Delta-Sigma encoding...

[Liz Tuddenham]

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

--
~ Liz Tuddenham ~
(Remove the \".invalid\"s and add \".co.uk\" to reply)
www.poppyrecords.co.uk

 

[John Larkin]

On Sun, 9 Apr 2023 21:45:39 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

 

[John Larkin]

On Sun, 9 Apr 2023 21:45:39 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

In the bipolar case, equal 1s and 0s is 0 volts = mid scale. That\'s
how audio does it.

The unipolar case is positive voltage proportional to duty cycle.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

A delta-sigma ADC will servo to get correct. The integrator integrates
the difference between the input and the duty-cycle feedback.

A d-s DAC is really a lowpass filter, which also recovers from a
disturbance. After a while, it forgets.

Some people like to use a sinc3 lowpass filter to recover the output
of a d-s data stream. It has differentiators and integrators inside
that will never recover from an error, the theory being that digital
systems never make errors.

But the sinc3, being a lowpass, will recover from a temporarily
trashed input data stream.

Delta-sigma is just duty-cycle modulation with some noise tricks.

There are some outrageous d-s ADCs and DACs. I don\'t understand how
they can be so good; there must be hidden semiconductor tricks.

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

I have a Spice model of a d-s ADC and corresponding DAC. I can
probably hunt it down if anybody is interested. I have a *lot* of
Spice files!

 

[John Larkin]

On Sun, 09 Apr 2023 15:20:09 -0700, John Larkin
<jjlarkin@highlandtechnology.com> wrote:

On Sun, 9 Apr 2023 21:45:39 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

I found out what was wrong with that reply. The keyboard wasn\'t
plugged in.

 

[Jasen Betts]

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid> wrote:

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

You have an integrator at both ends, one for encoding one for decoding.
For fidelity, make sure they\'re the same. For noise impulse recovery
make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise
that gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear after
finite time.

--
Jasen.
🇺🇦 Слава Україні

 

[Liz Tuddenham]

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid> wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

You have an integrator at both ends, one for encoding one for decoding.
For fidelity, make sure they\'re the same. For noise impulse recovery
make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise
that gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear after
finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.


--
~ Liz Tuddenham ~
(Remove the \".invalid\"s and add \".co.uk\" to reply)
www.poppyrecords.co.uk

 

[Anthony William Sloman]

On Monday, April 10, 2023 at 6:47:19 AM UTC+10, Liz Tuddenham wrote:

I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset?

No.

> Will drift in the integrator of the receiver mean it gradually develops an offset and eventually crashes into one of the power rails?

No. The offset generates a correction signal that pulps it back

> Will signal loss cause a crash even more quickly?

The output will move to match the no-signal input.

> Does this mean the integrator has to have a low frequency limit, so that the long-term average output stays centred around the half-way level?

No. It\'s feedback system. The integrator just integrates the error signal, which feeds back.

> I\'ve not come across that in any of the literature.

The literature has a nasty habit of skipping the fundamentals. Somebody taught a high level course on sigma-delta A/D conversion at Cambridge, and it\'s graduates thought that anybody who didn\'t use it\'s misleading jargon didn\'t know anything about the subject.

Happily, that didn\'t stop me from using a sigma-delta A/D converter because it was all buried inside the chip I bought.

https://iopscience.iop.org/article/10.1088/0957-0233/7/11/015/meta

--
Bill Sloman, Sydney

 

[server]

On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman
<bill.sloman@ieee.org> wrote:

On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset?

No.

Will drift in the integrator of the receiver mean it gradually develops an offset and eventually crashes into one of the power rails?

No. The offset generates a correction signal that pulps it back

Will signal loss cause a crash even more quickly?

The output will move to match the no-signal input.

Does this mean the integrator has to have a low frequency limit, so that the long-term average output stays centred around the half-way level?

No. It\'s feedback system. The integrator just integrates the error signal, which feeds back.

So this is the reason why many digital audio systems haven a low
frequency limit at about 3 Hz(and not DC) ?

I\'ve not come across that in any of the literature.

The literature has a nasty habit of skipping the fundamentals. Somebody taught a high level course on sigma-delta A/D conversion at Cambridge, and it\'s graduates thought that anybody who didn\'t use it\'s misleading jargon didn\'t know anything about the subject.

Happily, that didn\'t stop me from using a sigma-delta A/D converter because it was all buried inside the chip I bought.

https://iopscience.iop.org/article/10.1088/0957-0233/7/11/015/meta

 

[John Larkin]

On Mon, 10 Apr 2023 11:23:58 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid> wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

You have an integrator at both ends, one for encoding one for decoding.
For fidelity, make sure they\'re the same. For noise impulse recovery
make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise
that gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear after
finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.

The encoder end, the ADC, usually has two integrators in the loop.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

Fig 2.

 

[John Larkin]

On Mon, 10 Apr 2023 15:38:09 +0300, upsidedown@downunder.com wrote:

On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman
bill.sloman@ieee.org> wrote:

On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset?

No.

Will drift in the integrator of the receiver mean it gradually develops an offset and eventually crashes into one of the power rails?

No. The offset generates a correction signal that pulps it back

Will signal loss cause a crash even more quickly?

The output will move to match the no-signal input.

Does this mean the integrator has to have a low frequency limit, so that the long-term average output stays centred around the half-way level?

No. It\'s feedback system. The integrator just integrates the error signal, which feeds back.


So this is the reason why many digital audio systems haven a low
frequency limit at about 3 Hz(and not DC) ?

D-S works fine down to DC. Audio systems are generally AC-coupled, one
reason being not to fry speakers. The other reason is to be cheap.

We use d-s adc\'s and dac\'s in precision instrumentation, DC accurate.

I wonder why audio CDs weren\'t d-s encoded. It would have saved
gigabucks.

 

[Anthony William Sloman]

On Monday, April 10, 2023 at 10:38:20 PM UTC+10, upsid...@downunder..com wrote:

On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman
bill....@ieee.org> wrote:
On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals..
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset?

No.

Will drift in the integrator of the receiver mean it gradually develops an offset and eventually crashes into one of the power rails?

No. The offset generates a correction signal that pulps it back

Will signal loss cause a crash even more quickly?

The output will move to match the no-signal input.

Does this mean the integrator has to have a low frequency limit, so that the long-term average output stays centred around the half-way level?

No. It\'s feedback system. The integrator just integrates the error signal, which feeds back.

So this is the reason why many digital audio systems haven a low frequency limit at about 3 Hz(and not DC) ?

Probably not. Manufactures don\'t specify or measure anything they don\'t have to, and nobody in the audio market is interested in what the system does at frequencies humans can\'t hear.

I\'ve not come across that in any of the literature.

The literature has a nasty habit of skipping the fundamentals. Somebody taught a high level course on sigma-delta A/D conversion at Cambridge, and it\'s graduates thought that anybody who didn\'t use it\'s misleading jargon didn\'t know anything about the subject.

Happily, that didn\'t stop me from using a sigma-delta A/D converter because it was all buried inside the chip I bought.

https://iopscience.iop.org/article/10.1088/0957-0233/7/11/015/meta

--
Bill Sloman, Sydney

 

[Anthony William Sloman]

On Tuesday, April 11, 2023 at 12:40:24 AM UTC+10, John Larkin wrote:

On Mon, 10 Apr 2023 15:38:09 +0300, upsid...@downunder.com wrote:
On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman <bill.....@ieee.org> wrote:
On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:

<snip>

> I wonder why audio CDs weren\'t d-s encoded. It would have saved gigabucks.

Pretty much all of them were. How come you didn\'t know that?

--
Bill Sloman, Sydney

 

[Lasse Langwadt Christensen]

mandag den 10. april 2023 kl. 17.14.38 UTC+2 skrev Anthony William Sloman:

On Tuesday, April 11, 2023 at 12:40:24 AM UTC+10, John Larkin wrote:
On Mon, 10 Apr 2023 15:38:09 +0300, upsid...@downunder.com wrote:
On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman <bill.....@ieee.org> wrote:
On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:
snip
I wonder why audio CDs weren\'t d-s encoded. It would have saved gigabucks.
Pretty much all of them were. How come you didn\'t know that?

only the special SACDs are D-S encoded, normal CDs are 16bit PCM, though a lot of CD players use a D-S DAC

 

[Liz Tuddenham]

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 11:23:58 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid> wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

You have an integrator at both ends, one for encoding one for decoding.
For fidelity, make sure they\'re the same. For noise impulse recovery
make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise
that gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear after
finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.

The encoder end, the ADC, usually has two integrators in the loop.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

Fig 2.

The confusion is because those aren\'t integrators, they are low-pass
filters. They happen to integrate at the pulse frequncy but not at
audio. A true integrator, with a capacitor around an op-amp so as to
integrate an input current, would not work.

As Bill Sloman has pointed out, somebody has taught it using the wrong
terminology and this has persisted despite the confusion it causes.


--
~ Liz Tuddenham ~
(Remove the \".invalid\"s and add \".co.uk\" to reply)
www.poppyrecords.co.uk

 

[John Larkin]

On Mon, 10 Apr 2023 17:39:41 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 11:23:58 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid> wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in the
integrator of the receiver mean it gradually develops an offset and
eventually crashes into one of the power rails? Will signal loss cause
a crash even more quickly?

Does this mean the integrator has to have a low frequency limit, so that
the long-term average output stays centred around the half-way level?
I\'ve not come across that in any of the literature.

You have an integrator at both ends, one for encoding one for decoding.
For fidelity, make sure they\'re the same. For noise impulse recovery
make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise
that gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear after
finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.

The encoder end, the ADC, usually has two integrators in the loop.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

Fig 2.

The confusion is because those aren\'t integrators, they are low-pass
filters.

Fig 2 has integrators.

They happen to integrate at the pulse frequncy but not at
audio. A true integrator, with a capacitor around an op-amp so as to
integrate an input current, would not work.

As Bill Sloman has pointed out, somebody has taught it using the wrong
terminology and this has persisted despite the confusion it causes.

The second-order d-s modulator, the ADC side, has two integrators. See
the wiki figure.

Of course the decoder/dac end must be an effective lowpass filter.

Maybe the confusion is about which end of the process people are
talking about, the ADC or the DAC.

 

[Joe Gwinn]

On Mon, 10 Apr 2023 07:40:11 -0700, John Larkin
<jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 15:38:09 +0300, upsidedown@downunder.com wrote:

On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman
bill.sloman@ieee.org> wrote:

On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:
I\'m trying to get my mind around Delta-Sigma encoding of audio signals.
The principle seems straightforward enough until you start to look at
the D.C. level of the regenerated signal. Obviously the output can\'t go
negative, so the half-way point must be equivalent to 0v output. This
means exactly equal alternating numbers of \'0\'s and \'1\'s are needed to
give 0v output.

What happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset?

No.

Will drift in the integrator of the receiver mean it gradually develops an offset and eventually crashes into one of the power rails?

No. The offset generates a correction signal that pulps it back

Will signal loss cause a crash even more quickly?

The output will move to match the no-signal input.

Does this mean the integrator has to have a low frequency limit, so that the long-term average output stays centred around the half-way level?

No. It\'s feedback system. The integrator just integrates the error signal, which feeds back.


So this is the reason why many digital audio systems haven a low
frequency limit at about 3 Hz(and not DC) ?

D-S works fine down to DC. Audio systems are generally AC-coupled, one
reason being not to fry speakers. The other reason is to be cheap.

We use d-s adc\'s and dac\'s in precision instrumentation, DC accurate.

I wonder why audio CDs weren\'t d-s encoded. It would have saved
gigabucks.

Probably because CDs were developed in the late 1970s, long before
digital ICs (or semiconductors in general) were remotely fast enough
and cheap enough for a consumer mass-market product.

..<https://en.wikipedia.org/wiki/Compact_disc>

Joe Gwinn

 

[Liz Tuddenham]

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 17:39:41 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 11:23:58 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid
wrote: > I\'m trying to get my mind around Delta-Sigma encoding of
audio signals. > The principle seems straightforward enough until
you start to look at > the D.C. level of the regenerated signal.
Obviously the output can\'t go > negative, so the half-way point must
be equivalent to 0v output. This > means exactly equal alternating
numbers of \'0\'s and \'1\'s are needed to > give 0v output. > > What
happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in
the > integrator of the receiver mean it gradually develops an
offset and > eventually crashes into one of the power rails? Will
signal loss cause > a crash even more quickly?
Does this mean the integrator has to have a low frequency limit,
so that the long-term average output stays centred around the
half-way level? I\'ve not come across that in any of the
literature.

You have an integrator at both ends, one for encoding one for
decoding. For fidelity, make sure they\'re the same. For noise
impulse recovery make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise that
gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear
after finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.

The encoder end, the ADC, usually has two integrators in the loop.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

Fig 2.

The confusion is because those aren\'t integrators, they are low-pass
filters.

Fig 2 has integrators.

Fig 1b shows a \'proper\' integrator but Fig 2 shows only symbols, they
both show the modulator but not the demodulator.

Fig 3(1) has the symbol for an integrator in the demodulator, where it
would not work.

They happen to integrate at the pulse frequncy but not at
audio. A true integrator, with a capacitor around an op-amp so as to
integrate an input current, would not work.

As Bill Sloman has pointed out, somebody has taught it using the wrong
terminology and this has persisted despite the confusion it causes.

The second-order d-s modulator, the ADC side, has two integrators. See
the wiki figure.

Of course the decoder/dac end must be an effective lowpass filter.

Maybe the confusion is about which end of the process people are
talking about, the ADC or the DAC.

If I understand it correctly, the same \'integrator\' (of combination of
\'integrators\') should be used at both ends to make the modulator and
demodulator complimentary.


--
~ Liz Tuddenham ~
(Remove the \".invalid\"s and add \".co.uk\" to reply)
www.poppyrecords.co.uk

 

[John Larkin]

On Mon, 10 Apr 2023 20:36:22 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 17:39:41 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

John Larkin <jjlarkin@highlandtechnology.com> wrote:

On Mon, 10 Apr 2023 11:23:58 +0100, liz@poppyrecords.invalid.invalid
(Liz Tuddenham) wrote:

Jasen Betts <usenet@revmaps.no-ip.org> wrote:

On 2023-04-09, Liz Tuddenham <liz@poppyrecords.invalid.invalid
wrote: > I\'m trying to get my mind around Delta-Sigma encoding of
audio signals. > The principle seems straightforward enough until
you start to look at > the D.C. level of the regenerated signal.
Obviously the output can\'t go > negative, so the half-way point must
be equivalent to 0v output. This > means exactly equal alternating
numbers of \'0\'s and \'1\'s are needed to > give 0v output. > > What
happens if a small burst of interference corrupts a few of the
pulses? Does it give the output a permanent offset? Will drift in
the > integrator of the receiver mean it gradually develops an
offset and > eventually crashes into one of the power rails? Will
signal loss cause > a crash even more quickly?
Does this mean the integrator has to have a low frequency limit,
so that the long-term average output stays centred around the
half-way level? I\'ve not come across that in any of the
literature.

You have an integrator at both ends, one for encoding one for
decoding. For fidelity, make sure they\'re the same. For noise
impulse recovery make sure they drift towards AC 0V

If you use an R-C lowpass as the integrator, then any noise that
gets into the stream will decay exponetially. if you use FIR like
boxcar or trapezoid as the intregrator the noise will dissapear
after finite time.

That\'s where I was confused, I was thinking of a Blumlein/Miller
integrator around an op-amp, not a low-pass filter.

The encoder end, the ADC, usually has two integrators in the loop.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

Fig 2.

The confusion is because those aren\'t integrators, they are low-pass
filters.

Fig 2 has integrators.

Fig 1b shows a \'proper\' integrator but Fig 2 shows only symbols, they
both show the modulator but not the demodulator.

Fig 3(1) has the symbol for an integrator in the demodulator, where it
would not work.


They happen to integrate at the pulse frequncy but not at
audio. A true integrator, with a capacitor around an op-amp so as to
integrate an input current, would not work.

As Bill Sloman has pointed out, somebody has taught it using the wrong
terminology and this has persisted despite the confusion it causes.

The second-order d-s modulator, the ADC side, has two integrators. See
the wiki figure.

Of course the decoder/dac end must be an effective lowpass filter.

Maybe the confusion is about which end of the process people are
talking about, the ADC or the DAC.

If I understand it correctly, the same \'integrator\' (of combination of
\'integrators\') should be used at both ends to make the modulator and
demodulator complimentary.

The demod can have integrators inside, but as a black box it is a
lowpass filter.

 

[Anthony William Sloman]

On Tuesday, April 11, 2023 at 12:40:24 AM UTC+10, John Larkin wrote:

On Mon, 10 Apr 2023 15:38:09 +0300, upsid...@downunder.com wrote:
On Mon, 10 Apr 2023 03:47:35 -0700 (PDT), Anthony William Sloman
bill....@ieee.org> wrote:
On Monday, April 10, 2023 at 6:47:19?AM UTC+10, Liz Tuddenham wrote:

<snip>

So this is the reason why many digital audio systems haven a low
frequency limit at about 3 Hz(and not DC) ?
D-S works fine down to DC. Audio systems are generally AC-coupled, one
reason being not to fry speakers. The other reason is to be cheap.

We use d-s adc\'s and dac\'s in precision instrumentation, DC accurate.

I wonder why audio CDs weren\'t d-s encoded. It would have saved gigabucks.

I apologise for responding twice to same post, but John Larkin has made two mistakes here.

The audio data that was digitised and ended up on Compact Disks (and other digital media was mostly digitised by rather expensive sigma-delta A/D converters in recording studios. There weren\'t enough of them for anybody to save gigabucks by using them.

Compact Disk were just a means of distributing digital audio after the audio signal had been digitised.

The consumer electronics that played back the compact disks were produced in high volume, and using sigma-delta D/A converters in them probaby did save a lot of money.

The book \"The Art of Digital Audio\" by John Watkinson - ISBN 0-240-512270-7 first published in 1988, is a full bottle on the subject, and on digitisation.

It is still in print. I cited it in comment published in Rev. Sci. Instrum. in 1999

https://aip.scitation.org/doi/abs/10.1063/1.1150138

The author I was commenting on was bit resentful for being picked on for not knowing the audio literature - he was working in interferometry.

--
Bill Sloman, Sydney

 

[Liz Tuddenham]

Anthony William Sloman <bill.sloman@ieee.org> wrote:

[...]

The book \"The Art of Digital Audio\" by John Watkinson - ISBN
0-240-512270-7 first published in 1988, is a full bottle on the subject,
and on digitisation.

It is still in print. I cited it in comment published in Rev. Sci.
Instrum. in 1999

https://aip.scitation.org/doi/abs/10.1063/1.1150138

It was originally published as a series of articles in Wireless World
(UK) and can be downloaded from:

https://worldradiohistory.com/Wireless_World_Magazine.htm

Start at January 1985.



--
~ Liz Tuddenham ~
(Remove the \".invalid\"s and add \".co.uk\" to reply)
www.poppyrecords.co.uk