Anyone here familiar with so-called 'zipper noise' in audioremote/automation level control applicati

[Eeyore]

Anyone here familiar with so-called 'zipper noise' in audio
remote/automation level control applications ?

I am very familiar with it. The usual quick and dirty fixes are:

[1] Use lots of bits / high resolution.

Would do if I had them. Only 8 bits of control on the TI PGA2310.

[2] Slew-rate limit the fader so that is someone slams
the pot from bottom to top it doesn't skip too many steps.
This decreases zipper noise at the cost of increasing lag.

Unacceptable for mixing.

[3] If it's a gain applied to a analog signal, Do the
actual gain switching at zero crossing. )Note: may conflict
with [2])

It does conflict, especially at low frequencies. And it still causes an
audible change in dV/dt.

[3] If the entire signal chain is digital, de-zipper the
gain change by interpolating a bunch of small steps in software
between the larger steps from the fader hardware.

Yes, would do but it isn't. It's an analogue signal chain in fact.

[4] Watch the gain structure so the system isn't using
the bottom 1% of the fader and following it with high gain
elsewhere.

Definitely not !

From the other replies you'll probably get the gist of the problem.

In the early days of automated consoles especially, IIRC, 256 discrete
steps would cause zipper noise. Filtering the VCA control input could
make it acceptable but a PGA2310 isn't that kind of chip. Not sure what
control law was being used but if you still only have 256 steps, surely
the same problem exists regardless of the law, it simply moves the
problem around ?

So, digital fails to deliver. I could do it in 24 bit DSP but it would be
plain silly for a volume control. 48kHz sampling would most certainly NOT
be acceptable either so the low cost chips would be out.

Graham

 

[Martin Griffith]

That's already an alternative suggestion I've made. Thanks Martin. Believe it or not all
this was to try and get 'low cost' matched level controls. Yes, they can do that but
it's for mixing and lag or zipper noise is not acceptable. The originator of the idea
hadn't fully thought it through it seems. And don't forget the predominance of bass in
much music makes the zero-crossing method useless on account of the timeout.

Graham

Thats why I junked my 2310 project, potential zipper noise on the
proto, and it was just for the 5.1 monitoring chain, it had some
really neat features, maybe I should look at the files, junk the 8051
for a mega666, put VCA's in......nah, no more dead idea creep

The That2162 is cheap, and dual, that will probably track a bit
better, 'spose you will have to brush up on your tempco compensation


martin

 

[Rich Grise]

That's the kind of thing. It's caused by a succession of step changes in
amplitude.

So, what was the quesion, or have you already got sufficient answers?

Thanks,
Rich

 

[Eeyore]

So, what was the quesion, or have you already got sufficient answers?

The answers so far, do peruse them, do indeed confirm my suspicions. I just
wanted to make sure I wasn't going mad etc.

The part in question is being inappropriately promoted for certain uses and a
youngster got caught out by it.

Thankfully we have a Plan B and C !

Graham

 

[Eeyore]

On a sunny day it happened Eeyore wrote

happens.

Well, let us know.
But things below 20Hz may be present all by themselves too.
It does not have a brick wall filter at 20 I think, so 10Hz, at less amplitude could
happen too.

In the long ago past (1983 or so) I used some of those 8 bit National DACs with signal on the reference input,
as attenuator, making a zero crossing detector is simple too, some opamps.
But 120dB is difficult.

Have fun.
Better do it all in software?

Well one could but it would be total overkill for this simple requirement.

I think the young lad got a bit carried away with the technology. Always think BASICs first. He seems quite smart
otherwise but I do still feel a little uncomfortable with anyone who codes PICS in a mix of C and assembler. At
least he seems to understasnd what a finite state machine is. I've yet to discover if he knows an interrupt from
polling though (I got a real shock the first time I met a respected programmer who didn't and wondered why his
code would hang !).

Graham

 

[[email protected]]

You're referring to power rail modulation classicly caused by insuffient
local supply decoupling I take it ?

Zipper noise is another effect, essentially a fast series of 'Fourier
clicks' typically caused by a moderately rapid sequence of step gain
changes in audio level in a 'fade-out' for example, so called because it
sounds like the zip on your jeans or whatever. When using a VCA gain
element it can be 'smoothed' somewhat by RC filtering of the control
voltage but is inherent in DSP and can only be removed there by making the
step size very small and therefore all but inaudible.

Graham

Ask Jon Slaughter to design a tube with a level attenuation grid in
it.

 

[Phil Allison]

"Eeysore"

That's already an alternative suggestion I've made. Thanks Martin. Believe
it or not all
this was to try and get 'low cost' matched level controls. Yes, they can
do that but
it's for mixing and lag or zipper noise is not acceptable. The originator
of the idea
hadn't fully thought it through it seems. And don't forget the
predominance of bass in
much music makes the zero-crossing method useless on account of the
timeout.

** More wild guessing and BULLSHIT !!


..... Phil

 

[Rich Grise]

The answers so far, do peruse them, do indeed confirm my suspicions. I
just wanted to make sure I wasn't going mad etc.

Oh, heavens! Everyone's known for some time that you're quite mad, as
am I. ;-)

Cheers!
Rich

 

[Rich Grise]

Why, when one could use a servomotor to turn a potentiometer?

I was going to suggest an analog VCA and a filter on the output of the
DAC.

Cheers!
Rich

 

[[email protected]]

If so there's a quick question I'd like to run by you.

Graham

I have a suggestion:

Decrease the PGA's stepsize by applying dithering to it's 8Bit
control.
As it's update rate could be 390 KHz you have lots of room to increase
the resolution.

A steps increase of 4 Bit would mean the lowest frequency of the
dithering signal is 390KHz/2^4.
This would roughly make the PGA a equivalent linear Multiplier with 8
Bits (instead of 4), so the 8 Bit Quantization noise -48Db (-24Db) is
roughly the maximum zipper noise level.
You can optimize this dithering so that it's amplitude is highest with
highest frequency (bit-reversed counter) and even use more dithering.
Mixing products below 20KHz are then low in amplitude and more
dithering bits still may give audible improvement.

With this gained resolution and high enough update-rate of the
undithered volume control signal you may completely get rid of audible
zipper noises at the expense of mixing products above the audible
range, easily filtered out to some extent and with tolerable residual.

And mind that the dithering gives linear gain change, but the PGA is
exponential.

 

[Martin Griffith]

On Mon, 27 Oct 2008 13:04:32 -0700 (PDT), in sci.electronics.design

I have a suggestion:

Decrease the PGA's stepsize by applying dithering to it's 8Bit
control.
As it's update rate could be 390 KHz you have lots of room to increase
the resolution.

A steps increase of 4 Bit would mean the lowest frequency of the
dithering signal is 390KHz/2^4.
This would roughly make the PGA a equivalent linear Multiplier with 8
Bits (instead of 4), so the 8 Bit Quantization noise -48Db (-24Db) is
roughly the maximum zipper noise level.
You can optimize this dithering so that it's amplitude is highest with
highest frequency (bit-reversed counter) and even use more dithering.
Mixing products below 20KHz are then low in amplitude and more
dithering bits still may give audible improvement.

With this gained resolution and high enough update-rate of the
undithered volume control signal you may completely get rid of audible
zipper noises at the expense of mixing products above the audible
range, easily filtered out to some extent and with tolerable residual.

And mind that the dithering gives linear gain change, but the PGA is
exponential.

OK , low cost product.............

martin

 

[Eeyore]

I have a suggestion:

Decrease the PGA's stepsize by applying dithering to it's 8Bit
control.
As it's update rate could be 390 KHz you have lots of room to increase
the resolution.

Funny, I had that very same idea only either yesterday or early this
morning.

A steps increase of 4 Bit would mean the lowest frequency of the
dithering signal is 390KHz/2^4.
This would roughly make the PGA a equivalent linear Multiplier with 8
Bits (instead of 4), so the 8 Bit Quantization noise -48Db (-24Db) is
roughly the maximum zipper noise level.

That's still a bit high. Noise shaping might help.

You can optimize this dithering so that it's amplitude is highest with
highest frequency (bit-reversed counter) and even use more dithering.
Mixing products below 20KHz are then low in amplitude and more
dithering bits still may give audible improvement.

With this gained resolution and high enough update-rate of the
undithered volume control signal you may completely get rid of audible
zipper noises at the expense of mixing products above the audible
range, easily filtered out to some extent and with tolerable residual.

And mind that the dithering gives linear gain change, but the PGA is
exponential.

Yes, another curiosity requiring some form of 'mapping' from one to the
other.

Thanks for your thoughts btw !

As well as the laternative Plan B and C, I have now evolved a Plan D. All
of which will will be much simpler and appeal better to the audio purist.
Looks like the youngster PGA promoter may have to settle for more
conventional level control.

Graham

 

[Eeyore]

OK , low cost product.............

This is anything but that. When the product literature goes live I can give
a link.

Graham

 

[Jasen Betts]

Put some bass through it and consider how many zero crossings (and how
much time) it'll take to ramp the level down inaudibly. Delay is
unacceptable.

I can only see a viable option being running quickly through all 256 0.5
dB steps. But I'm convinced this will 'zipper'.

you've proven that 256 steps are insifficient for mastering when the
control will be manipulated during recording.

You need more steps, maybe you can dither the control signal and filter he
dithering out. should be be good for 4 more bits, enough?

 

[JosephKK]

You're referring to power rail modulation classicly caused by insuffient
local supply decoupling I take it ?

Zipper noise is another effect, essentially a fast series of 'Fourier
clicks' typically caused by a moderately rapid sequence of step gain
changes in audio level in a 'fade-out' for example, so called because it
sounds like the zip on your jeans or whatever. When using a VCA gain
element it can be 'smoothed' somewhat by RC filtering of the control
voltage but is inherent in DSP and can only be removed there by making the
step size very small and therefore all but inaudible.

Graham

Nice description. It is pretty rare though, in my experience. I have
heard it only once, and yes it is very obnoxious even at low levels.
It just sticks out, far worse that an abruptly sore thumb.

 

[JosephKK]

Hello Mr Rabbit

That what you call 'zipper noise', is familiar to me.
You find it for example in my sound blaster (EMU10) based sound card, when moving the slider.
Now a few interesting points:
This sound card also does zipper noise when silence, so you hear indeed
a zipper like noise when the slider moves.
So the design of the attenuator sucks (Hello Creative Labs).

So, first you should make sure changing attenuator steps (in hardware or software)
does not cause a signal by itself.

Second, and this is something I have only tested in a *light* related application,
perhaps you should, or could, switch amplitude only on the zero crossings of the audio
signal.
This would mean worst case for 20 Hz every 50 milli seconds,
So if the slider moved a lot in those 50 milli seconds, then you would have to move
a bit more at once, but during the zero crossing.
This does actually prevent higher harmonics switching noise.
I patent this solution hereby, and just because you insulted me and other people
in your previous rabbit posting, you own me 1000 Internet Credits for every
device you sell that uses my solution.

I am a bit surprised though that an audiophool genius like you cannot fix a simple
issue like that all by himself.

Greetings from
Dark Fader

Well, shucks, Jan. You do some programming in highly interactive but
non real time OS's with high latency GUI. Surely you of all people
can understand that part of the problem is created by how the software
is interacting with hardware. Part of the repeatable issue you have
found is created by this interface. The GUI and the underlying OS for
both most linux distributions and MS OSs are not real time. So the
software accumulates many small steps and sends a 15 to 40 step quanta
to the volume control hardware repeatedly as the visual slider is
moved. The excessive concentration on the visual smoothness results
in zipper noise on the audio path because of the clumping of the audio
control update rate. Correcting the analog attenuator equivalent
portion if the EMU10 design can correct this. It may even be possible
to just issue a software update to do the step interpolation.

 

[JosephKK]

Exactly the problem. 50ms is too slow. And the chip has a 16ms Timeout in this mode.

"ZERO CROSSING DETECTION
The PGA2310 includes a zero crossing detection function
that can provide for noise-free level transitions. The
concept is to change gain settings on a zero crossing of the
input signal, thus minimizing audible glitches.

NB - minimising not eliminating.

This function is enabled or disabled using the ZCEN input
(pin 1). When ZCEN is low, zero crossing detection is
disabled. When ZCEN is high, zero crossing detection will
be enabled.
The zero crossing detection takes effect with a change in
gain setting for a corresponding channel. The new gain
setting will not be latched until either two zero crossings
are detected, or a timeout period of 16ms has elapsed
without detecting two zero crossings. ****In the case of a
timeout, the new gain setting takes effect with no attempt
to minimize audible artifacts.****"



A step volume change even at zero cross still results in an instantaneous change in dV/dt which
is still audible and therefore does not solve the basic problem. Also it's not a fade. Fades are
required.


Graham

Some of that is the result of when the transitions between the analog
and the digital portions are made and how they are made. I have
already discussed some of the GUI / OS related issues. Converters
have usually carefully managed charge injection issues, not all analog
switches do. Some times the analog switches used in IC converters and
attenuators are not controlled for this property which can contribute
to zipper noise.

 

[JosephKK]

OK , low cost product.............

martin

Martin, dithering is used at and below the lowest single bit step,
about 0.5 db. Plus the noise shaping by hugely oversampling could
give you effectively 12 bits or <0.1 dB with the zipper noise shifted
out of band. Takes rather heavy control step to control step calc's
though. Optimally you want about 3 successive values in order to get
the first derivative and a hint on the second derivative to adjust
Simpson's rule coefficients to where the hugely oversampled 8 bit dac
is applied.

 

[Eeyore]

But this one isn't. It's high end pro-audio.

Martin, dithering is used at and below the lowest single bit step,
about 0.5 db. Plus the noise shaping by hugely oversampling could
give you effectively 12 bits or <0.1 dB with the zipper noise shifted
out of band. Takes rather heavy control step to control step calc's
though. Optimally you want about 3 successive values in order to get
the first derivative and a hint on the second derivative to adjust
Simpson's rule coefficients to where the hugely oversampled 8 bit dac
is applied.

One problem I see is the steps are all equally ratiometric i.e. 0.5dB.

You could whizz the gains about a bit from the desired value at that instant
in time but it would simply add noise as I see it.

Graham

 

[JosephKK]

Oddly, fingers squeaking on wound guitar strings have that
same effect on me, bt many folks don't seem to mind them.

I am not sure that i have ever heard that one. Could you send me a
clip and point out where is occurs? I cannot judge whether i like
something if i cannot identify it.