You are correct that if you increase the data acquisition duration the integral noise power should be equal (assuming white noise) - it’s more about what it “looks” like when plotted - increasing the frequency resolution will tend to smooth out (average) the noise across the frequency range and make things “look” quieter even though they aren’t.
some of us object strongly to poorly worded use of "averaging" - claims it is "inherent" in the Fourier Transform
"averaging" is poor wording because conventional use implies information loss
the fft, Fourier Transforms are Integral Transforms - but in fact the complete set of Fourier Domain frequency numbers (full complex representation) is an exact Mathematical Dual of the Time Series data
full complex valued fft contains all of the information of the time series up to round off errors - the data sets can be converted back and forth with accuracy only limited by your computer arithmetic wordlength
increasing record length, number of samples does increase resolution with smaller frequency bins - and you get "sharper" peaks of stationary periodic "noise" or "spurs" with the higher resolution
a lot of noise in electronics is "stationary" ( /= periodic ) in the statistical sense too - so it does get spread over more smaller frequency bins with longer series and is then lower in level in each bin
to add more confusion "averaging" is a valuable technique that can be used with fft - but the non obvious part is that the Fourier Transform is Linear - so it doesn't matter if you average by adding element wise the Fourier or Time series numbers of the multiple data sets you are averaging
"averaging" is poor wording because conventional use implies information loss
the fft, Fourier Transforms are Integral Transforms - but in fact the complete set of Fourier Domain frequency numbers (full complex representation) is an exact Mathematical Dual of the Time Series data
full complex valued fft contains all of the information of the time series up to round off errors - the data sets can be converted back and forth with accuracy only limited by your computer arithmetic wordlength
increasing record length, number of samples does increase resolution with smaller frequency bins - and you get "sharper" peaks of stationary periodic "noise" or "spurs" with the higher resolution
a lot of noise in electronics is "stationary" ( /= periodic ) in the statistical sense too - so it does get spread over more smaller frequency bins with longer series and is then lower in level in each bin
to add more confusion "averaging" is a valuable technique that can be used with fft - but the non obvious part is that the Fourier Transform is Linear - so it doesn't matter if you average by adding element wise the Fourier or Time series numbers of the multiple data sets you are averaging
Last edited:
Hi jcx, I read your comment when was writing this one. Couldn’t agree more.
I would just like to add that when one increases the number of points, the signal is sampled for a longer time: in case of the noise, one will be observing it during a longer period of time – which is a good thing.
The longer one looks into noise, the best one can characterize it: for example if the noisy signal “happens to” increase for a short period of time (there’s always a non-zero probability…), and we sample only during that period of time we will overestimate noise. If we sample for a longer period of time, including when noise takes values “around” its rms value, the estimation will be better.
Similar considerations can be made about the noise frequency content being better estimated by looking into noise for a longer time (or by running several FFTs and averaging them, as I mentioned above) - this is what Max H. is mentioning in #5362.
I understand some may say there is an “averaging” effect here. But in reality there isn’t, in the usual sense of the word (see jcx’s comment above). In fact a longer observation time will provide more information about the noise behavior which, in turn, will allow estimating more accurately its rms level, the probability of it taking an arbitrarily large value, and its frequency distribution.
Regards
I would just like to add that when one increases the number of points, the signal is sampled for a longer time: in case of the noise, one will be observing it during a longer period of time – which is a good thing.
The longer one looks into noise, the best one can characterize it: for example if the noisy signal “happens to” increase for a short period of time (there’s always a non-zero probability…), and we sample only during that period of time we will overestimate noise. If we sample for a longer period of time, including when noise takes values “around” its rms value, the estimation will be better.
Similar considerations can be made about the noise frequency content being better estimated by looking into noise for a longer time (or by running several FFTs and averaging them, as I mentioned above) - this is what Max H. is mentioning in #5362.
I understand some may say there is an “averaging” effect here. But in reality there isn’t, in the usual sense of the word (see jcx’s comment above). In fact a longer observation time will provide more information about the noise behavior which, in turn, will allow estimating more accurately its rms level, the probability of it taking an arbitrarily large value, and its frequency distribution.
Regards
inexpensive measurements of amps
Bob, this might be worth a chapter in your new book. Your championing of LTspice has made life a lot simpler for Amp aficianados. (both for those who understand the sims as well as those who don't
)
Perhaps you could look at the use of 'free' software for amp measurements too.
I think the contenders are ARTA and RMAA .. with a whole lot of caveats which need to be highlighted.
NwAvGuy: RightMark Audio Analyzer
NwAvGuy: Testing Methods
NWAVGuy is long winded and obviously has a chip on his shoulder but he's competent and painstaking.
I think the two most important caveats are that
I think RMAA 'measurements' are done 'correctly' as long as you understand the DSP behind them and deal with the above caveats.
Wayne Kirkwood of Pro Audio Design Forum • Index page uses & rates ARTA highly for electronic stuff
I've only used it for acoustic stuff.
Bob, this might be worth a chapter in your new book. Your championing of LTspice has made life a lot simpler for Amp aficianados. (both for those who understand the sims as well as those who don't
)Perhaps you could look at the use of 'free' software for amp measurements too.
I think the contenders are ARTA and RMAA .. with a whole lot of caveats which need to be highlighted.
NwAvGuy: RightMark Audio Analyzer
NwAvGuy: Testing Methods
NWAVGuy is long winded and obviously has a chip on his shoulder but he's competent and painstaking.
I think the two most important caveats are that
- RMAA has no sensible level calibration
- the need for a 'professional' soundcard with balanced i/ps and balanced o/ps (which don't mind feeding unbalanced inputs). RMAA are aware of the problem and they detail specific grounding strategies for certain soundcards.
I think RMAA 'measurements' are done 'correctly' as long as you understand the DSP behind them and deal with the above caveats.
Wayne Kirkwood of Pro Audio Design Forum • Index page uses & rates ARTA highly for electronic stuff
I've only used it for acoustic stuff.
Last edited:
Measurements & Sims for amps
If you could post a clean *.ASC for one of your amps, another *.ASC "crippled" ...
and also if the 'real life' amp could be measured with ARTA, RMAA & TrueRTA ... bearing in mind the caveats I mention and maybe more from the rest of yus true gurus.
Anyone living close to ostripper who has experience with these software packages and can help?
Making sims reflect 'real life' has been one of my Holy Grails for nearly 2 decades in da previous Millenium .. mainly with my own linear Circuit Analyser.
This century, after 1.5 decades as a beach bum, I'm a LTspice newbie. But working with astx, I'm astounded & amazed at what's possible .. though I'm still far from claiming my sims reflect 'real life'.
ostripper, perhaps you could help Bob write a new chapter for his book.
If you could post a clean *.ASC for one of your amps, another *.ASC "crippled" ...
and also if the 'real life' amp could be measured with ARTA, RMAA & TrueRTA ... bearing in mind the caveats I mention and maybe more from the rest of yus true gurus.
Anyone living close to ostripper who has experience with these software packages and can help?
Making sims reflect 'real life' has been one of my Holy Grails for nearly 2 decades in da previous Millenium .. mainly with my own linear Circuit Analyser.
This century, after 1.5 decades as a beach bum, I'm a LTspice newbie. But working with astx, I'm astounded & amazed at what's possible .. though I'm still far from claiming my sims reflect 'real life'.
Last edited:
Actually SYN08 , Andy C. (and other former guru's) were "cripplin'" sims 5 years ago.
I just followed their examples.
I did run across "windowing" in dsp/fft literature. If that's what you meant , I can
see where we came across wrong. I thought you meant the OS itself.
Some members actually DO think there is a "voodoo" like difference between
operating systems (or even processors) in the resulting SQ.
I'm sure they are also "rolling op-amps" , silver wire , 100$ caps - and all
those other superstitious audiophile habits.
OS
I just followed their examples.
I did run across "windowing" in dsp/fft literature. If that's what you meant , I can
see where we came across wrong. I thought you meant the OS itself.
Some members actually DO think there is a "voodoo" like difference between
operating systems (or even processors) in the resulting SQ.
I'm sure they are also "rolling op-amps" , silver wire , 100$ caps - and all
those other superstitious audiophile habits.
OS
Please bear with this LTspice newbie.
I've been 'crippling' sims for a loo.oong time too.
But I gather you've done some which are very near to 'real life'.
If you could post the before & after *.ASC for one of your amps and if we can get some people to measure them, this would indeed be a great leap forward for us unwashed masses.
Hi Harry,
Thanks for your kind words and thoughtfulness.
First and foremost, I have received a great deal of very helpful feedback, mostly from the great folks on this forum. That feedback has stimulated some new ideas as well, in addition to some of what I have already planned for the second edition.
The second edition has gone more slowly than I had hoped. Even having now been retired for 1.5 years I am as busy as ever and don't know where the time goes. I honestly look back and almost don't know how I ever got the first edition done with that darned distracting day job
.I'm well along the way with several new chapters and upgrades to many of the existing chapters. My target is still 750 pages, so the book will be about 25% larger.
About this time last year I had set a target publication date for September 2015, but I know I am not going to make that. Maybe Spring of next year if I can get my butt into a higher gear. Unplanned distractions have come in during retirement, and I'm hoping those will settle down now. BTW, the joy of being a grandfather for the first time has been one of those "distractions".
I hope to see you at AES in NY or RMAF in Denver.
Cheers,
Bob
Well now, Bob, you have a new person to dedicate the book to, your grandchild. It would be nice to have designs that take into consideration real available device, things that can be bought today without having to find rare to umobtanium parts. No more John Curl types designs where they can only be made because he has stashed away parts that can't be found anymore. I wouldn't suggest that we have to move to surface mount devices but it would be nice to take those devices into consideration, even we in the diy community will have to move forward at some point. Congratulations on becoming a Grandfather.
It would be nice to have designs that take into consideration real available device, things that can be bought today without having to find rare to umobtanium parts. No more John Curl types designs where they can only be made because he has stashed away parts that can't be found anymore. I wouldn't suggest that we have to move to surface mount devices but it would be nice to take those devices into consideration, even we in the diy community will have to move forward at some point. Congratulations on becoming a Grandfather.Well now, Bob, you have a new person to dedicate the book to, your grandchild.
Your comments are right on target. I even invite suggestions for IPS, VAS and other devices that you and others feel would be good candidates in terms of both performance and availability.
I do share your concern regarding unobtainium devices, but note that the Linear Systems LSK389 dual, LSK170 and LSJ74 are available for making the dual complementary JFET input stages like those John Curl and others like. Of course, there is no dual monolithic p-channel available at this time, so some not-too-difficult matching of LSK74s is required.
Cheers,
Bob
It doesnt imply info loss. you can use peak, peak to peak, rms and average value etal without information loss. Averaging random noise many times will tend toward zero level. Nice for seeing data below the noise accurately. [But of course that isnt what you hear coming out of your system].
The true brilliance of FFT is that it is a linear process which can measure non-linearity. Up until then, a very hard thing to do with math alone.
It gets away with this by sampling and averaging the data between points... the shorter the time between data points the more accurate it will be towards the final number.
If the nonlinear data plot is broken down into short enough times intervals... the line/distance/time between points is almost a straight line piece (linear) without curvature.
Thats the jist of it... the short version re meaning of inherantly averaging process. Its also obvious from inspection of the calculus used. Integration gives an averaging affect.
THx-RNMarsh
Last edited:
I assume you are talking about IPS/VAS where you use low power parts. Surface mount is a different world. In my work, we use all surface mount since the 90s already. It is much harder to work with. In my contract work right now, we have to sent out to the board house to do the assembly. Rework is 10 times harder, just the soldering equipments is a lot more expensive. You pretty much need two Metcal type soldering iron to do rework. Just the soldering stations is over $500 to start. If you start using SOT-23 and 0603 components, you need a 40X microscope to rework a lot of times. A lot of people still doing dip and etch pcb here, you have zero chance of using surface mount.
If you go to TSSOP and 0402........forget it!!!!
I am not looking forward to my pcb from work to come back. Troubleshoot is going to be a nightmare. getting too old to see all the small parts. Electronics is a different world from the 80s.
Alan,
I would think we will be using through hole output devices for a long time, as long as someone will keep making them. But others are already using the largest surface mount parts and doing it successfully and using a typical solder station with small thin tip. Yes I have heard how it can take two of those irons many times to remove a bad or misplaced component, it is being done on this diy site. Believe me my eyes aren't what they use to be either, but a nice set of magnifying visors goes a long way to make up for that. You understand as you must deal with this every day, but you are probably using the smallest surface mount devices that you can use in your packaging.
I would think we will be using through hole output devices for a long time, as long as someone will keep making them. But others are already using the largest surface mount parts and doing it successfully and using a typical solder station with small thin tip. Yes I have heard how it can take two of those irons many times to remove a bad or misplaced component, it is being done on this diy site. Believe me my eyes aren't what they use to be either, but a nice set of magnifying visors goes a long way to make up for that. You understand as you must deal with this every day, but you are probably using the smallest surface mount devices that you can use in your packaging.
Ha ha!! I am stocking parts. So far, I have not encounter much problem like that. Only time I encounter was looking for a matching pair JFET.
I have enough BC550 and 560, KSA992 and KSC1845 for a long time. I don't think the power devices is going to be a problem, I don't think SM work well for high power devices anyway, they still going to be through hole. For matching purpose, I already have in hand over 50 ea of the MJL3281 and 1302. I have enough parts for many amps!!!
I really think quality of the final amp really depends on the design and layout, not that much on one particular devices. This is particular true that the dominant source of distortion is the OPS. Distortion of a sensible IPS/VAS design is going to be an order of magnitude lower than the best OPS anyway. I don't think I'll sweat too much on one particular part.
I have enough BC550 and 560, KSA992 and KSC1845 for a long time. I don't think the power devices is going to be a problem, I don't think SM work well for high power devices anyway, they still going to be through hole. For matching purpose, I already have in hand over 50 ea of the MJL3281 and 1302. I have enough parts for many amps!!!
I really think quality of the final amp really depends on the design and layout, not that much on one particular devices. This is particular true that the dominant source of distortion is the OPS. Distortion of a sensible IPS/VAS design is going to be an order of magnitude lower than the best OPS anyway. I don't think I'll sweat too much on one particular part.
Last edited:
You are pretty much right about the difficulties with SMT of smaller dimensions. In my day job, I worked with mostly 0603 and 0402 components; any rework was done by our experienced tech, mostly under a microscope with serious soldering stations. However, if you get up to 0805 or 1206, things are not so bad; but you still have some difficulty with the SMT transistors and ICs. And then there were the 1500-ball BGA ICs.....
Cheers,
Bob