I have bought abt. 12 of the IF3602, and they are all individuals. My measurements look
completely different, they are in post #320 of the "Scott's take of the G=1000" thread.
For FETs, I tend to trust the Locky-Z curve tracer now; the results are in the specs
(which are addmittedly barn doors). They easily sank everything that my bipolar cascode
and active load could throw at them, and they would not even burp, even if it was excessive.
I have tried hard, but could not get my construction stable, at some frequencies Zin would
always be negative. Sometimes the network analyzer is cruel. In practical operation
it might not oscillate and one would be tempted to declare "DONE!" but then you
know that you have a skeleton in the closet...
I have shelved it until I get new ideas, the chopper is much more interesting for me anyway.
Cheers, Gerhard
PS
On a 2nd thought, is the 12 mA an artificial limitation?
completely different, they are in post #320 of the "Scott's take of the G=1000" thread.
For FETs, I tend to trust the Locky-Z curve tracer now; the results are in the specs
(which are addmittedly barn doors). They easily sank everything that my bipolar cascode
and active load could throw at them, and they would not even burp, even if it was excessive.
I have tried hard, but could not get my construction stable, at some frequencies Zin would
always be negative. Sometimes the network analyzer is cruel. In practical operation
it might not oscillate and one would be tempted to declare "DONE!" but then you
know that you have a skeleton in the closet...
I have shelved it until I get new ideas, the chopper is much more interesting for me anyway.
Cheers, Gerhard
PS
On a 2nd thought, is the 12 mA an artificial limitation?
Last edited:
12mA is a limitation of the PEAK curve tracer. I also have a Tek 576 -- but then I would have to take a picture and convert it to Excel.
One IF3602 = 1 bottle of very nice Bordeaux.
I pulled out the 8xBF862 and ran the Linear Audio 3 article with 4.8mA per device. There's a little tweaking which has to be done, but it didn't blow up. The input coupling cap is 1u/PP.
Attachments
THAT1510 is bipolar, so Zin is much, much lower. It's a slightly better SSM2019. Both are really great, however. The SSM2019 I purchased recently are <1nV/RtHz noise.
Some pictures. The first FFT is gain of 1,000 -- with 1mV signal, Second is amplifier with the input shorted, third uses the analog analyzer in "bandwidth" mode takes the square root of frequency, gain and bandpass factor.
It isn't as pretty as Sam's, it's just built on a protoboard at the moment. The line harmonics are much easier to reduce or eliminate if it is placed in multiple enclosures which will save for a later day.
It isn't as pretty as Sam's, it's just built on a protoboard at the moment. The line harmonics are much easier to reduce or eliminate if it is placed in multiple enclosures which will save for a later day.
Attachments
Remember these devices are mainly physics lab stuff IR bolometers, HUGE PD's, etc. The swap for equal voltage noise vis a vis bi-polar loses the importance of input source impedance.
The 1/f region is also very important in some research, the start of the black hole spin down at LIGO is still buried in 1/f noise.
Yes, but for everything but ribbon microphones or moving coil pickups you are
likely to need insane input coupling capacitors. You noted that effect yourself
with my 20xADA4898 preamp. I have changed that from foil to wet slug tantalum.
Or really, I added the tantal. It's impossible to remove the foil capacitor in a
destructionless way.
With FET's that's much easier. The filter effect of the input C grows linear
with bias resistor size, but the noise voltage grows only with the square root.
Somehow against the belly feeling.
BTW the IF3602 bias current cannot be ignored. I got 800 mV DC across
a 47 Meg resistor for 2 pairs of IF3602 in parallel, operating at 2.5V into
a cascode.
It is hard to get 10Meg & larger as thin film. Thick film / Cermet is worse
than carbon composition wrt 1/f noise at DC bias.
I have made a new iteration of the chopper. The switching spikes seem to
be different for the rising and falling clock edge, although everything is symmetrical.
That asymmetry is ugly because it creates an offset. Nevertheless, the
shielding boxes made the clock far less visible everywhere.
The chopper switches have their own floating ground and +/- 2V from
the rectified and filtered 500 KHz clock that is also isolated by a ferrite transformer.
I even have baluns on the input & before the step up transformer to enforce symmetry.
There is still place for 2 more ADA4898-2 in the stage after the transformer. That
could reduce the pressure on the step-up ratio.
The 2*4 FET switches stress my favorite homemade board process with 1 layer + ground.
Lots of bridges required.
cheers, Gerhard
likely to need insane input coupling capacitors. You noted that effect yourself
with my 20xADA4898 preamp. I have changed that from foil to wet slug tantalum.
Or really, I added the tantal. It's impossible to remove the foil capacitor in a
destructionless way.
With FET's that's much easier. The filter effect of the input C grows linear
with bias resistor size, but the noise voltage grows only with the square root.
Somehow against the belly feeling.
BTW the IF3602 bias current cannot be ignored. I got 800 mV DC across
a 47 Meg resistor for 2 pairs of IF3602 in parallel, operating at 2.5V into
a cascode.
It is hard to get 10Meg & larger as thin film. Thick film / Cermet is worse
than carbon composition wrt 1/f noise at DC bias.
I have made a new iteration of the chopper. The switching spikes seem to
be different for the rising and falling clock edge, although everything is symmetrical.
That asymmetry is ugly because it creates an offset. Nevertheless, the
shielding boxes made the clock far less visible everywhere.
The chopper switches have their own floating ground and +/- 2V from
the rectified and filtered 500 KHz clock that is also isolated by a ferrite transformer.
I even have baluns on the input & before the step up transformer to enforce symmetry.
There is still place for 2 more ADA4898-2 in the stage after the transformer. That
could reduce the pressure on the step-up ratio.
The 2*4 FET switches stress my favorite homemade board process with 1 layer + ground.
Lots of bridges required.
cheers, Gerhard
Attachments
Last edited:
I shortened the leads from the 9V batteries and sealed the "cookie tin" better -- here's #2 IF3602. Worse from 10Hz to 100Hz, better @10kHz. This amplifier with the second device required a lot more compensation to prevent oscillation.
You're certainly correct in that the devices vary from lot to lot and quite a bit!
Sam's amplifier with the BF862 has much better low frequency performance.
FWIW, the impedance measures ~122k
You're certainly correct in that the devices vary from lot to lot and quite a bit!
Sam's amplifier with the BF862 has much better low frequency performance.
FWIW, the impedance measures ~122k
Attachments
Hi, Scott, do you know where the signal comes from / what it looks like, that goes into your AD797s there? Is it DC centered?
What's the circuit you are using? I really doubt that you see the noise of the IF3602 there.
I also doubt Sam's 30 Hz (IIRC) 1/f corner from all I have seen from BF862. The pecking
order seems reversed.
With enough white noise you can do arbitrary things to the1/f corner.
Rubiola is not a friend of using the 1/f corner altogether.
< http://www.rubiola.org/pdf-articles/journal/2004rsi(rubiola)low-flicker-dc-amplifier.pdf >
There is a lot of good stuff on rubiola.org, btw.
If you have an active load at the drain, do you have enough emitter resistance? That was the second largest noise contributor in my case. The cascode practically adds nothing at all.
cheers, Gerhard
Thanks much for pointing out the paper -- looks like one of those Hammond boxes is in order.
This is approximately the circuit I have on the breadboard:
Attachments
First thing I'd do is to remove Q101. Its operating point is largely undefined with
no DC return for the drain. Its channel resistance is in series to the input and the
thermal noise of this resistance will dominate over anything that could come from
the IF3602s. Only 60 Ohms would be a nv/rtHz.
In case this is meant for protection, the IF3602 can probably take more than the BF862.
I would also use a much smaller bias resistor and a larger coupling capacitor.
Say, 10Meg * 22uF. You will have maybe half a volt of drop over the 200Meg
because of gate current, and that will produce 1/f noise over the resistor. 200Meg
thin film is hard to get and thick film/ cermet is worse than carbon.
Also 200 Meg and a plug-in test board may not go together well.
no DC return for the drain. Its channel resistance is in series to the input and the
thermal noise of this resistance will dominate over anything that could come from
the IF3602s. Only 60 Ohms would be a nv/rtHz.
In case this is meant for protection, the IF3602 can probably take more than the BF862.
I would also use a much smaller bias resistor and a larger coupling capacitor.
Say, 10Meg * 22uF. You will have maybe half a volt of drop over the 200Meg
because of gate current, and that will produce 1/f noise over the resistor. 200Meg
thin film is hard to get and thick film/ cermet is worse than carbon.
Also 200 Meg and a plug-in test board may not go together well.
This afternoon I reconfigured the IF3602 -- some bigger ferrites on the gates and adjusted the comp on the NE5534. These two charts illustrate the performance with a pair of 9V's alkaline cells and Jan's "silent switcher" with the leads dressed differently and 4x D-cells powering it.
The silent switcher shows somewhat diminished performance -- but not as bad as the prior run.
The silent switcher shows somewhat diminished performance -- but not as bad as the prior run.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.