Everybody seems to complain how JFETs are hard to bias and therefore a pain to build circuits around, especially cheap, mass produced ones, which can’t afford matched components or manual adjustment. I think there may be a reasonable way out, and I’m not talking about using an op amp bias servo, which is akin to killing a fly with a sledgehammer.

The circuit below, which is a derivative of the input stage and phase splitter from Pimped Alice, uses a single NPN transistor and a few passives to automate what the 1Meg potentiometer does manually in the classic Schoeps design (and Pimped Alice): set the gate voltage for the target drain current, but using a Miller Integrator.



Here's a simplified version of what's going on at DC, which somewhat resembles a two transistor NPN current source:


R3 sets ID (approx. 0.6V/0.47K = 1.27mA), R2 sees practically the same current (the BJT base current is a couple hundred nA), so (R2+R3)*ID sets the voltage at the JFET source. So these two resistors set the Q-point of the JFET, and most importantly their values are independent of the JFET parameters!?

To bias correctly in this configuration, the JFET only has to have:

• IDSS higher than the bias point (target IDS for this circuit is about 1.3mA, so any JFET having datasheet “min. IDSS” above 2mA is good)
• |VGSoff| lower than the target voltage at the source of the JFET (3.1V for the schematic below) minus the BJT’s VCEsat, so let’s say any JFET with datasheet “max. |VGSoff|” less than 2.8V.

These requirements are not strict at all, and certainly do not require measuring, sorting and matching individual parts.

The DC circuit is simple enough to build an intuitive understanding of what’s going on. Qualitatively, there is a tug-of-war between the JFET and the BJT, which settles at a “happy medium”. If the BJT were to conduct more than this “happy medium”, we’d have a high collector current and a high VBE (and equivalently a high current across R3). The high collector current would produce a high voltage drop across R1 and pull the JFET gate low, while the high current across R3 would push the source of the JFET to a high voltage. VGS would go more negative, reducing the JFET IDS, and therefore the current through R3 and consequently VBE, moving back to the “happy medium”. Conversely, if the BJT were to conduct less, VGS would be more positive, causing the JFET to conduct more, raising VBE and moving back to the “happy medium”.

The bias point is as reliable as any other circuit relying on a VBE as a reference (which is a large proportion of BJT circuits). As far as I can tell, thermal stability is the main drawback of the circuit: a variation of about +/- 20% in drain current over -25C to 85C, vs +/- 5% for a potentiometer-biased gate. Over a more realistic usage range of -10C to 50C, the NPN-based bias produces a variation of +/-10% in drain current. I think this is perfectly acceptable for the condenser (or electret) head amp application.

It gets slightly more complicated when we want an AC circuit for a head amp (without turning the thing into an oscillator). The answer is using the BJT as a Miller Integrator, which “amplifies” the time constant R6*C4, effectively limiting the gate voltage adjustment to very low frequencies.

Based on simulations with varying JFETs, BJTs, capacitances, temperatures, input signal amplitude (including dynamically switching from 0V to 2V), this seems reliable enough to build.

I know the devil’s in the details, but is there a significant drawback that I’m missing here? I plan to breadboard it when I have some time.

Haha, you had exactly the same thoughts and ideas as I had using a BJT to set the bias point! My circuit works quite similarly, though with a slight twist. And I'm not sure yet whether I want to share the schematics yet, as I intended to commercialize the idea somehow. On the other hand, I have several other circuit ideas that might be even better for that purpose. I'll think about it today and if I decide to share the circuit, I'll do that tonight when I have more time. Not to steal your thread, don't get me wrong. Just to supplement on it. Funny that sometimes people at different places on the world have the same ideas at the same time.?

Off to work now...

Jan

I'm not sure yet whether I want to share the schematics yet, as I intended to commercialize the idea somehow.

I had the same hesitation, but because I don't have any real contacts in this area, and from what I can tell any such improvement is likely to get copied by low cost manufacturers in less than 12 months (so any commercial advantage of the first implementer will fade quickly), I decided to just publish it.

Not to steal your thread, don't get me wrong. Just to supplement on it.

No issue here, I'd be ecstatic if this turns into a debate of pros and cons of different topologies, even if mine turns out in the "needs improvement" category!

开云体育

There are many ways to skin this proverbial cat.
One is given there:

It has the added advantage of minimizing output current offset.

I don't doubt there are endless variations on the theme.

On Tue, Jul 2, 2024 at 11:28 AM, sergio_logic wrote:

is there a significant drawback that I’m missing here?

It looks interesting. I have the following questions:

• One of the biggest hurdles to making FET impedance converters is the scarcity of suitably good FET transistors. Does this circuit/technique help to make more FETs "suitable"?
• Being a DIY group, adjusting the trimpot to set the bias on a pimped Alice board isn't a big drawback for most of us. Does the servo approach offer other advantages, like superior audio performance?
• Conversely... would the active servo itself be a potential source of additional noise?

Anyway, this looks intriguing, and something that could be beneficial in a mass-produced generic head amp for electret and condenser mics. I look forward to your results when you test some prototypes.

开云体育

You just hit the nail on the head with availability of good FETs. TI has a really nice one that is extremely low noise but is SMD is a weird package?

OK, so I decided to publish my impedance converter circuit as well. As depicted here, it operates from 16V. It looks quite similar to Sergio's circuit, doesn't it? However, in this circuit, there's also AC feedback from the output to the JFET Gate, which makes this essentially a charge amplifier, like the KM84. The circuit, and some other circuit features not depicted here, resolve several of the KM84 shortcomings, so I named it KM84+++ (I also designed a KM84+ and KM84++ btw, which stay closer to the original KM84 design). Output can be a single-ended, transformerless impedance-balanced circuit, or through a 1:1 transformer. I designed a PCB for the Takstar CM-63 with the Lundahl LL1968 transformer. PCB is ready and waiting to be populated. If I'm satisfied with the results, I will make the PCB and complete schematics available through PCBWAY.

Vbe of T2 and R16 define the JFET bias current. R2 provides DC feedback, like in Sergio's circuit. I had ~3.5mA bias current available for this circuit, which must be shared between JFET and BJT. In the design depicted here, they have approximately the same current, but you can play with other current distributions, depending on the JFET, noise requirements, and output impedance requirements.

This circuit enables the use of more different JFET types, because it depends less on high gm values than e.g. a KM84 circuit does. Not having to adjust the bias may not be such a huge benefit for the average DIY, but I've read enough posts from people who do seem to experience troubles. Small trimmers are also not very reliable and take a lot of board space. "But what's that P1 doing on your PCB?", I hear you saying. Well, that's optional trim pot to adjust the polarization voltage, so you can easily match the gain between two mics. It's not for JFET bias adjustment.

Another advantage of this circuit, or charge amplifiers in general, is that you can select the closed-loop circuit gain within a reasonable range by changing the value of C2. With Schoeps, it is more or less fixed and depends on the JFET chosen. Low-end roll-off is defined by R2 and C2 and to prevent noise, you want to increase R2 to higher values (e.g. 2, 3 or 5G) if C2 is decreased to obtain a higher gain. Due to the additional open-loop gain provided by the BJT, distortion is much lower compared to the KM84 and output drive capability has increased.

Enough for now. Once I have assembled the circuit, I will share the measured specifications (THD, Ein, etc.)

Jan
? ? ? ? ? ? ? ?

It's really nice that in Jan's charge amplifier topology the BJT is actually doing useful work at AC. For the Schoeps phase splitter we really want to keep it quiet at AC (hence turning it into a Miller Integrator). But yes, at DC they are using the exact same biasing principle.

Funny that sometimes people at different places on the world have the same ideas at the same time.

For me, in this case it's a validation that the idea is good, and that really makes me happy. I am certainly not nearly as experienced and/or qualified in this field as Jan.