You are most likely too pessimistic in this case.
The simmed 200pV/rtHz RTI is equivalent to 2R4 or 1R7 per transistor with the standard Diode Spice model.
Build and measure one, and we'll talk after. There is no Rx equivalent in the Spice model beyond RB, so those 0.2nV/rtHz have a different source, and the noise equivalence with the measured value of Rx=1.5ohm is a likely a pure coincidence.
Those models are perhaps good for switching applications otherwise, as Scott mentioned, only the bonding wire has over 0.1ohms; and three digit precision is simply baloney.
re the Joris MC-HA - the trannies should be matched to better than 10mV. The calc's then show about a 50% variation in Ic or about 60uA per mV at the currents being used. Obviously the tighter the better.
I have a few here. I will measure them next week to see what the spread is at ~1mA.
I have a few here. I will measure them next week to see what the spread is at ~1mA.
Last edited:
ok - that's true on the Ic - thanks for pointing it out. Obviously it would require tighter matching or some sacrifice in noise for ballasting.
re the cart coupling, I have a version in the pipeline that servo's the dVbe across the input device and the bias reference, so you could in theory direct couple the cart, but would still need to decouple the ref side to ground with a large cap for noise purposes (surprised you are worried about caps BTW . . . )
I've seen the JFET thing, played with it and wrote it up in my RIAA article way back - a development of Denis Collin's 'LP797' idea. I know Syn08 also ran with it as well. I'll resurrect that in due course for the compendium.
I see JFETs as a much more flexible solution, with virtually zero unforeseen problems. No need for input coupling electrolytic (or floating supply), can be used for both MM and MC with only a gain adjustment, etc... But then it's ultimately a matter of taste, I guess.
Servoing the input current always comes with a noise penalty, it's the stinkin' thermodinamics, you can balance the DC currents, but you cannot avoid injecting uncorrelated current noise with the DC current error signal. Input bias cancellation is impossible to make with any decent close to zero tempco in a discrete implementation (been there, tried that). Makes more sense to live with 1uA of cartridge current, but of course this may be outrageous for the purists.
What I find funny about any bipolar MC cartridge amp is that not even an input big film cap (like 100uF or more) would protect the cartridge; chances are, at power up/down, the cartridge gets a big current shock anyway. Of course, one can go as far as implementing power sequencing, interlock and delay the MC cartridge connect, etc... but this would make the whole setup indeed of an obscene complexity, only for the luxury of using bipolars?!
Oh, and low noise JFETs are still one order of magnitude cheaper than these ZTX devices .
Last edited:
I'm not fussed about current flowing in the coils. The wont burn out and it wont 'dampen' anything . Plus I wanted to be contrary
Neither am I, but we are in a minority .
The most common FUD story I heard is about the MC DC current magnetizing the core and then an imbalance in the channels levels, and/or a degradation of channel separation results. Clearly audible, of course.
Last edited:
Another point, some of these circuits run bi-polars into saturation. It might not matter but the distortion numbers from sims might not be very accurate.
We built a version of one of these on a test wafer (our first CB process) as a favor to a professor at MIT. It did not work do to quasi-sat IIRC.
Which brings up the point, since the researchers at Delft derived the optimization process to minimize rbb long ago why have none of the specialty houses gone past the 1nV type of discrete NPN/PNP's?
We built a version of one of these on a test wafer (our first CB process) as a favor to a professor at MIT. It did not work do to quasi-sat IIRC.
Which brings up the point, since the researchers at Delft derived the optimization process to minimize rbb long ago why have none of the specialty houses gone past the 1nV type of discrete NPN/PNP's?
The most common FUD story I heard is about the MC DC current magnetizing the core and then an imbalance in the channels levels, and/or a degradation of channel separation results. Clearly audible, of course.
We've avoided so far the fantasy that a cartridge motor/cantilever assembly is reciprocal.
The JFET circuits are nice and with a gm helper seem quite capable of decent performance with only 2 active devices. They are not as quiet as the bipolar designs. I've built 2 of them and they measure well and sound good. You do need to be prepared to deal with higher supply currents or use an opamp buffer arrangement (more complexity) as in the Joris MC-HA above. All the sims I did with single devices showed env of c. 1.2-1.5 nV/rt Hz which is of the same order as a 797 opamp (quite a few MC-HA's using those around now as well)
Which brings up the point, since the researchers at Delft derived the optimization process to minimize rbb long ago why have none of the specialty houses gone past the 1nV type of discrete NPN/PNP's?
Niche market, no relevant business case, I guess.
There's one running in Texas at the moment. What measurement are you interested in?
Exact schematic plus voltage and current noise as a function of frequency. But I guess I could dead bug one, as soon as I place my next DigiKey order and get the damn ZTXs .
Last edited:
All the sims I did with single devices showed env of c. 1.2-1.5 nV/rt Hz which is of the same order as a 797 opamp (quite a few MC-HA's using those around now as well)
A 2SK3557 is 20mS at 3mA so 10 in parallel off the same strip gets to theoretical ~.25nV. Adding the 1 Ohm feedback network and you should still be under .3nV. Cost $3.25
You do flirt with making an oscillator, but that is easy to work around.
Last edited:
The internals of a spice model are a box of Pandorra to me, no idea how this data is transformed into noise.Build and measure one, and we'll talk after. There is no Rx equivalent in the Spice model beyond RB, so those 0.2nV/rtHz have a different source, and the noise equivalence with the measured value of Rx=1.5ohm is a likely a pure coincidence.
Those models are perhaps good for switching applications otherwise, as Scott mentioned, only the bonding wire has over 0.1ohms; and three digit precision is simply baloney.
For me only the results are counting and they are very close to built and measured amps, let it be coincidence.
I therefore find this model in the mA range an excellent and very useful help.
It could very well be that for much higher currents results are unusable, but that’s no issue here.
Let’s be happy with what we have.
Hans
The internals of a spice model are a box of Pandorra to me, no idea how this data is transformed into noise.
For me only the results are counting and they are very close to built and measured amps, let it be coincidence.
I therefore find this model in the mA range an excellent and very useful help.
It could very well be that for much higher currents results are unusable, but that’s no issue here.
Let’s be happy with what we have.
I strongly disagree. Bad models is GIGO.
Have you built and measured an amp with the ZTXs? Can you share some results?
since the researchers at Delft derived the optimization process to minimize rbb long ago
Time frame and names Scott?
George
A 2SK3557 is 20mS at 3mA so 10 in parallel off the same strip gets to theoretical ~.25nV. Adding the 1 Ohm feedback network and you should still be under .3nV. Cost $3.25
You do flirt with making an oscillator, but that is easy to work around.
8 in parallel @10mA each plus an 1ohm common source resistor takes you around 0.4nV/rtHz flat (measured) and 3-400Hz noise corner frequency, a smidge worse than the now extinct BF862 (around 0.35nV/rtHz)
Same @1mA each are 0.5nV/rtHz and can be fed from two batteries. I’ve posted the schematics and Gerbers here and it is actually built by multiple members.
The internals of a spice model are a box of Pandorra to me, no idea how this data is transformed into noise.
For the JFET models the internal parasitic R and C are switched between the drain and source on the P devices, I doubt they fixed it.
Time frame and names Scott?
George
Albert Willemsen and Niko Bel - 1980 JSSC Vol. SC-15 No. 2
Total ENR of a single device 1.4 Ohms. I have two pairs waiting now 40yr. for when I buy a $$$ Ortophon cartridge.
EDIT - BTW they treat all sources of noise intrinsic and extrinsic RB and RE as well as interconnect.
Last edited:
How is the Lee circuit different from the Leach ?
Moving Coil Cartridge Head Amps
And the JFET version is simpler ?
Zen -> Cen -> Sen, evolution of a minimalistic IV Converter
Patrick
Moving Coil Cartridge Head Amps
And the JFET version is simpler ?
Zen -> Cen -> Sen, evolution of a minimalistic IV Converter
Patrick
- Home
- Source & Line
- Analogue Source
- Richard Lee's Ultra low Noise MC Head Amp