REW Update

[ruffrecords]

@alexc

I notice all the spectra you have included have a rising noise floor as frequency increases. All my tube circuit measurements using REW show the opposite - noise increases as frequency decreases. I notice also that your spectra are rather blocky at low frequencies so I wonder if this is an artefact of the RTA settings you are using?

Cheers

Ian

 

[trobbins]

i just use the spectrum mode setting, rather than an RTA setting like 1/48.

 

[alexc]

Hi Ian,

I've always seen a rising noise floor as frequency rises, at the top end.  I thought it was simply that there is more noise energy in the 'bins' at the higher freqs than lower  ...  and it takes off at 5K and up. I could well be wrong!

That includes many years with TrueRTA before I started using REW's RTA,  and with several different interfaces.

So, I think it is a basic characteristic of RTA apps ...  but I'm not an expert.

...

I have seen the differences in the REW 'spectra' mode (vs  '1/48'  mode) ...  they don't show the  'increase' except beyond 15K or so.

I prefer the 1/48 RTA display, personally.

Do you have a snap you could show ?  I'd be interested in your results.

...

My current REW rta settings are :

24bit, 44KHz sampling, RTA 1/48 display mode, 64K FFT, Hann windowing 50% overlap, 20Hz-22KHz BW filtering

I use update interval of 1, no averaging, no peak hold, and no historical traces ...

 

[alexc]

Here's a spectra of a G9 using pro gear  (not mine)

It doesn't have the rising noise response at high freqs that I always see    ..  just the opposite, like yours, I guess  ...

...

The answers are requiring more brain horses than I can muster! 

Maybe one of our more knowledgeable members can shed some light ?

 

[ruffrecords]

Can you list your RTA settings?

Cheers

Ian

 

[alexc]

I thought I did ?  a post above I think.

Here's a pic of the latest biz  .  50ohms to 100K and reverse .

One a smaller, for input.

 

[alexc]

And one a larger, for output.

For a 'grounded grid' input amp - in a differential setup, and  for cathode following.

The ip traffo is  planned to feed  a  'vari-gm cf' gain cell if I can get it working. I wrote about it a few years back. My thing is a hack of the real article ... it's hard to find those 'feed forward' delay cells with any kind of fidelity.
...

The ouput traffo ...  well it's kind of a 'high impedance speakers' kind of thing. 50ohms output driving a series string of speakers.

They do exist! Like that Bose from many years back  ..  8x mini full range drivers in a secksi cab  ...  all working happy like.

 

[ruffrecords]

I thought I did ?  a post above I think.

Oops, sorry, completely missed that. I will check them against my settings and get back to you.

Cheers

ian

 

[JohnRoberts]

The ouput traffo ...  well it's kind of a 'high impedance speakers' kind of thing. 50ohms output driving a series string of speakers.

They do exist! Like that Bose from many years back  ..  8x mini full range drivers in a secksi cab  ...  all working happy like.

FWIW the popular Bose 901 used 9 full range drivers in series parallel (3 series strings of 3 all in parallel) to maintain 8 ohm load...

You will notice such approaches use prime squares like 2^2, 3^2 (901), 4^2 (sweet sixteen), etc.

Later Bose tooled up their own speakers with 0.9 Ohm voice coils so they could be wired in series and still deliver 8 ohms.

JR

 

[trobbins]

Sort of like inter-leaved secondary windings on an output transformer  - but for your ears 

 

[ruffrecords]

My REW settings are attached. This is an older version so there are less options but the only real difference is I use a spectrum rather than 1/48th octave and I use a longer FFT.

Cheers

Ian

 

[trobbins]

I'm more with Ian on settings, and vary the averaging to suit the application.

I think the shape of the noise floor is more about the soundcard you are using - the floor level can vary if your soundcard has gain controls (like my EMU 0404 USB).  It may also change if you auto-calibrate for a flat frequency response (which I don't typically have engaged).

 

[ruffrecords]

I think I know the reason for the rising response. it is because you are using the 1/48 octave mode rather than spectrum. If you use spectrum mode and set the generator to pink noise, you get a steady decrease in amplitude as the frequency rises as expected. If you then change to 1/48 octave mode the line becomes horizontal. If you now change the oscillator to white noise but keep 1/48 octave mode you get a rising response. Keep the white noise and switch to spectrum mode and the line goes horizontal again.

So basically the 1/48 th octave mode shows the energy in different bandwidths and by definition the bandwidth gets bigger at higher frequencies. So as white noise has the same energy at all frequencies, 1/48th octave mode show it as a rising curve which simply reflects the larger bandwidths at higher frequencies.

The reason for the blockiness is unrelated to the above. It is caused simply byu checking the 'use bars on RTA' option in the RTA tools menu.

The interesting thing is that measured values (distortion. S/N etc) seem to remain unchanged.

As an aside, the periodic white noise (white PN) will lock to the FFT only if the window is set to rectangular. This means you can get an easy frequency response plot using spectrum white PN noise and a rectangular window.

Cheers

Ian

 

[alexc]

Thanks for the observations Ian - I think that jibes with what I thought.

With respect to observing unchanging thd measures etc  with the different modes ...  that is interesting, because I have generally observed (slightly) different results depending on the RTA settings  ...  certainly with different sample rates and window types, but even with some parameters like fft-length and fft-bit-length.

Not so much nowadays, as I've settled on my preferred settings with a 'high degree of comfort'  ...  but for sure,  a couple of years back, I would have a couple of different settings-scenarios and would have to check on each ...  to give myself confidence that was what being reported was 'sane' as well as improving my understanding of the measurement system as a whole.

...

Trobbins, I don't think the basic phenomenon of  '1/48th RTA mode  gently rising hf noise floor' is related to the audio interface or even choice of RTA app.

I see the same thing on all my interfaces (and there are a few) with this kind of spectral display (1/24th and 1/48th octave) and consistently with both TrueRTA and REW for many years.  Other RTA apps I have used, don't have the detail required to even notice it.

I do agree with the assertion that the audio interfaces can have (slightly) varying noise floor signatures across the 'mids' and the 'highs', but this 'issue' is something else I think. At the 'low' bands, modern interfaces 'hum' noise seems to have converged to very, very low levels. Older interfaces that I have, not so much - they can be pretty ordinary with big diffs in hum harmonics.

..

I mostly don't use the 'spectra' mode because the averaging that seems inherent in the hf display ..  it just muddies up the 'instantaneous' picture of things,  imho.

I like to 'take an action' and see the effect in 'real time'  (more or less, with maybe 1/3s delay)  and decide 'better or worse' 

Once I get to the end of testing, I do then apply some of the averaging and historical traces and what not.

The capabilities of REW in this area  ..  of 'multiple historical data set handling' and 'post processing display'  is pretty superb.

 

[alexc]

I'm getting very excited now    re. 'screen drive' mode for e83f

I've just knocked up a simple CF from an 6922  to set up the bias conditions ...  took only a couple of minutes to cobble it in.

I'll have to tune the CF cathode string,  to get the required grid and screen bias voltages at the e83f  ..  it's a little weird that way.

There also another slightly different way of arranging the screen-drive pentode, which I'll have a look into after I get this going ....

(circuit is a few posts back)

....

Just love the luxury of a 'tube test rig'  ..  it's pretty fast to get some results action  ...  which is just what's required ..

as my attention span doesn't support proper 'craftsman' values  ..  at the best of times, let alone when 'exploring' or 'baselining'  tube stuff. I'm more of a compulsive 'baseliner' than a true explorer, tho'. I just love standardised testing!

Here's where my topside is up to  ...

.....

I added the single triode 9pin 6S4A  next to the old  'telefunken 12K-6ohm se output traffo ...  that's going to be fun.

The 6S4A is my ultimate small power triode ... only the 6V6 beats it, and it's a sound thrashing too  ...and it does so with substantially increased plate rating and 30% less heater and more gain to boot. The 6V6 triooded is not as good on rp,  however.
And of course, the octal is a lot larger. 

6S4A is my choice for a 'm parallel se' amp  ...  I like 6 of them together, per channel, wired up with little 'bus bar' wires - it can be quite tidy if done right. Prev. best is 4 of them - it's the last two that really make it! 

Good if one has a switched mode supply for heater. I just do a series + parallel string with a 100VA toroid  at 12Vac.

One can even get 4-way 'bias adjust' pcbs on eby cheap .. small enough in size to use several ..  basically little trimmer networks for the cathode circuit of each parallel stage. 

Always a good idea to try to match those idle currents as best as you can and adjusting a little resistance helps fine tune to a very good degree.

...

Also to note, the big output traffo  ..  affordable, old 'medium fidelity' local unit 'A&R' with a nice cast can,  mid 50s-60s -  I would say  ... it's 5K, 3K primary for 15Wrms push-pull with a secondary taps from 500ohms  down to 100ohms. 

It cost about the same as a pair of edcor xsm from 10 years ago  ...  it kind of picks up, testing wise, where the edcor leaves off. It can take some decently high voltages a bit more safely than the humble edcor  with its exposed terminals and fragile wiring attachments.

It's not bad,  so far ..  I have a lot more detail testing to do with it.  It's just begging for a pair of 7591a  '19W plate' finals. 
 

Happily, they are right there, in place, glowing and ready to go.

Speaking of which, one of a quad nos 7591a (octal 19 watt plate, 800mA heater) turned out to borked  ...  I can't get any heater glow from it, but the heater still shows continuity. But heater wise, it's dead.  I couldn't believe my lying eyes  ..  I'm half believing I got it wrong.  I hope so, these old GE aren't cheap! 

-> update  ..  partial better .. not borked, just with the centering pin snapped off and wrong insertion.  D'oh!
-> so - better than borked but not as good as proper  ...  time to get another.

Oh well, s**ff happens  ..  I fritzed plenty of stuff myself, in my life, that's for sure!

I figure I'll buy a modern set (much less exe) and go forward from there ...  I'm interesting in seeing (and hearing)  how these compare with a high-power  6V6gt (14w plate) nos GE  (just awesome sounding) ,  and, the modern, higher-voltages-rated variant -  the fab and affordable  JJ 6V6s  ..  which I really like.

Other modern 'moderate  plate rated'  6V6gt  can be very fine also - I think I've used most of them at one time of other  ..  eh, sovtek, tung-sol, jj.

The 7591a can take some very high voltages and the screen can take a good beating  ..  just like the JJ 6V6s  ...  but at even higher currents. 

Of course the 7591a has nearly double the heater, for those extra ratings  ...  more like a low-powered 6L6  ie. 5881 and the like.

I'm not sure if the el34 is comparable, being a true pentode vs a beam-power type. These distinctions can get a little hazy  ..  at least in my mind.

I tried out the old russkie 'little 6L6'  type  ...  not impressed unfortunately ... prolly just bad luck (I only have 2) but the prices have increased a lot for them and it's just not worth bothering with.

Nothing beats the modern Tung Sol 7851a 'super 6L6' for 'big power' hifi greatness  ...  at a reasonable price  and with a very reasonable heater current of 900mA and  a generous 35W plate rating. 

It just gets in under the bar before one looks at kt88 and larger

Testing at those high power levels will have to wait until I 'bolt-on' a 'variac' and finish my Mk2 psu    This Tube Test Rig has limits around +370V at 150mA or so ...  but I have two of them available  ..  in my 1RU (+ 3RU of upper space required)

This current PSU has easy variation of HV to some degree, but one has to be very careful not to dial it down to0 far if HV current consumption is significant - the heatsinking needs to be up to what you demand of it.

I am running a fairly low regulator voltage drop, with low currents, so no probs.

For the Mk2, I'll have a 'higher power hv' cct seperately done  with the 500VA variac  ..  and the rest of the lower powered stuff can be done  the same as now.

 

[alexc]

So ..  onwards with trying to get a result on 'screen drive' ..

So far  .. no good      ..  but it's early hours yet ..

.....

I have a 'CF driver'  set up with a series string of resistance in the cathode to setup the required conditions    ...  but there is a lot of tuning to get it to work  - as I have read.  This is my first attempt to cobble together.

i) bias voltages (dc)  for the scrn_drv pentode  ie.  g2, g1, k
            -> some very particular values req'd

ii) ratio of signal voltages at g2, g1
            -> some very particular ratios req'd

For now,  I have  ..

-  ball park currents flowing in both the cf driver triode  and the screen-drv pentode

- 6922 driver cf is working fine at 1.5mA or so  with amazingly low thd  at 0.079% or something like that even with the rough and ready bias conditions I have and the small-ish 1Vpp starting signal 

-  the  scr_drv pentode is flowing current fine around 5mA or so

-  no 'gain' to speak of, at the the plate output,  of the scr_drv pentode.

The pentode output spectra looks like a regular gca stage but with it's grid at a low-ish Z,  mostly 'shorted out',  ..  resulting in a relatively tiny output signal  ...  in other words ..  not screen driving at all. Just a very crappy gca 

The basic gain structure is :

-  x1 into the driver cf,  about x0.95 out of the driver CF at the cathode

- cf driver cathode dc and signal voltages are divided to ground in a particular ratio
    like 60% to scrn_drv g2,  and 40% to scrn_drv g1 - so one is losing 6dB or so into the next stage g2 and even more into g1

-  scrn_drv plate is usually driving an se traffo primary at good high volts as is usual typically 3K or 5K to 10K and so on - I'm doing a plate load resistance, not a traffo and so my scrn_drv plate is way too low etc.

....

SO a *pitiful*  turnout for today  . ??? ..  except the eh 6922 triode stage as a simple CF.  Phew! It's a truly great performer, even in the crude hands of a  gardener  like me.

Time to hit the studies again tonight  ....  regroup for the next attempt 

A bit like with those rock monkeys in 2001,  really.  REW is my 'monolith' and the Tube Test Rig is my  'whirling sky bone'. 

And sure, I like  'tapir' just like any.

The golden teacher of experience will surely inspire me,  to somewhere interesting. I have no doubt.

It does tend to ramble, however  ..  I do know that.

....

*Attached* is another design along these lines posted from 'diyAudio'  ...  it shows an (the) alternate way of handling g1

It uses a 'totem pole' driver  and shows the scr_drv pentode's grid grounded thru 100K  (rather than the lowest leg of a resistive divider in the driver  cf cathode circuit - usually 10K or so).

That's basically the two main ways, afaik,  to get into this mode.

Sometimes the cathode voltage of the scrn_drv pentode is 'augmented' with a connection to a dc bias 'boost'  ..  coming from just about anywhere in the circuit that's handy. 

...

Here's a comment from 'diyAudio' contributor regarding why 'some pentodes are good for screen drive and some are not'. 

Note this is generally in the context of  TV 'power pentodes' with 'lower max screen voltage rating'  generally known as 'sweep tubes'.

ie. very high plate voltage rating, but quite low screen voltage rating and as well, a low-ish screen power rating.

..  I'm trying first with small signal ones to jump start my thinking processes  (needs several days notice at least!)

----

The critical factor for practical g2 screen drive, is the transconductance of grid 2. (high gm2 best) You just want reduced grid2 drive voltage basically.

gm2 = gm1 / Mu

This will generally show up in datasheets as a low max g2 voltage spec or a low g2/g1 Mu.

The g2/g1 Mu of 6L6 and similar KT tubes is around 8. While the more practical TV Sweep tubes have a g2/g1 Mu of around 3 or 4  and  so require about half the grid2 drive signal Volts of the 6L6 types.

----

So - I'm still pondering that  those things.
....

On top of my many  other  incorrects, I think I also  just have the wrong kind of pentode .. and that is compounding  things  ..  resulting in a very low gain at the plate output.  e83f has a high  'mu g2-g1'  spec of 38.

I don't just mean the disparity between 'power pentodes' and 'small signal pentodes' - it's more like low mu triodes vs hi mu triodes. But in pentodes.

Not quite sure - more data needed.

Either way, I still need to think a lot more about it   I have a few more basics to cover  .. the se traffo for example.

I could try to find a more suitable pentode from 'stock'  ...  at least I have several available 'tube socket' types on hand to play with.. 

otherwise I could try to get hold of a couple of sweep tubes and  get a 'contributed circuit'  going 'as shown'  ( read the manual  for once!)

I'm not game for a 'top cap' anode type except for in a 'proper build'  ...  I have a few candidate if I have to buy in.

I figure best to get some results first with what I have here
....

Screen Drive is very  intriguing and I know it does work  ..  but takes time and skill I guess.  The diyAudio guys like to publish 'scrn drive curves' which they use to properly bias  .. in detail.  Regular data sheets don't have these kinds of data.

 

[alexc]

and heres the prototype choice for a potential cheap 'buy in'  more suitable for screen drive  ...  in a 'lesser' el84 prototype as far as I can currently determine  ie. from my 'computer seat'

el86 with the 'low' mu g2-g1    mullard and other nos .. ge, rca  etc

Funnily enough, this one was quite popular down here in Aust (Oz) and us nos are still available  for better than soviet prices - a case of 'who cares'    the el84 is better!

There are also  a few old soviet types based on this 'slightly less statured' el84 avail affordably ...  6p15p, 6p43p, 6p18p.

These nos Soviet types  do have their staunch adherents,  as  regular moderate 'power pentode' types  ...  especially in triode mode ... where they have been touted as being very linear.

Not sure if they 'spec'  the mu g2-g1 parameter    ..  only the mullard, brimar  and a couple of holland pentodes seem to mention it.

Another one with the low mu(g2,g1) is the 6aq5  at 10..  which is like a standard 6v6 in a really little bottle 7pin.

Also el34 with low mu(g2,g1) at 10

 

[alexc]

and the el84 proper which I have 'in stock'  ...  apparently not as 'meritorious' in this metric  .. even discounting the operating conditions somewhat.

..  and the e83f is 'less' still more, with  mu g2-g1  at 38,  quite a high-ish value apparently

Its all very curious, really.  A reasonable low cost tube pursuit for those with time to  ...  pass.

 

[abbey road d enfer]

I think I know the reason for the rising response. it is because you are using the 1/48 octave mode rather than spectrum. If you use spectrum mode and set the generator to pink noise, you get a steady decrease in amplitude as the frequency rises as expected. If you then change to 1/48 octave mode the line becomes horizontal. If you now change the oscillator to white noise but keep 1/48 octave mode you get a rising response. Keep the white noise and switch to spectrum mode and the line goes horizontal again.

So basically the 1/48 th octave mode shows the energy in different bandwidths and by definition the bandwidth gets bigger at higher frequencies. So as white noise has the same energy at all frequencies, 1/48th octave mode show it as a rising curve which simply reflects the larger bandwidths at higher frequencies.

I'm actually quite surprized that such a basic mistake could be done.
Spectrum is based on a linear frequency scale; the bins have a constant absolute width (i.e. same f2-f1 expressed in Hz) and thus are dedicated to working with a noise source that has also constant Hz width, which is the definition of white noise.
OTOH, RTA is based on a constant relative BW (i.e. same f2/f1 ratio), which implies the use of noise with the same characteristic, i.e. pink noise.
REW is a very nice tool, but as with any tool, one must learn to use it and how not to misuse it.
IIRC there are enough explanations in the Help, but again who does RTFM?

 

[alexc]

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

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 used to post snaps from years back where I would annotate this 'gently rising hf noise signature'

My own 'gut feeling' tells me that it is reasonable to expect the trace to do exactly that - there is more energy in the hi freq register of the noise floor and the log display makes it inevitable that this should show up reasonably dramatically.

If the trace were 'flat' all the way out to the high freq limit, I am sure I would think that was 'massaged'.

....

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

The few snaps I have seen of the pro types don't seem particularly detailed one way or another.

But I'm a complete newby in today's high dollar hardware 'pro analyzer's.

Maybe one of our group with access to such, could post a snap of a simple loopback 'silence' in a similar display mode ?

..

And up till now, I've mostly been concerned with the quite simple sine type testing with a focus on the  'hum' region, whereas now I am interested in the rest.

I can't recall using white-noise or pink noise tests much, so I don't know if some types of analyzer are more tailored to that kind of thing.

....

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

I think someone mentioned an 'auto-cal' type of function somewhere  which does such  ... 

I suppose that would amount to a kind of 'normalisation' function. 

Perhaps  an  'RTA Mode  Raw/Normalised'  button   which could be useful for the purpose of 'baseline' performance measures that I like so much.

..

Here's an annotated  snap from TrueRTA of 6yrs back  ... 1/24th RTA I think ..  same thing  .. Motu 828MkII Analog Loopback

- you can also see that damnable hum spike I lived with for several years before binning that interface 

Today I would quit rather than live with that!

..

OK - off to read the manual.