I did point out that the proposed thermal noise coupling from the core bulk resistance seemed unlikely to be visible in low impedance transformers, unlike the different circuit conditions of tape head coils
thinking more about the 2nd order possibility of looking for a signal dependant "excess noise" in the core resistance when excited by eddy current from applied signal I think the "look for sidebands" idea may need an intermediate step
when I modulate white noise with a sine I can't see a peak at the sine frequency
this is of course the idea behind spread spectrum modulation - you can't see the "spread" signal frequency as a peak in the fft
you can "self-demodulate" by squaring the modulated noise - then you can see the doubled modulating frequency peak in the rectified waveform
sensitivity of the method is limited by other 2nd order nonlinearities in the system
I really don't expect JN's eddy current excess noise is measurable in laminated "bulk metal" cores -or at least not "separable" from magnetic material permeability nonlinearities
signal level dependance of the noise floor could be looked at (with signal frequency too) but again there are probably difficulties in attributing any "interesting features" in the measurements to specific mechanisms
thinking more about the 2nd order possibility of looking for a signal dependant "excess noise" in the core resistance when excited by eddy current from applied signal I think the "look for sidebands" idea may need an intermediate step
when I modulate white noise with a sine I can't see a peak at the sine frequency
this is of course the idea behind spread spectrum modulation - you can't see the "spread" signal frequency as a peak in the fft
you can "self-demodulate" by squaring the modulated noise - then you can see the doubled modulating frequency peak in the rectified waveform
sensitivity of the method is limited by other 2nd order nonlinearities in the system
I really don't expect JN's eddy current excess noise is measurable in laminated "bulk metal" cores -or at least not "separable" from magnetic material permeability nonlinearities
signal level dependance of the noise floor could be looked at (with signal frequency too) but again there are probably difficulties in attributing any "interesting features" in the measurements to specific mechanisms
Last edited:
I think the sideband thing is a little different. You were modulating the noise, whereas the sine will be rather huge here, with noise superimposed on it.
Do you happen to know of an idiots guide to modulation techniques I could find??
John Curl claims it is audible.
So I believe it is measureable..
Cheers, John
There is a possible snag with JN's proposed test. The silvering could introduce so much eddy current loss that it also markedly changes inductance. You are almost adding a shorted turn. This would make comparison impossible.
Core losses must remain smallish (i.e. not too much different from copper losses) otherwise they are not a small perturbation on normal transformer action but will dominate.
Core losses must remain smallish (i.e. not too much different from copper losses) otherwise they are not a small perturbation on normal transformer action but will dominate.
Excellent. That is why I said I may be correct... It is indeed solewhat like a shorted turn than a distributed element. But still, if there is a difference in unloaded noise, I believe you are correct.
As to shorted turn, it may store bulk energy and return it reactively with loss, which is a confounder as you said..
Cheers, John
Oh, ok. So it's a mic OUTPUT transformer.
Yeah, transformers transform both ways, but mic input trannies and output trannies are designed rather differently from each other (beyond just step-up or step-down).
More germane to trannies for moving coil step-up would be a 1:10 microphone input transformer.
As I said earlier, I'd be happy to provide one for measurement. And I think SY said he had some as well, though not sure if he's got any CineMag step-ups.
se
Hi SY,
The reference point I was referring to is at the input to the RIAA phono preamp, where the levels are like those of an MM cartridge. A single JFET input would not normally provide noise as low as we'd like for an MC cartridge at the 0.2mV level. We want to be in the range of 1 nV/rt Hz or better at that point in the system. Four LSK389 differential pairs should suffice (yes, unipolar and using differential pairs where one loses 3 dB in noise right off the bat compared to one single-ended device). Gate capacitance is on the high side, but impedances are low, and capacitance effects can be mitigated. I would be reluctant to use this with some of those "high-output" MC cartridges whose working impedance can be a rather high.
Cheers,
Bob
Hi Joachim,
I think the sound problems with paralleling are only in the example I cited where we were at the RIAA MM phono preamp input where the driving impedance from an MM cartridge is high. I'm sorry if my example created any confusion.
Cheers,
Bob
For MM cartridges (~5mV), the noise of a single 2SK170, for instance, isn't too high, but increased input capacitance by paralleling devices may be an issue.
For MC cartridges (0.5-0.2mV), the JFET noise posses a greater issue, while increased input capacitance by paralleling and cascoding devices is not a problem.
Forgot that the driver for my sound device has a bug wherein it goes into a -6dB mode (clipping at +-16384) and needs full power cycling to reset. Time for a new one anyway.
Right, look at it this way. The tape head has a DC Rs of 385 Ohms and the eddy losses are easy to see on top of this. That mic tranny was 915 Ohms but the secondary reflected back 20K so the losses only show up well on top of this. Does that sound resonable? It's like Nyquists energy exchange between two resistors.
Hi Joshua,
You're exactly right, and I think that is essentially what I said.
Cheers,
Bob
In order to keep everyone up to date with what we are discussing:
Paralleling devices is been known and used for at least 50 years, probably even more, to reduce input noise.
Kirkwood Rough, a member here, told me recently that it was a military secret in the early '60's. I would not be surprised.
I found the suggestion in an obscure journal called 'Electronics Letters', an I.E.E. publication, (not IEEE) in early 1967. It was a REVELATION!
What it essentially said, was that you could make a VIRTUAL TRANSFORMER by just paralleling devices, be it: tube, bipolar transistor, or fet.
At the time, the best bet for very low impedance sources was bipolar transistors, where paralleling reduced effect of the bulk resistivity of each transistor, reducing the noise 3dB by paralleling 2 devices, 6dB by paralleling 4 devices, etc.
This was also true with tubes and jfets, BUT jfets were very noisy then, and tubes took a lot of real estate and current. Often, a transformer was a better solution, but not always.
In today's case we are paralleling low noise jfets. This is by far the easiest way to lower noise without a transformer. In this example, 4 jfets in parallel would be a 6 dB lowering of noise, not bad for perhaps 1 sq inch of extra space, and some added current. 8 jfets would give us 9dB, and 16 jfets would give us 12dB or the equivalent of an ideal transformer with a gain of 4.
Now, what about input capacitance times Rin? Well, it would go up 16 times, and that is a lot. However, what if we used a X4 transformer? It would also go up 16 times. Therefore, it is a draw, with regards to nonlinear capacitance.
Now where COULD a transformer give a REAL advantage? IF you needed a BALANCED INPUT and you used a single ended source. Then, it would give it to you for free, without making a BALANCED fet driver stage. Tubes are ideal when used this way, for a number of reasons. A single fet would work, but not as well.
Paralleling devices is been known and used for at least 50 years, probably even more, to reduce input noise.
Kirkwood Rough, a member here, told me recently that it was a military secret in the early '60's. I would not be surprised.
I found the suggestion in an obscure journal called 'Electronics Letters', an I.E.E. publication, (not IEEE) in early 1967. It was a REVELATION!
What it essentially said, was that you could make a VIRTUAL TRANSFORMER by just paralleling devices, be it: tube, bipolar transistor, or fet.
At the time, the best bet for very low impedance sources was bipolar transistors, where paralleling reduced effect of the bulk resistivity of each transistor, reducing the noise 3dB by paralleling 2 devices, 6dB by paralleling 4 devices, etc.
This was also true with tubes and jfets, BUT jfets were very noisy then, and tubes took a lot of real estate and current. Often, a transformer was a better solution, but not always.
In today's case we are paralleling low noise jfets. This is by far the easiest way to lower noise without a transformer. In this example, 4 jfets in parallel would be a 6 dB lowering of noise, not bad for perhaps 1 sq inch of extra space, and some added current. 8 jfets would give us 9dB, and 16 jfets would give us 12dB or the equivalent of an ideal transformer with a gain of 4.
Now, what about input capacitance times Rin? Well, it would go up 16 times, and that is a lot. However, what if we used a X4 transformer? It would also go up 16 times. Therefore, it is a draw, with regards to nonlinear capacitance.
Now where COULD a transformer give a REAL advantage? IF you needed a BALANCED INPUT and you used a single ended source. Then, it would give it to you for free, without making a BALANCED fet driver stage. Tubes are ideal when used this way, for a number of reasons. A single fet would work, but not as well.
I remember some years ago seeing a mic preamp design that used a medium-power BJT as the input device (BD139 or BD140, IIRC). The idea was that Rbb is a significant source of noise, and that larger devices tend to have lower Rbb.
Measurements of a number of different devices were done, with interesting results. e.g. BC109, touted as "low noise", turned out to have high Rbb, and unacceptably high noise (in that application).
Anyway, I'm now wondering if in fet-land there may be advantages to using higher powered devices than normal for input stages? It's not just about size, which is taken care of by the paralleling approach.
If one looks beyond small-signal devices, there is a wider range of types of device to choose from - Various flavors of Mosfet, for example, or the new SiC devices Nelson's been playing with. Perhaps some of these would have useful characteristics?
Hi Godfrey,
Bear in mind that most small-signal BJTs that are touted as "low noise" are low noise for high-impedance applications. Such devices are generally characterized by high beta to keep the current noise low. They often have quite high base resistance, so are not suitable as low noise devices in low-impedance circuits.
One of the better BJTs for low noise in low-Z applications is the 2N4403, which is not touted as a low-noise device.
Power FETs will usually be MOSFETs, and they are pretty terrible for noise. The old National NPD5564 monolithic JFET dual matched pair was exceptional, but is no longer available.
Cheers,
Bob
The application of low-Rbb transistors to low-noise [at low impedance] audio was picked up by Ray Bradbury at Zetex. The E-line transistors called up in his DN11 are the same size as TO-92 parts, but with much greater capability.
Note the importance of higher-than-expected collector current, in this example.
The circuit shown is not much use for MC stages, on account of the input capacitor, but if the biasing problem is addressed, there are useful pointers in the note:
www.diodes.com/_files/design_note_pdfs/zetex/dn11.pdf
.
Really? I don't remember the spec, but if memory serves, it wasn't anything outstanding (maybe 10-15 nV/rtHz?). Were they actually better in practice?
Hi Rod,
Actually the first mention of it that I recall was from Marshall Leach in the late 1970's when he published his article on a moving coil preamp. He specifically mentioned the unusually low Rbb of the 2N4401/4403 pair.
Cheers,
Bob
- Status
- Not open for further replies.