Sign up Calendar Latest Topics Donate
 
 
 


Reply
  Author   Comment   Page 2 of 2      Prev   1   2
OLi

Sr. Member
Registered:
Posts: 1,918
Reply with quote  #16 
My view of this is not the same as yours as previously been described in detail in other topics.
1. You need in my book always obey Nyquist sampling theorem, 2.56 rule is a rule of thumb to cover some real world slope and decay of analog antialiasing filters. Using sigma delta converter can be a way to get closer t Nyquist 2x but not infinite close depending on filter characteristics.
2. In TWF you do have aliasing, Nyquist do not connect specific to FFT in my book, if you get aliasing if you have no antialising filteering you get aliasing and if you folded down signal are large enough you get the wrong measurement.
3. FFT square being "fake data not existing i IRL"  is fake truth, as long as you can tune a analog bandpass filter IRL and find those "not existing" signals live and kicking it is wrong to call them not existing in my book and I will not retract on that, we built handheld portable analog tunable filter vibrometer battery driven and digital tracking filters vector voltmeters since 1972 so I have a few those in my musem shelf and will happily show that on a square wave from a signal generator, any time. I can even measure the vector, amp and phase on those if you like.
4. Well in the box we still make (384 deliverd) with balancing, FFT and a couple of bearing cond incl demod we measure up 12kHz and obey Nyquist and rule of thumb and sample like 30.7 kHz.
As you say, length is a property of FFT resolution so in this we 3200 lines and TWF length to do that.
We do some quick stepping in that box as the ARM cpu was right on the limit to do the work so we do sample a block to do the parameter calcs like ISO, TOT and brg conditions. I really don't remember exactly the spec that block is sampled to but the brg params are to 12kHz and the unfiltered TOT param is 1Hz to 12kHz..... If 12kHz is required? It depend on how long before you need to know bad bearings, it may be 5kHz is ok, we do think about that. If you only measure out to 1kHz you get very short warning time or close to nothing but when it fails you can see it was due to bad bearings a few days before maybe, so it may warn and shut down just in time, but it will be bad, real bad, normally...
It is now nearly 20 years old design now and it is a hybrid, some filters are analog, some calculated and FFT actually a integer but you don't see that due to some fancy stepping.....
We now do it as a Android app and get the freedom of infinite power and memory so we can do it properly.... We still only go up to 20kHz and sample about 51kHz in both cases we use sigma delta A/D so we need not to bother about antialiasing filtering as such as it is built in to the A/D and very sharp and do behave as described in the datasheet, contrary to other suppliers....
So that is my view and it is hard earned and y need rely give hard data to make me move of these corner stones of my experience both in theory and IRL.
You know even soundchip A/D do care for antialising as the sound would otherwise sound like Cr***p they say. I really can't say that I listened to folded down music, it may be a special feature. It sure isn't Beach Boys.
So that is my view, take it as you like I stick to it as it is hard earned.


__________________
Good Vibrations since early 1950's, first patented vibrometer 1956 in the US.
http://www.vtab.se
BaartCM

Member
Registered:
Posts: 15
Reply with quote  #17 
I say tomato ...
OLi

Sr. Member
Registered:
Posts: 1,918
Reply with quote  #18 
I say you need to measure correct or the data you see will be wrong and that may make your analysis seriously more difficult, believe me....
So this is not to be fancy pancy, it is in your best interest seriously regardless if you care or not or I think, specific if you don't care, we care for ya and you seriously should too even if it makes the world difficult. Pls. read up on the Nyquist theorem as that is sure no tomato so y understand the reasons why it is so crucial IMHO or leave things like this to those who care.

__________________
Good Vibrations since early 1950's, first patented vibrometer 1956 in the US.
http://www.vtab.se
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #19 
I'll attempt to distill what Oli is saying:

the nyquist theorem is not only applicable to FFT. It is applicable to ALL quantities, scalar or not, that are calculated from a digital time waveform. If you use the TWF to calculate an overall value, the nyquist limit still applies.

It gets a little trickier when you forgo antialiasing filters. If your frequency components above the nyquist filter get folded into your signal, they would still be present, and therefore the overall value should still show the same regardless of the sampling rate right?

Yes and no, because there is another limit. Your ADC converter takes samples at a specific point in time, but it is not infinitessimally small. It lasts some micro, nano, pico seconds. Over that time period, the signal is averaged, and this effectively acts as a low-pass filter. Not to mention all of the random mechanical and electrical filters that exist in your measurement chain that will always exist.

But if you want to measure the overall value of a 2kHz signal by measuring at 1kS/s and can bypass the antialiasing filter, then yes, you will get about the same overall value. However, every other statistical quantity for complex signals, including pk-pk overalls, kurtosis, etc. will all be wrong.
BaartCM

Member
Registered:
Posts: 15
Reply with quote  #20 
@OLi I have already stated "I appreciate that selection of the correct sampling rate is CRUCIAL to getting meaningful information from your data. Hence my question what sample rate and data length do you suggest”

To sum up "rubbish in = rubbish out" I learned that 30 years ago.

I DO care about getting the correct data so I get meaningful information so please stop suggesting I don't care, I wouldn't have bothered asking the damned question if I didn't care! Please, no more diatribes or self promoting posts! Your units may be superb but that doesn't give me any indication of sampling rate and data length for my humble project. At the end of the day, that's what this post was for, to determine correct measurement procedure.

btw, I see you seem chuffed that they are growing wine in Sweden, I think you'll find they are called grapes!
MarkL

Avatar / Picture

Sr. Member
Registered:
Posts: 1,121
Reply with quote  #21 
I think the subjectiveness is getting lost in personal opinions and umbrage/offence taken. (Always taken never given)

I'll distil what I have read so far.
You want to use a theoretical 'Newish' well new in the sense of it's not commonly used why I wonder, (not throwing 'shade' just curious why it's not implemented by someone yet, except its existence outside of theoretical environments and considering the paper has been around a few years,) I digress.

OP is advised of the pitfalls, as we mostly do on here with new members who we have no prior Idea of capabilities experience etc.

So much jousting begins and tomato chucking...but we are still none the wiser, just some with bloodied noses..or maybe tomato puree /grapes of wrath stains.

Its interesting to see the intellectual 'mickey measuring' going on but let's not get away from the core of the discussion. Can it/will it work? 








Danny Harvey

Sr. Member / Moderator / Supporter
Registered:
Posts: 1,403
Reply with quote  #22 
I absolutely cannot add anything to this discussion. Or even follow it for that matter.

But I would like to remind all of us that we are here to discuss and learn.  Please keep it professional, congenial and never personal. 
OLi

Sr. Member
Registered:
Posts: 1,918
Reply with quote  #23 
I have many times above said that I use up to 12kHz and sample apprx 30 kHz and that is the compromise I currently are comfortable with since 20 years but may consider to test lower like 5 kHz but have not done so yet and can not confirm any results from that.
So what more can I do? I have written what I use several times.
Yes I do promote my things but also to point out that I currently have like 384 pcs delivered world wide including uni of Texas, Houston, uni Sagreb, uni Sthlm KTHx2 and I know some are in use since they (or the users) send me data and products are still sold (2pcs this week) and I think that indicates it is in some way working.
Vibrometers in China are mainly sold on a working bearing condition measurement parameter but that is the pure analog version used since 2004 in the "latest" pcb revision date.
I only object to the Nyquist problem as I am a bit edgy on that as it cost me 2 man year development time as also big component suppliers don't have a clue so excuse me for upholding the required specs for us to get correct data. It may make things complicated but is the only way to do things IMHO. 
We Viking "descendants" do take an interest in the wine, not so much in grapes y know...
My worst fear is that newbies start using flawed hardware in good faith, not having a clue, and use them the wrong way and flooding the market with cheap none functional devices, then we do have serious problems. I will have retired by then but you boys and girls still working will have problems when data get crazy.

__________________
Good Vibrations since early 1950's, first patented vibrometer 1956 in the US.
http://www.vtab.se
OLi

Sr. Member
Registered:
Posts: 1,918
Reply with quote  #24 
I have no idea, likely to complicated for me I like KISS it is normally bad enough.... But Primary I do request developer to do read up on what Nyquist really is about, it is not a theoretical wizardry it is a technical necessity IMHO and I am not guessing here I learned it the hard way.
__________________
Good Vibrations since early 1950's, first patented vibrometer 1956 in the US.
http://www.vtab.se
Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.