[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:

Ian Bell wrote:
Patrick Turner wrote:
Ian Bell wrote:
Patrick Turner wrote:

snip,

What is the maximum voltage gain at say 1kHz?

24dB

Gee, that's only about 15x no?

Correct but there is (will be) a 10:1 mic transformer at the input so
the overall gain is 10 times greater.

OK, but its the amp noise we are concerned with.

What is the noise at the amp output with maximum gain and with input
grid directly shunted to 0V close to the input?

Not possible to measure accurately at the moment as the LF blips whack
the meter needle all over the place one you try to see noise below 1mV.
That said, looking at it on a scope you can see the broadband noise
underneath the LF blips and I would estimate the noise at the output
with the input shorted as about 50uV rms.

If you have 50uV of noise at the output and gain is 15x, and input is
grounded, then you could have a total of 2uV grid input noise if the
input tube is a real good one.

Agreed and that is about -114dBV at the transformer secondary. At the
primary it is 20dB lower.

2uV gets amplified to make about 30uV at
the output, and some of that is LF noise. where does the rest of the
noise come from? By observation you should be able to see where the
noise is being generated and how, and find ways of stopping it without
much complexity and cost. In my MC phono amp without any GNFB with RIAA
correction, but using a passive RIAA, the LF gain at 20Hz is MUCH
greater than your 24dB yet the LF noise at the output is minimal,

I would be interested to know what the measured broadband noise is at
its output.

In my MC phono amp with j-fet input cascode plus µ-follower gain stage,
the LF gain at 20Hz is about 10,000x, or 80dB.
So 40uV of low bass signal becomes 400mV at the preamp output.
There is virtually no loss with the passive RIAA filter between the
cascode stage and following stage.
A signal of 0.4mV at 1kHz becomes 400mV at the output, and 4mV at 20kHz
also becomes 400mV at the output.

The RIAA filter has the effect of lowering the broadband noise of the
input stage as well as the noise from the vinyl.

Now when the amp is used with vinyl and turned up to good loud levels,
the noise of an unmodulated groove in the vinyl swamps the very low
amount of amplifier noise. If the arm is lifted off the record, there is
almost slence, and if the volume is turned up another 20dB to max, the
noise is a high pitched hiss with LF noise buried in there somewhere.
The j-fet gate is terminated at the input with the low impedance of the
MC cart bypassed with 470 ohms and 0.1uF in my case. ( You may wonder
why I have 0.1uF there, but it reduces the high distortion of high level
HF signals coming from the vinyl. The wanted HF content is unaffected. )

Now if you do the same test with an MM cart of 4mv 1kHz signal instead
of the 0.4mV with my MC, and with a 12AX7 input stage the noise of the
amp is barely below the noise of the unmodulated groove, but generally
quite low enough. If you lift the arm, and listen for noise, its low,
but present, and partly due to the 47k loading and higher cart
impedance. If the gain is turned up 20dB, the sound is a lower pitched
hiss with far more LF content because the tube input has much more
flicker noise than a j-fet. The overlall performance of the j-fet input
results in an unweighted SNR reduction of 20dB at least over what a good
12AX7 might ever do. Perhaps the j-fet performance is at least as good
as the tube with a step up tranny for MC, but i have never used a step
up tranny, so I cannot explain exact figures.
Denon invented the MC way back in 1949, and the DL103R cart is still
available and still well regarded, and I have one. Radio stations
preferred the MC because of better noise figures and possibly lower THD
and I sure found it better than MM Shure V15.

Broadband noise in phono amps is converted to noise with bandwidth of
very low F to 50Hz, the pole where the RIAA cuts off.
So the bandwidth is reduced by a factor of 50/20kHz = 1/400, and the
noise voltage becomes reduced by factor of the square root of the
bandwidth reduction, or 1/20 of the broadband noise. Noise below 100Hz
is less noticeable to the ear, so the noise we get with vinyl can be
weighted so in fact the noise with vinyl is rather good so all my pals
with record players say.

Vinyl isn't so good when the signal drops low on very quiet musical
passages, and then any hum if present or clicks and pops become real
irritating. But all the folks I know clean their records, and mostly
enjoy vinyl for jazz and the issue of noise doesn't exist and the sound
is usually better than cd versions of the same music from the same
master tapes.

Many commercially made integrated amps and preamps using solid state or
tubes I have had to work on have far worse noise performance than the
amps I have built. Often the response for each channel is very
different, and well away from the RIAA curve. Typical mass produced
crud! Hum levels are high, and if you examined the phono stage output
with a CRO, there you will see the trace flapping up and down to the LF
content from a poor amount of B+ filtering and complete absense of any
regulation.

I tried to measure the equivalent input noise of my MC amp with 2SK369
input. Seemed to me it was 0.14uV with gate to 0V.
It doesn't get much worse even with an unbypassed source resistance of
50ohms for MC.

The 2SK369 is a widely available j-fet with a slightly lower Pdd rating
than the identical 2SK147.
using them with Id = 5mAdc and Ed = 10Vdc gives gm = 40mA/V, and j-fet
input noise resistance is supposed to be proportional to 0.7/gm. With
triodes input noise resistance is proportional to 2.5/gm.

So its much more difficult to get a triode amp with high gain to have as
low an SNR as one with a humble tiny little j-fet at its input. In fact
the invention of the j-fet made tiny audio amps possible for use with
tiny microphones, and many have been used since in 1,001 spying
operations. Not all j-fets have high gm. Some have gm little better than
a 6AU6 or 6DJ8. But the 2SK369 has 10 times the gm, and as the formulas
above imply, EINR is 30 times lower than many tubes, and noise is at
least 1/sq.rt30 lower. The j-fets don't have as much LF noise as the
tubes. Mosfets on the other hand are not so good despite their much
highr gm. Lots of "popcorn noise".
Nobody uses a mosfet for a phono amp or mic amp input device.

Presumably, a transformer coupled microphone feeding a j-fet inputted
mic amp would give rather superb noise figures.



and
the result using 1 fet in cascode with 1 triode, then 2 triode
µ-follower gain stage produces an outcome equal or better than most
other phono amps I have tried including SS with opamps.


What happens if you temporally connect a spare 1,000uF or more to be in
parallel to the last 100uF cap in the filter line up, ie, the filter cap
giving the B+ supply to stage 1 of the mic amp?

Not tried that yet, I have a spare 470uF or two so I'll try that.

Noise should fall a lot with the extra C added where it'll do the most
good.

Think big, use enormous C values if you cannot bring yourself to make
what might be a very simple shunt regulator in your preamp.

Yes and no. I was using just a couple of RC stages using 470uF but then
I realised the five RC stage of 100Uf each would perform better. I
suppose I could go bananas and replace all the 100uF caps with 470uF ones.

The cost now of generic 470uF caps rated at 350Vdc is not huge, and far
cheaper than the 100uF caps were in real terms back in say 1960 when a
100uF cap was seen as a frivolous extravagance by bean conters in charge
of design teams at major manufacturers.

Agreed. You advised me of this about a year ago and I picked up a bunch
of 470uF 450 electrolytics as a result.

Keen diyers will *NEVER* try to
emulate the pausity of design by accountants among yesterday's people.

I think there is more to it than that. It is well known that a string of
five RC networks is better than a single RC network of five times the
capacitance and resistance. Employing that technique AND using much
larger caps should bring about a significant improvement.

Indeed it WILL make LF jitter much lower.

Consider 4k feeding 4,700uF, ( 10 x 470uF in parallel after a resevoir
cap.

The -3dB pole is at 159,000/4,700/4,000 = 0.0085Hz. If there was 5Vrms
of ripple at 100Hz at the resevoir C there'd be 0.423 mV at the 4,700uF.
If there was 5V of 1Hz ripple at the resevoir C, there would be 42.3mV
at the 4,700uF.

Now try having 4 sections of RC with 1k and 470uF. 1 section of 100Hz
ripple reduction factor will be 0.0034, and with 4 sections the ripple
reduction factor = 0.0034 x 0.0034 x 0.0034 x 0.0034 = 1.33 x 10 to the
-10.
You *will not* be able to measure 100Hz ripple. At 1Hz, the ripple
reduction factor of 1k and 470uF = 0.34, which isn't so hot, and after 4
filter sections is approximately 0.34 x 0.34 x 0.34 x 0.34 = 0.0134. So
with 5V at 1Hz at Cres, it becomes 0.0668V at the 4th filter cap.

The point of what I make is that with the same total amount of R, and
less capacitance, filtration of all F of concern is much better
because of the sectioning, but there is a limit of course to the
effectiveness of sectioning, for effective sectioning the ZC at the
wanted F should be ideally 1/10 of the preceding R in the R&C section.

Try doing the math with 4 sections using 100uF instead of 470uF. At 1Hz
ZC = 1,590 ohms, and 1Hz attenuation is barely 0.8,
and so after 4 sections attenuation is only maybe 0.5 times, so 5V of
1Hz at Cres is 2.5V after 4 sections.
So using 1k and 100uF just doesn't work. To be as good as the 470uF,
you'd need far more sections and a lot more R, and the voltage drop
across the extra R would be huge.

So try 3 sections of 1k2 plus 2x470uF, ie, 1k2 plus 940uF per section.
Each section has 1Hz attenuation = 0.141. # sections gives attenuation
of approximately 0.141 cubed, or 0.0028. So 5V at 1Hz at Cres becomes
14mV at C3.

In practice, you should find this to be plenty. The pole of each section
is 0.141 Hz, so after 3 sections the pole has moved down to around
0.07Hz, and by 1Hz, the rate of attenuation is 3rd order, so that
switching transients conveyed by mains F which contain many F will
severely flattened out. But DC is DC, and slow moving levels will still
get past you filter and the ONLY way to deal with them is with
regulation of some kind.



Beware using simple zener diode based shunt regs close to mic input
stages though. The LF noise of the zener will find its way into signal
paths.
Agreed. I have been looking at the Maida regulator as a means of
eliminating the LF noise *prior* to the normal RC string.

Using a regulator right after the resevoir C is OK and you can then make
RC filters after that to all stages without risk of LF motorboating.

That's the plan.

Always have series R after the Cres and the regulator, or else the shunt
reg or series reg may be attempting the impossible.
Ideally, the Pd in the reg should equal the Pd in the preceding feeding
resistance when the wanted Idc flows.
So if Idc increases, Pd in the pass device becomes lower, and it
survives the heating.

Active regs using SS devices have to be designed with care to prevent
them fusing if shorts from input or output occur to 0V. You should be
able to short the in and out to 0V repeatedly with a bar and not damage
the reg.

Patrick Turner.

And such LF oscillations may not be obvious at first. A PS and amp can
be right on the brink of oscillation at LW and the slightest noise
will become amplified by a the peak in the response if there is one
below 1Hz.

Zeners placed across the second cap in an CRCRCRC filter can reduce LF
content and any noise the zeners generate is less than the noise which
is shunted, and following RC stages filter the noise of all types.
Zeners have higher noise at lowish currents. So if you have a +375V B+
rail and held by 5 x 5watt x 75Vdc rated zeners, heatsink the zeners
with a wrap around strip of Al aor Cu and bolt to a chassis or sink and
allow the pda to be a safe 0.75Watts each, which means you'd have Izener
= 10mA at least. And and in a preamp, the simplest shunt reg that isn't
a simple zener string is to have the string feed a base of an npn bjt
with emitter to 0V and a current limiting R between collector and B+
rail being shunt regged. This shunt eg has much lower output resistance
than a plain zener string, and a much "sharper" threshold of turn on, as
the current in the zeners gets amplified by the bjt. The bjt needs a
high Vce rating, and such bjts have low hfe and a darlington pair is the
best solution, and with a limiting series base resistance and filter cap
at the base to 0V to filter out the zener noise. Such a shunt reg works
only at LF and simply keeps the Vdc stable while your large value
electros do the job on higher F. Shunt regs are good for low current
preamp supplies and screen voltage supplies in power amps and have the
advantage that in the case where the output becomes shorted or over
currented, then the regulator doesn't have any current and survives
while it is the low cost series R in RC section that cops the heat and
fails.

I have used such shunt regs in power amps with choke input supplies, so
that the shunt reg shunts enough anode supply dc current right after
turn on to stop the B+ soaring. As the input stages and output stages
turn on the "bleeder" current of the shunt reg reduces to a low level
enough to reg the B+ to stage 1. So thus the high current in a
permanently connected bleeder resistance is avoided.

Patrick Turner.


Cheers

Ian
Patrick Turner.

Cheers

IAn