Help with FETs for MXR Phase 45/90 (I used the "Search" button!!!)

[gigimarga]

Hello again,

I've tried some time ago to build a MXR Phase 45 and a Phase 90.
Because here it's impossible to find 2N5952, I've tried to use some matched BF245A/BF245B/2N5457/J201 instead.

I found in a couple of build reports that BF245A are the best and it was true: BF245B (matched around -2V or -2.5V) didn't work in P45 (4.7V Zenner), 2N5457 (matched around -1V, if I'm remebering well) made a very weak effect and J201 worked almost the same as 2N5457.

BF245A matched around -1.6V or -1.4V seems to be the best, but with the 250K trimpot maximized (or minimized). Anyway, for 2N5457 and J201 the trimpot must to be maximized/minimized, too.

But here http://www.lynx.bc.ca/~jc/pedalsPhase45.html (thx JC Maillet!) I found that isn't right (he said that these FETs aren't good). I've tried his method (to put only one FET and to adjust the trimpot for maximum sweep) with 4.3V, 4.7V, 5.1V and 5.6V Zenners, but no signifiant changes.

So, my questions is: which FETs to use from BF245A/B/C with what Zenner? Is it a clear method (some voltages) to tweak the FETs?

Thx a lot!

[frequencycentral]

I used 2n5457 in my (4 stage) P45. The matched Vgs were about -1v.

I used 2n5485 in my (8 stage) P90. The matched Vgs were about -1.8v.

Have a look at reply #62 in this thread: http://www.diystompboxes.com/smfforum/index.php?topic=76762.60 It's possible to use different combinations of LEDs and 1n4148 to set the vref at different voltages. If you alter the vref slightly you'll get a bit more 'play' in the bias trimmer.

These FET phasers can be very problematic. Try using R.G.'s FET matching fixture at GEOFEX. As I understand it there's more to matching FETs than just the Vgs, but the matching fixture should get you in the ballpark.

[R.G.]

I guess that my approach to most problems is that until you know what's really going on, you can't do much to solve the problem. Having someone simply tell you *their* answer doesn't make me happy, either. It only solves one instance of what may really be a complex problem.

All JFETs produce a source-to-drain resistance range of from infinity (that is, fully off, no conduction) to as low as their rds specification from the datasheet. This last is usually under 1K, and may be under 10 ohms for JFETs made for switching. It's much lower than you need in a phaser in any case. What is different about JFETs is the voltage on the gate needed to get it to change between fully off and fully on. All normal JFETs are as fully off as they can get at Vgs=0, or the gate and source at the same voltage. They are fully off when the gate is at Vgsoff as per the datasheet.

Vgsoff varies from as low as 0.1V for some J201s - which makes the 201 kinda remarkable in itself - to as much as 10V. And it depends not only on the type number, but also the specific JFET, which is why they need to be matched some way. The ways to match them can be either analytic (like the JFET matcher: measure them all, and pick the ones which match) or subjective (like trying one, and then throwing it away if it doesn't sound good).

I'm a measurement kinda guy. If I can measure something, I think I can make intelligent choices about which I want and what I do. So my favorite is to know what voltages I need and measure devices until I get what I want. But in this case the big question is what voltages do I (that is, *you*) need?

The answer is in the circuit. The P45/P90 circuit has limits on the voltages it can accept. The bias zener sets one of them: the drains and sources of all the JFETs are held at the zener voltage above ground. As we know, if the gate is up at this voltage, the JFET will be fully on, for all JFETs we can actually buy. But some JFETs will need a voltage 0.1 to 0.3V below this (i.e. the J201) and some will need as much as -8V or even more (MPF102 is like this) to get near turn-off. How do we pick?

The circuit tells us. The zener sets the maximum reverse gate voltage and at the same time the opamp bias. The opamps need to be in the middle of the nominally 9V power supply, so 3 to 6V is about the biggest range you can use without having the opamps start clipping on one side or another.The gates of the JFETs will be fed a DC voltage of the setting of the trimpot, and also an AC voltage equal to the swing on the LFO, which rides on the trimpot DC level.

How big is the LFO voltage? A little quick simulator work (or measurement of the circuit) shows its about 1.6V peak to peak at the 10uF cap to ground which acts as the integrator. It goes through the 3.3M resistor to the gates, but there is also a 1M resistor to the trimpot wiper. The 3.3M and 1M act as a voltage divider, making the LFO voltage be about 1/3 of the maximum, or about half a volt peak to peak. This half a volt of wiggling LFO can be set anywhere between 0V and the zener voltage by the trimpot. If you make the 1M a variable resistor, you can have a "width" control.

So the stock circuit limits you to JFETs with VGSoff of less than the zener voltage, which was 3.9V in the originals. The LFO voltage limits you to JFETs which have a range of useful voltages which cause a channel resistance of about 1K to 100K over a range of at most 0.5V. The original 2N5292 has a Vgsoff of around 3V. The 2N5485 has a range of  0.4V to 4.0V. I picked this one out as the best replacement for the 2N5292 about twelve years ago when I was first telling people how to make P90 clones. The BF245A has a range of 0.4 to 2.2, the BF245B has from 1.6 to 3.8. They can both work OK in the P45/P90 circuit if they (a) are properly matched and (b) the trimpot is properly tweaked. It is also possible that the 1M "width" pot may need to be tweaked to make a JFET's "hot zone" where its channel resistance changes most be fit best.

Knowing some of the answers, we can get back to your question. I would advise using 2N5485 if you can find it. It's in current production. But if you simply must use a BF245 (temporarily, while you find the best parts, maybe), use the BF245A. Match the JFETs for Vgsoff. I like most JFETs at about 1.4 to 1.8V in this circuit, so get a set where they all measure the same, and within that range. Then tweak the trimpot for best sound. You may need to tweak the value of the "1M" resistor too.

People who simply build and don't measure will report all kinds of things about which part worked best. They're usually telling the truth, because the one of that kind of part that they used did, in fact, work best for them. Remember the story of the blind men and the elephant?

[gigimarga]

Thx a lot both of you, frequencycentral and R.G.!

@frequencycentral: I've matched the FETs using the R.G.'s FET matcher from GEOFEX.
@R.G.: As usually, you answers are gems of knowledge (even I was able to understand the explanation). The only thing that I didn't understand it's if is it OK or not to have sweep only at the end of the trimpot or I must to find some BF245 which sweep around the half of the trimpot.

Thx again!

[DougH]

FWIW, I used J201's in my phase 90 and they worked fine. I don't remember the matching voltage and I may have adjusted the zener voltage (i.e. used a slightly different voltage z diode) but I don't remember.

I believe that, within the constraints that R.G. mentioned above, you can pretty much get any JFET to work as long as they are matched and you deal with the zener voltage properly.

[Mark Hammer]

It is also worth noting that you're really after producing notches within a certain range anyway, and the range they will occur in depends not oly on the FET properties, but the fixed resistor in parallel and the cap value.

Theoretically, it is possible to "correct" for any set of reasonably suitable matched FETs by altering the value of the parallel resistor to ground to set how much resistance change will occur, and to use a different value of cap to determine where the resulting notches are going to occur...roughly.  All of that is of course on top of the role that the trimpot adjustment and biasing plays.

[R.G.]

It is also worth noting that you're really after producing notches within a certain range anyway, and the range they will occur in depends not oly on the FET properties, but the fixed resistor in parallel and the cap value.

Theoretically, it is possible to "correct" for any set of reasonably suitable matched FETs by altering the value of the parallel resistor to ground to set how much resistance change will occur, and to use a different value of cap to determine where the resulting notches are going to occur...roughly.  All of that is of course on top of the role that the trimpot adjustment and biasing plays.

Right again, Mark.

The whole reason to select JFETs is to pick four that are all wiggled through the usable region of their channel resistance, that being from 1/10 to 10X the input resistor going to the (-) side of the opamps, at about the same LFO voltages, both in terms of the DC value and the AC value pulling them up and down from the DC value. The matching doesn't have to be perfect, which is good because it usually can't be without very extensive testing.

One could use less-well-matched JFETs if one used one DC trimpot per JFET and one "width" trimpot per JFET. This would let you trim in modestly poorly matched JFETs. However, JFETs cost about $0.10, and trimpots cost $0.20 to $0.75 each, so it gets expensive. Not to mention that an 8-trimpot setup is a technical nightmare. At least to me. 

[gigimarga]

Thx again all of you!

I found a 2N5952 with VGSoff=-1.6V and when I tested it in MXR45 I must to maximize the trimpot to get a sweep.
Using a 5.1V Zenner, I get voltages from 2V to 4.6V at the drains of the FETs. It's OK? Anyone has some good voltages?

[Fender3D]

I had some issues with my p90 replica and 5532 ICs,
it wouldn't work untill I swap them with 4558.
5532s altered the bias voltage and I couldn't achieve any sweep, but with 4558s everything's working as it should...
my 2 cents...

ps
just waiting for another batch of 5532... darn mouser....

[frequencycentral]

I had some issues with my p90 replica and 5532 ICs,
it wouldn't work untill I swap them with 4558.
5532s altered the bias voltage and I couldn't achieve any sweep, but with 4558s everything's working as it should...
my 2 cents...

ps
just waiting for another batch of 5532... darn mouser....

I used 4558's in my P45 (with two extra stages), I'm planning to try it with 5532's as it's a little hissy at times. I'll report back on that.....

I used two LM324 in my Phase 180, works fine. Out of curiosity I tried it with two TL084 and I got some quite insane ticking/thunking sounds. I guess I'm more than a little hazey on the differences between various opamp types, but I imagined it should work with TL084. 

[gigimarga]

Thx frequencycentral and Fender3D!
I used TL072 as is written in Tonepad's project...I will try to swap them with some 4558.

[GP]

I used two LM324 in my Phase 180, works fine. Out of curiosity I tried it with two TL084 and I got some quite insane ticking/thunking sounds. I guess I'm more than a little hazey on the differences between various opamp types, but I imagined it should work with TL084. 

I think it can go both ways, Rick. I'd originally intended to use a 324 in mine (just to get rid of one...not exactly the best specced quad opamp) but it was very ticky and noisy. Instead I used a TL074 and it was much much better.

I would imagine choice of opamp is vaguely important but less so than getting everything else (layout, DC conditions, blah blah) right and we all know how subjective the topic of which opamp sounds best is.

[gigimarga]

Maybe I am too tired, but it seems to me that on the PCB for Phase 45 from Tonepad the sources and the drains of the FETs are swapped (as I found on JC Maillet schematic and on GGG it seems to me that sources must to go to the trimpot and drains must to go to the "+ inputs" of the opamps...) for 2N5952/2N5457/BF245 type FETS (G-S-D).

I tested this observation in my Phase 45 and it seems that it doesn't matter anyway if source and drain are swapped: even the bias point remains the same (maybe as in Tonepad's PCB sounds a little fatter, but there is 1 AM in the morning and I tested that observation at a very low volume)!

[frequencycentral]

Drain and source are pretty much interchangeable in many cases. Though there must be situations where it does matter which is which.

[gigimarga]

Thx frequencycentral
Momentary I am almost satisified by the Phase 45 with a pair of 2N5457 matched around -0.6V and I hope that somebody will buy for me some 2N5952 from farnell.

[R.G.]

Many JFET datasheets have the note that drain and source are interchangeable.

There are some which are not.

[gigimarga]

Thx R.G.!

What it seems curious to me is, even having 4.5V on sources, some BF245B matched around -2V don't want to work.

[Eb7+9]

What it seems curious to me is, even having 4.5V on sources, some BF245B matched around -2V don't want to work.

If somebody bothered analyzing what that "simple" test circuit does on a graph they'd see one is measuring an "operating" point far away from the one that's really needed here - at least as far as setting the jFET bias for automated VCR applications goes ... measuring that voltage doesn't tell you what Vgs(off) is and SO says nothing about the compatibility of the zener voltage and the DC voltage divider circuit it sits in to your device's control range ...

We can show from a circuit designer's point of view what needs to occur in order to maximize jFET phasor depth by following the Vischay app note on jFET-VCR's - that's assuming that using matched devices indeed means maximizing the intensity of effect in a symmetric cascade ...

first, single out 4 jFET whose Vgs(off) and Idss are as close as you can get them - how you "infer" Vgs(off) is not that difficult but most people will never take the trouble to do it accurately or properly, and hence their measurement data would be meaningless - mere tail-chasing ... one must make sure Idss and Vgs(off) are taken at the same precisely controlled test-suppy voltage also ... and also for this, you need the ability to measure/infer a 10nA current under temp-controlled conditions (see Bob Pease's EDN articles on BJT matching) - probably ignored by 100% of the people who do the "simple" test above ...  Vgs(off) can't be measured directly and it is dependent on temperature ... temperature gradient being the dominating factor in all test methodologies

if you can do all that properly, you then set the mid-point of your LFO in a precise location inside the "infered" Vgs(off) limit (through a vertical voltage offset adjustment on the LFO) and then you establish how wide across the spread of voltages you're going to vary the bias and determine how much resistance swing you want  - the lower resistance limit set by Idss ... then you don't really need a paralleling (fudging) resistor to kill the response in case you hit the limit cause now you wont if things are properly set - the whole point of doing design work really ... the obvious reason why there's limiting resistors in the circuit in the first place is because MXR couldn't expect to match their devices well enough in production so they wouldn't go over the edge in-circuit ... in effect throwing out lots of phasor range as I've mentioned before - the p90 throws out slightly less range with 22k limiters

the resistance in parallel with the jFET only limits the max resistance in the total range - it is inserted as a protection in case you send the jFET into a fully closed channel state and you don't want to hear the resulting thumping ... the resistor does NOT average out the values below that nor, worse, does it compensate for bias range mismatches between devices - I would invite you to read my recent past posts on the subject and compare to what was written above

the approach in a productions environment necessitates taking shortcuts - in this case doing the reverse and choosing a biasing zener diode to provide a rough "bias" average to accommodate a larger number of available production devices within a spec range and given control adjustment scheme ... DIY'ing brings us the opportunity to do better (with the first method), but if one insisted on doing as well/bad as that, then a different test fixture would be required - fine ... as I point out on my phase45 page one "way" is to socket-in one device at-a-time in a fixed control circuit (ie.,the actual phasor circuit the devices are going into - ie., we match the devices to the control specs and not the other way around as described above) ... this is done while varying the bias control and carefully listening to the range ... devices whose perceived "middle" position (not necessarily Vgs(off)/2 either) individually appear in a similar location on the bias control AND have similar spreads are good enough for providing a full phasor effect noting that in op-amp based all-pass circuits we can have only one stage phasing while the others act like buffers with their jFETs taken out ...

The engineered approach (starting from Idss and Vgs(off)) gives data that has direct ramifications to the design itself and determines the specs of the control circuit in order to get good/best overall performance - less phasing can always be obtained by introducing an intensity control ... serving the design/idea instead of the authenticity of a memorable circuit - thus the aim of the Vischay app-note in presenting its theory in normalized terms ...

OTOH, the ballpark approach, ie. obtaining a single NFB-based number Vgs, has no demonstrated ramification to the circuit parameters - all that matters is that they are in some proximity without mention of sensitivity - which, as we see once again, leads to cases of non-applicability ...  

you can be a man of measurement all you want but blind measurement doesn't lead to understanding ...

<AN edited 10-30-09>

[R.G.]

who you gonna trust ? Vischay or Keen's ballpark ?
you can be a man of measurement all you want but blind measurement doesn't lead to understanding ...

J.C. Why is it that you can't say anything here without making a direct attack on me?

[Mark Hammer]

Please note the timestamps in the preceding 2 posts, and JC's voluntary retraction.

Okay, now let's carry on with civil discourse about those little black things with the three bits of metal sticking out of them.