FPGA Central

Vladimir Vassilevsky wrote:
>
>
> Scott Seidman wrote:
>
>
>> Most likely the latter. I can remember spending considerable time in
>> class figuring out just how must delay a feedback system can tolerate
>> before going unstable, but "almost" all of them could tolerate "some",
>> if I recall correctly, based upon just when the phase would kick a
>> nice negative feedback into the much less pleasant positive variety.
>> Of course, delay generally hurts performance, and often simple
>> stability is much less than what one needs.
>
> The feedback loop can't provide any correction at the frequencies higher
> then 1/(6*Tdelay).
> The conservative rule for the stability would be having a unity loop
> gain at 1/(8*Tdelay).
>
> (This assumes that the delay is the dominating source of the phase shift
> in the loop, the response is monotonic and the bandwidth of the loop is
> from 0 to Fmax).

...

Simply put, a one-sample delay amounts to 180 degreed phase shift at Fs.
That is the primary reason for needing high oversampling rates in
feedback systems. In turn, the oversampling makes it easier to skimp on
or dispense with an anti-alias filter.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Jerry Avins wrote:


>> The feedback loop can't provide any correction at the frequencies
>> higher then 1/(6*Tdelay).
>> The conservative rule for the stability would be having a unity loop
>> gain at 1/(8*Tdelay).
>>
>> (This assumes that the delay is the dominating source of the phase
>> shift in the loop, the response is monotonic and the bandwidth of the
>> loop is from 0 to Fmax).
>
> Simply put, a one-sample delay amounts to 180 degreed phase shift at Fs.

Even worse: one sample delay is 360 degrees at Fs.


> That is the primary reason for needing high oversampling rates in
> feedback systems. In turn, the oversampling makes it easier to skimp on
> or dispense with an anti-alias filter.

Unfortunately, it is not very easy. The typical oversampling delta-sigma
ADC creates a delay as long as 10...20 periods of the output sample
rate. This is due to the digital filtering inside. The ADCs with the
true direct sampling at the high speeds are expensive indeed.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Vladimir Vassilevsky wrote:
>
>
> Jerry Avins wrote:
>
>
>>> The feedback loop can't provide any correction at the frequencies
>>> higher then 1/(6*Tdelay).
>>> The conservative rule for the stability would be having a unity loop
>>> gain at 1/(8*Tdelay).
>>>
>>> (This assumes that the delay is the dominating source of the phase
>>> shift in the loop, the response is monotonic and the bandwidth of the
>>> loop is from 0 to Fmax).
>>
>> Simply put, a one-sample delay amounts to 180 degreed phase shift at Fs.
>
> Even worse: one sample delay is 360 degrees at Fs.
>
>
>> That is the primary reason for needing high oversampling rates in
>> feedback systems. In turn, the oversampling makes it easier to skimp
>> on or dispense with an anti-alias filter.
>
> Unfortunately, it is not very easy. The typical oversampling delta-sigma
> ADC creates a delay as long as 10...20 periods of the output sample
> rate. This is due to the digital filtering inside. The ADCs with the
> true direct sampling at the high speeds are expensive indeed.

Would you even consider a delta-sigma converter in a control system?
"Fast" for most mechanisms is well within reach of fairly inexpensive
successive-approximation converters. A .22 bullet goes only about an
inch in a tenth of a millisecond. Few machines need samplers that
greatly exceed audio rates.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Raymond Toy wrote:

...

> I think I understand what you're saying. Basically, you have some
> kind of a priori knowledge that the signal is bandlimited and you're
> sampling at a high enough frequency that aliasing isn't there. No
> problem with that.
>
> But that does bring up the question of how you got the a priori
> knowledge.
>
> But unless you do the experiment, it's kind of hard to tell if
> anti-aliasing filter effects are worse than aliasing effects. Maybe
> those EKG strips are mostly aliasing artifacts. :-)

In a servo system, inertia and available motive power set limits on
acceleration, and displacement is inversely proportional to acceleration
squared. That's an inherent 40 dB/decade rolloff. Other types of systems
have similar constraints.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Jerry Avins <[email protected]> wrote in
news:[email protected]:

> Vladimir Vassilevsky wrote:
>>
>>
>> Jerry Avins wrote:
>>
>>
>>>> The feedback loop can't provide any correction at the frequencies
>>>> higher then 1/(6*Tdelay).
>>>> The conservative rule for the stability would be having a unity
>>>> loop gain at 1/(8*Tdelay).
>>>>
>>>> (This assumes that the delay is the dominating source of the phase
>>>> shift in the loop, the response is monotonic and the bandwidth of
>>>> the loop is from 0 to Fmax).
>>>
>>> Simply put, a one-sample delay amounts to 180 degreed phase shift at
>>> Fs.
>>
>> Even worse: one sample delay is 360 degrees at Fs.
>>
>>
>>> That is the primary reason for needing high oversampling rates in
>>> feedback systems. In turn, the oversampling makes it easier to skimp
>>> on or dispense with an anti-alias filter.
>>
>> Unfortunately, it is not very easy. The typical oversampling
>> delta-sigma
>> ADC creates a delay as long as 10...20 periods of the output sample
>> rate. This is due to the digital filtering inside. The ADCs with the
>> true direct sampling at the high speeds are expensive indeed.
>
> Would you even consider a delta-sigma converter in a control system?
> "Fast" for most mechanisms is well within reach of fairly inexpensive
> successive-approximation converters. A .22 bullet goes only about an
> inch in a tenth of a millisecond. Few machines need samplers that
> greatly exceed audio rates.
>
> Jerry

Certainly sigma delta converters have considerable delay but I think they
can sometimes still be a good choice over SAR in control loops. The trick
is to over sample by a lot. For example, if you sample at 200k with a 20
sample delay, this would be the same delay as a SAR at 10k. You can then
use IIR antialiasing if needed. This would have about the same effect as
an analog filter of the same order for the SAR case. The advantage of the
sigma delta case is that you might not need the extra hardware. Today's
signal processors often have more than enough computation power to
perform the extra filtering.

--
Al Clark
Danville Signal Processing, Inc.
--------------------------------------------------------------------
Purveyors of Fine DSP Hardware and other Cool Stuff
Available at http://www.danvillesignal.com

Jerry Avins wrote:


>>> That is the primary reason for needing high oversampling rates in
>>> feedback systems. In turn, the oversampling makes it easier to skimp
>>> on or dispense with an anti-alias filter.
>>
>>
>> Unfortunately, it is not very easy. The typical oversampling
>> delta-sigma ADC creates a delay as long as 10...20 periods of the
>> output sample rate. This is due to the digital filtering inside. The
>> ADCs with the true direct sampling at the high speeds are expensive
>> indeed.
>
>
> Would you even consider a delta-sigma converter in a control system?

Why not. It will work for the feedforward adaptive system. Not for the
straightforward loop implementation, of course.

> "Fast" for most mechanisms is well within reach of fairly inexpensive
> successive-approximation converters. A .22 bullet goes only about an
> inch in a tenth of a millisecond. Few machines need samplers that
> greatly exceed audio rates.

And those rates are quite fast already. To provide for the sufficient
loop gain and stability, the sample rate has to be at least at the order
higher then the highest frq. of interest.
Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Vladimir Vassilevsky wrote:
>
>
> Jerry Avins wrote:

...

>> Would you even consider a delta-sigma converter in a control system?
>
> Why not. It will work for the feedforward adaptive system. Not for the
> straightforward loop implementation, of course.

I mostly do one-off things, and I like to keep them simple. You're
better at this than I am, and you can cur more corners (in my terms).

>> "Fast" for most mechanisms is well within reach of fairly inexpensive
>> successive-approximation converters. A .22 bullet goes only about an
>> inch in a tenth of a millisecond. Few machines need samplers that
>> greatly exceed audio rates.
>
> And those rates are quite fast already. To provide for the sufficient
> loop gain and stability, the sample rate has to be at least at the order
> higher then the highest frq. of interest.
> Vladimir Vassilevsky

Sure, but I've yet to have need to control something moving at bullet
speeds.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

>I'm trying to think of examples where using an anti-alias filter is a
>bad idea, or must at least be approached with extreme caution. I
>already know about control systems and video applications, and I believ

>that this is a big issue with EKG machines.
>
>Does anyone else have any examples that they can share?
>
>If someone has a sampled EKG output in some readable file format that
>they'd be willing to share I would be grateful -- I can whomp something
>up out of nothing, but then some doctor will read my article and know
>it's BS.
>
>--
>
>Tim Wescott
>Wescott Design Services
>http://www.wescottdesign.com
>
>Posting from Google? See http://cfaj.freeshell.org/google/
>
>"Applied Control Theory for Embedded Systems" came out in April.
>See details at http://www.wescottdesign.com/actfes/actfes.html
>







I’ve read this and re-read this and to be honest I’m having a difficul
time determining if you’re serious or not.

My background is instrumentation.

Personally I think the question was phrased wrong. It’s not a matter of
… where using an anti-alias filter is a bad idea …”, it’s more of how do
set the specifications of the anti-alias filter? The answer is that i
will be system dependent.

The fact is that the A-A filter is nothing more than an element in th
overall gain line up. Tradeoff is the name of the game. And it’s
question of “what do I want to accomplish” with this particular element.
How much “badness” am I willing tolerate as a result of relaxing th
rejection specs in order to meet some other requirement of the filter
system.

Take any generic acquisition system, it doesn’t matter if it’s an R
downconverter running from dc to daylight with an extremely wide baseban
bandwidth and a 100dB of dynamic range or a low bandwidth op-amp / A/
lineup in a navigational control system, the system architect needs t
have a clear understanding of his or her dynamic range requirements i
order to specify the attributes of the elements in the signal chain.

It’s that simple.

From a instrumentation stand point, the most problematic effect from an
filtering is flatness in both amplitude and phase. Sever group delay fro
narrow band filters is also a big issue. These issues are more easil
dealt with these days thanks in part to the ability to do corrections i
real time. In the “old days” all the bucks were sunk into buildin
incredibly expensive filters to achieve system performance.

Then on the other side of the spectrum you have low bandwidth contro
loops where simple RC filtering does the job because 30 db of dynami
range is all that is required.

It’s all about dynamic range and system requirements.


jd

Al Clark wrote:
> Jerry Avins <[email protected]> wrote in
> news:[email protected]:
>
>
>>Vladimir Vassilevsky wrote:
>>
>>>
>>>Jerry Avins wrote:
>>>
>>>
>>>
>>>>>The feedback loop can't provide any correction at the frequencies
>>>>>higher then 1/(6*Tdelay).
>>>>>The conservative rule for the stability would be having a unity
>>>>>loop gain at 1/(8*Tdelay).
>>>>>
>>>>>(This assumes that the delay is the dominating source of the phase
>>>>>shift in the loop, the response is monotonic and the bandwidth of
>>>>>the loop is from 0 to Fmax).
>>>>
>>>>Simply put, a one-sample delay amounts to 180 degreed phase shift at
>>>>Fs.
>>>
>>>Even worse: one sample delay is 360 degrees at Fs.
>>>
>>>
>>>
>>>>That is the primary reason for needing high oversampling rates in
>>>>feedback systems. In turn, the oversampling makes it easier to skimp
>>>>on or dispense with an anti-alias filter.
>>>
>>>Unfortunately, it is not very easy. The typical oversampling
>>>delta-sigma
>>> ADC creates a delay as long as 10...20 periods of the output sample
>>>rate. This is due to the digital filtering inside. The ADCs with the
>>>true direct sampling at the high speeds are expensive indeed.
>>
>>Would you even consider a delta-sigma converter in a control system?
>>"Fast" for most mechanisms is well within reach of fairly inexpensive
>>successive-approximation converters. A .22 bullet goes only about an
>>inch in a tenth of a millisecond. Few machines need samplers that
>>greatly exceed audio rates.
>>
>>Jerry
>
>
> Certainly sigma delta converters have considerable delay but I think they
> can sometimes still be a good choice over SAR in control loops. The trick
> is to over sample by a lot. For example, if you sample at 200k with a 20
> sample delay, this would be the same delay as a SAR at 10k. You can then
> use IIR antialiasing if needed. This would have about the same effect as
> an analog filter of the same order for the SAR case. The advantage of the
> sigma delta case is that you might not need the extra hardware. Today's
> signal processors often have more than enough computation power to
> perform the extra filtering.
>
You can always choose the sigma-delta which don't have a big delay.
There are quite a few that settle completely in 3 sample times, and some
that do it in 2. These are usually targetted as single shot conversion,
with a mux in front of the converter to allow sampling from several
channels. They have merit for control systems as well.

Steve

Vladimir Vassilevsky wrote:

>
>
> Jerry Avins wrote:
>
>
>>> The feedback loop can't provide any correction at the frequencies
>>> higher then 1/(6*Tdelay).
>>> The conservative rule for the stability would be having a unity loop
>>> gain at 1/(8*Tdelay).
>>>
>>> (This assumes that the delay is the dominating source of the phase
>>> shift in the loop, the response is monotonic and the bandwidth of the
>>> loop is from 0 to Fmax).
>>
>>
>> Simply put, a one-sample delay amounts to 180 degreed phase shift at Fs.
>
>
> Even worse: one sample delay is 360 degrees at Fs.
>
A pure delay of 1/Fs would contribute 360 degrees at Fs. But the actual
delay from sampling is somewhere between 0 and 1/Fs -- in fact, it
averages out to 180 degrees.

I figured that the real answer to this would be "the effect depends on
the plant", so I did some studies (using SciLab) to convert various
realistic (but quick) plant models into the sampled-time domain. It
appears that as long as they are effectively integrating at least once
at Fs/2 the sampling process contributes exactly 90 degrees of phase
shift -- which is consistent with Jerry's approximation.
>
>> That is the primary reason for needing high oversampling rates in
>> feedback systems. In turn, the oversampling makes it easier to skimp
>> on or dispense with an anti-alias filter.
>
>
> Unfortunately, it is not very easy. The typical oversampling delta-sigma
> ADC creates a delay as long as 10...20 periods of the output sample
> rate. This is due to the digital filtering inside. The ADCs with the
> true direct sampling at the high speeds are expensive indeed.
>
I think you're confusing "oversampling converter" with "converter that's
oversampling". In a control loop you're wise to sample way higher than
your intended bandwidth (between 4x and 400x, although the center is
10-20), but you'll often see SAR converters used here. Sigma-deltas, if
used, are used with care.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

jdc wrote:

...

> I’ve read this and re-read this and to be honest I’m having a difficult
> time determining if you’re serious or not.
>
> My background is instrumentation.
>
> Personally I think the question was phrased wrong. It’s not a matter of “
> … where using an anti-alias filter is a bad idea …”, it’s more of how do I
> set the specifications of the anti-alias filter? The answer is that it
> will be system dependent.

...

I think that Tim is asking for applications in which the aliasing that
is suffered without an A-A filter imposes a cost that is less than the
delay that it imposes. Since most of my experience is with controlled
systems, all the examples I can think of apply to them, and that's
already been covered.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Rune Allnor wrote:

(snip)

> I once attended a conference on DSP where somebody presented
> some scheme for processing EKG (that's the heart monitor, right?)
> signals.

> According to that presentation, ringing and transients are severe
> issues since physicians inspect the transient characteristic of the
> heartbeat in order to come up with a diagnostics. Any transient
> that is not "natural" (i.e. does not conform with text-book
> descriptions of normal heartbeat behaviour) is, by necessity,
> interpreted as an indicator of some non-healthy condition.
> It would be very, very dangerous to meddle with these
> systems such that the sensor system somehow alters
> the transient shape of the signal.

Weren't these originally done with an electro-mechanical system,
something like an analog voltmeter with a pen on the end drawing on
a moving strip of paper? There would then be some filter characteristic
determined by the mechanical system. I would also expect some
filtering to be done to the signal as it passed through the body
from the heart to the skin.

I would say, then, instead of no anti-aliasing filter that it has
a natural anti-aliasing filter. Other systems also have a natural
filter, such as scanning beam video camera tubes where the finite sized
beam spot acts as a filter, in addition to the optical system.

-- glen

glen herrmannsfeldt wrote:
> Rune Allnor wrote:
>
> (snip)
>
>> I once attended a conference on DSP where somebody presented
>> some scheme for processing EKG (that's the heart monitor, right?)
>> signals.
>
>> According to that presentation, ringing and transients are severe
>> issues since physicians inspect the transient characteristic of the
>> heartbeat in order to come up with a diagnostics. Any transient
>> that is not "natural" (i.e. does not conform with text-book
>> descriptions of normal heartbeat behaviour) is, by necessity,
>> interpreted as an indicator of some non-healthy condition.
>> It would be very, very dangerous to meddle with these
>> systems such that the sensor system somehow alters
>> the transient shape of the signal.
>
> Weren't these originally done with an electro-mechanical system,
> something like an analog voltmeter with a pen on the end drawing on
> a moving strip of paper? There would then be some filter characteristic
> determined by the mechanical system. I would also expect some
> filtering to be done to the signal as it passed through the body
> from the heart to the skin.

String galvanometer. My father-in-law had a portable one that he
sometimes took on house calls. I was in high-school when it was replaces
by a Gould pen recorder, and he lent it to me.

> I would say, then, instead of no anti-aliasing filter that it has
> a natural anti-aliasing filter. Other systems also have a natural
> filter, such as scanning beam video camera tubes where the finite sized
> beam spot acts as a filter, in addition to the optical system.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Jerry Avins skrev:
> My father-in-law ...
> I was in high-school when ... he lent it to me.

You married early, then?

Rune

Rune Allnor wrote:
> Jerry Avins skrev:
>> My father-in-law ...
>> I was in high-school when ... he lent it to me.
>
> You married early, then?

No. I married a long-time friend.

Jerry
--
"The rights of the best of men are secured only as the
rights of the vilest and most abhorrent are protected."
- Chief Justice Charles Evans Hughes, 1927
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯