lots of questions on fet matching for phasers

[Marcos - Munky]

So, it took me so much time to finally give it a go on building a phaser. I mean, I built 2 bad stones before, but it doesn't need any selected parts, so they were easy builds. This time, I'm thinking on building a phase 90. Which requires not 2 but 4 matched fets. Oh god.

I built this circuit, measured Vp of all my jfets (71, not so many of them) and wrote the values on a excel spreadsheet. Now, for the matching part, I did some researches, but still have lots of questions.

1- how close should the Vp voltages be for the jfets to be matched? I mean, I know they should be as close as possible, but how many mV should be the difference between them before I can call them not so well matched?
2- is there an optimal voltage range I should look for?
3- let's say I have a matched pair/quartet that's a bit outside of the optimal voltage range. Is there a way to make it work in a phaser?
4- let's say I have two jfets with very different parts number but the same-ish Vp. Can I call them matched, or they can't be matched due to the different parts number?

I guess that's all. For now .

[Mark Hammer]

The purpose of matching is to have all FETs changing drain-source resistance at the same time.  The range of control current that produces a systematic change in D-S resistance for FET 1 should also be "findable" in the range of FET 2, 3, and 4.  The maximum range of D-S change may be a little different in each one - 2 can go a little higher than 1, and 4 goes a bit lower than all the others, etc. - but there needs to be a common range of control current that "works" for all of them.

What that range needs to be will depend on the width of the sweep you want to achieve.  If the main use of the unit will be for medium fast to fast sweeps with modest sweep width, then it will be easy to find FETs that have overlapping ranges, even if you don't take steps to match them.  If the objective is to have fairly wide slow sweeps (i.e, you are aiming for a wide range of usable control-currents that produce D-S resistance change), then matching becomes important as a way of assuring that the expected range of change will be similar in all of them.

MXR did not include a sweep width/depth control on the Phase 90, so they aimed for a sweep width that would be suitable for both fastest and slowest speeds.  A noticeable, but not dramatic, increase in width of sweep can be achieved by replacing the 3M9 resistor from the LFO with 3M3.

[Marcos - Munky]

Thanks, Mark. I think I got some of the things sorted out.

Let's say I'm going for a fairly wide slow sweep, which I think it's the "hardest" case to achieve, because it needs well matched jfets. What's the voltage value I should look for? Now reading again my initial post, I should have said voltage value instead of range, but asking using the wrong word led me to understand a bit on how the sweep works.

I mean, I got a few different values, some around 400mV, some around 1V, the biggest one around 4V. Supposing I have a perfect matched quartet of different voltages (which of course I don't have ), how can I tell a too small voltage from a optimal voltage from a too big voltage?

[Rob Strand]

1- how close should the Vp voltages be for the jfets to be matched? I mean, I know they should be as close as possible, but how many mV should be the difference between them before I can call them not so well matched?

Around 50mV would be a good target.

2- is there an optimal voltage range I should look for?

Not really but if the original circuit wants VP=2V JFETs and you put in VP=1V JFETs the swing of the LFO is going to be be too large and will traverse a wider range of frequencies.  You can compensate without any side effects by changing the voltage divider ratio between the LFO and the JFET.

3- let's say I have a matched pair/quartet that's a bit outside of the optimal voltage range. Is there a way to make it work in a phaser?
4- let's say I have two jfets with very different parts number but the same-ish Vp. Can I call them matched, or they can't be matched due to the different parts number?

If the VPs of the JFETs are reasonably close you can simply run two bias circuits.   You are best bundling based on VP closeness that could mean three on one bias circuit and one on another instead of two and two.   If the VPs are way different it starts to get messy as you have to change the bias and the LFO swing to individual JFETs.

JC Maillet (user E7b9) has a method on his website (viva analog) to tune for different JFETs.   There's a few posts on the forum about it as well.

[Marcos - Munky]

Thanks, Rob. I have to check my spreadsheet, but I'm sure I have some pairs and quartets with less than 50mV one from the other.

The thing Is I have to check if they're readly suited for the lfo. I remember I have lots of matches around 400mV, but I have to check for other voltages. Another question, how can I know what Vp the lfo wants? I guess I could check the lfo swing using a scope, right? There's another way other than that?

Nice to suggest JC's page. I recall It's a great source of lots of info I wasn't ready to understand back then when I built my first pedal, and I'm sure I still can't understand it nowadays  . I'll check it for this info.

One thing I'm considering is to build the phase 45. Not because It's "easier" than the 90, I already have the board for the 90 almost done. But It's another nice phaser, maybe I can give my fets a good use, and one can't have enough phasers .

[kaycee]

The 45 is different enough to the 90 to be well worth the build, especially if you have taken the trouble to measure 70 jfets!

The Coloursound Phazex is another 4 stage jfet phaser, similar to the 90, but different enough to be worth building and my favourite of all the ones I have built. MH has put up plenty of mods for the 90 over the years, it's a deep rabbit hole once you have the first one going.

[Eb7+9]

hey Marcos, ...

I don't think there's anything preventing you from linearizing the p90 stages like the p45 and making the output 2/4 selectable ... the p45 does sound great for that reason

SIMPLY PUT

Vgs(off) and Idss are both key players on the control side, not to be separated ...

https://viva-analog.com/jfet-transistors-getting-to-know-vgsoff-and-idss/

normal way to engineer a recipe is to start with appropriate device specs and design a circuit around related constraints, with margins for error, yada yada ... the recipe tends to follow ingredients normally ... here it's backwards

AN INTERESTING ANALYSIS PROBLEM

personally, I prefer walking away from the semi-miraculous compactness of the original MXR designs and stretch out a little by way of an entirely new LFO - having adjustable SPEED, DEPTH and BIAS (color) - easy peasy ... but for those who insist on "cloning" the p90 verbatim let's have a closer look at things ...

ADAPTING INGREDIENTS TO THE RECIPE:

first we need to know first of all what the control side of the recipe is doing ..
have a look at the simulation results I posted here:

https://www.diystompboxes.com/smfforum/index.php?topic=116338.0

we have 300mVpp drive voltage at the gates ... verified in practices

next, as others pointed out, the jFET's are in shunting 24k resistors to gnd

let's make an assumption here, that we'd be very happy with the jFET's spanning 240k down to 2.4k ohms AC ... so, a decade above and below the grounding resistors ... this would leave us with an equivalent resistance spanning just under a decade ... making this a performance target

the question then becomes how do we match the 300mVpp bias swing from the LFO to these two VCR resistance limits

to answer this we can make use of the equations presented in Vishay  AN105

https://www.vishay.com/docs/70598/70598.pdf

rDS(on) = Vgs(off)^2/(2*Idss)

and

rDS(Vgs) = rDS(on)/(1-(|Vgs|/|Vgs(off)|))

let's make |Vgs1| correspond to 240k and |Vgs2| to 2.4k

from this we have |Vgs2| = |Vgs1| - 0.3v

if we plug these back into the rDS equation, and cancelling common terms (Idss) we get

2.4k*(|Vgs(off)|-|Vgs1|+0.3) = 240k*(|Vgs(off)|-|Vgs1|)

and we end up with |Vgs(off)|-|Vgs1| = 3.03mV

which is a very surprising result for a number of reasons

one, is it shows how fine of a control resolution is required to get that first number set

and then, even without knowing Vgs(off) per se the math tells us
the exact distance between Vgs(off) and Vgs1

of course we want to know the optimum Vgs(off) value

THIS IS WHERE THINGS TURN NOT-SO-SIMPLE

to isolate the target Vgs(off) values we need to know rDS(on)
or, in effect, Idss ...

so then, Idss plays a role in setting the optimum Vgs(off) values

we start out with a fixed LFO drive signal and resistance targets associated with its limits ... these are the only constraints to the problem

observation: since rDS(on) is a quadratic function of Vgs(off) and Idss follows a largely linear trend against Vgs(off) (https://viva-analog.com/characterizing-and-matching-2n5457-jfet-transistors/) we see that rDS(on) must be linear - or proportional - to Vgs(off) ... (see numbers down below)

if we plug in a value for rDS(on) in the rDS equation corresponding to Vgs1 we get

|Vgs(off)| = 3.03mV*(240k/Rds(on))

if we take some of my Fairchild 2n5457 data pairs as examples we get the following:

Vgs(off) = -1.01v ... Idss = 1.75mA ...> Rds(on) = 291 ohms AC
Vgs(off) = -1.44v ... Idss = 3.10mA ...> Rds(on) = 334 ohms AC
Vgs(off) = -1.61v ... Idss = 3.42mA ...> Rds(on) = 379 ohms AC

but according to the prior equation optimal |Vgs(off)| values are

2.49v, 2.18v and 2.15v correspondingly ...

as we can see the numbers are heading for a convergence ... somewhere around 2volts
but that's going from my small lot of Fairchild 2n5457's

with different device types (ie., different rDS(on) function profile) the optimum Vgs(off) values will vary ... bummer, I know

---

it's not hard to see that these targets are basically impossible to reach in real life unless we have a factory that can sort out devices for us, if we have an estimate for rDS(on) values ... a little more math would determine the range limits, etc ...

the bottom line, Idss will determine the optimal value of Vgs(off) so that a 2.4k to 240k resistance range will be produced from a 300mVpp variation in Vgs voltage

I said it wasn't simple ...

hate to say it, but because of these constraints anybody who tries cloning a p45 or p90 may very likely end up building a somewhat sub-par sounding circuit simply because these constraints cannot be met ... by how much though, that's a tougher problem especially since it then becomes so subjective ... going by math targets is the only way to say really

---

imo a somewhat better, but albeit less compact, approach to achieving precise limit operation from matched jFET's is to redesign the LFO so that everything is now independently variable and then tune BIAS and MAX DEPTH by ear ...

otherwise one can most unlikely get full benefit from the devices taking this reverse approach ... this is especially so going from a very restricted set of devices  ... imagine, MXR had a Vgs(off) target value to go along with their choice of device ie., with known Rds(on) targets ... it's the only way to make this circuit manufacturable AND make it sound strong while doing it

on the flip side, this is why I started looking at ways of making jFET phasors work using devices with largely un-matched (random) Vgs(off) values ... I now have a second way of doing it and, of course, it requires proper device specific adjustments on each stage ... considering how ludicrous it is to try matching jFET's with any serious accuracy I think this is a better approach to reliably nailing controlled resistance targets ... of course, accurate measurement of Vgs(off) is crucial here ...

but, we'll see how much work it is in the end ...

plus, as a bonus, I can easily get 4 orders of magnitude or more instead of the above 2, which means increasing the grounding resistors a getting much wider phasing ranges, etc ...

of course, variations in rDS(on) between devices can be normalized by simply shifting capacitor values ... but this aspect is forgiving anyway

as far as Cv control goes Vishay AN-105 makes it very clear that the only thing that matters is a Vgs voltage lying inside the |Vgs(off)| to zero range ... how one gets there the jFET cares naught

best of luck!

[Rob Strand]

Another question, how can I know what Vp the lfo wants? I guess I could check the lfo swing using a scope, right? There's another way other than that?

If you start with circuit that works with known a JFET the best you can do with that information is to look up the typical VP in the datasheet for the specified JFET.   

You then take your VP measurements and scale the LFO output up or down in proportion to your JFET's VP.    Suppose VP from the datasheet is 2V.  If you JFETs have VP 0.4V that means you need to reduce the LFO swing by a factor of 0.4V/ 2V = 0.2.   You need to adjust the voltage dividers to the JFET to be 0.2 *on top of* what is already there in the circuit.

One subtle point.  The JFET matcher on the runoffgroove site, will produce a Vgs measurement which is very close to JFET's VP value.
http://runoffgroove.com/fetzervalve.html
But for RG's JFET matcher
http://www.geofex.com/Article_Folders/fetmatch/fetmatch.htm
You need to multiply the Vgs measurement by about 1.4 to 1.5 to get an estimate of the JFET's VP value.

[Marcos - Munky]

Kaycee, the Colorsound PhaseX is indeed very interesting. I already have the Phase90 board almost populated, but I'll keep the the PhaseX in mind.

JC, I indeed could use a switch to select between 2 and 4 stages. But I prefer to make them two builds, to keep me busy for a longer time. I also have plans on using your Phase45 mods, and I'll probably won't be able to fit so many extra switch in a 1590A. And indeed the lfo can be improved/swapped, I'd just want to give it a go on building the original circuit because it's a "classic build" that I didn't built yet since I built my first pedal. So I can say I've tried it at least once.

Your explanation was so in depht, and some things were very far from my understanding lol. But indeed it's very complicated to achieve a very good result, and there are the design flaws on the original circuit.

Rob, got it. So basically, instead of picking a jfet for the lfo, I'll adapt the lfo to suit my jfets.

---

On my measurements, I found a pair of NTE458 with Vp around 550mV, two pairs of 2N5456 with Vp of 3.7V and 4V, 3 pairs of NTE312 with Vp of 1.7V, 2.2V and 2.6V, and a pair of J201 with Vp of 890mV. Those values using a difference of less than 50mV one from another.

For my 2N5457, the measurements were way better. They were so close from each other. I was able to get 3 quartets (430mV, 440mV and 480mV), one pair (470mV), one septet (420mV), one octet (450mV) and one group of 10 (460mV), that's with a difference of less than 10mV one from the other. For a difference of 20mV, I can match another 2 pairs. And for a difference of 60mV, I have a pair with Vp around 1.1V. I'll have to scale down the lfo output to use any of those jfets tho. And I unfortunally didn't measured any Idss.

[Eb7+9]

without getting into the details again, the two matching jigs referred to above aren't good for getting accurate Vgs(off) values, this has been discussed to death ... what these two tests are good for is getting very ballpark numbers

another way to get ballpark "yay/nay" is by sticking one jFET at a time anywhere in the phasor sections (use sockets) and playing with the bias control to see where onset of phasing starts to happen ... if you get two or more devices that kick in roughly at the same spot you can consider them "matched" ... well enough for the circuit in question anyway

I stumbled on this approach years ago when I was making p45's and I know others have tried it with success as it's very dependable and much quicker than building a jig, dealing with numbers, etc ...

http://www.lynx.net/~jc/pedalsPhase45.html

[Marcos - Munky]

I saw the approach on your page some time ago, and saw it again when I decided to give it a try on a jfet phaser. I'll probably go with it, since I'm not a theoretical/numbers guy I prefer to build it and see what happens.

Anyway, I already etched a Phase45 board yesterday, I'll cut it and drill it today. It'll be easy to find a roughly matched pair, after that I can try to get a quarted of roughly matched jfets. Still not sure if I have all the parts to build it, I know for sure I don't have 150K resistors. But, if I don't have parts for the Phase45, I'm almost sure I have everything for a DOD Phasor 201.

[POTL]

This topic appears very often, it would be cool to do one useful thing, to find the transistor model that will have the minimum parameter spread. Perhaps I would have assembled this pedal, with the mods I need, but I see no reason to buy about 100 transistors and spend the whole day sorting, there is MXR 95 in nature and I think buying it will be cheaper and better than spending money and time on assembly of outdated pedals.

[Marcos - Munky]

And where's the fun on just buying the pedal itself or buying proper jfets?

Don't take me the wrong way. I'm barely playing, and went from like 20-25 builds per year to 1-2 builds per year. I have lots of unused parts, so it's more expensive to buy more of then than trying to use what I already have, and a total waste of money to buy a (very expensive down here) pedal to add it to my pedalboard but not use it because I'm not playing. I agree it's a waste to buy 100 jfets looking for the magic pair/quartet, but the thing is I already have those ones, that's why I'll go thru the hard work on sort them one by one. To me it's more useful to try to use them than buying new ones. And also that's something I didn't tried before, so it'll be a new experience to me.

I'm doing nothing for the rest of week, we got a few days off at work because of covid. And I just went from the hospital to do the covid test, results will be out tomorrow, and if I get a positive I'll stay locked in my room for more 10 days, so I'll have plenty of time to test those jfets .

[Mark Hammer]

I'll just add that using a couple of fixed allpass stages to supplement swept stages is a convenient way to add more phase shift without having to match more FETs.

And once again, I'll note that the need to match FETs depends on how wide a sweep one seeks.  More modest sweep widths can do fine with unmatched or loosely-matched FETs.

[Marcos - Munky]

I think a modest sweep width is all I need, just want to give it a go on get the effects built the best they could be. Both boards (phase 90 and dod 201) almost finished, but I'm lacking some parts for both of them and stores are closed. Hopefully they'll be open the day after tomorrow, so I can get the missing parts.

[Marcos - Munky]

Just an update. After a few days, I finally managed to finish both boards and test them.

First, I built a Dod Phasor 201. I used sockets for the jfets, so I could try a few of them. I tested different pairs, and set for a pair of 2N5457 with Vp of 470mV. The ones with Vp around 2V gave me about the same results, so I saved them and used the 750mV ones because I have more of them than the other ones. I used JC's method of putting one at a time and check the bias trimpot to confirm they are matched enough for the circuit. It sounds so beautiful! I don't think it is subtle as some people claims. This one will get boxed soon.

Then I got a quartet of 2N5457 with Vp of 480mV. Checked the matching again by putting one at a time on my Phasor 201, and finished the Phase 90. I built the Block version, with the feedback resistor. To be really honest, I was expecting more out of it, since I liked the Phasor 201 so much. The phaser sound itself is kinda weak, not so pronounced. It's not the bias trimpot setting, so I guess it's just a matter of playing with the mixing resistors' value. But the most strange thing is the lfo ramp. The ramp up part is fine but the ramp down part is way faster, it seems like the ramp down have twice the speed of ramp up. That said, I used a 10uF cap for the lfo instead of a 15uF, so this could be what is causing this issue. I'll try to add a 4.7uF cap in series to see if it solves this.

[box]

Hi Marcos,
you can buy such a tester and choose the right one for your phaser:
https://www.elektor.com/lcr-t4-esr-meter-transistor-tester-diode-triode
Regards,
box

[Marcos - Munky]

Box, I have one of those, and indeed it's one of the most useful tools I got! It measures loads of stuff. But the jfet measurement can't be trusted at all, at least in mine. It reads some jfets as npns. That could indeed mean I got a bag full of fake jfets, but I checked some jfets I'm sure are real (some NTEs) and a few of them were identified as jfets while the other ones were identified as npns. That said, mine is an old version, I know there was at least one update that can read leakage for ge transistors and mine can't. And mine doesn't have a really fresh battery installed, it reads around 8.3V and I don't have any fresh batteries here to compare the results, so this wrong reading could be caused by a discharged battery, or it could just be something that was corrected on later versions. Anyway, I still say it's an awesome tool, it just need to be careful on some readings.

Back to the Phase 90, I installed a 4.7uF in parallel with the 10uF I have on the lfo, and now the ramp down is better. Still a bit faster than the ramp up, but way better. For the "weak phase sounds", I didn't played around with the mixing resistors, instead I turned up the volume of my amp a bit (well, actually a lot ), and it sounded way better than yesterday! I think the "issue" was the low volume due to be kinda late to play guitar when I tested it. So I'm pretty happy on how it sounds right now and I'm set with this build, just have to box it. Waiting until I make the artwork for the Phasor 201, so I'll box them on the same day.