stairstepped increase in circulating current, the Devil's
Staircase, until they saturate.

So I'm thinking I'll add a series blocking cap so I can ignore
any DC
problems. It will have to be big, 10s of millifarads at least.
Biggest
thing on the board. Maybe use a low voltage electrolytic with
antiparallel power diodes, or a shorted bridge, to protect it from
accidental forward or backwards over-voltage.

Lytics will be big, and supercaps don't seem to like ripple
current. I
think.

Any other ideas about driving a transformer from an audio amp?

Feed back a (heavily) filtered signal to the input?  The problem is
that they don't really tell you what the input of the 3255 looks
like,
IIRC (none do).  It's intended to be AC coupled.

TI does spec 60 mV max DC offset at the output, which is pretty good,
but even that could push a lot of DC into my transformer.

There is probably some way to tweak the input with a little DC.
Feedback loop or even a trimpot.



You can drive the transformer bridged but using current-sense feedback
instead of voltage feedback; put a small sense resistor in line with
each amp output going to each end of the transformer and take off
the DC
feedback to the opposite amp from the junction. That way the amps
should
act as their own servo to keep DC out of the transformer.

Lower power example like this for driving audio isolation transformer
for XLR cable:

https://www.dropbox.com/s/8d1flmr8lko2nf1/Screenshot_2019-06-01_00-12-49.png?dl=0




In the case of the TPA3255 I think you would put a low-offset op amp in
front of the inputs and AC couple into that and do something similar but
the 3255 just acts as a power buffer, I don't immediately see anything
in the datasheet that says you can't intentionally apply a small DC
feedback generated offset to its single input per channel as an error
signal.

It's not stated how the normally AC coupled inputs affect the DC
offset, but we have the eval board and I could have a scut bunny set
it up and try it.

One trimpot and 30 seconds of tech time, to turn it, is sure
appealing. If it turns out we don't need it, we can leave it off the
board.

The input resistance of the TPA3255 is 10K, and they want 10 uF input
coupling caps. Tau is 100 ms, which is 1.6 Hz corner. That suggests to
me that the input caps are also used as lowpass filters for the DC
feedback loop, which then suggests we could push the input pins gently
to change the output offset.

Datasheets tend to hide the good stuff.




Does LTC make any class D amp modules included with LTSpice? I'm curious
now about how using a DC servo loop into a first stage opamp driving the
D-amp as a power buffer, instead of a cap would work out in practice.

The problem with that XLR transformer driver circuit as drawn in a
modification is that while there are two feedback loops, one local to
the op-amps and one around the opposite side current sense resistor,
it's assumed that the phase of the AC signals on both inputs of the op
amps will be similar.

but the phase shift produced by the class D output filter complicates
things, where to put the sense resistor. If after the output filter need
to compensate the phase shift somehow and if before need to filter the
switching frequency down to DC. It may not be nearly as easily workable
in the class D bridged power-buffer topology as with linear amplifiers
driving the transformer, bridged.

Putting it after the LC also means have to take the (possibly not well-
defined) R of the L into account as a component of the current sense
impedance.


We plan to put the current sensor just before the customer output,
which is after the filter inductors and the isolation transformer.
Probably a resistive shunt, isolated by a small audio transformer.

If you're going to use transformer isolation for the shunt, you might as
well use a current transformer instead.

That way you can have less
voltage per turn of the core for a given customer-visible output
current, so a smaller core and/or better accuracy.

I have now realised that this might not be true. There could be
arbitrarily many turns on the isolation transformer after a shunt,
though I still don't think it is a better solution.

Probably better
frequency response too. You'd also face less modifications if you ever
needed to switch to a hall effect sort of current sensor, if you wanted
to extend the frequency range downwards.




A shunt+signal transformer is a lot smaller than a CT, and generally
has better frequency response. I could use a tiny surface-mount shunt
and a small step-up transformer, standard parts. 50 or 100 mV drop in
the shunt would be fine.

Although I am no longer convinced of the inferiority of your solution, I
would be surprised if there are not some small CTs available these days.
As I replied to myself above, I no longer think the CT necessarily
offers better linearity.

A possible advantage of the CT is that the manufacturer will expect it
to be used like this, whereas the audio transformer manufacturer will
not, so that the applications support and characterisation data
available might be better for the CT.

 

[Phil Hobbs]

On 6/3/19 2:17 PM, John Larkin wrote:

On Sun, 2 Jun 2019 19:15:52 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 6/1/19 11:44 AM, John Larkin wrote:
On Sat, 01 Jun 2019 10:59:58 +0300, upsidedown@downunder.com wrote:

On Fri, 31 May 2019 15:36:39 -0700, John Larkin
jjlarkin@highland_snip_technology.com> wrote:

I want to make a class-D audio amp, 150 watts or so, using a TI
TPA3255 maybe. It's good for 600 watts mono!

I'll use it full-bridge to drive a step-up transformer, probably a
custom toroid. But toroids are especially unhappy with any DC drive,
and the class D part will surely have some DC offset. The TI spec is
60 mV max output offset, which could be a problem into a good
transformer. Speakers don't mind a little DC, but transformers do. DC
can cause stairstepped increase in circulating current, the Devil's
Staircase, until they saturate.

Is it necessary to use a toroid, why not ordinary EI transformer with
possibly air gap ?

A toroid would be about half the footprint and half the weight of a
regular transformer, and would have lower output impedance. I know a
couple of guys who do nice toroids. One did this for us, lower power.

https://www.dropbox.com/s/0bl6jdkw4ljx4e2/400_Hz_Toroid.JPG?dl=0




Or how about a high inductance but low resistance external solenoid
across toroid primary ? Apparently the frequency is in the 400 Hz
ballpark, so even less solenoid inductance should suffice, compared to
50/60 Hz.

I'm guessing that the TI amp would fight to maintain its output offset
voltage. A series cap would be a lot easier. I could quit bitching and
use some big caps. Probably don't need the diodes.

Voltages are relative, so it might fight to maintain its output offset
compared to what? The ground pins, I expect. If you used only one
output per package, you could maybe servo the grounds to eliminate the
offset.

Cheers

Phil Hobbs

The output offset is what a speaker (or my transformer) would see, at
the full-bridge output. There's probably a feedback loop from the
outputs to the input diffamp; the numbers look about right.

Right, I knew that. Moving right along....

Cheers

Phil Hobbs

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

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net