Make that 1.4nV/rthz, raise the JFET corner frequency at about 500Hz and we may agree on this as well
Best JFETs are at 1nV/rthz and a diff stage will always degrade the noise by 3dB.BTW, 1/f noise always decreases at 10dB/decade, to the corner frequency, independent of the device type, otherwise it's not 1/f noise. If that's minor or not, it depends on the application.
You also forgot to mention the very low frequency corner of bipolars.
Actually, the 1K input noise of the Toshiba 2SK170 and its complement the 2SJ74, averages out to be about 0.8nV/rt Hz at 10ma.
If you both series these parts and then parallel them for quad comp differential operation, the final IDEAL noise is about 0.8nV/rt Hz. However adding bias and gain set resistors, can move the noise to 1nV/rt Hz in best case.
This is comparable to the AD797 with a realistic feedback resistor used, but the fet version will hold this noise over a large impedance range. The 797 will not.
It is true that the 1/f noise can come in somewhat earlier than 100 Hz, but not to any audible extent, as the ear becomes insensitive to noise below 700 Hz at a rapid rate. The spot noise of the input stage might get to 2-3nV/rt Hz at 10 Hz, but that is not audible without extensive bass boost, above the 6dB/oct. that you might expect from RIAA or analog tape EQ.
If you both series these parts and then parallel them for quad comp differential operation, the final IDEAL noise is about 0.8nV/rt Hz. However adding bias and gain set resistors, can move the noise to 1nV/rt Hz in best case.
This is comparable to the AD797 with a realistic feedback resistor used, but the fet version will hold this noise over a large impedance range. The 797 will not.
It is true that the 1/f noise can come in somewhat earlier than 100 Hz, but not to any audible extent, as the ear becomes insensitive to noise below 700 Hz at a rapid rate. The spot noise of the input stage might get to 2-3nV/rt Hz at 10 Hz, but that is not audible without extensive bass boost, above the 6dB/oct. that you might expect from RIAA or analog tape EQ.
Last edited:
Now you are talking about paralleling JFETs
So you are replacing one JFET with four JFETs in differential-complementary configuration, for the same noise.
Nice design decision
I don't understand what you're saying. Where should the additional phase lag come from, compared to a "traditional" input diff pair? From cob of the "biasing transistors" in the diamond buffer? But they are there also in the noninverting input.
One pair of emitter followers in non-inverting input VS 2 pairs in inverting one. In non-inverting input there are no 2'nd (output) pair of followers; it is a common emitter stage, the same stage is a common base stage for an inverting input. Don't compare to "traditional" input diff pair, compare what you see on the picture.
Anyway, I have to repeat that the picture looks nice and very originally. I like it. Work of art.
It depends on what you are driving. I am always driving jfet input power amps, so 1K drive impedance is not too high.
However, cables are a different story, and though many here would scoff at any cable differences, it is possible that the connecting cable between the preamp and the power amp could cause a problem with a 1K drive impedance. It is a matter of trade-offs. Low Z is good, but look at the extra cost and compromise of adding an extra stage?
It so happens that the Audible Illusions line preamp has an effective output impedance of 1800 ohms or so, and it is successful in the marketplace, so 1K is not too far, out of line.
However, cables are a different story, and though many here would scoff at any cable differences, it is possible that the connecting cable between the preamp and the power amp could cause a problem with a 1K drive impedance. It is a matter of trade-offs. Low Z is good, but look at the extra cost and compromise of adding an extra stage?
It so happens that the Audible Illusions line preamp has an effective output impedance of 1800 ohms or so, and it is successful in the marketplace, so 1K is not too far, out of line.
Typical for JC running out of arguments.
Using differential configurations in low noise designs (like RIAA, analog tape EQ) is nonsense.
It eliminates DC offset of a single device. I am referring in my design examples with differential input. That is difficult to do with a single device, or even pairs of complementary devices like the Vendetta Research SCP-1 or SCP-2. This makes differential pairs necessary. With differential in and out system, like the Parasound JC-2 or the CTC Blowtorch (in principle) differential pairs are mandatory. Of course, for single ended in and out, complementary pairs, or even a single device is possible.
For some reason you think that you have come up with a new way to look at input stages, but I was doing it 25 years ago, and selling existing products with complementary paralleled matched jfet pairs all over the world, at that time.
Bringing that up here, just confuses the situation.
First, THINK DIFFERENTIAL PAIR.
That is the topic here.
For some reason you think that you have come up with a new way to look at input stages, but I was doing it 25 years ago, and selling existing products with complementary paralleled matched jfet pairs all over the world, at that time.
Bringing that up here, just confuses the situation.
First, THINK DIFFERENTIAL PAIR.
That is the topic here.
The point is, which peace of which published (or not) picture is more optimal for certain application. Syn08 says that john's picture is not. Loss of vital parameters in order to decrease DC offset... oh my God... NO COMPROMISE DESIGN?
Last edited:
The point is that the BT gain stage is a primitive, suboptimal design from almost whatever perspective you are looking at. It is an excellent proof on how objectively poor our hearing is and how much marketing hype is in the high end audio industry.
No more than a tiger is the equivalent of a camel. LT1362 is elegant, it is completely different from typical opamps derived from vacuum tube analog computers. Your trick is in eliminating of current sources in tails of traditional diffpairs, while LT's trick is in creative usage of diamond buffers.
Single JFET is primitive, with regards of cost/noise BT gain stage isn't optimal,
With regards linearity it is much linear than single device, with regards output current it doesn't exhibit symmetrical saturation as differential stage and asymmetrical saturation as single device. Intrinsic linearity of this stage allow you to use your loop gain to improve something else.
I don't understand how something optimal can be more primitive, while something suboptimal can be less primitive. It is beyond any logic, Dimitri.
Dimitri and PMA, you both know and understand the tradeoffs in differential circuit design.
Actually, even a single N channel differential input stage is better than the single sided complementary or complementary differential input stage in one MAJOR way. Lower input capacitance. This could be very important with 1H inductance sources such as pro tape heads, or some moving magnet cartridges, especially ones made in the past. In this case complementary design is worse, because the nonlinear input capacitance is perhaps 4 times the N channel only equivalent. That is why I used single differential input stages with a topology similar to what Bob Cordell uses, but much simpler, for magnetic tape reproduction EQ circuits for both Mobile Fidelity (1979), Wilson Audio (1983), and the phono input stage for both MC and MM for the JC-80 phono stage (1981). In these cases, lowish input capacitance was necessary to reduce distortion.
Actually, even a single N channel differential input stage is better than the single sided complementary or complementary differential input stage in one MAJOR way. Lower input capacitance. This could be very important with 1H inductance sources such as pro tape heads, or some moving magnet cartridges, especially ones made in the past. In this case complementary design is worse, because the nonlinear input capacitance is perhaps 4 times the N channel only equivalent. That is why I used single differential input stages with a topology similar to what Bob Cordell uses, but much simpler, for magnetic tape reproduction EQ circuits for both Mobile Fidelity (1979), Wilson Audio (1983), and the phono input stage for both MC and MM for the JC-80 phono stage (1981). In these cases, lowish input capacitance was necessary to reduce distortion.
I think you should look closer.
Both inputs have 4 transistors.
All common collector. In fact, there are two diamond buffers (unity gain buffers) and in between the 500 Ohm resistor.
And with JFET's, you need only 4 active devices for the same thing.
Tino
- Status
- Not open for further replies.