[jamont]

I've got a bag of Fairchild 2N5486's, all with the same date code. I'm checking their Idss values, using a breadboard setup as described here on tangent's web site. I'm getting essentially the same value (9.08 mA) for all of them. I expected a little scatter in the values, does this seem reasonable?

 

[jeffreyj]

It is very reasonable if they all came from the same batch. Having the same date code does not necessarily mean they were all made on the exact same *day* - the date code invariably specifies the year and month only.

A better question might be: why use a FET for a constant current source in the first place???

 

[jamont]

Quote:

Originally posted by jeffreyj It is very reasonable if they all came from the same batch. Having the same date code does not necessarily mean they were all made on the exact same *day* - the date code invariably specifies the year and month only.

Thanks, Jeffrey!

Quote:

A better question might be: why use a FET for a constant current source in the first place???

You would prefer bipolar? I'd be interested to hear your analysis of the pros and cons.

 

[jeffreyj]

Quote:

Originally posted by jamont ... You would prefer bipolar? I'd be interested to hear your analysis of the pros and cons.

Which is "best" depends on what parameters are most important, of course, but my opinion that bipolar is better considered those parameters most relevant to the particular application at hand (biasing an op-amp's output stage into Class A). If sheer simplicity is paramount, then you can't beat a FET with its gate tied to its source. If you want to make sure multiple current sources track each other closely - so that each channel exits Class A at the same output level - then the bipolar current mirror is definitely the way to go. The much higher transconductance of the bipolar transistor means that its dynamic output impedance comes closer to the ideal of infinity when used as a current source. I believe morsel explained that FETs were chosen for the META42 and PPA because they supposedly have lower capacitance (from where to where??) but that's true only in specific comparisons, not as a general rule. Certain forms of transistor construction result in higher collector to emitter capacitance, yes, but since we are usually talking about 1-6pF worth no matter what (5pF at 20kHz is 1.6M of capacitive reactance) this is a totally superfluous argument. Given that bipolar transistors are almost always cheaper than similarly rated FETs and aren't nearly as sensitive to ESD, I'd pretty much go with bipolar.

Of course, if someone says that using a FET for the current source sounds better, well, there's no arguing that. Subjective experience, no matter how objectively valid it might be, is not arguable!

edit: hilarious typo corrected...

 

[jamont]

Quote:

Originally posted by jeffreyj Of course, if someone says that using a FET for the current source sounds better, well, there's no arguing that. Subjective experience, no matter how objectively valid it might be, is not arguable!

Well, people can (and do) hear almost anything. If I'm going to build or buy equipment, I want a design based on sound engineering, not tweako nonsense.

Thanks, Jeffrey, you'll make us into engineers yet...

 

[aos]

You would suggest a current source between channels? But what about interchannel crosstalk that would ensue? FETs are incredibly cheap (come on less than 10 cents), schematics are simple and it works well. It is also possible to match them to within a few hundred uA. How are you going to bias the BJT? With current mirror? With a LED? FET one will have a great efficiency - all current through it is used to bias the amp, what about BJT, current goes through LED or other side of mirror (unless it's tied to the other channel, but crosstalk?), it's wasted, no? I mean, if you are taking argument that capacitance with BJT is low enough and the variation is irrelevant, you must also take the argument that the FET impedance is high enough and the variation is irrelevant. You sound like you owned a BJT factory that was forced into bankruptcy by a FET factory. Give poor FETs a chance man, it's bad enough they're barely used these days, leave them in peace for this one role they're good for.

By the way use of JFETs as current sources as such was not invented by Morsel or ppl (however biasing an opamp with them is ppl's idea as far as I know). It's probably as old as JFETs themselves. Borbely Designs (high end audio) uses JFETs (cascoded as well) as current sources, and they're even sold by manufacturers such as Vishay Siliconix for specific use as current sources. For example:

http://www.vishay.com/document/70711/70711.pdf

 

[jeffreyj]

Quote:

Originally posted by aos You would suggest a current source between channels?

Good lord, no, aos! Surely you know better than to assume I meant *that*?! I specifically stated using a current mirror.

Quote:

[A] FET one will have a great efficiency - all current through it is used to bias the amp, what about BJT, current goes through LED or other side of mirror (unless it's tied to the other channel, but crosstalk?), it's wasted, no?

A valid point, if somewhat specious. If I am biasing the output of an op-amp into class A with 2mA of current and the beta of the BJT is 200, then the current bias current for the mirror is a rather insignificant 10uA per channel. Shaving that off the power consumption of a Class A amp is surely an unimpressive feat!

Quote:

I mean, if you are taking argument that capacitance with BJT is low enough and the variation is irrelevant, you must also take the argument that the FET impedance is high enough and the variation is irrelevant.

True, but why settle for less performance for more money?

Quote:

You sound like you owned a BJT factory that was forced into bankruptcy by a FET factory. Give poor FETs a chance man, it's bad enough they're barely used these days, leave them in peace for this one role they're good for.

ROTFL!! You're killing me, aos!

I don't hate FETs, I just don't think they are a panacea as so many in the audio field do. I use MOSFETs almost exclusively in switching power supplies because that is one are they are demonstrably better than BJTs. Another are would be when high input impedance is truly needed for an amplifier. Other than that, there just isn't any compelling reason from an engineering perspective.

Quote:

By the way use of JFETs as current sources as such was not invented by Morsel or ppl (however biasing an opamp with them is ppl's idea as far as I know).

Of course, but it was morsel, I believe, who last justified their use.

Don't get me wrong, aos (et al.) - JFETs work just fine as current sources in such a non-critical application, but a BJT current mirror, requiring but 3 BJTs and one resistor, can give near-perfect tracking of currents for two independent channels, especially if Vcc is regulated. The equivalent current regulation performance from a FET-based design would require using the cascode (4 FETs, 2 resistors minimum) but there still isn't any good way to ensure tracking between separate channels.

But mostly I make these posts to be the Loyal Rebel, as tangent so aptly christened me in a recent post

 

[aos]

Quote:

Good lord, no, aos! Surely you know better than to assume I meant *that*?! I specifically stated using a current mirror.

That was a typo

. I mean, current *mirror* not *source*, where one BJT is left channel, the other is right channel? Why not draw the schematics then, because the way I imagine it, you have two BJT's (standard mirror configuration with tied bases and emitters), their collectors one to each channel.

Quote:

A valid point, if somewhat specious. If I am biasing the output of an op-amp into class A with 2mA of current and the beta of the BJT is 200, then the current bias current for the mirror is a rather insignificant 10uA per channel. Shaving that off the power consumption of a Class A amp is surely an unimpressive feat!

As I mentioned above, the way I imagined it, unless you want crosstalk, you'd need to have two current mirrors (2 BJTs each) with one end of each unused, and that would go to waste. I'm sure you have different schematics in mind now, so when you draw it it will become clear.

Quote:

True, but why settle for less performance for more money?

This I don't get, as 9 cents per JFET surely isn't anything to talk about? You need to add a resistor, so total of 2 JFETs and 2 resistors. Or in cascode, double that. I'm accustomed to buying BJT's that are 5 per dollar in local store but I don't doubt Mouser is cheaper for that too. Still when you're talking 9 cents it only matters if you're a big company designing a ultrahigh volume product.

And this brings me to another point - whatever schematics you have in mind, I doubt it can be any simpler than this. Besides, you can use SOT-23 JFETs as linked by datasheet, making this thing ultrasmall. So it's simple, few parts, small size, and cheap. Another benefit is that even if you want to model it mathematically, you only have that one JFET to use.

Quote:

I just don't think they are a panacea as so many in the audio field do.

Actually most poeple don't - as you say, modern engineering doesn't use FETs very much. I'm trying to save them from extinction here!

Also I am not certain how much importance there is in tracking the bias of channels. For that to work you need to have identical chips. Are the opamps so well controlled that the point they enter A/B from A is nearly identical for two chips? I guess the process is so good these days that the answer is yes. But you still have to show that there's no crosstalk with your schematics - you have an interconnected circuit after all so this is definitely a potential issue. The other practical issue is PCB layout - you may find it harder to lay out the parts now that you have more parts that are common to both channels.

 

[jeffreyj]

Quote:

Originally posted by aos ... As I mentioned above, the way I imagined it, unless you want crosstalk, you'd need to have two current mirrors (2 BJTs each) with one end of each unused, and that would go to waste. I'm sure you have different schematics in mind now, so when you draw it it will become clear.

Schematic is attached (I think.. we'll see).
Indeed it is the configuration you imagined. It appears that crosstalk would be a problem, but in my informal testing it seems to only lower the separation between channels by about 10dB when powered by dual polarity regulated supplies and 0dB when powered by a TLE rail splitter. This confirms what I would expect: the virtual ground's high-ish impedance dominates the crosstalk contribution.


Quote:

Also I am not certain how much importance there is in tracking the bias of channels. For that to work you need to have identical chips. Are the opamps so well controlled that the point they enter A/B from A is nearly identical for two chips?...

I'm not certain, either, but it seems like a worthwhile pursuit. The difference in current I was thinking of, w/r/t using JFETs in the current sources, was on the order of 0.5mA or more. The difference between any two op-amp output stages of the same part number should be at least an order of magnitude less, so I didn't assign it any relevance. But who knows?!

 

[aos]

Isn't current through the R1 going to be the same as through L and R collectors, by the virtue of current mirror? That's what I referred to as loss of efficiency.

 

[jeffreyj]

Yes, the current through R1 is the programming current and so it is essentially wasted. I see the point you were making earlier; I thought you were pointing out that the practically non-existent gate current was a significant improvement over the neglible base current in the BJT mirror. Guess I ignored the obvious on this one

I still like the BJT mirror over individually self-biased JFETs (even/especially with source degeneration resistance in place to force closer tracking of the currents) because performance is superior and predictable, programming the current for multiple channels can be done simultaneously by changing one resistor and all channels track the program current exceptionally well. Oh, and the average BJT for this job costs about $0.04 each in 100's versus $0.07 - $0.10 for those 5400 series JFETs - a significant cost savings if you ask me

 

[Arzela]

I'm currently matching my class A
cascodes. With 100 ohms in R8, and
random 5486's in Q2, various 5484's
in Q1 give readings, across R9, of 1.1-2.0 ma.

Good thing I bought 50 of the 5484's!

 

[guzzler]

just curious as to where you get all your semiconductors... in the quantities that most of us deal with, a 2N3904 sets me back around 25p each, the 2N5486 around 70p (although the price decreases rapidly with the FETs)

g

 

[Arzela]

Mine came from Newark. 7 cents
american for 5484/86.

 

[jeffreyj]

Quote:

Originally posted by guzzler just curious as to where you get all your semiconductors... in the quantities that most of us deal with, a 2N3904 sets me back around 25p each, the 2N5486 around 70p (although the price decreases rapidly with the FETs) g

Well, guzzler, living in the hinterlands of Ye Old World has its economic downfall, I see

I just checked Farnell - one of the two main suppliers of electronic components on your side of the pond - and the lowest price for the 2N3904 in single quantities is 10p, which isn't so bad for 1's. The other major supplier is RS Components but their price wasn't better (11.4p each for a min. quantity of 5). The price for the 2N5486 JFET was a different story - it is amazingly more expensive there than it is here: Farnell wants 58p each in 1's. That's almost a dollar?! Wow - guess it's even more compelling for you UK'ers to go bipolar, eh?