Gain

agreed, we don't need dual-matched NPNs

I really appreciate that, Phil! You saying "if" rather than *when* LOL! Seriously, you're so far above me I wouldn't even be able to see you with that fancy new telescope they've just put into space. :-D More and more I'm tempted to go down the IC route...

In earlier days, realism was the reason for regenerative and superregenerative circuitry. All the gain you need without the expensive second transistor.

... and without the third through thirtieth transistor

Yeah, the 'expensive' word was, but is no longer, applicable.

Renesas / ex Intersil HFA3046, HFA3096, HFA3127, HFA3128

They even contain 5.5 GHz PNPs, still available the last time I needed them, even acceptable for space.

There is a data sheet from jan 2019, relatively fresh.

On Semi MBT6429DW1T1G NST45010MW6T1G UMZ1NT1G AD MAT14 SSM2210 SSM2220 Fairchild MMPQ6700 Diodes/ex Zetex ZXTC2061E6

(from my .pdf collection)

Cheers, Gerhard

Probably true for most applications, especially for higher frequencies where drive impedances are low. The gm is going to be significantly larger at the same current.

The classic application that jfets win is the guitar pickup front end. Base current noise is a problem when dropped across a 2 H inductor at 5kHz. even the dc resistance is a tad on the high side, maybe up to 10k. One also really requires an input resistance > 1M Ohm. 100k isn't enough.

The 2SK162 for audio is actually very good. Gets down to a noise equivalent of 26 ohms, matching bipolars

Sure, with off the shelf ics there are specialised gaps that they cant fill.

If there is a large enough market, there is, nothing that can't be done by a full ic implementation. It might require more than one ic though.

Most off the shelf ics such as op-amps are going to be "general purpose" not ASIC (Application Specific)

If one takes an iPhone, its 100s of "custom" ICs.

Its certainly difficult to get all types of devices in one process. However, even Analog Devices might supply what looks like one ic package, but actually has several die from different processes.

One can integrate any discrete device. An integrated circuit is just more than one different discrete devices on one die.

ASIC design is fundamentally different to discrete. Transistors are cheap, as is calibration. I would have to examine what your product spec is to evaluate how one would do it in an ic.

I haven't checked all the combinations available in current processes, but there are certainly "standard" (mos & bipolar) component processes that have GaN FETs, and "standard" processes with SiGe 200GHz devices, and cmos processes with pHEMT. There are a LOT of processes out there.

If the market size is there, its always possible to integrate anything. For small markets, there may be be a few applications where discrete is the way to go.

There are fundamental advantages to ic implementation. Routing capacitances are at the ff level for starters.

-- Kevin Aylward

Yes. I was doing such things about 15 years ago. The 2 H and 10 Kohms (DC) values are typical. The coil has something like 10,000 turns of AWG #42 enameled wire. An electric guitar pickup is a low-Q inductor with a self resonant frequency around 3 KHz when loaded by the guitar cable between pickup and the JFET input stage. The resonant Q might be

2. I do have one detail to add:

It's useful to have a 100 K film resistor in series with the JFET gate, to prevent output peak inversion on very strong input transients, where the gate junction diode is driven solidly into conduction.

Such transients are relatively common when the guitar is strummed hard. The peak is some volts, as I recall.

I don't think it damages the JFET, but the peak inversion is probably audible, more so than simply clipping the peak.

Do you have any information on its flicker noise performance?

Joe Gwinn

Joe Gwinn wrote: "Kevin Aylward"

----------------------------

-----------------------------------------------------------

...... Phil

All true, but the artists seem to like the resonance lower than 6 KHz (the no-cable value), and add cable to achieve. These folk do have golden ears.

Adds some hiss for sure, benefit on balance is matter of opinion.

I thought it blurred the attack transients, but never made the behind-curtain tests necessary to settle the issue.

Yes. Which is why I clipped it, once the inversion was prevented.

Joe Gwinn

Joe Gwinn is a D*****ad wrote: "Kevin Aylward"

---------------------------

Been making and testing/repairing commercial guitar amps for all my 50 year career and must have dealt with 500 different guitar players. ( The ones I made all had JFET inputs )

FYI pal:

There is NO consensus among such "artists".

..... Phil

Well, you can always reduce the collector current and stabilize almost anything. ;) <snip>

To get a gain of 400 at 10 kHz, you'll need both sections of that LM358A. ;)

To do it with discretes, you need to do a bit of algebra to understand the issues. It's pretty simple though--there's no need to collect all the effects into one giant opaque expression. It's better to follow all the separate terms--the way the transconductance sets the maximum gain, the Miller (C-B) rolloff due to source impedance and C-B capacitance, and the Early effect that leads to an effective shunt resistance at the collector.

You can ignore their interactions (e.g. the way the Early effect reduces the gain, which reduces the Miller effect) because when any one effect gets big enough for that to be an issue, you're already hosed. ;)

Doing the math that way is about the fastest way to generate intuition about how the circuit actually works.

Cheers

Phil Hobbs

Miller effect; Early effect. I've got some reading up to do, clearly! Seriously, I already have a future project in mind for an audio amp using good old tubes just for the hell of it and will hold back til then from studying the finer points. For *this* particular problem (the one under discussion here; just getting this vintage radio working again) however, I'm coming around to your suggestion of the quickest solution and just going for a suitable IC. I believe I've found one in my copious treasury of old bits. The only reservation I have is if it's suitable to use in battery powered equipment. I'll post about this under a fresh thread title shortly.

That mathematics don't lie, but hand calcs can be very, very time consuming in certain circs. That's the outstanding benefit of the Spice programs IMHO.

If you do them wrong, e.g. building that huge opaque expression that has all the effects at once, that's true. But there's no need to do that in most cases.

What I'm talking about is like this:

1. Miller effect: The capacitive current at the base due to Miller effect is

i_Miller = -C_cb * d V_cb / dt = (|A_V| + 1) C_cb dV_be / dt,

where as usual A_V is the voltage gain of the stage (-400 in this case).

Thus the effective input capacitance of an amp with a gain of -400 is

C_in ~= 401 C_cb.

So you have a rolloff with a 3-dB corner at

f_Miller = 1/(2 pi C_cb R_s) * 1/401.

With a 400-ohm source and a 3-pF C_cb, that comes out to 330 kHz. So Miller doesn't do a lot at 10 kHz, and we can ignore him henceforward.

1. Early effect is similar, but more or less frequency independent: it looks like a differential resistance in parallel with the collector, and is characterized by the Early voltage V_A, which is the X-axis intercept of the I_C vs. V_CE curve. (It's not a real DC resistance, so it doesn't matter where the other end goes as long as it's signal ground.)

R_Early = (V_CE + V_A) / I_C.

Here all all quantities are assumed positive, even though the X intercept is usually far out on the negative V axis. (That's why V-CE gets added in--the slope is the rise over the run.) This shunt resistance appears in parallel with the external collector resistor, and reduces the gain exactly as you'd expect if you had a real resistor there.

We talked about how to handle the transconductance earlier.

So from a math POV it's dead simple. You just need a clear idea of what each effect actually does to the circuit, so you know what you can ignore and what you have to work around.

Cheers

Phil Hobbs

Your faith is touching, but misplaced. There are a lot of discrete GaN FETs, IGBTs, and so on that aren't going away any time very soon. Many front end things are like that too.

Sure, but that's a hybrid, not an IC. My front end would probably fit in a 40-pin CERDIP too.

If you can figure out how to get power GaN FETs really monolithically integrated with silicon CMOS, you could make a lot of money. Ain't happening soon.

Just the lattice matching, strain, and process compatibility issues are horrendous.

(Yes, people have published papers on it, and have begun looking at doing wafer bonding. The yield problems with that approach would make a brave man blench.)

Sure, I've collaborated on chip designs, so I know the drill. Getting 6 GHz f_max, very low noise (300 pV /sqrt(Hz) in the flatband, with sub-nanoamp gate leakage on the one hand, with ~5-ns switching over a

SiGe, I believe. The others are all wafer-bonding or die-on-wafer AFAICT.

And I have a bridge to sell you.

Absolutely. And for most things those sorts of advantages, together with monolithic matching and so forth, are enough that a clever designer with a sufficiently-big budget can do a good job.

That time stretcher thing I was talking about works a lot better in an IC--the proto used eight 2-GHz CFAs, each with three pHEMT T/Hs hung on it.

Each T/H hold cap was connected directly to an input of a simultaneous-sampling ADC, and the channels were sampled sequentially at much lower speed. The hold capacitors thus had to be several picofarads, which confused the CFAs when the switches opened, so we had to interleave the sampling to give the amps time to recover a bit. Sucked power like anything, but did demonstrate that you can get multiple well-behaved sub-nanosecond time slices in a single shot without needing a big-iron digitizer/FPGA solution.

We're in the process of integrating that and scaling it up to 4096 APDs x 24 samples per shot.

High performance front ends, not so much. A couple of CPH3910s, or a SAV-551+ cascoded with a BFP640, can do things no available IC can touch, even with no high voltage requirement.

Cheers

Phil Hobbs

volt

OT: There are no CPH3910s in stock at Digikey. First microcontrollers and now this.

I have 30k of them in stock. ;)

Cheers

Phil Hobbs

TBF most MMICs that have only a Darlington in the signal path also have an op-amp circuit that controls the bias around the designed operating current, and may also have impedance matching for the input & output.

Bias stabilisation is worth having on-die because they can temperature-compensate it using the die temperature.

CH

And Mouser has over 1,000 in stock.