Another way to match JFETs?

[ashcat_lt]

I do, though, also wonder what the effect of a mismatch can be.  If one stage delays a given frequency by, say, 90 degrees, and the next shifts that same frequency by 100, you wont get full attenuation at that particular frequency, but doesn't there almost have to be some frequency where the two filters add to 180?  Things are a little fuzzy here today, so maybe my thinking is a bit off.  

See "The Technology of Phasers and Flangers" from 1999 at geofex: http://geofex.com/Article_Folders/phasers/phase.html

I've read that a couple times in past and just re-read to be sure.  Doesn't answer the question I asked:

As long as all the phase shifts through the stages add up at some (set of) frequency to 180 + (n * 360) where n is an integer, why does it matter that the phase shift at a given frequency is the same in all stages? 

Which then leads to the follow up:  Assuming it does matter, how close is close enough?  Or how far off is too far?  If I was at home I could run some tests.  Since I'm bored at work, I'm contributing to the derailment of this thread. 

[slacker]

I'd suggest it doesn't matter that much. If you do the "univibe mod" to a phaser so all the stages have different cap values, meaning the stages are completely unmatched, it still sounds like a phaser. Maybe not as intense an effect as one using all the same cap values but it still works.
Turkey's calling.........

[R.G.]

I've read that a couple times in past and just re-read to be sure.  Doesn't answer the question I asked:

As long as all the phase shifts through the stages add up at some (set of) frequency to 180 + (n * 360) where n is an integer, why does it matter that the phase shift at a given frequency is the same in all stages? 

There was actually a scholarly paper in JAES in about 1974 that addressed exactly your question, in the abstract of where the phase shift put the notches. The question that paper answered was whether in a multi-stage phaser, should all the capacitors be equal, concentrating the phase shift at one part of the spectrum, or spaced out to distribute the phase shift contributions over the audio spectrum.

The specific answer to your question as I understand it now is that the phase shift at a given frequency does not need to be the same for each phase stage to get notches. However, the sum of phase shifts contributed by each stage does cause a notch in frequency response where it adds up to a phase reversal compared to the original signal.  Having the phase shift the same at a given frequency for all stages produces a different spacing of cancellation notches in the frequency spectrum.

Like all musical effects, the results are good or bad only by personal taste. But having them all be similar is usually interpreted as a more intense phase effect, and spaced out, as in the Univibe more diffuse and less focused, to bend a few adjectives.

In the smaller world of JFET based phasers with identical phase caps it matters for the different reason that you want all the JFETs to be at very similar resistance all through the sweep because when one of them bottoms or tops out and the others do not, the resulting notches move or hang in odd ways in the frequency response spectrum/sweep. While having a notch move oddly in the frequency spectrum or hang at one end or the other of the range while the sweep is going on might be desirable just because it's different sounding, usually we're listening for smooth movements of the phase notches.

Really, what we're listening for (mostly!) is each JFET continuing to sweep together, not having one JFET top or bottom out so that its part of the phase contribution stops or starts suddenly before the sweep reverses. This is different from the capacitors being spaced out and the maximum contribution of the phase shift from it being at a different frequency.

[Perrow]

I've followed this thread and have a question that I don't remember being asked or answered (but I might well be wring there).

Doesn't the lfo need to be "matched" to our selected JFETs? It's all well and good to have matched FETs, but if the lfo doesn't take them from low to high resistance matching them seems a bit of a waste of time.

[armdnrdy]

Doesn't the lfo need to be "matched" to our selected JFETs? It's all well and good to have matched FETs, but if the lfo doesn't take them from low to high resistance matching them seems a bit of a waste of time.

I think that a better word would be "adjusted".

We match JFETs in hope of finding groups that "operate" in a similar fashion within the movement of the LFO. We adjust the LFO to make it work better with the set that we've selected, sort of "fine tuning" it to the Jfets. This area of "fine tuning" is generally a small window where one can move the frequency of the LFO up and down a bit causing the overall sound of the phasing to increase in intensity and change tonally.

Perrow,
You stated, "It's all well and good to have matched FETs, but if the lfo doesn't take them from low to high resistance matching them seems a bit of a waste of time."

Yes it would be a waste of time! But the LFO does take them from low to high as long as we select a set that "plays nice" within the confines of the LFO. Unfortunately, we are not privy to information pertaining to parameters that each manufacturer chose in their individual designs. For all the words that have been written on the subject of JFET matching, to me "selected" remains a very broad term. We can only go by sound samples and by "what sounds right."

The point has been brought up about "matching to what degree, and will it really make a difference."

All I can say is that if we match them to a "large" degree of points and it makes a "sonic" difference, then we can experiment to see if matching to a lesser degree is sufficient. There was a gentleman that posted a beautifully executed graph on this site. After studying the relationship of vgs to rds of this information, it appears that there are many examples clustered together where the JFETs start and end in the same zone. If we look to the middle of the chart, we find that there is little deviation. What this means to me is a three point match may be sufficient. This is yet to be seen.

Oh..And by the way, Happy Holidays everyone!

[ashcat_lt]

Thanks RG, it's kind of what I figured.  The notches in a multi-stage phaser with well-matched components should come out in harmonic series, or some other simple mathematical relationship.  If unmatched they start to get away from even multiplies of one another.  The first could be argued to be more musical because our ears like simple math.

All of this points to the reason I have never gotten into transistors.  I started looking at them for a couple of things recently, but then got wind of this thing called an OTA which can be made to do those things without all this rigamarole.  

[armdnrdy]

Ashcat,

Look to Vactrols as well. One of the best sounding phasers I've made is the Mutron Phasor II. It has six phase stages using Vactrols. It sounds very (for lack of a better word) liquid. Very lush and swirly.


I'm beating my head against the wall with this matching because I want to reproduce the sound of a couple of very unique sounding designs. To make matters worse, these designs have eight phase stages. What this means is matching eight JFETs instead of two (Phase 45) or four. (Phase 90) Needless to say, there is a lot more room for error with eight JFETs implemented as variable resistors.

[Perrow]

Yes it would be a waste of time! But the LFO does take them from low to high as long as we select a set that "plays nice" within the confines of the LFO. Unfortunately, we are not privy to information pertaining to parameters that each manufacturer chose in their individual designs. For all the words that have been written on the subject of JFET matching, to me "selected" remains a very broad term. We can only go by sound samples and by "what sounds right."

The point has been brought up about "matching to what degree, and will it really make a difference."

Seems to me that if we want to improve on RGs tester we really should test for two points, high resistance and low resistance, and then select a set with nicely overlapping ranges, then tune the LFO to work (mostly) within the selected range. With a on-off-on DPDT added to RGs tester we should be able to test that and a midpoint. I suspect that testing more than that will give very little ROI. Now the interesting part is probably to find what high and low is, and if and how much it differs in different phasers.

[R.G.]

Seems to me that if we want to improve on RGs tester we really should test for two points, high resistance and low resistance, and then select a set with nicely overlapping ranges, then tune the LFO to work (mostly) within the selected range. With a on-off-on DPDT added to RGs tester we should be able to test that and a midpoint. I suspect that testing more than that will give very little ROI. Now the interesting part is probably to find what high and low is, and if and how much it differs in different phasers.

Yep. You can switch out the fixed resistor for a higher and a lower value and have a three-point match. Probably something like 1K/10K/100K would work or 10K/51K/100K.

[PRR]

> test for two points, high resistance and low resistance

Yes but...

R.G. said:

> JFET theory somewhere that would say what the curves would look like, but I have this distrust of JFETs

I'm more trusting, but when in doubt (as we are), I'd test the MID-point as well.

I'd also pull back a bit from the 100K end. It's well-defined by the resistor. I'd find the 50K point. You may disagree (especially in light of the ill-effects when one FET hits the wall before others).

So 50K and 5K limits, the mid-point (geometric) is 15K.

--------------------

But then again, it seems aw-ful un-likely that they did computer matching in the 1970s. Not even low-paid women with voltmeters.

[alparent]

So the very non-technical question is...
What did they do in the late 70' ?
How did they pick what JFET to put in?

[armdnrdy]

I agree!

Although many good sounding phasers have been constructed utilizing this method, the way that we have been "matching" JFETs definitely needs to be reviewed. I hope to find an alternative process that doesn't include spending four hours to match eight JFETs. I suspect that was not how it was done back in the day....But....I could be wrong!

Here is part of the post from 2005 showing actual Jet Phaser JFET values as measured with R.G.s matcher.

Ok, so I pulled out the 8 "matched" transistors on the Jet Phaser in the LFO circuit, and I was expecting at least a few of them to be in a 3V sweep range (I don't know why I was expecting 3V except RG told me so  :wink: ).  Instead, I got this with RG's JFET matching circuit...
-0.791
-0.943
-0.818
-0.826
-0.879
-0.827
-0.807
-0.786

As I mentioned in the first post of this thread, I wouldn't even use this set!
I wonder if it would be feasible to choose a target JFET parameter and design a matcher that tells you yay or nay as you test individual JFETs to the target sample.

[R.G.]

I'd also pull back a bit from the 100K end. It's well-defined by the resistor. I'd find the 50K point. You may disagree (especially in light of the ill-effects when one FET hits the wall before others).

Doh! I was not clear (again) - I'd not leave the 100K in parallel when doing the testing unless I was trying to to a semi-in-circuit test. For a test rig, I'd just test for the JFET itself. 'Course, the paralleled Req could be washed out in the math, or ignored if its constant.

Although many good sounding phasers have been constructed utilizing this method, the way that we have been "matching" JFETs definitely needs to be reviewed. I hope to find an alternative process that doesn't include spending four hours to match eight JFETs. I suspect that was not how it was done back in the day....But....I could be wrong!

I strongly suspect that they did a binning test of some kind. You have someone sit down with a test circuit, a voltmeter, a bucket of JFETs to sort, and several boxes marked with the limits that some guy figured would sound good enough together. Then someone filters JFETs through the circuit and tosses each one into the appropriate bin, and all devices in the bin are considered to be matched, good enough anyway. The obvious issue is how big or small to make the ranges you mark on the bins. Most production JFETs are sorted by the manufacturer into three bins as they come out. The Japanese tend to make the same part type number with a color or letter modifier to indicate this. US makers tended to give the crude bins different part numbers, as in 2N5484, 5485, and 5486. The manufacturers tried to leave you with as big a bin range as they could sell.

Ok, so I pulled out the 8 "matched" transistors on the Jet Phaser in the LFO circuit, and I was expecting at least a few of them to be in a 3V sweep range (I don't know why I was expecting 3V except RG told me so  :wink: ). 

Sigh. There went another perfectly good speculation all shot to blazes. 

Instead, I got this with RG's JFET matching circuit...
-0.791
-0.943
-0.818
-0.826
-0.879
-0.827
-0.807
-0.786
As I mentioned in the first post of this thread, I wouldn't even use this set!

Sure you would, if you had 500 phasers to turn out by the end of the day and your boss had the food-stand hot dog with extra sauerkraut for lunch. Well, OK, if you had a bin with JFETs from 0.75V to 1.0V to work from for the 500 phasers that kept your job.    Besides, the customers will be the ones that try it out.  It's all in the size of the bin you'll accept - for whatever reason.

I wonder if it would be feasible to choose a target JFET parameter and design a matcher that tells you yay or nay as you test individual JFETs to the target sample.

You bet. Alls you gots to do is define what parameter you want to test for and what the limits on that parameter you're willing to accept.

In setting up my primal tester, I chose to test for Vgs at some current flow, nominally similar to a 10k resistor. You could as easily say that you're going to put in some Vgs, and sense what channel resistance you want. I would pick the center of my nominal bias and LFO range, and bin by resistances.

Actually, you could set any two parameters of Vgs, rds, Idss, Yfs, etc. and pick for one of the other measured things you want to know. When I did the tester, I reasoned that it was important to get the right resistance for the phase resistor, and to select on the basis of the Vgs that gave that.

The math on JFET semiconductor physics has a fair amount of interrelations of yfs, Vgs, Idss, and channel resistance, and I expect someone will pop in here in a minute  with a treatise on that. But I'm working at a much more rough and ready level. If I'm looking for phaser JFETs, I want to know what JFETs will give me the right resistance for a voltage I can put on a gate, and I don't want to spend a few minutes per JFET for a batch of 100+ JFETs trying to figure it out. Generally, a match from the original JFET matcher produces phasing, while sometimes just using unselected JFETs won't.

Getting the perfect sound from a particular circuit is a more refined requirement. However, if anyone can tell me what to look for, I'll volunteer to design a test circuit, especially if someone will build it and try it out.   

[gritz]

I agree!

Although many good sounding phasers have been constructed utilizing this method, the way that we have been "matching" JFETs definitely needs to be reviewed. I hope to find an alternative process that doesn't include spending four hours to match eight JFETs. I suspect that was not how it was done back in the day....But....I could be wrong!

Here is part of the post from 2005 showing actual Jet Phaser JFET values as measured with R.G.s matcher.

Ok, so I pulled out the 8 "matched" transistors on the Jet Phaser in the LFO circuit, and I was expecting at least a few of them to be in a 3V sweep range (I don't know why I was expecting 3V except RG told me so  :wink: ).  Instead, I got this with RG's JFET matching circuit...
-0.791
-0.943
-0.818
-0.826
-0.879
-0.827
-0.807
-0.786

As I mentioned in the first post of this thread, I wouldn't even use this set!
I wonder if it would be feasible to choose a target JFET parameter and design a matcher that tells you yay or nay as you test individual JFETs to the target sample.

Do you have a link to that thread?

And why wouldn't you "even use this set"? Gut feeling? In the absence of any technical data re: matching from Roland those fets are all you have to go on! You can draw umpteen point curves, but as has been said, it would be smart to think about their area of operation in this particular application. What range of Vgs does the LFO subject the fets to? Where does all the good sounding stuff happen (a bit subjective, but there we go)? Where's the mid point, what the extremes and what is the "typical" range? Do you still have the original fets?

I think I said this before, but a bit of general archaeology might help. Is there much general information from the time regarding component matching? Perhaps a dig through other contemporary Roland circuits might reveal some of the designers "signature" techniques. What is the tolerance of the capacitors in the allpasses? Do the factory fets have e.g. paint blobs or other identifiers on them? If not then that suggests that they were not "binned" - graded into different groups that would be used together, but simply given a "go/no go" test and the ones that failed muster were rejected (or used for another application). I can't vouch for how random factories tested components in the 1970's, but evidence of one technique vs the other might be a suggestion of how closely ~whatever params~ needed to be matched. Do we actually have any evidence that they were graded at all? Or did they just buy a certain transconductance grade?

Without any proper information it's all guesswork anyway - and you'll never be able to answer your own question to your satisfaction. But some sensible and pragmatic reverse engineering will get you ballpark figures.

[armdnrdy]

color=navy]http://www.diystompboxes.com/smfforum/index.php?topic=29076.0;wap2[/color]

And why wouldn't you "even use this set"? Gut feeling?

When I match JFETs with R.G.s matcher or the R.O.G. matcher I try to shoot for voltages that are much closer than those listed. ie: all in the .82X range instead of .732, .956, .801 etc.

In the absence of any technical data re: matching from Roland those fets are all you have to go on! You can draw umpteen point curves, but as has been said, it would be smart to think about their area of operation in this particular application. What range of Vgs does the LFO subject the fets to? Where does all the good sounding stuff happen (a bit subjective, but there we go)? Where's the mid point, what the extremes and what is the "typical" range? Do you still have the original fets?

With the two matchers mentioned above, I have extensively experimented with different ranges, different JFET types, zener values, and bias settings. That is why I started this thread. I'm out of things to try with the resources at hand.

I think I said this before, but a bit of general archaeology might help. Is there much general information from the time regarding component matching? Perhaps a dig through other contemporary Roland circuits might reveal some of the designers "signature" techniques. What is the tolerance of the capacitors in the allpasses? Do the factory fets have e.g. paint blobs or other identifiers on them? If not then that suggests that they were not "binned" - graded into different groups that would be used together, but simply given a "go/no go" test and the ones that failed muster were rejected (or used for another application). I can't vouch for how random factories tested components in the 1970's, but evidence of one technique vs the other might be a suggestion of how closely ~whatever params~ needed to be matched. Do we actually have any evidence that they were graded at all? Or did they just buy a certain transconductance grade?

There is no information that I am aware of that outlines the manner in which JFETs were matched. I've already dug through Roland's schematic archives in search of info. There is no mention anywhere that passive components were matched. In pictures I've noticed matched components marked with a dot of paint.

The evidence that they were graded is the difference in phasing. There are many sound samples on Youtube

http://www.youtube.com/watch?v=kaueqA4wNNg

that demonstrate what the Jet Phaser should sound like. My build is close....but not quite there. The phasing needs to be more intense. Even in the clean sound samples without the "jet" turned on, one can here the intensity of the phasing. As I've stated....the matching I have done with the available matchers doesn't get the phasing where it needs to be. That's why I started this thread.

Without any proper information it's all guesswork anyway - and you'll never be able to answer your own question to your satisfaction. But some sensible and pragmatic reverse engineering will get you ballpark figures.

After much research and experimentation on this build the only real question I have is the title of this thread, "Another way to match JFETs?" My question will be answered to my satisfaction when I find a different way to match JFETs, throw a set in the phaser and hear that it makes a big difference.

[armdnrdy]

R.G.

I'll build it and test it!

Here's a redrawn schematic of the Jet Phaser.
http://www.aronnelson.com/gallery/main.php/v/diyuser/Roland+AP-7+Jet+Phaser+Redraw.jpg.html

When I have my build sounding pretty close, the JFET gate voltage is -1.150 min. and -.870 max.

You'll notice that the resistors that run parallel with the JFET's source and drain are much larger than in the MXR designs.

At this point I wouldn't mind trying to match different points of resistance to VGS. That method of matching sounds logical to me.

[armdnrdy]

I thought that I would pull this thread back up with some new information.

I took RDS measurements of the "matched" JFETs in the Jet Phaser circuit. These are the readings.

Q4 min 7.7K
    max 54.6K
Q5 min 6.8K
    max 43.8K
Q6 min 6.9K
    max 17.7K    
Q7 min 7.7K
    max 19.8K
Q8 min 8.4K
    max 21.3K
Q9 min 5.5K
    max 12.9K
Q10 min 8.8K
      max 22.3K
Q11 min 6.1K
      max 14.7K

The minimum RDS doesn't appear to be that far apart but the maximum RDS has quite a bit of difference. 12.9K compared to 54.6K!

[armdnrdy]

I just contacted the member Tonemonger who lives in Tasmania.....yes Tasmania!

He has an original Roland Jet Phaser and has helped me verify the actual components.

I asked him if he would be so kind as to take some measurements of the JFET gate min/max voltage and the RDS off of the JFETs.

I'll report back when I get the measurements. This should be very telling!

[PRR]

> I'd not leave the 100K in parallel when doing the testing

Understood. What I meant was: the FET's 100K impedance is not very important. 80K or 120K is only a 10% difference when you got 100.0K in parallel. (Also through the 1970s that "100K" was 5% at best, often 10%?). The FET's ~~50K point is more important, because there the 100K R is not swamping it.

But whatever. I doubt there's any great difference 50K or 100K test. If the ~~5K and the 50K/100K points are matched, the curve is surely smooth between 50K and 100K.

> interrelations of yfs, Vgs, Idss, and channel resistance, and I expect someone will pop in here

Cold. Snowy.

It looks like *for the same area/geometry*, Idss and Vgs(on) move in the same direction. A high Vgs part will have a high Idss. The hi-Vgs part will have lower resistance for all values of bias.

Which suggests that a simple sort on Vgs (*assuming* all parts from the same mask and batch) would do the job. But that's not clear from DIY reports. So maybe the "small" interactions are emphasized in phazer use. I can see that if driven very near Vgs(off), some stages will stop sweeping before others, causing a pole in the phase/volt curve. And since we never have enough phase-stages, the "sticking" of even one would be annoying.

> -0.943 -0.786 ....I wouldn't even use this set!

I don't like it because they are all too low. The maximum audio swing is a small fraction of Vgs, 1/10th or even 1/100th. Signal must be near 10mV to stay clean. True, in Phaser only the highs of the signal appear on the FET, so it isn't the full 20mV-200mV of the guitar direct. However "highs" can be well down in the guitar fundamentals when swept to lowest pitch. I'm wondering if you want to find Vgs up in the 2V-3V range. (Which conflicts with what we pick JFETs for in Amplifier duty, and RF amplifiers is the main remaining market for JFETs.)

[armdnrdy]

I connected three jumpers from one of the JFET sockets of my Jet Phaser build to a bread board and have been testing RDS off, in circuit.

Many of the 2SK30s were previously labeled with a small square of tape marked with the VGS off according to R.G.s tester. I'm seeing quite a correlation between RDS off and VGS off, but then there are the odd ball readings that don't follow the pattern.

Paul,
I definitely see high VGS = low RDS.

The RDS measurements that I posted above are not exactly correct. The Jet Phaser has a "Depth" switch that chooses resistors from the output of the LFO to the confluence of the bias and gates. The above readings were taken with the depth switch in the shallow position. The actual RDS off readings are higher.

You had mentioned trying JFETS with higher VGS. I tried 2N5457s and 2N5458s which had higher VGS and I changed the zener to accommodate the difference but the result was less dramatic phasing then the lower VGS 2SK30s.

I'm going to find some RDS off matches in different ranges to see if it makes a difference.