I am suitably reprimanded Mark. I'm sinking even further into beach bum mode and haven't been active on any forum for years. If you have a direct link to Wayne's detailed measurements, please post them.
I'm not sure a supa dupa design with 2sk170 bla bla will outperform an optimised NJM 2068 in noise, or even a vintage Mullard/Philips NE5532 hand made by Southampton virgins, ... optimised circuit and real MM cartridge. If you have one, please show measurements, preferably noise spectrum plots. Wayne may have one such beast.
Alas, 5532s are nowhere as good as the originals which ALL easily met the higher 5534AN spec. I must have measured at least a 100 myself in my time at Calrec.
I'm not sure a supa dupa design with 2sk170 bla bla will outperform an optimised NJM 2068 in noise, or even a vintage Mullard/Philips NE5532 hand made by Southampton virgins, ... optimised circuit and real MM cartridge. If you have one, please show measurements, preferably noise spectrum plots. Wayne may have one such beast.
Alas, 5532s are nowhere as good as the originals which ALL easily met the higher 5534AN spec. I must have measured at least a 100 myself in my time at Calrec.
The reason I mentioned trying the 5087 is the graphs indicated to me the 5087 would do slightly better at 47K, being able to achieve a NF of .5db whereas the 4403 only achieves a 2db NF.
Like I wrote to Mark, the noise figure with 47 kΩ source is only relevant when you are interested in the noise with open input, otherwise it is mostly the cartridge that determines the impedance. 12 kΩ would be a reasonable estimate for the RIAA- and A-weighted average impedance of a 500 mH cartridge.
(If you want to know where that 12 kΩ comes from, see "Noise and moving-magnet cartridges", Electronics World October 2003, pages 38...43, https://worldradiohistory.com/UK/Wireless-World/00s/Electronics-World-2003-10-S-OCR.pdf Mind you, Electronics World drew one of the sections of the gain switch in the wrong state in figure 5 and I mixed up the terms spectral density and power spectral density.)
Last edited:
A very exciting examination of the topic. Unfortunately, I can't contribute any more.
Many thanks,
HBt.
Many thanks,
HBt.
And I am back to the sparse datasheets I have. The 5087 can achieve .5db NF from around 8K to 70K by varying Ic at 100Hz 1Hz BW and an even broader range at 1KHz. At 12Kohm it looks like the best the 4403 can do is 1.5db. I wish my datasheet for the 4403 had the constant NF graph that the 5087 had. The 5087 sheet also has a wideband graph for NF. It would be much simpler. My read of the two datasheets says the 4403 is better if source R is less than maybe 2K while the 5087 is better at source impedance greater. Small potatoes though of maybe 1db difference. And of course, 47K is not the source impedance. As you say, add a cart and it will go down significantly. Easy to show, I've never seen a phono pre that noise did not go down after connecting the cart.Like I wrote to Mark, the noise figure with 47 kΩ source is only relevant when you are interested in the noise with open input, otherwise it is mostly the cartridge that determines the impedance. 12 kΩ would be a reasonable estimate for the RIAA- and A-weighted average impedance of a 500 mH cartridge.
(If you want to know where that 12 kΩ comes from, see "Noise and moving-magnet cartridges", Electronics World October 2003, pages 38...43, https://worldradiohistory.com/UK/Wireless-World/00s/Electronics-World-2003-10-S-OCR.pdf Mind you, Electronics World drew one of the sections of the gain switch in the wrong state in figure 5 and I mixed up the terms spectral density and power spectral density.)
And note to kgrlee, 2068 is not yet unobtainium. Picked some up from digikey recently, and they are showing almost 6000 in stock now. The DIP is marked as not for new designs though.
The 4401/4403 were intended for high speed switching. This forced a low R_bb process. NF and 1/f were not key production parameters, so what you get is a lottery between brands.
The Ferranti/Zetex ZTX... transistors were good, low R_bb for high currents and a process that had decent gain at low currents.
The Ferranti/Zetex ZTX... transistors were good, low R_bb for high currents and a process that had decent gain at low currents.
I put together the little perl script attached below, to find transistor parameters which exactly match the NE5534A's datasheet "typical" noise specifications. For low current noise, you need high beta. For low voltage noise, you need low r_bb. But exactly how much of each? The plot below may help.
Why did they force you to learn to code software in school? To enable you to write little quick-and-dirty programs like this one, long after you finish school.
_
Code:
#
# same_as_5534A.pl written 17 Dec 2023 mgj
#
# for each Beta, solve for required r_bb that gives
# same noise current and same noise voltage as 5534A
#
$KBOLTZ = 1.38065E-23 ;
$QELECTRON = 1.6022E-19 ;
$ROOMTEMP = 3.0E+2;
$REQUIRED_VN = 3.5E-9 ;
$REQUIRED_IN = 0.4E-12 ;
for($beta=80.0; $beta<=800.0; $beta+=10.0) {
# noise current and beta determine I_collector
$temp = ($REQUIRED_IN) * ($REQUIRED_IN);
$i_collector = $temp * $beta / (2.0 * $QELECTRON) ;
# noise voltage determines r_bb
$temp = ($REQUIRED_VN) * ($REQUIRED_VN);
$lhs = $temp / (8.0 * $KBOLTZ * $ROOMTEMP) ;
$ratio = ($KBOLTZ * $ROOMTEMP) /
(2.0 * $QELECTRON * $i_collector) ;
$r_bb = $lhs - $ratio ;
printf "%.0f , %.2f , %.3le\n", $beta, $r_bb, $i_collector
}
Why did they force you to learn to code software in school? To enable you to write little quick-and-dirty programs like this one, long after you finish school.
_
Attachments
Last edited:
Hi-five for programming! My main hobby is programming.
The article is very informative.
BTW, I use 2N5210/2N5087 in my MM phono pre-amp. My experience is that MM pre-amp noise is far too low to worry about the last decibel. Even record surface noise is below audibility at my typical listening levels.
Ed
The article is very informative.
BTW, I use 2N5210/2N5087 in my MM phono pre-amp. My experience is that MM pre-amp noise is far too low to worry about the last decibel. Even record surface noise is below audibility at my typical listening levels.
Ed
This thread is a nice look back at John Curls work. Glad the transistors are still available.
Then just measure the ones you buy next timeYes for those transistors measured hfe was high, but that does not tell you what you get next time you buy.
While MPS8099/8599 are listed as obsolete, they come up well in the AoE3 listing. Anyone who wants a dozen of each, will part with for the postage.
I have a reel of each
I have a reel of each
If we use the cute simplification from Marcel's WW article {linked in post #22 above}, the MM cartridge can be modeled as a 12K resistor.
Then for the NE5534A: TotalNoise = SQRT( (3.5E-9)^2 + ((4E-13 * (12K//47K))^2 ) = 5.18 nV/rtHz
Now let's use a differential pair of Rohm 2SD2226KT146W transistors (in stock today at both Mouser and DigiKey), biased at 120 microamps per transistor. Datasheet Beta for the "rank W" part is 1200 to 2700 but we'll assume 1200. The Spice model on Rohm's website says r_bb = 16.7 ohms and we'll round up to 17. Then for the Rohm device
Vnoise = 2.03nV/rtHz and Inoise = 0.155pA/rtHz .
Thus for the Rohm discretes: TotalNoise = SQRT( (2.03E-9)^2 + ((1.55E-13 * (12K//47K))^2 ) = 2.75 nV/rtHz
Quite an improvement.
_
Then for the NE5534A: TotalNoise = SQRT( (3.5E-9)^2 + ((4E-13 * (12K//47K))^2 ) = 5.18 nV/rtHz
Now let's use a differential pair of Rohm 2SD2226KT146W transistors (in stock today at both Mouser and DigiKey), biased at 120 microamps per transistor. Datasheet Beta for the "rank W" part is 1200 to 2700 but we'll assume 1200. The Spice model on Rohm's website says r_bb = 16.7 ohms and we'll round up to 17. Then for the Rohm device
Vnoise = 2.03nV/rtHz and Inoise = 0.155pA/rtHz .
Thus for the Rohm discretes: TotalNoise = SQRT( (2.03E-9)^2 + ((1.55E-13 * (12K//47K))^2 ) = 2.75 nV/rtHz
Quite an improvement.
Code:
*
* Q2SD2226K NPN BJT model
* Date: 2006/12/04
.MODEL Q2SD2226K NPN
+ IS=400.00E-15
+ BF=1.3544E3
+ VAF=69.931
+ IKF=.16193
+ ISE=400.02E-15
+ NE=1.6742
+ BR=9.6205
+ VAR=100
+ IKR=1.1285
+ ISC=725.25E-15
+ NC=2.0260
+ NK=.74645
+ RE=.3
+ RB=16.704
+ RC=.12233
+ CJE=27.484E-12
+ MJE=.37516
+ CJC=6.5540E-12
+ MJC=.31941
+ TF=474.50E-12
+ XTF=38.562
+ VTF=63.249
+ ITF=2.3187
+ TR=79.789E-9
+ XTB=1.5000
_
Attachments
Last edited:
What about darlington's? Beta would certainly improve, but how would r_bb stack up? Could you combine a device with low r_bb and medium gain with one that has only medium r_bb but high beta? Any other downsides?
The first transistor of the Darlington would determine the base shot noise and both of them would contribute to the equivalent input noise voltage. If you biased the second transistor of the Darlington pair at a much larger collector current than the first, and if the base resistance of the second transistor would be low enough, you would roughly end up with the noise performance of the first transistor of the Darlington pair.
If we use the cute simplification from Marcel's WW article {linked in post #22 above}, the MM cartridge can be modeled as a 12K resistor.
Then for the NE5534A: TotalNoise = SQRT( (3.5E-9)^2 + ((4E-13 * (12K//47K))^2 ) = 5.18 nV/rtHz
Now let's use a differential pair of Rohm 2SD2226KT146W transistors (in stock today at both Mouser and DigiKey), biased at 120 microamps per transistor. Datasheet Beta for the "rank W" part is 1200 to 2700 but we'll assume 1200. The Spice model on Rohm's website says r_bb = 16.7 ohms and we'll round up to 17. Then for the Rohm device
Vnoise = 2.03nV/rtHz and Inoise = 0.155pA/rtHz .
Thus for the Rohm discretes: TotalNoise = SQRT( (2.03E-9)^2 + ((1.55E-13 * (12K//47K))^2 ) = 2.75 nV/rtHz
Quite an improvement.
Code:* * Q2SD2226K NPN BJT model * Date: 2006/12/04 .MODEL Q2SD2226K NPN + IS=400.00E-15 + BF=1.3544E3 + VAF=69.931 + IKF=.16193 + ISE=400.02E-15 + NE=1.6742 + BR=9.6205 + VAR=100 + IKR=1.1285 + ISC=725.25E-15 + NC=2.0260 + NK=.74645 + RE=.3 + RB=16.704 + RC=.12233 + CJE=27.484E-12 + MJE=.37516 + CJC=6.5540E-12 + MJC=.31941 + TF=474.50E-12 + XTF=38.562 + VTF=63.249 + ITF=2.3187 + TR=79.789E-9 + XTB=1.5000
_
Interesting, I never knew you could buy discrete transistors with such high hFE anywhere, and they even have reasonable breakthrough voltages (even a remarkably high emitter-base breakthrough voltage) and a low base resistance, if you can trust the Spice model. It's also a good sign that hFE doesn't roll off quickly at low currents; according to M. Koolen and J. Aerts, "The influence of non-ideal base current on 1/f noise behaviour of bipolar transistors", Proceedings on Bipolar Circuits and Technology Meeting, 1990, https://www.nxp.com/wcm_documents/models/bipolar-models/mextram/bctm90koolen.pdf , the non-ideal base current that makes the hFE roll off has a lot of 1/f noise associated with it. Apparently their only disadvantage is the very low Early voltage, but you can always cascode them if needed.
I usually neglect the effect of the 47 kΩ on the source impedance that the input transistor or valve sees, because the 12 kΩ from the cartridge is mostly reactive. I later calculated that you can account for it by calculating the impedance of the cartridge and resistor at 3852 Hz. If the cartridge had a frequency-independent ESR and inductance (which it hasn't), a 1 kΩ and 500 mH cartridge would have an impedance of 12142.66209 Ω at 3852 Hz without 47 kΩ termination and 11528.93936 Ω at 3852 Hz with 47 kΩ termination.
Is anyone going to do a noise spectrum plot of da various implementations of LN MM preamps WITH A REAL LIFE CARTRIDGE? Especially of Marcel's supa dupa version.
A single number is practically meaningless (contrary to da proponents of various weightings) MM noise is complex, due to the varying Rs and the effect of RIAA EQ. I don't think using a single 12k resistor is meaningful except to get yet another meaningless number.
A good 'conventional' (ie not Marcel's supa version) preamp with MM cartridge shows a somewhat 'whiter than pink' noise spectrum above 1kHz. A good MC cartridge with SOTA noise preamp will be 'redder than pink' above 1kHz and this is less objectionable. Common MC cartridges with SOTA preamps will also show significantly smaller 'meaningless' numbers cos they usually have high power output to go with their low voltage output
But I want to see the noise spectrum of Marcel's beast with a good MM cartridge 🙂
A single number is practically meaningless (contrary to da proponents of various weightings) MM noise is complex, due to the varying Rs and the effect of RIAA EQ. I don't think using a single 12k resistor is meaningful except to get yet another meaningless number.
A good 'conventional' (ie not Marcel's supa version) preamp with MM cartridge shows a somewhat 'whiter than pink' noise spectrum above 1kHz. A good MC cartridge with SOTA noise preamp will be 'redder than pink' above 1kHz and this is less objectionable. Common MC cartridges with SOTA preamps will also show significantly smaller 'meaningless' numbers cos they usually have high power output to go with their low voltage output
But I want to see the noise spectrum of Marcel's beast with a good MM cartridge 🙂
It's not a 12 kohm resistor but a 12 kohm mostly inductive reactance. A resistor would produce much more thermal noise of its own. Then again, record surface noise is not included in my analysis at all, so it's all about the noise level in between records rather than with a record playing.
Anyway, if you dislike high-frequency noise more than the common noise weighting graphs indicate, that's an extra reason to design for low equivalent input noise current. The transfer from the noise current to the output includes the cartridge impedance that increases with frequency. The best you can then do is using a JFET input stage and either an electrically cold input resistance or Nick Sukhov's non-standard input termination impedance (he uses far higher resistances and far lower capacitances than recommended by the cartridge manufacturers, see https://www.diyaudio.com/community/...ono-stage-85-dba-sn-ratio.387375/post-7051721 ).
By the way, that 3852 Hz becomes 5179 Hz with ITU-R 468- rather than A-weighting.
Anyway, if you dislike high-frequency noise more than the common noise weighting graphs indicate, that's an extra reason to design for low equivalent input noise current. The transfer from the noise current to the output includes the cartridge impedance that increases with frequency. The best you can then do is using a JFET input stage and either an electrically cold input resistance or Nick Sukhov's non-standard input termination impedance (he uses far higher resistances and far lower capacitances than recommended by the cartridge manufacturers, see https://www.diyaudio.com/community/...ono-stage-85-dba-sn-ratio.387375/post-7051721 ).
By the way, that 3852 Hz becomes 5179 Hz with ITU-R 468- rather than A-weighting.
High Vceo and Vebo dont normally go with high Hfe'. This looks a lot like a darlington. Compare with the KSP13Interesting, I never knew you could buy discrete transistors with such high hFE anywhere, and they even have reasonable breakthrough voltages (even a remarkably high emitter-base breakthrough voltage) and a low base resistance, if you can trust the Spice model. It's also a good sign that hFE doesn't roll off quickly at low currents; according to M. Koolen and J. Aerts, "The influence of non-ideal base current on 1/f noise behaviour of bipolar transistors", Proceedings on Bipolar Circuits and Technology Meeting, 1990, https://www.nxp.com/wcm_documents/models/bipolar-models/mextram/bctm90koolen.pdf , the non-ideal base current that makes the hFE roll off has a lot of 1/f noise associated with it. Apparently their only disadvantage is the very low Early voltage, but you can always cascode them if needed.
The base-emitter voltage and collector-emitter saturation voltage graphs don't look like those of a Darlington.
I agree that Vcesat looks like a single, darlingtons don't go much below 0.7VThe base-emitter voltage and collector-emitter saturation voltage graphs don't look like those of a Darlington.
This must be a very strange structure with heavily doped base and lightly doped emitter to get such a high Vebo. The vast majority of diffused transistors are around 7V, RF devices even lower.
Is anyone going to do a noise spectrum plot of da various implementations of LN MM preamps WITH A REAL LIFE CARTRIDGE? Especially of Marcel's supa dupa version.
But I want to see the noise spectrum of Marcel's beast with a good MM cartridge 🙂
You can perform the experiments yourself, of course. But consider this: perhaps nobody else wants to see that, except you. If you do nothing while hoping someone else might eventually do the work, it could turn out to be a very long wait.
- Home
- Source & Line
- Analogue Source
- Discrete 2N4403 from 1976, same noise performance in Moving Magnet RIAA preamp, as NE5534A. And better performance than the dual (5532A)