REW Update

[abbey road d enfer]

So what does this 'rising hf noise'  amount to ?

It just shows that the analyzed signal is closer to white noise than pink.

Is it fair to say the '1/48 RTA'  apps that I and others have observed  (TrueRTA, REW) are doing things not quite optimally?

I think it's not fair to implicate the tool when the issue is likely to be an operator's mistake. My understanding is that someone tried to evaluate a frequency response  and noticed a difference between Spectrum analysis (in fact a derivative of FFT) and RTA. Both are usable but they must be used in conjunction with the correct stimuli: white noise for spectrum, pink noise for RTA.

Again, I've not compared with the pro types  ..  just the best of the low or no dollar PC based apps.

The big issue with those is they don't come with the necessary metrologic knowledge.

Maybe we need a filter or some kind of 'pre-emphasis' to 'flatten it out' 

No, one just needs to use the proper settings according to the type of measurement.
BTW the difference between white and pink noise is just a 3dB/octave low pass filter.

 

[alexc]

In the meanwhile, I want to rattle off some '6922 footprint'    gca REW snaps ....  in this brave new world of uber-measurement

I'm thinking some comparisons of the usual suspects  ..  eh6922, some amperex, some nos russkies.

Once I get the bias dialled in, it should be fairly quick to get an updated picture of this family.

 

[alexc]

OK I think I get it a bit better now, here's my current 'take'  on the issues    ..  still struggling a little, so thanks for the patience Abbey   

I do think you have the important things here very correct and I do want to understand the measurement process better.

.....

Because of the kind of  testing that I like to do    ..  ie  'single tone'  and  'noise floor' signatures  ... 

which are closer in their 'nature' (energy distribution in the  freq domain)

to 'white noise' ,

I should use settings in the analyzer which are best suited to that 

ie. 'spectrum' mode rather than 'rta mode', as an example - I'm guessing there are others ..

that does not 'obfiscate' results in the  'hi freq' register ,  leading to a more 'reality based' interpretation of the data

...

That does make more sense to me now  - thank you. 

I can do some more reading of the manual and do some more exploration of the different settings available in REW to get a better understanding of the measurement capabilities - particularly in the area of these baselines that I like.

 

[alexc]

I had a read of the manual regarding the 'spectra' and 'rta' modes  ..

It  all seems reasonable, as does the rest of their details on how to setup REW ...

....

For this specific type of profiling of  'line-in/line-out testing : single tone and silence' for a 20-20KHz view

I still think it basically comes down to a preference of the displayed detail.

I tried most of the options out, windows and such - I still think 'my settings'  give the most useful interpretation of what I'm trying to measure. 

Especially in an 'instantaneous' type of view with the desired 10Hz to 22KHz viewed as the default 'zoom' setting.

I can tweak the 'averaging' and 'windowing' and 'fft-len' within each of the spectra/rta modes and see varying details come into or out of focus sure enough. 

Certainly, when one zooms in on smaller displayed freq ranges, 'detail' and 'resolution' issues can come into play a lot more  .... 

but with a standard view, such as my settings give, well I think :

..

'what's there is actually there'

in a given 1/3 of a sec or so and to an amplitude error of 0.1 dB when everything is stable 

-  there can be some uncertainty to the tune of 0.3dB (when stable) rising thru 0.5dB and up to 3dB  ..  when the bottom end is twitching about  .. from an unstable diy-under-test

- it's that 'hum floor crest factor' (my terminology)  I go on about  ..  and it is a good measure of just that  ..  low freq stability  ..  it part-answers the question 'apart from  (subtracting) the peak, what is the relative importance of the  low end in the schmutz that remains  ?'

- as does  the 'peak-rms-factor'  (my terminology) figure in the floating 'rms input' headsup windowlet

- if things are going right each of these are very low  ...  approaching 0.1dB or so  ..  saying literally 'nothing, apart from the peak, to see here'.

I'm pretty sure that's as good as one could ask for, generally speaking. I would say I'm 0.2dB to 0.3dB now, in 'real life test' like here.

..

The basic signature of the hi-freq register noise floor,  at the end of the day,  is different for the two modes - consistent within each mode but distinct from each other.

I get no refinement to answering the basic question of 'what's my unit *really* doing, noise-wise,  at silence, right now ?'   

by using one mode over the other. 

Visually, and with making the most of the 'single measurement cursor'

....    apart from the obvious difference  in the  'envelope' of the hi-freq response,  all that 'shaded black' trace to me is not useful and masks where any real instantaneous action might be happening - it amounts to more 'averaging' as far as I can see.

....

SO - one take's their pick of 'mode',  then tweaks 'settings' within it to suit their testing scenario as best as possible for today and the near future,  and carries on. 

Who knows where it (REW)  might go  in another year or two.  I'm sure there are better instruments for metrology, but using *this* one I don't think I'm demonstrating any operator error  .. nowadays   

I sure have, without question, historically - but *now* I think I'm getting good results and have always enjoyed putting them out there for others to look at.

I know there are better 'protocols' and 'supporting external test tools' to improve overall test methodologies for sure

.....

Remembering, my objective is to do  'long term collection of baselines'  of noise, single tone tests and thd measures and frequency response sweeps so as to 'keep it real' and 'with traceability'  for my diy stuff as best I can. On the cheap!

....

Of course I remain really interested to see how other analysers, particular the really good ones, handle a similar display RTA mode.

I would definitely expect them to have various normalizing functions - as to their way of handling what I think is a basic property of the 'RTA mode' and similar implementations - allowing some amazing and comprehensive (ai ?) dataset mining, baselining, cross-correlating  and ultimately, convolutions thereof ... no doubt.

'Gold standards' and metrologists are always great while you can keep your hands on them - they don't come cheap. I used to have access to them alot for medical instrumentation  when I was working for a living. Nowadays, not so much.

...

This 'diy thing of ours' is much more fun, although just as detailed in the required info levels.

Thanks to all the people  who likes to discuss this diy stuff and share their knowledge!

 

[abbey road d enfer]

For many years, designers have relied on sensitive voltmeters and dedicated weighting filters for noise optimization.
Spectrum analysers, whether RTA or FFT-based) simply were not commonly available, and actually nor really necessary.
Most noise in analog audio gear is either Johnson noise or hum residuals. Assessing them just takes a sensitive meter and adequate filters.
Distortion can be assessed satisfactorily for most needs with a distortion analyser, which is, after all, a glorified notch filter.
Spectrum analysers have the advantage of providing a more detailed array of data that allow extracting the ususal single-figure results we're used to see.
However, this takes a little math.
The single-figure result is the integration over the desired BW of all the individual bins (FFT) or bands (RTA).
The more bins (FFT length) or bands, the lower the level in each.
In fact, many RTA's use an actual FFT and some math.
The availability of FFT has made the analysis of distortion much easier,prompting a new interest in tube distortion.

 

[alexc]

All very true.

 

[ruffrecords]

For many years, designers have relied on sensitive voltmeters and dedicated weighting filters for noise optimization.
Spectrum analysers, whether RTA or FFT-based) simply were not commonly available, and actually nor really necessary.

It is said a picture speaks a thousand words. An FFT plot from an RTA with derived values  contains a host of information not available from voltmeters and filters. You can see 1/f noise. You can see mains hum harmonics so as to determine if you have a ripple or screening issue. You can see all distortion harmonics and their values. Using a periodic white noise generator (which REW has built in) you can see a direct plot of frequency response and see exactly how an EQ changes response as you alter its controls. And of course you can save and print the display so others can see.

I started out with an audio millivoltmeter and a distortion analyser that gave only THD. I later bought a tuneable one so I could see individual harmonics. Now I have REW I would never go back to them.

Cheers

Ian

 

[abbey road d enfer]

It is said a picture speaks a thousand words. An FFT plot from an RTA with derived values  contains a host of information not available from voltmeters and filters. You can see 1/f noise. You can see mains hum harmonics so as to determine if you have a ripple or screening issue. You can see all distortion harmonics and their values. Using a periodic white noise generator (which REW has built in) you can see a direct plot of frequency response and see exactly how an EQ changes response as you alter its controls. And of course you can save and print the display so others can see.

I started out with an audio millivoltmeter and a distortion analyser that gave only THD. I later bought a tuneable one so I could see individual harmonics. Now I have REW I would never go back to them.

Cheers

Ian

I certainly don't disagree with that; spectrum analysis is a powerful tool, but, as it is with any tool, the more powerful it is, the more competence it takes.
I've seen too many cases of spectrum analysis yielding incongruous results because of misinterpretation or improper setting.
One should be alert to the many caveats.
For example, evaluating an EQ with an RTA is a very common mistake. Or assessing a HPF with a 1024-point FFT.
Clearly, you're one of those who read the manual before plugging; this is far to be the case in many instances.

 

[ruffrecords]

I've seen too many cases of spectrum analysis yielding incongruous results because of misinterpretation or improper setting.

Me too. I rememeber one case where an audiophile published the spectrum of his new tube preamp. It had the noise floor bumping along at -100dBu. This, he claimed meant his preamp had a -100dBu noise floor. I explained that  was probably as much as 43dB out but I did not receive a reply.

Cheers

Ian

 

[abbey road d enfer]

Me too. I rememeber one case where an audiophile published the spectrum of his new tube preamp. It had the noise floor bumping along at -100dBu. This, he claimed meant his preamp had a -100dBu noise floor. I explained that  was probably as much as 43dB out but I did not receive a reply.

Cheers

Ian

Alas!  That's a perfect example of what an abundance of resources, shed on lazy Philistines can generate.
In the past, only a few idle rich could do that, and were easily identified. Today, thanks to democratization, there are so many of them it's much more difficult to filter them.
Do I sound elitist?  8)

 

[ruffrecords]

Alas!  That's a perfect example of what an abundance of resources, shed on lazy Philistines can generate.
In the past, only a few idle rich could do that, and were easily identified. Today, thanks to democratization, there are so many of them it's much more difficult to filter them.
Do I sound elitist?  8)

Thanks to the internet there is so much more of everything - good and bad. I would not be without it because it makes available a wealth of information but the price is exposure to a large number of idiots.

I remember back in the 80s when we first started emailing PCB layouts and mechanical drawings to manufacturers in the Far East. It seemed a revelation at that time. I also rememeber sending a customer a telex from my Commodore 64 around midnight on a Sunday. He was bowled over I was still working for him that late (in fact I had just had a light bulb moment so decided to put it on paper before I forgot).

When automobiles were first invented they were only available to a select few. Today out roads are clogged with them.

When I think about the technological progress in the last few decades (in 1980 I ran the project that developed the Dragon 32 8 bit home computer for the UK, then Europe than the USA) I wonder what the next few decades will bring.

Cheers

Ian

 

[alexc]

One of the things I most love about the 'real time' responsiveness of REW, is tuning the harmonic structure of a tube stage and with amazing dexterity.

The e83f is a fine tube ..  I think a 'precision' one, so to speak.  I've run about  a tenth of  it's different capabilities so far.

Its a privilege to have a nice one under the the lens of REW. Truly made for each other!

In a typical gca type stage, with a 500ohm trimmer in the cathode circuit, it's amazing how one can turn up or down the harmonics  ..  the even set and the odd set.

I like to find the minima, then adjust around there for the least odds, generally.

Also of interest, the effects of different B+ voltages on thd ...  varying at the hv regulator module and adjusting the bias to hold .. while watching REW and how gain changes, harmonic structure, hum etc.

Of course, be sure to use a  'well insulated screw driver'  for those pcb trimmers ..  one slip and who knows where the tip might land.

 

[alexc]

Just for some basic education, Abbey ..  could you elaborate a bit on the problem with using an RTA for some kinds of eq testing ?

I could use a refresher

I mean I know that doing any sort of eq increases thd  ..  by definition , generally speaking.

And also the log scale of the frequency axis can scale the characteristics of a filters slopes.

Are there issues with doing basic sine freq sweeps with eqs ?

I suppose this is where one starts using the more advanced (from my perspective) capabilites  ..,

... white noise, pink noise and so on.

Thanks

 

[abbey road d enfer]

Are there issues with doing basic sine freq sweeps with eqs ?

Sine sweep is actually the most accurate measurement for filters, provided the sweep rate is low enough to allow the whole circuit to stabilize (that includes the meter's rectifier). If the sweep is too fast, the measurement is actually that of a past event.

... white noise, pink noise and so on.

Spectrum peasurements imply the existence of one or more filters. Their slope comes in addition to the slope of the DUT, so the resulting graphs are always too wide.
It may look like FFT does not suffer this, because the BW is split in numerous individual narrow bands, but in fact a 1024 point FFT has a BW of 19.5 Hz, which means the 20Hz bin has a relative BW of one octave.

 

[alexc]

Nice  ..  I indirectly knew of those things but it's always good to do some 'relinking' of packets of info.

I think REW does allow durations of sweeps, so I can do some trys of that.

On my bench, I have the 'filter' section of a gPultec racked up awaiting some new makeup amps ..

I think I'll hook that up to my tube test rig and try some of the white/pink and variable duration sweeps.
Could be quite nice with this ef83 I think

....

It's always good to consider  the filtering aspects of the spectral analysis ..  and what it means in practice. I'm relearning this stuff. I did study it but that was 35yrs ago and mostly now 'archived'.

I have the book 'Scientists and Engineer's Guide to DSP by Steven W Smith  1997'  which is *way* better than the texts for signal analysis were in 1984 when I started that study.

....

I did use those studies, at one period, and wrote some 'vibration analysis' tools .. similar to the more basic  RTA type tools  ..  for the rail system in aust.

We (engineering teams where I was a 'cadet') would wire up trains of 4 to 6 locomotives and about 1km of various trailing stock  ..  like coal carriers etc  ..  every transducer imaginable, kms of wiring, combining networks, multitrack recorders and our own digital converts  ( I did that once too). 

Then 'hit the outback' for weeks at a go  ... there were sleeping carriages for the staff and we could pull up anywhere we wanted to breaks and meals etc.

You haven't really lived as a 'wire jagger' until you've 'roadied' inside the engine compartments of a 3000hp diesel electric loco going strongly ..  and I don't mean idling  ..  pulling a huge train. 

It's extremely hot and noisy next to a V16 super charged with a blower half the size of my car.  The 'perks' was to get a much desired 'seat' in the loco cab most of the time  ..  air conditioned, air cushioned with space to nap some.

And then later, back in the labs to analyse it all  ..  vibration analysis was the main thing but also all the electrical parameters and many mechanical and hydraulic things as well.

Literally 50 or so engineers, techs, loco drivers, signals guys  ..  metallurgists    ... the works.

There were even 'pyro engineers' who were incredibly expert in all aspects of 'burning stuff'.

 

[alexc]

and back on the 'screen driven pentode' thing  ..

Here is attached an article from 25yrs back ....  mr Tim de Paravicini of EAR  ...  he published the design and so on back then for 'people like us'  diy folk.

It's prolly the most well known example of this kind of thing ..

...

has example with the el509 power pentode, with grids grounded thru 10K and screen driven from a totem pole driver.

 

[ruffrecords]

Spectrum peasurements imply the existence of one or more filters. Their slope comes in addition to the slope of the DUT, so the resulting graphs are always too wide.
It may look like FFT does not suffer this, because the BW is split in numerous individual narrow bands, but in fact a 1024 point FFT has a BW of 19.5 Hz, which means the 20Hz bin has a relative BW of one octave.

Does anyone use 1024 point FFTs any more? REW allows ones two orders of magnitude longer.

Cheers

Ian

 

[abbey road d enfer]

Does anyone use 1024 point FFTs any more? REW allows ones two orders of magnitude longer.

That is true; there are still apps and hardware that use 1024 points. My comment was to illustrate the fact that all systems have their limitations. I haven't even started on windowing. 

 

[ruffrecords]

That is true; there are still apps and hardware that use 1024 points. My comment was to illustrate the fact that all systems have their limitations. I haven't even started on windowing. 

Windowing is a whole other ball game.  For instance, I have discovered the REW periodic white noise only works properly with a rectangular window.

Cheers

Ian

 

[abbey road d enfer]

Windowing is a whole other ball game.  For instance, I have discovered the REW periodic white noise only works properly with a rectangular window.

Cheers

Ian

FFT misunderstanding is a subject that comes so often on the LTspice group...