[Eeyore]

wrote:

Eeyore wrote:
wrote:

But I don't know if it would be a good idea to make both stages exactly the
same... that would give a severe upper limit phase shift and may induce
oscillation...

No.

How about you explain why you think that and I can correct any mis-assumptions you made
?


I think that because I learned it in school! I have more than 1 text book that
mentions it.

If you have gain circuits in series that have the same phase shift, you can get
a 180 degree shift in the feedback loop at the high end where the gain is above
1.

Which = oscillator.

Only if that final output is connected back to the input. Which it isn't.


Two stages with 90 degrees, three stages with 60 degrees, four with 45, etc etc.
This affects high gain stages more, obviously.

So its best to change the response (phase/gain) slope of each stage so they
don't line up.

It makes no difference at all.

Graham

[Patrick Turner]

Phil Allison wrote:




I think that because I learned it in school! I have more than 1 text book
that
mentions it.

If you have gain circuits in series that have the same phase shift, you
can get
a 180 degree shift in the feedback loop at the high end where the gain is
above
1.

Which = oscillator.

** WRONG !!!!!!!!!!!!!!

Where does this fool pluck this drivel from ??

His bum ?

........ Phil

Hmm, I thought that for the transition from stable to unstable
occurs where the open loop gain where phase shift is 180 degrees is
multiplied
by ß, the fraction of output fed back = 1.0.

Ie, for borderline stability, find gain, A, where phase shift is 180d,
then maximum ß for borderline stability = 1 / A.

An example in a tube amp is where the gain without NFB, open loop gain,
or OLA,
is say 200 at 1kHz, where OLA is maximal, and phase shift = 0.0degrees.
But where phase shift = -180d could be at 70kHz, and OLA could be only
40.
Borderline stability would be where ß = 1 / 40, ie, a sample of the
output voltage
equal to Vo x 1/40 is fed back to one of the input ports of the amp.

At this equation point of 1/A, the fed back voltage becomes greater than
the voltage between the two input
ports to cause the output voltage, and if ß is any more than 1/40, say
1/20, the amp
will surely oscillate.
But it won't oscillate if ß = 1/80.

However, nobody should ever use this equation to establish stability
because the formula
would make any amp on the very brink of oscillation begin oscillation
when the phase shift is increased a bit by say adding a capacitance to
the output
which usn't large enough to reduce A, but large enough to cause say 30
extra
degrees of phase shift, and the amp oscillates.

So to be sure that oscillation won't happen, you need to select the A
where phase shift is 90 degrees,
so ß becomes a lower number because A is higher.

Or else to enable ß to be say 0.1, then gain where 180 degrees phase
shift occurs
needs to be reduced well below its normal level.

At my website which has numerous samples of tube amp schematics which
have all been tried and tested
and which will not and cannot oscillate under any load condition
including not having any load connected at all, there is often one or
more series R&C networks
within the amp to reduce the open loop gain at say 50kHz by typically
10dB
without increasing the phase shift, so that
frequency where the -180 typical phase shift does occur is pushed up to
about 200kHz,
and where open loop gain has become quite low, and the ß can be the
wanted fraction
to ensure stability but also render the amp with sufficiently low output
resistance
as well as reducing THD/IMD, noise and phases shift within the audio
band to negligible levels.

In addition to farnarkling around with OLA mainly *outside* the audio
band,
there is also the option of advancing the phase of the fed back signal
to compensate for the lagging phase of the OLA at HF.
Hence a cap is placed across the FB resistance from output to input tube
cathode.
It has zero effect at 1 kHz, but the cap can produce an extra 45d phase
advance to
help make the phase of the fed back signal at V1 cathode be more nearly
equal to that
of the input grid signal.

Its never a perfect solution, but perfection isn't needed, since we do
not want high levels
of NFB to be operating above 20kHz, and in fact would prefer a much
reduced amount of NFB to operate by 100kHz.
The same general thinking applies to LF phase lead in the open loop
phase character.

Its extremely commom for DIYERs who build their own audio amps
or who build their own RF amateur band amps to suffer huge problems of
parasitic oscillations,
and the more daft DIYers don't even realise they have problems, or that
they could have problems.


So one may infact have some hook up of NFB which works as NFB at the mid
frequencies, say
1kHz, but at 200kHz, the NFB has become PFB, and for any PFB to exist,
a special set of rules exist; and I leave to others to teach us the the
positive feedback formulas.
I build amps, not oscillators, so PFB formulas get little use here.

The man without problems makes nothing.

And never forget

COWPAT = 1 / N squared.

Patrick Turner.

[Iain Churches]

"jack" wrote in message
. ..
"Iain Churches" wrote in message
. fi...
I was grateful for information given by Patrick and others
on a recent thread, on the use of a low noise Op-Amp to
measure the floor noise of a cathode follower stage. Using
my AC voltmeter on its lowest range close to the limits of
its resolution had resulted in spurious readings.

Bandwidth limit the output -- I have found the best to be the HP3581 wave
analyzer which has selectable bandwidths of 3 to 300 Hz a really
economical
way to go. The HP3581 allows you to sweep the measurements from a few Hz
to
50kHz -- this is the information you are really looking for. The HP3581
has
a very good RMS detector but it is helpful to take your time with the
measurements -- a data logger is useful since you really want the envelope
over time (i.e. integral).

Hi Jack.

I have both an HP 3580 and 3581. Thanks for the idea.

Iain

[Iain Churches]

"Phil Allison" wrote in message
...


Audio noise levels ( unlike THD measurements ) are * always * limited
to the audio band - normally to 16kHz or 20 kHz.

You must have an audio band filter after the 60dB pre-amp to define
the
upper frequency limit to do this.

I have just such a filter, BBC type which conforms to the ITU 468
standard.
With this in place, the noise floor is 12µV as I stated originally.



** You did NOT state any such damn thing.

Your comment was the word " weighted " !!!!!!!!!!

Entirely separate from audio band limiting the response.

A "weighting" filter has a non flat response - ie i * weights * some
parts of the spectrum more than others.

The BBC unit is marked:
ITU-R 468 Weighting Filter.



** Right - so it is 100% definitely ** NOT ** an audio band
limiting filter.


Hello Phil

But it *IS* the curve used by the BBC and all facilites within the EBU
for noise measurement of audio amplifiers. So it becomes my method
of choice.

Iain

[Iain Churches]

"Eeyore" wrote in message
...


Iain Churches wrote:

"Phil Allison" wrote

You must have an audio band filter after the 60dB pre-amp to define the
upper frequency limit to do this.

I have just such a filter, BBC type which conforms to the ITU 468
standard.
With this in place, the noise floor is 12µV as I stated originally.

Better than using a weighting filter is a *band limiting* filter in the
first
place that cuts out the influence of noise that's inaudible to humans but
that
the equipment can measure.

The normal measuring bandwidth for audio noise these days is 22Hz-22kHz
(-3dB
points).

Graham


Hi Graham,

The ITU 468 curve is approx 20dB down at 20kHz.
The old "A weighting" curve was "not specified" above 20kHz.
ITU 468 was the curve introduced by the BBC for noise floor
measurement, which is now used throughout the EBU.

I have a number of band equalisers. I could set one of these
22Hz to 22kHz bandpass and place it across the output of the
60dB measuring amp to see how it compares with the
wideband, ITU468 and A weighted results. It might be
interesting

Thanks for your help.
Much appreciated.

Iain

[Phil Allison]

"Iain Cheurchs"


Audio noise levels ( unlike THD measurements ) are * always *
limited
to the audio band - normally to 16kHz or 20 kHz.

You must have an audio band filter after the 60dB pre-amp to define
the upper frequency limit to do this.

I have just such a filter, BBC type which conforms to the ITU 468
standard.
With this in place, the noise floor is 12µV as I stated originally.



** You did NOT state any such damn thing.

Your comment was the word " weighted " !!!!!!!!!!

Entirely separate from audio band limiting the response.

A "weighting" filter has a non flat response - ie i * weights * some
parts of the spectrum more than others.

The BBC unit is marked:
ITU-R 468 Weighting Filter.



** Right - so it is 100% definitely ** NOT ** an audio band
limiting filter.



But it *IS* the curve used by the BBC and all facilites within the EBU
for noise measurement of audio amplifiers.


** YOU are still WRONG - ****head !!!!

That is an unusual * WEIGHTING * curve used in very few places - so its
use MUST be specified alongside the figure quoted. This you did not do.

Any UNWEIGHTED ( or FLAT ) noise test MUST use an audio band limiting
filter - or the number quoted will be impossible to compare with others
where such a filter will invariably have been used.


You still have not got that through your

INCORRIGIBLY THICK ****ING head.




........ Phil

[Eeyore]

Iain Churches wrote:

"Phil Allison" wrote
Iain Churches wrote

The BBC unit is marked:
ITU-R 468 Weighting Filter.

** Right - so it is 100% definitely ** NOT ** an audio band
limiting filter.

Hello Phil

But it *IS* the curve used by the BBC and all facilites within the EBU
for noise measurement of audio amplifiers. So it becomes my method
of choice.

Why ?

It is barely used outside those realms and makes it impossible to compare your
results with any equipment measured in the normal way (unweighted 22Hz-22kHz or
A-weighted).

http://en.wikipedia.org/wiki/Weighting_filter
http://en.wikipedia.org/wiki/A-weighting
http://en.wikipedia.org/wiki/ITU-R_468_noise_weighting

Graham

[Patrick Turner]

Eeyore wrote:

Iain Churches wrote:

"Phil Allison" wrote
Iain Churches wrote

The BBC unit is marked:
ITU-R 468 Weighting Filter.

** Right - so it is 100% definitely ** NOT ** an audio band
limiting filter.

Hello Phil

But it *IS* the curve used by the BBC and all facilites within the EBU
for noise measurement of audio amplifiers. So it becomes my method
of choice.

Why ?

It is barely used outside those realms and makes it impossible to compare your
results with any equipment measured in the normal way (unweighted 22Hz-22kHz or
A-weighted).

http://en.wikipedia.org/wiki/Weighting_filter
http://en.wikipedia.org/wiki/A-weighting
http://en.wikipedia.org/wiki/ITU-R_468_noise_weighting

Graham



Suppose one found the noise from an amp was 1mV
between 20Hz and 22kHz was the same as pink noise in its
energy distribution.

It appears the filter units designed to have the 468 slope
effectively has a bandwidth from 3kHz to 10kHz.
below 3kHz, the attenuation is 6dB/octave.

Thus the 1mV of pink noise would be much reduced
to 0.6mV, because effectively because filter bw is 7kHz instead of
20kHz.

In the case of a phono stage where the amp bw is only say from 20Hz to
50Hz,
and LF noise is therefore likely to mostly appear,
using the 468 filter would give ridiculous results, no?

But then if we weighted the results used in the flat measurement,
ie, boosted F measured at +6dB/octave on a line passing
through a reference point of say 20Hz, then the very low levels of phono
stage HF noise
would be raised by 40dB above the 20Hz level by 2kHz, and if this was
the method used, it'd make it much harder to get a decent SNR.

In a line stage, buffer stage, power amp, the bw is flat from 20Hz to
20kHz.
If the the actual noise present without any boosting of measurements
means that noise is inaudible with gain fully up and with zero signal,
and with normal speakers then isn't is quiet enough, regardless
of whatever the measurements?
If a system can produce 120dB SPL with say 28Vrms applied to 8 ohms,
and we have 1mV of flat noise at this SPL, then the SNR = -89dB, and
probably acceptable to most people.
If the noise is 1mV without signal then it will not be heard.

Noise from PSUs etc can increase with increasing levels and
perhaps an AB amp might have 10dB more noise at the 100 watt level
so the SNR slips to -79dB but at idle its quite OK.
Many tube amps achieve these figures.
But were we to say we will increase the measurement by weighting
the actual measurements, the noise figures look a lot worse.

Some say it matters not if there is low F noise present because our
ears are not sensitive to it therefore if there is a high 50Hz hum
content to the measured noise
spectrum we can reduce what we measure and quote some figure that looks
a lot better than reality.
This is a weighting reduction of what we measure.
I prefer the unweighted measurement, or worse looking measurement, and I
don't like hum either.
And often hum is 100Hz, and switching harmonics of the rectifiers, ie,
pulses at 100Hz and they are visible on a CRO because they are larger
than the
other noise prsent.

I just try to minimise all noise, and after building rather a lot of
amplifiers,
one knows what to expect, and I know that its impossible to get things
any quieter,
lest I throw away the tubes and use all solid state immersed in liquid
nitrogen.
The only case where tubes can give dissapointing noise performances
are in microphone amps or phono amps and where the input
signal is rather low, at say 0.1mV.
If noise is 1uV, the SNR is only -40dB.
So the use of a j-fet can improve things if its noise is 0.1uV,
and then SNR becomes -60dB, and not so bad.
Typical single j-fets can give a real world performance of 0.14uV
of unavoidable input noise which does not have much LF rumbles
and compares very favourably to an exceptionally
well chosen 12AX7 which may barely measure as low as 1.5uV,
if you can find one without LF rumbles and sputterings.

The old way of dealing with noise at low level signals
was to transform the signal up.
Usually low level voltage signals are from low impedance sources
of say 12ohms for an MC cart, or similar from a ribbon mic,
and so the noise from such low resistances is so low,
that the signal can be transformed up from
say 0.1mV to 1mV, and also the noise with it, but because the
noise was so low initially at the 12ohm R, when transformed 10 times it
still remains
below the 1uV level and thus the -60dB snr is achieved.
So after Denon invented the MC in 1949, perhaps there was a Denon 1:10
step up tranny
that went with the cart, and voila, we had 3mV of signal
and its noise plus the noise of the amp input amounted to
less than 3uV, giving snr = -60dB, a not impossible figure, and
the following RIAA filter would have a noise reducing capability.
Usually the MC should give better SNR than MM.
Why did broadcast stations use MC?
The MC must have been better on most counts.
And I have a Denon 103R MC and its rated at 0.37mV output, 1 kHz, but
have no tranny
because I have a single 2SK369 input fet, and its noise is very low.
I tried a tubed input, and the amp noise was clearly
audible above the noise of an un-modulated groove on the record when
the gain was set for normal listening levels.
Late last year I paralleled 3 x 2SK369 together which theoretically
should lower noise by a factor = 1 / sq.root of 3, ie, reduce noise x
0.58 times, nearly -4.7dB,
4 fets give -6dB, and 8 give -9dB and 10 fets give -10dB.

But I have unbypassed source resistors on the 3 fet front end, and such
resistors
create noise, and I have not calculated by how much the noise is
increased,
mainlt because when finished the amp was so darn quiet.

Fet equivalent input noise resistance = 0.7 / gm, so where a 2SK369 ( or
2SK147 )
have 5mA idle current, gm = 0.04 A/V so einr = 17.5 ohms.
If there are 3 fets, einr = 5.8 ohms.

Triode einr is supposed to be ideally 2.5 / gm, so a 12AX7
should have einr = 2.5 / 0.0012 = 2,083 ohms.
Its input noise will be greater than the single fet by a factor
of square root of (2,083 / 17.5), ie, 10.9 times greater.
This is never realised because of the nature of tube noise,
which often includes more LF crap than the fet, so the fet is great for
RIAA.
Now if we used a 6DJ8 with each half paralleled, we would get
einr = 2.5 / 0.013 A/V = 190 ohms, and the noise SHOULD
be 3.3 times that of the fet with 17.5 ohms.
Or if you like the 6DJ8 gives 1/3 of the noise of the 1/2 12AX7.
Getting any 6DJ8 phono stage to perform this well is
slightly less likely than teaching pigs to fly.
Its more than adequate for MM, and is quiter than very many
thousands of generic consumer grade phono amps made with bjt inputs.
But when the input signal is low, the fet is the best answer, imho.
Even the 6C45pi, a russian triode with 40mA/V, same as a single iddy
biddy j-fet
the theoretical einr = 63 ohms, so noise is near twice the fet noise,
but in practice its probably a lot more; I have not
built a phono stage with 6C45pi, and afaik, NOBODY else in the world has
done
any serious report of exactly what the real truth about noise is for
such tubes.

I leave you all to work out what amount of noise is generated in a 5.8
ohm R
over the 20Hz to 22kHz bw range, and don't let speaches about weighting
any thing
distract or confuse you. Make sure you have a meter that reads Vrms
over the 20Hz to 22kHz bw range. Typical DVM have a range from
10Hz to 1.5kHz if you are lucky.

When considering distortion artifacts in addition to noise, the picture
becomes often one dominated by the distortions, as anyone will discover
when they monitor what they are trying to measure with a CRO.
In good gear, where noise is low, and remains low at high power levels,
THD will usually rise well above the noise.
If the SNR = -90dB, and THD = 0.01%, or at -80dB, the THD will clearly
become seen on the CRO, depending on how sensitive you have
made your set up for monitoring such things.
at 28Vrms, 0.005% THD = 1.4mV, a poofteenth of a voltage to measure
in amoungst such a large signal but I assure its as easily viewable on
the CRO
At these signal/THD levels LF PS noise can be greater, so I have a hum
filter
with 12dB/octave below 300Hz to show the THD without the LF noise trash.
Then any switching spikes become biewable as well.
For measuring down to 0.01% THD at the 1V level, the THD = 0.1mV,
and impossible to see on the CRO, so I have an amp to raise this to 1mV,
and the filter is 1.5kHz to 10kHz, with steep slope below 1.5kHz
to remove traces of 1kHz test signals left after passing
through the initial bridged T LCR notch filter with a notch depth of
over 75dB when adjusted for maximum rejection of 1 kHz.
Both the oscilator and filter are adjusted for phase accuracy within a
fraction
of a hertz, because such things are prone to slight phase errors with
respect to
levels applied.

If I want to, I can then go on to apply a potentiometer controlled
tunable filter with Q =50 between 2kHz and 11kHz, and
discover what the first 10 harmonics are of the 1kHz test signal.
Often the THD is so darn low that measuring the H beyond 3H is
impossible because
the noise is above the levels of the higher H.

I spend half my creative life being a policeman to prevent or reduce
noise, distortions, and bull****.



Patrick Turner.

[Iain Churches]

"Phil Allison" wrote in message
...

"Iain Cheurchs"


Audio noise levels ( unlike THD measurements ) are * always *
limited
to the audio band - normally to 16kHz or 20 kHz.

You must have an audio band filter after the 60dB pre-amp to define
the upper frequency limit to do this.

I have just such a filter, BBC type which conforms to the ITU 468
standard.
With this in place, the noise floor is 12µV as I stated originally.



** You did NOT state any such damn thing.

Your comment was the word " weighted " !!!!!!!!!!

Entirely separate from audio band limiting the response.

A "weighting" filter has a non flat response - ie i * weights *
some
parts of the spectrum more than others.

The BBC unit is marked:
ITU-R 468 Weighting Filter.



** Right - so it is 100% definitely ** NOT ** an audio band
limiting filter.



But it *IS* the curve used by the BBC and all facilities within the EBU
for noise measurement of audio amplifiers.


** YOU are still WRONG - ****head !!!!

That is an unusual * WEIGHTING * curve used in very few places - so its
use MUST be specified alongside the figure quoted. This you did not do.

Any UNWEIGHTED ( or FLAT ) noise test MUST use an audio band limiting
filter - or the number quoted will be impossible to compare with others
where such a filter will invariably have been used.


You still have not got that through your

INCORRIGIBLY THICK ****ING head.




....... Phil




Hello Phil.

It is always *so* rewarding to have a meaningful and informative
discussion with a man of your literary skill and intellect:-)

Let's hope we meet very soon.

Cordially,

Iain

[Patrick Turner]

Hello Phil.

It is always *so* rewarding to have a meaningful and informative
discussion with a man of your literary skill and intellect:-)

Let's hope we meet very soon.

Cordially,

Iain

I can understand why you are so attracted to Phil and want desperately
to meet him.
Its all that charm and down to earth manliness that has you drooling.

( There is no accounting for taste, eh )

So exactly how do you measure noise, and what does the use of the ITU
468 filter gadget do?

The BBC unit is marked:
ITU-R 468 Weighting Filter.

After measuring the real world real actual noise voltages between 20Hz
and 22kHz for a given amplifier,
what is the purpose of the above weighting filter?

Does it make the noise measurement greater, and hence worse because it
weights a real measurement,
for example, tells us that noise in the 2kHz to 3kHz region is far more
than what it actually is,
or does it reduce the real levels at LF because it doesn't matter
because the ear won't notice?

I make no apology for posting more questions than answers.

I just measure the real world noise voltages, and don't weight anything.
If I have ALL noise as low as possible, then no amount of mucking around
with weighted figurings
will improve the amp or make it better.

Patrick Turner.