Adjustable current mirror

We are today trying to modify an existing board that has an LC filter after a DDS DAC. The Ls are all 1206s, and we want a 100 KHz lowpass. It looks very nasty, as we need tens of millihenries and the Qs go to hell.

John

And the BFT25 c-b junction makes an astonishing fA diode.

John

John Larkin a écrit :

What camera did you use?

The near view seems extremely detailed. Is it plain view or did you crop the image?

I wish I had one for my small sensor right now.

--
Thanks,
Fred.

I seem to recall reading a parametric equalizer circuit, dating back to the

70s or so -- when chokes were still cheaper than op-amps. It had chokes up to a few henry inside, little RFC style hash chokes simply with enough turns of dinky wire to get there. I think you can still get ~1H signal chokes from Mouser. Let's see, there's a Fastron 0.1H, 490 ohm, 34mA choke, and a 0.1H, 278 ohm, 29mA one from JW Miller. I'm not seeing anything larger (without going to the Hammond Power section, where you might find a 10H 100mA choke, but it's designed for power, not signal).

For all the tedium of building an EQ (i.e., build the same circuit ten times over, then double it for stereo), op-amps do make it easier.

Is there enough space on the board to scratch off the traces, and using conductive ink, draw a circuit for use with one of those really small op-amps? ;-) Not as crazy as it sounds -- they make epoxies with not terrible resistivity, so if you could, say, inkjet the stuff precisely into place, then stick the chips and whatnot into the goo, it might actually work when you turn it on.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

That would be too ugly. Williams' filter book has tables of predistorted "lossy-L" filter designs, but they can't seem to handle seriously bad inductors.

The real problem is noise coupling between some layer 2 traces (uP data bus lines) and some analog components. Lowering the filter cutoff frequency would help a lot, but the details look bad. It's looking like The Brat will just have to spin the board and shuffle some planes around, move a power pour to layer 2 and the bus stuff to 3, to block the couplings. Then we can leave the existing filter alone.

John

On a sunny day (Fri, 19 Feb 2010 18:29:25 -0600) it happened "Tim Williams" wrote in :

LOL, I think I did see some article many years ago about drawing a circuit with pencil. The carbon conducts. Maybe it was an April 1 joke...

Huh? It's monolithic, undoubtedly a square array. DI means a micron of SiO2, the other way to make a multiple transistor uses 25 microns of polyimide. Also, small-signal dice (20 mils) will need at least 30 mils of center-center spacing, and the heatpath goes twice through the

20 mils of Si vertical distance plus the heatsink before they connect.

Monolithic is a big win here.

The heat radiated from an op amp output stage can interact with the thermal dipole of a dual transistor, but not so much with the quadrupole lowest thermal moment of a cross-quad. For a four-transistor array, there's six dipoles, it's VERY complicated to get rid of all those dipole moments simultaneously. MAT-04 doesn't, I'm certain.

The datasheet shows the layout. Not a square array.

DI means a micron

Only a *big* win if it's thermally balanced.

John

message=20

the=20

chokes up=20

turns=20

chokes=20

and a=20

=20

=20

times=20

=20

into=20

work=20

Cool; your brat learning the hard way.

I can't blame a layout person for a problem like this. I should have noticed the potential coupling myself when I checked the board.

The board is this waveform generator...

and one customer (in Italy) just noticed the noise on two channels, so we'll have to replace their boards.

Luckily, we had another customer who wanted a completely different function, some long-interval digital timers. All we needed was a VME interface and an FPGA, so we did it using this same PCB, but don't stuff the analog parts. So we don't have to throw away any bare boards.

LC filters are nice, small and reliable and low parts count, as long as the frequency range is moderate.

John

The THAT380 is not even a quad transistor intended for transconductance multipliers, but a mixed array of four PNP and four NPN. There's nothing explicit in the datasheet for the quad transistors, but a square array is the likely configuration.

Are you two both trashing this part? First, in most applications for this part there's no heat source, like an opamp output stage, so a simple four NPN transistors in a square pattern, without the increasingly-rare inter- digitization, is still quite useful. Second, the 500uV matching spec is not to be taken lightly. It applies to all four parts. Third, the 0.8nV/rt-Hz noise spec (1mA) means the transistors have a low Rbb' (spec'd at 30 ohms, en = 0.7nV), also not to be taken lightly. Sounds good.

But I wonder, how easy are they to purchase?

--
 Thanks,
    - Win

_I_ don't trash this part, Both Hobbs and I have used it (essentially together with the same client) in a rather dicey medical instrumentation application _successfully_ !! ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.